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
Subcellular localization critically influences protein function , and cells control protein localization to regulate biological processes . We have developed and applied Dynamic Organellar Maps , a proteomic method that allows global mapping of protein translocation events . We initially used maps statically to generate a database with localization and absolute copy number information for over 8700 proteins from HeLa cells , approaching comprehensive coverage . All major organelles were resolved , with exceptional prediction accuracy ( estimated at >92% ) . Combining spatial and abundance information yielded an unprecedented quantitative view of HeLa cell anatomy and organellar composition , at the protein level . We subsequently demonstrated the dynamic capabilities of the approach by capturing translocation events following EGF stimulation , which we integrated into a quantitative model . Dynamic Organellar Maps enable the proteome-wide analysis of physiological protein movements , without requiring any reagents specific to the investigated process , and will thus be widely applicable in cell biology . The hallmark of eukaryotic cells is their compartmentalization into distinct membrane-bound organelles . Protein function is critically determined by subcellular localization , as organelles offer different chemical environments and interaction partners . In order to regulate protein activity , many biological processes involve changes in protein subcellular localization . Prominent examples include the endocytic uptake of activated plasma membrane signalling receptors , to terminate the signalling process ( Jones and Rappoport , 2014 ) , and the nucleo-cytoplasmic shuttling of many transcription factors , to regulate their access to DNA ( Plotnikov et al . , 2011 ) . The ability to monitor changes in organellar composition would provide a powerful tool to investigate cell biological processes at the systems level . While transcriptomic ( Curtis et al . , 2012 ) and proteomic abundance profiling approaches ( Deeb et al . , 2015 ) have yielded valuable insights into changes in gene or protein expression , they lack the important spatial dimension . Microscopy-based approaches can provide spatial information on individual proteins ( Uhlen et al . , 2015 ) , but are limited by the availability of specific antibodies , and are very labour-intensive for analysing complete proteomes ( Marx , 2015 ) . Genome-wide GFP-tagging in yeast circumvents the need for antibodies ( Huh et al . , 2003 ) , but tags may inadvertently alter protein subcellular localisation , which is difficult to control for; in addition , serial imaging of cells for comparative purposes remains experimentally challenging ( Breker et al . , 2013 ) . Mass spectrometry-based proteomics has much enhanced our understanding of cellular composition ( Larance and Lamond , 2015 ) . Although sophisticated approaches for organellar proteomics have been available for over a decade ( Andersen et al . , 2003; Christoforou et al . , 2016; Dunkley et al . , 2004; Foster et al . , 2006; Gilchrist et al . , 2006; Smirle et al . , 2013 ) , there is currently no proteomic method that allows global dynamic mapping of protein subcellular localization . The main reason for this deficiency is the high variability between spatial proteomics experiments , which renders the identification of genuine organellar transitions very difficult ( Gatto et al . , 2014 ) . Here , we have developed a rapid proteomic profiling workflow for the generation of highly reproducible organellar maps . We use the method to assemble a comprehensive database of protein subcellular localization and abundance information from HeLa cells , allowing us to build a quantitative model of cellular anatomy . We then apply organellar maps to capture the protein translocation events triggered by EGF stimulation , demonstrating the dynamic capabilities of our approach . The principle of our approach is to separate organelles partially with a minimum number of fractionation steps , and to generate organellar profiles by high-accuracy quantification of each fraction against an invariant reference . Metabolically labelled HeLa cells , SILAC light or heavy ( Ong et al . , 2002 ) , were mechanically lysed following gentle hypo-osmotic swelling ( Figure 1A ) . Damage to organelles was minimal , as assessed by leakage of lumenal contents ( Figure 1—figure supplement 1A ) . Post-nuclear supernatant from light cells was fractionated by a series of five differential centrifugation steps , whereas a total organellar ‘reference’ fraction was obtained in a single centrifugation step from heavy post-nuclear supernatant . This procedure is highly reproducible , as assessed by protein recovery ( Figure 1—figure supplement 1B , C ) . Each light sub-fraction was then combined with an equal amount of the heavy reference , subjected to tryptic digest and analysed by LC-MS/MS . For each protein , we obtained an abundance distribution profile across the sub-fractions . In a typical experiment , approximately 4500 proteins were profiled . Proteins associated with the same organelle have similar profiles , and organelles can be distinguished from one another ( Figure 1B ) . In parallel , the global distributions of proteins across the nuclear , organellar , and cytosolic fractions were obtained by label-free quantification mass-spectrometry , typically covering 8000 proteins ( Figure 1C ) . 10 . 7554/eLife . 16950 . 003Figure 1 . Generation of organellar maps through fractionation profiling . ( A ) Metabolically labelled HeLa cells were mechanically lysed to release organelles . Light labelled lysate was then subjected to differential centrifugation at the indicated speeds ( RCFMAX ) and times ( in minutes ) . Heavy-labelled lysate was centrifuged twice , once at low speed to generate a nuclear-enriched pellet , and again at high speed to generate the organellar pellet; the supernatant was the cytosolic fraction . The heavy organellar ‘reference’ fraction was combined with equal protein amounts of each of the five light membrane sub-fractions and analysed by mass spectrometry , generating SILAC ratios for each protein in all fractions . ( B ) The SILAC ratios were converted to enrichment over reference . Median values of organellar marker proteins were plotted , showing clearly distinct profiles . ( C ) In a parallel analysis , the heavy-labelled nuclear , organellar and cytosolic fractions were subjected to label-free mass spectrometric analysis , revealing the global distribution of proteins across these three fractions . Examples of normalized profiles of marker proteins for the nucleus ( Histone H3 ) , lysosome ( Cathepsin D ) and the cytosol ( Pyruvate Kinase ) are shown . Bars show mean ± SD ( n = 6 ) . Please refer to Figure 1—figure supplement 1 for organellar leakage analysis and evaluation of fractionation yield reproducibility . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 00310 . 7554/eLife . 16950 . 004Figure 1—figure supplement 1 . Organellar leakage analysis ( A ) and fractionation reproducibility ( B , C ) . ( A ) Leakage of lumenal contents from endoplasmic reticulum , mitochondria and lysosomes was calculated by quantifying the cytosolic pool of lumenal marker proteins ( see Figure 1C , and Materials and methods for further details ) . For each organelle , a distribution of values was obtained . In all cases there was a large pool of proteins that showed no leakage ( <1% ) . These are probably attached to the organellar membrane , or part of a larger assembly , and thus cannot leak . Conversely , there was a very small number of proteins with very high values ( >20% ) ; these are likely to have a genuine cytosolic pool , possibly caused by a cytosolic splice variant not discriminated by the mass spectrometry . In between , there were proteins showing a range of values ( 1–20% ) ; they are likely to reflect actual organellar leakage . Averages calculated from these middle values are 8 . 3% for ER , 3 . 9% for mitochondria , and 2 . 3% for lysosomes . These very low values suggest that organelles are largely intact . ( B ) The protein yields of each of the differential centrifugation fractions ( see Figure 1A ) were calculated using a BCA assay . Yields were converted to % by dividing each fraction by the total yield . This allowed independent experiments to be combined; error bars show the small standard deviations of 6 experiments , highlighting the high reproducibility of the fraction yields . ( C ) The protein yields of the nuclear , membrane and cytosolic fractions were calculated as in ( B ) , and the small standard deviations of six experiments reveal a similarly high yield reproducibility . In B and C , bars show mean + SD , n=6 . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 004 Following this scheme , we prepared six replicate fractionations , in batches of two , on three different days . We first considered the set of 3766 proteins common to all replicates , and applied principal component analysis ( PCA ) to their abundance profiles . The PCA scores plot was then overlaid with established organellar markers ( Supplementary file 1 ) , which clustered into distinct regions of the plot ( Figure 2 ) . Resolved compartments included plasma membrane , endoplasmic reticulum , ERGIC , Golgi apparatus , endosome , lysosome , peroxisomes and mitochondria , as well as a cluster of diverse large protein complexes , such as ribosomes and proteasomes . Closer inspection of the marker proteins suggested further sub-organellar resolution , revealing a partial divide between ER membrane and lumen , as well as division of mitochondria into matrix , inner membrane , and outer membrane ( Figure 2—figure supplement 1A ) . For independent validation , we overlaid the scores plot with UniProt annotation for subcellular targeting features , including signal peptides , mitochondrial transit peptides and transmembrane domains , and observed near-complete agreement with our maps ( Figure 2—figure supplement 1B ) . To assess the reproducibility of the method , we next analysed the six individual maps by PCA; all had very similar patterns ( Figure 2—figure supplement 2A ) , with organellar clusters occupying stable positions between maps . The SILAC ratios of replicate fractions were also highly reproducible ( average correlation > 0 . 95; Figure 2—figure supplement 2B ) . 10 . 7554/eLife . 16950 . 005Figure 2 . Visualization of an organellar map . Thirty SILAC ratios from six replicate fractionation experiments were combined and subjected to principal component analysis to achieve dimensionality reduction . Projections along the first ( x-axis ) and third ( y-axis ) principal components ( PCs ) provided the optimal separation of clusters . Each scatter point represents a protein . Proximity of proteins indicates similar fractionation behaviour . Marker proteins for organelles are coloured as indicated in the legend , and reveal clustering of proteins belonging to the same organelle . Non-marker proteins are shown as small grey dots . PCs 1–3 account for 64% , 21% , and 12% of the variability in the data , respectively . Please note that the actual resolution of the map is much higher than is apparent in this 2D representation of the full 30-dimensional data set , and most of the seemingly overlapping clusters are in fact separated . Please refer to Figure 2—figure supplement 1 for more detailed cluster annotation , and overlays with external protein sequence feature predictions . Figure 2—figure supplement 2 shows the reproducibility analysis of six replicate organellar maps . The complete organellar assignments , spatial and abundance information can be found in Supplementary file 1 ( compact format ) and 4 ( interactive database ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 00510 . 7554/eLife . 16950 . 006Figure 2—figure supplement 1 . Organellar map with full cluster annotation ( A ) ; overlay of an organellar map with external protein sequence feature predictions ( B ) . ( A ) Close inspection of the map shown in Figure 2 reveals sub-clustering within the main clusters . The mitochondrial cluster shows separation into outer membrane , inner membrane and matrix proteins; the endoplasmic reticulum is separated into lumenal and membrane proteins . Furthermore , numerous large protein complexes show very tight clustering; a few examples are annotated . In many cases the resolution of this map is sufficiently high to predict constituents of a complex by a ‘neighbourhood analysis’ ( Supplementary file 4 ) . Note that the actual resolution of the map ( in full dataspace ) is much higher than apparent from this 2D principal component analysis . ( B ) The organellar map shown in Figure 2 was coloured according to UniProt annotations for proteins with transmembrane domains , mitochondrial transit peptide , or signal peptide . The transmembrane domain annotation is almost completely absent from the large protein complex area of the plot , as would be expected . Moreover , the mitochondrial transit peptide annotations cluster , and overlap with our independently derived mitochondrial cluster . Conversely , the signal peptide annotation overlaps with membrane organellar markers , except mitochondria , as would be expected . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 00610 . 7554/eLife . 16950 . 007Figure 2—figure supplement 2 . Reproducibility analysis of organellar maps . ( A ) Six individual maps ( with five SILAC ratios each ) were visualized by PCA as described in Figure 2 . Maps were made in pairs ( 1&2 , 3&4 , 5&6 ) , on three separate days . Notice the highly reproducible pattern of all maps . ( B ) Pearson correlation of log2 SILAC ratios in equivalent subcellular fractions from six replicate maps . The average fraction correlation is reported for all 15 pairwise comparisons . The correlation is very high in all cases , and almost identical for intra-day ( bold text ) and inter-day comparisons . ( C ) Map concordance , ie the proportion of identical organellar predictions between two replicate maps , shown as a function of prediction confidence . The averages from the three intra-day comparisons ( black dots ) , and 12 inter-day comparisons ( red dots ) are shown . For example , for two maps made on the same day , 93 . 7% of all predictions are identical . If a stringency filter is introduced ( eg confidence score >8 ) , which retains 77% of the predictions , concordance is increased to 98% . Remarkably , concordance for maps made on different days is almost as high . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 007 For the rigorous assignment of proteins to organellar clusters , we used a support vector machine ( SVM ) -based supervised learning approach . Briefly , SVMs allow non-linear separation of clusters ( Varmuza and Filzmoser , 2009 ) . Optimal boundaries between organellar clusters are determined using marker proteins , with cross-validation to prevent over-fitting . Non-marker proteins falling within the boundaries of a particular cluster are then assigned to that organelle . Since a suitable canonical set of organellar markers was not available , we manually curated a set of over 1000 proteins ( Supplementary file 1 ) . We chose markers based on their expression in HeLa cells and their well-documented ( and ideally unimodal ) localization to a particular organelle . Clustering of these markers was visually confirmed with several PCA-based pilot maps ( not included in this study ) . Where necessary , we specifically chose further established markers near the edges of organellar clusters , as these are particularly important for defining boundaries . We applied SVM classification to all six maps individually . The mean prediction accuracy for marker proteins was 94 . 7% ( with full cross validation ) , demonstrating the exceptionally high level of organellar resolution achieved ( Supplementary file 2 ) . While marker prediction accuracy does not provide a direct measure of overall prediction accuracy , it nevertheless serves as a useful estimate ( see Methods for further details ) . The average proportion of identical organellar assignments between maps , referred to as concordance , was 93 . 7% for all proteins , and >98% for three-quarters of the proteins ( Figure 2—figure supplement 2C ) . Collectively , these data show that fractionation profiling is effective for generating high resolution organellar maps . The remarkable level of reproducibility enables comparative applications ( see below ) . We combined the predictions from the six replicate maps into a single output ( see Methods for details ) . In total , we derived organellar profiles for 5265 proteins , of which 2423 were assigned to 9 membranous organelles , with 96 . 5% of marker proteins predicted correctly ( 92 . 7% average per membrane-bound organelle; Table 1 ) . To validate the novel predictions , we removed the organellar markers from the set , and annotated the remaining proteins with UniProt subcellular localization information . A Fisher’s exact test showed that for eight of the nine compartments , the most significantly enriched localization term corresponded to our own organellar classification ( Supplementary file 3 ) . Furthermore , we compared our mitochondrial predictions with the MitoCarta database of experimentally validated mitochondrial proteins ( Calvo et al . , 2016 ) ; the overall concordance was 97% ( 92 . 3% for non-marker proteins ) . These data provide strong independent support for the high quality of our organellar assignments . Organellar maps deliberately exclude the cytosolic fraction , since numerous peripheral membrane proteins have a soluble as well as a membrane-bound pool . Inclusion of the cytosol in the maps would reveal which proteins are predominantly cytosolic but sacrifice information on the precise localization of the membrane-associated fraction . Therefore , the maps were augmented by an auxiliary workflow , which reveals the nuclear-organellar-cytosolic distribution ( Figure 1C ) . In total , this global profile analysis extends to 8710 proteins , including 1999 cytosolic , 1133 nuclear , and 672 nucleo-cytosolic proteins ( Supplementary file 1 ) . 10 . 7554/eLife . 16950 . 008Table 1 . Prediction output and performance of HeLa organellar maps . The table shows the combined organellar prediction output from six replicate maps from HeLa cells . Prediction performance is judged by the proportion of correctly assigned organellar marker proteins . Please also refer to Supplementary file 1 ( compact format ) and 4 ( complete database ) , which contain detailed information for all 8710 proteins covered in this study , including nuclear and cytosolic predictions . Supplementary file 2 shows the performance of each individual map . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 008CompartmentNumber of marker proteinsCorrectly predicted markersAll proteins predicted in this compartmentNumber%Endosome857588 . 2%304ER127127100 . 0%530ER , high curvature1111100 . 0%45ERGIC/cisGolgi262596 . 2%73Golgi332987 . 9%190Lysosome434195 . 3%88Mitochondrion24223998 . 8%658Peroxisome211571 . 4%25Plasma membrane12712396 . 9%510All organellar proteins71568595 . 8%2423Average per organelle92 . 7%Large Protein Complexes36135397 . 8%2739Total1076103896 . 5%5162 We combined all data into a database , which contains three layers of information . At the global level , it includes the distribution of each protein between nuclear , organellar , and cytosolic pools , as well as copy numbers per cell and cellular concentrations ( calculated with the ‘proteomic ruler’ approach [Wisniewski et al . , 2014] ) . At the organellar level , predictions of compartment associations are provided , with confidence scores . Furthermore , maps have high local resolution; this third level of information provides the ‘neighbourhood’ of a protein , revealing which other proteins have similar fractionation profiles . In many cases , this allows identification of stable protein complexes . The database is accessible via a web interface ( www . MapOfTheCell . org ) , and as an interactive Excel file ( Supplementary file 4 ) ; Supplementary file 1 contains a compact summary of the organellar predictions and copy numbers . The website allows visual exploration of the individual organellar maps . Combined knowledge of protein subcellular localization and abundance enables construction of a model of HeLa cell composition . We calculated the protein mass of each organelle by multiplying the molecular weights of constituent proteins by their estimated copy numbers ( Figure 3 ) . This revealed that the endomembrane system contributes approximately 16% to total cellular protein mass , dominated by mitochondria ( 6 . 6% ) , ER ( 4 . 4% ) , and plasma membrane ( 3 . 1% ) , with relatively minor contributions from endosomes , lysosomes , peroxisomes and Golgi ( Figure 3A ) . The mitochondria , ER and plasma membrane are themselves dominated by a few highly abundant proteins ( Figure 3B ) . In each case , the 20 most abundant proteins constitute at least 40% of organellar protein mass ( Figure 3C–E , and Supplementary file 5 ) . For example , the most abundant plasma membrane protein is the 4F2 cell-surface antigen heavy chain ( SLC3A2 ) , with 15 million copies/cell . This versatile protein can heterodimerize with several other proteins ( eg SLC7A5 , another very abundant protein , three million copies ) to form amino acid transporters . This predominance probably reflects the adaptation of HeLa cells for fast nutrient uptake to support rapid growth . Supporting this view , all plasma membrane transporters combined ( 40 million copies ) contribute approximately 25% of the total compartment protein mass . Other integral membrane proteins ( such as adhesion and signalling receptors ) contribute ~30 million copies . Assuming a cell surface area of ~1600 μm2 typical of adherent HeLa cells ( Fisher and Cooper , 1967 ) yields an estimated density of 4–5 integral membrane proteins per 100 nm2 , in excellent agreement with a previous biochemically derived estimate of 3 for baby hamster kidney ( BHK ) fibroblasts ( Quinn et al . , 1984 ) . Within the endoplasmic reticulum , proteins involved in protein folding and quality control predominate ( 20% chaperones , 10% protein disulfide isomerases ) . A similar abundance of chaperones was observed in the mitochondria ( 14% ) , which exceeds the collective contribution of citric acid cycle components ( 9% ) . We detected the five members of the mitochondrial ATP synthase F0 catalytic complex with the expected stoichiometry of ~3:1 for subunits A/B to C/D/E , and estimate the number of complexes at ~3 million per cell ( 5% of mitochondrial protein mass ) . Thus , a picture of HeLa cell anatomy emerges from the quantitative subcellular localization information . 10 . 7554/eLife . 16950 . 009Figure 3 . Quantitative anatomy of a HeLa cell . ( A ) Schematic diagram of a cell where compartments are approximately scaled by their relative contributions to total cell protein mass ( not by their volumes ) . All membranous organelles combined ( excluding the nucleus ) contribute ca . 16% . For comparison , ribosomes and proteasomes contribute 6% and 1 . 3% , respectively . The proportion of the ER fraction would increase from 4 . 4% to ca . 5 . 4% if attached ribosomes were included . ( B ) Proteins of major organelles were ranked in order of decreasing abundance , and plotted against their cumulative mass . Very few proteins contribute the majority of organellar protein mass in all three cases . ( C–E ) Top 20 most abundant proteins in each of the three major organelles , plotted against their contribution to protein organelle mass . The complete quantitative composition of ER , mitochondria , and plasma membrane are shown in Supplementary file 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 009 The very high reproducibility of our approach opens the possibility to compare maps under different physiological conditions , to identify protein translocation events . To test this , we investigated the well-characterized process of epidermal growth factor receptor ( EGFR ) uptake . Following stimulation with EGF , EGFR autophosphorylates , binds downstream factors , and is rapidly endocytosed from the plasma membrane to an endosomal compartment ( Jones and Rappoport , 2014 ) . The translocation process is readily imaged using fluorescently-labelled EGF ( Figure 4A , B ) . We prepared organellar maps from untreated ( control ) HeLa cells and from HeLa cells continuously stimulated with EGF for 20 min , in biological triplicate ( Figure 4C , D; Figure 4—figure supplement 1 provides a schematic of the experimental design; Figure 4—figure supplement 2A shows all six maps ) . Overall map morphology from treated and control cells was unchanged , however EGFR , which localized to the plasma membrane cluster in control cells , was localized to the endosomal cluster upon EGF treatment , as expected . To identify subcellular translocations in an unbiased manner , we developed a two-stage statistical analysis . For each protein the magnitude of translocation ( Movement score ) as well as the consistency of the direction of the translocation across biological repeats ( Reproducibility score ) were assessed . The two metrics were then combined in a ‘MR’ plot ( Figure 4E , F ) to identify proteins undergoing consistent translocations . To derive stringent score cut-offs , we took advantage of the maps used to generate our subcellular localization database ( Figure 2—figure supplement 2A ) . We treated these six maps as a mock experiment in which we would not expect to detect any specific changes , by assigning three maps as 'controls' and three as 'mock-treated' . We determined the most stringent score cut-offs from the MR plot of the mock experiment by defining a region where no false positives were obtained . Applying these cut-offs to the EGF treatment experiment identified four proteins as significantly translocating; EGFR , GRB2 , SHC1 and PKN2 . Both GRB2 and SHC1 are recruited to EGFR upon EGF stimulation and constitute the first step in EGFR signaling ( Oda et al . , 2005 ) . Inspection of the maps ( Figure 4C , D , Figure 4—figure supplement 2A ) , and classification with support vector machines ( Supplementary file 6 ) showed that all proteins had moved to the endosome/lysosomal compartment , as expected . Therefore , our approach correctly and specifically identified the major translocation events following EGF stimulation . For a deeper exploratory analysis , we then relaxed score cut-offs to allow an FDR of 10% , and identified 14 further significantly translocating proteins ( Supplementary file 7 ) . These included numerous known downstream targets of EGF , such as RPS6KA3 , PIK3C2B and ROCK2 , as well as several new candidates ( see Discussion ) . These results validate the use of dynamic organellar maps for the systematic detection of subcellular translocation events . 10 . 7554/eLife . 16950 . 010Figure 4 . Dynamic organellar maps reveal protein localization changes following EGF stimulation . ( A , B ) Fluorescently tagged EGF ( green ) was pre-bound to HeLa cells on ice , and imaged by confocal microscopy . Lysosomal compartments were visualized with Lysotracker ( red ) . Most of the EGF was at the cell surface ( A ) . Cells were then shifted to 37°C , and incubated for 30 min . EGF had been cleared off the cell surface , and localized predominantly to an endosomal compartment , with little lysosomal co-localization ( B ) . Scale bars = 10 μm . ( C ) Organellar maps were prepared from untreated HeLa cells , and ( D ) from cells following 20 min of continuous stimulation with 20 ng/ml EGF . The translocation of the EGFR receptor ( star symbol ) from plasma membrane to endosomes was captured . Colours indicate organelles as in Figure 2 . Maps show the combined data from three replicates each . ( E , F ) Unbiased identification of significant translocation events triggered by EGF stimulation . Each protein is scored for magnitude of translocation ( M score , x-axis ) and reproducibility of translocation direction ( R score , y-axis ) across the three replicates . A MR plot reveals significant translocations in the top right quadrant . Score cut-offs for FDR-control were determined by analysis of a triplicate mock experiment where no genuine translocations are expected ( E ) . Ultra-stringent cut-offs ( corresponding to an FDR of 0 ) were then applied to the EGF treatment experiment ( F ) . Four significant translocations were detected , including EGFR and two known binding partners , GRB2 and SHC1 . As the maps in C , D reveal , all move to the endolysosomal cluster . Figure 4—figure supplement 1 provides a schematic of the experimental design . Please refer to Figure 4—figure supplements 2 and 3 for further in-depth analysis of protein localization changes following EGF stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 01010 . 7554/eLife . 16950 . 011Figure 4—figure supplement 1 . Dynamic organellar maps ( EGF-treatment ) – overview of the experimental workflow . Starting with SILAC light and heavy cells in both conditions , lyse each batch of cells separately . Subject the lysates to differential centrifugation , generating membrane sub-fractions with light lysate and global fractions with heavy lysate . To identify moving proteins with precise location ( follow grey lines ) , mix light fractions 1:1 with global membrane fraction of identically treated cells , to obtain ratios , which are visualised in PCA space . Weight SILAC L/H ratios by protein amount in the light fraction . Subtract equivalent weighted ratios of the untreated samples from the treated samples to obtain a difference profile of five differences for each protein . Repeat this three times , apply statistical test to identify moving proteins ( MR plot ) . Use SVM-based machine learning to identify the new location of proteins that have moved . To identify proteins moving within the global membrane , nuclear and cytosolic fractions ( red lines ) , measure the heavy fractions and quantify using MaxLFQ ( Cox et al . , 2014 ) . Perform a T-test on triplicate data to reveal protein abundance changes in the global fractions . For copy number changes ( green lines ) , multiply the intensity data by the protein yields , and use the sum of these values in the proteomic ruler ( Wisniewski et al . , 2014 ) to obtain total copy numbers . Multiply the copy numbers by the change in the proportion of a protein in a global fraction to obtain copy numbers entering or leaving this fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 01110 . 7554/eLife . 16950 . 012Figure 4—figure supplement 2 . Protein localization changes following EGF stimulation . ( A ) Organellar maps were prepared from untreatedHeLa cells ( control , left side ) , and from cells following continuous stimulation with EGF for 20 min ( +EGF , right side ) . The individual maps from triplicate biological repeats are shown , visualized by PCA . Organellar clusters are colour coded as in Figure 2 . Major translocating proteins are shown as unique symbols . CBL and UBASH3B were identified in only one of the +EGF maps; they are mostly cytosolic before EGF treatment , and hence not identified in control maps . ( B ) Detection of EGF-induced global profile changes . Nuclear , membrane and cytosolic fractions from the experiments described in A ) were subjected to mass-spectrometric analysis using label-free quantification ( LFQ ) . Mean Log2 LFQ values from triplicate control experiments were subtracted from triplicate EGF stimulation experiments and plotted against Student’s ( two-sided ) t-test p-value for that difference ( a ‘volcano’ plot ) . Proteins that increase in abundance in the relevant compartment following EGF stimulation are found on the right-hand side of the plots . Proteins undergoing significant translocations are shown in red , based on cut-offs determined as follows . First , the protein must show a minimum two-fold change in abundance ( absolute log-difference >1 ) . Second , the protein must constitute at least 10% of the total pool , either before or after EGF stimulation , in the compartment where it is shown to be changing ( as determined from the protein’s global intensity profile; see Figure 1A ) . Finally , the p-value cut-off was FDR-controlled using the six control maps generated in Figure 2—figure supplement 2A as a mock experiment , in which no true positives would be expected . Three maps were assigned as mock-treated , three as control . For each compartment , a p-value cut-off was chosen such that no false positives would be detected in the mock experiment , but changes could still be detected in the genuine experiment ( FDR = 0 ) . This was possible for cytosolic and membrane fractions ( -log10 p=2 . 0 and 3 . 1 , respectively ) . In the case of the nucleus ( -log10 p=2 . 6 ) , two false positives are expected among the 13 positives ( FDR ≈ 15% ) . Two relevant changes ( shown in grey ) narrowly missed significance with our extremely stringent cut-offs ( SHC1 in the organellar fraction , and MAPK1 in the nuclear fraction ) . While their p-values were sufficiently high to reach significance , their fold-changes were just below two . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 01210 . 7554/eLife . 16950 . 013Figure 4—figure supplement 3 . Global protein distribution profile changes induced by EGF treatment . For proteins undergoing significant localization changes ( Figure 4—figure supplement 1 , Supplementary file 7 ) , the distribution between nuclear , organellar and cytosolic fractions is shown before and after EGF treatment ( bars show mean ± SD , n=3 ) . Many proteins show transitions between nuclear and cytosolic fractions ( eg CIC , NAA40 ) . Several are recruited to the organellar fraction , from the cytosolic pool ( eg CBL , GRB2 , and SHC1 ) . VASN shows overall degradation . Please refer to the Methods for full details on the interpretation of global distribution profiles and their changes . Furthermore , note that in each case , the three control fractions are normalized to a sum of 1 . If EGF treatment changes the overall abundance of a protein , the sum of the three +EGF fractions will be different from 1 ( eg <1 , if the protein is degraded ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 013 In addition to intra-organellar translocation events , EGF signalling also involves cytosol/membrane as well as cytosol/nuclear transitions . To capture these events , we compared the abundance of proteins in membrane , nuclear and cytosolic fractions ( as prepared in Figure 1C ) from control and EGF-stimulated cells , based on label-free quantification ( Cox et al . , 2014 ) ( Figure 4—figure supplement 2B ) . Stringent FDR controls were derived using a mock experiment of our six database maps , as above , identifying 26 significant changes in the EGF experiment ( Supplementary file 7 , and Figure 4—figure supplement 3 ) . In agreement with the organellar maps , we detected substantial recruitment of GRB2 and SHC1 to membranes . In addition , we observed recruitment of CBL and UBASH3B , which are also known to bind to activated EGFR ( Grovdal et al . , 2004; Raguz et al . , 2007 ) . Consistent with that , CBL and UBASH3B were not detected in control maps but were found in individual EGF-treated maps , in the endosome/lysosome ( Figure 4—figure supplement 2A ) . Among others , we identified the known translocation of RPS6KA3 into the nucleus , as well as a surprising number of transcriptional regulators leaving the nucleus ( Supplementary file 7 ) . These included ZCCHC8 and RBM7 , the unique components of the nuclear exosome targeting ( NEXT ) complex ( Lubas et al . , 2011 ) , which targets the exosome to promoter upstream transcripts ( PROMPTS ) for their degradation . Therefore our data suggest that EGF may induce modulation of the non-coding transcriptome . Finally , we combined the identified translocation events with our estimates of absolute protein abundances . For each translocating protein , we calculated the number of molecules in cytosol , nuclear , and organellar fractions , before and after EGF treatment . Differences were then interpreted as the number of proteins moving between compartments ( summarized in Supplementary file 7 ) . For example , our data show a significant overall loss of EGFR upon EGF treatment ( from 700 , 000 to 620 , 000 copies per cell , p=0 . 0022 ) , suggesting that a proportion of endocytosed EGFR has already been degraded in lysosomes . Approximately 500 , 000 copies of GRB2 are recruited onto endosomes/EGFR , suggesting a stoichiometry of ~1:1 with EGFR . In contrast , CBL ( 10 , 000 copies ) and UBASH3B ( 30 , 000 copies ) are recruited sub-stoichiometrically , as would be expected of enzymatically acting proteins . SHC1 ( 100 , 000 copies ) is also recruited sub-stoichiometrically . The cell loses three quarters of its Vasorin ( a negative regulator of TGFB signalling; 30 , 000 copies ) , most likely through plasma membrane shedding ( Malapeira et al . , 2011 ) . Over 300 , 000 copies of the actin regulator Palladin are released into the cytosol , and the Rho-effector PKN2 is shifted to endosomes , indicating major cytoskeletal rearrangements . Our data thus begin to provide a quantitative , integrated view of EGF-triggered subcellular translocations at the protein level ( Figure 5 ) . 10 . 7554/eLife . 16950 . 014Figure 5 . Quantitative mapping of EGF-triggered subcellular translocation events . Summary of key protein translocations in HeLa cells following 20 min of continuous stimulation with EGF . All depicted changes were detected by organellar maps in this study; they include numerous previously known as well as novel translocation events . Numbers on arrows indicate how many copies of a protein undergo the indicated movement ( per cell ) . These estimates were also calculated from the mass spectrometry data , using the proteomic ruler approach ( Wisniewski et al . , 2014 ) . Figure 4—figure supplement 3 and Supplementary file 6 ( interactive database ) and 7 ( compact summary ) show additional translocations not included here . DOI: http://dx . doi . org/10 . 7554/eLife . 16950 . 014 Here , we provide localization information for 8710 proteins in HeLa cells . The database is accessible via an Excel file ( Supplementary file 4 ) and a website ( www . MapOfTheCell . org ) , which provide complementary features for analysing the data . Both contain information on protein abundance ( copy numbers per cell ) , global cellular distributions ( eg cytosolic vs membrane pools ) , and predicted organellar associations . In addition , the website offers visualization and interactive exploration of the maps . Supplementary file 4 provides an extra local ‘neighbourhood analysis’ identifying proteins with highly similar fractionation profiles ( useful for identifying potential protein complexes ) , and also allows easy annotation of whole protein families via its batch submission option . The complexity of the HeLa proteome has been estimated at around 10 , 000 proteins ( Beck et al . , 2011; Nagaraj et al . , 2011 ) ; a substantial proportion of this is covered by our database . Importantly , it accounts for the vast majority of protein cell mass ( as can for example be seen from the cumulative mass plots in Figure 3B , which all reach a stable plateau ) ; further identifications would mostly correspond to low abundance proteins , with minimal contributions to organellar composition . In this respect , our database approaches comprehensive coverage , and offers a quantitative view of cellular architecture ( Figure 3 ) . The relative sizes of organelles differ significantly between cell types; the approach presented here allows a comparatively rapid characterization at a level previously only achievable through extensive morphological studies . A future comparison of different cell types will substantially enhance our understanding of cellular identity , by uncovering universal features and specific adaptations . In addition to the organellar level , it will also give new insights for individual proteins , by revealing cell- or species-specific localization differences , and thus potentially new regulatory or functional aspects . The profiling approach presented here maximizes speed and simplicity of the subcellular fractionation procedure . This ensures reproducibility , and at the same time keeps organelles as intact as possible . Since the preparative aspects are straightforward , several fractionations can be carried out in parallel on the same day , allowing multiplexing and complex experimental designs . Relative to the previous LOPIT approach ( localization of organelle proteins by isotope tagging; Christoforou et al . , 2016 ) , our fractionation protocol is five times faster ( 4 hr vs ~20 hr ) , and requires an order of magnitude less starting material ( 107 cells vs 108 cells ) . Most importantly , our method can be used comparatively , and also offers quantitative data on protein abundance; a comparative application of LOPIT has yet to be demonstrated . The peptide labelling strategy allows very flexible use of LOPIT; our method requires metabolic labelling ( SILAC ) , currently rendering it most suitable for dividing cells in culture . However , an application of fractionation profiling to mammalian tissues is possible , since mice can be kept on a SILAC diet ( Zanivan et al . , 2012 ) ; alternatively , a representative mix of SILAC-labelled cell lines may be used to generate the reference fraction ( SuperSILAC approach [Geiger et al . , 2010] ) . In addition , mass tagging is in principle compatible with our approach , too , and may thus extend its range of applications in future . A detailed comparison of the methods’ relative advantages and requirements is presented in Supplementary file 8 . Our organellar assignments are in excellent agreement with independent external data ( Figure 2—figure supplement 1 , Supplementary file 3 ) . Furthermore , we made a direct comparison with a recent analysis of the mouse stem cell spatial proteome using LOPIT ( Christoforou et al . , 2016 ) . 2397 homologous proteins were classified in both studies , of which 2196 had identical compartment predictions ( 91 . 6%; Supplementary file 3 ) . This exceptionally high level of agreement , across species and cell types , reciprocally supports the very high accuracy of predictions in both datasets . Organellar maps based on subcellular fractionation profiles reflect protein steady state localizations . Proteins predominantly associated with a single organelle have closely matching profiles , and can be assigned unambiguously . In contrast , proteins equally split over two ( or more ) compartments have mixed profiles , which may be difficult to interpret ( Gatto et al . , 2014 ) . Here , we assign each protein to the most likely compartment , but potential secondary assignments are also indicated ( Supplementary file 4 ) . Furthermore , our two-tiered profiling approach considerably alleviates the dual-localization problem , by separating organellar predictions from quantifying a protein’s nuclear , cytosolic and membrane pools . This allows , for example , the accurate characterization of nuclear-cytosolic shuttling proteins: instead of showing an ambiguous ‘in between’ state , our approach precisely determines how these proteins are distributed over the two compartments . Similarly , for proteins with a cytosolic and an organellar pool , it allows quantification of the distribution , in addition to identification of the membrane compartment . Of note , our dynamic implementation of maps is generally unaffected by multiple localization difficulties , since we uncouple the detection of protein translocations from organellar assignments ( Figure 4 ) . Thus , our approach allows the identification of translocation events , even if they only involve partial organellar transitions . Here , we have used organellar maps to analyse cellular events following EGF stimulation . We correctly captured the endosomal transition of EGF receptor , and recruitment of signalling adaptors . Remarkably , the translocations were detected with extremely stringent FDR control , using cut-offs where we expect no false positives . This supports that our approach is capable of identifying translocation events de novo , without having to filter results based on prior knowledge . Furthermore , in combining the translocation data with protein copy number estimations , we provide a genuine systems-biology approach to EGF signalling at the protein level ( Figure 5 ) . Unlike transcriptomic or proteomic profiling , our approach allows detection of cellular rearrangements at very early time points after stimulation , long before changes in protein abundance occur . The entire experiment ( triplicate comparisons , six maps ) required only five days of mass spectrometry measuring time . In total , our analysis identified 40 translocation events , including numerous previously unreported movements . Among them are ten major regulators of actin dynamics , such as the kinases ROCK2 , PKN2 , PIK3C2B , and their downstream targets ADD1 and CTNN1 , as well as PALLD , LASP1 , and UTRN , suggesting that re-arrangement of the cytoskeleton is one of the major immediate effects of EGF signalling in HeLa cells . For several of these proteins , this study provides the first experimental evidence that they are targets of the EGF pathway ( Supplementary file 7 ) . Our data also reveal an unexpected cross-talk with other signalling pathways; Vasorin-shedding , AHNK and PDCD4 rerouting are all likely to counteract anti-proliferative TGFB signalling , and may serve to enhance EGF activity . Strikingly , we observed several transcriptional regulators leaving the nucleus . While nuclear import of proteins , such as ERK2/MAPK1 , is a common downstream effect of signalling , nuclear protein export has been reported comparatively rarely . A possible explanation is that this type of movement is more difficult to detect with conventional approaches , such as microscopy: protein import concentrates the signal in the nucleus , whereas export diffuses it . Taken together , these observations highlight the power of the holistic proteomic approach , which identifies the co-ordinated behaviour of functionally linked groups of proteins , and thus uncovers cellular response modules . This study demonstrates that dynamic organellar maps can shed new light even on relatively well-studied processes , such as EGF uptake . We propose that they will be similarly suitable in the fields of autophagy , membrane trafficking and cellular differentiation , providing a powerful complement to imaging-based techniques . Since they offer an unbiased approach to studying cellular dynamics that does not require prior knowledge , they will also be an effective tool for exploratory investigations of poorly characterized processes . The possibility to combine maps with high-throughput phosphoproteomics data ( Humphrey et al . , 2015 ) promises to provide unprecedented views of signalling , by linking the movement of substrates to their phosphorylation status . Moreover , as we have shown here , organellar maps will pave the way for quantitative process modelling in cell biology . The following is an overview protocol for the rapid generation of organellar maps , focusing on essential steps , and including an approximate time frame . Detailed descriptions may be found in the corresponding sections below . StepDescriptionTime requirementsStarting material: SILAC heavy and light labelled HeLa cells ( 1 x15 cm dish each , 50% confluent , ie 2 x 10 million cells ) . 1Mechanical cell lysis , and differential centrifugation subcellular fractionation → the actual ‘Fractionation Profiling’4 hr2Protein assay of fractions , overnight tryptic digest , peptide clean-up4 hr hands-on , + overnight digestion3Mass spectrometry analysis ( Thermo Q-Exactive HF ) 20 hr ( fast protocol ) 4MaxQuant data processing ( free software ) , data filtering< 24 hr ( processor with 8 cores e . g . intel i7 ) 5Visualization of maps by PCA , check clustering ( using eg SIMCA software ( free demo ) , or Perseus software ( free ) 1 hr6Prediction of protein subcellular localization by SVM classification ( Perseus , free software ) 1 hr→From cells to map in 3 days Visit www . MapOfTheCell . org for interactive exploration of the data provided in this study . Please refer to Supplementary file 10 , which shows a screenshot of the webpage , with annotations on how to use it . We would like to thank Matthias Mann for his generous support of this project . This work was funded by the German Research Foundation ( DFG/Gottfried Wilhelm Leibniz Prize ) ; the Louis-Jeantet Foundation; and the Max Planck Society for the Advancement of Science . We are very grateful to Korbinian Mayr , Igor Paron , and Gabriele Sowa for outstanding technical support . We would also like to thank Sebastian Schuck for a critical discussion of the manuscript , and all the members of the Mann Department for valuable feedback .	The interior of every cell is highly organised , and contains many compartments , called organelles , that are dedicated to specific roles . Proteins are the tools and machines of the cell , and each organelle has its own set of proteins that it requires to work correctly . Each cell contains ten or more organelles , and several thousand different types of proteins . The exact location of proteins in the cell is important; once we know what compartment a protein is in , it is easier to narrow down what it might be doing . The location of many proteins in a cell is unclear or simply not known . Moreover , since changing the location of a protein can change its activity , it is also important to be able to detect changes in the location of proteins under different circumstances , such as before and after drug treatment . Itzhak et al . set out to develop a method that reveals the locations of all the proteins in a cell at any given time . The resulting technique maps the location of most of the proteins in a human cancer cell line and , in addition , determines how many copies of each protein there are . Combining these two types of information produces a model of the cell’s architecture . Importantly , Itzhak et al . were able to compare such a model of the cell under normal circumstances to a model made after the cell had been stimulated with a growth factor . This revealed which proteins had changed location , identifying these proteins as important for the cell’s response to the growth factor . The new mapping method could be used in the future to analyse the anatomy of different cell types , such as nerve cells and cells of the immune system . Itzhak et al . also want to investigate the differences between healthy cells and cells from people with neurological disorders to understand how such diseases arise .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods", "Acknowledgments" ]	[ "computational", "and", "systems", "biology", "cell", "biology", "tools", "and", "resources" ]	2016	Global, quantitative and dynamic mapping of protein subcellular localization
During their final maturation in the cytoplasm , pre-60S ribosomal particles are converted to translation-competent large ribosomal subunits . Here , we present the mechanism of peptidyltransferase centre ( PTC ) completion that explains how integration of the last ribosomal proteins is coupled to release of the nuclear export adaptor Nmd3 . Single-particle cryo-EM reveals that eL40 recruitment stabilises helix 89 to form the uL16 binding site . The loading of uL16 unhooks helix 38 from Nmd3 to adopt its mature conformation . In turn , partial retraction of the L1 stalk is coupled to a conformational switch in Nmd3 that allows the uL16 P-site loop to fully accommodate into the PTC where it competes with Nmd3 for an overlapping binding site ( base A2971 ) . Our data reveal how the central functional site of the ribosome is sculpted and suggest how the formation of translation-competent 60S subunits is disrupted in leukaemia-associated ribosomopathies . The assembly of eukaryotic ribosomes involves the concerted action of over 200 trans-acting assembly factors . Following their assembly in the nucleus , pre-60S ribosomal subunits are exported to the cytoplasm where they are converted to translation-competent particles . The pre-60S particle attains export competence following release of the GTPase Nog2 by binding of the adaptor protein Nmd3 that recruits the nuclear export receptor Crm1 ( Xpo1 ) through a C-terminal leucine-rich nuclear export signal sequence ( Ho et al . , 2000; Thomas and Kutay , 2003; Trotta et al . , 2003 ) . Crm1-independent adaptors , including Arx1 ( Hung and Johnson , 2006; Bradatsch et al . , 2007 ) , Bud20 ( Bassler et al . , 2012 ) , Mex67 ( Yao et al . , 2007 ) and Ecm1 ( Yao et al . , 2010 ) facilitate pre-60S nuclear export . Once the ribosomal precursor enters the cytoplasm , the final assembly factors are removed and the last remaining ribosomal proteins integrated . The AAA-ATPase Drg1 initiates the final cascade of cytoplasmic events by recycling the assembly factors Rlp24 and Nog1 ( Kappel et al . , 2012; Pertschy et al . , 2007; Lo et al . , 2010 ) . Downstream cytoplasmic maturation events include the release and recycling of additional shuttling proteins including the export factors Arx1 , Mex67 and Nmd3 , removal of the GTPase Lsg1 as well as incorporation of late joining ribosomal proteins . The Arx1-Alb1 heterodimer is bound at the end of the peptide exit tunnel ( Bradatsch et al . , 2012; Leidig et al . , 2014; Wu et al . , 2016 ) , from where it is released by the concerted action of the zinc-finger protein Rei1 and the cytosolic J protein Jjj1 ( human DNAJC21 ) that stimulates the ATPase activity of the Hsp70 chaperone protein Ssa ( Hung and Johnson , 2006; Lebreton et al . , 2006; Demoinet et al . , 2007; Meyer et al . , 2007; Meyer et al . , 2010; Lo et al . , 2010 ) . Incorporation of Rei1 requires prior release of Nog1 , whose C-terminal tail seals the exit tunnel as the particle transitions from the nucleolus to the cytoplasm ( Wu et al . , 2016 ) . Interestingly , Rei1 also inserts into the exit tunnel ( Greber et al . , 2012; Greber et al . , 2016 ) from where it is displaced during later cytoplasmic maturation steps by Reh1 ( Ma et al . , 2017 ) . However , it is unclear when Reh1 is exchanged for Rei1 and how long Reh1 persists on the particle . A parallel branch of the cytoplasmic maturation pathway involves Yvh1-dependent exchange of uL10 for the placeholder protein Mrt4 to form the P-stalk ( Kemmler et al . , 2009; Lo et al . , 2009; Rodríguez-Mateos et al . , 2009 ) . The subsequent cytoplasmic maturation steps are crucial to correctly shape the peptidyltransferase centre ( PTC ) , evolutionarily the oldest part of the ribosome , in a strictly controlled sequence of events . Single-particle cryo-electron microscopy ( cryo-EM ) has identified the binding sites for Nmd3 and Lsg1 on the intersubunit face of the 60S subunit ( Ma et al . , 2017; Malyutin et al . , 2017 ) . Nmd3 spans the tRNA corridor from the uL1 protein at the L1 stalk through the tRNA exit site and the PTC to interact with the anti-association factor eIF6 ( yeast Tif6; herein called eIF6 ) at the sarcin-ricin loop ( SRL ) , while Lsg1 embraces H69 . The timing and mechanism of Nmd3 release remains unclear , but is dependent on the GTPase Lsg1 ( Hedges et al . , 2005 ) and the integration of ribosomal proteins eL40 ( Fernández-Pevida et al . , 2012 ) and uL16 ( Hedges et al . , 2005 ) . The eIF6 protein prevents premature joining of the ribosomal 60S and 40S subunits by binding to the SRL and ribosomal proteins uL14 and eL24 ( Gartmann et al . , 2010 ) . Release of eIF6 in the cytoplasm is catalysed by EFL1 ( elongation factor like GTPase 1 , an EF-2 homolog ) and its cofactor SBDS ( Shwachman-Bodian-Diamond syndrome , yeast Sdo1 ) ( Bécam et al . , 2001; Senger et al . , 2001; Menne et al . , 2007; Finch et al . , 2011; Wong et al . , 2011; Weis et al . , 2015 ) . The recruitment of SBDS to the 60S subunit depends on the prior integration of uL16 into the PTC in vivo ( Weis et al . , 2015 ) . Although recent structural studies suggest that eIF6 is released after Nmd3 ( Weis et al . , 2015; Ma et al . , 2017 ) , an alternate model posits that the binding of uL16 breaks the interaction of the Nmd3 N-terminus with eIF6 , allowing the recruitment of SBDS to promote eIF6 removal prior to the eviction of Nmd3 ( Lo et al . , 2010; Malyutin et al . , 2017; Patchett et al . , 2017 ) . Maintaining the fidelity of late cytoplasmic 60S subunit maturation is crucial for all eukaryotic cells as defects in this process cause human developmental defects and cancer predisposition . Multiple mutations associated with human bone marrow failure and leukaemia target this pathway . For example , recurrent somatic mutations in the RPL10 gene ( encoding the ribosomal protein uL16 ) have been identified in 10% of cases of paediatric T-cell acute lymphoblastic leukaemia ( T-ALL ) ( De Keersmaecker et al . , 2013 ) . The T-ALL associated uL16-R98S missense variant impairs the release of both Nmd3 and eIF6 ( De Keersmaecker et al . , 2013 ) . However , the underlying mechanism is unknown . Inherited mutations in the SBDS gene have been identified in 90% of individuals with Shwachman-Diamond syndrome ( SDS ) , an autosomal recessive disorder characterized by poor growth , exocrine pancreatic insufficiency , skeletal abnormalities and bone marrow failure with an increased risk of progression to myelodysplastic syndrome ( MDS ) and acute myeloid leukaemia ( AML ) ( Boocock et al . , 2003; Warren , 2018 ) . SDS is also associated with mutations in DNAJC21 , the human homologue of yeast JJJ1 ( Dhanraj et al . , 2017; Tummala et al . , 2016; D'Amours et al . , 2018 ) and EFL1 ( Stepensky et al . , 2017 ) . To elucidate the mechanism of PTC completion and understand how this process in corrupted by leukaemia-associated mutations , we used single-particle cryo-EM and cross-linking mass spectrometry to delineate the sequential steps that lead to assembly of the key functional site of the ribosome . We show how integration of the final ribosomal proteins eL40 and uL16 initiates a hierarchical sequence of RNA and protein rearrangements that result in release of the essential nuclear export adaptor Nmd3 , a key conserved step in PTC formation . Defective completion of the PTC causes developmental disorders associated with an increased propensity for malignant transformation . Hence , our atomic models not only illuminate the mechanism of PTC assembly but also suggest how mutations found in leukaemia disrupt this process . We set out to determine the mechanism of cytoplasmic 60S subunit maturation and completion of PTC assembly by affinity purifying native pre-60S particles from S . cerevisiae using tandem affinity purification ( TAP ) -tagged Lsg1 as bait ( Figure 1—figure supplement 1A ) and subjecting them to immunoblotting ( Figure 1—figure supplement 1B ) , cryo-EM analysis ( Figure 1A , Figure 1—figure supplement 1C–F , Figure 1—figure supplements 2–4 , Supplementary file 1A , B ) and crosslinking mass spectrometry ( XL-MX ) ( Figure 1—figure supplement 5A , B , Supplementary file 2 ) . Immunoblotting revealed enrichment of the assembly factors Lsg1 , Nmd3 , Arx1 and eIF6 in the purified particles , but under-representation of ribosomal proteins uL16 and uL10 ( P0 ) ( Figure 1—figure supplement 1B ) . Analysis by single-particle cryo-EM with extensive 3D classification yielded a series of structures ( hereafter termed states I-VI ) that likely reflect sequential snapshots of final cytoplasmic pre-60S maturation ( Figure 1A ) . The ability to capture state VI ( lacking both Nmd3 and Lsg1 ) likely reflects ongoing maturation of the particles or possibly the dissociation of Lsg1 and Nmd3 during immunopurification . We refined all six pre-60S cryo-EM reconstructions to average resolutions of 3 . 1–3 . 9 Å with the local resolution extending to 2 . 3 Å in the core of the particles in states I and III ( Figure 1—figure supplement 1E , F ) . The maps allowed us to fit and refine atomic models for all six assembly states , including the biogenesis factors Arx1 , Lsg1 , Nmd3 , Rei1 , Reh1 , eIF6 , ribosomal proteins eL40 , uL16 and uL11 together with the 5S , 5 . 8S and 25S rRNA ( Figure 1B , Figure 1—figure supplements 2–4 , Supplementary file 1A , B ) . In addition , we identified density corresponding to the phosphatase ( Baßler et al . , 2017 ) and zinc finger ( Zhou et al . , 2019 ) domains of Yvh1 situated between uL11 in the P-stalk and eIF6 ( Figure 1A , states I-II ) . Sequential maturation of the cytoplasmic pre-60S subunit couples dramatic conformational rearrangements of two long flexible helices ( the L1 stalk and H38 ) to recruitment of the final ribosomal proteins ( eL40 and uL16 ) and assembly factor displacement ( Figure 1A ) . In states I-III , the L1 stalk is displaced inwards with H38 in the ‘closed’ position . Rearrangement of H38 to the mature position ( state IV ) is followed by partial ( state V ) then full ( state VI ) retraction of the L1 stalk . Focused classification of the local density around the P-stalk ( Figure 1—figure supplement 2 ) revealed that binding of ribosomal protein eL40 does not occur until state III , prior to the recruitment of uL16 in state IV . Although there is clear density for uL11 ( state I ) at the base of the P-stalk , uL10 was poorly defined in our maps , consistent with the paucity of uL10 protein in the Lsg1-TAP particles as detected by immunoblotting ( Figure 1—figure supplement 1B ) . Focused classification around the P-stalk revealed that Yvh1 is predominantly bound in state I , is present at low occupancy in state II but is absent in state III . Thus , Yvh1 and eL40 do not appear to bind simultaneously to the same particle . The homologous C-termini of Reh1 and Rei1 occupy the PET sequentially ( Figure 1A ) . While the departure of Rei1 seems to be concurrent with Arx1 release , the C-terminus of Reh1 unexpectedly persists in the exit tunnel even after the departure of Nmd3 and Lsg1 . The timing of the exchange of Rei1 for Reh1 ( between states I-II ) is consistent with XL-MS analysis that yielded crosslinks between Rei1 , Arx1 and eL24 , while Reh1 yielded crosslinks to eL24 , but not to Rei1 or Arx1 ( Figure 1—figure supplement 5A , B and Supplementary file 2 ) . These data are also consistent with a previous structural snapshot ( Ma et al . , 2017 ) and with co-immunoprecipitation analysis ( Parnell and Bass , 2009 ) . We conclude that the exchange of Reh1 for Rei1 occurs at the time of Arx1 release , but that surveillance of the PET by Reh1 continues throughout the entire process of cytoplasmic pre-60S maturation . While the overall structures of Rei1 and Reh1 are consistent with previous reports ( Greber et al . , 2016; Ma et al . , 2017 ) , our maps reveal additional density that corresponds to the N-termini of both Rei1 and Reh1 extending up across the surface of eL24 to directly interact with eIF6 subdomains C and D and with a loop that extends out from the α-helical domain of Lsg1 ( residues G438-T456 ) ( Figure 1A , state I ) . The Lsg1 α-helical domain ( residues G474-D479 ) also interacts with uL14 . We unambiguously distinguished the helical C-termini of Reh1 and Rei1 within the PET based on specific side chain densities ( Ma et al . , 2017 ) . Interestingly , the position of the extreme C-terminal leucine and glutamine residues of Rei1 and Reh1 differs compared with previous reports ( Ma et al . , 2017; Greber et al . , 2016 ) : the side chains of Q393 ( Rei1 ) and Q432 ( Reh1 ) form an electrostatic interaction with the base of U2875 ( H89 ) , while the backbone interacts with the base of U2978 ( H93 ) . The coexistence of eIF6 and Reh1 on the same particle ( Figure 1B ) suggests that state VI is not simply a product of eIF6 rebinding but is a bona fide late pre-60S subunit maturation intermediate that lies downstream of Nmd3 and Lsg1 release . These data support the hypothesis that eIF6 is evicted after Nmd3 ( Weis et al . , 2015 ) . We set out to understand the mechanism of Nmd3 release , a key event in the completion of PTC assembly . Nmd3 extends across the entire tRNA binding cleft on the intersubunit face of the 60S subunit from uL1 at the L1 stalk through the E , P and A sites to contact Lsg1 and eIF6 at the SRL ( Figure 1A , B ) . The cryo-EM density allowed us to build a complete atomic model for Nmd3 ( residues T16-R404 ) including backbone atoms and side chains ( Figure 2A ) , revealing a multi-domain architecture that includes two treble clef zinc fingers ( that superimpose with an RMSD of 1 Å over 13 Cα atoms; Figure 2A , inset ) , two alpha-beta domains and two beta-barrel domains . The N-terminal treble clef zinc finger ( residues 16–42 , C19-C22-C35-C38 ) is grafted into an alpha-beta domain ( residues 43–154 ) that is structurally related to ribosomal protein eL31 . The second alpha-beta domain ( residues 155–250 ) is structurally related to ribosomal protein eL22 , while the two C-terminal beta-barrel domains have SH3 ( 251–310 ) and OB ( 311-400 ) folds , respectively . Although the combination of the SH3 and OB domains is similar in sequence and structure to eIF-5A-1 , the domains are oriented differently with respect to each other . Thus , Nmd3 comprises a modular assembly of existing structural blocks associated with the ribosomal machinery combined together to form a functionally distinct protein . We sought to understand how Nmd3 maintains the ‘closed’ orientation of the L1 stalk ( states I-IV ) and H38 ( states I-III ) ( Figure 1A ) . The Nmd3 OB domain holds the L1 stalk in the closed position , while the eL22-like , SH3 and OB domains , together with an extended C-terminal loop , encircle and distort the tip of H38 ( Figure 2B , C ) . Specifically , two loops ( β1–2 and β3–4 ) from the OB domain stabilize a base-flip of A1025 . The side chain and backbone atoms of N332 contact the base and backbone phosphate of A1025 . H364 contacts the backbone phosphate of A1027 , while the side chains of Y402 and R404 contact the backbone phosphates of A1026-G1029 . Within the SH3 domain , the side chain of K253 contacts the phosphate backbone of C1023 . Helix α1 from the eL22-like domain ( particularly the side chains of R169 and Q176 ) packs against the bases and backbone of H38 ( G1020-C1023 ) . Throughout the steps of pre-60S subunit maturation visualised herein , H69 adopts a conformation that differs from the mature ribosome ( Ben-Shem et al . , 2011 ) but that surprisingly persists even in the absence of Nmd3 and Lsg1 ( state VI ) ( Figure 3A , B ) . Lsg1 stabilises a base flip at G2261 ( H69 ) ( Figure 3A ) , while the side chain of Lsg1 W142 stacks against the base of A2256 at the tip of H69 ( Figure 3B ) . The altered conformation of H69 is also promoted by a β-hairpin in the SH3 domain of Nmd3 that stabilises a base flip in U2269 at the junction between H68 and H69 ( Figure 3C ) . The Nmd3 SH3 domain makes additional interactions with the 25S rRNA using two short α-helices ( SH3-α1 and SH3-α2 ) ( Figure 3D ) . As the altered conformation of H69 is maintained even in the absence of both Nmd3 and Lsg1 ( state VI ) , we suggest that H69 may only adopt the mature conformation after joining of the 60S and 40S subunits . Compared with the reconstituted Nmd3-Lsg1-60S particle ( Malyutin et al . , 2017 ) , the Nmd3 eL31-like domain is rotated ( 60° ) away from H89 towards the long α-helix of the Lsg1 GTPase domain in each of the native intermediates carrying Lsg1 and Nmd3 ( states I-V ) ( Figure 1A ) . The interaction involves a stacking interaction between the side chain of Nmd3 residue W104 and the side chain of Lsg1 residue R163 ( Figure 3E ) . Although focused classification of state III identified a subset of particles in which the Nmd3 eL31-like domain is rotated towards H89 , this class lacks Lsg1 and Arx1 , with Reh1 present in the PET . We suggest that the rotated conformation of the Nmd3 eL31-like domain in this subset of state III particles is a consequence of Lsg1 dissociation during sample preparation rather than a physiologically relevant ‘pre-Lsg1’ state . We next assessed the dynamic properties of the N-terminus of Nmd3 in solution using NMR spectroscopy . Heteronuclear 1H {15N} NOE analysis of the Archaeoglobus fulgidus Nmd3 ( residues 22–150 ) indicates that the N-terminal zinc finger ( residues 22–43 ) has a higher degree of motion than the attached eL31 domain on the picosecond timescale ( Figure 3—figure supplement 1A , B ) . Exchange broadened residues indicate the presence of a flexible linker between the two domains that is undergoing segmental motion . These data indicate that the Nmd3 N-terminal zinc finger has dynamic mobility that allows it to sample a range of positions in solution . However , the cryo-EM data indicate that the Nmd3 zinc finger remains bound to eIF6 subdomains D and E ( total buried surface area of 107 Å2 ) throughout states I-V ( Figure 3F ) , providing a click-lock that fixes the Nmd3 N-terminus in position throughout pre-60S maturation . Taken together , our data illustrate how the modular architecture of Nmd3 confers it with the flexibility to modulate its conformation depending on the specific stage of pre-60S maturation . In the late nuclear pre-60S particle purified using epitope-tagged Nog2 , the N-terminal domain of Nog1 separates H89 into two strands , H89 adopting an upright position ( Wu et al . , 2016 ) . We therefore sought to understand how H89 accommodates into its mature conformation to form one face of the u16-binding site . Focused classification around the H89 density revealed that in the absence of eL40 ( states I-II ) , H89 adopts a range of conformations even with Yvh1 present ( Figure 4A ) . However , following the departure of Yvh1 , eL40 stabilises H89 in its near-mature conformation by forming two major contacts with the opposing H91 , including a stacking interaction of A2847 ( H89 ) with G2898 ( H91 ) and base pairing between C2844 ( H89 ) and G2898 ( H91 ) ( Figure 4B ) . An observed contact between Yvh1 and the side chain of uL6 K141 overlaps with the binding site for the N-terminus of eL40 in state III , raising the possibility that the recruitment of eL40 may destabilise Yvh1 to promote its departure from the pre-60S particle . In turn , the docking of uL16 to the upper surface of H89 promotes a dramatic ~65° rotation of H38 away from Nmd3 ( Figure 4C ) , sandwiching uL16 in a cleft between H38 and H89 ( Figure 4D ) . In mature , actively translating ribosomes with tRNA bound , the eukaryotic-specific loop ( residues 102–112 ) of uL16 extends into the PTC ( Schmidt et al . , 2016 ) . However , in states I-IV the eL22-like domain of Nmd3 extends into the PTC where it is surrounded by helices H64 , H69-71 , H80 , H90 and H92-93 ( Figure 5A ) . Residues on the surface of the eL22-like domain β-sheet make extensive interactions with the backbone of H80 ( Figure 5B ) . Residue N205 ( α2 helix ) stabilises a stacking base pair interaction between C2308 ( H64 ) and C2284 ( H70 ) ( Figure 5C ) . Furthermore , the β4–5 loop of Nmd3 stabilises the ‘closed’ conformation of base A2971 through specific interactions with the side chains of K204 ( helix α2 ) , K224 , F242 ( strand β5 ) , Y240 and the backbone amide of S238 ( β4–5 loop ) ( Figure 5D ) . As a result , the uL16 P-site loop is unable to extend into the PTC in state IV because of a steric clash with the Nmd3 β4–5 hairpin loop . The initial docking of uL16 is fully compatible with retention of Nmd3 on the pre-60S particle ( Figure 1A , Figure 6A , B ) . However , uL16 liberates H38 from the C-terminus of Nmd3 promoting partial ( 20° ) ( state V ) and subsequently full ( 56° ) retraction ( state VI ) of the L1 stalk ( Figure 6B , C ) . Partial retraction of the L1 stalk ( state V ) is accompanied by a conformational switch in Nmd3 in which the OB , SH3 and eL22-like domains are displaced upwards and outwards from the P and E sites by ~20 Å through a pivot point between the flexible helical linker connecting the eL22- and eL31-like domains , while the N-terminus remains anchored to eIF6 and Lsg1 ( Figure 6B , D ) . Displacement of the eL22-like domain breaks the interaction of residue N205 ( helix α2 ) with C2308 ( H64 ) and C2284 ( H70 ) ( Figure 5C ) , allowing helix α2 to form a new contact with H69 ( Figure 6—figure supplement 1 ) . Displacement of the β4–5 hairpin loop from the entrance to the PTC ( state V ) permits the uL16 P-site loop to extend into the PTC where the highly conserved , essential residue R110 coordinates A2971 in an ‘open’ flipped-up conformation ( Figure 6B , C ) . Genetic and biochemical analysis are consistent with an essential role for the uL16 P-site loop in Nmd3 release ( Hofer et al . , 2007; Bussiere et al . , 2012; Sulima et al . , 2014; Weis et al . , 2015; Patchett et al . , 2017 ) . Furthermore , an nmd3-N205D allele rescued the fitness defect of yeast cells expressing the T-ALL-associated uL16-R98S mutant as the sole copy of uL16 ( Figure 6—figure supplement 4 ) . These data support a functional role for the interaction between Nmd3 residue N205 and helices H64 and H70 ( Figure 5C ) in stabilising the binding of Nmd3 to the 60S subunit . We conclude that the uL16 P-site loop drives the conformational equilibrium in favour of Nmd3 release by competing with Nmd3 for an overlapping binding site within the PTC . The cryo-EM structures reported here allow us to propose a mechanism for the completion of assembly of a functional PTC ( Figure 7 , Video 1 ) . The sequential incorporation of eL40 and uL16 couples large-scale rRNA rearrangements to a conformational switch in Nmd3 that allows the P-site loop of uL16 to fully accommodate into the PTC to push the conformational equilibrium towards Nmd3 dissociation ( Figure 6A–D , Figure 6—figure supplement 1 ) . The incorporation of eL40 ( state III ) stabilises H89 to form one face of the uL16-binding platform and may also facilitate Yvh1 release by disrupting its interaction with uL6 . While the initial docking of uL16 to the upper surface of H89 is still compatible with the retention of Nmd3 on the particle , it promotes a ~ 65° rotation of H38 away from Nmd3 to adopt its mature conformation , thereby sandwiching uL16 in the cleft formed by H38 and H89 . Loss of the stabilising interactions between Nmd3 and the tip of H38 promotes partial retraction of the L1 stalk ( state V ) . This is coupled to a conformational switch in the C-terminus of Nmd3 that displaces the SH3 and OB domains from the E site and the eL22-like domain from the P-site . The uL16 P-site loop now fully accommodates into the PTC by competing with Nmd3 for an overlapping binding site at base A2971 , thus driving the conformational equilibrium in favour of Nmd3 dissociation . The side chain of uL16 residue R110 stabilises the flipped-out conformation of base A2971 to complete the assembly of a functional PTC . Our structures explain the essential roles of eL40 ( Fernández-Pevida et al . , 2012 ) and the u16 P-site loop ( Patchett et al . , 2017; De Keersmaecker et al . , 2013; Hofer et al . , 2007 ) in Nmd3 release . The extension of the uL16 P-site loop into the PTC that is observed in the mature ribosome in the presence of tRNA ( Schmidt et al . , 2016 ) is accomplished by a conformational switch in Nmd3 that allows uL16 to bypass the steric clash with the β4–5 loop of the Nmd3 eL22-like domain ( Figure 6—figure supplement 2A ) . Importantly , the link between the Nmd3 N-terminus and eIF6 remains unbroken at this stage of the maturation process , contrary to a previous model ( Malyutin et al . , 2017 ) . Indeed , the links between Nmd3 , Lsg1 and eIF6 are maintained throughout states I-V . Consistent with the studies on nucleolar pre-60S subunit maturation ( Kater et al . , 2017; Sanghai et al . , 2018 ) , our work does not support parallel 60S assembly pathways as suggested for the bacterial 50S subunit ( Davis et al . , 2016 ) . In contrast , our data strongly support the concept that eukaryotic cytoplasmic 60S maturation proceeds in a linear , stepwise manner . The precise role of Lsg1 in Nmd3 release and the mechanism of Lsg1 GTPase activation remain unclear . It has been proposed that coupling of the flipped-out base G2261 in H69 to the Lsg1 Switch one region ( that coordinates a magnesium ion in the active site ) may regulate Lsg1 GTP hydrolysis ( Malyutin et al . , 2017 ) . However , the G2261 base-flip appears to persist throughout all the stages of cytoplasmic 60S maturation visualised , even in the absence of Lsg1 ( Figure 3A ) . By contrast , the Nmd3 eL22-like domain establishes a new contact with H69 as the L1 stalk retracts ( Figure 6D , Figure 6—figure supplement 1 ) . We speculate that this interaction of Nmd3 with H69 may transiently reposition base G2261 , thereby relaying the change in conformation to the Switch one region to activate Lsg1 GTP hydrolysis . The identification of a crosslink between the N-terminus of Lsg1 and the P-site loop of uL16 ( Figure 1—figure supplement 5A , B and Supplementary file 2 ) raises the intriguing possibility that Lsg1 may proofread accommodation of the uL16 P-site loop into the PTC as the Nmd3 β4–5 loop is retracted . In this model , Lsg1 may act as a gatekeeper that licenses further progression of 60S subunit maturation by dissociating only once it senses that uL16 is correctly integrated . The disengagement of Lsg1 from the N-terminus of Nmd3 will further destabilise the interaction of Nmd3 with the 60S subunit , allowing retraction of the L1 stalk to pull Nmd3 completely off the intersubunit face . Unexpectedly , our data reveal that the C-terminus of Reh1 is retained in the PET in a late assembly intermediate carrying eIF6 that is downstream of Lsg1 and Nmd3 release ( state VI ) ( Figure 1A ) . Identification of this particle provides compelling evidence that eIF6 release by SBDS and EFL1 occurs after the dissociation of Nmd3 and Lsg1 as proposed ( Weis et al . , 2015 ) ( Figure 7 ) . This timing of events is consistent with the requirement for uL16 integration to allow SBDS binding both in vivo ( Weis et al . , 2015 ) and in vitro ( Sulima et al . , 2014 ) and with recent cryo-EM data ( Weis et al . , 2015; Ma et al . , 2017; Malyutin et al . , 2017 ) . Importantly , we also note that the binding sites for SBDS , Nmd3 and Lsg1 are incompatible , precluding the removal of eIF6 before the departure of Lsg1 and Nmd3 ( Figure 6—figure supplement 2B ) . We conclude that Nmd3 and Lsg1 are released prior to SBDS and EFL1 recruitment and that Lsg1 and Nmd3 function at least in part as placeholders for SBDS . This study reinforces the concept that SBDS and EFL1 interrogate the structural and functional integrity of the completed PTC by licensing removal of the anti-association factor eIF6 ( Weis et al . , 2015 ) ( Figure 7 ) . The side chain of SBDS residue R11 and the backbone of K62 interact with the side chain of R110 within the uL16 P-site loop that stabilises the flipped-out conformation of base A2971 ( Figure 6—figure supplement 3A ) . In addition , there are interesting parallels between the interactions of the uL16 P-site loop with the N-terminus of SBDS ( pdb 6qkl ) ( Weis et al . , 2015 ) and the contacts of uL16 with the P-site tRNA during decoding ( pdb 5gak ) ( Schmidt et al . , 2016 ) ( Figure 6—figure supplement 3B ) . The PET of the large ribosomal subunit appears to be directly monitored initially in the nucleus by Nog1 ( Wu et al . , 2016 ) and subsequently throughout the entire process of cytoplasmic assembly by Rei1 ( Greber et al . , 2012; Greber et al . , 2016 ) and Reh1 ( Ma et al . , 2017 ) in turn . Indeed , the direct interaction between the α-helical domain of Lsg1 and the N-termini of both Rei1 and Reh1 ( Figure 1A ) suggests how events at the PET and PTC might be coupled during cytoplasmic maturation . Docking analysis raises the possibility that the C-terminus of Reh1 may interact directly with the N-terminus of SBDS during late 60S maturation ( Figure 6—figure supplement 3C ) . However , the functional relevance of this observation remains to be tested . It is conceivable that Reh1 may not be removed from the exit tunnel until the pioneer round of translation , but the precise timing and mechanism of Reh1 release require elucidation . Our data define a coherent pathway that is targeted by multiple mutations in sporadic and inherited forms of leukaemia ( Figure 7 ) . The association of the SDS clinical phenotype with mutations in several components of the 60S maturation pathway , including SBDS ( Boocock et al . , 2003 ) , EFL1 ( Stepensky et al . , 2017 ) and DNAJC21 ( Tummala et al . , 2016; Dhanraj et al . , 2017 ) provides compelling support for the hypothesis that SDS is a ribosomopathy ( Menne et al . , 2007; Finch et al . , 2011 ) . Importantly , given the high degree of conservation in the amino acid sequence and structure of the uL16 protein ( Figure 7—figure supplement 1A , B ) , our cryo-EM data provide a mechanistic model that allows us to interpret the consequences of the recurrent uL16-R98S mutation found in paediatric T-ALL ( De Keersmaecker et al . , 2013 ) . We propose that the uL16-R98S mutation may increase the flexibility of the P-site loop , reducing its ability to effectively compete with the Nmd3 eL22-like domain for the overlapping binding site at base A2971 in the PTC , thereby driving the equilibrium towards Nmd3 release . Our study suggests that Yvh1 dissociates from the pre-60S particle prior to the binding of eL40 and uL16 ( Figure 1A ) . However , while this work was under review , Yvh1 was reported to bind cytoplasmic pre-60S particles carrying eL40 and uL16 ( Zhou et al . , 2019 ) . Perhaps explaining this discrepancy , our study has exclusively analysed native pre-60S particles , while Zhou et al examined particles purified from Rlp24-mutant cells or following treatment with the inhibitor diazaborine ( Zhou et al . , 2019 ) . Interpretation of the structural intermediates in the Zhou study should take also account of the heterogeneity in the deposited maps due to the classification strategy used . In conclusion , we have used single-particle cryo-EM to demonstrate the conformational changes during cytoplasmic 60S subunit maturation that couple incorporation of the ribosomal proteins eL40 and uL16 to the release of Nmd3 in what is likely to be a universal mechanism in eukaryotes . Our data not only reveal how the central functional site of the ribosome is assembled , but provide a framework to interpret the consequences of mutations linked to leukaemia-associated ribosomopathies . The cryo-EM density maps have been deposited in the Electron Microscopy Data Bank with accession numbers EMD-10068 , EMD-10071 , EMD-4560 , EMD-4636 , EMD-4884 and EMD-4630 . Atomic coordinates have been deposited in the Protein Data Bank , with entry codes 6RZZ , 6S05 , 6QIK , 6QTZ , 6RI5 and 6QT0 . Proteins present in crude extracts and purified pre-60S particles were separated on precast 4% to 12% NuPAGE gels ( Novex life technologies ) and blotted on a PVDF membrane ( Carl Roth GmbH ) using a tank-blot device ( Hoefer ) . For immunoblotting , primary antibodies directed to pre-ribosome maturation factors or ribosomal proteins and a secondary goat anti-rabbit antibody ( Sigma Aldrich ) were used . All antibodies were described previously ( Zisser et al . , 2018; Loibl et al . , 2014 ) , except the eIF6 antibody that was purchased from GeneTex . Chemiluminescence signals were detected using the ChemiDoc Touch Imaging System ( Bio-Rad ) and the Clarity Western Blotting Detection Reagent as substrate . For the isolation of late cytoplasmic pre-60S particles Lsg1-TAP ( Nissan et al . , 2002 ) was used as bait protein . Purification was performed in the absence of GTPase inhibitor . Affinity purification was performed using rabbit IgG covalently linked to magnetic beads as described ( Zisser et al . , 2018 ) . Briefly , the Lsg1-TAP strain was grown in two litres of YPD complete medium ( 2% ( w/v ) peptone , 1% ( w/v ) yeast extract , 2% ( w/v ) glucose , 0 . 002% ( w/v ) adenine ) to an OD600 of 1 . 2 and cells were harvested by centrifugation for 2 min at 4000 x g . Cells were washed once in lysis buffer 1 ( LB1; 20 mM HEPES , pH 7 . 5 , 10 mM KCl , 2 . 5 mM MgCl2 , 1 mM EGTA , 0 . 5 mM PMSF , 1 mM DTT and FY-protease inhibitor ( Serva ) . Cell lysis was performed in LB1 after addition of 1 . 5 volumes of glass beads by vigorous shaking for 4 × 30 s in a bead mill ( Merkenschlager ) with constant CO2 cooling . After centrifugation for 30 min at 40 , 000 x g , the cleared lysates were loaded on 200 µl magnetic beads covalently coupled with rabbit IgG ( Oeffinger et al . , 2007 ) and incubated at 4°C for 90 min under constant mixing using an overhead rotator . After washing twice with 8 ml LB1 containing 1 mM DTT and once with 8 ml LB1 containing 100 mM NaCl and 1 mM DTT , beads were transferred to 0 . 5 ml reaction vials and pre-ribosomal particles eluted by overnight TEV cleavage in 150 µl LB1 containing 100 mM NaCl , 0 . 5 mM DTT and 2 µg of purified , RNAsin treated TEV protease . After removal of the resin by centrifugation for 5 min at 5000 x g , the pre-60S particles present in the supernatant were spotted onto grids and processed for cryo-EM freezing . SDS-PAGE and immunoblotting was used to monitor the quality of the isolated particles . Pre-60 ribosomal TAP-tagged Lsg1 particles purified from S . cerevisiae were incubated 10 min at 4°C in the presence of 0 . 5% ( v/v ) glutaraldehyde ( Sigma-Aldrich ) to reduce preferential orientation and DTT added to a final concentration of 6 mM . EM grids were prepared by adding 3 μL pre-60 ribosomal purified TAP-tagged Lsg1 particles ( 40 nM ) to freshly glow-discharged Quantifoil R2/2 grids ( PELCO easyGlow ) . Grids were blotted and flash frozen in liquid ethane at 100 K using a Vitrobot Mark IV ( FEI Company ) . Grids were screened on a Tecnai T12 microscope ( FEI Company ) and data acquisition performed under low-dose conditions on a Titan Krios microscope ( FEI Company ) operated at 300 kV over two sessions of ~70 hr each . The two datasets were recorded on a Falcon III detector ( FEI Company ) at a nominal magnification of 75 , 000x ( effective pixel size of 1 . 065 Å on the object scale ) with a defocus range of −0 . 8 to −3 . 2 μm and a dose of ~63 e−/Å2 . The acquisition of 7957 and 15 , 923 movies for each session was performed semi-automatically using EPU software ( FEI Company ) . Movies were corrected for the effects of beam-induced motion using MotionCor2 ( Zheng et al . , 2017 ) . Contrast transfer function ( CTF ) parameters were estimated using GCTF ( Zhang , 2016 ) . All subsequent data processing was performed in RELION ( Scheres , 2012b; Scheres , 2012a; Kimanius et al . , 2016 ) . Electron micrographs showing signs of drift or astigmatism were discarded , resulting in a dataset of 7279 and 12 , 261 images . A total of 696 , 991 ( dataset 1 ) and 724 , 890 ( dataset 2 ) particles were selected automatically in RELION . Extracted particles were subjected to two rounds of 2D and 3D classification to discard defective particles , resulting in 370 , 687 ( dataset 1 ) and 512 , 903 particles ( dataset 2 ) . 3D auto-refinement resulted in an initial cryo-EM reconstruction with an overall resolution of 3 . 1 Å and 3 . 4 Å for the two datasets . After movie refinement and particle polishing the ‘shiny’ particles were subjected to further 3D auto-refinement and post-processing to yield maps with an overall resolution of 2 . 9 Å ( dataset 1 ) and 3 . 2 Å ( dataset 2 ) based on the gold-standard Fourier Shell Correlation ( FSC ) criterion calculated within RELION ( van Heel and Stöffler-Meilicke , 1985; Scheres and Chen , 2012 ) . However , the final maps were clearly heterogeneous in composition . We therefore sorted the images into subsets by a succession of 3D classifications using signal subtraction in RELION ( Penczek et al . , 2006 ) ( Figure 1—figure supplement 2 ) . A mask with a voxel value of one inside and zero outside extended by four pixels with a soft edge of ten pixels was applied to the intersubunit interface , containing Lsg1 , Nmd3 , eIF6 , uL16 , H38 and H89 , was used for the first round of focused classification , providing 10 classes for each dataset ( Figure 1—figure supplement 2 , mask 1 ) . In dataset 2 , a mask was applied around the L1 stalk , uL1 and the OB , SH3 and eL31-like domains of Nmd3 to separate the ‘open’ and ‘closed’ L1 stalk conformations ( Figure 1—figure supplement 2 , mask 2 ) . Finally , identical classes from both datasets were merged to improve the overall resolution . A mask was applied around the P stalk in states I and II to reveal different conformations of H89 and the presence of Yvh1 ( Figure 1—figure supplement 2 , mask 3 ) . Similarly , further classification ( using masks 2 and 3 ) was performed in states III and IV to reveal eL40 ( State III , subclass 1 ) and the retracted L1-stalk ( State V , subclass 1 ) , respectively . Image processing converged to yield six distinct classes ( states I-VI ) with global resolutions ranging from 3 . 1 to 3 . 9 Å ( Figure 1—figure supplement 1E ) . Local resolution was estimated to range from 2 . 3 to 6 . 3 Å using ResMap ( Kucukelbir et al . , 2014 ) ( Figure 1—figure supplement 1F , Figure 1—figure supplement 4 ) . As an initial starting model , the 3 . 0 Å crystal structure of the mature 60S subunit ( Ben-Shem et al . , 2011 ) ( pdb 4v88 ) from S . cerevisiae was initially fitted as a rigid body into the cryo-EM map of state I using UCSF-Chimera ( Pettersen et al . , 2004 ) . Atomic coordinates for Nmd3 , Lsg1 , eIF6 and uL1 ( backbone atoms only ) were taken from pdb code 5t62 ( Malyutin et al . , 2017 ) ; Rei1 and Arx1 from pdb code 5apn ( Greber et al . , 2016 ) ; Reh1 from pdb code 5h4p ( Ma et al . , 2017 ) ; eL40 , uL16 and the mature conformation of H38 from pdb code 4v88 ( Ben-Shem et al . , 2011 ) . Models were manually adjusted in Coot ( Emsley and Cowtan , 2004 ) and further refined using Phenix ( Adams et al . , 2010 ) and REFMAC v5 . 8 adapted for EM-refinement ( Amunts et al . , 2014 ) . Model evaluation was performed in MolProbity ( Chen et al . , 2010 ) ( Supplementary file 1A ) . Cross-validation against overfitting was performed as described ( Weis et al . , 2015 ) ( Figure 1—figure supplement 3 ) . Buried surface areas were calculated using the gmx sasa routine in GROMACS ( Van Der Spoel et al . , 2005 ) using the double cubic lattice method ( Eisenhaber et al . , 1995 ) with a probe radius of 0 . 14 nm . Molecular visualization was performed in UCSF-Chimera ( Pettersen et al . , 2004 ) , ChimeraX ( Goddard et al . , 2018 ) , Pymol ( The PyMOL Molecular Graphics System , Version 2 . 0 . 6 Schrödinger , LLC ) and VMD ( Humphrey et al . , 1996 ) . Flexible fitting of atomic models was initially performed using MDFF ( Trabuco et al . , 2008 ) . The system was set up in vacuo and subjected to energy minimization for 50 , 000 steps ( 50 ps ) to relax any steric clashes . A production run of 1 , 000 , 000 steps ( one ns ) was followed to fit the atoms into the EM density . The magnitude of the forces applied to the atoms ( scaling factor ξ ) was adjusted to 0 . 3 kcal/mol . To prevent overfitting , harmonic restraints were applied to maintain the secondary structure with a force constant of 200 kcal mol−1 rad−2 . Default values were used to restrain hydrogen bonds , cis-peptide bonds and chiral centres . All steps were performed using the VMD visualization tool ( Humphrey et al . , 1996 ) . The model was optimized in vacuo using NAMD2 ( Phillips et al . , 2005 ) and the CHARMM36 force field ( Best et al . , 2012 ) for proteins and nucleic acids . Pre-ribosomal particles were purified from S . cerevisiae BY4741 cells by tandem affinity purification . Genomically expressed Lsg1-TAP was used as bait protein to isolate ribosome assembly intermediates . Therefore , 12 L YPD medium was inoculated from 300 ml overnight culture with an OD600 of 0 . 1 and grown at 30°C to OD600 = 0 . 8–1 . 0 . Cells were harvested by centrifugation at 4300 x g and 4°C for 12 min . Cell pellets were resuspended in 80 ml cold lysis buffer ( LB-P , 50 mM HEPES , pH 7 . 4 , 100 mM KCl , 1 . 5 mM MgCl2 , 0 . 1% ( v/v ) NP-40 , 5% ( v/v ) glycerol , pefabloc 1:100 , aprotinin and leupeptin 1:1000 ) and centrifuged at 4000 x g and 4°C for 5 min . Washed cell pellets were resuspended in 20 ml LB-P and dripped into liquid nitrogen . Frozen droplets of cell suspension were stored at −80°C until milling in a pre-cooled Retsch ball mill MM400 at 30 Hz for 2 × 60 s . 150 ml ice cold LB-P was added to the frozen cell powder which was thawed on a rolling mixer at 4°C . Cell debris was separated from S . cerevisiae lysate by centrifugation at 30 , 000 x g and 4°C for 20 min . The lysate was incubated with 1 . 2 ml equilibrated IgG sepharose beads ( GE Healthcare ) at 4°C for 3 hr . IgG beads were washed 3x with LB-P and 1x with LB-DTT ( 50 mM HEPES , pH 7 . 4 , 100 mM KCl , 1 . 5 mM MgCl2 , 0 . 1% ( v/v ) NP-40 , 5% ( v/v ) glycerol , 1 mM DTT ) . IgG beads were loaded onto a 5 mL Polyprep column using 3 × 10 mL LB-DTT . The column was closed and IgG beads were incubated in 4 . 5 mL LB-DTT with 175 µl TEV protease ( produced in-house , 1 . 5 µg/µL in 10% glycerol ) at 4°C over night on a rolling incubator . IgG eluate was incubated with 1 mL equilibrated calmodulin affinity resin ( Agilent ) in 15 mL LB-CaCl2 ( 50 mM HEPES , pH 7 . 4 , 100 mM KCl , 1 . 5 mM MgCl2 , 0 . 02% ( v/v ) NP-40 , 5% ( v/v ) glycerol , 2 mM CaCl2 ) at a final CaCl2 concentration of 2 mM on a rolling mixer at 4°C for 3 hr . Calmodulin beads were loaded onto a 5 mL Polyprep column using 2 × 20 mL LB-CaCl2 and washed with 1 × 10 mL LB-CaCl2 . The column was closed and calmodulin beads were incubated with 550 µL LB-EGTA ( 50 mM HEPES , pH 7 . 4 , 100 mM KCl , 1 . 5 mM MgCl2 , 0 . 01% ( v/v ) NP-40 , 5% ( v/v ) glycerol , 5 mM EGTA ) for 20 min at 4°C on a rolling incubator . The eluate was collected and the elution was repeated 3x with 450 µL LB-EGTA . Eluates 1–4 were concentrated using an Amicon Ultra 10 K 0 . 5 mL filter ( Merck Millipore ) to a final volume of ca . 100 µl in crosslinking buffer ( 20 mM HEPES , pH 8 . 3 , 5 mM MgCl2 ) . XL-MS was carried out essentially as described ( Leitner et al . , 2014 ) . In short , roughly 100 µg of eluate was directly cross-linked with 1 . 5 mM disuccinimidyl suberate d0/d12 ( DSS , Creativemolecules Inc ) , digested with trypsin and subsequently enriched for cross-linked peptides . LC-MS/MS analysis was carried out on an Orbitrap Fusion Tribrid mass spectrometer ( Thermo Electron , San Jose , CA ) . Data were searched using xQuest in iontag mode against a database containing ribosomal proteins and known assembly factors ( total of 380 proteins ) of S . cerevisiae with a precursor mass tolerance of 10 ppm . For each experiment , only unique cross-links were considered and only high-confidence cross-linked peptides that were identified with a delta score ( deltaS ) below 0 . 95 and an Id-Score above 32 , translating to an FDR of 0 . 2 ( Erzberger et al . , 2014 ) , were selected for this study . Crosslinks were visualised by xiNET software ( Combe et al . , 2015 ) . All NMR data was collected at 298 K on an Avance II + 700 MHz spectrometer , equipped with a cryogenic triple-resonance TCI probe . 2D BEST-Trosy and standard 3D triple-resonance experiments were acquired with a sample of 100 μM 15N , 13C labelled A . fulgidus Nmd3 ( residues 22–150 ) in PBS buffer with 1 . 5 mM DTT at pH 7 . 2 . Data were processed using Topspin 3 . 0 ( Bruker ) and analysed using SPARKY ( T . D . Goddard and D . G . Kneller - University of California , San Francisco ) . PCR was used to amplify the coding sequence for wild-type uL16 and Nmd3 plus 500 bp of upstream and downstream of the coding sequence using yeast genomic DNA as template . PCR products were cloned into vectors pRS316 ( CEN , URA ) and pRS313 ( CEN , HIS ) using NEBuilder HiFi DNA Assembly Master Mix ( New England Biolabs ) . Partially overlapping primers containing the mutation were used to perform site-directed mutagenesis . For plasmids and primers , see Supplementary files 3A , B . Haploid yeast cells ( strain NE0206 , see Supplementary file 3C ) transformed with plasmids expressing wild type or mutant Nmd3 or vector alone were spotted in ten-fold serial dilutions onto solid synthetic defined -Ura -His medium containing glucose as carbon source for 2 days at 37 ˚C .	Biological machines called ribosomes make proteins in the cells of our body . Mammalian cells build roughly 7 , 500 new ribosomes every minute , each one containing 80 proteins and four RNA molecules . Problems that prevent ribosomes from assembling correctly have been linked to cancers such as leukemia , and a class of disorders called ribosomopathies that increase the likelihood of someone developing cancer . Understanding how ribosomes assemble could therefore help to develop new treatments for these diseases . Ribosomes are mostly constructed in the cell nucleus , but the final stages of assembly occur in the cytoplasm of the cell . A protein called Nmd3 binds to the partly constructed ribosome to export it out of the nucleus . Then , the final ribosomal proteins integrate into the structure to form a key site called the peptidyltransferase centre ( PTC ) , which is where the ribosome joins together amino acids when making new proteins for the cell . Questions remained about how these final assembly steps occur , and how Nmd3 is removed from the ribosome . Kargas et al . have now examined how the PTC forms by using a method known as cryo-electron microscopy to determine the structures that the ribosome forms at different stages of assembly . This revealed that when the last two ribosomal proteins integrate into the ribosome , the ribosomal RNA goes through large shape changes that evict Nmd3 from the PTC . Quality control factors then check the structure of the newly formed ribosome and , if it passes their checks that it works correctly , license it to start making cell proteins . This stage of ribosome assembly is likely to occur in the same way in all plant , animal and other eukaryotic species . The results presented by Kargas et al . will also help researchers to better understand the consequences of the mutations that affect ribosomal proteins in cancer cells . Ultimately , this knowledge may help to uncover new ways to treat cancer and ribosomopathies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2019	Mechanism of completion of peptidyltransferase centre assembly in eukaryotes
Variation in the risk and severity of many autoimmune diseases , malignancies and infections is strongly associated with polymorphisms at the HLA class I loci . These genetic associations provide a powerful opportunity for understanding the etiology of human disease . HLA class I associations are often interpreted in the light of ‘protective’ or ‘detrimental’ CD8+ T cell responses which are restricted by the host HLA class I allotype . However , given the diverse receptors which are bound by HLA class I molecules , alternative interpretations are possible . As well as binding T cell receptors on CD8+ T cells , HLA class I molecules are important ligands for inhibitory and activating killer immunoglobulin-like receptors ( KIRs ) which are found on natural killer cells and some T cells; for the CD94:NKG2 family of receptors also expressed mainly by NK cells and for leukocyte immunoglobulin-like receptors ( LILRs ) on myeloid cells . The aim of this study is to develop an immunogenetic approach for identifying and quantifying the relative contribution of different receptor-ligand interactions to a given HLA class I disease association and then to use this approach to investigate the immune interactions underlying HLA class I disease associations in three viral infections: Human T cell Leukemia Virus type 1 , Human Immunodeficiency Virus type 1 and Hepatitis C Virus as well as in the inflammatory condition Crohn’s disease . Genetic associations provide a powerful opportunity for understanding the etiology of human disease since , unlike most human observational studies , once linkage disequilibrium is corrected for , associated genes can be assumed to be causal rather than simply correlative . The HLA region is a well-known hotspot for disease associations: it comprises just 0 . 3% of the genome yet contains 6 . 4% of the significant SNP associations in the EMBL-EBI genome wide association study ( GWAS ) catalog ( MacArthur et al . , 2017 ) . Multiple candidate gene and GWAS studies have identified significant associations between polymorphisms in the classical HLA class I genes and the risk and/or severity of infectious disease , a range of autoimmune conditions and a number of forms of cancer ( Matzaraki et al . , 2017 ) . A search of the literature yields over 2000 papers reporting HLA class I disease associations . Some of the most striking odds of disease are seen with autoimmune conditions such as the association between ankylosing spondylitis and possession of HLA-B27 ( odds ratio >100 ) or between Behçet’s Disease and HLA-B51 ( accounts for 32–50% of cases ) ( Matzaraki et al . , 2017; de Menthon et al . , 2009 ) . Amongst infectious pathogens , HIV-1 has some of the best studied genetic associations: HLA-B57 is strongly associated with reduced viral load and slow progression of disease in multiple cohorts whilst HLA-B35Px is associated with high viral load and poor prognosis ( Pereyra et al . , 2010; Carrington and O'Brien , 2003; Martin et al . , 2007; Carrington et al . , 1999; Kiepiela et al . , 2004 ) . However , interpretation of HLA class I disease associations is problematic since the classical HLA class I molecules ( HLA-A , -B and –C ) , which bind cytosolic peptides ( typically of length 8–11 amino acids ) have multiple functions . HLA class I molecules are the ligands for several different receptors expressed by different immune cells including CD8+ T cells , NK cells and dendritic cells . CD8+ T cells recognise HLA:peptide via their T cell receptor ( TCR ) . TCR-HLA:peptide binding is exquisitely specific and depends both on the HLA allele and the sequence of the bound peptide . The affinity of an HLA class I molecule for a peptide is a significant determinant of the CD8+ T cell response elicited by that peptide ( Chen et al . , 2000; Müllbacher et al . , 1999; Yewdell and Bennink , 1999; Deng et al . , 1997; Boggiano et al . , 2005 ) ; it has been shown that 85% of epitopes bind their HLA molecule with an affinity of 500 nM or stronger ( Assarsson et al . , 2007 ) . However the relationship between HLA binding and immunogenicity is nontrivial , 50–66% of peptides that bind do not elicit a response ( Lee et al . , 2004; Hoof et al . , 2010 ) and conversely cases where a peptide has undetectable binding but still elicits a response have also been described ( Lee et al . , 2004 ) . A second crucial determinant of immunogenicity is binding of the HLA:peptide complex to the T cell receptor . It has been established that peptide positions P4-8 are most likely to be in close contact with the TCR ( Rudolph et al . , 2006; Garboczi et al . , 1996; Calis et al . , 2012 ) . However , which of these peptide positions are critical for a T cell’s ability to bind have not been comprehensively mapped in humans; additionally non-contact positions can also impact on TCR specificity ( Hausmann et al . , 1999 ) . Most studies only investigate one or two HLA molecules; the residues identified as important in these studies ( for TCR recognition rather than HLA binding ) include P4 , 6 and 8 ( Lee et al . , 2004 ) , P3 , 5 , 6 and 8 ( Tynan et al . , 2005 ) , P3-5 ( Wooldridge et al . , 2010 ) and P3-6 and 8 ( Hausmann et al . , 1999 ) . A comprehensive study of murine data by the Kesmir group convincingly identified P4-6 ( Calis et al . , 2013 ) but it is unclear how this translates to human HLA:peptide as sparsity of data prevents a similar analysis in humans . NK cells bind HLA class I molecules via two distinct groups of receptors , killer immunoglobulin-like receptors ( KIRs ) and CD94:NKG2 . KIRs are a family of inhibitory and activating receptors that are expressed mainly on the surface of NK cells and also some T cells . KIRs recognise broad groupings of HLA class I molecules sharing structural motifs . For instance KIR2DL1 binds HLA-C molecules with an asparagine at position 80 ( designated the C2 group of alleles ) . Whereas KIR2DL3 binds HLA-C molecules with a lysine at position 80 ( designated the C1 group ) ( Trowsdale , 2001; Moesta et al . , 2008 ) . Exceptions to these broad rules have been described ( Sim et al . , 2017 ) . KIR binding also shows some dependence on the HLA-bound peptide , particularly positions 7 and 8 , though this specificity is weak compared to that of the TCR ( Fadda et al . , 2010; Maenaka et al . , 1999a; Maenaka et al . , 1999b; Boyington et al . , 2000 ) . The second way in which NK cells ( and to a lesser extent , T cells ) survey HLA class I expression is via the CD94:NKG2 family of receptors . Particularly interesting is the inhibitory CD94:NKG2A receptor which ligates the monomorphic non-classical HLA-E loaded with peptide from the leader sequence of HLA-A , -C and a subset of –B molecules ( Braud et al . , 1998; López-Botet et al . , 2000 ) and which has been recently shown to play a key role in NK cell education ( Horowitz et al . , 2016 ) . Finally , HLA class I molecules are also the ligands for the leukocyte immunoglobulin-like receptors ( LILR ) of which LILRB1 and LILRB2 are the best characterised . Different HLA allotypes bind LILRB1 and LILRB2 with varying affinity , this is particularly true for LILRB2 which shows considerable variation across the HLA alleles ( Jones et al . , 2011 ) . LILRB1 and LILRB2 are inhibitory receptors which are expressed mainly on myeloid cells including dendritic cells and macrophages; signalling via LILR affects the activation of these antigen presenting cells ( Bashirova et al . , 2014 ) . To the best of our knowledge the impact of HLA-bound peptide on LILR signalling has not been investigated . As a result of these diverse functions of HLA class I molecules the biological mechanisms underlying HLA associations with disease outcome are difficult to infer and contradictory interpretations of the same associations are common . Consider , for example , HLA-B57-associated protection in the context of HIV-1 infection . It has been suggested that HLA-B57 is protective because it preferentially presents CD8+ T cell epitopes from the highly conserved Gag p24 protein which is less susceptible to escape mutations ( Borghans et al . , 2007; Kiepiela et al . , 2007; Miura et al . , 2009 ) and that T cell responses to Gag are specifically associated with a reduced HIV-1 viral load ( Kiepiela et al . , 2007 ) . Indeed , the association between B57 and low viral load is widely cited as evidence that CD8+ T cells are important in controlling HIV-1 ( Boppana and Goepfert , 2018; Walker and McMichael , 2012 ) . However , it has also been argued ( Flores-Villanueva et al . , 2001 ) that HLA-B57 is protective because of its role as a KIR ligand ( binding both the inhibitory receptor KIR3DL1 and the activating receptor KIR3DS1 ) ; an argument which , though disputed ( O'Brien et al . , 2001 ) , does appear to be supported by subsequent studies ( Martin et al . , 2007; Martin et al . , 2002; Pelak et al . , 2011 ) . It has also been suggested that the unusually weak binding of HLA-B57:01 for LILRB2 contributes to control of HIV-1 viral load due to its reduced inhibitory regulation of dendritic cells ( Bashirova et al . , 2014 ) . Finally , recent papers have called for a re-evaluation of HLA class I clinical associations , including HLA-B57 , taking into account the fact that HLA-B57 , by virtue of having a methionine at position 21 provides peptides for HLA-E thus supplying CD94:NKG2A ligands ( Horowitz et al . , 2016; Yunis et al . , 2007 ) . In short , the HLA-B57 protective effect may be attributable to CD8+ T cells , to inhibition of NK cells via KIR3DL1 , to activation of NK cells via KIR3DS1 , to inhibition of NK cells via CD94:NKG2A and/or to reduced inhibition of DCs via LILRB2 . Observational and in vitro studies to investigate the mechanism underlying the B57-protective effect have yielded inconclusive results . The interpretation of observational studies are problematic since , for example , a preponderance of polyfunctional CD8+ T cells in HLA-B57+ elite controllers of HIV-1 may be because polyfunctional CD8+ T cells are responsible for elite control . But it is hard to rule out the possibility that polyfunctionality is a consequence of low viral load . In vitro functional work is also difficult to interpret , with CD8+ T cells , natural killer cells and DCs all playing a role depending on the in vitro experimental conditions , with no obvious means to infer the relative importance of these different factors in vivo . The aim of this study is to develop an immunogenetic approach for identifying and quantifying the relative contribution of different receptor-ligand interactions to a given HLA class I disease association . We applied this approach to investigate well-described associations between single HLA class I alleles and disease in 3 viral infections: Human T cell Leukemia Virus type 1 ( HTLV-1 ) , Human Immunodeficiency Virus type 1 ( HIV-1 ) and Hepatitis C Virus ( HCV ) ( Carrington and O'Brien , 2003; Kim et al . , 2011; Jeffery et al . , 1999 ) . We then extended the scope of this work by using the method to investigate the association between the multi-gene ancestral MHC 8 . 1 haplotype and good prognosis in Crohn’s disease ( Lee et al . , 2017 ) . We focus on receptor-ligand pairs which are polymorphic and well-characterised . Specifically , we investigate TCR-HLA:peptide , inhibitory KIR-HLA:peptide , activating KIR-HLA:peptide , LILRB1-HLA and LILRB2-HLA . The strategy was first to develop a metric for quantifying the proximity or similarity of HLA class I alleles in terms of their TCR binding ( i . e . a metric in ‘CD8+ T cell recognition space’ ) , metrics for quantifying the proximity of HLA class I alleles in terms of their activating and inhibitory KIR binding ( i . e . distance metrics in ‘NK cell recognition space’ ) and metrics for quantifying the proximity of HLA class I alleles in terms of their LILRB1 and LILRB2 binding ( i . e . distance metrics in ‘DC recognition space’ ) . Next , for the HLA class I allele with the disease association of interest ( henceforth the ‘index’ allele ) , the similarity to all other HLA class I alleles in terms of TCR binding , inhibitory KIR binding ( iKIR ) , activating KIR ( aKIR ) , LILRB1 and LILRB2 binding was estimated . Multivariate regression was used to quantify the association between similar HLA class I alleles and clinical outcome . We hypothesised that , if an HLA class I disease association is attributable to CD8+ T cells then other HLA class I alleles with similar TCR-HLA:peptide binding to the index allele would have similar disease associations whereas HLA class I alleles with similar KIR-HLA:peptide and LILR-HLA binding would show no disease associations . Conversely , if the HLA class I allele disease association is attributable to NK cells then HLA class I alleles that are near in KIR-HLA:peptide binding space but not HLA class I alleles which are near in TCR-HLA:peptide binding space or LILR-HLA binding space would be associated with disease . And similarly for LILR binding . Inclusion of combinations of distance metrics as predictors in the regression also allows us to quantify the relative contribution of different receptor-ligand interactions and whether or not they behave independently in the case where more than one interaction was identified as playing a role . Details of the distance metrics are provided in the Methods , a brief , more intuitive , summary is provided below . Figure 1 illustrates the approach . We studied a case-control cohort of 392 HTLV-1-infected individuals from Kagoshima in Southern Japan . 178 subjects were asymptomatic carriers of the virus , and 214 subjects were diagnosed with HTLV-1-associated myelopathy/tropical spastic paraparesis ( HAM/TSP ) according to World Health Organisation criteria . There are well-documented associations between HLA-A02 , HLA-C08 and HLA-B54 and outcome ( Jeffery et al . , 1999; Jeffery et al . , 2000 ) . HLA-A02 and HLA-C08 are protective: they are associated with a reduced risk of HAM/TSP whilst HLA-B54 is detrimental: it is associated with an increased risk of HAM/TSP . There are also reported associations with proviral load , but these associations suffer from poor robustness ( next section ) so we do not investigate them further . We first identified which HLA alleles at the 4 digit level were driving these disease associations . A02:06 and A02:07 are associated with a reduced risk of disease , C08:01 was weakly associated with a reduced risk of disease and B54:01 was associated with an increased risk of disease ( Appendix 3—table 1 ) . To assess which immune interactions were responsible for these associations with HAM/TSP , for each of these 4 HLA alleles , we performed 5 regressions , one for each of the similarity metrics ( TCR . FS , aKIR . FS , iKIR . FS , LILRB1 . S and LILRB2 . S ) , in each case the index allele ( together with age and gender ) were included as covariates . Results are given in Table 1 . We found that for each of the index HLA class I alleles considered , TCR . FS was strongly associated with risk of disease and in the same direction as the index allele i . e . possession of alleles with similar TCR binding to A02:06 , A02:07 and C08:01 are associated with a large decrease in the risk of disease whilst possession of alleles near B54:01 in TCR binding space is associated with a significant increase in the risk of disease . In every case inclusion of TCR . FS in the multivariate analysis strengthened the effect of the index allele ( i . e . increased the magnitude of the coefficient ) indicating that removal of near alleles from the baseline made the ‘background’ alleles more dissimilar to the index . None of the other metrics were significant for any of the index alleles considered . We conclude that for all 4 HLA class I alleles studied in HTLV-1 infection the protection/susceptibility associated with those alleles is best explained by their TCR binding properties and therefore is most likely to be attributable to CD8+ T cells . Having investigated the ‘extreme case’ alleles that are most strongly associated with protection or susceptibility in HTLV-1 infection we next sought to analyse the larger group of ‘average’ alleles associated with intermediate risk of HAM/TSP . Here we define an ‘average allele’ as one that is not significantly associated with outcome ( p>0 . 05 ) , is represented in the cohort at a sufficient frequency ( N > 15 ) and has sufficient near alleles to permit an analysis ( N > 15 with 50% or more similarity ) . It is possible that there is no meaning in the rank order of protection associated with different average alleles . That is , it is possible that , other than the extremes , most HLA class I alleles confer very similar levels of protection and the order of protection that we see is simply a function of the particular cohort studied ( and that analysis of another cohort would yield a different rank order ) . However , a subsampling strategy revealed that this was not the case and that rank order of intermediate alleles was robust and significantly more informative than random when considering the risk of disease but not when considering proviral load ( Appendix 2 Supplementary Results ‘Are average HLA class I associations robust’ , Appendix 4—figure 2 ) . Consequently , it is meaningful to analyse the impact of average alleles on risk of HAM/TSP but not on proviral load . For each HLA class I allele in the HTLV-1 cohort that was sufficiently frequent ( N ≥ 15 ) we calculated the risk of HAM/TSP associated with that allele . We then calculated the risk of HAM/TSP associated with alleles with similar TCR binding , similar inhibitory KIR binding , similar activating KIR binding , similar LILRB1 binding and similar LILRB2 binding . Alleles which were underpowered ( <15 alleles with greater than 50% similarity ) were discarded . The results were striking ( Figure 2 ) . Across all alleles there was a very strong positive correlation between the protection conferred by an allele and the protection conferred by other alleles with similar TCR binding ( Rs = + 0 . 76 p=5×10−6 , Spearman Correlation two tailed ) . 29 alleles were considered , in every case if the index was protective then alleles with similar TCR binding were also protective and if the index allele was detrimental then alleles with similar TCR binding were also detrimental ( p=4×10−9 , Binomial test ) . No such association was seen for any of the other measures of similarity ( Figure 2 ) . However , if we restricted the NK analysis to KIR binding alleles ( ie . alleles with a C1 , C2 or Bw4 motif ) than a weak association was also seen for iKIR ( Rs = 0 . 6 , p=0 . 07 , Spearman Correlation two tailed ) but not for aKIR . The protection/susceptibility associated with an allele’s nearest neighbours in CD8+ T cell recognition space was a significant determinant of protection/susceptibility ( p=0 . 0006 ) even when all other metrics were included in the model . We conclude that , in HTLV-1 infection , the peptide:TCR binding properties of an allele is a significant determinant of the risk of disease associated with that allele; this is true not only for the extreme case alleles which are associated with significant protection or susceptibility but also for the intermediate ‘average’ alleles without significant associations . We studied a well-characterised cohort of HIV-1 seroconverters from sub-Saharan Africa who were identified when seronegative and followed under Protocol C of IAVI ( International AIDS Vaccine Initiative , 2020; Kamali et al . , 2015 ) . Two allele groups were significantly associated with clinical outcome in this cohort , consistent with findings in several other cohorts ( Carrington et al . , 1999; Gao et al . , 2001 ) . HLA-B35Px was associated with an increased viral load set point but did not have a significant impact on progression . HLA-B57 was associated with a low early viral load set point and slow progression to low CD4 count . We first focussed on B35Px and identified which HLA alleles at the 4 digit level were driving the detrimental association . The detrimental effect was entirely due to B53:01 which was significantly associated with an increased early viral load set point , all other B35Px alleles were infrequent in this cohort; ( Appendix 3—table 2 ) . HLA-B53:01 is unusual in that its nearest allele out of all the 151 alleles in the cohort , in terms of both TCR and KIR recognition was distant ( only 46% similarity ) . We were therefore unable to perform the similarity analysis for TCR and KIR binding as there were no similar alleles . In contrast , we were powered to study LILRB1 and LILRB2 as there were a large number of similar alleles by both these metrics . However , alleles similar to B53:01 in terms of LILRB1 and LILRB2 binding behaved very differently in terms of their impact on clinical outcome: near alleles had no significant impact either on early viral load set point nor on time to low CD4+ cell count; moreover , if anything , near alleles tended to be slightly protective rather than detrimental ( Table 2 ) . We conclude that the detrimental effects of B53:01 are independent of its LILRB binding properties but that we were not powered to study whether TCR or KIR binding effects were determinants of susceptibility . Moving on to the protective B57 allele group , we found that both B57:02 and B57:03 were significantly associated with a reduced early viral set point ( Appendix 3—table 2 ) . We also studied alleles similar to B57:01 since , although B57:01 is very infrequent in this African cohort ( N = 1 carrier ) , it is well described to be protective in other cohorts and whilst there was no power to study B57:01 directly in this cohort there was power to study near alleles by all five metrics . We found a clear picture that alleles with similar aKIR binding to B57:01 , B57:02 and B57:03 were significantly protective ( Table 2 ) . There was also a trend for alleles with similar TCR binding and similar LILRB2 binding to also be protective when these metrics were considered in isolation . But when they were considered in a multivariate analysis with aKIR . FS only aKIR . FS retained significance . We conclude that the main reason for the protective effect of HLA-B57 in this cohort is attributable to its activating KIR binding properties . B57:01 , B57:02 and B57:03 all contain the Bw4-80I KIR binding motif and are thought to bind KIR3DS1 . So , the aKIR . FS for the B57 alleles will be 0 for individuals who are KIR3DS1― or who do not have an allele with a Bw4-80I motif and between 0 and 1 ( depending on the degree of similarity of their alleles to B57 ) for individuals who have the compound genotype KIR3DS1:Bw4-80I ( i . e . possession of KIR3DS1 together with an HLA allele containing the Bw4-80I binding motif ) . So our finding that HLA-B57 is protective because of its aKIR binding properties is consistent with the report , in an independent cohort , that KIR3DS1:Bw4-80I is protective ( Martin et al . , 2007 ) . Interestingly , we found that aKIR . FS was more protective ( Coeff = −0 . 41 p=0 . 006 * ) than KIR3DS1:Bw4-80I ( Coeff = −0 . 22 p=0 . 04 * ) . And , in a model including both terms , aKIR . FS remained protective ( Coeff = −0 . 52 p=0 . 07 . ) whilst the compound genotype KIR3DS1:Bw4-80I loses significance and becomes , if anything , detrimental ( Coeff = +0 . 09 , p=0 . 6 ) . This indicates that though aKIR . FS and KIR3DS1:Bw4-80I are related , they are capturing slightly different features and that aKIR . FS is a stronger determinant of protection . To investigate the difference between aKIR . FS and KIR3DS1:Bw4-80I we plotted the two variables against each other for each individual in the cohort ( Figure 3A ) . This identified three distinct groups of people that all had identical KIR3DS1:Bw4-80I status ( all being KIR3DS1+ HLA-Bw4-80I+ ) but with high ( 0 . 6–1 ) , medium ( 0 . 4–0 . 6 ) or low ( 0–0 . 4 ) aKIR . FS . Despite having the same KIR3DS1:Bw4-80I status these groupings are associated with very different effects on viral load ( Figure 3B , Appendix 3—table 4 ) . In particular people with KIR3DS1 and the group I alleles ( aKIR . FS >0 . 6 ) are strongly protected; an effect that is entirely dependent on the presence of KIR3DS1 ( group I allele with KIR3DS1 Coeff = −0 . 4 p=0 . 007 *; group I allele without KIR3DS1 Coeff = −0 . 05 p=0 . 5 ) . Whereas KIR3DS1 with the group II or group III alleles offers no protection . Pooling the group II and group III individuals ( to increase the numbers ) did not change the finding that group I but not group II+III is associated with a significantly reduced early viral load set point ( Figure 3B ) . Downsampling showed that this difference could not be explained simply by group size ( Figure 3C ) . This finding supports the utility of the fraction shared approach . Here we find that aKIR . FS confirms what is known about KIR3DS1:Bw4-80I but extends it beyond a simple binary descriptor adding extra information that is clearly biologically relevant . Finally , since there was a trend for alleles with similar TCR binding and similar LILRB2 binding to B57 to also be protective , we investigated whether there was any evidence for a protective effect of B57 independent of KIR3DS1 . To this end we excluded everyone with KIR3DS1 from the cohort . HLA-B57:02 and B57:03 both remained significantly protective amongst KIR3DS1― individuals ( Coeff = −0 . 58 p=0 . 027 , Coeff = −0 . 39 p=0 . 019 * ) indicating that not all of the B57:02 and B57:03 protective effect is attributable to their role as KIR3DS1 ligands . Amongst KIR3DS1― individuals the best model to explain the residual protective effect of B57:02 and B57:03 was provided by their TCR binding , i . e . alleles with similar TCR binding were significantly protective ( Appendix 3—table 5 ) . We conclude that HLA-B57 alleles are protective for two reasons: firstly ( and most importantly ) because of their activating KIR binding properties; secondly , their TCR binding properties . As for HTLV-1 infection , we next investigated the interactions responsible for the early viral load associations of average alleles . As before , we first investigated whether the viral load associations of average alleles was meaningful . We found that rank order of intermediate alleles was highly robust and significantly more informative than random ( p<10−16 two sample Kolmogorov-Smirnov Test two-tailed ) Appendix 2 Supplementary Results ‘Are average HLA class I associations robust’ , Appendix 4—figure 2 ) . Unlike HTLV-1 infection , and more in line with expectation , the picture was mixed with no single interaction able to explain all HLA associations ( Appendix 4—figure 3 ) . The only significant correlation was between the protection offered by an allele and the protection offered by alleles with similar LILRB2 binding . However , the correlation was weak ( Rs = 0 . 29 p=0 . 048 , Spearman two-tailed ) and there were plenty of alleles which did not conform to the pattern ( e . g . A2902 is detrimental but alleles with similar LILRB2 binding are protective ) . We conclude that , in HIV-1 infection , different interactions are responsible for the protection conferred by different HLA alleles . For the well-described protective effect of the B57 alleles the dominant effect was explained by binding to activating KIR with a weaker effect attributable to TCR binding ( revealed once individuals with KIR3DS1 were removed from the cohort ) . It is worth noting that in our cohort , unlike in white US cohorts , KIR3DL1:Bw4 was not associated with protection . We studied a case-control cohort of 782 HCV-seropositive individuals . 257 subjects had spontaneously cleared the virus , and 525 subjects were chronically infected . A number of published studies have repeatedly found that the HLA-B57 alleles are associated with increased odds of spontaneous clearance and reduced viral load amongst those chronically infected ( Kuniholm et al . , 2013; Kuniholm et al . , 2010; Thio et al . , 2002 ) . Other alleles ( including A11:01 , A23:01 , C01:02 , C04:01 ) have also been associated with outcome in some studies but are not consistently replicated ( and in particular these effects have not been reproduced in the studies with the largest cohorts ) so we focus solely on the B57 alleles . In the cohort we studied , the B57 protective effect appeared to be attributable to B57:02 ( with B57:03 and B57:05 showing similar effects but were not statistically significant due to low carrier numbers ) ; the more frequent B57:01 allele did not appear to be protective ( Appendix 3—table 6 ) . Unlike in HIV-1 infection , where the B57 protective effect is mainly due to aKIR interactions , in HCV infection there was no evidence that the aKIR-binding properties of the B57 alleles contributed to their protection ( model 4 in Table 3 all the P values for aKIR . FS are very high , all the Coefficients are close to zero ) despite adequate power . Instead the protective effect was explained by TCR . FS ( model 2 ) indicating that CD8+ T cells are most likely responsible for the protective effect of HLA-B57 in HCV infection . The models used contained a number of nominal covariates ( 4 ) , it has been shown that under some circumstances , inclusion of such covariates in logistic regression can reduce power ( Pirinen et al . , 2012 ) . We therefore repeated the analysis of models 1–6 omitting the covariates . This strengthened the finding that the protective effect was attributable to TCR . FS in every case leading to an increase in the size of the coefficient and a decrease in the P value ( Coeff = 1 . 09 , p=0 . 03 , Coeff = 1 . 00 p=0 . 04 * , Coeff = 1 . 25 p=0 . 02* for TCR . FS with B57:02 , B57:03 and B57:05 respectively ) . In contrast aKIR . FS never became significant . Looking across all alleles , there was again substantial evidence that the rank order of protection conferred by average alleles was robust ( p<10−16 two-sample Kolmogorov-Smirnov Test , 93 . 8% of runs significant , Appendix 2 Supplementary Results ‘Are average HLA class I associations robust’ , Appendix 4—figure 2 ) . As for HIV-1 infection , the picture was mixed with no single interaction able to explain all HLA associations ( Appendix 4—figure 4 ) . The strongest positive correlation was seen for alleles with similar TCR binding ( Rs = 0 . 21 p=0 . 09 , Spearman two-tailed ) but the correlation is weak and not significant indicating that although the protection conferred by some alleles is attributable to TCR binding there are many alleles where the protection is better explained by another factor . The ancestral haplotype AH8 . 1 ( also known as MHC 8 . 1 ) is a multigene haplotype consisting of HLA-A01:01 -B08:01- C07:01 -DRB103:01 -DQA105:01 -DQB102:01 . Genes from AH8 . 1 or the complete haplotype have been associated with risk or severity of disease in a number of inflammatory and autoimmune conditions including autoimmune hepatitis , myasthenia gravis , systemic lupus erythematosus , type 1 diabetes , a range of myositis phenotypes and Crohn’s disease ( Lee et al . , 2017; Price et al . , 1999; Manabe et al . , 1993; Miller et al . , 2015; Gorodezky et al . , 2006 ) . We investigated the association between good prognosis in Crohn’s disease and the HLA class I alleles of the AH8 . 1 haplotype . In a cohort of 2650 Crohn’s disease cases , HLA-B08:01 and HLA-C07:01 ( which are in tight linkage disequilibrium with each other and with the class II genes of the haplotype ) but not HLA-A01:01 were significantly associated with good prognosis ( Appendix 3—table 7 ) . Strikingly , despite good power in every case , we did not find that alleles with similar TCR binding , similar iKIR binding , similar aKIR binding or similar LILRB1 and LILRB2 binding to either B08:01 or C07:01 were associated with good prognosis ( Table 4 ) . The results were very clear , in every case the P values were very high and in some cases the direction of association was if , anything , reversed ( nearby alleles were detrimental ) . This suggests that the protection marked by B08:01 and C07:01 is either attributable to other genes in this extended haplotype or that B08:01 and C07:01 confer protection by a mechanism other than immune receptor binding ( Elahi et al . , 2011; Candore et al . , 1995 ) . We have developed an approach , based on biologically-plausible similarity metrics , to help identify the immune interactions responsible for the protection or susceptibility associated with a given HLA class I allele . First we applied this approach to investigate HLA disease associations in 3 viral infections . We studied a total of 11 HLA alleles from 6 allele groups . In every case , with the exception of B53:01 in HIV-1 infection ( where we had insufficient power to study TCR , iKIR and aKIR ) we were able to successfully identify the most likely cause of the protective or detrimental effect . In HTLV-1 infection , the pattern was remarkably skewed . All 4 HLA disease associations were best explained by TCR binding . This pattern actually extended to all HLA alleles , with a very strong correlation between the protection conferred by an allele and the protection conferred by alleles with similar TCR binding . This is a wholly unexpected result that implies that , in HTLV-1 infection , the most important immune response in determining protection via all HLA alleles is overwhelmingly the CD8+ T cell response . In HIV-1 infection , the picture was more balanced . The protective effect of B57:01 , B57:02 and B57:03 was mainly due to aKIR with evidence for a weaker effect of TCR binding . Across all alleles no single interaction was responsible for the degree of HLA-mediated protection conferred . In HCV infection the B57 alleles were also protective , but unexpectedly for a different reason to in HIV-1 . In contrast to HIV-1 infection , there was no evidence that protection was attributable to aKIR instead TCR binding appeared to be the main determinant . We then applied the method to investigate the association between the classical HLA class I alleles of the AH8 . 1 haplotype and good prognosis amongst cases of Crohn’s disease . The resulting picture was very clear: none of the immune interactions investigated explained the protective effect . We conclude that either HLA-B08:01 and C07:01 mark a protective allele but are themselves not protective or that they protect via a mechanism independent of their receptor binding . A number of studies have reported that the AH8 . 1 haplotype is associated with impaired immune activation , perhaps due to a defect in the TCR signal transduction pathway ( Candore et al . , 1995; Lio et al . , 1995; Egea et al . , 1991 ) in agreement with our conclusion that the protection associated with HLA-B08:01 and C07:01 is not attributable to their receptor binding . It is interesting to note that in HIV-1 infection all the B57 alleles with high carrier frequency ( B57:01 , B57:02 , B57:03 ) are associated with protection both in the cohort we study and in the work of others ( Carrington and O'Brien , 2003; Kiepiela et al . , 2004; Costello et al . , 1999; Fellay et al . , 2007 ) . However , in HCV infection , B57:02 and B57:03 are protective , but the evidence for B57:01-mediated protection is less clear . B57:01 is not significantly protective in the cohort we study despite a large number of carriers and this has also been reported by others , for example , Kuniholm et al . found both B57:01 and B57:03 were protective in a univariate analysis but in a multivariate analysis B57:01 lost significance ( presumably due to linkage with other HLA alleles ) whilst B57:03 retained significance ( Kuniholm et al . , 2013 ) . Our metrics provide an explanation for this divergent behaviour . The fraction shared metrics all depend upon the proteome of interest . For the HIV-1 proteome , all the B57 alleles are very similar ( e . g . when the index is B57:02 the TCR . FS with B57:01 is 0 . 74 and only 2 alleles –B58:02 and B57:03- are more similar ) . In contrast for HCV , whilst B57:02 and B57:03 are similar , B57:01 is more distant ( e . g . when the index is B57:02 the TCR . FS with B57:01 drops to 0 . 58 and 10 alleles are more similar to B57:02 than B57:01 ) . This provides a plausible explanation for why B57:01 confers a similar degree of protection to B57:02 and B57:03 in the context of HIV-1 infection but appears to have less similarity in the context of HCV infection . Another interesting observation was that , for all three viral infections , the rank order of ‘average’ alleles was robust , that is , they could be robustly classified as protective or detrimental . Although there are a large number of reported HLA class I associations , it is not known which immune interactions are responsible for any of these associations so there does not exist a ‘gold standard’ test data set with which we can formally validate our approach . However , a number of observations suggest that our approach is identifying biologically meaningful features . Firstly , the TCR . FS metric finds that alleles within the same supertype are more similar ( have a higher TCR . FS ) than alleles between supertypes ( p=3×10−7 ) . Secondly , for HTLV-1 infection we find a very strong positive correlation between the risk of HAM/TSP associated with an allele and the risk of HAM/TSP associated with similar alleles ( by the TCR . FS metric ) . Such a strong correlation is unlikely to be generated by random . Thirdly , in HIV-1 infection we found that similarity in terms of activating KIR binding ( aKIR . FS ) revealed clinically relevant subtleties lost within the traditional KIR3DS1:Bw4-80I protective group , essentially splitting this group into 3 groups with decreasing levels of protection; again it is difficult to see how such a result could be generated other than by a method that was reflecting biological reality . We stress that this method should not be used in isolation as the sole method for determining the mechanism underlying an HLA association . Rather it provides a line of evidence to be used alongside other lines of evidence to triangulate to the most likely answer . In common with classical disease association approaches ( which use HLA alleles as predictors ) , this method should be used with care . Many of the problems arising in classical HLA disease association studies , such as linkage disequilibrium , unmeasured confounding variables or population stratification are less of a problem with these metrics . This is because many alleles will be similar and contribute to the similarity metric and it is unlikely that all similar alleles will all be in linkage disequilibrium with the same polymorphism or all correlated with the same confounder . Nevertheless linkage disequilibrium and other correlations can and will distort the results and should be investigated . In short , the method should not be used blindly , it should be used with care by someone with knowledge both of the biological problem and the structure of the dataset . Another limitation of this approach is the relatively simple definition of the KIR binding groups which does not take into account the KIR allele or variations in KIR-HLA binding that break the C1/C2 and Bw4/Bw6 rules . Nevertheless , these simplistic groupings have proved very powerful in other studies ( Martin et al . , 2007; Martin et al . , 2002; Pelak et al . , 2011; Vince et al . , 2014; Khakoo et al . , 2004; Ahlenstiel et al . , 2008; Nakimuli et al . , 2015 ) , indicating that , to a first approximation , they are informative . We do not study all known receptor-HLA interactions; instead we focus on receptor-ligand pairs which are polymorphic and well-characterised . Specifically we investigate TCR-HLA:peptide , inhibitory KIR-HLA:peptide , activating KIR-HLA:peptide , LILRB1-HLA and LILRB2-HLA . Inclusion of other receptors-ligand interactions would follow the same approach but requires more data and better characterisation of the receptors . In particular we did not investigate the CD94:NKG2A-HLA interaction since , with current knowledge , we could only split alleles in a binary fashion into binders or non-binders which would provide no power in subsequent analysis . The strength of binding is known to be more subtle than this and to depend on the peptide interaction with both CD94 and NKG2A ( Petrie et al . , 2008 ) but there is no comprehensive data to allow us to characterise this binding . We do not study atypical effects of the HLA molecules ( Elahi et al . , 2011 ) . Monomorphic receptor-ligand pairs are not studied as these cannot explain polymorphic HLA associations . In addition to identifying immune interactions underlying HLA class I disease associations this approach could also be used alongside classical GWAS or candidate gene approaches as a tool for investigating identified HLA associations . Lack of association between ‘near’ alleles and outcome may indicated that the identified allele is a passenger marking the causal variant or a false positive . Our approach differs from another important attempt to move from HLA associations to understanding function by Raychaudhuri et al . ( 2012 ) . Raychaudrhuri et al extended the usual approach of identifying alleles associated with disease traits and instead fine mapped associations down to the level of amino acids . This high resolution approach identified 5 amino acids , all in the peptide binding grove of HLA molecules , that were associated with seropositive rheumatoid arthritis . However , the authors assumed that variation in the CD8+ T cell response elicited must explain the observed associations . Studies of human disease are typically observational and correlative in nature . An exception to this is genetic association studies which provide a rare and powerful opportunity to uncover causal factors . Here we provide an approach for interrogating some of the most important genes for disease associations: the HLA class I genes . Ultimately , knowing the causal factors will lead to a better understanding of the molecular pathways involved in disease pathogenesis and help to identify potential therapeutic targets . The HTLV-1 , HIV-1 and HCV cohorts were previously KIR and HLA genotyped ( Martin et al . , 2002; Jeffery et al . , 1999; Thio et al . , 2002; Khakoo et al . , 2004; Seich Al Basatena et al . , 2011; Prentice et al . , 2014; Boelen et al . , 2018 ) . For the Crohn’s cohort , subjects were genotyped by imputation using the Immunochip ( Illumina ) , GeneChip 500K array ( Affymetrix ) and the UK Axiom Biobank array ( Affymetrix ) . HLA genotypes were imputed previously ( Lee et al . , 2017 ) . For KIR imputation we first estimated haplotypes for all individuals using SHAPEIT with parameters states = 500 , burn = 10 , prune = 10 , main = 50 and the HapMap b37 recombination map ( Delaneau et al . , 2013 ) . Then , for 984 individuals who were genotyped with the Immunochip , KIR imputations was performed using 231 SNPs on chromosome 19 . For 818 individuals typed using GeneChip 500K Array , KIR imputation was based on 141 SNPs on chromosome 19 . For 410 individuals genotyped with the UK Axiom Biobank array , KIR imputations was based on 145 SNPs on chromosome 19 . In all cases KIR imputation was performed using KIRIMP ( Vukcevic et al . , 2015 ) with a probability threshold of 0 . 5 . The KIR haplotype imputation accuracy has been reported as 92% at a 0 . 5 probability threshold with a call rate greater than 90% . However , this accuracy is for copy-number imputation , we used KIRIMP for imputing presence or absence alone , suggesting we would achieve an accuracy greater than 92% . We constructed 5 similarity metrics to quantify how similar two alleles are in terms of their TCR binding , iKIR binding , aKIR binding , LILRB1 binding and LILRB2 binding . The T cell receptor fraction shared ( TCR . FS ) is defined as the fraction of peptides bound by the index allele whose motifs appear in the peptides bound by the query ( non-index ) allele . The motif coordinates for TCR . FS are flexible , in the results reported here it is determined by the anchor positions 2 and the C terminus and the TCR contact residues position 3–6 . Peptide length is also a variable of the equation . In the results reported here a peptide length of 9 amino acids was used . In the GitHub script ( see key resources table above ) peptide lengths of 8–11 amino acids are allowed . TCR . FS ranges from 0 to 1 . TCR . FS of 0 means that none of the motifs in the index-bound peptides appear in the peptides bound by the non-index allele . If the TCR . FS is 1 , then all the motifs in the index-bound peptides appear in the peptides bound by the non-index allele . Specifically , TCR . FS for an index allele Ai and a query allele Aj is defined asTCR . FS ( Ai , Aj ) =\| Mi∩Mj \|\|Mi\| Where Mi is the list of TCR recognition motifs of index allele Ai , Mj is the list of TCR recognition motifs of query allele Aj , Mi∩Mj is the motifs in the list Mi that are also in the list Mj and \|Mk\| denotes the number of motifs in the list Mk . Note the measure is asymmetric . As a basic test of the TCR . FS metric we investigated the hypothesis that the TCR . FS would be higher between alleles of the same supertype than between alleles of different supertypes . This hypothesis was strongly supported ( Supplementary Results , Appendix 4—figure 1 ) . The inhibitory KIR fraction shared ( iKIR . FS ) is based on the inhibitory KIRs: KIR2DL1 , KIR2DL2 , KIR2DL3 , KIR3DL1 and KIR3DL2 . The activating KIR fraction shared ( aKIR . FS ) is based on the activating KIRs: KIR2DS1 , KIR2DS2 , KIR2DS4 and KIR3DS1 . If the index allele and query allele bind distinct inhibitory KIR ( based on positions 77 and 80 ) ( Trowsdale , 2001; Moesta et al . , 2008 ) their iKIR . FS is set to zero; similarly if the index and query bind distinct activating KIRs , their aKIR . FS is set to 0 . If the index and query allele are both not known to interact with any inhibitory KIRs , their iKIR . FS is assigned a value of 1 and the same being true for activating KIRs and aKIR . FS . Otherwise the iKIR . FS and aKIR . FS is set to a value to reflect the number of shared KIR recognition peptide motifs similar to the TCR . FS . KIR recognition motifs being defined by the KIR contact positions 7 and 8 and the peptide anchor positions 2 and C . Specifically , iKIR . FS for an index allele Ai and a query allele Aj is defined asAiKIR . FS ( Ai , Aj ) ={0AiandAjbinddifferentiKIR0AiandAjbindthesameiKIR , individualisiKIR−1neitherAinorAjbindanyiKIRMi∩Mj\|Mi\|AiandAjbindthesameiKIR , individualisiKIR+ Where Mi is the list of KIR recognition motifs of index allele Ai , Mj is the list of KIR recognition motifs of query allele Aj , Mi∩Mj is motifs in the list Mi that are in the list Mj and \|Mk\| denotes the number of motifs in the list Mk . aKIR . FS is defined similarly . The LILRB1 and LILRB2 similarity scores are based on the LILR-HLA binding data reported by Bashirova et al . in their Supplementary Table S1 ( Bashirova et al . , 2014 ) . The LILRB1 similarity score ( LILRB1 . S ) between an index allele Ai and a query allele Aj is defined as:LILRB1 . S ( Ai , Aj ) =1−\|B ( LILRB1 , Ai ) −B ( LILRB1 , Aj ) \|max{\|B ( LILRB1 , Ak ) −B ( LILRB1 , Am ) \|}where B ( LILRB1 , Ai ) is the binding score between LILRB1 and the HLA allele Ai reported ( Bashirova et al . , 2014 ) and alleles Ak and Am are all HLA alleles whose LILRB1 binding score has been measured ( Bashirova et al . , 2014 ) . The denominator ( which is the same for all allele pairs Ai , Aj ) normalises the score to the maximum difference in binding scores observed so that LILRB1 . S is on the same scale as the other metrics; i . e . 1 corresponds to maximum similarity and zero corresponds to maximum disparity . The LILRB1 similarity scores are calculated on a locus by locus basis . If the LILRB1 binding score for a particular HLA allele has not been measured by Bashirova et al . ( 2014 ) then the similarity score was calculated by taking the mean of all similarity scores for alleles in the same 2-digit group . The LILRB2 similarity scores are defined in an analogous way . Worked examples of the similarity score calculations are provided in Appendix 1 Supplementary Methods . NetMHCpan version 4 . 0 was used to predict the peptides bound by a given HLA class I molecule . The stand-alone software package was used to perform binding affinity predictions ( -BA ) for peptides of length 8 to 11 . A percentile rank of 2% was used as the threshold for bound peptides with peptides below this rank being considered to be bound ( Nielsen et al . , 2007; Jurtz et al . , 2017 ) . Whole proteomes were obtained from Uniprot , 2018 . For HTLV-1 , the subjects were from Kagoshima and so the HTLV-1 subtype prevalent in Japan was used ( Uniprot accession: J02029 ) . For IAVI ( subjects from East and South Africa ) , Zambian HIV-C subtype was used ( Uniprot accession: AB254142 ) . For HCV ( subjects from US and UK ) HCV subtype 1a was used ( Uniprot Accession: M62321 ) . For Crohn’s Disease 100 random human proteins were used . For small proteomes ( e . g . the viral proteomes considered here which were all less than 5000 amino acids ) then results differ depending on the choice of proteome ( e . g . alleles which are similar in terms of binding to HCV are not necessarily similar in terms of binding to HIV-1 ) but once the proteome becomes large , then the exact choice of proteome does not impact the results . Each individual in each cohort was allocated 5 measures representing the distance of the nearest allele in their genotype to the index allele in TCR recognition space , iKIR recognition space etc ( for all 5 similarity metrics ) . That is , an individual heterozygote at all 3 loci ( HLA-A , HLA-B , HLA-C ) would have 6 different TCR . FS values for a specific index allele; the maximum of these six TCR . FS values ( i . e . the nearest allele in their genotype ) would be allocated to the individual . This methodology was applied for all metrics . Multivariate linear , logistic and Cox regression was used to quantify the impact of the index allele and near alleles by the different similarity metrics on the outcome of interest . Potentially confounding covariates were identified and included in the analysis ( listed in Appendix 1 Supplementary Methods ) . The analysis was of the form:Outcome∼Indexallele+FSmetric+covariates And was repeated for each of the 5 metrics . Combinations of metric were also considered to determine independence and relative sizes . In the scenario where the same index HLA allele is protective or detrimental via different mechanisms in different people , then this would be detectable as two independently protective ( or detrimental ) FS metrics ( i . e . that did not lose significance when included in the regression together ) ; as an example of this please see the results section”HIV-1 infection: why are HLA-B*57 alleles protective’ in which different mechanisms of protection are identified in different people . All reported P values are two tailed . Calculations were performed using R v3 . 4 . 1 ( R Development Core Team , 2014 ) . An implementation of this method is available on Github at https://github . com/bjohnnyd/fs-tool ( Debebe et al . , 2020; copy archived at https://github . com/elifesciences-publications/fs-tool ) .	When considering someone’s risk of disease , every person is different but some similarities can be found when looking across populations . Some people are more likely to develop a certain disease , while others are protected in some way . Part of this variation is explained by the individual’s genes , while their lifestyle and environment are other factors . Numerous studies have looked for associations between different versions of genes , known as gene variants , and the occurrence of disease to identify who is at risk . There is one cluster of genes called the HLA genes that is a well-known hotspot for disease associations . The HLA cluster is named for the group of proteins it encodes , called the human leukocyte antigen ( HLA ) complex . These cell-surface proteins regulate the immune system in humans . These proteins are present on the surface of cells , and they help the immune system distinguish foreign invaders such as viruses and bacteria from the body’s own cells . Variants in the HLA genes are associated with more than 100 diseases , including infectious diseases like HIV , autoimmune conditions such as multiple sclerosis , and some cancers . However , while identifying which genetic variants are associated with an increased or decreased risk of disease is relatively simple , understanding why those genetic variants are associated with a particular disease is much harder . Debebe et al . have developed a new method to find out why certain gene variants in the HLA cluster are associated with disease in humans . They used this method to investigate known genetic variants associated with three viral infections: HIV , hepatitis C , and human leukemia virus – and one inflammatory disease: Crohn’s disease . Critically , Debebe et al . looked at the interactions between different immune cells and the cell-surface proteins encoded by the HLA gene variants in different cases of these diseases . In doing so , the analysis was able to identify which cells of the immune system were responsible for the associations between gene variants and diseases . In principle , this method could be applied to study any disease in any species . It could also be used in classic gene association studies to test for false positive results and “passenger” mutations , two common problems that beset sound interpretations from these studies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "immunology", "and", "inflammation" ]	2020	Identifying the immune interactions underlying HLA class I disease associations
The Arabidopsis mutant wrky33 is highly susceptible to Botrytis cinerea . We identified >1680 Botrytis-induced WRKY33 binding sites associated with 1576 Arabidopsis genes . Transcriptional profiling defined 318 functional direct target genes at 14 hr post inoculation . Comparative analyses revealed that WRKY33 possesses dual functionality acting either as a repressor or as an activator in a promoter-context dependent manner . We confirmed known WRKY33 targets involved in hormone signaling and phytoalexin biosynthesis , but also uncovered a novel negative role of abscisic acid ( ABA ) in resistance towards B . cinerea 2100 . The ABA biosynthesis genes NCED3 and NCED5 were identified as direct targets required for WRKY33-mediated resistance . Loss-of-WRKY33 function resulted in elevated ABA levels and genetic studies confirmed that WRKY33 acts upstream of NCED3/NCED5 to negatively regulate ABA biosynthesis . This study provides the first detailed view of the genome-wide contribution of a specific plant transcription factor in modulating the transcriptional network associated with plant immunity . Necrotrophic fungi including Botrytis cinerea , Fusarium oxysporum , and Alternaria brassicicola are the largest class of fungal phytopathogens causing serious crop losses worldwide ( Łaźniewska et al . , 2010 ) . These pathogens extract nutrients from dead host cells by producing a variety of phytotoxic compounds and cell wall degrading enzymes ( Williamson et al . , 2007; Mengiste , 2012 ) . B . cinerea has a broad host-range , causes pre- and postharvest disease , and is the second most agriculturally important fungal plant pathogen ( Dean et al . , 2012 ) . Plant immunity towards B . cinerea appears to be under complex poorly understood genetic control ( Rowe and Kliebenstein , 2008 ) . Apart from the Arabidopsis thaliana RESISTANCE TO LEPTOSPHAERIA MACULANS 3 ( RLM3 ) , no major R-gene has been associated with resistance to necrotrophs . However , over the past two decades numerous genes that influence the outcome of B . cinerea—host interactions have been identified ( Mengiste , 2012 ) . Among these are several transcription factors ( TFs ) consistent with the large transcriptional reprogramming observed in host cells upon Botrytis infection ( Birkenbihl and Somssich , 2011; Birkenbihl et al . , 2012; Windram et al . , 2012 ) . In Arabidopsis , several MYB-type TFs regulate distinct host transcriptional responses towards B . cinerea . BOS1 ( BOTRYTIS SUSCEPTIBLE 1 ) /MYB108 appears to restrict necrosis triggered by B . cinerea and A . brassicicola , and loss-of-BOS1 function increased plant susceptibility ( Mengiste , 2012 ) . In response to stress and B . cinerea infection , BOS1 physically interacts with and is ubiquitinated by BOI , a RING E3 ligase that contributes to defense by restricting the extent of necrosis ( Luo et al . , 2010 ) . MYB51 is involved in the transcriptional activation of indole glucosinolate biosynthetic genes , which also contributes to resistance towards necrotrophs ( Kliebenstein et al . , 2005; Sánchez-Vallet et al . , 2010 ) . In contrast , the MYB-related genes ASYMMETRIC LEAVES 1 ( AS1 ) and MYB46 appear to play a role in disease susceptibility as such mutants show increased disease resistance towards B . cinerea ( Nurmberg et al . , 2007; Ramírez et al . , 2011 ) . Ethylene and jasmonic acid ( ET , JA ) signaling are critical for host immunity to necrotrophic pathogens , and several transcriptional activators and repressors of the ET and JA pathways impact resistance to B . cinerea ( Glazebrook , 2005; Bari and Jones , 2009 ) . In particular the TFs ERF1 , ORA59 , ERF5 , ERF6 , and RAP2 . 2 , have regulatory functions in host susceptibility to this fungus . ( Berrocal-Lobo et al . , 2002; Pré et al . , 2008; Moffat et al . , 2012; Zhao et al . , 2012 ) . Transgenic Arabidopsis lines overexpressing ERF1 or ORA59 confer resistance to B . cinerea ( Kazan and Manners , 2013 ) , whereas RNAi-ORA59 silenced lines were more susceptible ( Berrocal-Lobo et al . , 2002; Pré et al . , 2008 ) . Both ERF1 and ORA59 appear to be the key integrators of the ET- and JA-signaling pathways ( Pieterse et al . , 2009 ) . In contrast , the bHLH transcription factor MYC2/JIN1 is a master regulator of diverse JA-mediated responses by antagonistically regulating two distinct branches of the JA signaling pathway in response to necrotrophs ( Kazan and Manners , 2013 ) . The WRKY family of TFs modulates numerous host immune responses ( Pandey and Somssich , 2009 ) . In particular , WRKY33 is a key positive regulator of host defense to both A . brassicicola and B . cinerea ( Zheng et al . , 2006; Birkenbihl et al . , 2012 ) . WRKY33 was directly phosphorylated in vivo by the MAP kinases MPK3 and MPK6 upon B . cinerea infection and subsequently activated PAD3 expression by direct binding to its promoter ( Mao et al . , 2011 ) . PAD3 encodes a key biosynthetic enzyme required for the production of the antimicrobial compound camalexin . Moreover , WRKY33 directly interacted with its own promoter , suggesting a positive feedback regulatory loop on WRKY33 expression . WRKY33 was also found to interact with the VQ-motif containing protein MAP KINASE SUBSTRATE1 ( MKS1/VQ21 ) and to form a ternary complex with the MAP kinase MPK4 within the nucleus of resting cells ( Andreasson et al . , 2005; Qiu et al . , 2008 ) . Upon challenge with the hemibiotrophic pathogen Pseudomonas syringae or upon elicitation by the microbe-associated molecular pattern ( MAMP ) flg22 , the active epitope of the bacterial flagella , activated MPK4 phosphorylates MKS1 thereby releasing WRKY33 from the complex and leading to its detection at the PAD3 promoter . We previously reported that activation of Arabidopsis WRKY33 resulted in rapid and massive host transcriptional reprogramming upon infection with B . cinerea strain 2100 ( Birkenbihl et al . , 2012 ) . Compared to resistant wild-type ( WT ) plants , susceptible wrky33 mutants displayed early inappropriate activation of salicylic acid ( SA ) -related host responses , elevated SA and JA levels , and down-regulation of JA-associated responses at later infection stages . Consistent with these results ChIP analysis demonstrated that WRKY33 directly binds to the regulatory regions of JAZ1 and JAZ5 , two genes encoding repressors of JA signaling , but also to the ERF class TF gene ORA59 involved in JA-ET crosstalk , and to two camalexin biosynthesis genes CYP71A13 and PAD3 ( Birkenbihl et al . , 2012 ) . Although pad3 plants are susceptible to B . cinerea 2100 , wrky33 mutants are more highly susceptible . Genetic studies revealed that altered SA responses at later infection stages may contribute to the susceptibility of wrky33 to B . cinerea , but were insufficient for WRKY33-mediated resistance ( Birkenbihl et al . , 2012 ) . Thus , WRKY33 apparently targets additional genes whose functions are critical for establishing full WRKY33-dependent resistance towards this necrotroph . In this paper , we performed ChIP-seq and RNA-seq analyses to identify WRKY33-regulated target genes in the A . thaliana genome following infection with B . cinerea 2100 . The study uncovered numerous targets many of which are associated with the regulation of hormonal signaling pathways . Expression of the majority of WRKY33 direct targets is down-regulated upon infection , but some notably genes of camalexin biosynthesis are strongly up-regulated , indicating that WRKY33 is a dual functional TF acting in a promoter-context dependent manner . Subsequent genetic and hormonal studies verified components of abscisic acid ( ABA ) biosynthesis as being critical for WRKY33-dependent resistance towards this necrotrophic fungus . This study provides the first genome-wide view of the gene regulatory network underlying plant immunity governed by a host specific TF . To gain a deeper insight into how WRKY33 regulates plant immunity towards B . cinerea 2100 , we performed ChIP-seq for genome-wide in vivo identification of WRKY33 DNA-binding sites . For this , a transgenic wrky33 null mutant expressing an HA epitope-tagged WRKY33 construct under the control of its native promoter ( PWRKY33:WRKY33-HA ) was used . This line complemented the B . cinerea 2100 susceptibility phenotype of wrky33 plants resulting in resistance similar to WT Col-0 plants ( Birkenbihl et al . , 2012 ) . Rosette leaves of 4-week old plants , mock treated or spray inoculated with spores of B . cinerea 2100 , were collected at 14 hr post inoculation and used to perform ChIP-seq . The 14 hr timepoint was selected based on the induced WRKY33-HA protein levels observed in western blots ( Figure 1A ) . No WRKY33-HA protein was detected in the absence of infection . Besides the non-induced sample , we used identically treated WT plant tissue lacking WRKY33-HA as an additional negative control . Two biological replicates each were analyzed . The previously identified WRKY33 in vivo target genes , CYP71A13 and PAD3 , were used to monitor by ChIP-qPCR specific enrichment in samples used for library construction and sequencing ( Birkenbihl et al . , 2012 ) . 10 . 7554/eLife . 07295 . 003Figure 1 . Genome-wide identification of Arabidopsis WRKY33 binding sites . ( A ) Western-blot analysis of WRKY33-HA protein levels after mock treatment or spray-inoculation of PWRKY33:WRKY33-HA transgenic plants with B . cinerea 2100 spores . Plant material selected for ChIP-seq is boxed . ( B ) Relative binding-peak distribution across genomic regions . The 1 kb region upstream of the transcription start site is defined as promoter . The fraction of nucleotides in the complete At genome associated with each annotation type is included in the figure as background control ( At genome ) . ( C ) Distribution of identified WRKY33 binding sites relative to the TSS . ( D ) Conserved DNA elements enriched within the 500 bp WRKY33 binding peak regions identified by DREME motif search . The TTGACT/C motif represents the well-established W-box , whereas T/GTTGAAT is an identified new motif . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 00310 . 7554/eLife . 07295 . 004Figure 1—figure supplement 1 . Conserved DNA elements within the 500 bp WRKY33 binding peak summit regions identified by MEME . ( A ) W-box with 5′ extended motifs . ( B ) W-box with 3′ extended motifs . ( C ) Additional conserved sequence GACTT/ATTC element . ( D ) Venn diagram illustrating the number of overlapping peaks containing both the W-box and the newly identified motif T/GTTGAAT . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 00410 . 7554/eLife . 07295 . 005Figure 1—figure supplement 2 . WRKY33 does not bind to the G/TTTGAAT motif . EMSA was performed using recombinant WRKY33 and biotin labeled DNA probes . A DNA oligonucleotide containing 3 W-box elements ( Mao et al . , 2011 ) or a W-box mutated version ( W-boxmut ) hereof served as a positive and as a negative control , respectively ( A ) . The 45 bp M-3 was derived from the PROPEP3 ( At5g64905 ) gene promoter and contained three copies of the G/TTTGAAT motif , whereas the 40 bp M-7 was derived from the WAKL7 ( At1g16090 ) gene promoter and contained one G/TTTGAAT motif . No binding of WRKY33 was observed to the W-boxmut probe and to both the M-3 and M-7 probes ( B ) . Specificity of W-box binding was shown by competition assays using 250-fold ( W-box ) and 500-fold ( W-box; M-3 , M-7 ) excess of unlabeled probes . Protein lysates derived from IPTG-induced bacteria harboring the empty expression vector pMCSG48 served as an additional control . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 005 We identified 1684 high confidence WRKY33 binding sites common to both replicates , which are associated with 1576 genes ( Figure 2C , Supplementary file 1 ) . WRKY33 binding to all detected sites was dependent on prior infection with B . cinerea 2100 . Over 78% of the identified peak regions were located in promoter ( 1 kb region upstream of the transcription start site ) or 5′ intergenic regions ( Figure 1B ) and 15 . 4% were located near transcription termination sites . Less than 1% and 5% of the peaks were located in exons and intronic regions , respectively ( Figure 1B ) . The genome-wide local distribution of peak regions relative to genes showed clear accumulation of WRKY33 binding at about −300 bp from the transcription start sites ( Figure 1C ) . The fidelity of the ChIP-seq data was subsequently confirmed by ChIP-qPCR for a number of genes ( Supplementary file 2 ) . Moreover , nearly all previously reported WRKY33 in vivo targets including PAD3 , CYP71A13 , ACS2 , JAZ1 , ORA59 , TRX-h5 , and WRKY33 itself were successfully identified in our ChIP-seq dataset ( Mao et al . , 2011; Birkenbihl et al . , 2012; Li et al . , 2012 ) . 10 . 7554/eLife . 07295 . 006Figure 2 . WRKY33-regulated direct target genes in response to B . cinerea 2100 infection . ( A ) Number of differentially expressed genes ( ≥ twofold; p ≤ 0 . 05 ) between WT and wrky33 ( w33 ) at 14 hr after mock treatment ( mo ) or spray inoculation with spores of B . cinerea 2100 ( B . c . ) identified by RNA-seq . Indicated are total numbers ( boxed ) and numbers of up-regulated ( ) and down-regulated genes ( ) between treatments or genotypes . ( B ) Venn diagram illustrating the total numbers and the number of common genes affected in WT and wrky33 14 hr post B . cinerea 2100 inoculation . ( C ) Venn diagram showing the numbers of genes common to WRKY33-regulated genes and WRKY33 target genes . ( D ) Percentage of WRKY33-repressed and WRKY33-induced target genes ( in total 318 ) . ( E ) Enrichment of specific Gene Ontology ( GO ) terms related to defense response , kinase activity , cell death , and hormone responses among WRKY33-regulated target and non-target genes ( compared to the overall genome ) . The y-axis indicates the percentages of genes associated to each GO category in each gene set . Asterisks indicate significant enrichment ( adj . p value < 0 . 05 ) of genes associated to the respective GO term within a gene set as determined by GO term enrichment analysis with goseq ( null distribution approximated as Wallenius distribution; correction for potential count biases via probability weighting ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 00610 . 7554/eLife . 07295 . 007Figure 2—figure supplement 1 . GO graph visualization of top GO terms enriched among WRKY33-regulated WRKY33 target genes . ( A ) GO sub-graph induced by the top 10 GO terms in the category ‘Biological Process’ . ( B ) GO sub-graph induced by the top 5 GO terms in the category ‘Molecular Function’ . Boxes indicate the 10 ( A ) and 5 ( B ) most significant terms , respectively . Box color represents the relative significance , ranging from dark red ( most significant ) to light yellow ( least significant ) . GO term enrichment was determined using goseq , with the null distribution approximated as Wallenius distribution and correcting for potential count biases via probability weighting . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 00710 . 7554/eLife . 07295 . 008Figure 2—figure supplement 2 . Analysis of WRKY33-regulated target genes associated to GO-terms ‘hormone responses and cell death’ . ( A ) Venn diagram showing the overlap between the WRKY33-regulated target genes responsive to SA and associated with cell death . ( B ) The overlap of WRKY33-regulated target genes associated with the hormone pathways SA , ET , ABA , and JA . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 008 Numerous studies have revealed that WRKY proteins specifically bind to a DNA motif , TTGACT/C , termed the W-box ( Rushton et al . , 2010 ) , although adjacent bases ( W-box extended motifs ) can also influence binding ( Ciolkowski et al . , 2008 ) . Using the DREME/MEME software , we determined conserved consensus sequences within high confidence WRKY33 binding sites across the genome . Of the 1684 identified WRKY33 binding regions , 80% contained the well-established W-box motif ( Figure 1D ) . We also found W-box extended sequence motifs within the WRKY33 binding regions ( Figure 1—figure supplement 1A , B ) . These W-box extended motifs also included the core sequence GACTTTT ( Figure 1—figure supplement 1C ) , which was reported to be bound by Arabidopsis WRKY70 and to be required for WRKY70-activated gene expression ( Machens et al . , 2014 ) . Apart from the W-box and W-box variants , we found one additional sequence motif , T/GTTGAAT that occurs in 60% of the WRKY33 binding regions ( Figure 1D ) . More than 48% ( 817 out of 1684 ) of WRKY33 binding peaks contained both this new motif and the W-box ( Figure 1—figure supplement 1D ) . We performed electrophoresis mobility shift assays ( EMSA ) using recombinant WRKY33 protein to determine whether this newly identified DNA element is bound by WRKY33 . Two DNA oligonucleotide probes were synthesized whose sequences were derived from two WRKY33 targets ( WAKL7 , PROPEP3 ) containing either one or three copies of the motif , respectively . A previously described DNA oligonucleotide containing three W-boxes ( Mao et al . , 2011 ) and a W-box mutated version hereof W-boxmut served as positive and negative controls . A clear interaction ( mobility shift ) was observed between WRKY33 and the labeled W-box probe but not with W-boxmut and the two probes harboring the T/GTTGAAT motifs ( M-3 and M-7; Figure 1—figure supplement 2 ) . Specificity of W-box binding was confirmed in competition experiments , wherein only an excess of the W-box probe was able to compete for binding of the protein . Thus , T/GTTGAAT does not appear to be a WRKY33 direct binding site and its functionality remains unclear . ChIP-seq studies in different organisms have revealed that the majority of binding sites bound by specific TFs in vivo do not result in altered expression levels of associated genes ( MacQuarrie et al . , 2011; Chang et al . , 2013; Fan et al . , 2014 ) . To investigate the impact of WRKY33 binding on target gene expression , we performed RNA-seq and examined WRKY33-mediated gene expression changes in mock and B . cinerea 2100 ( 14 hpi ) treated 4-week old wrky33 and WT plants . Three independent biological replicates were generated and analyzed allowing us to identify genes with consistently altered expression after inoculation . In WT plants , the expression of 6101 genes was altered twofold or more ( p ≤ 0 . 05 ) compared to non-infected plants , with 3048 genes being up-regulated and 3053 genes being down-regulated ( Figure 2A ) . In wrky33 , upon infection , the expression of 7441 genes was altered more than twofold , 3583 of them being up-regulated and 3858 down-regulated . A common set of 4686 genes showed changes upon infection in both genotypes ( Figure 2A , B ) . Comparative profiling of mock treated plants identified 705 genes that were differentially expressed between wrky33 and WT in the absence of the pathogen , 458 of them being up-regulated and 247 down-regulated ( Figure 2A ) . Comparing the expression profiles of B . cinerea infected wrky33 and WT plants ( wrky33 B . c vs WT B . c ) , we identified 2765 differentially expressed genes dependent on WRKY33 , of which 1675 were up-regulated in the mutant ( termed WRKY33-repressed genes ) and 1090 were down-regulated in the mutant ( termed WRKY33-induced genes; Figure 2A , C ) . We then compared the WRKY33-dependent differentially expressed gene set obtained by RNA-seq with the WRKY33 target gene set revealed by ChIP-seq . This comparison identified 318 WRKY33-regulated target genes that were both bound by WRKY33 and exhibited WRKY33-dependent altered gene expression ( Figure 2C ) . Of these , 240 ( 75% ) were repressed upon infection while 78 ( 25% ) were induced ( Figure 2C , D ) . We named those genes WRKY33-repressed targets and WRKY33-induced targets , respectively . Based on this analysis , WRKY33 appears to have a prominent repressive role on the transcription of many specific host genes indicating a negative regulatory function of WRKY33 in mediating immunity to this pathogen . Genes displaying altered expression in the wrky33 mutant compared to WT but showing no binding of WRKY33 at their respective gene loci were defined as WRKY33-dependent non-targets ( 1435 WRKY33-repressed non-targets and 1012 WRKY33-induced non-targets; Figure 2C ) . The overlap between observed WRKY33 binding and altered expression of the associated genes upon fungal infection was around 20% ( 318 of 1576 ) . This fraction is similar to values reported for other plant TFs such as EIN3 , HBI1 , and BES1 ( Yu et al . , 2011; Chang et al . , 2013; Fan et al . , 2014 ) . Compared to the entire genome the identified WRKY33-regulated targets were significantly enriched in gene onthology ( GO ) categories involved in diverse biological processes and molecular functions related to different forms of stress , external and endogenous stimuli , signal transduction , transport , metabolic processes and catalytic activity ( p < 0 . 05; Figure 2—figure supplement 1 ) , and many of these genes are repressed upon B . cinerea infection ( Figure 2E ) . For example , genes related to ‘defense response’ were highly overrepresented among WRKY33-repressed targets ( 38% ) and in the WRKY33-repressed non-target sets ( 17% ) , suggesting that WRKY33 mainly functions as a repressor of plant defense responses . However , it is important to note that nearly 18% of the WRKY33-induced targets were associated to defense responses compared to only 3% of the WRKY33-induced non-targets . This indicates that WRKY33 can also act as a direct activator of defense gene expression , very likely in a promoter-context dependent manner . Particularly prominent among the WRKY33-induced targets are genes associated with responses to the phytohormone ethylene ( ET; 21% ) . Apart from hormonal pathways discussed below , genes associated with the GO terms ‘cell death’ or related to diverse ‘kinase activities’ were markedly enriched among WRKY33-repressed targets and non-targets ( Figure 2E ) . 42 out of 318 WRKY33-regulated targets are involved in cell death , and 38 of these appear to be repressed by WRKY33 ( Supplementary file 3 ) . This WRKY33-mediated repression may be an important feature required to reinforce resistance towards the necrotroph B . cinerea that depends on dead host tissue to complete its life cycle . Furthermore , 41 of the WRKY33-regulated target genes encode for various kinases , and again the majority of these genes appear to be negatively regulated by WRKY33 ( Supplementary file 4 ) . For the WRKY33-regulated target LecRK VI . 2 , a critical role in resistance against hemibiotrophic P . syringae pv . tomato DC3000 and necrotrophic Pectobacterium carotovorum bacteria has been demonstrated ( Singh et al . , 2012; Huang et al . , 2014 ) . Several TF gene families involved in defense were targeted by WRKY33 . In total , WRKY33 binding was found at 133 TF gene loci . Predominant among these are members of the AP2/ERFs , MYBs , WRKYs , and NACs families ( Figure 3A ) . However , expression of only 16% ( 21 of 133 ) of these genes was directly modulated in a WRKY33-dependent manner after B . cinerea infection ( complete list see Supplementary file 5 ) . WRKY factors are predicted to form a highly interconnected regulatory sub-network ( Llorca et al . , 2014 ) . Indeed , 18 WRKY genes were identified as direct targets of WRKY33 ( Figure 3A ) . However , only seven genes , WRKY33 , WRKY38 , WRKY41 , WRK48 , WRKY50 , WRKY53 , and WRKY55 , showed altered expression upon WRKY33 binding at 14 or 24 hpi ( Figure 3B–G; Figure 3—figure supplement 1 , Supplementary file 5 ) . Binding to the WRKY33 promoter is consistent with reports suggesting a positive autoregulatory feedback loop resulting in high-level accumulation of WRKY33 in response to B . cinerea ( Mao et al . , 2011 ) . In addition , WRKY33-dependent altered transcription of 18 other WRKY genes with no detectable WRKY33 binding was observed following fungal infection , indicating that these genes are indirectly regulated by WRKY33 ( WRKY33-regulated non-targets; Figure 3A ) . WRKY33 function negatively affected expression of most of these WRKY targets . 10 . 7554/eLife . 07295 . 009Figure 3 . WRKY33-regulated transcription factor families commonly associated with stress responses . ( A ) WRKY , MYB , NAC , and AP2/ERF TF family genes are dominant targets of WRKY33 after B . cinerea 2100 infection . The total number of members for each TF family is given in parenthesis next to name . The number of WRKY33 directly or indirectly regulated family members are indicated . ( B , C ) Integrative Genomics Viewer ( IGV ) images of ChIP-seq data revealing high infection-dependent WRKY33 binding at the promoters of Arabidopsis WRKY33 ( B ) and WRKY41 ( C ) . Images for mock and B . c . treatment of both biological repetitions are shown ( 1 and 2 ) . Structure of the targeted genes is indicated below along with the position of all W-box motifs within the loci . Arrows indicate direction of transcription . ( D , E ) qRT-PCR analysis of B . cinerea 2100-induced expression of WRKY33 ( D ) and WRKY41 ( E ) in WT and wrky33 mutant plants at indicated time points post fungal spore application . All data were normalized to the expression of At4g26410 and fold induction values of all genes were calculated relative to the expression level of mock treated ( mo ) WT plants set to 1 . Error bars represent SD of three biological replicates . Asterisks indicate significant differences between WT and wrky33 ( * , p < 0 . 05; ** , p < 0 . 001; two-tailed t-test ) . ( F , G ) Validation of ChIP-seq data by ChIP-qPCR showing WRKY33 binding to its own promoter region ( F ) and to the WRKY41 promoter ( G ) . WRKY33-HA ( 33HA ) plants were spray inoculated with spores of B . cinerea 2100 ( Bc ) or mock treated ( mo ) for 14 hr . Input DNA before immune precipitation ( IN ) and immune-precipitated DNA using an anti-HA antibody ( IP ) was analyzed by qPCR employing gene-specific primer pairs ( p ) indicated in the IGV graph . Shown is the fold enrichment of bound DNA relative to a non-bound DNA fragment from At2g04450 . As a control for primer efficiency purified genomic DNA was included in the analysis . Each ChIP experiment was repeated at least twice with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 00910 . 7554/eLife . 07295 . 010Figure 3—figure supplement 1 . Validation of WRKY33 directly regulated WRKY genes . Shown for each of the WRKY genes 38 ( A ) , 48 ( B ) , 50 ( C ) , 53 ( D ) , and 55 ( E ) , are; ( i ) the ChIP-seq data , visualized in the IGV browser , revealing strong infection-dependent WRKY33 enrichment at the corresponding promoter , ( ii ) qRT-PCR analysis of B . cinerea 2100-induced expression of the respective WRKY gene in WT and wrky33 at indicated time points post spore inoculation , and ( iii ) ChIP-qPCR confirmation of WRKY33 binding to the respective WRKY promoter . Amplicons used are indicated in the IGV images ( p , p1 , p2 , p3 ) . For detailed descriptions on how qRT-PCR and ChIP-qPCR were performed see legend to Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 010 Genes encoding components of pathways related to the key phytohormone signaling molecules SA , JA , ET , and ABA were highly enriched in the WRKY33-regulated gene set ( Figure 2E ) . Genes involved in SA response were overrepresented in WRKY33-repressed targets and non-targets ( Figure 2E ) . This is consistent with our previous transcriptomic profiling showing that WRKY33 directly or indirectly repressed the expression of genes in SA biosynthesis and SA-mediated signaling ( Birkenbihl et al . , 2012 ) . More than 80% ( 34 out of 42 ) of the SA-response targets are also associated with the GO term ‘cell death’ ( Figure 2—figure supplement 2A ) , suggesting that WRKY33 repression of the SA pathway is linked to modulation of host cell death responses . In contrast to SA signaling genes , genes responsive to ET were highly enriched in the WRKY33-induced target dataset , among them are ACS6 , ORA59 , and ERF5 ( Figure 2E ) . ACS6 is involved in Botrytis-induced ethylene production and plays an important role in plant immunity ( Han et al . , 2010; Li et al . , 2012 ) . ORA59 and ERF5 belong to the AP2/ERF TF family with ORA59 acting as an integrator of JA and ET signaling and as a positive regulator of resistance against B . cinerea , while ERF5 also regulates ET signaling and is a key component of chitin-mediated immunity ( Pré et al . , 2008; Moffat et al . , 2012 ) . Genes responsive to JA and ABA were also overrepresented in our GO term analysis , but in this case similar fractions of genes were identified among WRKY33-induced targets , WRKY33-repressed targets and WRKY33-repressed non-targets ( Figure 2E ) . Some WRKY33-regulated targets were associated to more than one hormone response ( Figure 2—figure supplement 2B ) , suggesting the involvement of WRKY33 in hormonal co-regulation or crosstalk . In conclusion , our global analysis revealed that WRKY33 influences various hormonal responses upon infection with B . cinerea 2100 , and that WRKY33 had both a positive and a negative functional relationship with a fraction of its direct targets . Our previous genetic analyses excluded a major role of SA , JA and ET signaling in WRY33-dependent resistance towards B . cinerea 2100 ( Birkenbihl et al . , 2012 ) . Here , two additional genes , GH3 . 2 and GH3 . 3 , encoding acyl-acid-amide synthetases capable of conjugating amino acids to JA and auxin were identified as being WRKY33-repressed targets ( Figure 8—figure supplement 3A ) . GH3 . 3 controls JA homeostasis in seedlings , and gh3 . 2 mutants showed increased resistance to B . cinerea ( González-Lamothe et al . , 2012; Gutierrez et al . , 2012 ) . Thus , we generated wrky33 gh3 . 2 gh3 . 3 triple mutant plants but did not observe restoration of WT-like resistance towards B . cinerea 2100 , indicating that they are not critical for WRKY33-dependent defense against this fungal strain ( Figure 8—figure supplement 3B ) . Interestingly , our global binding studies also revealed that WRKY33 binds to the promoter region of NCED3 and to the 3′UTR region of NCED5 ( Figure 4A , D ) , two major genes encoding 9-cis-epoxycarotenoid dioxygenase , a key enzyme in the biosynthesis of ABA ( Leng et al . , 2014 ) . The precise role of ABA in host defense remains enigmatic and ABA can positively or negatively impact the outcome of plant–microbe interactions , depending on the pathogens' lifestyle ( Robert-Seilaniantz et al . , 2011 ) . WRKY33 binding to both gene loci was confirmed by ChIP-qPCR ( Figure 4C , F ) . Transcript levels of NCED3 and NCED5 both increased in the wkry33 mutant upon B . cinerea infection suggesting direct negative regulation by WRKY33 ( Figure 4B , E ) . WRKY33 also bound to the CYP707A3 promoter , a gene involved in ABA catabolism ( Leng et al . , 2014 ) , but its expression decreased in the wrky33 mutant and increased in WT plants after infection indicative of positive regulation by WRKY33 ( Figure 4G–I ) . 10 . 7554/eLife . 07295 . 011Figure 4 . WRKY33 directly regulates target genes encoding ABA biosynthetic ( NCED3 , NCED5 ) and metabolic ( CYP707A3 ) enzymes by binding to their promoters or 3′UTR after B . cinerea 2100 treatment . ( A , D , G ) IGV visualization of ChIP-seq data revealing infection-dependent WRKY33 enrichment at the Arabidopsis NCED3 ( A ) , NCED5 ( D ) and CYP707A3 ( G ) loci with features described in Figure 3 . ( B , E , H ) qRT-PCR analysis of B . cinerea 2100-induced expression of NCED3 ( B ) NCED5 ( E ) and CYP707A3 ( H ) in WT and wrky33 as described in Figure 3 . ( C , F , I ) Validation of the ChIP-seq data for WRKY33 binding to the promoters of NCED3 ( C , with primer pairs p1 and p2 ) and CYP707A3 ( I ) , and to the 3′UTR of NCED5 ( F ) performed as described the legend to Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 011 These results suggest that WRKY33 represses ABA levels during B . cinerea 2100 infection , and that this repressor function is an important component in host resistance to this pathogen . To clarify the involvement of ABA in WRKY33-mediated host defense to B . cinerea 2100 , we analyzed ABA mutants with respect to their phenotypes after fungal infection . Previous reports have shown that aba2-12 ( Adie et al . , 2007 ) , aba3-1 ( Léon-Kloosterziel et al . , 1996 ) , and nced3 nced5 ( Frey et al . , 2012 ) accumulated much less ABA than WT plants . Indeed , the aba2-12 , aba3-1 , nced3-2 , nced5-2 , and nced3 nced5 mutants were nearly as resistant as WT plants to B . cinerea 2100 ( Figure 5B; Figure 5—figure supplement 1 ) . To test whether WT resistance towards this necrotroph is due to WRKY33-mediated repression of NCED3 and NCED5 expression we generated wrky33 nced3 , wrky33 nced5 double , and wrky33 nced3 nced5 triple mutants , and tested their infection phenotypes . 10 . 7554/eLife . 07295 . 012Figure 5 . WRKY33 controls ABA-mediated plant susceptibility to B . cinerea 2100 . ( A ) Growth phenotypes of WT , wrky33 , nced3 nced5 , and wrky33 nced3 nced5 Arabidopsis plants at 4 weeks under short day conditions . ( B ) B . cinerea infection phenotypes 3 days post inoculation of WT , wrky33 , nced3 , nced5 , nced3 nced5 , and wrky33 nced3 nced5 . ( C ) B . cinerea biomass quantification on indicated Arabidopsis genotypes . For fungal biomass determination , the relative abundance of B . cinerea and Arabidopsis DNA was determined by qPCR employing specific primers for BcCutinase A and AtSKII , respectively . ( D ) Exogenous application of ABA ( 10 μM ) directly to infection droplets on wrky33 nced3 nced5 leaves partially rendered plants susceptible to B . cinerea 2100 . Upon completion of the infection experiments ( 3 dpi ) , leaves were detached and photographed . For the infections , one or two 2 μl droplets containing 2 . 5 × 105 spores were applied to each leaf . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 01210 . 7554/eLife . 07295 . 013Figure 5—figure supplement 1 . B . cinerea 2100 infection phenotypes of mature 4-week old leaves derived from aba2-12 and aba3-1 mutant plants . Pictures were taken 3 days post inoculation of two 2 μl droplets containing 2 . 5 × 105 spores to each leaf ( leaves were detached after completion of the experiment for photographic purposes only ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 01310 . 7554/eLife . 07295 . 014Figure 5—figure supplement 2 . B . cinerea 2100 infection phenotypes of mature 4-week old leaves derived from wrky33 nced3 and wrky33 nced5 mutant plants . Pictures were taken 3 days post inoculation of two 2 μl droplets containing 2 . 5 × 105 spores to each leaf ( leaves were detached after completion of the experiment for photographic purposes only ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 01410 . 7554/eLife . 07295 . 015Figure 5—figure supplement 3 . Phenotype of WT Col-0 plants treated with ABA . Infection droplets with ABA ( 10 μM; red arrows ) or without ABA ( black arrows ) were applied to 4-week old leaves . No alteration of the WT resistant phenotype was observed . Picture was taken 3 days post inoculation of two 2 μl droplets containing 2 . 5 × 105 spores to each leaf . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 01510 . 7554/eLife . 07295 . 016Figure 5—figure supplement 4 . B . cinerea 2100 infection phenotypes of mature 4-week old leaves derived from cyp707a1 , cyp707a2 and cyp707a3 mutant plants . Pictures were taken 3 days post inoculation of two 2 μl droplets containing 2 . 5 × 105 spores to each leaf ( leaves were detached after completion of the experiment for photographic purposes only ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 016 ABA deficiency severely affects plant growth leading to stunted phenotypes , as observed in nced3 nced5 , aba2-12 and aba3-1 mutants . A strong reduction of rosette diameter was also observed in the wrky33 nced3 nced5 mutant under short day conditions ( Figure 5A ) . However , unlike the wrky33 mutant , the wrky33 nced3 nced5 triple mutant showed clear resistance to B . cinerea 2100 similar to WT plants ( Figure 5B ) . In contrast , the wrky33 nced3 and wrky33 nced5 double mutants were as susceptible as wrky33 ( Figure 5—figure supplement 2 ) . Consistent with the observed resistance phenotype , qPCR analysis revealed strongly reduced fungal biomass in wrky33 nced3 nced5 compared to wrky33 plants at 3 days post infection ( Figure 5C ) . This clearly indicates that increased expression levels of NCED3 and NCED5 in the wrky33 mutant contribute to susceptibility toward B . cinerea 2100 , and that a key function of WRKY33 in host immunity towards this pathogen is to repress ABA biosynthesis . Since nced3 nced5 mutants have reduced ABA levels , we tested whether exogenous application of ABA to the mutants could revert the resistant phenotype . Indeed , application of ABA together with the fungal spore droplet to leaves of the wrky33 nced3 nced5 triple mutant partially rendered plants susceptible to B . cinerea 2100 ( Figure 5D ) . Similar tests on WT plants did not alter host resistance ( Figure 5—figure supplement 3 ) . CYP707A mutants affecting ABA metabolism were reported to accumulate more ABA than lines overexpressing ABA biosynthetic genes ( Finkelstein , 2013 ) . We therefore also tested the phenotypes of cyp707a1 , cyp707a2 , and cyp707a3 following infection . Interestingly , all of these mutants remained as resistant as WT plants towards the fungus ( Figure 5—figure supplement 4 ) . Taken together , our genetic analysis combined with our ChIP-seq and expression results strongly suggests that WRKY33-mediated control of NCED3 and NCED5 expression plays a critical role in host resistance towards B . cinerea 2100 . Hormonal signaling appears to be affected in the susceptible wrky33 mutant compared with the resistant WT ( Figure 2E ) , and wrky33 nced3 nced5 triple mutants restore WT-like resistance . Thus , we hypothesized that WRKY33 plays a critical regulatory role in hormone homeostasis . To test this , we measured hormonal levels in WT , wrky33 , nced3 nced5 , and wrky33 nced3 nced5 plants following infection with B . cinerea 2100 at various time points . It is important to note that we have previously shown that up to 40 to 48 hpi no differences in fungal biomass , hyphal expansion , or other phenotypic criteria were observed between resistant WT and susceptible wrky33 plants ( Birkenbihl et al . , 2012 ) . As expected , ABA and SA levels increased strongly in susceptible wrky33 compared to resistant WT plants during fungal infection ( Figure 6A , B ) . However , JA and ACC ( precursor of ET ) levels also increased strongly in wrky33 compared to WT plants ( Figure 6C , D ) . Interestingly , the elevated SA levels observed in wkry33 appear to be a direct consequence of increased ABA levels as SA levels were clearly reduced in the resistant ABA-deficient wrky33 nced3 nced5 compared to wrky33 plants post infection . Moreover , the levels of JA and ACC were also reduced in wrky33 nced3 nced5 at later infection stages . This suggests that ABA signaling exerts a positive role on the biosynthesis of these other hormonal components . 10 . 7554/eLife . 07295 . 017Figure 6 . Hormone levels in different genotypes during B . cinerea 2100 infection . Concentrations of the hormones ABA ( A ) , SA ( B ) , ACC ( C ) , and JA ( D ) were measured at 8 , 14 , 24 , and 40 hpi in leaves of indicated Arabidopsis genotypes spray inoculated with spores of B . cinerea 2100 . Mock treated plants ( mo , 14 hr ) served as a control . The data show the average values and SDs of the combined data from three independent experiments with up to four replicates each . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 017 Taken together , our data indicate that a key function of WRKY33 in B . cinerea strain 2100 challenged WT plants is to limit ABA levels . Loss of WRKY33 function affects hormonal homeostasis in the plant during infection , leading to elevated ABA activity and subsequently resulting in altered hormone signaling . Over 75% of WRKY33-regulated target genes showed elevated expression in the susceptible wrky33 mutant after B . cinerea 2100 infection ( Figure 2D ) . To test these genes for altered expression in resistant wrky33 nced3 nced5 plants , we performed qRT-PCR analyses . The transcript levels of several highly expressed SA-related genes observed in wrky33 at 24 hpi decreased in the wrky33 nced3 nced5 plants , often returning to WT states . These included: ICS1 , NPR1 , NPR3 , NPR4 , TRX-h5 , and FMO1 ( Figure 7 ) . However , not all SA-related genes were similarly affected as illustrated for EDS1 , PAD4 , NIMIN1 , PR1 , and PR2 , whose expression levels remained significantly higher than in WT ( Figure 7 ) . These results imply that simultaneous mutations of NCED3 and NCED5 in the wrky33 genotype partially impair SA biosynthesis and signaling . 10 . 7554/eLife . 07295 . 018Figure 7 . Expression of numerous genes up-regulated in infected wrky33 plants showing WT-like levels in the wrky33 nced3 nced5 triple mutant . Heatmap showing expression levels of genes , differentially expressed in RNA-seq and analyzed by qRT-PCR in WT , wrky33 ( w33 ) , nced3 nced5 ( n3n5 ) , and wrky33 nced3 nced5 ( w33n3n5 ) after mock treatment or 24 hr post B . cinerea ( B . c . 2100 ) infection . Genes showing high expression levels in wrky33 but reduced levels in the triple mutant are boxed . All values were normalized to the expression of At4G26410 . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 018 Reduced transcript levels were also observed for other genes in the wrky33 nced3 nced5 mutant at 24 hpi such as the TF genes ERF1 , NAC019 , NAC055 , NAC061 , NAC090 , WRKY41 , WRKY48 , WRKY53 , and WRKY55 , indicating a positive effect of ABA on their expression ( Figure 7 ) . In contrast , expression of WRKY38 and WRKY50 increased in wrky33 nced3 nced5 plants to even higher levels than observed in wrky33 ( Figure 7 ) suggesting a negative effect of ABA on these genes . The Botrytis-induced expression of other ABA response genes including ACS2 , BIR1 , CDPK1 , MPK11 , and CRK36 was also restored to WT levels in the wrky33 nced3 nced5 mutant ( Figure 7 ) . Interestingly , CDPK1 , MPK11 , CRK36 , and NPR3 are also responsive to SA and these genes are associated with the GO term ‘cell death’ , suggesting that ABA has a positive effect on cell death responses ( Figure 2—figure supplement 2A , Supplementary files 3 , 4 ) . In summary , WRKY33 suppresses the expression of many of its target genes by negatively regulating ABA responses . The subset of genes showing restoration of WT-like expression levels in the resistant triple mutant constitutes prime candidates whose functions may be causal for WRKY33-mediated resistance against this necrotrophic fungus . The role of ABA in biotic stress responses is complex and currently ill-defined . The ability of Arabidopsis to restrict penetration by the non-host barley pathogen Blumeria graminis was shown to be dependent on the NAC TF ATAF1-mediated repression of ABA biosynthesis ( Jensen et al . , 2008 ) . In contrast , overexpression of ATAF1 resulted in enhanced susceptibility of Arabidopsis plants to B . cinerea ( Wang et al . , 2009 ) . ATAF1 was shown to directly bind to the NCED3 promoter , which positively correlated with increased NCED3 expression and ABA levels ( Jensen et al . , 2013 ) . Moreover , transcriptomic studies using 4-week old detached Arabidopsis leaves infected with B . cinerea strain pepper revealed that genes involved in the suppression of ABA accumulation and signaling were up-regulated at early infection stages ( Windram et al . , 2012 ) . Our study clearly demonstrates that increased expression of WRKY33 target genes associated with ABA biosynthesis ( NCED3 and NCED5 ) is causal for the susceptibility of wrky33 to B . cinerea 2100 , and the ABA deficient wrky33 nced3 nced5 mutant restored WT-like resistance towards this necrotroph . Hence , our findings reveal a novel role of WRKY33 in modulating host resistance to B . cinerea by suppressing ABA accumulation/signaling ( Figure 8—figure supplement 1 ) . Interestingly , resistance to the necrotrophic fungus Plectosphaerella cucumerina is also negatively impacted by ABA , and wrky33-1 mutant plants exhibited an enhanced susceptible phenotype towards this pathogen ( Sánchez-Vallet et al . , 2012 ) . Stimulating NCED3 expression and ABA biosynthesis has also been described as an important virulence strategy employed by the hemi-biotroph P . syringae DC3000 in Arabidopsis ( de Torres-Zabala et al . , 2007 ) . Virulence to this pathogen is strongly reduced in ABA mutants . Whether WRKY33 is involved in modulating ABA signaling during this host–bacterial interaction , and whether P . syringae DC3000 suppresses WRKY33 expression is unknown . In our experiments , the expression of several NAC TF genes associated with ABA regulation was affected upon Botrytis infection in a WRKY33-dependent manner . In particular , increased expression of NAC002 ( ATAF1 ) , NAC019 , NAC055 , NAC061 , NAC068 ( NTM1 ) , and NAC090 was observed in the wrky33 mutant . Like ATAF1 , transgenic Arabidopsis lines overexpressing NAC019 or NAC055 displayed enhanced susceptibility to B . cinerea ( Bu et al . , 2008 ) . In contrast , the nac019 nac055 double mutant showed increased resistance to B . cinerea compared with WT plants . ABA has been shown to induce NAC019 and NAC055 expression ( Jiang et al . , 2009; Zheng et al . , 2012 ) . Whether any of these NAC factors , apart from ATAF1 , can also target the NCED genes and thereby enhance ABA biosynthesis in the wrky33 mutant remains to be tested . ABA can repress SA- , ET- , and JA/ET-dependent signaling but also positively affect some JA responses ( Asselbergh et al . , 2008; Ton et al . , 2009 ) . Our genetic and phytohormone studies showed that elevated ABA levels in the susceptible wrky33 mutant resulted in concomitant increases in SA , JA , and ACC levels upon B . cinerea 2100 infection , implying a positive effect of ABA on these hormone signaling components . Increased SA levels per se in the wrky33 mutant however do not contribute to susceptibility as wrky33 sid2 double mutant plants are as susceptible to B . cinerea 2100 as the single mutant ( Birkenbihl et al . , 2012; Figure 8—figure supplement 1 ) . Interestingly , concurrent increases in ABA , JA/ET and SA have also been observed in the interaction of Arabidopsis with P . syringae DC3000 and with the vascular oomycete pathogen Pythium irregular ( Adie et al . , 2007; de Torres-Zabala et al . , 2007 ) . In the case of P . irregular , however , host resistance correlated with high ABA levels , whereas ABA mutants were clearly more susceptible . Our molecular analysis of Botrytis-challenged wrky33 and wrky33 nced3 nced5 plants confirmed that elevated ABA levels mainly activate NPR1-dependent SA signaling while not affecting the upstream EDS1-PAD4 pathway ( Figure 7 ) . Increased ABA also activated ACS2 , ACS6 , ERF1 , and ORA59 , targets involved in ET/JA signaling . In the wrky33 nced3 nced5 mutant as in WT plants expression of ACS6 , ORA59 , but also ERF5 and PDF1 . 2 ( data not shown ) was significantly reduced . However , as both genotypes are resistant to B . cinerea these ET/JA components appear not to be essential in maintaining plant resistance . How elevated ABA levels trigger activation of these hormone signaling cascades and specific TFs remains to be elucidated . The receptors for ABA are known ( Miyakawa et al . , 2013 ) , but the molecular mechanisms linking downstream ABA signaling to the other hormonal pathways require further investigation . It is conceivable that the increased ABA levels in wrky33 during Botrytis infection trigger the activation of currently unidentified downstream ABA-response factors that bind to the ABA response elements ( ABRE , ACGTGG/T ) or G-boxes ( CACGTG ) present in some gene promoters , resulting in transcriptional activation . Indeed , several genes including NAC019 , NAC061 , FMO1 , GH3 . 2 , and GH3 . 3 contain such conserved motifs , which could respond to and be activated by ABA . In addition , the elevated levels of SA in wrky33 during Botrytis infection may in part be responsible for the strong expression of the glutaredoxin gene GRX480/ROXY19 observed in this mutant ( RNA-seq data in this study; Birkenbihl et al . , 2012 ) . GRX480 binds to class II TGA factors , thereby activating TGA-regulated SA responses while preventing their participation in JA-mediated signaling ( Zander et al . , 2014 ) . Plants ectopically overexpressing GRX480 are susceptible to B . cinerea 2100 ( Birkenbihl et al . , 2012 ) . In animals and humans , several TFs can act either as transcriptional activators or repressors , depending on DNA-binding sequences or interaction with additional co-factors ( Alexandre and Vincent , 2003; Berger and Dubreucq , 2012; Sakabe et al . , 2012; Zhu et al . , 2012 ) . Moreover , many human TFs function as repressors as often as they act as activators ( Cusanovich et al . , 2014 ) . In plants , few factors with dual functions have been unequivocally characterized . In tomato , the transcriptional activator Pti4 can repress the expression of PR10-a by forming a complex with the SEBP repressor ( Gonzalez-Lamothe et al . , 2008 ) . In Arabidopsis , the TF WUSCHEL acts mainly as a repressor in stem cell regulation , but can function as an activator of AGAMOUS ( AG ) during floral patterning ( Ikeda et al . , 2009 ) . Also WRKY proteins can act as activators or repressors , and selected family members in diverse plant species have been identified as key regulators in diverse plant processes ( Rushton et al . , 2010 ) . WRKY53 can activate or repress the expression of genes , depending on the nature of the target promoter sequence ( Miao et al . , 2004 ) . WRKY6 activates PR1 expression while suppressing the expression of its own gene , and that of its closely related family member WRKY42 ( Robatzek and Somssich , 2002 ) . Our data show that also WRKY33 is a bi-functional TF that can act as an activator or as a repressor in a promoter-context dependent manner ( Figure 8 ) . WRKY33 positively regulates genes involved in camalexin biosynthesis such as CYP71A13 and PAD3 by directly binding to their promoter regions ( Mao et al . , 2011; Birkenbihl et al . , 2012 ) . Our study confirmed these observations and identified two additional camalexin biosynthetic genes , AMT1 and CYP71A12 that are positively and directly regulated by WRKY33 ( Figure 8—figure supplement 2; Supplementary file 2 ) . Mutants of CYP71A13 and PAD3 are susceptible to the necrotrophs A . brassicicola and B . cinerea ( Zhou et al . , 1999; Nafisi et al . , 2007 ) . Beyond this , several Botrytis-induced ET response genes were targeted and positively regulated by WRKY33 ( Figure 2E ) . Still , WRKY33 had a negative regulatory relationship on the expression of >75% of all targets , implying that it mainly acts as a direct repressor of many defense genes following pathogen challenge . 10 . 7554/eLife . 07295 . 019Figure 8 . Dual regulatory role of WRKY33 in modulating host defenses to B . cinerea 2100 . WRKY33 positively regulates target genes involved in camalexin biosynthesis thereby contributing to host resistance towards B . cinerea 2100 . Target genes involved in ET/JA biosynthesis and signaling can either be positively or negatively regulated by WRKY33 . On the other hand , WRKY33 negatively regulates ABA levels by directly targeting and repressing NCED3 and NCED5 expression , or inducing expression of CYP707A3 , a gene involved in ABA metabolism . Thus , WRKY33 has both activator and repressor functions that may depend on promoter context . Red arrows indicate positive regulation , whereas black bars indicate negative regulation . The curved red arrow highlights positive feedback regulation of WRKY33 on its own gene promoter . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 01910 . 7554/eLife . 07295 . 020Figure 8—figure supplement 1 . Schematic representation showing WRKY33-dependent host immunity towards the necrotroph B . cinerea 2100 through repressing the ABA network . In WT Arabidopsis Col-0 plants ( upper panel ) , B . cinerea inoculation induces WRKY33 expression and protein accumulation that controls ABA levels by repressing the ABA biosynthetic target genes NCED3 and NCED5 , and by inducing the ABA metabolic gene CYP707A3 . WRKY33 also directly or indirectly represses expression of other genes in different hormone signaling pathways as well as genes encoding WRKY , and NAC TFs . WRKY33 also positively affects resistance towards B . cinerea 2100 by directly targeting and inducing the expression of several genes involved in camalexin biosynthesis . In the wrky33 mutant ( lower panel ) , B . cinerea infection fails to activate host camalexin biosynthesis genes and increases ABA levels by up-regulation of NCED3 and NCED5 , which activates downstream ABA-dependent genes and ABA signaling . Among the ABA-dependent genes are genes involved in SA signaling , WRKY TFs , NAC TFs , and genes associated with auxin or JA conjugation indicating that ABA may act as a key sub-node in WRKY33-dependent host immunity to B . cinerea . Whereas WT-like resistance is restored in the wrky33 nced3 nced5 triple mutant , all currently tested mutants of WRKY33-regulated genes acting downstream of ABA failed to restore resistance in the wrky33 mutant background ( marked by an asterisk ) . Thus , additional crucial genes causal for WRKY33-dependent resistance to B . cinerea 2100 remain to be discovered , as do ABA signaling components affecting WRKY33 expression ( indicated by the question marks ) . The solid lines with arrows indicate induction or positive modulation , the bar heads indicate suppression . Direct WRKY33 targets are indicated by the green color . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 02010 . 7554/eLife . 07295 . 021Figure 8—figure supplement 2 . AMT1 and CYP71A12 expression is positively regulated by WRKY33 . ( A–B ) IGV images of ChIP-seq data revealing high infection-dependent WRKY33 binding at the promoters of AMT1 ( A ) and CYP71A12 ( B ) . ( C–D ) qRT-PCR analyses of AMT1 ( C ) and CYP71A12 ( D ) in WT and wrky33 at the indicated time points after B . cinerea 2100 spore inoculation . qRT-PCR data were normalized to the expression of At4g26410 . Error bars represent SD of three biological replicates ( n = 3 ) . Asterisks indicate significant differences between WT and wrky33 ( * , p < 0 . 05; ** , p < 0 . 001; two-tailed t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 02110 . 7554/eLife . 07295 . 022Figure 8—figure supplement 3 . The role of GH3 genes in WRKY33-mediated host resistance to B . cinerea 2100 . ( A ) Expression levels of GH3 . 2 and GH3 . 3 in WT and wrky33 plants determined by qRT-PCR at the indicated time points after B . cinerea 2100 spore inoculation . Data were normalized to the expression of At4g26410 . Error bars represent SD of three biological replicates ( n = 3 ) . Asterisks indicate significant differences between WT and wrky33 ( * , p < 0 . 05; ** , p < 0 . 001; two-tailed t-test ) . ( B ) B . cinerea 2100 infection phenotypes of WT , wrky33 , gh3 . 2 , gh3 . 3 , gh3 . 2 gh3 . 3 , and wrky33 gh3 . 2 gh3 . 3 plants are shown at 3 days post infection . Two 2 μl droplets containing 2 . 5 × 105 spores were applied to each leaf . DOI: http://dx . doi . org/10 . 7554/eLife . 07295 . 022 How WRKY33 exerts its dual regulatory functions mechanistically requires further research . The simplest hypothesis would be that WRKY33 is recruited to distinct repressor and activator complexes at defined promoter sites . For instance , several WRKY33-interacting proteins containing a VQ motif have been discovered that influence defense gene expression ( Lai et al . , 2011; Cheng et al . , 2012; Pecher et al . , 2014 ) . VQ proteins appear to act as suppressors of defense-related genes via their interaction with WRKY factors . Indeed , the protein VQ4/MVQ1 appears to function as a negative regulator of WRKY-type transcriptional activators including WRKY33 . In a transient Arabidopsis protoplast assay , stimulation by MAMPs resulted in degradation of VQ4/MVQ1 following MAPK-mediated phosphorylation enabling WRKY33 to activate transcription of a defense-related reporter gene ( Pecher et al . , 2014 ) . Two other proteins , SIB1 and SIB2 , have been reported to interact with WRKY33 via their VQ motifs . This interaction was required to stimulate WRKY33 DNA-binding activity and very likely to positively regulate WRKY33-mediated resistance to necrotrophic fungi ( Lai et al . , 2011 ) . Interestingly , four VQ protein genes ( VQ8 , VQ22/JAV1 , VQ28 , VQ33 ) are also direct targets of WRKY33 and show altered expression upon Botrytis infection in wrky33 compared to WT . VQ22/JAV1 functions as a negative regulator of JA-mediated defenses , and transgenic VQ22/JAV1 RNAi lines showed enhanced resistance to B . cinerea ( Hu et al . , 2013 ) . Thus , elevated levels of these VQ proteins in wrky33 may contribute to the suppression of JA signaling and susceptibility to this fungus . Finally , many members of the WRKY TF gene family , including WRKY33 itself , are also targets of WRKY33 . The majority of WRKY genes are transcriptionally activated during immune signaling ( Pandey and Somssich , 2009 ) . Experimental and bioinformatics analyses have revealed that WRKY factors form a complex and highly interconnected regulatory sub-network that is positively and negatively affected by auto- and cross-regulation by various WRKY factors . This WRKY web appears to be deeply interconnected to various hormonal pathways at multiple levels , probably to ensure rapid and efficient signal amplification while allowing for tighter control in limiting the extent of the host immune response ( Eulgem , 2006; Llorca et al . , 2014 ) . In summary , genome-wide binding analysis and transcriptional profiling have identified potential targets of WRKY33 , a key transcription factor involved in mediating resistance towards the necrotroph B . cinerea 2100 . This study revealed that genes involved in ABA biosynthesis and directly regulated by WRKY33 act at crucial nodes in this signaling cascade . Due to the complexity of the highly interconnected hormonal signaling networks targeted by WRKY33 , the precise molecular mechanisms underlying this resistance remain to be fully elucidated . In this respect , global transcriptional profiling of infected wrky33 nced3 nced5 plants should prove extremely valuable to further narrow down key genes and sub-signaling pathways required to re-establish WT-like levels of resistance in this mutant . For all experiments , A . thaliana ecotype Columbia ( Col-0 ) was used . Besides WT , the following genotypes were employed: wrky33 ( GABI_324B11 ) , nced3-2 , nced5-2 , nced3 nced5 , aba2-12 , aba3-1 , cyp707a1-1 , cyp707a2-1 , cyp707a3-1 , wrky33 sid2-1 , wrky33 npr1-1 , wrky33 wrky70 , gh3 . 2 , gh3 . 3-1 . The double or triple mutants; wrky33 nced3 , wrky33 nced5 , wrky33 nced3 nced5 , gh3 . 2 gh3 . 3-1 , and wrky33 gh3 . 2 gh3 . 3 were generated by crossing single or double mutants followed by PCR-based verification using appropriate primers ( Supplementary file 6 ) . Plants were grown for 4 weeks under short-day conditions in closed cabinets ( Schneijder chambers: 16 hr light/ 8 hr dark cycle at 22–24°C , 60% relative humidity ) on 42 mm Jiffy-7 pots ( Jiffy ) to prevent contaminations from garden soil . Before sewing , the Jiffy pot peat pellets were re-hydrated in water containing 0 . 1% liquid fertilizer Wuxal ( Manna ) . B . cinerea strain 2100 was cultivated on potato dextrose plates at 22°C for 10 days . Spores were collected , washed , and frozen at −80°C in 0 . 8% NaCl at a concentration of 107 spores ml−1 . For inoculation of Arabidopsis plants , the spores were diluted in Vogel buffer prepared as previously described ( Birkenbihl et al . , 2012 ) . For droplet inoculations , 2 μl of 2 . 5 × 105 spores ml−1 were applied to single leaves of 4-week old intact plants . Leaves were excised from plants only for photographic documentation . The same spore concentration was used for spray inoculations of 4-week old intact plants . For mock treatment , Vogel buffer was used . Plants were kept prior to and during infection under sealed hoods at high humidity . 4-week old WT plants or plants expressing WRKY33-HA from the native WRKY33 promoter ( PWRKY33:WRKY33-HA ) were spray inoculated or mock treated for 14 hr . ChIP assays were performed as previously described ( Birkenbihl et al . , 2012 ) following the modified protocol by Gendrel et al . ( 2005 ) , using rabbit polyclonal antibodies to HA ( Sigma-Aldrich , St Louis , MO ) . ChIP DNA was purified using a QIA quick PCR Purification kit ( Qiagen , Germany ) and subjected to a linear DNA amplification ( LinDA ) protocol ( Shankaranarayanan et al . , 2011 ) which included two rounds of ‘in vitro transcription’ by T7 RNA polymerase . The resulting LinDA DNA was used to generate sequencing libraries bearing barcodes using a NEBNext ChIP-seq Library Pre Reagent Set for Illumina kit ( New England Biolabs , Ipswich , MA ) . Sequencing was performed on Illumina HiSeq2500 at the Max Planck Genome Centre Cologne and resulted in about 10 million 100 bp single-end reads per sample . ChIP-qPCR validation of WRKY33 target genes was performed using gene specific primers ( Supplementary file 7 ) . Before mapping , remaining LinDA adapters and low quality sequences were removed from the sequencing data using a two-step procedure . In this procedure , first Bpm and t7-Bpm sites were trimmed from the 5′ end using cutadapt ( version 1 . 2 . 1 ) ( Martin , 2011 ) with options–e 0 . 2 , -n 2 and–m 36 ( otherwise default settings were used ) , and subsequently poly-A and poly-T tails and low quality ends were trimmed and reads with overall low quality or with less than 36 bases remaining after trimming were removed using PRINSEQ lite ( version 0 . 20 . 2 ) ( Schmieder and Edwards , 2011 ) with options–trim_qual_right/left 20 , trim_tail_right/left 3 –min_len 36 , -min_qual_mean 25 . After this pre-processing steps , the remaining high quality reads were mapped to the A . thaliana reference genome TAIR10 ( http://www . arabidopsis . org ) using Bowtie ( version 0 . 12 . 7 ) ( Langmead et al . , 2009 ) with options–best–m 1 to extract only uniquely mapped reads and allowing two mismatches in the first mapping steps ( default settings ) . The ChIP-seq data sets used in this study have been deposited at the GEO repository ( GSE66289 ) . To identify genomic DNA regions enriched in sequencing reads in the ChIP sample compared to input control as well as in inoculated compared to mock treated samples ( ‘peak regions’ ) , the peak calling algorithm of the QuEST program ( version 2 . 4 ) ( Valouev et al . , 2008 ) was applied using the TF mode ( option ‘2’ ) , with permissive parameter settings for the peak calling ( option ‘3’ ) . Each of the two biological replicates was first analyzed separately and additionally , to obtain more exact peak locations for the consistent peaks , the mapped reads of the two replicates were pooled and peaks were also called for the pooled samples . To annotate the peak location with respect to annotated gene features in TAIR10 the annotatePeaks . pl function from the Homer suite ( Heinz et al . , 2010 ) was used with default settings . To extract consistent peaks between the replicates , a custom R ( http://www . r-project . org ) function ( Source code 1 ) was used that identified overlapping peak regions between the replicates . Two peak regions were counted as overlapping , if they overlapped by at least 50% of the smaller region and a peak region was counted as consistent , if it was found to be overlapping between the two individual replicates as well as the pooled sample . To search for conserved binding motifs in the consistent WRKY33 binding regions , for each consistent peak the 500 bp sequence surrounding the peak maximum was extracted and submitted to the online version of MEME-ChIP ( Machanick and Bailey , 2011 ) . MEME-ChIP was run with default settings , but a custom background model derived from the Arabidopsis genome was provided and ‘Any number of repetitions’ of a motif was allowed . For visualization , prominent motifs identified within MEME-ChIP by either MEME or DREME were chosen . To extract the number/percentage of peak regions that contain a certain motif , the online version of FIMO ( Grant et al . , 2011 ) was run with the peak sequences and the motif of interest ( MEME/DREME output ) as input and a p-value threshold of 0 . 001 . Total RNA was extracted from mock treated ( 14 hpi ) and B . cinerea infected ( 14 hpi ) 4-week old plants ( WT and wrky33 ) using the RNeasy Plant Mini kit ( Qiagen ) according to the manufacturer's instructions , and mRNA sequencing libraries were constructed with barcodes using the TrueSeq RNA Sample Preparation Kit ( Illumina ) . Three biological replicates were sequenced on Illumina HiSeq2500 by the Max Planck Genome Centre Cologne , resulting in 25–45 million 100 bp single end reads per sample . Total reads were mapped to the Arabidopsis genome ( TAIR10 ) under consideration of exon-intron structures using the splice-aware read aligner TopHat ( version 2 . 0 . 10 ) ( Kim et al . , 2013 ) with settings–a 10 –g 10 and known splice sites provided based on TAIR10 gene annotations . The RNA-seq data sets used in this study have been deposited at the GEO repository ( GSE66290 ) . The mapped RNA-seq reads were transformed into a count per gene using the function coverageBed of the bedTools suite ( Quinlan and Hall , 2010 ) with option–split to consider exon-intron structures . Genes with less than 50 reads in all samples together were discarded , and subsequently the count data of the remaining genes were TMM-normalized and log2-transformed using functions ‘calcNormFactors’ ( R package EdgeR ) ( Robinson et al . , 2010 ) and ‘voom’ ( R package limma ) ( Law et al . , 2014 ) . To analyze differential gene expression between genotypes ( WT , wrky33 ) and treatments ( mock treated , B . cinerea infected ) , we fitted a linear model with the explanatory variable ‘genotype_treatment’ ( i . e . , including both genotype and treatment ) using the function lmFit ( R package limma ) . Subsequently , we performed moderated t-tests over the four contrasts of interest . Two contrasts compare B . cinerea infected vs mock treated samples within each genotype and the other two contrasts compare wrky33 vs WT Col-0 plants within each treatment . In all cases , the resulting p values were adjusted for false discoveries due to multiple hypothesis testing via the Benjamini–Hochberg procedure . For each contrast , we extracted a set of significantly differentially expressed genes between the tested conditions ( adj . p value ≤ 0 . 05 , \|log2FCΙ ≥ 1 ) . GO term enrichment analysis on the gene sets of interest was performed using the R package goseq ( Young et al . , 2010 ) with custom GO term mappings obtained from org . At . tairGO2ALLTAIRS within the R package org . At . tair . db ( Carlson , 2010 ) . To identify enriched GO terms , the Wallenius distribution was used to approximate the null distribution and a probability weighting function was applied to correct for potential count biases in the analyzed gene sets . The resulting p values were adjusted for false discoveries due to multiple hypothesis testing via the Benjamini-Hochberg procedure and for each subset the significantly over-represented GO terms were extracted ( adj . p value < 0 . 05 ) . For the set of all WRKY33-regulated targets , the R package topGO ( Alexa and Rahnenführer , 2010 ) was used to visualize the GO sub-graphs induced by the 10 most significantly enriched GO terms in the category ‘Biological Process’ and the five most significantly enriched GO terms in the category ‘Molecular Function’ , respectively . Total RNA was isolated from leaves at 8 , 14 , 24 , and 48 hpi with B . cinerea spores as described above and reverse transcribed with oligo ( dT ) primer to produce cDNA using the SuperScript First-Strand System for Reverse-Transcription PCR following the manufacturer's protocol ( Invitrogen , Grand Island , NY ) . cDNAs corresponding to 2 . 5 ng of total RNA were subjected to qPCR with gene-specific primers ( Supplementary file 8 ) using the SYBR Green reagent ( Bio-Rad , Hercules , CA ) . The qPCRs were performed on the iQ5 Multicolor Real-Time PCR Detection System ( Bio-Rad ) with three biological replicates . The relative expression was normalized to At4g26410 that was described as being highly constant under varying stress conditions ( Czechowski et al . , 2005 ) . Data shown are means ± SD from the three biological replicates . Sample processing , data acquisition , instrumental setup , and calculations were performed as described ( Ziegler et al . , 2014 ) . Instrument specific parameters for the detection of SA are shown in the Table below . ( 3 , 4 , 5 , 6-D4 ) -SA was obtained from Campro Scientific ( Veenendal , The Netherlands ) and used as internal standard for SA quantification ( 1 . 5 ng per sample ) . MS parameters for MRM-transition of salicylic acids ( SA ) . HormoneMRM transitionsDeclustering potential ( DP ) , VEntrance potential ( EP ) , VCell entrance potential ( CEP ) , VCollision potential ( CE ) , VCell exit potential ( CEX ) , VSA137 → 93−25−5 . 5−14−220137 → 65−25−5 . 5−14−440SA-D4141 → 97−25−5 . 5−14−220Quantifier and qualifier transitions are indicated in bold and italics , respectively . Quantification of fungal biomass relative to plant biomass by qPCR was basically performed as previously described ( Gachon and Saindrenan , 2004 ) . Leaves of the indicated Arabidopsis lines were inoculated with two 2 μl droplets of B . cinerea spores and DNA extracted 3 days later from whole leaves of similar fresh weight . The relative amounts of B . cinerea and Arabidopsis DNA were determined by qPCR employing specific primers for cutinase A and SKII , respectively . Full-length WRKY33 protein fused with NusA and an 8xHis-tag was expressed in the vector pMCSG48 in the Escherichia coli strain BL21 ( DE3 ) Magic ( kindly provided by Dr Michal Sikorski and Marta Grzechowiak , Institute of Bioorganic Chemistry , Poznan , Poland ) . Bacteria containing the WRKY33 expression construct or the empty vector were induced with 0 . 5 mM isopropylthio-β-galactisude for 3 hr at 18°C . The His-tagged protein was purified using nickel affinity columns ( QIAexpress Ni_NTA Fast Start , Qiagen , Germany ) following the instructions of the manufacturer , and subsequently dialyzed against 20 mM Tris–HCl , pH 7 , 5 overnight at 4°C . The following DNA oligonucleotide probes were synthesized and biotin labeled by Sigma–Aldrich ( Germany ) : W-box probe , 5′-CGTTGACCGTTGACCGAGTTGACTTTTTA-3′; W-boxmut , 5′-CGTTGAACGTTGAACGAGTTGAATTTTTA-3′; M-3 , 5′-AATTTGAATAATCAAAGATCTTCCTTTGAATTACCTATTCAACAT-3′ ( derived from the PROPEP3 promoter sequence ) ; and M-7 , 5′-GTCCACGCTTGTTTGAATTTTCAGCCTTTGCAGGCAAGGT-3′ ( derived from the WAKL7 promoter sequence ) . Two complementary strands of the oligonucleotides were annealed by heating probes to 95°C for 3 min and then allowing probes to cool to room temperature overnight . Freshly prepared recombinant WRKY33 protein ( 1 μg ) was incubated with the DNA probe ( 20 fmol ) for 20 min at room temperature using the LightShift Chemiluminescent EMSA kit ( Thermo Fisher , Germany ) in the presence or absence of unlabeled competitor DNA . The resulting protein-DNA complexes were electrophoresed on 5% ( wt/vol ) polyacrylamide gels , and then transferred to N+ nylon membranes in 0 . 5% Trisborate/EDTA buffer at 380 mA at 4°C for 60 min . Biotin labeled DNA detection was done according to the instructions of the manufacturer ( Thermo Fisher ) . Bands were visualized using the BioRAD ChemiDoc MP Imaging System .	Crop yields can be badly affected by diseases caused by some fungi and other microbes . One fungus called Botrytis cinerea is able to infect many different species of crop plants—including tomatoes and grapes—and can cause severe damage both before and after harvest . This fungus belongs to a group of microbes that kill the plant cells they invade and then extract the nutrients from the dead cells . Some plants are able to resist infection by B . cinerea and researchers have identified several proteins that are involved in this resistance . One such protein is called WRKY33 , which is able to bind to DNA to regulate the activity of particular genes . However , it was not clear exactly which genes were involved in the response to B . cinerea . Arabidopsis thaliana is a small flowering plant that is often used in research . Mutant A . thaliana plants lacking WRKY33 are very susceptible to infection with B . cinerea . Here , Liu et al . use several genetic techniques to find out which genes WRKY33 regulates when A . thaliana plants are exposed to the fungus . The experiments indicate that WRKY33 can alter the activity of over 300 genes . Some of these genes had previously been shown to be targets of WRKY33 and are involved in cell responses to plant hormones and the production of an antimicrobial molecule called camalexin . Liu et al . also show that two genes called NCED3 and NCED5—which are required for the production of a plant hormone called abscisic acid—are repressed by WRKY33 . Mutant plants lacking WRKY33 had increased levels of abscisic acid and further experiments suggested that the repression of NCED3 and NCED5 by WRKY33 is important to resistance against the fungus . Liu et al . 's findings provide the first detailed view of which genes in A . thaliana are regulated by WRKY33 when the plant is exposed to B . cinerea . A future challenge is to understand how blocking the production of abscisic acid protects plants against B . cinerea and other similar fungi .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2015	Negative regulation of ABA signaling by WRKY33 is critical for Arabidopsis immunity towards Botrytis cinerea 2100
Host–parasite interactions are often embedded within complex host communities and can be influenced by a variety of environmental factors , such as seasonal variations in climate or abiotic conditions in water and soil , which confounds our understanding of the main drivers of many multi-host pathogens . Here , we take advantage of a combination of large environmental data sets on Mycobacterium ulcerans ( MU ) , an environmentally persistent microorganism associated to freshwater ecosystems and present in a large variety of aquatic hosts , to characterize abiotic and biotic factors driving the dynamics of this pathogen in two regions of Cameroon . We find that MU dynamics are largely driven by seasonal climatic factors and certain physico-chemical conditions in stagnant and slow-flowing ecosystems , with an important role of pH as limiting factor . Furthermore , water conditions can modify the effect of abundance and diversity of aquatic organisms on MU dynamics , which suggests a different contribution of two MU transmission routes for aquatic hosts ( trophic vs environmental transmission ) depending on local abiotic factors . Despite increased understanding of infectious disease ecology and dynamics , recent decades have seen an upsurge in the emergence and re-emergence of multiple infectious diseases ( Jones et al . , 2008 ) . Most emerging pathogens are zoonotic and have a very broad range of hosts ( Woolhouse et al . , 2001; Woolhouse and Gowtage-Sequeria , 2005 ) , and as a result , host–parasite interactions are often embedded within complex host communities ( Plowright et al . , 2008; Roche et al . , 2013b ) . Biotic interactions between hosts in the community may play an important role on disease transmission , either promoting or diluting the overall prevalence of the pathogen ( LoGiudice et al . , 2003; Keesing et al . , 2010; Johnson et al . , 2013; Roche et al . , 2013a ) . In addition , the effect of seasonal variations in climate on abiotic conditions in water and soil can influence the composition of host communities and in turn have an impact on the ecological dynamics of the pathogens ( Ostfeld et al . , 2008 ) . Because of the intertwined nature of biotic and abiotic drivers , disentangling their respective contribution to pathogen dynamics and transmission through complex and different-scale processes remains a challenge ( Plowright et al . , 2008 ) . Nonetheless , identifying the underlying ecological mechanisms driving the emergence and persistence of diseases is essential to reduce disease risk in human populations ( Roche and Guégan , 2011 ) . An illustrative example is the case of Mycobacterium ulcerans ( MU ) , an environmentally persistent microorganism associated to freshwater ecosystems in tropical countries and present in a large variety of aquatic hosts ( Benbow et al . , 2008; Marion et al . , 2010; Garchitorena et al . , 2014 ) . From a public health perspective , MU is the agent responsible for Buruli ulcer ( BU ) , a devastating skin disease with great health and socio-economic consequences in tropical and subtropical countries ( WHO , 2008 ) . Emergence , distribution , and risk factors for BU in many parts of the world are associated with stagnant and slow-flowing ecosystems ( Brou et al . , 2008; Wagner et al . , 2008; Jacobsen and Padgett , 2010; Marion et al . , 2011 ) . The environmental factors that favour MU persistence and transmission within these ecosystems are still poorly understood but the environmental and multi-host nature of the pathogen suggests that its environmental dynamics can be the result of a complex interplay between environmental factors and biotic interactions ( Garchitorena et al . , 2014; Morris et al . , 2014 ) . MU is broadly present across taxa in aquatic communities over space and time ( Benbow et al . , 2008; Marion et al . , 2010; Garchitorena et al . , 2014 ) , suggesting that a multiplicity of hosts can play a role in MU persistence in the environment . Biotic interactions between hosts are thought to be a pathogen transmission route between organisms ( Merritt et al . , 2010 ) , with MU being integrated in the aquatic community from the environment thanks to filter feeder , herbivorous , and scavenger organisms and then transmitted across the community through predation ( Marsollier et al . , 2002 , 2007a , 2007b; Mosi et al . , 2008 ) . As a result , community-level factors such as biodiversity or abundance of aquatic organisms could drive the environmental load of MU through amplification or dilution effects , as demonstrated for other pathogens ( Ezenwa et al . , 2006; Suzán et al . , 2009; Keesing et al . , 2010; Johnson et al . , 2013 ) . Furthermore , some keystone species could play an overwhelming role in the transmission and overall MU prevalence in host communities ( Roche et al . , 2013a ) . Some specific water conditions , physical or chemical , could also favour MU environmental persistence and dynamics in aquatic ecosystems . MU seems to grow better under laboratory conditions with low oxygen , high temperature , and mildly acidic pH ( Portaels and Pattyn , 1982; Palomino and Portaels , 1998; Palomino et al . , 1998; Dega et al . , 2000; Marsollier et al . , 2004 ) . Genomic studies show that MU is sensitive to UV light ( Stinear et al . , 2007; Doig et al . , 2012 ) so turbid or protected environments could promote MU persistence . Many of these conditions are generally met in swamps and other stagnant and slow-flowing ecosystems , and therefore , if optimal abiotic conditions are met , MU could grow and persist in these ecosystems as free-living stages in the water and infect aquatic organisms directly , without the need for a trophic transmission to take place . Besides , numerous abiotic factors within aquatic ecosystems can influence host community structures and assemblages ( Eric Benbow et al . , 2013; Garchitorena et al . , 2014 ) , since aquatic invertebrate and vertebrate taxa have different ranges of optimal water conditions ( Dickens and Graham , 2002 ) . This could represent an indirect influence of abiotic conditions on MU transmission within aquatic communities . A direct transmission of MU driven mostly by abiotic factors and a trophic transmission driven by biological interactions are not two mutually exclusive routes but rather could complement each other to allow persistence of MU under a wide range of environments . Identifying the contribution of such transmission routes requires a deep understanding of the dynamics of MU within aquatic communities through space and time . A recent characterization of MU dynamics with unprecedented detail in two BU endemic areas with very distinct environmental conditions ( Garchitorena et al . , 2014 ) offers the variability needed to address this question . Indeed , in Bankim , a region located in a transition zone between forest and savannah , swamps had remarkably higher MU positivity , as initially expected , whereas in Akonolinga , where rainforest is the prevailing landscape , all ecosystems had similar MU positivity . These regional differences suggested that savannah swamps had unique favourable conditions for MU that were not found elsewhere , but MU was still able to persist in unfavourable environments . Furthermore , temporal fluctuations in MU presence in Akonolinga suggested a potential role of seasonal climatic events as drivers of MU dynamics . Relying on this work , the aim of this paper is to study for the first time the contribution of ecological factors , both biotic and abiotic , to the dynamics of MU in the aquatic environment . More specifically , we attempt to identify a set of abiotic conditions that could be optimal for MU growth and allow direct transmission to aquatic organisms , likely in stagnant waters and , in the absence of these , explore which biotic factors could still allow MU to persist , potentially through trophic transmission . Insights into the ecological mechanisms allowing for MU growth and persistence over space and time , while accounting for the potential impact of seasonal climatic events , may have profound implications for understanding BU risk to human populations . To address these questions , we model MU positivity in 32 aquatic communities over time with generalized linear mixed models ( GLMMs ) , including all relevant seasonal , abiotic , and biotic factors as fixed effects . We use cutting-edge multi-model selection procedures and information theory to identify and quantify the most important predictors of MU dynamics , using a genetic algorithm to screen multiple models from all potential combinations of explanatory variables and making inference from a set of weighted best-performing models . In addition , we back the results of this novel approach , which deals with the uncertainty associated with model selection , by comparing them to those obtained by classical model selection procedures . We then discuss the implications of disentangling biotic and abiotic factors for host/parasite interactions and the importance of rigorously analysing the underlying drivers of pathogen dynamics mediated through complex and different-scale processes . Seasonality was investigated by including sine and cosine functions as independent predictors of MU positivity . The presence of MU in aquatic ecosystems in Akonolinga was associated to seasonal variations with a single annual cycle as revealed by the positive effect of the sine function on MU ( b = 0 . 36; 95% confidence interval ( CI ) [0 . 04 , 0 . 67] , wi = 1 ) and the presence of this variable in all best models ( Table 2 ) . On the contrary , the seasonal effect in Bankim was not apparent , where only 4 months of collection were available , and none of the sine and cosine functions were important in the final models for this region ( Table 3 ) . Among all abiotic conditions , included in the models both as individual physico-chemical variables or through their combined effect as PCs , pH had a significant positive effect in all best models in Akonolinga ( Table 2 ) , either through the effect of component 2 that is directly correlated with pH ( b = 0 . 49; 95% CI [0 . 21 , 0 . 76] , wi = 0 . 56 ) or as individual variable in the remaining models ( b = 7 . 15; 95% CI [2 . 56 , 11 . 73] , wi = 0 . 44 ) . In Bankim , however , the most important abiotic factor was water flow ( Table 3 ) . Lentic water bodies ( low water flow ) had significantly lower MU positivity than stagnant waters ( b = −1 . 80; 95%CI [−3 . 04 , −0 . 56] , wi = 1 ) , and lotic water bodies ( high water flow ) had the lowest MU positivity ( b = −3 . 63; 95% CI [−5 . 35 , −1 . 91] , wi = 1 ) . It is important to note that water flow in most environments was directly correlated with pH , and thus each region provides contrasting results for this abiotic factor . The impact of aquatic communities on MU was studied through the effect of both individual aquatic taxa and community-level factors such as abundance and diversity . In Akonolinga , we found a negative association between total abundance and MU presence in all final models ( b = −0 . 71; 95% CI [−1 . 07 , −0 . 35] , wi = 1 ) and individual effects of several taxa . Individual taxa inversely correlated with MU were Gastropoda in all models ( b = −0 . 58; 95% CI [−0 . 92 , −0 . 24] , wi = 1 ) and Decapoda ( b = −1 . 10; 95% CI [−1 . 84 , −0 . 35] , wi = 0 . 59 ) , Hemiptera ( b = −0 . 54; 95% CI [−0 . 94 , −0 . 13] , wi = 0 . 54 ) , and Anura ( b = −0 . 41 , 95% CI [−0 . 73 , −0 . 09] , wi = 0 . 41 ) , with lower importance in the final models ( Table 2 ) . Finally , the orders Oligochaeta , Odonata , and Hydracarina had a large importance in the final models ( wi > 0 . 80 ) , but their effect on MU presence was not significant . Results for Bankim revealed a significant positive effect in all models for Shannon's diversity index ( b = 4 . 29; 95% CI [1 . 93 , 6 . 66] , wi = 1 ) and nearly significant for total abundance ( b = 0 . 86; 95% CI [−0 . 06 , 1 . 79] , wi = 1 ) . In addition , most taxonomic orders appeared in the final models for this region , but their effect was unclear ( Table 3 ) . The estimates for taxonomic groups with a relative importance over 0 . 8 ( Gastropoda , Anura , Trichoptera , Odonata , Fish , Coleoptera , and Diptera ) had considerable uncertainty , with upper and lower CIs of opposite sign . The only two orders with a significant CI were Hydracarina ( b = −1 . 42 , 95% CI [−2 . 50 , −0 . 33] ) and Ephemeroptera ( −0 . 84; 95% CI [−1 . 33 , −0 . 35] ) , but their relative importance was relatively low ( wi = 0 . 58 and wi = 0 . 13 , respectively ) . Understanding how environmental factors influence host–pathogen interactions in complex natural systems , where multiple feedbacks between biotic and abiotic factors take place , is especially important in the context of multi-host and environmentally persistent pathogens . In this study , we identify abiotic and biotic drivers that may promote or block MU transmission in aquatic communities in two climatically distinct regions of Cameroon through a comprehensive multi-model selection procedure . In Akonolinga , we show that MU follows seasonal dynamics and is mainly present in waters with higher pH and within low abundance communities , notably those with low abundance of Gastropoda and other orders such as Decapoda , Hemiptera , or Anura . In Bankim , we show that MU is most prevalent in stagnant ecosystems and those with low water flow , with highly diverse ( and abundant ) communities . A seasonal effect for MU presence in Akonolinga remains in our final models after accounting for abiotic and biotic parameters in the water bodies , which also vary seasonally . This suggests that seasonal fluctuations in MU presence might be directly related to climatic pressures ( Figure 1 ) . Indeed , while the seasonal effect is not directly linked to rainfall dynamics ( Pearson's correlation test , p = 0 . 45 ) , it is highly correlated with the 3-month mean rainfall accumulation in the region ( Current month , plus two previous months; Pearson's correlation test , p < 0 . 01 ) . As a result , we propose that the cumulative effect of rainfall over several months , increasing water levels in the environment either boosts MU growth or washes it off from other environmental matrices ( mud , soil , plants ) to aquatic ecosystems , as previously suggested from epidemiological evidence ( Morris et al . , 2014 ) . Furthermore , given the slow growth of MU ( Palomino and Portaels , 1998; Stinear et al . , 2007 ) , the 2-month delay between the peaks in the dynamics of rainfall and those of the seasonal effect could represent the time that takes MU to grow and/or be transmitted through the aquatic community once the suitable habitats have been created . Unfortunately , the less frequent sampling in Bankim ( only 4 months instead of 12 ) may have prevented to capture seasonal variations appropriately , explaining the lack of associations with MU in this region . 10 . 7554/eLife . 07616 . 006Figure 1 . Link between the seasonal effect for M . ulcerans and the rainfall dynamics in Akonolinga . ( A ) Represents the monthly values for the seasonal effect ( red ) , the mean rainfall for the period under study and the 3-month cumulative rainfall ( blue ) . ( B ) Shows a clear linear relationship between the values of the seasonal effect and the 3-month cumulative rainfall . A graphical representation of the different seasonal effects tested can be found in Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07616 . 00610 . 7554/eLife . 07616 . 007Figure 1—figure supplement 1 . Values for the different seasonal effects tested in the statistical models . The seasonal effect was tested through sin ( A ) and cosine ( B ) functions with frequencies of 12 and 4 months ( solid and dashed lines , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07616 . 007 Our results for Bankim support the hypothesis that , under certain circumstances , conditions in stagnant and slow-flowing ( lentic ) bodies of water are favourable for MU presence . After controlling for all the other abiotic and biotic factors , sites with stagnant waters in this area have higher MU positivity than those with lentic waters ( slow flow ) , and these have in turn higher positivity than sites with lotic waters ( faster flow ) . PCA on physico-chemical parameters of these ecosystems provides some potential explanations ( Figure 2 ) . Sites with stagnant or lower water flows have higher temperatures ( PC1 and PC2 ) and most have lower oxygen ( PC1 ) and lower pH ( PC1 ) , all of which seem to promote MU growth in experimental studies ( Portaels and Pattyn , 1982; Palomino and Portaels , 1998; Palomino et al . , 1998; Dega et al . , 2000; Marsollier et al . , 2004 ) . Indeed , a previous field study suggested that some water characteristics may be important for the presence of mycobacteria in water and biofilms throughout the year ( Hennigan et al . , 2013 ) . 10 . 7554/eLife . 07616 . 008Figure 2 . Impact of water flow on physico-chemical characteristics of the water and M . ulcerans prevalence in aquatic communities ( Bankim ) . ( A ) Links between water conditions in the first two principal components obtained through principal component analysis ( PCA ) . Comp . 1 , explaining more than 50% of the variation in physico-chemical conditions in Bankim , reveals that ecosystems with lower water flows have less dissolved oxygen , more acidic pH , and higher temperature . ( B ) MU positivity in each type of ecosystem as described by the first component of the PCA , which takes into account variations in all physico-chemical characteristics ( each category has equal number of points and increasing values of Comp . 1 ) . Stagnant ecosystems in Bankim have higher MU positivity than lentic , and these have in turn higher MU positivity than lotic ecosystems . ( C ) Difference in values for the various water conditions in MU positive and MU negative sites in Bankim . As a result of the association of water flow with the other physico-chemical conditions , similar patterns for MU positivity can be observed for most abiotic conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 07616 . 008 While stagnant waters with lower pH contribute significantly to MU presence in Bankim , the results for Akonolinga show a positive association between pH and MU in this region . Significantly lower pH values in Akonolinga than in Bankim ( t-test , p < 0 . 001 ) may be behind the disparity between the model results for each region ( Figure 3 ) . Indeed , pH range for slow-growing mycobacteria has been estimated between 5 . 8 and 6 . 5 ( Portaels and Pattyn , 1982 ) , which corresponds to the lower range of pH in Bankim , associated with stagnant waters . Because in Akonolinga , this optimal range corresponds to the upper range of values , stagnant waters with intolerably low pH might not meet all the optimal conditions for MU growth , which would explain the lack of association with these ecosystems ( Garchitorena et al . , 2014 ) . The role of pH on MU growth in combination with other abiotic conditions needs to be urgently assessed , since it might be an important limiting factor in the environment . 10 . 7554/eLife . 07616 . 009Figure 3 . Distribution of relevant biotic and abiotic variables for Akonolinga and Bankim . For the construction of histograms ( A–D ) , the relative frequency of the variable within each region is normalized by dividing each frequency by its maximum frequency . It can be noted that the distribution of pH is radically different for both regions , with much more acidic pH in aquatic environments from Akonolinga . For the community composition ( E and F ) , the area an order has in the pie chart is proportional to the mean relative abundance of the order for all sites and months for each region . Only orders representing more than 1% of the overall community are labelled . DOI: http://dx . doi . org/10 . 7554/eLife . 07616 . 009 Biotic interactions seem to have an important effect on MU positivity in the local aquatic communities , especially in Akonolinga . Less abundant communities are associated with a reduction of MU in this region , and individual taxa have an independent effect on MU . Higher relative abundance of aquatic snails ( Gastropoda ) , shrimps ( Decapoda ) , water bugs ( Hemiptera ) , and tadpoles ( Anura ) is associated with reduced MU prevalence in the aquatic community . The protective role of aquatic snails is supported by experimental infections , where MU has been unable to grow within these organisms ( Marsollier and Sévérin , 2004 ) , but this is not the case for Hemipteran water bugs , where MU can grow and even colonize their salivary glands after they have fed on infected prey ( Marsollier et al . , 2002 , 2005; Mosi et al . , 2008 ) . Even though water bugs can host MU and allow its growth , they are voracious predators of aquatic organisms , and therefore , an increase in water bugs in the community may result in a decrease of infected prey available to other predators such as Coleoptera or Odonata; this could result in an overall reduction in MU positivity . This is an example of how considering the full breadth of factors taking place in real ecological systems can provide unexpected insights on this type of host–pathogen interactions . Furthermore , differences in community composition may partly explain the different effects of biotic factors in the two regions ( Figure 3C–F ) . Total abundance in aquatic communities was significantly higher in Akonolinga ( Mann–Whitney test , p < 0 . 001 ) , and the relative abundance of more than half of the taxa included in our model was significantly different between Akonolinga and Bankim ( Appendix 1 , section 2 ) . These different distributions of biotic and abiotic factors can nevertheless yield a hypothesis to explain the contrasting results between the two regions . Indeed , two transmission routes , not competitively exclusive , may coexist for MU colonization of aquatic organisms , through a trophic transmission ( Roche et al . , 2013a ) and/or a pathogen transmission through infection with free-living stages present in favourable aquatic environments ( Merritt et al . , 2010 ) . In our study , community abundance has opposite effects in Akonolinga and Bankim , while water flow and pH suggest contrasted influence of stagnant waters in these regions . These results could suggest that the prevailing transmission modes could be different within these two environmentally distinct regions . Transmission could be mainly environmentally mediated in Bankim , since stagnant waters , through weak water flows and optimal physico-chemical conditions , are strongly associated with MU presence . Furthermore , a lack of association of MU abundance with specific taxa in addition to strong positive associations with host diversity and abundance under these favourable conditions , suggests that infection probability in these environments is density dependent , a characteristic feature of this type of transmission ( Codeço , 2001 ) . Conversely , in Akonolinga , trophic transmission may be expected since optimal abiotic conditions are not met in stagnant ecosystems , and host abundance has a negative impact on MU presence , suggesting that presence of some taxa , at least Gastropoda and Hemiptera , can limit transmission in aquatic communities . These alternative transmission routes proposed for MU to persist and thrive in aquatic ecosystems could partly explain why BU distribution in humans is greatly associated with stagnant ecosystems but expands over larger geographical regions ( Brou et al . , 2008; Wagner et al . , 2008; Jacobsen and Padgett , 2010; Marion et al . , 2011 ) . Our results demonstrate the complex interplay between abiotic and biotic factors driving the dynamics of multi-host/multi-environment diseases . By studying and comparing savannah- and rainforest-like regions , we provide a comprehensive ecological picture of MU , that is , a unified framework that reconciles the many contrasting findings observed during the last decade that could apply to a broader geographical area in the tropics and could help us understand the risk of BU for human populations . This study provides a new illustration of emerging infectious diseases for which further investigations looking for a ‘bigger picture’ are clearly needed in order to cope with the complexity of local and regional environmental situations , and different-scale processes . Judging by the number and importance of multi-host and environmentally persistent pathogens in the total number of emerging infectious diseases appeared in the last four decades such hypotheses deserve to be rigorously tested across multiple epidemiological systems and diverse local conditions . More comprehensive environmental studies in other contexts are needed to assess the generalizability of our findings . Data were collected as described in Garchitorena et al . ( 2014 ) . Briefly , between June 2012 and May 2013 , periodic sampling in aquatic ecosystems was performed monthly in Akonolinga and every 3 months in Bankim , two regions in Cameroon where BU is endemic . Akonolinga health district is located in the Centre Province , where rainforest is predominant all across the region . Bankim , on the other hand , is a health district located in the Adamaoua Province , near the border with Nigeria , in a transition zone between forest and savannah . In all , 32 water sites were selected ( 16 in each region ) , including a large variety of streams , rivers , swamps , and flooded areas .	Mycobacterium ulcerans is a slow-growing bacterium that causes a rare tropical disease in humans called Buruli ulcer . The infection affects the skin and underlying tissues , initially causing small painless lesions that can develop into large open sores or ulcers that are responsible for significant handicap in rural areas of sub-Saharan Africa . Disease outbreaks generally occur in close association with stagnant and slow-flowing aquatic ecosystems , mostly after floods or other large environmental disturbances . It is believed that contact with water sources that contain the bacteria causes infection in humans , but the specific mode of transmission remains a mystery . By defining the factors that influence the bacteria's presence in the environment , public health officials could develop initiatives that would reduce an individual's risk of infection when conditions support a M . ulcerans outbreak . To identify the environmental conditions that affect the prevalence of M . ulcerans in two regions of Cameroon where Buruli ulcer is present , Garchitorena et al . have now analyzed a large amount of ecological data about the bacteria using cutting-edge statistical techniques . This revealed that the amount of M . ulcerans varies following seasonal changes in climate , at least in the region dominated by tropical rainforest . In this region , the bacteria are also generally present in waters that are more alkaline and contain fewer animals , especially from certain species that could prevent the infection spreading to other aquatic hosts . In the other region , dominated by a savannah landscape , the bacteria are most abundant in stagnant or slowly moving waters that have optimal physical and chemical conditions and contain many diverse species of potential animal hosts . The discovery of contrasting results for the two regions suggests that there are at least two ways that M . ulcerans can persist in the environment and infect the aquatic animals . The prevailing method—through environmental transmission or through interactions between hosts—depends on the properties of the water . Many other infectious diseases are caused by pathogens that , like M . ulcerans , infect many different hosts and persist in the environment for long periods . Future research following methods like those used by Garchitorena et al . would help to reveal whether these pathogens are affected by environmental factors in similar ways to M . ulcerans .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "epidemiology", "and", "global", "health" ]	2015	Mycobacterium ulcerans dynamics in aquatic ecosystems are driven by a complex interplay of abiotic and biotic factors
Serotonin's function in the brain is unclear . One challenge in testing the numerous hypotheses about serotonin's function has been observing the activity of identified serotonergic neurons in animals engaged in behavioral tasks . We recorded the activity of dorsal raphe neurons while mice experienced a task in which rewards and punishments varied across blocks of trials . We ‘tagged’ serotonergic neurons with the light-sensitive protein channelrhodopsin-2 and identified them based on their responses to light . We found three main features of serotonergic neuron activity: ( 1 ) a large fraction of serotonergic neurons modulated their tonic firing rates over the course of minutes during reward vs punishment blocks; ( 2 ) most were phasically excited by punishments; and ( 3 ) a subset was phasically excited by reward-predicting cues . By contrast , dopaminergic neurons did not show firing rate changes across blocks of trials . These results suggest that serotonergic neurons signal information about reward and punishment on multiple timescales . Reward and punishment play critical roles in shaping animal behavior over short and long timescales ( Rolls , 2005; Somerville et al . , 2013 ) . On short timescales , moment-to-moment expectations of reward or punishment increase or decrease motivation to perform a specific action associated with the reward or punishment . On long timescales , repeated exposure to reward or punishment can elicit long-lasting positive or negative emotional states ( often called ‘mood’ ) , that can increase or decrease the frequency of performing reward-seeking actions more generally ( Niv et al . , 2006; Cools et al . , 2011; Somerville et al . , 2013; Wang et al . , 2013 ) . The midbrain raphe nuclei contain the majority of forebrain-projecting serotonergic neurons in mammals ( Jacobs and Azmitia , 1992 ) . This small population of neurons ( approximately 9000 in the mouse dorsal raphe; Ishimura et al . , 1988 ) projects to almost the entire forebrain ( Azmitia and Segal , 1978; Moore et al . , 1978; Steinbusch , 1981; O'Hearn and Molliver , 1984; Vertes , 1991; Gagnon and Parent , 2014 ) . The diffuse projection targets of the serotonergic system have led to many theories about its function . Serotonin has been proposed to be involved in processing reward and punishment ( Maswood et al . , 1998; Daw et al . , 2002; Maier and Watkins , 2005; Nakamura et al . , 2008; Ranade and Mainen , 2009; Tops et al . , 2009; Cools et al . , 2011; Amo et al . , 2014 ) . One theory proposes that serotonin regulates aversive learning and negative motivation in response to punishments ( Soubrié , 1986; Deakin and Graeff , 1991; Daw et al . , 2002; Dayan and Huys , 2009 ) . According to this theory , serotonin opposes the positive reinforcement and behavioral activation regulated by dopamine . Whereas dopaminergic neurons signal appetitive prediction errors ( Schultz et al . , 1997 ) , serotonergic neurons could signal punishments , thereby adjusting future behavior to avoid those punishments or inhibiting specific actions that are associated with punishments . This theory has support from lesion , stimulation , tryptophan depletion , and pharmacological studies ( Tye et al . , 1977; Graeff and Silveira Filho , 1978; Liu and Ikemoto , 2007; Crockett et al . , 2009; Shin and Ikemoto , 2010 ) , but there is little evidence that serotonergic neurons signal punishments in awake animals ( cf . Aghajanian et al . , 1978; Montagne-Clavel et al . , 1995; Schweimer and Ungless , 2010 ) . A second theory proposes that serotonin signals global reward states , such as tracking average reward ( Daw et al . , 2002 ) and modulating mood ( Savitz et al . , 2009 ) . Here , serotonin is thought to provide its diffuse targets with long-term signals about the value of the environment , which , at the extreme , is correlated with changes in mood . These long-term signals inhibit behavior or increase the vigor of taking actions in a relatively action-general manner . This theory has support from clinical observations ( Fava and Kendler , 2000 ) and genetic studies ( Donaldson et al . , 2013 ) . However , there is little neurophysiological evidence for such a function . A third theory proposes that serotonin is involved in waiting for reward ( Miyazaki et al . , 2011a , 2011b , 2014; Fonseca et al . , 2015 ) . In this theory , activation of serotonergic neurons promotes patience ( Miyazaki et al . , 2014 ) or slows movements that allow an animal to wait for a delayed reward . This theory could explain how waiting for reward could be linked to behavioral inhibition ( Soubrié , 1986; Fonseca et al . , 2015 ) . Clarifying whether and how serotonin exerts these functions requires understanding how serotonergic neuron firing correlates with both global features of the environment , such as changes in average reward value , and with punishments . These data have been challenging to collect because of the heterogeneity of neurons in and around the raphe nuclei . About two-thirds of dorsal raphe neurons are serotonergic , but others contain GABA , glutamate , dopamine , acetylcholine , or various peptides ( Hökfelt et al . , 2000; Commons , 2009; Fu et al . , 2010; Hioki et al . , 2010 ) . Many previous studies have relied on spike waveform characteristics to identify serotonergic neurons in extracellular recordings . This approach may lead to false positives and misses , however ( Allers and Sharp , 2003; Kirby et al . , 2003; Marinelli et al . , 2004; Urbain et al . , 2006; Hajós et al . , 2007; Schweimer et al . , 2011 ) . Given this chemical diversity and the difficulty of identifying neuron types in extracellular recordings , it would be unsurprising to find corresponding physiological diversity . Indeed , previous studies found significant heterogeneity in the activity of dorsal raphe neurons in relation to behavioral tasks ( Fornal et al . , 1996; Nakamura et al . , 2008; Ranade and Mainen , 2009; Bromberg-Martin et al . , 2010; Miyazaki et al . , 2011a; Inaba et al . , 2013; Li et al . , 2013 ) . One set of experiments demonstrated that dorsal raphe neurons fired in a reward-value-dependent manner during trials of a saccade-to-target task , but that the modulation disappeared before the next trial ( Nakamura et al . , 2008; Bromberg-Martin et al . , 2010 ) . Another set of experiments showed tonic firing rate modulations correlating with waiting for reward from putative dorsal raphe serotonergic neurons ( Miyazaki et al . , 2011a ) . These studies suggested that serotonergic neurons signal reward expectation or waiting time ( or patience ) on the scale of hundreds of milliseconds , but does not determine whether serotonin could be involved in longer-term changes in value . In the present study , we sought to test whether dorsal raphe serotonergic neuron firing correlated with rewards and punishments on different timescales . To address this question , we used a task in which reward value changed on slow and fast timescales . Importantly , we recorded from optogenetically-identified serotonergic neurons . Our results show that serotonergic neurons signal information about reward and punishment across multiple timescales . We classically conditioned head-fixed , thirsty mice with different odor cues that predicted a reward ( water ) , neutral outcome ( nothing ) , or punishment ( a puff of air delivered to the animal's face; Cohen et al . , 2012 ) . Each behavioral trial began with a conditioned stimulus ( CS; an odor , 1 s ) , followed by a 1-s delay and an unconditioned stimulus ( US; the outcome; Figure 1A ) . Mice licked toward the water-delivery tube in the delay period before rewards arrived , but not during neutral or punishment trials , indicating that they had learned the CS-US associations ( analysis of variance ( ANOVA ) , t-tests , p < 0 . 001 for each session; Figure 1B ) . We varied the structure of the task so that animals received blocks of 10 reward trials , alternating with 10 punishment trials , with a tone indicating transitions between blocks ( or , in 35% of sessions , block order was random; Figure 1C; block duration mean ± S . D . , 1 . 36 ± 0 . 30 min ) . Mice licked significantly more during the tone preceding reward vs punishment blocks . Because the tone was the same preceding both block types , this indicates that they attended to the block structure ( Wilcoxon signed-rank test , p < 0 . 01; Figure 1—figure supplement 1 ) . In each block of some sessions , the reward or punishment trial was replaced by a neutral trial with probability 0 . 1 . To ensure that the time of onset of each trial could not be predicted following the end of the previous one , we used an exponentially-distributed ( flat hazard rate ) inter-trial interval ( ITI ) . The ITIs lasted , on average , longer than twice the duration of trials ( ITI , 6 . 41 ± 1 . 88 s , mean ± S . D . ) . Using this task , we could study neuronal responses on fast ( CS and US ) and slow ( across blocks ) timescales . 10 . 7554/eLife . 06346 . 003Figure 1 . Behavioral task . ( A ) Structure of individual trials . ( B ) Average lick rate for all animals during each trial type . ( C ) Representative sequence of trials from one experiment . Each point represents an odor cue . Shaded regions indicate reward blocks . ( D ) Schematic of midbrain indicating recording sites ( shaded ) , with low and high magnification of 5-HT labeling ( cyan ) , ChR2-EYFP ( magenta ) , nuclei ( DAPI ) , and their overlay . Scale bars are 100 µm and 10 µm for low and high magnification , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 00310 . 7554/eLife . 06346 . 004Figure 1—figure supplement 1 . Histogram of lick rate during tones indicating block transitions , for experiments in which reward blocks alternated with punishment blocks . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 00410 . 7554/eLife . 06346 . 005Figure 1—figure supplement 2 . Mice treat air puffs as punishments . ( A ) Mice performed a two-alternative forced choice task , in which they chose between a water reward and a water reward together with an air puff , indicating their response by moving to the associated response port . ( B ) Task timing , in which an odor cue in a central port signals a choice , followed by an outcome . ( C ) Median ( horizontal line ) , interquartile range ( box ) , and 1 . 5 times interquartile range ( whiskers ) of proportion of water choices for each mouse , across five experiments of 300 trials each . Each mouse chose water over water together with air puffs significantly more than chance ( Tukey's Honest Significant Difference tests , p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 005 To determine whether mice treated air puffs as punishments , we performed a behavioral experiment in which mice were given free choices between a water reward and a water reward delivered simultaneously with an air puff . Mice reliably chose the water reward without the air puff , confirming that air puffs act as a negative reinforcement ( Figure 1—figure supplement 2 ) . We recorded the activity of 149 dorsal raphe neurons while mice performed the conditioning task ( 6 mice , 24 . 8 ± 6 . 1 neurons per mouse , mean ± S . E . M . ) . We expressed channelrhodopsin-2 ( ChR2 ) , a light-gated cation channel , in serotonergic neurons by injecting an adeno-associated virus containing FLEX-ChR2 ( AAV5-EF1α-DIO-hChR2 ( H134R ) -EYFP-WPRE-pA ) into the dorsal raphe of transgenic mice expressing Cre recombinase under the control of the promoter of the serotonin transporter gene ( Slc6a4; Sert-Cre mice ) . Expression was specific ( of 260 ChR2-EYFP-positive cells , 6 cells , or 2 . 3% , were 5-HT-negative ) and efficient ( of 314 5-HT-positive cells , 254 , or 80 . 9% , were ChR2-EYFP-positive; Figure 1D ) . For each neuron , we measured the response to light stimulation and the shape of spontaneous spikes ( Figure 2A , B ) . We reasoned that for a neuron to be identified as responding to light stimulation , it must first show a significant response to the stimulus ( quantified here as light-evoked energy , defined as the integral of the squared voltage values ∫v2dt ) . Second , to ensure that the neuron under observation , rather than a population of nearby ChR2-expressing serotonergic neurons , responded to light stimulation , we verified that the response to light stimulation matched the shape of spontaneous spikes . We calculated the distance between the spontaneous spike waveform and light-evoked voltage response and plotted it against the energy of light-evoked response for each recording ( Figure 2B ) . Using an expectation-maximization clustering method , we observed two distinct clusters: one that showed significant responses to light pulses and one that did not . 29 neurons fell into the former cluster ( filled cyan points in Figure 2B ) . Three points in that cluster were not considered identified serotonergic neurons because they did not reliably respond to light stimulation . Consistent with direct light activation rather than indirect synaptic activation , all 29 neurons showed fast light-evoked spikes ( Figure 2C ) and followed high-frequency stimulation ( Figure 2D , E ) . These properties indicate that these 29 neurons expressed ChR2 ( henceforth called ‘serotonergic neurons’; 5 mice , 5 . 8 ± 1 . 5 neurons per mouse , mean ± S . E . M . ) . 10 . 7554/eLife . 06346 . 006Figure 2 . Identifying serotonergic neurons . ( A ) Example voltage trace from 10 pulses of 10-Hz light stimulation ( cyan bars; light duration , 5 ms ) . Each light-triggered spike is shown below . The lower right is the first two principal components of all waveforms from one tetrode wire , showing the neuron's ( ‘unit’ ) isolation quality , with 100 randomly-chosen light-evoked spikes in cyan . ( B ) Quantification of light-evoked responses to identify serotonergic neurons ( filled points ) . Abscissa: energy ( integral of the squared voltage values ) of the light-evoked response from each neuron . Ordinate: Euclidean distance between the mean spontaneous spike and the light-evoked response . Example neurons are shown to the right ( black , spontaneous spikes; cyan , light-evoked voltages; SD of spike waveforms are smaller than line thicknesses ) . Note that three unfilled points in the lower-right cluster are not considered identified serotonergic neurons because of low probability of firing in response to light stimulation . ( C ) Mean and SD of the light-evoked spike latency for identified serotonergic neurons . Probability ( D ) and latency ( E ) of light-evoked spikes from serotonergic neurons as a function of stimulation frequency ( filled points are means across neurons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 006 The structure of the behavioral task allowed us to ask whether serotonergic neuron firing correlated with short-term ( rewards , punishments , and the cues that predicted them within a trial ) or long-term ( blocks of reward vs punishment trials ) changes in the environment . We first asked whether serotonergic neuron firing rates were significantly modulated during the behavioral task . We performed an ANOVA on the trial-by-trial firing rates during the baseline period ( 1 s before odor onset ) , CS period ( from odor onset to odor offset ) , delay ( from odor offset to outcome onset ) , and US period ( from outcome onset to 500 ms after outcome onset ) . The factors were task epoch ( baseline , CS , delay , or US ) and outcome type . All 29 serotonergic neurons exhibited task-related modulations in firing rate ( ANOVA , all p < 0 . 01 ) . Next , we examined the responses of serotonergic neurons in detail . We observed three main features of the firing patterns of serotonergic neurons . First , a large fraction ( 41% ) of serotonergic neurons fired at a higher or lower rate during the ITIs of reward blocks vs punishment blocks ( Figure 3A , B; Figure 3—figure supplement 1 ) . That is , even before a particular trial began , these neurons fired at a rate that correlated with the value ( reward vs punishment ) of the block . Remarkably , this response persisted across minutes . To quantify this observation , we calculated the firing rate in the 2 s before the start of each trial during reward and punishment blocks . 12 of 29 serotonergic neurons showed significantly different pre-trial firing rates between reward and punishment blocks: 7 were more excited during reward blocks , 5 were more excited during punishment blocks ( Figure 3C , D; Wilcoxon rank sum tests , p < 0 . 05 ) . Interestingly , this tonic signal appeared to build up or down slowly within blocks , rather than sharply increasing or decreasing in response to block transitions ( Figure 3—figure supplement 2 ) . This tonic signal did not depend on the duration of ITIs ( Wilcoxon rank sum tests , all p > 0 . 3 ) . In addition , 16 of 29 serotonergic neurons displayed gradually decreasing firing rates over the course of the experiment ( Figure 3—figure supplement 3 ) . 10 . 7554/eLife . 06346 . 007Figure 3 . A population of serotonergic neurons is more or less active during blocks of reward trials than punishment trials . ( A ) Average firing rates of four example serotonergic neurons during reward trials ( black ) and punishment trials ( orange ) . Shaded regions denote S . E . M . Note the higher pre-trial firing rate during reward trials than punishment trials in the top two neurons and the higher firing rate during punishment vs reward blocks in the third neuron . ( B ) Firing rate of the same four neurons across the timecourse of the experiment . Note the slow ( across minutes ) fluctuations in firing rate in the top three neurons that correlated with block type ( reward: black , shaded regions; punishment: orange ) . ( C ) Mean ±95% confidence intervals around firing rates during the baseline epoch for punishment vs reward blocks for each serotonergic neuron ( significant data points are filled ) . Examples from ( A ) are labeled . ( D ) Average firing rates of serotonergic neurons with significantly higher baseline firing rates during reward ( top ) or punishment ( bottom ) blocks . ( E ) Average firing rate of an example identified dopaminergic neuron during reward and punishment trials . ( F ) Firing rate across the timecourse of the experiment for the dopaminergic neuron in ( E ) . ( G ) Mean ±95% confidence intervals around firing rates during the baseline epoch for punishment vs reward blocks for each dopaminergic neuron ( identified in cyan , putative in white ) . ( H ) Average firing rate of dopaminergic ( identified and putative ) neurons during reward and punishment trials . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 00710 . 7554/eLife . 06346 . 008Figure 3—figure supplement 1 . Raster plots showing spike times during 40 trials for the four example neurons in Figure 3A . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 00810 . 7554/eLife . 06346 . 009Figure 3—figure supplement 2 . ( A ) Normalized mean ± S . E . M . firing rate within reward and punishment blocks for the positive-coding serotonergic neurons . Note the build-up and build-down activity within blocks . ( B ) Raw firing rates for reward and punishment blocks for six example serotonergic neurons with tonic firing rate differences across blocks . ( C ) Mean ± S . E . M . lick rate during the delay between CS and US as a function of trial within block . Note the lack of resemblance to the firing rates in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 00910 . 7554/eLife . 06346 . 010Figure 3—figure supplement 3 . ( A ) Trial-by-trial scatter plot of lick rate against spike rate for an example serotonergic neuron . ( B ) Pearson correlation coefficient of trial-by-trial lick rate and spike rate across serotonergic neurons . The two neurons with significant correlations are indicated in gray . ( C ) Firing rate across reward trials for two example serotonergic neurons , with linear fits in dashed gray . The top neuron had a significantly decreasing firing rate across trials , whereas the bottom neuron did not . ( D ) Slope of firing rate across reward trials for all serotonergic neurons , with slopes significantly different from zero indicated in gray . ( E–H ) Slope of firing rate across reward ( E and F ) and punishment ( G and H ) trials for all serotonergic neurons , during CS ( odor onset to odor offset; E and G ) and US ( US onset to 500 ms after US onset; F and H ) , with slopes significantly different from zero indicated in gray . ( I ) Difference between background firing rates during reward blocks and pre- ( top panel ) or post-session ( top panel ) activity , for serotonergic neurons . ( J ) Difference between background firing rates during punishment blocks and pre- ( top panel ) or post-session ( top panel ) activity , for serotonergic neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 01010 . 7554/eLife . 06346 . 011Figure 3—figure supplement 4 . Example activity of a dopaminergic neuron with a smaller response to predicted reward than unpredicted reward-predicting cue . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 011 To compare this effect to the response of dopaminergic neurons—which have been proposed to be involved in long-term value-related signaling ( Niv et al . , 2006; Cools et al . , 2011 ) —in the same task , we recorded the activity of 28 ventral tegmental area ( VTA ) neurons , 15 of which were identified as dopaminergic ( by ChR2 tagging , as described above ) , and 13 of which were putatively dopaminergic based on their task-related responses ( 4 mice , 7 . 0 ± 2 . 0 neurons per mouse , mean ± S . E . M . ) . None of the putative or identified dopaminergic neurons showed significantly different pre-trial firing rates between reward and punishment blocks ( Wilcoxon rank sum tests , all p > 0 . 5; Figure 3E–H ) . Most dopaminergic neurons were excited by predicted rewards to varying degrees ( see Figure 3—figure supplement 4 for a more ‘canonical’ example ) , similar to prior observations in classical conditioning tasks with trace delays in mice ( Cohen et al . , 2012 ) and monkeys ( Fiorillo et al . , 2008 ) . The second main feature of serotonergic neuron activity we observed was their response to punishments ( Figure 4A–C , Figure 4—figure supplement 1 ) . Previous studies found identified serotonergic neurons to be excited or inhibited by punishments in anesthetized animals ( Aghajanian et al . , 1978; Montagne-Clavel et al . , 1995; Schweimer and Ungless , 2010 ) . To test whether punishments modulated serotonergic firing , we calculated the area under the receiver operating characteristic ( auROC ) curve in sliding 100-ms windows for each neuron , comparing each window during the trial to the baseline firing rate ( 1 s before punishment trials; Figure 4B ) . The auROC quantifies the discriminability of the two firing rate distributions . Values of 0 . 5 ( black ) indicate no change in firing rate relative to baseline . Values greater than 0 . 5 ( yellow ) indicate increases in firing rate relative to baseline , while values less than 0 . 5 ( blue ) indicate decreases in firing rate relative to baseline . Most serotonergic neurons ( 28 out of 29 ) responded phasically to punishments: 22 were excited , 6 were inhibited ( Figure 4A–C; Wilcoxon rank sum tests in the 500 ms after punishment onset , p < 0 . 05 ) . This response was transient , lasting less than 500 ms for most neurons ( 315 ± 140 ms , mean ± S . D . ) . 10 . 7554/eLife . 06346 . 012Figure 4 . Serotonergic neurons are briefly excited or inhibited by punishments or reward-predicting cues . ( A ) Average firing rates of two example serotonergic neurons during punishment trials . CS and US analysis windows are shaded in gray . ( B ) Area under the ROC curve for punishment trials for all serotonergic neurons , sorted by the sum of the auROC for reward trials in ( E ) . Yellow indicates excitation , blue indicates inhibition , and black indicates no change relative to baseline . ( C ) Histogram of changes in firing rate relative to baseline during the CS and US epochs of punishment trials . ( D ) Average firing rates of two example serotonergic neurons during reward trials . ( E ) Area under the ROC curve for reward trials for all serotonergic neurons , sorted by their sum . ( F ) Histogram of changes in firing rate relative to baseline during the CS and US epochs of reward trials . ( G ) Area under the ROC curve for free reward for dopaminergic and serotonergic neurons . ( H ) Average firing rate of dopaminergic and serotonergic neurons around free reward ( shaded curves show S . E . M . ) . ( I ) Histograms of average differences between mean firing rates during expected vs unexpected rewards for serotonergic and dopaminergic neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 01210 . 7554/eLife . 06346 . 013Figure 4—figure supplement 1 . ( A ) Area under the ROC curve for reward ( water ) vs punishment ( air puff ) trials for serotonergic neurons , sorted by the sum of the auROC . Yellow indicates excitation , blue indicates inhibition , and black indicates no change relative to baseline . ( B ) Histogram of changes in firing rate during the CS and US epochs of reward vs punishment trials . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 013 The third main feature of serotonergic neuron activity we observed was their response to reward-predicting cues ( Figure 4D–F ) . We calculated the auROC , comparing the baseline ( 1 s before reward trials ) to each 100-ms window during the trial . About half of serotonergic neurons ( 15 out of 29 ) were phasically excited by reward-predicting cues ( Figure 4D–F; Wilcoxon rank sum tests during the 1 s of odor presentation , p < 0 . 05 ) . Very few neurons ( 2 out of 29; p < 0 . 05 ) were inhibited by reward-predicting cues . The duration of excitation was brief , lasting less than 500 ms ( 235 ± 194 ms , mean ± S . D . ) . For 9 of these 15 neurons , the response to a reward-predicting CS was significantly greater than the response to a punishment-predicting CS ( Wilcoxon rank sum tests , p < 0 . 05 ) . The time of peak response during a reward-predicting CS was significantly shorter for serotonergic than dopaminergic neurons ( mean ± S . E . M . , 331 ± 15 . 7 ms for serotonergic neurons , 388 ± 12 . 7 ms for dopaminergic neurons , Wilcoxon rank sum test , p < 0 . 05 ) . Recent work has provided conflicting views on the role of the dorsal raphe in reward behavior ( Liu and Ikemoto , 2007; Shin and Ikemoto , 2010; Liu et al . , 2014; McDevitt et al . , 2014; Miyazaki et al . , 2014; Qi et al . , 2014; Fonseca et al . , 2015 ) . To compare the phasic reward-related responses during the task to responses to unexpected rewards , we examined dopaminergic and serotonergic responses to unexpected reward ( delivered at random times prior to the task in 5 Slc6a4-Cre mice and 3 Slc6a3-Cre mice ) . Whereas dopaminergic neurons showed a large , phasic excitation in response to unexpected rewards , a subset of serotonergic neurons was weakly and slowly , but significantly , excited ( Wilcoxon rank sum tests , p < 0 . 05 for 9 of 10 dopaminergic neurons and 11 of 29 serotonergic neurons; Figure 4G , H ) . We compared neuronal responses to unexpected rewards to responses to expected rewards in the context of the task . Dopaminergic neurons showed larger responses to unexpected vs expected rewards ( Figure 4I ) , similar to previous observations ( e . g . , Schultz et al . , 1997; Cohen et al . , 2012 ) . Interestingly , serotonergic neurons showed a weak but significantly larger response to unexpected vs expected rewards ( Figure 4I ) . Next , we tested whether these three features were correlated within the population of serotonergic neurons . We found that the difference in pre-trial firing rate during reward vs punishment blocks positively correlated with the difference in response to reward- vs punishment-predicting CS; neurons with higher excitation for reward than punishment CS tended to fire at a higher rate before reward than punishment trials ( Figure 5A; Pearson's r = 0 . 85 , t27 = 8 . 56 , p < 0 . 001 ) . This demonstrates that the same serotonergic neurons can multiplex different signals about reward and punishment on different timescales . 10 . 7554/eLife . 06346 . 014Figure 5 . Correlations between serotonergic neuron response features . ( A ) Difference in firing rate during the 2-s pre-trial epoch vs the difference in firing rate during the CS ( reward − punishment ) , corrected for baseline differences . ( B ) Area under the ROC curve for the 2-s pre-trial epoch during reward vs punishment trials plotted against area under the ROC curve during the punishment US epoch . ( C ) Area under the ROC curve for during the reward CS vs punishment US epochs . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 014 The difference in pre-trial firing rate ( reward vs punishment ) did not significantly correlate with the response to punishment ( Figure 5B; Pearson's r = 0 . 03 , t27 = −0 . 17 , p > 0 . 8 ) . The response to reward CS was significantly positively correlated with the response to punishment ( Figure 5C; Pearson's r = 0 . 60 , t27 = 3 . 89 , p < 0 . 01 ) . To test whether differences between firing rates during reward vs punishment blocks reflected value , as opposed to salience , we compared the firing rates during neutral trials to those during reward and punishment trials . Neurons with responses to neutral stimuli that fall in between those to rewards and punishments are defined as value-coding . Those with responses to rewards and punishments that are either both larger , or both smaller , than responses to neutral stimuli , are defined as salience-coding ( Matsumoto and Hikosaka , 2009 ) . We asked whether serotonergic neurons were value-coding across blocks . For 16 serotonergic neurons , we modified the task to include blocks of neutral trials randomly interspersed among reward and punishment blocks ( with equal probability ) . Eight of these displayed significantly different pre-trial firing rates during reward vs punishment blocks ( Figure 3C ) . For 7 of these 8 neurons ( 3 of which displayed higher firing rates during reward blocks , 5 of which displayed higher firing rates during punishment blocks ) , pre-trial firing rates during neutral blocks did not fall outside of the bounds of those during reward and punishment blocks ( Figure 6A; Tukey's Honest Significant Difference tests , p > 0 . 05 ) . Only 1 of the 16 neurons had both significantly larger pre-trial firing rates during neutral blocks than during reward and punishment blocks . Thus , 15 of 16 serotonergic neurons were not salience-coding across blocks of trials . 10 . 7554/eLife . 06346 . 015Figure 6 . Serotonergic neuron background firing rates signal graded value . ( A ) Median ( horizontal line ) , interquartile range ( box ) , and 1 . 5 times interquartile range ( whiskers ) of pre-trial firing rates during reward ( black ) , neutral ( gray ) , and punishment ( orange ) blocks . Brackets indicate significant differences . ( B ) Mean ±95% confidence intervals pre-trial response during blocks of three reward sizes for 13 serotonergic neurons with strictly increasing ( blue ) or decreasing ( red ) firing rates as a function of reward size . ( C ) Mean ± S . E . M . firing rate of two example serotonergic neurons during reward ( water or chocolate milk ) and punishment ( air puff , orange; quinine , green ) trials . ( D ) Mean ±95% confidence intervals around firing rates during the baseline epoch for punishment ( quinine ) vs reward ( water or chocolate milk ) blocks for each serotonergic neuron ( significant data points are filled ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 01510 . 7554/eLife . 06346 . 016Figure 6—figure supplement 1 . Behavioral task . ( A ) Structure of individual trials . ( B ) Average lick rate for all animals during each trial type . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 01610 . 7554/eLife . 06346 . 017Figure 6—figure supplement 2 . ( A ) Area under the ROC curve for reward , air-puff punishment , and quinine punishment trials for 21 serotonergic neurons . ( B ) Area under the ROC curve for the punishment period for quinine vs air-puff punishments . Pearson correlation with significance test is shown . ( C ) Mean ± S . E . M . firing rate of serotonergic neurons in the task that included different reward sizes . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 017 So far , we have described differences in background firing rates as being value-related due to their differences between reward ( water ) , neutral ( no stimulus ) , and punishment ( air puff ) conditions . We asked whether responses across blocks were truly value-related , or , rather , due to differences in responses to the different sensory modalities for reward and punishment ( gustatory vs somatosensory ) . We performed two experiments to address this . We recorded from 13 additional identified serotonergic neurons in which blocks contained trials with one of three reward sizes: zero , small ( 1 µl ) , or big ( 4 µl; Figure 6—figure supplement 1 ) . If pre-trial , tonic responses are value-related , they should vary monotonically with reward size . Indeed , for each of the 13 neurons , firing rates were either strictly increasing ( 5 neurons ) or strictly decreasing ( 8 neurons ) with reward size ( Figure 6B; Tukey's Honest Significant Difference tests , p > 0 . 05 ) . Next , we recorded from 21 additional identified serotonergic neurons ( 4 mice , 5 . 3 ± 1 . 0 neurons per mouse , mean ± S . E . M . ) in a similar task with four types of blocks: water or chocolate milk reward , neutral , air-puff punishment , and quinine punishment . Quinine , a bitter-tasting solution , is a punishment of the same sensory modality as a water reward . We found that 8 of these 21 neurons were tonically more ( 6 neurons ) or less ( 2 neurons ) active in water vs quinine blocks , during ITIs ( Figure 6C , D ) . These neurons also showed a positive correlation in their response to the two different punishments , albeit with a weaker response to quinine ( Figure 6—figure supplement 2 ) . This could be due to the longer timecourse or smaller magnitude of the aversiveness of quinine compared to air puffs . Across both experiments , 20 of 50 identified serotonergic neurons showed tonic differences in firing for reward vs punishment blocks of trials . Thus , background firing rate in serotonergic neurons signal value across different magnitudes of reward and different types of punishment . Next , we asked whether serotonergic neurons were value-coding within trials . We compared the firing rates during the CS , rather than US , for two reasons . First , there was no US during neutral trials . Second , outcomes were predicted by the CS , therefore neuronal responses to the US could be confounded by expectation . For each of the 23 serotonergic neurons with sufficient data for this analysis , the CS-induced firing rate as a function of value was monotonic ( Figure 7A , B ) . As a population , the CS firing rate was larger for reward than punishment ( paired t-test , t22 = 2 . 6 , p < 0 . 05 ) , larger for reward than neutral ( t22 = 2 . 9 , p < 0 . 05 ) , and larger for neutral than punishment ( t22 = 2 . 2 , p < 0 . 05; Figure 7C–E ) . This suggests that firing rate differences between reward and punishment trials reflected the value , not the salience , of those stimuli . 10 . 7554/eLife . 06346 . 018Figure 7 . Serotonergic neurons signal value , not salience , in response to the CS . ( A ) Mean ± S . E . M . firing rate of serotonergic neurons during all three trial conditions . ( B ) Mean ± S . E . M . firing rate during the CS epoch ( region bounded by dashed lines in A ) for each trial condition . ( C–E ) Mean ±95% confidence intervals around firing rates during the CS epoch for neutral vs reward ( C ) , punishment vs reward ( D ) , and punishment vs neutral ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 018 As with identified serotonergic neurons , we observed many unidentified neurons with firing-rate fluctuations from block to block ( Figure 8A , B ) . We note that this population of unidentified neurons likely contains serotonergic as well as non-serotonergic neurons because of incomplete ChR2 expression or our strict criteria for identification . 10 . 7554/eLife . 06346 . 019Figure 8 . A population of unidentified neurons is more or less active during blocks of reward trials than punishment trials . ( A ) Mean ± S . E . M . firing rates of three example neurons during reward trials ( black ) and punishment trials ( orange ) . Note the higher pre-trial firing rate during reward trials than punishment trials in the bottom two neurons . ( B ) Firing rate of the same three neurons across the timecourse of the experiment . Note the slow ( across minutes ) fluctuations in firing rate in the bottom two neurons that correlated with block type ( reward: black , shaded regions; punishment: orange ) . ( C ) Mean ±95% confidence intervals around firing rates during the baseline epoch for punishment vs reward blocks for each unidentified neuron ( significant data points are filled ) . ( D ) Mean ± S . E . M . firing rates of unidentified neurons with significantly higher baseline firing rates during reward ( top ) or punishment ( bottom ) blocks . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 01910 . 7554/eLife . 06346 . 020Figure 8—figure supplement 1 . Response profiles across all neurons and all trial types . The figure shows the heterogeneity of unidentified neuron responses relative to serotonergic neurons . Area under the ROC curve for reward ( left ) and punishment trials ( right ) for each neuron . Neurons are sorted by the sum of the auROC values from reward trials . Serotonergic neurons are marked with arrows to the left of the plot . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 02010 . 7554/eLife . 06346 . 021Figure 8—figure supplement 2 . Serotonergic neurons cannot be identified based on firing properties in this data set . ( A ) Firing rate vs spike duration with marginal density histograms . Serotonergic neurons had significantly longer spike duration than unidentified neurons ( Wilcoxon rank sum test , p < 0 . 05 ) . ( B ) Example inter-spike interval ( ISI ) density histograms for three serotonergic neurons and three unidentified neurons . ( C ) Mean spontaneous spikes ( black ) and light-evoked voltages ( cyan ) from all serotonergic and unidentified neurons . Asterisks indicate the three unidentified neurons from the lower-right cluster in Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 06346 . 021 We calculated the firing rate in the 2 s before the start of each trial during reward blocks and punishment blocks . 29 of 120 unidentified neurons showed significantly different pre-trial firing rates between reward and punishment blocks: 10 were more excited during reward blocks , 19 were more excited during punishment blocks ( Figure 8C , D , Wilcoxon rank sum tests , p < 0 . 05 ) . In addition to these slow firing-rate fluctuations across minutes , 92 of 120 unidentified neurons showed task-related responses during the trial ( ANOVA , all p < 0 . 01; Figure 8—figure supplement 1 ) . These neurons were either excited or inhibited by rewards , punishments , reward-predicting cues , and punishment-predicting cues to varying degrees and durations . Although it is likely that there were false negatives ( serotonergic neurons contained in the populations of unidentified neurons; Figure 8—figure supplement 1 ) , there were significant differences between serotonergic and unidentified neurons . For instance , the duration of the significant response ( excitation or inhibition ) was significantly longer for unidentified neurons than for serotonergic neurons . That is , serotonergic neuron responses during the trial epoch ( from CS to US ) tended to be more phasic than those of the 92 task-responsive unidentified neurons ( Fisher's exact test on the proportion of 100-ms bins significantly different from baseline , odds ratio = 0 . 643 , p < 0 . 01 ) . Finally , as has been observed previously in anesthetized cats and rats , we found both serotonergic and unidentified neurons that displayed ‘clock-’ and ‘non-clock-like’ firing patterns ( Figure 8—figure supplement 2B; Nakahama et al . , 1981; Schweimer and Ungless , 2010 ) . Serotonergic neurons are typically identified extracellularly based on their wide spike shapes and low firing rates ( Bramwell , 1974; Aghajanian et al . , 1978; Baraban et al . , 1978; Baraban and Aghajanian , 1980; Aghajanian and Vandermaelen , 1982; Gallager , 1982; Heym et al . , 1982; Wang and Aghajanian , 1982; Chiang and Pan , 1985; Fornal et al . , 1985 , 1987; Cunningham and Lakoski , 1988; Levine and Jacobs , 1992; Ceci et al . , 1994; Guzmán-Marín et al . , 2000; Celada et al . , 2001; Sakai and Crochet , 2001; Waterhouse et al . , 2004; Miyazaki et al . , 2011a ) . These criteria have recently been called into question , however ( Park , 1987; Allers and Sharp , 2003; Kirby et al . , 2003; Marinelli et al . , 2004; Kocsis et al . , 2006; Urbain et al . , 2006; Hajós et al . , 2007; Ranade and Mainen , 2009; Bromberg-Martin et al . , 2010; Schweimer and Ungless , 2010; Schweimer et al . , 2011; Gocho et al . , 2013; Li et al . , 2013 ) . We asked whether we could have identified serotonergic neurons in our data set in this way . Serotonergic neurons had significantly longer spike duration than unidentified neurons ( Wilcoxon rank sum test , p < 0 . 05 ) , although there was significant overlap in the distributions ( Figure 8—figure supplement 2A ) . There was no significant difference between the mean firing rate ( Figure 8—figure supplement 2A; Wilcoxon rank sum test , p > 0 . 6 ) or coefficient of variation of the inter-spike interval distributions ( Wilcoxon rank sum test , p > 0 . 1 ) between serotonergic and unidentified neurons . Neither an expectation-maximization nor a k-means clustering algorithm could classify serotonergic neurons based on spike duration , firing rate , or shape of the inter-spike interval distribution . Reward and punishment can exert their effects on behavior on multiple timescales . It has been proposed that the average reward rate on relatively long timescales ( or state value ) regulates the vigor of behavioral responding in a manner relatively non-specific to actions ( Niv et al . , 2006; Wang et al . , 2013 ) . However , the neural mechanisms that regulate this process remain unclear . Much attention has been paid to a potential involvement of tonic dopaminergic firing in this process ( Niv et al . , 2006; Cools et al . , 2011 ) . In contrast to this proposal , we did not find that dopaminergic neurons changed their baseline firing according to state values . Our data showed , instead , that 40% of serotonergic neurons changed their baseline firing depending on the state value of the environment . We observed this effect using rewards and punishments of the same sensory modality , and using brief punishments ( air puffs ) , in which the acute aversiveness likely did not persist into the ITI . These firing rate changes were relatively small in magnitude ( around 1–2 spikes s−1 ) , but given the low baseline firing rates of serotonergic neurons , these changes corresponded to around 20–100% increases in firing rates , which could have led to substantially higher serotonin release . Our data also showed that serotonergic neurons exhibited transient activations associated with various task events ( Nakamura et al . , 2008; Ranade and Mainen , 2009 ) . Serotonergic neuron responses to reward- or punishment-predictive cues as well as reward or punishment were transient , typically lasting less than 500 ms , and relatively small in magnitude ( 5–10 spikes s−1 ) , in agreement with previous studies ( Nakamura et al . , 2008; Ranade and Mainen , 2009; Bromberg-Martin et al . , 2010; Miyazaki et al . , 2011a; Inaba et al . , 2013 ) , though substantially smaller than a recent one ( Li et al . , 2013 ) . Previous studies observed that dorsal raphe neurons exhibited sustained activities that appeared to track moment-to-moment changes in value triggered by sensory cues and outcomes , within a trial ( Nakamura et al . , 2008; Bromberg-Martin et al . , 2010 ) . Although these activities lasted for several hundreds of milliseconds to seconds , our identified serotonergic neurons showed relatively transient activities within trials . In contrast , our data showed that many unidentified neurons showed sustained activities within trials , suggesting that sustained activities may be more common in non-serotonergic neurons . Another recent set of studies found that putative serotonergic neurons showed firing modulations lasting up to several seconds , during a task in which rats waited for a reward ( Miyazaki et al . , 2011a ) , and that manipulations of serotonergic signaling altered waiting behavior ( Miyazaki et al . , 2014; Fonseca et al . , 2015 ) . This raises the interesting possibility that both time and reward value modulate the firing of serotonergic neurons , increasing the flexibility of the serotonergic signal . Indeed , the tonic signal we observed could be the subjective value of waiting ( i . e . , waiting for punishments elicits a low-value state , whereas waiting for rewards elicits a high-value state ) . A third study found that dorsal raphe neurons , on average , fired at a higher rate before cues that predicted rewards relative to cues that predicted no reward ( Li et al . , 2013 ) . In this work , trials were also delivered in a block-wise fashion , though it was not possible to identify serotonergic neurons , nor was there an analysis of the slow modulations in firing rate in individual neurons . The sign of value-dependent changes in tonic firing varied across serotonergic neurons: some increased and others decreased during periods of high state values although , on average , high-value blocks were associated with higher tonic firing rates . Nevertheless , the topography of projections from the raphe ( Imai et al . , 1986; Vertes , 1991; Lowry et al . , 2000; Chandler et al . , 2013 ) and the physiological topography within the raphe ( Lowry et al . , 2000; Crawford et al . , 2010 ) suggest the potential for a specific mapping between subsets of serotonergic neurons and their functions . Our results cannot distinguish these subpopulations of serotonergic neurons . It remains to be examined whether different firing patterns of serotonergic neurons correspond to specific subpopulations . Together , these results suggest that background serotonin could serve as an explicit signal of state values . It is important to note that our data demonstrate a relative value code during the task ( blocks of trials of different values ) , but do not speak to the possibility that serotonergic neurons signal an absolute state value ( cf . Figure 3—figure supplement 3 ) . As discussed above , reward and punishment affect behavior on multiple timescales . Our finding that tonic serotonergic firing tracks values over long timescales raises the possibility that tonic serotonin regulates long-lasting affective states . In addition to the changes in tonic firing , our data showed that a large fraction of serotonergic neurons were excited by reward-predictive cues and unpredicted reward on shorter timescales ( i . e . , within trials ) . Although there was substantial diversity in serotonergic neuron firing patterns , during reward trials , on average , their response to reward-predicting cues resembled those of dopaminergic neurons signaling reward prediction error ( Schultz et al . , 1997; Cohen et al . , 2012 ) . It is important to note , however , that the magnitude of increases was smaller , and did not appear to signal prediction errors for most neurons ( Figures 3D , H , 4I ) . In addition , we found that many serotonergic neurons were excited by punishments . It is important to disentangle whether the phasic responses to air puffs we observed are due to aversiveness or other factors such as saliency or relief from punishment ( Heym et al . , 1982; Waterhouse et al . , 2004; Dayan and Huys , 2009; Cools et al . , 2011 ) . The short response latency suggests that it is unlikely to be relief from punishment , but the nature of phasic air puff responses remains to be clarified . Of course , all of these signals could be expressed by the population of serotonergic neurons . In addition , it will be important to understand why the phasic response to air puffs was stronger than the phasic response to air-puff-predicting cues . It has been proposed that serotonergic and dopaminergic neurons work largely in an opponent manner ( Daw et al . , 2002 ) . This idea was supported by pharmacology and intracranial self-stimulation ( Redgrave , 1978 ) , as well as a few recording studies that suggested phasic activations of serotonergic neurons by aversive stimuli ( Aghajanian et al . , 1978; Montagne-Clavel et al . , 1995; Schweimer and Ungless , 2010 ) . However , in these studies , responses were measured in anesthetized animals or the effect of reward was not examined , leaving unanswered whether phasic serotonin signals purely aversive information . Our data showed that many single serotonin neurons were activated by both reward-predictive cues and punishment ( Figure 5C ) . Although it remains unclear whether these phasic firing patterns can be unified to encode a particular variable , or whether the response to punishment was due to its sensory nature or the relief arriving at the end of the punishment , these firing patterns appear to be inconsistent with the idea that serotonergic neurons send an opponent signal compared to dopaminergic neurons , though median raphe serotonergic neurons could provide such a signal ( Daw et al . , 2002 ) . This lack of pure opponency is consistent with recent studies measuring putative serotonergic firing and serotonin concentration in a waiting task ( Miyazaki et al . , 2011a , 2011b ) . Our results suggest that serotonergic neurons multiplex information about reward and punishment on multiple timescales . We propose that slow , value-related firing could represent state value , whereas phasic responses to CS or US could encode a different variable . This slow signal appears to require some time to change ( Figure 3—figure supplement 2 ) . Future studies should clarify how downstream neurons read out tonic vs phasic serotonin signals . It is possible that tonic serotonin has very different effects at target neurons and on behavior than phasic serotonin due to receptors with different affinities or other mechanisms ( Daw et al . , 2002 ) . In addition , serotonergic neurons are known to contain other transmitters ( Varga et al . , 2009; Liu et al . , 2014 ) , suggesting that the slow and fast timescales could correspond to the action of different transmitters on target neurons , or to downstream circuit effects with differing durations . The function of serotonin could be target-dependent ( Deakin and Graeff , 1991 ) , timing-dependent ( Daw et al . , 2002; Dayan and Huys , 2009 ) , or dependent on co-release of other transmitters ( Dayan and Huys , 2009; Varga et al . , 2009; Liu et al . , 2014; McDevitt et al . , 2014 ) . Another possibility is that serotonin combines with other circuits to form logical combinations postsynaptically ( for example , serotonin AND dopamine codes for reward , whereas serotonin AND NOT dopamine codes for punishment ) . How do serotonergic neurons compute the signals over short ( hundreds of milliseconds ) and long ( minutes ) timescales ? Serotonin release is controlled by many types of afferents . The densest include frontal cortex , basal forebrain areas ( bed nucleus of the stria terminalis , substantia innominata , ventral pallidum ) , hypothalamic nuclei ( preoptic nucleus , lateral and posterior nuclei ) , lateral habenula , and several midbrain and brainstem structures ( Peyron et al . , 1998; Pollak Dorocic et al . , 2014; Ogawa et al . , 2014; Weissbourd et al . , 2014 ) . Neurons in the VTA and substantia nigra pars compacta ( SNc ) , some dopaminergic , provide input to raphe serotonergic neurons ( Beckstead et al . , 1979; Pollak Dorocic et al . , 2014; Ogawa et al . , 2014; Weissbourd et al . , 2014 ) . These neurons are phasically excited by rewards or reward-predicting cues ( Schultz et al . , 1997; Cohen et al . , 2012 ) , but their activity appears not to correlate with longer-term changes in value ( Figure 3 above; Matsumoto and Hikosaka , 2009 ) . They may excite serotonergic neurons via D2 receptors ( Haj-Dahmane , 2001; Aman et al . , 2007; but see; Martín-Ruiz et al . , 2001 ) . Interestingly , dopaminergic stimulation of putative serotonergic neurons in vitro caused excitation that persisted beyond the application of dopamine ( Haj-Dahmane , 2001 ) . It is possible that serotonergic neurons could accumulate information about rewards via short pulses of input from dopaminergic neurons , although our observation of shorter reward-predicting CS response latencies in serotonergic vs dopaminergic neurons suggests the opposite . There are several possible explanations for why we could not use spike width to clearly identify serotonergic neurons . First , unidentified neurons likely contained serotonergic neurons that we could not identify using our stringent criteria . Second , the shape of extracellulary-recorded spikes depends on the position of the electrode tip to the soma and dendrites of the recorded neuron ( Schultz , 1986; Harris et al . , 2000; Henze et al . , 2000; Buzsáki et al . , 2012 ) . Given that the geometry of dendritic and axonal processes in the dorsal raphe is diverse , with fusiform , multipolar , and ovoid somata ( Diaz-Cintra et al . , 1981 ) , it is potentially difficult to optimize the location of electrode tip relative to cell bodies across the population ( cf . Schultz , 1986 ) . Indeed , we did not attempt to optimize this position , once we found an identified serotonergic neuron . Third , our nichrome wires could have introduced recording bias away from small cells or certain cell shapes , potentially less of an issue for glass pipette recordings ( Towe and Harding , 1970; Shoham et al . , 2006; O'Connor et al . , 2010 ) . Indeed , in our previous study in the ventral tegmental area , using the same type of electrode , we did not find the oft-reported difference between spike width for dopaminergic vs non-dopaminergic neurons ( Cohen et al . , 2012 ) . Careful simultaneous intra- and extracellular measurements ( Harris et al . , 2000; Henze et al . , 2000 ) in the dorsal raphe are needed to resolve this . The present study revealed that serotonergic neurons can use tonic as well as phasic firing to convey reward information . This raises questions as to whether these different modes of firing convey distinct information , whether they have different impacts on target neurons ( notably , on dopaminergic neurons [Fibiger and Miller , 1977; Watabe-Uchida et al . , 2012; Ogawa et al . , 2014] ) , and how they are calculated . Although our data likely do not reveal the full diversity of serotonergic neuron firing dynamics , our results suggest distinguishing these distinct modes is crucial to teasing apart the seemingly complex functions of serotonin in various brain functions and disorders . For dorsal raphe recordings , we used nine adult male mice , backcrossed with C57BL/6J mice , heterozygous for Cre recombinase under the control of the serotonin transporter gene ( Slc6a4tm1 ( cre ) Xz; Zhuang et al . , 2005 ) . For VTA recordings , all of which came from the task in Figure 1 , we used four adult male mice backcrossed with C57BL/6J mice , heterozygous for Cre recombinase under the control of either the dopamine transporter ( 3 mice ) or tyrosine hydroxylase gene ( 1 mouse ) ( Slc6a3tm1 . 1 ( cre ) Bkmn/J and B6 . Cg-Tg ( Th-cre ) 1tmd/J , respectively , The Jackson Laboratory; Savitt et al . , 2005; Bäckman et al . , 2006 ) . We did not observe any differences between the different genotypes during the behavioral task . For cell counting , we used three additional adult male Sert-Cre mice . Animals were housed on a 12 hr dark/12 hr light cycle ( dark from 06:00–18:00 ) and each performed the conditioning task at the same time of day , between 07:00 and 19:00 . All surgical and experimental procedures were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the Harvard or Johns Hopkins Institutional Animal Care and Use Committees . Mice were surgically implanted with a head plate and a microdrive containing electrodes and an optical fiber . During a prior surgery , we injected 200–500 nl adeno-associated virus ( AAV ) , serotype 2/5 , using the EF1α promoter , carrying an inverted ChR2 ( H134R ) -EYFP flanked by double loxP sites ( Nagel et al . , 2003; Boyden et al . , 2005; Atasoy et al . , 2008 ) into the dorsal raphe stereotactically ( from bregma: 4 . 4–4 . 7 mm posterior , 0 . 1–0 . 2 mm lateral , 1 . 9–2 . 3 mm ventral ) , or into VTA as previously described ( Cohen et al . , 2012 ) . All surgery was performed under aseptic conditions with animals under ketamine/medetomidine ( 60 and 0 . 5 mg/kg , I . P . , respectively ) or isoflurane ( 1–2% at 0 . 5–1 . 0 l/min ) anesthesia . Analgesia ( ketoprofen , 5 mg/kg , I . P . ; buprenorphine , 0 . 1 mg/kg , I . P . ) was administered postoperatively . After at least 1 week of recovery , mice were water-deprived in their home cage . Weight was maintained within 90% of their full body weight . Animals were head-restrained using a custom-made metal plate and habituated for 1–2 day while head-restrained before training on the task . Odors were delivered with a custom-made olfactometer ( Uchida and Mainen , 2003 ) . Each odor was dissolved in paraffin oil at 1/10 dilution . 30 µl of diluted odor was placed inside a filter-paper housing . Odors were isoamyl acetate , 1-butanol , N-citral , eugenol , ( + ) limonene , ( − ) carvone , ( + ) carvone , allyl tiglate , eucalyptol , acetophenone , hydroxymethylpentylcyclohexenecarboxaldehyde , 3-hexanone , pentyl acetate , 1-hexanol , p-cymene , and ethyl butyrate , and differed for different animals . Odorized air was further diluted with filtered air by 1:10 to produce a 500 ml/min total flow rate . Licks were detected by breaks of an infrared beam placed in front of the water tube . We delivered one of three odors , selected pseudorandomly , for 1 s , followed by a delay of 1 s and an outcome . Each odor predicted a different outcome: a drop of water ( 4 µl ) , no outcome , or an air puff delivered to the animal's face . ITIs were drawn from an exponential distribution with a rate parameter of 10 ( i . e . , P ( ITI ) = 1/10 exp ( −x/10 ) ) . This resulted in a flat ITI hazard function , ensuring that expectation about the start time of the next trial did not change over time . A 15 kHz tone lasting 1 s signaled block changes , ending 1 s before the start of the next trial . Data were obtained from 141 sessions ( 19–28 sessions per animal ) . For 17 identified serotonergic neurons , we omitted rewards during 10% of big-reward trials . Animals performed between 400 and 700 trials per day ( 533 ± 120 trials , mean ± SD ) . Free rewards were delivered before the start of the conditioning task . We did not observe differences in lick rate ( Figure 1 ) or sniff rate ( counted manually during 1-s intervals of 20 reward trials and 20 punishment trials for two mice; t-tests , t38 = 1 . 28 , p > 0 . 20 , t38 = 1 . 05 , p > 0 . 29 for each mouse ) during ITIs of different block types . For 16 of 29 serotonergic neurons , block type varied randomly , whereas for 13 of 29 serotonergic neurons , and all dopaminergic neurons , block type alternated between reward and punishment . For 16 of 29 serotonergic neurons , neutral blocks were included , in which case an additional odor was used as a CS . For 2 of 29 serotonergic neurons , block size was 20 trials . We trained each animal on the task with randomly-interleaved trials for 5 days before beginning the blocked structure . For the task including quinine as a US , we used water ( 4 µl ) or chocolate milk ( 1% , Hood , Lynnfield , Massachusetts , 4 µl ) as a reward US and quinine HCl ( 0 . 5–1 . 0 mM , 4 µl ) as a punishment US . To ensure that animals ingested the quinine , which is known to be aversive ( Schoenbaum et al . , 1998; Berridge , 2000; Peciña and Berridge , 2005; Roitman et al . , 2005 ) , we placed the delivery tube at the entry to their mouths . Other task parameters were the same as above . For the freely-moving behavioral task , we trained five adult male C57BL/6J mice to perform a two-alternative forced choice task as follows ( Figure 1—figure supplement 1 ) . An odor cue ( 1-hexanol ) delivered at a central port was a start cue that signaled the mouse to choose between one peripheral port that contained water ( 4 µl ) and a second that contained water ( 4 µl ) and an air puff . After an ITI ( same distribution as for the tasks described above ) , the next trial began . Mice performed 300 trials of this task for 5 days , after 6 days of shaping and training . We recorded extracellularly from multiple neurons simultaneously using custom-built 200-µm-fiber-optic-coupled screw-driven microdrives with eight implanted tetrodes ( four nichrome wires wound together , Sandvik , Palm Coast , Florida ) . Tetrodes were glued to fiber optics with epoxy or cyanoacrylate . The ends of tetrodes were 400–600 µm from the ends of fiber optics . Neural signals and time stamps for behavior were recorded using DigiLynx recording systems ( Neuralynx , Bozeman , Montana ) . Broadband signals from each wire were filtered between 0 . 1 and 9000 Hz and recorded at 32 kHz . To extract the timing of spikes , signals were bandpass-filtered between 300 and 6000 Hz and sorted online and offline . To verify that our recordings targeted serotonergic or dopaminergic neurons , we used ChR2 to observe stimulation-locked spikes ( Cardin et al . , 2009; Lima et al . , 2009; Cohen et al . , 2012 ) . The optical fiber was coupled to a diode-pumped solid-state laser with analog amplitude modulation ( Laserglow Technologies , Toronto , Canada ) . For each neuron , we delivered trains of 10 light pulses , each 5 ms long , at 1 , 5 , 10 , 20 , and 50 Hz at 473 nm at 5–20 mW/mm2 , before and after the experimental session . Higher intensities typically resulted in light-evoked spike waveforms that did not match spontaneous ones . Therefore , we adjusted the light intensity after observing the responses at the beginning of experiments . The increasing latency of light-evoked spiking as a function of stimulation frequency indicates that the response was not due to photochemical artifact ( Figure 2E ) . Spike shape was measured using a broadband signal ( 0 . 1–9000 Hz ) sampled at 32 kHz . We used two criteria to include a neuron in our data set . First , the neuron must have been recorded within 500 µm of an identified serotonergic neuron , to ensure that all neurons came from the dorsal raphe ( except for dopaminergic neurons , which all came from VTA ) . Second , the neuron must have been well isolated . To measure isolation quality , we calculated the L-ratio ( Schmitzer-Torbert and Redish , 2004 ) , which approximates the fraction of ‘contaminated’ spikes . Smaller L-ratios indicate better isolation . All neurons in the data set had L-ratios < 0 . 05 and signal-to-noise ratios of > 5 dB . Identified serotonergic neurons , of which there were 50 across experiments , came from all nine mice ( a range of 1–10 per mouse ) . To measure firing rates , peristimulus time histograms ( PSTHs ) were constructed using 10-ms bins . To calculate spike density functions , PSTHs were convolved with a function resembling a postsynaptic potential , ( 1 − exp ( −t ) ) exp ( −t/200 ) , for time t . For display ( but not analysis ) , we smoothed spike density functions with a smoothing spline with 30 degrees of freedom ( Kass et al . , 2005 ) . To determine whether a neuron showed a significant task-related response , we calculated an ANOVA on the trial-by-trial firing rates during the baseline period ( 1 s before odor onset ) , CS period ( from odor onset to odor offset ) , delay ( from odor offset to outcome onset ) , and US period ( from outcome onset to 500 ms after outcome onset ) . The factors were task epoch ( baseline , CS , delay , or US ) and outcome type . Normality was tested by Kolmogorov–Smirnov tests and quantile–quantile plots . All two-group comparisons were two-sided . Effect sizes for each experiment were determined post-hoc using Cohen's U3 ( cf . Hentschke and Stüttgen , 2011 ) . Light-evoked spikes were detected during the 10 ms after light onset . If less than 20% of light pulses evoked a spike ( defined as a waveform that matched that of the isolated unit ) during the 10 ms after light onset ( upper left points in Figure 2B ) , the maximum absolute voltage during that interval was used as the light-evoked ‘response’ . Euclidean distances between spontaneous and light-evoked spike waveforms were calculated by aligning the larger of the positive or negative peak of each waveform , averaging separately , and aligning the peaks of the averages . The distance was calculated using the full duration of the spontaneous spike ( spike duration was measured as the first time until the last time at which the voltage was significantly different from baseline using Wilcoxon rank sum tests ) . The energy of the light-evoked waveform is defined as the integral of the squared voltage values ( ∫v2dt ) . ROC curves were calculated by comparing the distribution of firing rates ( spike density functions ) across trials in 100-ms bins to the distribution of baseline firing rates ( 1 s before odor onset , using 100-ms bins ) . The duration of significant responses were calculated using Wilcoxon rank sum tests comparing the baseline firing rate to the firing rate in the response window of interest , bin by bin . The number of bins in which p < 0 . 05 was taken as the duration of the significant response ( after Bonferroni corrections ) . Expectation-maximization clustering was performed using hierarchical clustering for parameterized Gaussian mixture models , setting the number of clusters to 2 ( one cluster of ‘identified serotonergic neurons’ and one of ‘unidentified neurons’ ) , with model selection by Bayesian Information Criterion . All statistical tests were done with Bonferroni corrections for multiple comparisons . Analyses were done with R ( http://www . r-project . org/ ) . After recording , which lasted between 19 and 28 days , mice were given an overdose of ketamine/medetomidine , exsanguinated with saline , perfused with 4% paraformaldehyde , and brains were cut in 50–100 µm coronal sections . Sections were immunostained overnight with a primary antibody to 5-HT ( 1:200; S5545 , Sigma–Aldrich , St . Louis , Missouri; Steinbusch et al . , 1978; Bras et al . , 1986; Dai et al . , 2008 ) , then incubated with an Alexa594-coupled secondary antibody for 2 hr ( 1:200; Invitrogen , Carlsbad , California ) . Sections were further counterstained with 4′ , 6-diamidino-2-phenylindole ( DAPI ) to visualize nuclei . Recording sites were identified and verified to be amid EYFP expression and 5-HT staining in the dorsal raphe .	Rewards and punishments can both encourage animals to change their immediate behavior and influence their mood over a longer term , particularly when given repeatedly . A region of the brain that increases its activity in response to rewards and punishments also contains many neurons that communicate with each other by releasing a chemical called serotonin . This chemical is commonly thought to produce feelings of happiness; however , it remains unclear exactly how these particular ‘serotonergic’ neurons help to process rewards and punishments . The ideal way to work out the role that a type of neuron plays in a behavior is to measure its electrical activity as the behavior is being performed . However , it is difficult to distinguish the activity of serotonergic neurons from the activity of the non-serotonergic neurons around them . To overcome this problem , Cohen et al . used viruses to force serotonergic neurons to make a type of ion channel that produces electrical currents in response to light . Shining light on these neurons via optical fibers and then measuring the neurons' responses helped to develop criteria that can identify which responses are generated by the serotonergic neurons . Cohen et al . then recorded the activity of serotonergic neurons in thirsty mice as they experienced a series of rewards ( for example , a drop of water ) or punishments ( such as a puff of air to the eye ) . Each reward or punishment was preceded by a distinct odor , so that the mice learned to anticipate what was coming . These experiments revealed that serotonergic neurons respond to rewards and punishments by changing two aspects of their electrical activity: by producing short bursts of high activity , and by altering their baseline activity . Some of the serotonergic neurons fired rapidly in response to punishments , but not rewards; others fired rapidly when the mice detected a scent that meant that a reward was about to be given . The average level of reward or punishment the mice received also affected the baseline activity of many of the serotonergic neurons; this effect lasted for several minutes . Overall , Cohen et al . suggest that serotonergic neurons can affect how mice respond to rewards or punishments in both the short and long term . Future experiments should aim to understand the diversity of the signals that Cohen et al . observed , and to determine how these signals are used to drive behavior . Ultimately , understanding how neural circuits made up of different types of cells work may aid in understanding the neural basis of behavior .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Serotonergic neurons signal reward and punishment on multiple timescales
CDK9 is the kinase subunit of positive transcription elongation factor b ( P-TEFb ) that enables RNA polymerase ( Pol ) II's transition from promoter-proximal pausing to productive elongation . Although considerable interest exists in CDK9 as a therapeutic target , little progress has been made due to lack of highly selective inhibitors . Here , we describe the development of i-CDK9 as such an inhibitor that potently suppresses CDK9 phosphorylation of substrates and causes genome-wide Pol II pausing . While most genes experience reduced expression , MYC and other primary response genes increase expression upon sustained i-CDK9 treatment . Essential for this increase , the bromodomain protein BRD4 captures P-TEFb from 7SK snRNP to deliver to target genes and also enhances CDK9's activity and resistance to inhibition . Because the i-CDK9-induced MYC expression and binding to P-TEFb compensate for P-TEFb's loss of activity , only simultaneously inhibiting CDK9 and MYC/BRD4 can efficiently induce growth arrest and apoptosis of cancer cells , suggesting the potential of a combinatorial treatment strategy . The proper control of eukaryotic gene expression is fundamental for normal development and cellular response to environmental challenges . The control frequently occurs at the level of transcription , where RNA polymerase ( Pol ) II is employed to execute a series of interconnected stages that collectively constitute the transcription cycle . In the past , the early stages of this cycle involving the recruitment of Pol II to gene promoters and assembly of active pre-initiation complexes were considered the primary step where transcription is controlled ( Kuras and Struhl , 1999; Ptashne , 2005 ) . However , recent evidence indicates that the subsequent stages are also frequently targeted to regulate gene expression . For example , genome-wide analyses from Drosophila to mammals have shown that promoter-proximal pausing of Pol II is a prevalent feature of many genes and that the regulated release of Pol II is essential for synchrony and robustness of their induction ( Guenther et al . , 2007; Muse et al . , 2007; Zeitlinger et al . , 2007; Levine , 2011; Zhou et al . , 2012 ) . During transcription , the extensive and dynamic modifications of the Pol II C-terminal domain ( CTD ) have been linked to specific stages of the transcription cycle and mRNA processing . Among these , the CTD Serine-2 phosphorylation , which is a hallmark of productive elongation and RNA processing , is catalyzed by the positive transcription elongation factor b ( P-TEFb ) , which is composed of CDK9 and its cyclin partner T1 ( CycT1 ) or the minor forms T2a and T2b . Additionally , P-TEFb also phosphorylates the SPT5 subunit of DSIF and the NelfE subunit of NELF , which antagonizes the inhibitory actions of these two negative elongation factors and promotes the release of paused Pol II and transition into productive elongation ( Zhou et al . , 2012 ) . The importance of P-TEFb in transcriptional elongation requires that its activity be tightly controlled in the cell . Indeed , under normal growth conditions , the majority of P-TEFb is sequestered in the inactive 7SK snRNP , in which the CDK9 kinase activity is suppressed by HEXIM1 or 2 in a 7SK snRNA-dependent manner ( Nguyen et al . , 2001; Yang et al . , 2001; Yik et al . , 2003 ) . The remaining P-TEFb is catalytically active and present in a BRD4-containing complex and the super elongation complex ( SEC ) ( Zhou et al . , 2012 ) . In the former , the BET bromodomain protein BRD4 serves to recruit P-TEFb to the promoters of many primary response genes ( PRGs ) through binding to acetylated chromatin or the transcriptional mediator complex ( Jang et al . , 2005; Yang et al . , 2005 , 2008 ) . The SEC , on the other hand , is a target of the Tat protein encoded by the HIV-1 virus or the MLL ( mixed lineage leukemia ) fusion proteins created by chromosomal translocations to stimulate transcriptional elongation of HIV-1 and MLL-target genes , respectively ( Mueller et al . , 2009; He et al . , 2010; Lin et al . , 2010; Sobhian et al . , 2010; Yokoyama et al . , 2010 ) . A number of reagents and conditions that can globally impact growth and/or induce stress response have been shown to cause the release of P-TEFb from 7SK snRNP and formation of the BRD4-P-TEFb complex for stimulation of transcriptional elongation ( Zhou and Yik , 2006; Zhou et al . , 2012 ) . In HIV-1 infected cells , however , Tat has been shown to directly extract P-TEFb from 7SK snRNP to assemble the Tat-SEC complex on the viral promoter ( Barboric et al . , 2007; Sedore et al . , 2007; Lu et al . , 2014 ) . Multiple lines of evidence support the notion that the 7SK snRNP represents a cellular reservoir of unused P-TEFb activity , which can be withdrawn in response to various signals to form active P-TEFb complexes for activation of cellular and viral genes ( Zhou et al . , 2012; Lu et al . , 2013 ) . The proto-oncogene MYC occupies a central position downstream of many growth-promoting signal transduction pathways . As an immediate early response gene activated by many membrane-associated ligand–receptor complexes , MYC links growth factor stimulation to regulated cellular proliferation and cell cycle progression under normal conditions ( Levens , 2013 ) . Because of this property , it is one of the most frequently amplified genes in tumors , a major genetic change that leads to uncontrolled proliferation of cancer cells ( Dang , 2012 ) . The expression of MYC is normally controlled at almost every possible level to achieve a proper concentration of the protein for optimal cell growth . Prior to our current understanding of the pervasiveness of elongation control , MYC was in fact one of the first few cellular genes found to be regulated by this particular mechanism ( Bentley and Groudine , 1986 ) . Recent studies employing the BET bromodomain inhibitors such as JQ1 and iBET-151 have provided fresh mechanistic insights into how this is accomplished ( Filippakopoulos and Knapp , 2014; Shi and Vakoc , 2014 ) . The published data indicate that BRD4 and its recruitment of P-TEFb to the MYC promoter region play a particularly important role in MYC expression in human cancer cells . JQ1 and iBET-151 are shown to bind to the BRD4 bromodomains and displace the BRD4-P-TEFb complex from acetylated chromatin to inhibit MYC transcription , which in turn induces differentiation and growth arrest of cancer cells ( e . g . , acute myeloid leukemia , multiple myeloma , and Burkitt's lymphoma ) that are addicted to MYC ( Filippakopoulos and Knapp , 2014; Shi and Vakoc , 2014 ) . Because the inhibition of BRD4-mediated recruitment P-TEFb to MYC and other PRGs by JQ1 and iBET-151 can suppress cancer growth and progression , it is tempting to speculate that the direct inhibition of P-TEFb itself will produce a similar or perhaps even more focused effect . In support of this idea , a recent study implicates CDK9 inhibition as an effective therapeutic strategy for MYC-overexpressing liver tumors ( Huang et al . , 2014 ) . Although small molecule pan CDK inhibitors ( e . g . , flavopiridol and SNS-032 ) with potent anti-CDK9 activities already exist and have been valuable tools for exploring the functions of P-TEFb ( Chao et al . , 2000; Chen et al . , 2009 ) , they display pleiotropic effects indicative of interference with other cellular enzymes/pathways ( Bible et al . , 2000 ) . At this moment , there are few verified and well-characterized compounds that show high selectivity and potent inhibitory activity against CDK9 . In light of the tremendous interest in P-TEFb as a potential therapeutic target , we describe here the development and characterization of a potent and selective CDK9 inhibitor called i-CDK9 that efficiently suppresses P-TEFb's phosphorylation of the Pol II CTD and the DSIF subunit SPT5 and causes widespread Pol II pausing at gene promoters . Although the vast majority of genes display a drastic i-CDK9-induced reduction in gene expression , a small group of genes show a surprising increase in expression after the drug treatment , and the proto-oncogene MYC and several other key PRGs are among them . In the current study , we explore the molecular mechanism as well as biological significance of i-CDK9-induced MYC expression . Our data reveal the essential dual roles played by BRD4 in MYC induction and indicate that the elevated expression in i-CDK9-treated cells is part of the cellular compensation for the loss of CDK9 . Because of this compensatory mechanism , our data demonstrate that the simultaneous inhibition of both CDK9's catalytic activity and MYC's expression or function causes synergistic induction of growth arrest and apoptosis of cancer cells . High throughput screening combined with crystal structure-enabled lead compound optimization has led to the identification of a novel and potent CDK9-selective inhibitor called i-CDK9 ( the screen and a co-crystal structure of i-CDK9 bound to CDK9 will be described elsewhere ) . i-CDK9 has a N2′- ( trans-4-aminocyclohexyl ) -5′-chloro-N6- ( 3-fluorobenzyl ) -2 , 4′-bipyridine-2′ , 6-diamine scaffold that is structurally distinct from flavopiridol and all the other known non-selective CDK inhibitors ( Figure 1A ) . It occupies the ATP-binding pocket of the CDK9 kinase domain as revealed by a co-crystal structure solved at 2 . 6 Å resolution ( manuscript in preparation ) . 10 . 7554/eLife . 06535 . 003Figure 1 . i-CDK9 is a potent and selective CDK9 inhibitor that elicits cellular responses indicative of P-TEFb inhibition . ( A and B ) Structures and selectivity profiles of i-CDK9 ( A ) and falvopiridol ( B ) . The numbers refer to the concentrations ( µM ) of the two compounds that resulted in 50% inhibition of the enzymatic activity of the indicated CDK–cyclin pairs in the AlphaScreen ( PerkinElmer ) -based kinase assay ( A ) or 50% inhibition of the bindings of the indicated CDKs to the immobilized ligands in the KINOMEscan platform ( B ) . ( C ) In vitro kinase reactions containing affinity-purified CDK9-F-CycT or CDK12-F-CycK and GST-CTD as a substrate were conducted in the presence of the indicated concentrations of i-CDK9 . pSer2 and the Flag-tagged kinase in each reaction were detected by Western blotting with anti-pSer2 and anti-Flag antibodies , respectively . ( D ) HeLa cells were treated for 8 hr with DMSO or the indicated concentrations of i-CDK9 . Total cell lysates were examined by immunoblotting for the proteins labeled on the left . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 00310 . 7554/eLife . 06535 . 004Figure 1—source data 1 . Selectivity profile of i-CDK9 . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 00410 . 7554/eLife . 06535 . 005Figure 1—figure supplement 1 . Recombinant CDK12-CycK is less sensitive to inhibition by i-CDK9 . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 005 Using an in vitro AlphaScreen ( PerkinElmer , Inc ) -based kinase assay , i-CDK9 was shown to potently inhibit the CDK9-CycT1 catalytic activity with an IC50 value below the detection limit of 0 . 0004 µM ( Figure 1B ) . In the same assay , flavopiridol showed an IC50 value of 0 . 007 µM . Furthermore , compared to its inhibition of the CDK9-CycT1 kinase activity , i-CDK9 exhibited at least 600-fold lower activity toward CDK1-CycB , CDK2-CycA , CDK4-CycD1 , CDK7-CycH-MAT1 and CDK8-CycC ( Figure 1B ) . In contrast , flavopiridol displayed only 2 . 9- to 28 . 6-fold lower activity toward the same CDK-cyclin pairs compared to CDK9-CycT1 ( Figure 1B ) . As there are no commercial or in-house kinase assays available for CDK3 , CDK5 , CDK11 and CDK13 , the inhibitory abilities of i-CDK9 toward these CDKs were thus evaluated in the DiscoveRx KINOMEscan assay , which is based on a proprietary active site-directed competition-binding platform ( Fabian et al . , 2005 ) . At 1 µM , i-CDK9 almost completely blocked the binding of CDK9 to the immobilized ligands ( only 5 . 1% CDK9 captured on solid support; Figure 1A ) , but displayed essentially no effect on CDK5 ( 100% binding to immobilized ligands ) , CDK11 ( 94% ) , CDK13 ( 100% ) and only a partial inhibition of the binding of CDK3 ( 57% ) . Among all members of the CDK super family , CDK12 deserves special attention because of its reported phosphorylation of the Pol II CTD on Ser2 , an ability that is shared with CDK9 ( Bartkowiak et al . , 2010 ) . To determine whether i-CDK9 also affects CDK12 kinase activity , we examined the abilities of affinity-purified Flag-tagged CDK12 ( CDK12-F ) , CDK9 ( CDK9-F ) and their associated cyclin partners to phosphorylate GST-CTD in the presence of increasing amounts of i-CDK9 . Phosphorylation of Ser2 ( pSer2 ) was detected by Western blotting with a specific antibody . While Ser2 phosphorylation by CDK9-F was efficiently inhibited by i-CDK9 with an estimated IC50 of ∼2 nM under the current experimental conditions , no obvious inhibition of CDK12-F was detected even at 80 nM of the inhibitor ( Figure 1C ) . To confirm this result using materials from a different source , the sensitivity of the two kinases to i-CDK9 was also compared in reactions containing the baculovirus-produced recombinant CDK9-CycT1 and CDK12-CycK ( SignalChem ) . While Ser2 phosphorylation by CDK9-CycT1 was mostly inhibited by 80 nM i-CDK9 , CDK12-CycK was not significantly inhibited until 640-1200 nM i-CDK9 was added into the reactions ( Figure 1—figure supplement 1 ) . Thus , between CDK9 and CDK12 , i-CDK9 displayed markedly higher selectivity against the former . It is interesting to note that a similar finding has also been made with the pan-CDK inhibitor flavopiridol ( Bosken et al . , 2014 ) . Finally , the selectivity profile of i-CDK9 toward other non-CDK kinases was also assessed using the KINOMEscan platform . Among the >400 kinases evaluated , only 9 ( highlighted yellow in Figure 1—source data 1 ) consistently showed less than 40% binding to the immobilized ligands in the presence of 1 or 10 µM of i-CDK9 . Except for CLK4 , for which no suitable assay was available , the rest of the kinases plus a selected group of others were re-examined for the concentration-dependent effect of i-CDK9 in either functional kinase assays or binding assays . Again , i-CDK9 was found to display great selectivity for CDK9 with more than 100-fold difference detected between CDK9 and the next two best targets DYRK1A and DYRK1B ( IC50 < 0 . 0004 µM for CDK9 vs IC50 = 0 . 055 and 0 . 047 for DYRK1A and B , respectively; column D in Figure 1—source data 1 ) . The fact that DYRK1A can also be inhibited by i-CDK9 albeit with reduced efficiency suggests that the structure and/or function of this kinase may resemble that of CDK9 to a certain degree . Indeed , it has recently been reported that DYRK1A acts as a Pol II CTD Kinase at its target gene promoters ( Di Vona et al . , 2015 ) . In contrast to i-CDK9 , flavopiridol displayed strong but non-selective interactions with a far greater number of kinases in the DiscoveRx panel ( see http://www . discoverx . com/ReferenceTreeImages/Flavopiridol . htm ) , some of which ( e . g . , with ICK , CDK4-CycD1 and CDKL5 ) had affinities that are similar or even higher than that for CDK9 . Consistent with its specific binding to and potent inhibition of purified CDK9-CycT1 in vitro , i-CDK9 markedly reduced the CDK9-mediated pSer2 in the Pol II CTD and Thr775 ( pThr775 ) in the DSIF subunit SPT5 in a dose-dependent manner in HeLa cells ( Figure 1D ) . Inhibition of phosphorylation of both Pol II CTD and SPT5 is expected to block transcriptional elongation , which will effectively decrease the production of labile proteins from short-lived transcripts . Indeed , correlating with the decrease in cellular levels of pSer2 and pThr775 , i-CDK9 markedly down-regulated the expression of the short-lived anti-apoptotic protein MCL-1 and at the same time induced proteolytic cleavage of poly ( ADP-ribose ) polymerase ( PARP ) , which is considered a major hallmark of apoptosis and caspase activation ( Figure 1D ) . Collectively , these data demonstrate that i-CDK9 is a highly potent and selective inhibitor of CDK9 that is able to elicit cellular responses indicative of P-TEFb inhibition . To investigate the global impact of CDK9 inhibition on transcriptional elongation by Pol II , chromatin immunoprecipitation followed by massively parallel DNA sequencing ( ChIP-seq ) was performed to examine the genome-wide occupancy of Pol II before and after HeLa cells were treated with i-CDK9 . Similar to the situation reported in other cell types ( Zeitlinger et al . , 2007; Rahl et al . , 2010; Liu et al . , 2013 ) , among the total of 11 , 197 genes with detectable Pol peaks , 7805 ( 70% ) showed a traveling ratio ( TR; also called the pausing index; ( Zeitlinger et al . , 2007; Rahl et al . , 2010 ) greater than 2 . 0 in the control DMSO-treated HeLa cells . Since TR is defined as the relative ratio of Pol II density in the promoter-proximal region vs the gene body ( Figure 2A ) , the above numbers suggest that the majority of the genes were experiencing promoter-proximal pausing by Pol II in the control cells . Importantly , upon treatment with i-CDK9 for 2 and 8 hr , 6339 ( 56 . 6% ) and 7658 ( 68 . 4% ) genes displayed an increase of at least 1 . 5-fold in Pol II TR , respectively ( Figure 2B ) , indicating significantly elevated promoter-proximal pausing by Pol II upon CDK9 inhibition . Figure 2C shows four representative genes ( SMUG1 , TMEM115 , SEC13 and CSNK1D ) with significantly elevated Pol II TR upon 8 hr of i-CDK9 treatment . 10 . 7554/eLife . 06535 . 006Figure 2 . i-CDK9 causes widespread promoter-proximal pausing by Pol II and the biggest decrease in expression for genes involved in regulation of transcription and RNA metabolic process . ( A ) Schematic diagram illustrating calculation of the Pol II traveling ratio ( TR ) . ( B ) Distribution of Pol II-bound genes with a given TR as determined by ChIP-seq under the various conditions as indicated . The pie charts below describe the percentages of genes with 1 . 5-fold increase , 1 . 5-fold decrease , or no change in TR after exposure to i-CDK9 for 2 or 8 hr as compared to DMSO . ( C ) Occupancy of Pol II as revealed by ChIP-seq across 4 representative genes with increased TR after CDK9 inhibition . The read coverage is shown for the entire gene plus a margin on either side equal to 7% of the gene length . ( D ) Distribution of Pol II-bound genes with a given TR as determined by ChIP-seq . The genes are grouped by expression changes induced by i-CDK9 . Up: the 138 genes that showed at least twofold increase in expression after exposure to i-CDK9 or 8 hr . Other: genes whose expression was either unaffected or affected less than twofold by i-CDK9 . ( E ) Enrichment of GO biological processes by DAVID . Only top 4 gene sets are shown for top 500 genes with the biggest increase in TR at 8 hr treatment with i-CDK9 ( top ) and top 500 genes with the largest decrease in gene expression at 8 hr i-CDK9 treatment ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 006 To investigate how the changes in TR value may correlate with changes in expression of the genes , we conducted DNA microarray studies to determine the effect of selective CDK9 inhibition on global gene transcription . A comparison between the two data sets indicates that TR values for those 138 genes whose expression was up-regulated at least twofold by i-CDK9 were significantly less affected compared to the rest of the genes ( Figure 2D ) . Furthermore , the top 4 gene sets with the largest increase in TR show an enrichment in biological processes ranging from RNA processing , DNA and non-coding RNA metabolic process to transcription ( Figure 2E ) . Largely consistent with this TR-based analysis , the top 4 sets , whose expression had the biggest drop upon i-CDK9 treatment , are populated by genes involved in regulation of transcription and RNA metabolic process , indicating that these processes and their control are particularly susceptible to inhibition of CDK9 . Among the small group of genes that showed at least twofold increase in expression after 8 hr of i-CDK9 treatment , the proto-oncogene MYC has caught our special attention because of its extreme importance in cell growth control and oncogenic transformation and also the fact that its expression has been shown to depend on CDK9 ( Kanazawa et al . , 2003; Rahl et al . , 2010; Huang et al . , 2014 ) . Microarray analysis performed at three different time points ( 2 , 8 and 16 hr ) indicates that MYC expression in i-CDK9-treated HeLa cells displayed an interesting biphasic responses with a small initial decrease ( within 2 hr ) followed by a dramatic rebound ( around 8 hr and continue at 16 hr ) in the presence of continuous CDK9 inhibition by i-CDK9 ( Figure 3A ) . This result was further confirmed by qRT-PCR analysis of MYC mRNA levels in cells treated with i-CDK9 or DMSO ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 06535 . 007Figure 3 . Induction of MYC mRNA production in response to sustained inhibition of CDK9 by i-CDK9 and the requirement of MYC's natural genomic structure in this process . ( A ) The indicated tumor cell lines were treated with i-CDK9 ( 0 . 5 μM ) for 2–16 hr and the MYC mRNA levels , which were divided by those in the DMSO-treated cells and averaged from three independent replicates of DNA microarray analysis , were shown as log2 values . ( B ) HeLa cells were treated with the indicated concentrations of i-CDK9 for 8 hr and the various proteins in the total cell lysates were detected by immunoblotting as indicated . Quantifications of the levels of MYC protein and Ser2-phosphoryalted Pol II ( pSer2 ) in the lysates were shown above and below the immunoblots , respectively . ( C ) HeLa cells were treated with 0 . 3 µM i-CDK9 for the indicated number of hr and cell lysates were obtained and analyzed as in B . ( D and E ) HeLa cells were treated with either 0 . 3 µM flavopiridol for the indicated time periods ( D ) or 8 hr with the indicated concentrations of flavopiridol ( E ) and analyzed by immunoblotting as in B . ( F ) Cells were pretreated with ( + ) or without ( − ) cycloheximide ( CHX ) prior to incubation with i-CDK9 . The levels of the indicated proteins were examined by immunoblotting . ( G ) A HeLa-based cell line containing the stably transfected , doxycycline ( Dox ) -inducible MYC-F-expressing plasmid driven by the CMV promoter was pretreated with ( + ) or without ( − ) Dox prior to incubation with i-CDK9 or DMSO . MYC and MYC-F expressed from the endogenous MYC locus and the transfected plasmid , respectively , were detected by immunoblotting . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 00710 . 7554/eLife . 06535 . 008Figure 3—figure supplement 1 . Biphasic response of MYC mRNA production throughout the course of CDK9 inhibition by i-CDK9 . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 00810 . 7554/eLife . 06535 . 009Figure 3—Figure supplement 2 . HEXIM1 expression is continuously suppressed throughout the entire course of i-CDK9 treatment of five different tumor cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 009 In addition to HeLa cells , a similar biphasic expression pattern was also observed in four other cell lines representing a wide spectrum of human cancers and MYC amplification states ( Figure 3A; http://www . broadinstitute . org/ccle/home ) . Notably , the extent of the initial decrease and then subsequent induction of MYC mRNA production varied among the cell lines . For example , after 2 hr of i-CDK9 ( 0 . 5 µM ) treatment , the MYC mRNA levels in U87MG ( MYC copy number [CN] = 2 ) and HeLa ( CN = 4 ) cells showed only a slight decrease compared to the DMSO control , whereas they were significantly reduced in A375 ( CN = 2 ) , NCIH441 ( CN = 8 ) , and A2058 ( CN = 4 ) cells . Furthermore , the prolonged treatment for 8–16 hr caused the MYC mRNA to reach levels much higher than those in the control cells in all cell lines except NCIH441 ( Figure 3A ) . Despite these variations , which could well be caused by differences in MYC mRNA stability in difference cell lines , it is clear that the induction of MYC expression in response to prolonged CDK9 inhibition by i-CDK9 is a general phenomenon likely caused by a common mechanism independent of the MYC amplification/expression levels . In contrast to MYC , another short-lived gene HEXIM1 , which was previously shown to depend on active CDK9 for expression ( He et al . , 2006 ) , was continuously suppressed throughout the entire 16 hr of i-CDK9 treatment in all five cell lines ( Figure 3—figure supplement 2 ) . How can the sustained global inhibition of P-TEFb , a general transcription elongation factor , cause such a surprising and dramatic induction in MYC's expression ? We decided to use HeLa cells , which are a convenient and reliable model system for conducting many biochemical analyses , to determine the molecular mechanism behind this phenomenon . While most of the experiments were done in this cell line , a few key conclusions were also confirmed in other cell types ( see below ) . As the initial reduction in MYC mRNA level was not prominent in HeLa cells , we focused our attention on the major up-regulation of MYC expression detected upon 8–16 hr of treatment with i-CDK9 . First , the induction of MYC protein level was confirmed in HeLa cells treated with i-CDK9 . Upon exposure to just 0 . 1 µM i-CDK9 , the MYC level increased 3 . 6-fold whereas the global pSer2 level was only decreased by 23% ( Figure 3B ) . It took 1 . 0 µM i-CDK9 to cause up to 85% reduction in pSer2 . In a separate time-course analysis , the 8-hr incubation with i-CDK9 ( 0 . 3 μM ) induced the MYC level by 3 . 3-fold , whereas the same condition reduced the pSer2 level by only 40% ( Figure 3C ) . Thus , MYC expression was highly responsive to i-CDK9 treatment and occurred at a time point earlier and drug concentration lower than those required to achieve a significant suppression of Ser2 phosphorylation on the Pol II CTD . Notably , a very similar MYC induction pattern was also produced by the pan-CDK inhibitor flavopiridol . Just like i-CDK9 , flavopiridol also significantly elevated the MYC protein level and this effect required a shorter exposure time ( e . g . , 4 hr ) and lower drug concentration ( e . g . , 0 . 1 μM ) than those needed to fully reduce the global pSer2 level ( Figure 3D , E ) . The nearly identical activities displayed by i-CDK9 and flavopiridol , which are very different in their structures , suggest that the inhibition of CDK9 but not some other unknown enzymes was the cause of MYC induction . To determine whether the induction of MYC expression by i-CDK9 requires continuous protein synthesis , we treated cells with or without the protein synthesis inhibitor cycloheximide ( CHX ) prior to incubation with i-CDK9 . Without the pre-treatment , i-CDK9 significantly increased the MYC protein level as expected ( Figure 3F , lanes 1–2 ) . In the presence of CHX , however , MYC production was essentially wiped out in both the i-CDK9-treated and i-CDK9-untreated cells ( lanes 3 and 4 ) . This result , together with the above demonstration that sustained i-CDK9 treatment dramatically increased the MYC mRNA level ( Figure 3A ) , indicates that MYC is a labile protein and that its prominent accumulation in i-CDK9-treated cells was most likely due to its enhanced mRNA production but not protein stability . Interestingly , although i-CDK9 efficiently induced expression from the endogenous MYC locus , it failed to activate production of the Flag-tagged MYC ( MYC-F ) from a stably transfected , doxycycline-inducible expression construct driven by the CMV promoter ( Figure 3G ) , indicating that i-CDK9 cannot activate MYC transcription from this heterologous promoter . A number of stress-inducing agents/conditions are known to cause the release of P-TEFb from 7SK snRNP , which serves as the principal reservoir of uncommitted P-TEFb activity in the nucleus ( reviewed in Zhou et al . , 2012 ) . The released P-TEFb then joins the bromodomain protein BRD4 to form the BRD4-P-TEFb complex for activation of many PRGs ( reviewed in Zhou et al . , 2012 ) . To determine whether i-CDK9 could also induce the transfer of P-TEFb from 7SK snRNP to BRD4 , we examined the interactions of immunoprecipitated Flag-tagged CDK9 ( CDK9-F ) with HEXIM1 , a signature component of 7SK snRNP , and BRD4 by Western blotting ( Figure 4A ) . CDK9-F was stably expressed in the HeLa-based F1C2 cells ( Yang et al . , 2001 ) that were treated with a wide range of i-CDK9 concentrations for 8 hr . 10 . 7554/eLife . 06535 . 010Figure 4 . Activation of MYC transcription by i-CDK9 depends on induced transfer of kinase-active P-TEFb from 7SK snRNP to BRD4 , binding of the BRD4-P-TEFb complex to acetylated MYC chromatin template , and BRD4-mediated increase in CDK9's catalytic activity and resistance to inhibition . ( A ) The HeLa-based F1C2 cells stably expressing CDK9-F were incubated with the indicated concentrations of i-CDK9 . Nuclear extracts ( NE ) and the anti-CDK9-F immunoprecipitates ( IP ) derived from NE were analyzed by immunoblotting to detect the indicated proteins . ( B and C ) Lysates of HeLa cells expressing the BRD4-specific shRNA ( shBRD4; B ) or pretreated with JQ1 or the control enantiomer ( ctl . ; C ) were incubated with ( + ) or without ( − ) i-CDK9 and analyzed by immunoblotting for the indicated proteins . ( D–H ) Cells were first transfected with plasmids expressing F-PID ( D , E and F ) , D167N-F ( D ) , or shCycT1 ( G ) or transfected with siRNAs specific for CDK9 ( G ) or CDK12 ( H ) and then treated with i-CDK9 or DMSO ( − ) . NE ( D , G and H ) and immunoprecipitates ( IP ) obtained from NE with anti-Flag mAb ( E ) or anti-CDK9 antibodies or rabbit total IgG ( F ) were examined by immunoblotting for the indicated proteins . The relative CDK12 mRNA levels were analyzed by qRT-PCR at the bottom in H , with the level in cells transfected with a control siRNA set to 1 . ( I and J ) In vitro kinase reactions containing a synthetic Pol II CTD peptide ( CDK7tide ) as the substrate and CDK9-CycT1 ( Invitrogen ) as the kinase were conducted in the presence of the indicated amounts of WT or ∆PID BRD4 . Phosphorylation of the peptide was measured over the indicated time periods and plotted in I , with the error bars representing mean ± SD from three independent experiments . The indicated amounts of i-CDK9 were added to the reactions in J , and its inhibition of CDK9 phosphorylation of the peptide was measured and plotted with the inhibitory IC50 shown . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 01010 . 7554/eLife . 06535 . 011Figure 4—figure supplement 1 . i-CDK9 ( 0 . 3 μM ) induces disruption of 7SK snRNP at a time point much earlier than that required to cause about 50% reduction in global pSer2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 01110 . 7554/eLife . 06535 . 012Figure 4—figure supplement 2 . JQ1 decreases associations of both BRD4 and CDK9 with the MYC locus . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 01210 . 7554/eLife . 06535 . 013Figure 4—figure supplement 3 . JQ1 blocks the i-CDK9-induced MYC expression in H1792 and A2058 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 01310 . 7554/eLife . 06535 . 014Figure 4—figure supplement 4 . MYC induction by 0 . 3 μM i-CDK9 can be subsequently shut off by 2 μM of the drug . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 01410 . 7554/eLife . 06535 . 015Figure 4—figure supplement 5 . Examination of the purity and concentrations of WT and ∆PID BRD4 used in the CDK9 kinase assay . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 015 Indeed , i-CDK9 gradually decreased the amounts of HEXIM1 bound to CDK9-F and at the same time increased the BRD4-CDK9-F binding in a dosage-dependent manner ( Figure 4A ) . Just like the induction of MYC expression ( Figure 3B ) , the disruption of 7SK snRNP and formation of BRD4-P-TEFb were highly sensitive to i-CDK9 and required as little as 0 . 1 μM of the drug ( Figure 4A , compare lane 8 to lane 6 ) . At this concentration , i-CDK9 only slightly decreased the global pSer2 level ( Figure 4A , compare between lanes 1 and 3 ) but began to markedly induce MYC expression ( Figure 3B ) . In a time course analysis , i-CDK9 ( 0 . 3 μM ) -induced disruption of 7SK snRNP at a time point ( 1 hr ) that was much earlier than that required to cause about 50% reduction in global pSer2 ( 8 hr; Figure 4—figure supplement 1 ) . Together , these results reveal the remarkable efficiency with which i-CDK9 induces the transfer of P-TEFb from 7SK snRNP to BRD4 . Given the above demonstration that the i-CDK9-induced MYC transcription occurred under the same conditions that stimulated the transfer of P-TEFb from 7SK snRNP to BRD4 , it is important to confirm that BRD4 is indeed required for the MYC induction . Toward this goal , BRD4 knockdown ( KD ) was performed with a specific shRNA ( shBRD4 ) , whose expression was induced by Cre recombinase ( Yang et al . , 2008 ) . The loss of BRD4 , verified by Western blotting , dramatically reduced both basal and the i-CDK9-induced MYC protein production ( Figure 4B ) . To determine if the binding of BRD4 to acetylated chromatin is required for i-CDK9 to induce MYC expression , we tested whether the BET bromodomain inhibitor JQ1 could block the induction . JQ1 is known to competitively bind to the acetyl-lysine recognition pocket within BRD4's bromodomains , leading to the dissociation of the BRD4-P-TEFb complex from acetylated chromatin ( Zuber et al . , 2011; Li et al . , 2013 ) . Consistent with the results obtained in other cell types ( Mertz et al . , 2011; Zuber et al . , 2011 ) , JQ1 decreased the associations of both BRD4 and CDK9 with the MYC locus in HeLa cells ( Figure 4—figure supplement 2 ) . More importantly , JQ1 also completely abolished the MYC induction by i-CDK9 in HeLa ( Figure 4C ) as well as the lung cancer cell line H1792 and the melanoma cell line A2058 ( Figure 4—figure supplement 3 ) , whereas the control enantiomer was ineffective in this regard . These observations , in conjunction with the shBRD4 result above , confirm that the interaction between BRD4 and acetylated chromatin is required for i-CDK9 to induce MYC expression in diverse cell types . We next asked whether the enhanced interaction between BRD4 and P-TEFb as a result of i-CDK9 treatment ( Figure 4A ) is also required for the MYC induction . To this end , the ability of i-CDK9 to induce MYC protein production was tested in the presence of the overexpressed P-TEFb-interacting domain ( PID; aa1209-1362 ) of BRD4 ( Bisgrove et al . , 2007 ) . Western analyses indicate that the PID efficiently suppressed the induction ( Figure 4D , lanes 3 and 4 ) , likely due to its own strong interaction with CDK9 and CycT1 in i-CDK9-treated cells ( Figure 4E ) , which in turn interfered with the interaction of endogenous BRD4 with P-TEFb ( Figure 4F ) . Not only was the enhanced physical interaction between P-TEFb and BRD4 essential for the i-CDK9-induced MYC expression , more importantly , the wild-type ( WT ) kinase activity of the BRD4-bound P-TEFb was also required . This point is illustrated by the demonstration that the overexpressed kinase-inactive CDK9 mutant D167N , which binds to BRD4-like WT CDK9 ( Yang et al . , 2005 ) and acts dominant-negatively to suppress the activity of endogenous CDK9 ( Garber et al . , 2000 ) , prevented i-CDK9 from inducing MYC ( Figure 4D , lanes 5 and 6 ) . The dependence on catalytically active P-TEFb for MYC induction was further demonstrated by the observation that the initial MYC induction by a low level ( e . g . , 0 . 3 μM ) of i-CDK9 was subsequently shut off by a higher i-CDK9 concentration ( e . g . , 2 μM , see Figure 4—figure supplement 4 ) . No MYC induction was detected also when 2 μM i-CDK9 was added into the medium at time zero ( see Figure 7A below ) . Finally , the RNAi-mediated KD of cellular CDK9 and CycT1 levels also completely abolished the i-CDK9-induced MYC expression ( Figure 4G ) . In contrast , KD of CDK12 , another reported CTD kinase , failed to prevent i-CDK9 from inducing MYC ( Figure 4H ) . Taken together , these data strongly support the notion that the i-CDK9 induction of MYC expression depends on the enhanced interaction of BRD4 with catalytically active P-TEFb released from 7SK snRNP as well as the binding of the BRD4-P-TEFb complex to acetylated MYC chromatin template . The above data have revealed an apparent paradox concerning the role of CDK9 kinase during i-CDK9 induction of MYC expression . Although the induction was caused by a highly specific CDK9 inhibitor , the actual process was found to require the interaction of BRD4 with catalytically active CDK9 . One possible explanation for these seemingly contradictory observations is the above demonstrations that the MYC induction was highly sensitive and occurred at a i-CDK9 concentration lower and time point earlier than those required to completely abolish the nuclear CDK9 kinase activity . In addition to this kinetic advantage displayed by the very sensitive MYC induction process , we also investigated whether the i-CDK9-induced BRD4-P-TEFb binding could also directly affect the kinase activity of CDK9 . To this end , the ability of recombinant CDK9-CycT1 ( Invitrogen ) to phosphorylate a synthetic Pol II CTD peptide ( termed CDK7tide; Bio-Synthesis , Inc . ) was measured in the presence of either WT BRD4 or a BRD4 mutant lacking the C-terminal P-TEFb-interacting domain ( ∆PID ) . The BRD4 proteins were affinity-purified from transfected HEK293T cells under highly stringent conditions to strip away their associated factors ( Figure 4—figure supplement 5 ) . While WT BRD4 was able to increase CDK9 kinase activity dose dependently up to 2 . 2-fold , ∆PID lacked this ability ( Figure 4I ) , indicating that the physical interaction between BRD4 and P-TEFb was essential for the elevated CDK9 activity . It is worth pointing out that different from a recent study reporting that BRD4 is an atypical kinase that can directly phosphorylate the Pol II CTD on Ser2 ( Devaiah et al . , 2012 ) , neither WT nor ∆PID BRD4 alone was able to cause phosphorylation of the CTD peptide in the absence of P-TEFb ( Figure 4I ) . In addition to increasing CDK9's catalytic activity , BRD4 was also found to render CDK9 less sensitive to inhibition by i-CDK9 , whereas BRD4∆PID was less effective in this regard ( Figure 4J ) . For example , in the absence of BRD4 , it took only 4 . 4 ± 0 . 7 nM i-CDK9 to achieve 50% inhibition of CDK9 ( i . e . , IC50 = 4 . 4 ± 0 . 7 nM ) . However , the addition of WT BRD4 into the reaction protected CDK9 against i-CDK9 and increased IC50 to 14 . 7 ± 2 . 1 nM . BRD4∆PID , on the other hand , displayed a much weaker effect by changing the IC50 value to 7 . 2 ± 0 . 6 nM ( Figure 4J ) . These data indicate that the i-CDK9-mediated MYC induction benefited not only from efficient release of P-TEFb from 7SK snRNP , but the drug-enhanced BRD4-P-TEFb interaction can also directly promote CDK9's kinase activity and resistance to inhibition , in addition to its recruitment of more P-TEFb to the MYC locus . To further probe the molecular basis underlying MYC induction by i-CDK9 , we performed the classic ChIP-qPCR analysis to examine the interactions of BRD4 , CDK9 , total Pol II , and the Ser2-phosphoryalted Pol II with the MYC locus before and after treatment with 0 . 3 μM i-CDK9 for 8 hr . The first noticeable major change caused by i-CDK9 is the significant increase in BRD4's occupancy across the entire MYC locus ( Figure 5A , B ) , which explains why the MYC induction was highly sensitive to the BRD4 inhibitor JQ1 ( Figure 4C ) and shBRD4 ( Figure 4B ) . Correlating with this increase and consistent with the i-CDK9-induced BRD4-P-TEFb interaction , there was also significantly elevated CDK9 binding to the MYC locus ( Figure 5B ) . Although the elevation in CDK9 binding was mostly BRD4 dependent ( Figure 5—figure supplement 1 ) , the distribution pattern of CDK9 was somewhat different from that of BRD4 ( Figure 5B ) . This difference could be caused by P-TEFb's dissociation from BRD4 and joining the Pol II elongation complex once it is recruited to the MYC chromatin template . 10 . 7554/eLife . 06535 . 016Figure 5 . Treatment with i-CDK9 ( 0 . 3 μM for 8 hr ) increases the levels of P-TEFb , BRD4 , total Pol II , Pol II with pSer2 CTD and acetyl-H3/H4 at the MYC locus . ( A ) Genomic structure of the MYC locus . Arrows indicate the positions and direction of the four MYC promoters P0 to P3 . The small horizontal bars labeled with letters A to K mark the positions of 11 amplicons generated by quantitative PCR ( qPCR ) analysis of the ChIP DNA . ( B and C ) HeLa cells were treated with either i-CDK9 or DMSO and subjected to ChIP-qPCR analysis to determine the levels of the indicated factors bound to the MYC locus . The signals were normalized to those of input; and the error bars in all panels represent mean ± SD from three independent experiments . ( D ) HeLa cells were treated with 0 . 3 μM i-CDK9 for the indicated time periods and subjected to ChIP-qPCR analysis to determine the levels of the indicated factors bound to the MYC locus at position C and the HEXIM1 locus at position L ( see Figure 5—figure supplement 2 ) . The signals were normalized to those of input; and the error bars represent mean ± SD from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 01610 . 7554/eLife . 06535 . 017Figure 5—figure supplement 1 . The i-CDK9-induced increase in CDK9's binding to the MYC locus is mostly BRD4-dependent . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 01710 . 7554/eLife . 06535 . 018Figure 5—figure supplement 2 . Treatment with i-CDK9 ( 0 . 3 μM for 8 hr ) decreases the levels of both total Pol II and Pol II with pSer2 CTD at the HEXIM1 locus . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 01810 . 7554/eLife . 06535 . 019Figure 5—figure supplement 3 . i-CDK9 does not affect the cellular levels of acetylated histones H3 and H4 . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 019 Consistent with the demonstration that the i-CDK9-induced MYC expression is due to increased transcription by Pol II ( Figure 3A , F ) , i-CDK9 also markedly enhanced the concentration of Pol II across the entire MYC locus ( Figure 5B ) . Furthermore , despite the incubation with 0 . 3 μM i-CDK9 for 8 hr , a condition that was shown above to decrease the global Ser2 phosphorylation on Pol II CTD by about 40% ( Figure 3B ) , the pSer2 level was found to still increase at the MYC locus ( Figure 5B ) . Comparing to CDK9 and total Pol II , the pSer2 distribution shifted toward the 3′ end of the gene , which is consistent with the patterns detected on many other actively transcribed genes ( Rahl et al . , 2010; Zhou et al . , 2012 ) . In contrast to the situation observed on the i-CDK9-induced MYC gene , on HEXIM1 , a gene that was permanently inhibited by i-CDK9 ( Figure 3—figure supplement 2 ) , the same treatment decreased the levels of both total Pol II and Pol II with pSer2 CTD ( Figure 5—figure supplement 2 ) . Finally , while the levels of acetylated histones H3 ( Ac-H3 ) and H4 ( Ac-H4 ) in whole cell extracts remained unchanged upon exposure to i-CDK9 ( Figure 5—figure supplement 3 ) , they were found to increase across the MYC locus especially at locations surrounding the major P1 promoter ( Figure 5C ) . Although the underlying mechanism for this increase is unclear , it explains well the drug-promoted BRD4 binding to the MYC locus . Interestingly , even at early time points of i-CDK9 treatment ( 0 , 1 and 2 hr ) when MYC transcription was yet to be induced , the acetylation state of the MYC promoter was already different from that of the HEXIM1 promoter . As shown in Figure 5D , the levels of both Ac-H3 and Ac-H4 at the MYC promoter began to increase at this early stage , with the more robust increase observed for Ac-H4 . In contrast , at the HEXIM1 promoter , i-CDK9 caused a drastic decrease in the Ac-H3 level but a small increase of Ac-H4 . As for the BRD4 level , it displayed a marked decrease at the MYC promoter at 1 hr post i-CDK9 treatment , but began to rebound by 2 hr . At the HEXIM1 promoter , however , it showed a sustained reduction throughout the entire period ( Figure 5D ) . It is likely that these early differences at the chromatin level predispose a transiently repressed gene such as MYC and a permanently repressed one such as HEXIM1 to completely different expression states later on during the i-CDK9 treatment . Taken together , the data presented thus far indicate that the i-CDK9-induced MYC expression is likely caused by a combination of events: ( 1 ) the induced transfer of P-TEFb from 7SK snRNP to BRD4; ( 2 ) BRD4's promotion of CDK9's kinase activity and resistance to inhibition; and ( 3 ) favorable chromatin changes at an early stage of the i-CDK9 treatment that lead to subsequent increase in the occupancy of the BRD4-P-TEFb complex at the MYC locus , promotion of Ser2 phosphorylation on the Pol II CTD , and finally productive transcriptional elongation . MYC is a classic example of the so-called PRGs , which are a set of genes that can be induced in response to both extracellular and intracellular signals without the requirement for de novo protein synthesis ( Fowler et al . , 2011 ) . Given the above demonstrations that BRD4 and its interaction with P-TEFb played an essential role in i-CDK9-induced MYC expression , we next investigated whether other BRD4-dependent primary response ( BDPR ) genes such as those identified in bone marrow-derived macrophages ( Figure 6A; [Hargreaves et al . , 2009] ) might behave similarly as MYC in their response to i-CDK9 . To this end , we performed gene set enrichment analysis ( GSEA ) of expression microarray data obtained from HeLa cells treated with either DMSO or i-CDK9 for 2 and 8 hr ( Figure 6B , C ) . At the 2-hr time point , GSEA reveals a marked i-CDK9-induced reduction of expression ( normalized enrichment score or NES = −2 . 28 ) for essentially all the 27 genes within the gene set ( Figure 6A , B ) . However , at the 8-hr time point , the trend was completely reversed with virtually all the genes displaying significant up-regulation by i-CDK9 ( Figure 6C; NES = +2 . 26 ) . Thus , the biphasic transcriptional signature of the curated BDPR genes is the same as that of MYC in their response to i-CDK9 . 10 . 7554/eLife . 06535 . 020Figure 6 . i-CDK9 affects the expression of other BRD4-dependent primary response genes similarly as it does to MYC . ( A ) The list of 27 curated BRD4-dependent primary response ( BDPR ) genes identified in bone marrow-derived macrophages is displayed in alphabetical order . The 23 genes in bold face type had detectable Pol II signals in HeLa cells as revealed by ChIP-seq analysis . ( B and C ) GSEA results for the 27 BDPR genes at 2 hr ( B ) and 8 hr ( C ) post CDK9 inhibition . NES: Normalized Enrichment Score; FDR: False Discovery Rate . ( D ) Distribution of Pol II-bound genes with a given TR as determined by ChIP-seq . The genes are grouped by the indicated gene types and treatment conditions . The top panel compares the 23 BDPR genes to the remaining Pol II-bound genes in the genome , and the bottom compares the BDPR genes to 23 randomly selected genes . ( E , F , G ) Occupancy of Pol II across three representative BDPR genes as revealed by ChIP-seq . The read coverage is shown for the entire gene plus a margin on either side equal to 7% of the gene length . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 020 Another interesting observation about the 23 BDPR genes that had detectable Pol II ChIP-seq signals in HeLa cells ( Figure 6A , bold face type ) is that they showed little change in their Pol II TR values after the exposure to i-CDK9 for 8 hr ( Figure 6D ) . In contrast , a significant increase in TR was found for most of the remaining ( Figure 6D upper panel ) or 23 randomly selected genes ( lower panel ) in the ChIP-seq database . A close examination of the distribution patterns of Pol II on three well-studied PRGs , MYC , FOS , and JUNB , which happen to be highly important proto-oncogenes , reveals that i-CDK9 caused a significant and fairly uniform increase in Pol II occupancy across the entire length of these genes ( Figure 6E–G ) . Notably , the distribution of Pol II on MYC as revealed by ChIP-seq is remarkably similar to that obtained by the classic ChIP-qPCR ( Figure 5B ) . These results are not only consistent with the induced expression of these genes by i-CDK9 , but also explain why their TR values underwent little change in the process . Taken in aggregate , the data so far strongly support the notion that the induced transfer of P-TEFb from 7SK snRNP to BRD4 and the BRD4-mediated protection and enhancement of P-TEFb's activity and recruitment to chromatin templates are the primary driving force behind the induction of MYC and other BDPR genes by i-CDK9 . Having identified the mechanism of MYC induction by i-CDK9 , we also wanted to determine the biological significance of this phenomenon . Given that MYC induction was triggered by inhibition of cellular CDK9 kinase , we hypothesized that MYC may normally facilitate CDK9's function and that the elevated MYC expression in i-CDK9-treated cells is therefore a cellular attempt to compensate for the loss of CDK9 . To test this hypothesis , we first determined whether the shRNA-mediated MYC KD would affect the global pSer2 levels before and after the treatment with i-CDK9 . Indeed , even though the KD was incomplete , it markedly reduced pSer2 levels in both untreated cells as well as in cells exposed to a range of i-CDK9 concentrations ( Figure 7A ) , revealing a requirement of MYC for CDK9 to phosphorylate the Pol II CTD under both normal and inhibitory conditions . 10 . 7554/eLife . 06535 . 021Figure 7 . Simultaneous inhibition of CDK9 and MYC synergistically induces growth arrest and apoptosis of cancer cells due to the fact that MYC facilitates P-TEFb phosphorylation of Pol II CTD and increases binding to BRD4-P-TEFb upon CDK9 inhibition . ( A ) Lysates of HeLa cells expressing the indicated shRNA and exposed to increasing concentrations of i-CDK9 were analyzed by immunoblotting for the indicated proteins . ( B and C ) Lysates of HeLa cells treated with the indicated drugs and their concentrations were analyzed by immunoblotting , with quantification of the pSer2 signals shown at the bottom . ( D and E ) Nuclear extracts ( NE ) of HeLa-based cells expressing MYC-F and untreated ( − ) or treated with i-CDK9 or DMSO were subjected to anti-Flag immunoprecipitation . The immunoprecipitates ( IP ) were examined by immunoblotting for the indicated proteins . ( F , G , H , and I ) HeLa ( F and H ) and H1792 ( G and I ) cells were incubated with JQ1 or i-CDK9 alone or together at various concentrations . The concentrations of each drug ( IC50 ) , either used as a single agent or in combination , that caused 50% of cells to show growth inhibition in Celltiter-Glo assay ( F and G ) or produce Caspase 3/7 ( H and I ) were plotted using the isobologram method . The dotted lines denote the IC50 values of i-CDK9 and JQ1 had the effects of the two compounds been simply additive . ( J and K ) HeLa cells were treated with the indicated concentrations of i-CDK9 plus JQ1 ( J ) or i-CDK9 plus 10 , 058-F4 ( K ) and measured by flow cytometry for propidium iodide ( PI ) -stained sub-G1 population . The error bars represent mean ± SD from three independent measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 06535 . 021 Not only did the KD of MYC expression negatively affect the ability of CDK9 to phosphorylate Ser2 , the treatment of cells with either JQ1 , the BET bromodomain inhibitor that suppressed the BRD4-dependent MYC expression ( Figure 4C ) , or 10 , 058-F4 , a small molecule inhibitor that inhibits the MYC-MAX interaction to prevent transactivation of genes targeted by this heterodimer ( Rahl et al . , 2010 ) , also slightly decreased the cellular pSer2 levels ( Figure 7B , lane 4; Figure 7C , lane 2 ) . More importantly , when JQ1 or 10 , 058-F4 was used in combination with i-CDK9 , global pSer2 dropped to levels that were much lower than those caused by i-CDK9 alone ( Figure 7B , lane 3; Figure 7C , lane 4 ) . All together , these results are consistent with the idea that MYC normally functions to facilitate P-TEFb's phosphorylation of the Pol II CTD on Ser2 and that decreasing the expression or activity of MYC eliminates this beneficial effect . The observation that MYC can bind and recruit P-TEFb to numerous MYC-target genes to activate transcription ( Kanazawa et al . , 2003; Rahl et al . , 2010 ) provides a plausible explanation for its promotion of P-TEFb's pSer2 . In addition to enhancing MYC's expression , we also investigated whether i-CDK9 may also directly affect the interaction of MYC with BRD4-P-TEFb . Indeed , the amounts of CDK9 , CycT1 , and BRD4 bound to the immunoprecipitated MYC-F were found to increase upon the exposure to i-CDK9 ( Figure 7D , E ) , revealing a dual mechanism used by cells to compensate for the loss of CDK9 activity by targeting the BRD4-P-TEFb complex to essential genes via a MYC-dependent pathway . Because the loss of CDK9 activity in i-CDK9-treated cells can be compensated by an increase in MYC's expression and interaction with P-TEFb , we reasoned that any meaningful antitumor effect caused by inhibiting CDK9 must involve the simultaneous suppression of the BRD4-dependent MYC expression . To test this hypothesis , we first performed cell proliferation assays and used the isobologram method to determine whether the combined effects of i-CDK9 and JQ1 were additive , synergistic , or antagonistic ( Tallarida , 2006 ) . Data matrices were generated by measuring anti-proliferative effects ( IC50 ) of i-CDK9 and JQ1 as a single agent or in combination on both HeLa and H1792 , a non-small cell lung cancer cell line with significant MYC amplification ( CN = 8 ) . Our data clearly show that the combination of i-CDK9 and JQ1 synergistically inhibited growth of both cell lines ( Figure 7F , G ) . In addition to inhibition of cell growth , the combination of i-CDK9 and JQ1 also synergistically induced apoptosis of both HeLa and H1792 cells as demonstrated by isobolograms that measure the production of Caspase 3/7 in treated cells ( Figure 7H , I ) . Furthermore , strong and synergistic apoptosis caused by i-CDK9 and JQ1 was also observed by flow cytometry measurement of propidium iodide ( PI ) -stained sub-G1 population of HeLa cells that contained fragmented DNA ( Figure 7J ) . Finally , in addition of JQ1 , the MYC inhibitor 10 , 058-F4 was also found to synergize with i-CDK9 to induce potent cell death ( Figure 7K ) . Taken all together , the data are consistent with the notion that simultaneous inhibition of CDK9's kinase activity and MYC's expression or function leads to synergistic induction of growth arrest and apoptosis of cancer cells . The aberrant activation of cyclin-dependent kinases ( CDKs ) and dysregulation of cell cycle progression is a hallmark of many human diseases that include cancer , cardiac myopathies , and inflammatory processes . Because of this , manipulation of cell cycle progression by means of small molecule inhibitors has long been suggested as a therapeutic avenue to cure these diseases especially cancer . There are now more than 50 pharmacological CDK inhibitors that have been described ( Knockaert et al . , 2002; Fisher , 2010 ) . Among these , flavopiridol has been promoted as a potent inhibitor of CDKs with a notable bias for CDK9 ( Senderowicz and Sausville , 2000; Chao and Price , 2001; Blagosklonny , 2004 ) . Treatment with flavopiridol has been shown to block cell cycle progression , promote differentiation , and induce apoptosis in various types of cancerous cells ( Senderowicz and Sausville , 2000; Blagosklonny , 2004 ) . However , as indicated in the current study , flavopiridol displayed only 2 . 9 to 28 . 6-fold lower inhibitory activity toward other CDK-cyclin pairs compared to CDK9-CycT1 ( Figure 1B ) . This relatively weak selectivity for CDK9 , coupled with the demonstrations that flavopiridol also binds strongly to duplex DNA ( Bible et al . , 2000 ) and many other non-CDK kinases ( http://www . discoverx . com/ReferenceTreeImages/Flavopiridol . htm ) , raise the possibility that some of the effects attributed to this compound may not be caused by the inhibition of CDK9 . Because simultaneous inhibition of both CDK9 and cell cycle CDKs makes the data interpretation difficult and confusing , a highly selective CDK9 inhibitor that can specifically block the P-TEFb-dependent elongation phase of Pol II transcription is thus very desirable and urgently needed . In this regard , the current development and characterization of i-CDK9 , which demonstrated far superior selectivity against CDK9 than did flavopiridol , is both timely and valuable for our efforts to further investigate CDK9 biology as well as the therapeutic potential of P-TEFb inhibition . One key finding from the present study is that long-term inhibition of P-TEFb by i-CDK9 potently up-regulated the expression of a number of PRGs that include MYC , FOS , JUNB , ERG1 , and others . Notably , a subset of the same PRGs ( FOS , JUNB , EGR1 , and GADD45B ) has also been shown to be potently down-regulated before they were significantly up-regulated following the treatment with the pan-CDK inhibitor flavopiridol ( Keskin et al . , 2012 ) . As for MYC , although several mechanisms have been indicated as responsible for its elevated expression , including gene amplification , chromosomal translocation , and alteration of protein stability ( Meyer and Penn , 2008 ) , the BRD4-dependent mechanism identified here and used by cells to up-regulate MYC and other PRGs in response to CDK9 inhibition represents a previously uncharacterized alternative method that appears to be equally effective in promoting MYC expression . In addition to serving as an epigenetic reader and recruitment agent to deliver active P-TEFb released from 7SK snRNP to acetylated chromatin at the MYC locus , BRD4 is also shown in the present study to use its C-terminal P-TEFb-interaction domain ( PID ) to directly increase the catalytic activity of CDK9 despite the fact that itself failed to display any kinase activity toward the Pol II CTD . This result is consistent with the observations reported recently by the Geyer laboratory , which employed both the synthetic CTD peptides as well as the full-length human CTD containing all 52 repeats in the kinase assays ( Itzen et al . , 2014 ) . Likely employing a similar mechanism for activating CDK9 , BRD4 is further shown in our study to also render CDK9 more resistant to inhibition when bound to P-TEFb ( Figure 4J ) . At this moment , the precise mechanism used by BRD4 PID to promote CDK9's kinase activity and protect against i-CDK9 is yet to be determined . It is interesting to note that the HIV-1 Tat protein appears to possess a similar ability to activate CDK9 upon binding to the isolated , recombinant P-TEFb ( Garber et al . , 2000 ) . While Tat is known to bind mostly to the surface of CycT1 ( Schulze-Gahmen et al . , 2014 ) , the PID appears to contact both CycT1 and CDK9 ( Yang et al . , 2005; Itzen et al . , 2014 ) . Thus , Tat and BRD4 PID may not contact P-TEFb in exactly the same manner . Despite this possible difference , the fact that the PID exhibits a sequence motif composition that is similar to that of Tat , and moreover , the conserved Tat-like motifs are also required for P-TEFb activation ( Itzen et al . , 2014 ) , suggests that the interaction of P-TEFb with either Tat or BRD4 PID may induce a similar conformational change that is beneficial for CDK9's interaction with substrates . The data presented in the present study support the model that the i-CDK9-induced MYC expression and interaction with P-TEFb is important to compensate for the loss of P-TEFb activity in treated cells . Given the fact that P-TEFb is a general transcription factor ( Zhou et al . , 2012 ) and that MYC is traditionally viewed as a sequence-specific transcription factor existing in a heterodimer with MAX ( Amati et al . , 1993 ) , how is the elevated MYC level and binding to P-TEFb able to sustain the expression of a vast array of cellular genes that are normally dependent on P-TEFb for proper transcription ( the current study and Chao and Price , 2001; Shim et al . , 2002 ) ? The answer to this question has recently come from studies showing that MYC is a transcriptional factor much more promiscuous in terms of the number and types of its target genes than what is suggested by the simple distribution and abundance of the MYC cognate sequence called the E-box in the genome . For example , when investigating the MYC-induced global amplification of transcription in lymphocytes and ES cells , the Levens group noticed that the activation of genes that contain the E-box , which are traditionally defined as the MYC-target genes , makes up only a modest fraction of the total number of genes activated in these cells ( Nie et al . , 2012 ) . Consistent with the idea that sequence-specific DNA binding is largely dispensable for most of MYC's transactivation function , the Price laboratory reported that the positions occupied by the MYC-MAX dimer across the human genome correlate with the Pol II transcription machinery rather than E-box elements ( Pufall and Kaplan , 2013 ) . Furthermore , a significant percentage of MYC-MAX is located slightly upstream of nearly all the promoter-proximally paused Pol II . These results , together with the previous demonstrations that MYC can recruit P-TEFb to its target genes to activate Pol II transcriptional elongation ( Kanazawa et al . , 2003; Rahl et al . , 2010 ) , implicate a critical role for this transcription factor in globally controlling the release of Pol II from pausing . This general but hitherto unknown function of MYC also explains well our present observations that blocking MYC's expression by shMYC or JQ1 or inhibiting the MYC-MAX function by 10 , 058-F4 further exacerbated the i-CDK9-induced global decrease in Pol II CTD phosphorylation on Ser2 ( Figure 7A–C ) . While the inhibition of CDK9 leads to elevated MYC expression and binding to P-TEFb in order to mitigate the impact of loss of cellular CDK9 activity , previous reports have shown that treating cells with JQ1 , which blocks the BRD4-dependent MYC expression , causes the release and activation of P-TEFb from 7SK snRNP ( Bartholomeeusen et al . , 2012; Li et al . , 2013 ) . These observations suggest the existence of a complementary and compensatory relationship between P-TEFb and MYC , and that the two transcription factors are likely part of a redundant mechanism used by cells to ensure optimal expression of key jointly controlled genes . Consistent with this idea , we show in the present study that simultaneous inhibition of CDK9's kinase activity and MYC's expression or function synergistically induced growth arrest and apoptosis of cancer cells . The importance of this result lies in its implication for future clinical applications of i-CDK9 and JQ1 or other BET bromodomain inhibitors . Since small molecule inhibitors such as i-CDK9 and JQ1 , no matter how selective they are , will inevitably produce off-target effects when used at high concentrations , the demonstration that they can produce dramatic anti-tumor effects when used together at levels much lower than either alone holds the great promise of minimizing these unwanted off-target effects in future therapy .	Cancers are often caused by mutations in genes that allow cells to proliferate uncontrollably . One gene that is frequently mutated in many cancers encodes a protein called MYC . The activity of this gene is normally tightly controlled , but the mutations found in human cancer cells mean that this gene is constantly switched on , and so too much MYC protein is produced . Previous studies have shown that a protein complex called ‘positive transcription elongation factor b’ ( or P-TEFb for short ) is essential to control the expression of the gene for MYC . P-TEFb works with an enzyme called RNA polymerase II to copy the instructions contained in protein-coding genes into long molecules called messenger RNAs . This process is called transcription and it involves a number of stages . P-TEFb is needed to start of one these stages , which is known as the ‘elongation’ step . P-TEFb consists of two protein subunits; one of which—an enzyme called CDK9—is the catalytic subunit . Most of the P-TEFb complexes in a cell are held in an inactive form , in which the activity of the CDK9 subunit is suppressed . If CDK9 is active when it should not be , certain proteins—including the MYC protein—can be produced in abnormally high amounts . This means that inhibiting CDK9 has been investigated as one way to reduce the production of the MYC protein . While some CDK9 inhibitors already exist , these compounds have the undesirable effect of also inhibiting other related enzymes and thus killing normal cells . Hence , new and more selective inhibitors of CDK9 are urgently needed . Lu , Xue et al . have now developed a new inhibitor of CDK9 , called i-CDK9 . The experiments show that i-CDK9 can potently inhibit CDK9 activity; and in human cells , very low levels of i-CDK9 prevented RNA polymerase II carrying out elongation for many genes . Unexpectedly , Lu , Xue et al . observed that more messenger RNA molecules that encode MYC were produced after cells were treated with low levels of i-CDK9 . Further investigation revealed that this unexpected result occurred because the P-TEFb complexes were released from the inactive form and brought to the MYC gene by another protein called BRD4 . This stimulated production of the MYC messenger RNAs . When P-TEFb was bound by BRD4 , the CDK9 activity was also protected against inhibition by i-CDK9 . Moreover , the reason that the MYC expression was induced by i-CDK9 is because the cells can compensate for the loss of CDK9 by using MYC to maintain the production of messenger RNAs of many key genes; these genes include the gene for MYC itself . These results suggest that CDK9 and MYC must be simultaneously inhibited in order to effectively treat cancers .	[ "Abstract", "Introduction", "Results", "Discussion" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2015	Compensatory induction of MYC expression by sustained CDK9 inhibition via a BRD4-dependent mechanism
The nuclear pore complex ( NPC ) is the principal gateway between nucleus and cytoplasm that enables exchange of macromolecular cargo . Composed of multiple copies of ~30 different nucleoporins ( Nups ) , the NPC acts as a selective portal , interacting with factors which individually license passage of specific cargo classes . Here we show that two Nups of the inner channel , Nup54 and Nup58 , are essential for transposon silencing via the PIWI-interacting RNA ( piRNA ) pathway in the Drosophila ovary . In ovarian follicle cells , loss of Nup54 and Nup58 results in compromised piRNA biogenesis exclusively from the flamenco locus , whereas knockdowns of other NPC subunits have widespread consequences . This provides evidence that some Nups can acquire specialised roles in tissue-specific contexts . Our findings consolidate the idea that the NPC has functions beyond simply constituting a barrier to nuclear/cytoplasmic exchange as genomic loci subjected to strong selective pressure can exploit NPC subunits to facilitate their expression . The main gateway between the nucleus and the cytoplasm is the nuclear pore complex ( NPC ) , a large multi-protein assembly spanning the nuclear envelope . The NPC is composed of multiple copies of ~30 proteins , termed nucleoporins ( Nups ) , arranged into an eightfold symmetric ring ( Beck and Hurt , 2017; Hampoelz et al . , 2019; Kim et al . , 2018 ) . Small molecules can freely diffuse across the NPC , whilst particles larger than 40 kDa or 5 nm require active transport . Transcripts that have passed nuclear quality control steps are actively trafficked across the NPC towards their target sites ( Tutucci and Stutz , 2011 ) and dedicated protein networks ensure that transcripts going through the NPC reach their correct cytoplasmic destinations ( Köhler and Hurt , 2007; Tutucci and Stutz , 2011 ) . The NPC has been implicated as more than a simple gateway , serving also as an active player in gene regulation ( Köhler and Hurt , 2010; Strambio-De-Castillia et al . , 2010 ) . Some Nups associate with chromatin , displaying preferences for certain epigenetic modifications ( Capelson et al . , 2010; Gozalo et al . , 2020; Iglesias et al . , 2020; Kalverda et al . , 2010; Vaquerizas et al . , 2010 ) , inducible genes sometimes re-locate proximally to the NPC upon activation ( Blobel , 1985; Dieppois et al . , 2006; Luthra et al . , 2007; Rohner et al . , 2013; Strambio-De-Castillia et al . , 2010 ) , and other Nups contribute to heterochromatin organisation and epigenetic inheritance ( Holla et al . , 2020; Iglesias et al . , 2020 ) . Notably , altered expression or mutation of certain Nups can cause human diseases that only affect specific tissues , despite the NPC being ubiquitous ( Beck and Hurt , 2017 ) . This suggests that some Nups might have evolved tissue-specific functions , though the nature of these remains elusive . Transposable element ( TE ) silencing in animal gonads is accomplished primarily through the action of piRNAs ( Czech et al . , 2018; Ozata et al . , 2019 ) . These 23- to 30-nt small RNAs guide PIWI-clade Argonaute proteins to recognise and silence active TEs . piRNAs originate from discrete genomic loci , termed piRNA clusters , largely composed of TE remnants ( Aravin et al . , 2006; Brennecke et al . , 2007; Mohn et al . , 2014 ) . In Drosophila melanogaster , dual-strand clusters produce RNAs from both genomic strands in the nurse cells of the ovary ( Figure 1—figure supplement 1A ) and rely on non-canonical transcription and export mechanisms ( Andersen et al . , 2017; ElMaghraby et al . , 2019; Kneuss et al . , 2019; Mohn et al . , 2014; Zhang et al . , 2014 ) . Specification of these transcripts for piRNA production takes place in perinuclear structures , namely nuage ( Lim and Kai , 2007; Malone et al . , 2009; Senti et al . , 2015 ) . Uni-strand clusters instead are transcribed from only one genomic strand and appear to be conventional RNA polymerase II transcripts ( Brennecke et al . , 2007; Dennis et al . , 2016; Goriaux et al . , 2014; Mohn et al . , 2014 ) . Of the two Drosophila uni-strand clusters , flamenco ( flam ) is the principal source of piRNAs in the somatic follicle cells that enclose egg chambers ( Figure 1—figure supplement 1A; Brennecke et al . , 2007; Malone et al . , 2009 ) and was originally identified as a master regulator of gypsy retrotransposons ( Mével-Ninio et al . , 2007; Pélisson et al . , 1994; Prud'homme et al . , 1995 ) . Being an unusually large transcriptional unit , flam covers up to ~650 kb of pericentromeric heterochromatin of chromosome X and depends on conventional RNA export mechanisms , centred on the nuclear export factor heterodimer Nxf1/Nxt1 ( Dennis et al . , 2016; Herold et al . , 2001; Tutucci and Stutz , 2011 ) . Upon export , flam transcripts localise to perinuclear Yb-bodies , where they are thought to be licensed for piRNA biogenesis ( Hirakata et al . , 2019; Murano et al . , 2019; Qi et al . , 2011; Saito et al . , 2010 ) , though the underlying molecular mechanisms are not fully understood . Yb-bodies are cytoplasmic , perinuclear condensates of the DEAD-box RNA helicase Yb , encoded by the fs ( 1 ) Yb gene , which is exclusively expressed in somatic follicle cells ( Figure 1—figure supplement 1A; Hirakata et al . , 2019; Olivieri et al . , 2010; Qi et al . , 2011; Saito et al . , 2010; Szakmary et al . , 2009 ) . Yb is essential for somatic piRNA production , and its assembly into Yb-bodies does not depend on any known piRNA biogenesis factor , and therefore is at the apex of the piRNA biogenesis protein network ( Hirakata et al . , 2019; Ishizu et al . , 2015; Ishizu et al . , 2019; Murota et al . , 2014; Olivieri et al . , 2010; Saito et al . , 2010 ) . Yb-bodies have been reported to possess biophysical properties typical of phase-separated condensates , which likely facilitate the biochemical processes happening in these foci ( Hirakata et al . , 2019 ) . Given that flam is the major piRNA source locus in somatic follicle cells , we and others have found that the formation of Yb-bodies depends on flam RNA ( Figure 1—figure supplement 1; Dennis et al . , 2016; Hirakata et al . , 2019; Sokolova et al . , 2019 ) . Two previously described flam mutant alleles ( Brennecke et al . , 2007; Malone et al . , 2009; Mével-Ninio et al . , 2007 ) disrupt piRNA cluster transcription via P-element insertions near the 5′ end: flamBG , carrying an insertion in the putative promoter , and flamKG , carrying an insertion immediately downstream of the TSS ( Figure 1—figure supplement 1B ) . The strongest effect on flam expression is observed in trans-heterozygous flies ( flamBG/KG ) , obtained through crosses between the two alleles . These mutants show strong de-repression of somatic , gypsy-family TEs ( Brennecke et al . , 2007; Malone et al . , 2009; Mével-Ninio et al . , 2007; Figure 1—figure supplement 1C , D ) , which is accompanied by a disassembly of Yb-bodies , despite some of the protein still being present ( Figure 1—figure supplement 1E–G ) . Notably , the production of other classes of piRNAs , such as those derived from the 3′ UTR of coding genes , is unchanged in these mutants ( Hirakata et al . , 2019; Sokolova et al . , 2019 ) , further underscoring a link between flam and Yb-bodies . Knockdowns of nxf1/nxt1 in ovarian somatic cells ( OSCs ) , a cell line derived from the somatic compartment of the ovary that expresses a functional piRNA pathway ( Saito et al . , 2009 ) , also compromise Yb-bodies formation ( Figure 1—figure supplement 1A , H ) , with similar results reported in soma-specific knockdowns in ovaries ( Dennis et al . , 2016; Sokolova et al . , 2019 ) . These results , together with previous findings ( Dennis et al . , 2016; Hirakata et al . , 2019; Sokolova et al . , 2019 ) , suggest that the production and localisation of flam transcripts to Yb-bodies and the assembly of those structures are interdependent . Nonetheless , it is still unknown how the transcript is specifically directed to Yb-bodies and , from there , licensed for processing . Here , by investigating the export and licensing of flam , we uncover a requirement of specific channel Nups for TE silencing in the somatic cells of the Drosophila ovary . We find that depletion of some NPC subunits compromises the assembly of Yb-bodies and that loss of Nup54 and Nup58 specifically impacts flam export , but not that of bulk mRNAs . We show that Nup54 and Nup58 physically associate with Nxf1/Nxt1 as well as Yb , implying the existence of an export-coupled localisation mechanism specifying flam as a piRNA precursor . Considered together , our results suggest that genomic loci under strong selective pressure can co-opt NPC subunits to facilitate expression , thus expanding the repertoire of processes in which Nups play a role . Various Nups have been ascribed gene regulatory functions , often via chromatin binding ( Strambio-De-Castillia et al . , 2010 ) , and a subset of NPC subunits has been genetically implicated in transposon control in Drosophila ovaries ( Czech et al . , 2013; Handler et al . , 2013; Muerdter et al . , 2013; Figure 1A ) . To understand whether any of these Nups play a specific role in piRNA-guided TE silencing , we systematically assessed the effect of their depletion on cell viability , TE expression , and Yb-body formation in OSCs ( Figure 1A , B ) . Knockdown of most Nups resulted in pronounced cell death and disassembled Yb-bodies , with little to no effect on TEs ( Figure 1B , Figure 1—figure supplement 2A–F , Figure 1—figure supplement 3 ) . Instead , loss of subunits of the Nup62 sub-complex ( Nup54-Nup58-Nup62 ) and their scaffold Nup93-1 ( Chug et al . , 2015; Stuwe et al . , 2015; Ulrich et al . , 2014 ) caused strong TE de-repression ( Figure 1B , Figure 1—figure supplement 2A–F ) . Among these , only the depletion of Nup54 and Nup58 resulted in TE up-regulation without severely affecting cell viability ( Figure 1B , Figure 1—figure supplement 2A ) , potentially hinting to an effect distinct from general nuclear transport . Yb-bodies were also dispersed in siNup54 and siNup58 and residual Yb was visible only at increased laser power , despite an overall minor effect on Yb protein levels ( Figure 1B , C , Figure 1—figure supplement 2C–G ) . Of note , TE de-repression caused by knockdown of nup54 and nup58 was comparable to that observed upon depletion of Nxt1 ( Figure 1B , Figure 1—figure supplement 2A ) , reported to also function in the co-transcriptional gene silencing branch of the piRNA pathway ( Batki et al . , 2019; Fabry et al . , 2019; Murano et al . , 2019; Zhao et al . , 2019 ) . These data indicate that loss of TE silencing upon depletion of most Nups is likely a result of the general NPC function in gene expression , whereas Nup54 and Nup58 seem to have more specific roles in transposon control . To test this hypothesis , we carried out RNA-seq from OSCs depleted of the Nup62 complex subunits ( Nup62-Nup54-Nup58 ) , the scaffold protein Nup93-1 or the piRNA biogenesis factor Yb ( Olivieri et al . , 2010; Saito et al . , 2010; Szakmary et al . , 2009 ) . Cells depleted of Nup54 or Nup58 showed a strong increase in the expression of gypsy-family TEs , which are known to be expressed in the somatic compartment of the ovary and regulated by flam ( Figure 1D–F , Figure 1—figure supplement 1C–D; Lécher et al . , 1997; Pélisson et al . , 1994; Prud'homme et al . , 1995 ) . Both the spectrum of TEs affected and the magnitude of de-repression were very similar to those observed in yb knockdowns . In contrast , we observed only a moderate impact on protein-coding genes , with 130 genes de-regulated in siYb ( more than fourfold and adjusted p value < 0 . 05 ) , 42 in siNup54 , and 42 in siNup58 ( Figure 1D–F , Figure 2—figure supplement 1A ) . A substantial fraction of those genes up-regulated by more than fourfold is found nearby transposon insertions that become de-silenced when the piRNA pathway is compromised ( 49/126 or 39% in siYb , 16/39 or 41% in siNup54 , 13/40 or 33% in siNup58 ) ( Figure 1D–F ) . One such example is the expanded ( ex ) gene on chromosome 2L ( Figure 1D–F ) . This strongly suggests that most of the gene expression changes observed upon these knockdowns are in fact a consequence of TE re-activation . Although knockdown of nup62 and nup93-1 also caused de-repression of some flam-regulated TEs , we found much more pronounced mis-expression of protein-coding genes , with 207 genes de-regulated in siNup62 and 417 in siNup93-1 ( more than fourfold and adjusted p value < 0 . 05 ) ( Figure 1G , H ) , which could not be explained by proximity to nearby TE insertions . Furthermore , this was accompanied by de-repression of TEs that are not normally subject to piRNA-mediated silencing in somatic cells , for example , R2-element ( Figure 1G , H ) . Of note , siNup62 and siNup93-1 resulted in expression changes of other Nups , RNA export factors , and Yb , presumably via indirect effects on nuclear transport and/or gene expression ( Figure 1I ) . Considered together , our results indicate that , in somatic follicle cells , Nup54 and Nup58 play specialised roles dedicated to transposon silencing , distinct from Nup62 and Nup93-1 . This functional specialisation of the two proteins , especially from their closest binding partners in the NPC , is highly surprising , particularly considering that Nup54 and Nup58 are integral components of an essential and ubiquitous protein complex and so presumed to have general functions across the animal . Nup54 and Nup58 belong to the highly conserved class of ‘FG-Nups’ ( Beck and Hurt , 2017; Figure 2A ) and constitute the heterotrimeric Nup62 complex ( Nup54-Nup58-Nup62 ) ( Chug et al . , 2015; Stuwe et al . , 2015; Ulrich et al . , 2014 ) that lines the inner channel of the NPC ( Beck and Hurt , 2017; Grandi et al . , 1995; Kim et al . , 2018 ) . The phenylalanine-glycine ( FG ) repeats of the Nup62 complex subunits contribute to the selective permeability barrier of the NPC and interact with nuclear transport receptors , such as Nxf1 ( Köhler and Hurt , 2007 ) . Our RNA-seq showed a reduction in steady-state levels of flam transcript upon nup54 , nup58 , nup62 , nup93-1 , and yb knockdowns in OSCs ( Figure 1D-H , Figure 2B , Figure 2—figure supplement 1B ) . This was both specific to flam as other somatic piRNA source loci ( e . g . the protein-coding gene tj or the piRNA cluster 20A ) were unaffected ( Figure 2—figure supplement 1C , D ) , and unexpected since prior studies had shown accumulation of flam transcripts in cases where its conversion into piRNAs was disrupted , for example , by knockdown of the ribonuclease zuc ( Murota et al . , 2014 ) . To probe the underlying mechanism , we first asked whether the impact was uniform throughout the locus . To address this question , we divided the flam genomic region into non-overlapping 1 kb bins and extracted those reads that could be mapped with high confidence ( see Materials and methods ) . Plotting fold-changes in these 1 kb bins following zuc knockdown , which prevents flam processing into piRNAs , showed an increased precursor abundance that was uniform across the entire locus ( Figure 2B , C ) . In contrast , depletion of Yb , Nup54 , and Nup58 revealed a reduction that was more pronounced towards the 3′ end of flam ( Figure 2B , C ) . In nup62 and nup93-1 knockdowns , reduced RNA levels were uniform , again highlighting a different role for these Nups ( Figure 2B , C ) . As expected , we observed uniformly reduced RNA levels in flamBG/KG trans-heterozygous mutants that impair transcription of the entire locus ( Figure 2B , Figure 2—figure supplement 1E; Brennecke et al . , 2007; Goriaux et al . , 2014; Malone et al . , 2009 ) . All these analyses showed similar patterns using 100 kb sliding windows ( Figure 2—figure supplement 1F ) , thus our results were consistent regardless of the window size . Next , to determine whether the observed reduction stems from decreased transcription initiation , we examined nascent RNA at the flam locus via PRO-seq . PRO-seq in control cells ( siGFP ) revealed one major transcription initiation peak , as expected ( Brennecke et al . , 2007; Goriaux et al . , 2014 ) , and detected a second , previously unidentified , minor peak ~1 kb further downstream ( Figure 2—figure supplement 1G ) . Knockdown of nup54 , nup58 , or yb had little to no effect on either signal around the transcription initiation site or within the first 10 kb and 40 kb of flam , whereas a more pronounced decrease in siYb cells was observed upon inspection of the entire locus ( Figure 2—figure supplement 1G , H ) . In contrast , global PRO-seq signal from protein-coding genes and cluster 20A was unchanged ( Figure 2—figure supplement 1I ) . Overall , these observations are consistent with a hypothesis that loss of Nup54 or Nup58 reduces the stability of flam transcripts , with larger effects on regions distal from the transcription initiation sites . To analyse the effects of Nup54 and Nup58 depletion on piRNA populations , we sequenced small RNAs from OSC knockdowns . As previously reported ( Hirakata et al . , 2019 ) , depletion of Yb caused a collapse in the antisense , TE-targeting piRNA population , but leaving 21-nt siRNAs unaffected ( Figure 2—figure supplement 2A , B ) . Knockdown of nup54 and nup58 resulted in a approximately threefold decrease in antisense piRNAs , but this impact was highly specific to those derived from flam ( Figure 2D , Figure 2—figure supplement 2A , B ) . In contrast , piRNAs derived from tj and cluster 20A were unaffected or slightly more abundant ( Figure 2E , Figure 2—figure supplement 2C ) . Whilst siYb had a general impact on piRNA production , in line with its role as key biogenesis factor ( Hirakata et al . , 2019; Ishizu et al . , 2015; Ishizu et al . , 2019; Murota et al . , 2014; Olivieri et al . , 2010; Saito et al . , 2010 ) , siNup54 and siNup58 only showed a reduction of flam-derived piRNAs ( Figure 2F , G , Figure 2—figure supplement 2D ) . Binning analysis of the flam locus showed that piRNAs were lost homogenously along the entire cluster in siYb ( Figure 2H , Figure 2—figure supplement 2E ) , unlike the precursor transcript levels measured by RNA-seq , indicating that no processing can occur in the absence of Yb . In contrast , piRNA loss upon nup54 and nup58 knockdown was more pronounced towards the 3′ region ( Figure 2H , Figure 2—figure supplement 2E ) , mirroring the precursor transcript reduction observed by RNA-seq . These data are in agreement with a defect in precursor specification upon siYb and suggest that Nup54 and Nup58 play a role in flam piRNA biogenesis that is distinct from that of Yb . RNA-FISH for flam in OSCs typically shows discrete foci on the nuclear rim and in the cytosol ( Dennis et al . , 2016; Murota et al . , 2014; Figure 2L , Figure 2—figure supplement 3A ) . Depletion of Nup54 and Nup58 resulted in clustering of the signal in one predominant focus within the nuclear envelope ( Figure 2I–L , Figure 2—figure supplement 3A ) . Since the flam DNA locus is located at the nuclear periphery ( Figure 2—figure supplement 3B , C ) , this RNA nuclear focus likely corresponds to flam RNPs stalled prior to nuclear export ( Dennis et al . , 2016; Dennis et al . , 2013 ) . Nonetheless , this positions the flam locus in close proximity to the NPC , possibly underscoring the need of a specialised transcription-coupled export machinery linked directly to piRNA production . Of note , neither nup54 nor nup58 knockdown affected the distribution of bulk polyadenylated mRNAs ( Figure 1—figure supplement 2C , Figure 1—figure supplement 3 , Figure 2L , Figure 2—figure supplement 3A ) , unlike depletion of Nxf1 or Nxt1 , which instead resulted in nuclear retention of newly synthesised mRNA ( Figure 1—figure supplement 2C , Figure 1—figure supplement 3 , Figure 2M ) , as reported ( Herold et al . , 2001 ) . Considered together , these results indicate that transposon silencing defects resulting from loss of Nup54 or Nup58 arise from their role in facilitating flam export from the nucleus . In OSCs and follicle cells of the ovary , this activity dominates any general function in NPC biology since cells are viable and distributions of bulk mRNAs are largely unaffected upon their depletion . Given that the effect is most prominent on the 3′ end of the transcript , we hypothesise that Nup54 and Nup58 might be required to ensure processivity of nuclear export of this , otherwise unstable , long transcript . In this scenario , residual flam molecules ( likely corresponding to the 5′ portion of the cluster ) that reach the cytosol upon siNup54 and siNup58 might still be processed by Yb , although with lower efficiency than within properly formed Yb-bodies . This role of Nup54 and Nup58 could also , directly or indirectly , affect transcriptional elongation and termination of the piRNA cluster; however , further work will be required to test this hypothesis . The FG-repeats of Nup54 and Nup58 protrude into the inner channel to form the NPC permeability barrier and interact with nuclear transport receptors , such as Nxf1 , making these regions obvious candidates for regulating flam export . We therefore designed deletion mutants targeting Nup54 and Nup58 domains and assayed their ability to interact with other Nups and to rescue TE de-repression in OSCs ( Figure 3A ) . These constructs lack either the amino-terminal region , which in both Nups carries the FG-repeats , or the carboxy-terminal part , which mediates the interaction of these proteins with each other and with the rest of the pore ( Figure 3A , B; Chug et al . , 2015; Stuwe et al . , 2015 ) . We depleted Nup54 or Nup58 individually in OSCs and then re-introduced either an siRNA-resistant full-length ( FL ) or deletion construct of Nup54/Nup58 , or a negative control ( mCherry ) , and assayed their ability to restore transposon repression . As expected , FL Nup54 and Nup58 rescued mdg1 up-regulation to levels comparable to siGFP ( Figure 3C ) . Likewise , deleting the FG-repeats in Nup54 and Nup58 ( ∆FG ) had little effect on their TE silencing capability compared to FL Nups . In contrast , Nup54 and Nup58 lacking the C-terminal domain ( ∆C ) failed to rescue transposon de-repression ( Figure 3C , Figure 3—figure supplement 1A ) . These results suggest that the ability to interact with the other Nups is required to ensure TE silencing , and thus that Nup54 and Nup58 carry out this function from within the NPC . With very few exceptions , Nup null mutants are generally not viable in Drosophila . One such exception , the nup54MB003363 ( nup54MB ) allele , produces a truncated protein lacking the carboxy-terminal region due to a Mi{ET1} transposon insertion ( Nallasivan et al . , 2020 ) . This shortened Nup54 protein lacks the Nup54-family domain ( Figure 3A , D , Figure 3—figure supplement 1B ) and fails to co-precipitate with Nup58 ( Figure 3E ) , as expected from earlier reports ( Chug et al . , 2015 ) . Thus , it resembles the carboxy-terminal truncation ( Nup54∆C ) that is unable to sustain transposon repression in OSCs ( Figure 3A–C ) . We sought to determine if this hypomorphic allele phenocopies the molecular phenotype of the nup54 knockdown in OSCs . Trans-heterozygous flies ( nup54MB/9B4 ) carrying the nup54MB allele over the nup549B4 deficiency spanning the entire nup54 locus ( Nallasivan et al . , 2020 ) were viable , although at reduced Mendelian ratios , and had smaller , but not rudimentary , ovaries ( Figure 3—figure supplement 1C ) . To minimise differences in TE content between fly strains , we crossed the nup54MB allele to w1118 flies and compared nup54MB/9B4 trans-heterozygote mutants to nup54MB/w1118 heterozygotes and to w1118 controls . RNA-seq from ovaries of trans-heterozygous nup54MB/9B4 flies showed TE de-repression and reduced flam RNA levels ( Figure 3F , G , Figure 3—figure supplement 1D ) . Expression levels of major dual-strand piRNA clusters ( 42AB , 38C , and 80F ) , which rely on different , specialised transcription and export pathways , were unchanged , if not slightly higher , in the case of 80F ( Figure 3F ) . Of note , the increased expression of flam-regulated transposons ( e . g . gypsy ) was already evident in nupMB/w1118 heterozygous flies ( Figure 3G , Figure 3—figure supplement 1D ) . In this in vivo setting , we observed a broader impact on the expression of protein-coding genes than for knockdowns in cell culture ( Figure 3—figure supplement 1D , E ) , likely reflecting a more general function of Nup54 in the germline tissue of the ovary . In this regard , also germline TEs were up-regulated in nup54 mutants ( Figure 3G , Figure 3—figure supplement 1D ) , confirming an earlier genetic link to TE silencing in germ cells ( Czech et al . , 2013 ) . Close inspection of the levels of flam RNAs using the previously described binning strategy showed that the down-regulation was more prominent towards promoter-distal regions of the cluster ( Figure 3—figure supplement 1F ) , thus recapitulating our results from knockdowns in OSCs . Lastly , Yb-bodies were reduced in nup54MB/9B4 trans-heterozygous flies ( Figure 3H , Figure 3—figure supplement 2 ) . The yb transcript is mildly up-regulated in nup54MB/9B4 ( Figure 3—figure supplement 1E ) , thus loss of Yb-bodies likely arises as a result of compromised flam export . Overall , these data confirm a requirement of Nup54 and Nup58 for the expression and TE silencing activity of flam in both OSCs and in vivo . We find that the ability of Nup54 and Nup58 to form a complex is critical for their function but the integrity of the FG-repeat regions of each protein individually is not . Of note , the FG-repeats of their yeast homologs Nup49 and Nup57 are also dispensable for cell survival , leading to the suggestion that not all FG-Nups are equivalent and some can facilitate distinct translocation pathways ( Iovine et al . , 1995; Strawn et al . , 2004 ) . To understand the molecular basis of this specificity , we explored three avenues: ( 1 ) physical proximity of the flam genomic locus to cytosolic Yb-bodies , ( 2 ) direct binding of Nup54 and/or Nup58 to flam RNA , or ( 3 ) an interaction between Nup54/Nup58 and flam RNP export complexes . Although the flam locus is at the nuclear periphery , we failed to detect any consistent correlation with the position of Yb-bodies on the opposite side of the nuclear envelope in both OSCs and ovaries ( Figure 2—figure supplement 3C; Figure 2—figure supplement 3D ) thus the first hypothesis seems unlikely . To test the second possibility , we carried out CLIP-seq for Nup54 , Nup58 , and the mRNA exportin Nxf1 , which was previously implicated in flam export ( Dennis et al . , 2016 ) . Nxf1 was shown to be loaded onto nascent mRNAs upon splicing , to interact with FG-Nups and to function itself as a mobile Nup ( Ben-Yishay et al . , 2019; Derrer et al . , 2019; Köhler and Hurt , 2007 ) , therefore representing the most probable link between the nascent flam transcript and the NPC . CLIP-seq experiments with HALO-tagged Nup54 and Nup58 failed to enrich for flam RNA , whereas independent experiments with amino- and carboxy-terminally HALO-tagged Nxf1 showed an interaction towards the 5′ end of the flam transcript ( Figure 4—figure supplement 1A ) , which was reported to undergo splicing ( Goriaux et al . , 2014 ) . Since the association of Nxf1 with a cargo transcript is believed to follow a splicing event , this may implicate its co-transcriptional loading onto the 5′ spliced region of flam as the initial signal for export . So far , no splicing has been reported in the downstream regions of flam . Next , we sought to determine whether the so far identified factors participating in flam export ( i . e . Nxf1 , Nup54 , Nup58 , and Yb ) physically interact with each other . We first searched for protein interactions between Yb and the NPC via BASU-mediated proximity labelling , followed by mass spectrometry ( PL-MS ) in OSCs ( Kim et al . , 2014; Munafò et al . , 2019; Ramanathan et al . , 2018; Roux et al . , 2012 ) . In this experiment , the protein of interest is fused to a biotin ligase ( BASU ) and expressed in OSCs . Upon biotin supplementation , the fusion protein produces activated biotin-AMP intermediates that covalently attach to accessible lysine residues of proteins in close spatial proximity ( Roux et al . , 2012 ) . Biotinylated proteins are subsequently recovered by streptavidin pulldown and identified by quantitative mass spectrometry . Yb PL-MS enriched , among others , for the known Yb interactors Armi and Piwi ( Hirakata et al . , 2019; Saito et al . , 2010 ) but not for mitochondrial piRNA pathway proteins ( Figure 4—figure supplement 1B , Figure 4—figure supplement 1—source data 1 ) , in line with the current model for piRNA biogenesis that postulates shuttling of Armi between Yb-bodies and mitochondria but not that of Yb itself ( Ge et al . , 2019; Munafò et al . , 2019; Yamashiro et al . , 2020 ) . Only one Nup , Nup214 , which localises to the cytoplasmic filaments and whose yeast homolog ( Nup159 ) cooperates with the DEAD-box helicase Dbp5 in disassembling export complexes , was detected as significantly enriched . This result indicates that Yb is not an integral NPC component but may nonetheless localise proximally to the cytosolic filaments of the pore . Instead , PL-MS for Nup54 and Nup58 ( Figure 4A , B , Figure 4—source data 1 , Figure 4—source data 2 ) enriched for various subunits of the NPC , including all the components of the Nup62 sub-complex and the scaffold Nup93-1 . We detected a modest enrichment of Nxf1 , which is known to interact with FG-Nups and to export flam RNA ( Dennis et al . , 2016; Segref et al . , 1997 ) . Notably , Yb was among the most highly enriched proteins in PL-MS for both Nups ( Figure 4A , B ) , thus indicating that Yb contacts the ( cytoplasmic side of the ) NPC in OSCs . No other known piRNA pathway factor was detected in proximity to Nup54 and Nup58 with this approach . Since neither of the Nups was detected by Yb PL-MS , we sought to probe a putative interaction between Yb and the NPC via alternative methods . Using immunofluorescence , we could often observe overlapping and/or adjacent signals between Yb and Nup54/Nup58 ( Figure 4C ) , as approximated by biotinylation staining from TurboID fusion proteins ( Branon et al . , 2018 ) ; however , the Nup signal was found along the entire nuclear envelope and not exclusively in association with Yb-bodies ( Figure 4C ) . Conversely , Yb-bodies were often distinct from Nup foci , thus indicating that Yb is not stably anchored to the NPC but rather dynamic and possibly explaining why only the cytosolic Nup214 was detected by Yb PL-MS . These data were not an artefact due to protein over-expression as we observed similar staining patterns using an antibody specific for FG-Nups ( Figure 4D , Figure 4—figure supplement 1C ) . This suggests that physical associations between Yb and Nup54/Nup58 within the NPCs are likely transient and that Nup54/Nup58 are not confined only to a discrete site on the nuclear envelope . With Nup54 and Nup58 expected to being present in every pore , flam export might only be initiated where they contact Yb , and this in turn would nucleate Yb-body assembly . We cannot formally exclude the hypothesis that a pool of Nup54 and Nup58 is also present within Yb-bodies and carries out its function independently of the NPC , although in this case one might have expected them to be detected in Yb PL-MS experiments . To test whether Yb binds directly to Nup54 and Nup58 , we carried out co-immunoprecipitation experiments in Drosophila Schneider 2 ( S2 ) cells , which do not express a functional piRNA pathway and represent a convenient system to investigate protein-protein interactions that are not bridged by other piRNA pathway components . We used 3xFLAG-tagged Yb , Nup54 , or Nup58 as baits and probed for the corresponding HA-tagged versions . As expected , Nup54 and Nup58 recovered a substantial amount of their respective partner Nup ( Figure 4E ) . We also detected interactions with Yb and , to a lesser extent , with Nxt1 and Nxf1 ( Figure 4E ) . Co-immunoprecipitation of Nup54/Nup58 and Yb in OSCs was insensitive to RNase I treatment ( Figure 4—figure supplement 1D ) , thus indicating that their interaction is not mediated by flam RNA , and to knockdown of endogenous Yb , thus ruling out over-expression artefacts ( Figure 4F , Figure 4—figure supplement 1F ) . Reciprocal pull-down experiments in OSCs confirmed these findings , with 3xFLAG-Yb recovering HA-tagged Nup54 and Nup58 and only modest amounts of Nxt1 , but not Nxf1 ( Figure 4—figure supplement 1E ) . The amino-terminal region of Yb contains a HelC domain , which is required exclusively for flam piRNA production as opposed to piRNA biogenesis in general ( Hirakata et al . , 2019 ) . Its deletion in OSCs recapitulates the biogenesis phenotype caused by nup54 and nup58 knockdown . We observed that this deletion weakened , though did not completely abolish , the interaction between Yb and the Nups ( Figure 4F , Figure 4—figure supplement 1F ) , thus indicating that it is at least partially responsible for anchoring Yb to the NPC . Co-immunoprecipitation of HA-tagged FL Nup54/Nup58 or versions carrying the FG or carboxy-terminus domain deletions showed that Yb interacts with the FG-Nups only when they are able to assemble into the NPC as deletion of either Nup carboxy-terminus , which ablates interaction between the two Nups ( Figure 3B ) , reduced the interaction ( Figure 4—figure supplement 1G ) . Although we cannot exclude the involvement of additional adaptor proteins that remained undetected by our approach , these data suggest that Yb associates with the cytoplasmic side of the NPC and binds to exiting flam transcripts . Because binding to RNA triggers the aggregation of flam-Yb RNPs into phase-separated Yb-bodies ( Hirakata et al . , 2019 ) , we hypothesise that this provides the directionality to flam transport . Here , we find that Nup54 and Nup58 are specifically required for TE regulation in the follicle cells of Drosophila ovaries by enabling export and subsequent piRNA production from flam transcripts . This functional requirement is distinct from the general role of the NPC as other Nups , even those most proximal to Nup54 and Nup58 within the pore , are broadly required for gene expression and cell survival . These findings consolidate our view of Nups as dynamic players in various cellular processes and expand the variety of roles ascribed to the Nups . Though flam bears canonical features common to other mRNAs , it is nonetheless specifically recognised and processed into piRNAs . A dedicated export-coupled licensing , involving Nup54 and Nup58 , may allow this long transcript to be escorted directly from its genomic origin to the Yb-bodies to facilitate piRNA production . If this machinery is disrupted , for example , by loss of Nup54 and Nup58 , flam transcripts are confined to the nucleus and destabilised , thus underscoring the need for a unified process from transcription to licensing . It is interesting to note that siYb caused a slight decrease of PRO-seq signal across the entire piRNA cluster locus , possibly indicating that disruption of flam export-coupled licensing negatively affects its transcription via a yet-unknown feedback mechanism . We find that Nup54 and Nup58 interact with both the nuclear ( Nxf1/Nxt1 ) and cytosolic ( Yb ) components of flam expression and thus suggest that these factors bridge nuclear export to cytosolic fate specification . Although several aspects of the molecular mechanism remain to be elucidated ( i . e . how this links to upstream transcription and whether additional factors are involved in the nucleus or in the cytosol ) , we propose a tentative model whereby this Nxf1-NPC-Yb axis coordinates initiation , processivity , and directionality of flam export ( Figure 4G ) , directly feeding the transcript into the piRNA biogenesis route via Yb-bodies . Our rescue experiments and protein-protein interaction studies argue for this TE silencing function of Nup54 and Nup58 to be carried out from within the pore , especially since neither Nup is among the most dynamic components of the NPC ( Rabut et al . , 2004 ) . However , we cannot completely exclude a function for Nup54 and Nup58 in the cytosol , possibly as components of Yb-bodies . We find that Nup54 and Nup58 interact directly with both Yb and Nxf1; however , it is presently unclear how the transcript is released from the Nxf1/Nxt1 export complexes and handed over to Yb , especially since we did not detect physical interactions between Yb and Nxf1 . It cannot be excluded that such interaction exists and remained undetected in our over-expression experiments or that additional adaptor proteins take part in this process . Alternatively , Nup54/58 may form independent complexes with Yb and Nxf1; however , our lack of enrichment for flam transcripts associated with Nup54/58 makes this seem unlikely . Super-resolution imaging of the NPC will be required to clarify the relative position of each component of this export route and to understand whether Nup54 and Nup58 function also outside the NPC . Of note , cytosolic Nup358 has been shown to be required for piRNA production from the dual-strand cluster 42AB in nurse cells ( Parikh et al . , 2018 ) . Its depletion leads to prominent de-localisation of Piwi , which we did not observe upon loss of Nup54/Nup58 , thus suggesting a different mode of action . Nonetheless , this further underscores that different NPC subunits can be co-opted for TE silencing in tissue-specific contexts . The proposed mechanism for flam export-coupled licensing relies on a tissue-specific effector ( Yb ) and could represent a broader paradigm for co-option of FG-Nups for specific transcript trafficking routes . We envision that in principle any transcript subject to strong selective pressure could evolve a dedicated export machinery via adapting FG-Nup functions in a cell-type-specific manner . Such mechanisms may have gone unnoticed previously because of the general role of the NPC . Interestingly , several Nup genes in Drosophila show signs of rapid adaptive evolution that result in hybrid incompatibilities ( Presgraves et al . , 2003; Tang and Presgraves , 2009 ) , which is often a hallmark of genes involved in genetic conflicts , such as transposon control . We further speculate that a tissue-specific control of export routes might contribute to explain the molecular mechanisms underlying so-called ‘nucleoporopathies’ , human syndromes caused by mutated Nups ( Beck and Hurt , 2017; Braun et al . , 2018; Miyake et al . , 2015 ) . In these diseases , mutation or expression changes of a Nup present in all cells of the organism leads to tissue-specific phenotypes . We hypothesise that this might stem from specific roles of those Nups in regulating genes that are essential for the functionality of that particular tissue , which would in turn make the said tissue especially susceptible to the loss of the Nup . Future investigation will shed light on how widespread these mechanisms might be . OSCs were a gift from Mikiko Siomi and were cultured as described ( Niki et al . , 2006; Saito , 2014; Saito et al . , 2009 ) . Drosophila S2 cells were purchased from Thermo Fisher Scientific and were grown at 26°C in Schneider media supplemented with 10% FBS . Cells were routinely tested for mycoplasma infection by an in-house facility . All flies were kept at 25°C on standard cornmeal or propionic food . Control w1118 flies were a gift from the University of Cambridge Department of Genetics Fly Facility . Nup54 mutant lines were provided by M . Soller ( Nallasivan et al . , 2020 ) . A full list of fly stocks used in this study is provided in Supplementary file 1 . Knockdowns and nucleofections in OSCs were carried out as previously described ( Saito , 2014 ) using the Cell Line Nucleofector Kit V ( Lonza VVCA-1003 ) on a Nucleofector II device ( program T-029 ) . OSC transfections were carried out using Xfect transfection reagent ( Takara Bio 631317 ) , as previously described ( Saito , 2014 ) . All constructs used in cells were expressed from the Drosophila act5c promoter . A full list of siRNAs used in this study is provided in Supplementary file 1 . S2 cells were transfected using Effectene ( Qiagen ) , according to the manufacturer’s instructions . PL-MS experiments in OSCs were performed as previously described ( Munafò et al . , 2019 ) . Briefly , 4 × 106 OSCs were transfected with 20 μg of plasmid expressing an HA-BASU fusion or HA-ZsGreen . 48 hr after transfection , the media was supplemented with 200 μM biotin for 1 hr . Cell pellets were lysed in 1 . 8 ml lysis buffer ( 50 mM Tris , pH 7 . 4 , 500 mM NaCl , 0 . 4% SDS , 1 mM DTT , 2% TritonX-100 with cOmplete protease inhibitors ) and sonicated using a Bioruptor Pico ( Diagenode , 3× cycles 30 s on/30 s off ) . Sonicated lysates were diluted 2× in 50 mM Tris , pH 7 . 4 , and cleared for 10 min at 16 , 500 g . Lysates were pre-cleared for 1 hr at 4°C with 100 μl of Protein A/G Dynabeads ( Thermo Fisher Scientific 10015D ) and the supernatant collected to a fresh tube . Biotinylated proteins were isolated by incubation with 200 μl of Dynabeads ( MyOne Streptavidin C1; Life Technologies ) overnight at 4°C . Beads were washed 2× in 2% SDS , 2× in Wash Buffer 1 ( 0 . 1% deoxycholate , 1% Triton X-100 , 500 mM NaCl , 1 mM EDTA , and 50 mM 4- ( 2-hydroxyethyl ) −1-piperazineethanesulfonic acid , pH 7 . 5 ) , 2× with Wash Buffer 2 ( 250 mM LiCl , 0 . 5% NP-40 , 0 . 5% deoxycholate , 1 mM EDTA , and 10 mM Tris , pH 8 ) , and 2× with 50 mM Tris . Beads were rinsed twice with 100 mM ammonium bicarbonate . BASU-Nup54 , BASU-Nup58 , and BASU-Yb pulldowns were subjected to TMT-labelling followed by quantitative mass spectrometry on a nano-ESI Fusion Lumos mass spectrometer ( Thermo Fisher Scientific ) . On-bead Trypsin digestion , TMT chemical isobaric labelling and data analysis were performed by the CRUK-CI proteomics core as previously described ( Papachristou et al . , 2018 ) . Spectral raw files from PL-MS of BASU-Yb , Nup54 , and Nup58 were processed with the SequestHT search engine on Thermo Scientific Proteome Discoverer 2 . 1 . Data was searched against a custom FlyBase database ( ‘dmel-all-translation-r6 . 24’ ) at 1% spectrum-level FDR ( False Discovery Rate ) criteria using Percolator ( University of Washington ) . MS1 mass tolerance was constrained to 20 ppm , and the fragment ion mass tolerance was set to 0 . 5 Da . TMT tags on lysine residues and peptide N termini ( +229 . 163 Da ) were set as static modifications . Oxidation of methionine residues ( +15 . 995 Da ) , deamidation ( +0 . 984 ) of asparagine and glutamine residues , and biotinylation of lysines and protein N-terminus ( +226 . 078 ) were included as dynamic modifications . For TMT-based reporter ion quantitation , we extracted the signal-to-noise ratio for each TMT channel . Parsimony principle was applied for protein grouping , and the level of confidence for peptide identifications was estimated using the Percolator node with decoy database search . Strict FDR was set at q-value <0 . 01 . Downstream data analysis was performed on R using the qPLEXanalyzer package ( Kishore and Eldridge , 2018 ) as described ( Papachristou et al . , 2018 ) . Only proteins with more than one unique peptide were plotted . S2 cells or OSCs were transfected with 3xFLAG- and HA-tagged constructs . After 48 hr , cells were lysed in 250 μl of Pierce IP lysis buffer supplemented with cOmplete protease inhibitors ( Roche ) . Equal amounts of lysate for each sample were diluted to 1 ml with IP lysis buffer and incubated with 30 μl of anti-FLAG M2 magnetic beads ( Sigma M8823 ) overnight at 4°C . For anti-HA pulldowns , lysates were incubated with 30 µl of anti-HA ( Thermo Fisher Scientific 88836 ) beads overnight at 4°C . Beads were washed 3 × 15 min in 1× Tris-buffered Saline ( TBS ) with protease inhibitors , then resuspended in 2xNuPAGE LDS Sample Buffer ( Thermo Fisher Scientific ) and boiled for 3 min at 90°C to elute immunoprecipitated proteins . Western blots were carried out using standard protocols . Protein lysates were run on NuPAGE 4–12% pre-cast gels and transferred to nitrocellulose membranes using a dry blotting system ( iBlot2 . 0 ) . Membranes were blocked for 1 hr at RT with 1× LiCor blocking buffer diluted in Phosphate Buffered Saline ( PBS ) and primary antibodies incubated overnight at 4°C . Following 3× washes in PBS + 0 . 1%Tween , secondary antibodies conjugated to infrared dyes ( and/or streptavidin; LiCor925-32230 ) were incubated for 45 min at room temperature ( RT ) . Images were acquired on an Odyssey CLx scanner ( LiCor ) . The following primary antibodies were used: anti-HA ( C29F4; Cell Signaling Technology ) , anti-FLAG ( Sigma #F1804 ) , anti-Piwi ( Brennecke et al . , 2007 ) , anti-Yb ( Saito et al . , 2010 ) , anti-Yb ( Handler et al . , 2011 ) ( used on ovary lysates ) , and anti-tubulin ( ab18251 ) . Cells were plated 1 day in advance on fibronectin-coated coverslips , fixed for 15 min in 4% Paraformaldehyde ( PFA ) , permeabilised for 10 min in PBS , 0 . 2% Triton X-100 , and blocked for 30 min in PBS , 0 . 1% Tween-20 ( PBST ) , and 1% BSA . Primary antibodies were diluted 1:500 in PBST and 0 . 1%BSA ) and incubated overnight at 4°C . After 3 × 5 min washes in PBST , secondary antibodies were incubated for 1 hr at RT . After 3 × 5 min washes in PBST , DAPI was incubated for 10 min at RT and washed twice in PBST . Coverslips were mounted with ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific #P36961 ) and imaged on a Leica SP8 confocal microscope ( 100× oil objective ) . For TurboID labelling , cell culture media was supplemented with 500 µM biotin for 1 hr . Detection was carried out using streptavidin conjugated to AlexaFluor-555 ( Thermo Fisher Scientific ) . The following antibodies were used: anti-Piwi ( Brennecke et al . , 2007 ) , anti-FLAG tag ( Sigma #F1804 ) , anti-HA tag ( ab9110 ) , anti-Yb ( Saito et al . , 2010 ) , anti-lamin ( DSHB , ADL67 . 10 ) , and anti-FG-Nups ( Biolegend mAb414 ) . RNA-FISH was performed with hybridisation chain reaction ( HCR ) , similar as reported ( Ang and Yung , 2016; Choi et al . , 2014 ) . OSCs were seeded on fibronectin-coated coverslips , fixed for 15 min in 4% PFA , washed 2 × 5 min with PBS , and permeabilised for at least 24 hr in 70% ethanol at −20°C . Ethanol was removed and slides were washed 2 × 5 min in 2× saline-sodium citrate buffer ( SSC ) . Samples were incubated for 10 min in 15% formamide in 2× SSC . HCR probes were diluted to 1 nM each in hybridisation buffer ( 15% formamide , 10% dextran sulfate in 2× SSC ) and incubated overnight at 37°C in a humidified chamber . Samples were washed twice in 2× SSC and 10 min in 30% formamide at 37°C . HCR hairpins conjugated to AlexaFluor-647 or oligo-dT probes conjugated to AlexaFluor-488 were heat-denatured and diluted to 120 nM in 5× SSC and 0 . 1% Tween-20 ( SSCT ) . HCR amplification was carried out for 2 hr at RT in the dark and washed 3 × 10 min with 5× SSCT . Nuclei were stained with DAPI ( 1:10 , 000 in SSCT ) for 10 min , followed by 3 × 10 min washes in 5× SSC . Slides were mounted with ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific ) and imaged on a Leica SP8 confocal microscope ( 100× oil objective ) . Probes were purchased from IDT , and all sequences are provided in Supplementary file 1 . Ovary samples or cell pellets were lysed in 1 ml TRIzol , and RNA was extracted according to the manufacturer’s instruction . 1 μg of total RNA was treated with DNAseI ( Thermo Fisher Scientific ) , and reverse transcribed with the Superscript III First Strand Synthesis Kit ( Thermo Fisher Scientific ) , using oligo-dT20 primers . Real-time PCR ( qPCR ) experiments were performed with a QuantStudio Real-Time PCR Light Cycler ( Thermo Fisher Scientific ) . Transposon levels were quantified using the ∆∆CT method ( Livak and Schmittgen , 2001 ) , normalised to rp49 , and fold-changes were calculated relative to the indicated controls . All primer sequences are listed in Supplementary file 1 . Fly ovaries were dissected in ice-cold PBS , fixed for 15 min in 4% PFA at RT , and permeabilised with 3 × 10 min washes in PBS with 0 . 3% Triton X-100 ( PBS-Tr ) . Samples were blocked in PBS-Tr with 1% BSA for 2 hr at RT and incubated overnight at 4°C with primary antibodies in PBS-Tr and 1% BSA . After 3 × 10 min washes at RT in PBS-Tr , secondary antibodies were incubated overnight at 4°C in PBS-Tr and 1% BSA . After 4 × 10 min washes in PBS-Tr at RT ( DAPI was added during the third wash ) and 2 × 5 min washes in PBS , samples were mounted with ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific #P36961 ) and imaged on a Leica SP8 confocal microscope . Images were deconvoluted using Huygens Professional . The following antibodies were used: anti-Piwi ( Brennecke et al . , 2007 ) , anti-Yb ( Saito et al . , 2010 ) , and anti-FG Nups ( Biolegend mAb414 ) . Acquired images were analysed on Fiji using custom scripts ( Source code 1 and 2 ) . Briefly , for Yb-body area measurements we extracted the relative channel , applied a threshold , and analysed particle number and size . A similar number of images was processed for all samples . For flam RNA-FISH analysis , we identified nuclei from the lamin staining applying a difference of Gaussian filter . We then isolated the RNA-FISH spots and counted the number present inside the nuclear envelope versus the total amount . Cytoplasm was identified via oligo-dT staining . A similar number of images was processed for all samples . Ribosomal RNAs were depleted using riboPOOLs against D . melanogaster rRNAs ( siTOOLs Biotech ) , according to the manufacturer’s instructions . Fly riboPOOLs were hybridised to 1 µg of RNA by adding 1 µl of resuspended riboPOOLs ( 100 µM ) , 5 µl of hybridisation buffer ( 10 mM Tris-HCl pH 7 . 5 , 1 mM EDTA , 2 M NaCl ) , and 1 µl of RNAse Inhibitor Plus ( Promega ) , incubating for 10 min at 68°C and cooling down slowly to 37°C . 80 µl of MyOne Streptavidin C1 beads ( Thermo Fisher ) for each sample were washed twice in 100 µl of Beads Resuspension Buffer ( 0 . 1 M NaOH , 0 . 05 M NaCl ) and twice in 100 µl of Beads Wash Buffer ( 0 . 1 M NaCl ) . The beads were resuspended in 160 µl of Depletion buffer ( 5 mM Tris-HCl pH 7 . 5 , 0 . 5 mM EDTA , 1 M NaCl ) and divided into two 80 µl aliquots . Hybridised riboPOOLs were added to 80 µl of washed beads , mixed well , and incubated for 15 min at 37°C , followed by 5 min at 50°C . The supernatant was transferred to the second tube with 80 µl of washed beads and incubated for 15 min at 37°C , followed by 5 min at 50°C . rRNA-depleted samples were transferred to a fresh tube and purified using Agencourt RNAClean XP beads ( Beckman Coulter A63987 ) . RNA-seq libraries were prepared using the NEBNext Ultra Directional Library Prep Kit for Illumina ( NEB #E7760 ) , according to the manufacturer’s instructions for ribosome-depleted RNA . DNA libraries were quantified with KAPA Library Quantification Kit for Illumina ( Kapa Biosystems ) and sequenced on an Illumina HiSeq 4000 instrument . Small RNA libraries from OSCs were generated as described previously with slight modifications ( McGinn and Czech , 2014 ) . Briefly , 19- to 28-nt small RNAs were purified by PAGE from 15 μg of total RNA from OSCs . Next , the 3' adapter ( containing four random nucleotides at the 5' end ) was ligated using T4 RNA ligase 2 , truncated KQ ( NEB ) . Following recovery of the products by PAGE purification , the 5' adapter ( containing four random nucleotides at the 3' end ) was ligated to the samples using T4 RNA ligase ( Ambion ) . Small RNAs containing both adapters were recovered by PAGE purification , reverse transcribed , and PCR amplified . Libraries were sequenced on an Illumina HiSeq 4000 instrument . PRO-seq was performed as described previously ( Mahat et al . , 2016 ) . 4 × 106 OSCs treated with siRNAs for 96 hr were used for nuclei isolation . Isolated nuclei were resuspended in storage buffer and stored at −80°C until further processing . Nuclear run-on reactions were performed with biotin-11-CTP and biotin-11-UTP and unlabelled ATP and GTP . Purified RNA samples were fragmented for 10 min on ice in 0 . 2 N NaOH and purified using Bio-Spin P30 columns ( Bio-Rad ) . Biotin-labelled RNA was purified using MyOne Streptavidin C1 Dynabeads ( Thermo Fisher Scientific ) , decapped using the RppH enzyme ( NEB ) , and purified by phenol/chloroform extraction . 3′ linkers and then 5′ linkers ( same as for small RNA cloning ) were ligated and biotinylated ligation products purified after each ligation using MyOne Streptavidin C1 Dynabeads . Reverse transcription was carried out using SuperScript III ( Thermo Fisher Scientific ) , and libraries were amplified using HS Phusion Flex polymerase . The libraries were sequenced on an Illumina HiSeq 4000 . DNA-FISH was carried out as described in Kishi et al . , 2019 . DNA-FISH probes were designed against a 10 kb region spanning the DIP1 and flamenco genomic loci ( chrX:21624796–21634619 ) using Oligominer ( Beliveau et al . , 2018 ) . The final set of 74 probes were completed with the addition of the SABER primer sequences at their 3′ ends ( tttCAACTTAAC ) and purchased from IDT . PER amplification was carried out as described by Kishi and colleagues ( Kishi et al . , 2019 ) and used directly for DNA-FISH . OSCs were plated on fibronectin-coated slides and fixed for 10 min in 4% PFA , permeabilised for 10 min in PBS , 0 . 5% TritonX-100 , and washed twice in PBS with 0 . 1% Tween-20 ( PBST ) . If necessary , DNase treatment was carried out at this stage by incubation with 4 µl of Turbo DNase in 100 µl of 1× Turbo DNase buffer for 30 min at 37°C . Cells were incubated 5 min in 0 . 1 N HCl , washed twice in PBST , and incubated in 2× SSCT ( 2× SSC with 0 . 1% Tween-20 ) with 50% formamide for 2 hr at 60°C . Cells were hybridised in 80 μl of ISH solution consisting of 2× SSCT , 50% formamide , 10% dextran sulfate , 400 ng/μl RNase A , and each PER extension at a final concentration of ~67 nM ( 1:15 dilution from 1 μM PER ) . After denaturation for 3 min at 80°C , cells were incubated overnight at 44°C in a humidified incubator . Hybridised samples were washed 4 × 5 min in prewarmed 2× SSCT at 60°C and then twice at RT . 80 µl of fluorescent hybridisation solution consisting of 1× PBS and 1 μM fluorescent imager strands were added to the samples and incubated for 1 hr at 37°C . Cells were washed 3 × 5 min in prewarmed PBS at 37°C , stained for 10 min at RT with DAPI ( 1:1000 dilution in PBS ) , and mounted using ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific #P36961 ) . Samples were imaged on a Leica SP8 confocal microscope ( 100× oil objective ) . 1 × 107 OSCs were nucleofected with 5 μg of the desired plasmid ( HALO-tagged Nup54 , Nup58 , Nxt1 , or a HALO-MCS control ) and crosslinked on ice with 400 mJ/cm2 at 254 nm . HALO-tag CLIP-seq was performed as previously described ( Munafò et al . , 2019 ) . Briefly , cell pellets were lysed in 300 µl of lLysis buffer ( 50 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 1% Triton X-100 , 0 . 1% deoxycholate , Protease Inhibitor [1:50 Promega] , and RNasin Plus [1:500 , Promega] ) , for 30 min at 4°C . DNase digestion was performed by adding 2 μl ( 4U ) of Turbo DNase to the cell lysate and immediately placing the samples at 37°C for 3 min , shaking at 1100 rpm . Samples were transferred to and kept on ice for >3 min , then cleared by centrifugation at top speed for 20 min at 4°C . Cell lysates were diluted up to 1 ml with 100 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , and incubated with 200 µl of Magne-HaloTag ( Promega G7282 ) beads overnight at 4°C . Beads were washed 2× in Wash Buffer A ( 100 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 0 . 05% IGEPAL CA-630 ) , 3× in Wash Buffer B ( PBS , 500 mM NaCl , 0 . 1% Triton X-100 , RNasin Plus 1:2000 ) , 3× in PBS , 0 . 1% Triton X-100 , and rinsed in Wash Buffer A . For release of the bait protein from the tag , beads were resuspended in 100 μl of 1X ProTEV Buffer , 1 mM DTT , and RNasin Plus ( 1:50 ) and 25 units of ProTEV Plus Protease ( Promega V6101 ) and incubated 2 hr at 30°C , shaking at 1300 rpm . The supernatant containing the eluted protein and the crosslinked RNA was transferred to a fresh tube , and 15 µl Proteinase K in 300 µl PK/SDS buffer ( 100 mM Tris , pH 7 . 5; 50 mM NaCl; 1 mM EDTA; 0 . 2% SDS ) were added to the eluate and incubated 1 hr at 50°C . RNA was isolated via phenol/chloroform extraction , resuspended in 8 µl of nuclease-free water , and used for library preparation . Library preparation for CLIP-seq samples was carried out with the SMARTer Stranded RNAseq kit ( Takara Bio 634839 ) , according to the manufacturer’s instructions . For small RNA-seq , adapters were clipped from raw fastq files with fastx_clipper ( adapter sequence AGATCGGAAGAGCACACGTCTGAACTCCAGTCA ) keeping only reads with at least 23 bp length . The first and last four bases were trimmed using seqtk ( https://github . com/lh3/seqtk; Ramírez et al . , 2016 ) . After removal of cloning markers and reads mapping to rRNAs and tRNAs ( list downloaded from FlyBase ) , high-quality reads were aligned to the D . melanogaster genome release 6 ( dm6; downloaded from FlyBase ) ( Hoskins et al . , 2015 ) using STAR ( Dobin et al . , 2013 ) . For transposon-wide analysis , genome multi-mapping reads were randomly assigned to one location using option '--outFilterMultimapNmax 1000 --outSAMmultNmax 1 --outMultimapperOrder Random' and non-mapping reads were removed . Small RNA-seq reads were normalised to miRNA reads in the control library ( set to rpm ) . Only high-quality small RNA reads with a length between 23 and 29 bp were used for the piRNA genome browser shots . piRNA distribution was calculated and plotted in R . For RNA-seq , raw fastq files generated by Illumina sequencing were analysed by a pipeline developed in-house . In short , the first five bases of each 50 bp read were removed using fastx trimmer ( http://hannonlab . cshl . edu/fastx_toolkit/ ) . After removal of reads mapping to Drosophila rRNA using STAR , high-quality reads were aligned to the D . melanogaster genome release 6 ( dm6; downloaded from FlyBase ) ( Hoskins et al . , 2015 ) using STAR ( Dobin et al . , 2013 ) . For transposon-wide analysis , genome multi-mapping reads were randomly assigned to one location using option '--outFilterMultimapNmax 1000 --outSAMmultNmax 1 --outMultimapperOrder Random' and non-mapping reads were removed . For genome-wide analyses , multi-mapping reads were removed to ensure unique locations of reads . Normalisation was achieved by calculating rpm using the deepTools2 ( Ramírez et al . , 2016 ) bamCoverage . Differential expression analysis was performed using DESeq2 ( Love et al . , 2014 ) and plotting was done in R using ggplot2 ( https://ggplot2 . tidyverse . org ) . For genes carrying ‘nearby transposon insertions’ , we considered only gene promoters within −10 kb to +15 kb of a TE insertion on the same genomic strand . Coordinates used for the uni-strand piRNA clusters are chrX:21631891–22282863 ( flam ) and chrX:21520428–21556793 ( 20A ) . Coordinates used for the dual-strand piRNA clusters were from Andersen et al . , 2017 . For piRNA clusters 1 kb bin analysis , each cluster was divided into non-overlapping 1 kb bins and only those with a mappability score above 0 . 8 were retained . Uniquely mapped reads were counted using HTSeq ( Anders et al . , 2015 ) , normalised to the total number of genome-mapped reads , and the per-window log2 fold-change between each knockdown and its control was calculated and plotted in R . The mappability was calculated as described in Derrien et al . , 2012 . For OSC RNA-seq , bins with 0 rpm in more than one sample were discarded . For ovary RNA-seq analysis , a pseudo-count of 0 . 01 was added to each bin and the bins with 0 rpm only in the control were discarded . For flam 100 kb bins analysis , the dm6 genome was divided into 100 kb sliding windows using 1 kb steps . Mappability for a window was defined as the fraction of all possible 50-mers derived from the window that aligned uniquely to it using STAR . Windows with mappability >0 . 05 and located fully within flam ( n = 285 ) were kept for the analysis . For each sample , reads aligning uniquely to the sense strand and at least 50% within a window were counted and subsequently normalised to reads per 1 million uniquely aligned reads . The per-window log2 fold-change between each knockdown and its control was calculated using a pseudo-count of 1 . Results from four individual replicates per knockdown were highly consistent with the results shown from the pooled analysis . Data visualisation and analyses were done using R and the following packages: ggplot2 , DEseq2 , qPLEXanalyzer . The UCSC genome browser was used to display high-throughput sequencing data and to prepare coverage plots shown in the article . n indicated in the figure legends refers to the number of independent biological replicates . Bar graphs display average of n biological replicates and standard deviation ( SD ) , and p values were calculated with an unpaired t-test using GraphPad . Box plots display median , first , and third quartiles ( box ) and highest/lowest value within 1 . 5 interquartile range ( whiskers ) ; dots represent potential outliers beyond 1 . 5 * interquartile range .	Transposons are genetic sequences , which , when active , can move around the genome and insert themselves into new locations . This can potentially disrupt the information required for cells to work properly: in reproductive organs , for example , transposon activity can lead to infertility . Many organisms therefore have cellular systems that keep transposons in check . Animal cells comprise two main compartments: the nucleus , which contains the genetic information , and the cytosol , where most chemical reactions necessary for life take place . Molecules continually move between nucleus and cytosol , much as people go in and out of a busy train station . The connecting ‘doors’ between the two compartments are called Nuclear Pore Complexes ( NPCs ) , and their job is to ensure that each molecule passing through reaches its correct destination . Recent research shows that the individual proteins making up NPCs ( called nucleoporins ) may play other roles within the cell . In particular , genetic studies in fruit flies suggested that some nucleoporins help to control transposon activity within the ovary – but how they did this was still unclear . Munafò et al . therefore set out to determine if the nucleoporins were indeed actively silencing the transposons , or if this was just a side effect of altered nuclear-cytosolic transport . Experiments using cells grown from fruit fly ovaries revealed that depleting two specific nucleoporins , Nup54 and Nup58 , re-activated transposons with minimal effects on most genes or the overall health of the cells . This suggests that Nup54 and Nup58 play a direct role in transposon silencing . Further , detailed analysis of gene expression in Nup54- and Nup58-lacking cells revealed that the product of one gene , flamenco , was indeed affected . Normally , flamenco acts as a ‘master switch’ to turn off transposons . Without Nup54 and Nup58 , the molecule encoded by flamenco could not reach its dedicated location in the cytosol , and thus could not carry out its task . These results show that , far from being mere ‘doorkeepers’ for the nucleus , nucleoporins play important roles adapted to individual tissues in the body . Further research will help determine if the same is true for other organisms , and if these mechanisms can help understand human diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "cell", "biology" ]	2021	Channel nuclear pore complex subunits are required for transposon silencing in Drosophila
Centrioles are cylindrical cell organelles with a ninefold symmetric peripheral microtubule array that is essential to template cilia and flagella . They are built around a central cartwheel assembly that is organized through homo-oligomerization of the centriolar protein SAS-6 , but whether SAS-6 self-assembly can dictate cartwheel and thereby centriole symmetry is unclear . Here we show that Leishmania major SAS-6 crystallizes as a 9-fold symmetric cartwheel and provide the X-ray structure of this assembly at a resolution of 3 . 5 Å . We furthermore demonstrate that oligomerization of Leishmania SAS-6 can be inhibited by a small molecule in vitro and provide indications for its binding site . Our results firmly establish that SAS-6 can impose cartwheel symmetry on its own and indicate how this process might occur mechanistically in vivo . Importantly , our data also provide a proof-of-principle that inhibition of SAS-6 oligomerization by small molecules is feasible . Eukaryotic flagella and cilia have essential roles in cell locomotion , fluid movement , and sensing . Their formation strictly requires the presence of centrioles , cylindrical organelles with a radially ninefold symmetric , peripheral microtubule array ( Bettencourt-Dias and Glover , 2007; Azimzadeh and Marshall , 2010; Gonczy , 2012 ) . During ciliogenesis/flagellogenesis , centrioles ( then referred to as basal bodies ) dock to the cell membrane via their distal ends and nucleate the flagellar/ciliar axoneme from their microtubule array ( Sherwin and Gull , 1989; Dawe et al . , 2007; Ishikawa and Marshall , 2011; Avasthi and Marshall , 2012; Chemes , 2012 ) . The symmetry and diameter of centrioles ( and thereby of flagella/cilia ) is , to a large extent , established through scaffolding by the centriolar cartwheel , a structure with a central ring-like hub and nine radially projecting spokes that contact the peripheral centriolar microtubules ( Azimzadeh and Marshall , 2010; Brito et al . , 2012; Gonczy , 2012 ) . The cartwheel is organized through the homo-oligomerization of the highly conserved centriolar protein SAS-6 ( Nakazawa et al . , 2007; Kitagawa et al . , 2011; van Breugel et al . , 2011 ) . High-resolution crystal structures of zebrafish and C . reinhardtii SAS-6 fragments together with biochemical and biophysical characterizations ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) demonstrated that two dimerization interfaces in SAS-6 mediate this oligomerization: a coiled-coil domain that forms a rod-like parallel dimer; and a globular N-terminal domain that forms a curved head-to-head dimer . Modeling these two dimer interactions together in silico resulted in ring assemblies that were compatible with the symmetry and dimension of cartwheels observed in vivo . In these models , the N-terminal head domains constitute the cartwheel hubs from which the coiled-coil rods project to form the cartwheel spokes ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) . However , so far , no high-resolution structure of the SAS-6 cartwheel is available . Although rotary shadowing EM studies suggested that SAS-6 might be able to form cartwheels , the available resolution was insufficient to determine the symmetry of the observed assemblies directly ( Kitagawa et al . , 2011 ) . Another , higher resolution , EM study with recombinant SAS-6 revealed cartwheel-like assemblies with an eightfold , not a ninefold symmetry ( van Breugel et al . , 2011 ) , while a third EM study showed the presence of SAS-6 tetramers ( Gopalakrishnan et al . , 2010 ) . Furthermore , the available biochemical and biophysical data do not provide evidence for efficient cartwheel formation by SAS-6 in solution ( Gopalakrishnan et al . , 2010; Kitagawa et al . , 2011; van Breugel et al . , 2011 ) , raising the question of whether SAS-6 alone is sufficient to organize ninefold symmetric cartwheels and , if so , under what conditions . SAS-6 is a highly conserved protein that is found in all eukaryotes that have cilia/flagella during some of their life-cycle stages ( Carvalho-Santos et al . , 2010; Hodges et al . , 2010 ) . Inhibiting SAS-6 self-assembly by point mutations in vivo abolishes the formation of centrioles ( Kitagawa et al . , 2011; van Breugel et al . , 2011; Lettman et al . , 2013 ) and thereby flagella ( van Breugel et al . , 2011 ) . Thus , targeting SAS-6 oligomerization by inhibitors could be a strategy to disable flagellogenesis in organisms that cause human disease and rely on flagella for their pathogenicity . The Trypanosomatids are of special interest in this respect . They consist of parasitic , eukaryotic protozoa with a single flagellum and comprise members that cause major human diseases , such as sleeping sickness ( Trypanosoma brucei ) , Chagas disease ( Trypanosoma cruzi ) , and Leishmaniasis ( Leishmania spec . ) ( Simpson et al . , 2006 ) . These organisms display complex life cycles in which they shuttle between an insect vector and human or animal hosts . Their flagellum provides key roles in this life cycle through its multiple functions in essential cellular processes such as motility , signalling , sensing , and attachment ( Vaughan and Gull , 2003; Ralston et al . , 2009; Gluenz et al . , 2010 ) . Furthermore , the flagellar pocket , a membrane invagination from which the flagellum emerges , is the exclusive site of vesicular membrane traffic in these organisms ( Overath et al . , 1997; Landfear and Ignatushchenko , 2001; Field and Carrington , 2004 , 2009; Overath and Engstler , 2004 ) . It is therefore not surprising that flagella have been demonstrated or are proposed to be key factors in the pathogenicity of these organisms ( Vaughan and Gull , 2003; Ralston et al . , 2009; Gluenz et al . , 2010 ) . As in other eukaryotes , the flagellum of the Trypanosomatids is templated from centrioles ( basal bodies ) . Tomographic EM studies demonstrated the presence of canonical , ninefold symmetric cartwheel structures in the centre of their centrioles ( Lacomble et al . , 2009 ) . Currently , no high-resolution structures of SAS-6 homologues from Trypanosomatids are available , and it is unclear whether oligomerization of their SAS-6 homologues could be inhibited . The genomes of Trypanosoma brucei , Trypanosoma cruzi , and Leishmania major have recently been sequenced ( Berriman et al . , 2005; El-Sayed et al . , 2005; Ivens et al . , 2005 ) . BLAST searches identified the likely SAS-6 homologues in these organisms with similar domain architectures ( Figure 1A , Figure 1—figure supplement 1 ) and sequence identities to human SAS-6 of 21 . 0 ± 1 . 4% in 459 ± 21 aligned residues . Multiple sequence alignment of their N-terminal domains ( Figure 1—figure supplement 1B ) shows that key residues are well conserved compared to zebrafish SAS-6 , the closest homologue of human SAS-6 for which high-resolution structures are available ( van Breugel et al . , 2011 ) . Different from zebrafish SAS-6 , they have long N-terminal extensions ( Figure 1A , Figure 1—figure supplement 1A , B ) that vary in length and are poorly conserved . Since these extensions hindered our crystallization attempts , we largely removed them in the constructs used in this manuscript . The part of the extensions still present did not show electron density in our crystal structures . 10 . 7554/eLife . 01812 . 003Figure 1 . Structural and biophysical characterization of L . major SAS-6 . ( A ) Domain overview of L . major SAS-6 . Lines indicate constructs that were used in this work . ( B and C ) Left: ribbon presentation of the head-to-head dimers of L . major SAS-6’s N-terminal domain present in the SAS-697–274 crystal . Shown are the dimers formed between chain B and chain C ( B ) and chain A and symmetry-related chain A ( C ) . α-helices ( α ) and β-sheets ( β ) are numbered sequentially . Right: detailed views of the corresponding dimerization interfaces . Interface residues are labelled and shown in sticks , dotted orange lines indicate hydrogen bonds . The two dimers show largely identical side-chain orientations in their interfaces . Note , however , that F257 ( ringed in blue ) in the B–C dimer inserts into a hydrophobic pocket , while in the A–A dimer Y215 is flipped into this pocket and displaces F257 . To better illustrate this , a semi-transparent molecular surface of one of the subunits is also presented ( grey ) . ( D ) Sedimentation-equilibrium analytical ultracentrifugation data for 400 µM L . major SAS-697–274 wild-type ( blue circles ) and F257E mutant ( green circles ) and Y215K mutant ( red circles ) obtained at 11300 , 17000 , and 21200 rpm . Data for the F257E mutant were fitted to an ideal single-species model ( solid line ) . Analysis of multiple concentrations gave a molecular weight of 17 , 727 ± 219 Da , close to the expected molecular weight for the monomer of 19 , 640 Da . As initial fits to a similar model for the WT and Y215K data gave higher molecular weights of 27 , 589 ± 209 Da and 30 , 951 ± 595 respectively , the data were fitted to a monomer–dimer equilibrium model ( solid line ) giving dissociation constants , KD , of 622 ± 70 µM for the WT and 190 ± 27 µM for Y215K mutant . The plots on the right show the residuals of the fits to the data for the wild-type ( blue circles ) and the corresponding F257E mutant ( green circles ) and Y215K mutant ( red circles ) . ( E ) Left: ribbon presentation of the L . major SAS-697–320 F257E coiled-coil dimer structure ( chain A: red , chain B: green ) . Right: detailed view of the region boxed on the left . Interaction interface between N-terminal head domain and the coiled-coil stalk . Residues that make contact are labelled and are shown as sticks , dotted orange lines indicate hydrogen bonds . DOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 00310 . 7554/eLife . 01812 . 004Figure 1—figure supplement 1 . L . major SAS-6 and Danio rerio SAS-6 are highly similar . ( A ) Schematic representation of Leishmania major and Danio rerio SAS-6 . Both proteins show a similar overall architecture with an N-terminal head domain and a coiled-coil domain of comparable lengths . The bar indicates the aligned SAS-6 region shown in ( B ) . ( B ) Multiple sequence alignment of the N-terminal head domains of SAS-6 from Trypanosoma brucei , Trypanosoma cruzi , Leishmania major , and Danio rerio . The numbering refers to L . major SAS-6 . The alignment is colored according to the Clustal coloring scheme . Red stars indicate key residues of the interaction interface of the homo-dimer of this domain . ( C ) Overlay of the structures of L . major SAS-697–274 ( green ) and Danio rerio N-SAS-61–156 ( blue ) . Face-on view onto the hydrophobic pocket into which a highly conserved phenylalanine ( F257 in L . major , F131 in D . rerio SAS-6 ) is inserted in the homo-dimerized form of this domain . Residues of this pocket are labelled and are shown as sticks . ( D ) Detailed view of the interaction interface between N-terminal head domain and the coiled-coil stalk of SAS-6 . Overlay of this interface from L . major SAS-697–320 F257E ( green ) with that from Danio rerio N-SAS-61–179 F131D ( blue ) . Residues that make contact are labelled and are shown as sticks , dotted yellow lines indicate hydrogen bonds . DOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 00410 . 7554/eLife . 01812 . 005Figure 1—figure supplement 2 . The closed Y215 conformation is observed in the low affinity head-to-head homo-dimer of the F257E mutant . ( A ) Ribbon presentation of the SAS-6 octamer present in the L . major SAS-697–320 F257E crystal . The ASU of the crystal contained four SAS-6 monomers that are labelled from A–D . The arrow specifies the view direction shown as a close-up in ( B ) . ( B ) Left: ribbon presentation of the B–C head-to-head homo-dimer in the L . major SAS-697–320 F257E crystal . Right: corresponding detailed view of the homo-dimer interface . The other head-to-head dimers in the crystal showed similar arrangements to the ones presented here . Interface residues are labelled and shown in sticks . Dotted yellow lines indicate hydrogen bonds . Highlighted in lemon and ringed in blue is the mutated E257 residue . Note that Y215 is swung into the hydrophobic pocket , whereas E257 points outwards . The remainder of the interface residues show similar side-chain orientations as observed in the wild-type SAS-697–274 crystal ( Figure 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 005 To elucidate the structural organization of L . major SAS-6 , we solved the structure of its N-terminal domain ( Lm SAS-697–274 ) by X-ray crystallography to a resolution of 2 . 2 Å ( Table 1 , Table 2; Figure 1B ) . The asymmetric unit ( ASU ) of the crystal contained three molecules that were virtually identical to each other ( 139 ± 2 selected pairs superpose with an rmsd of 0 . 70 ± 0 . 13 Å in secondary structure matching ) . Our structure demonstrates a high similarity of L . major SAS-6 to previously solved SAS-6 structures ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) ( secondary structure matching to the D . rerio SAS-6 head domain results in an rmsd of 1 . 59 ± 0 . 09 Å with 133 ± 2 selected pairs ) . Like these , the N-terminal domain of L . major SAS-6 consists of a 7-stranded β-barrel , capped by a helix-turn-helix motif , that forms a curved cross-handshake homo-dimer within the crystal through a highly conserved interaction interface ( Figure 1B , Figure 1—figure supplement 1B , C ) . In this dimer , the β-hairpins formed by β-strands β6 and β7 pack antiparallelly against each other . Phenylalanine 257 at the tip of this hairpin is inserted into a conserved hydrophobic pocket that is constituted by the helix-turn-helix motif and the base of this hairpin in the B-C homo-dimer ( formed by chain B and chain C in the crystal , Figure 1B ) . Dimerization is also observed in solution , as judged by equilibrium ultracentrifugation , and then depends on the presence of F257 ( Figure 1D ) . The measured KD of this dimerization is ∼600 μM and therefore approximately 5- to 10-fold weaker than seen for other SAS-6 homologues ( Kitagawa et al . , 2011; van Breugel et al . , 2011 ) . 10 . 7554/eLife . 01812 . 006Table 1 . Native dataset analysis and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 006L . major SAS-697–274 WTL . major SAS-697–320 F257EL . major SAS-697–320 WTBeamlineDiamond I04ESRF ID29ESRF BM14Space GroupP43212C121H3Wavelength ( Å ) 0 . 97940 . 900 . 97813Monomers in the asymmetric unit346Unit Cell dimensions ( Å ) a = 84 . 25 b = 84 . 25 c = 239 . 94 α = 90 . 0 β = 90 . 0 γ = 90 . 0a = 108 . 9 b = 81 . 25 c = 133 . 1 α = 90 . 0 β = 91 . 5 γ = 90 . 0a = 482 . 7 b = 482 . 7 c = 43 . 13 α = 90 . 0 β = 90 . 0 γ = 120 . 0Resolution ( Å ) 48 . 9–2 . 246 . 91–2 . 9/3 . 4 ( anisotropy ) 66 . 9–3 . 5/4 . 2 ( anisotropy ) Completeness ( overall/inner/outer shell ) 99 . 9/98 . 9/10099 . 9/99 . 1/99 . 9100/99 . 1/100Rmerge ( overall/inner/outer shell ) 0 . 144/0 . 064/1 . 5050 . 152/0 . 032/2 . 5700 . 335/0 . 074/3 . 061Rpim ( overall/inner/outer shell ) 0 . 065/0 . 030/0 . 6570 . 061/0 . 013/1 . 0330 . 073/0 . 016/0 . 704Mean I/σI ( overall/inner/outer shell ) 7 . 6/18 . 6/1 . 410 . 6/41 . 4/0 . 98 . 2/33 . 3/1 . 5Multiplicity ( overall/inner/outer shell ) 5 . 9/5 . 5/6 . 07 . 2/6 . 7/7 . 122 . 3/21 . 9/19 . 8Number of reflections45 , 56924 , 09048 , 521Number of atoms353656538777Waters14100Rwork/Rfree ( % data used ) 20 . 8/24 . 5 ( 5 . 0% ) 23 . 7/25 . 6 ( 5 . 0% ) 22 . 4/24 . 2 ( 5 . 1% ) rmsd from ideal values: bond length/angles0 . 010/1 . 3490 . 008/1 . 2550 . 006/1 . 348Mean B value51 . 592 . 8150 . 6Average Real-space correlation coefficient0 . 9760 . 9240 . 881Molprobity Score1 . 14 ( 100th percentile ) 1 . 33 ( 100th percentile ) 1 . 41 ( 100th percentile ) 10 . 7554/eLife . 01812 . 007Table 2 . SeMet L . major SAS-697−274 WT dataset analysisDOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 007BeamlineDiamond I03Space groupP43212Wavelength ( Å ) 0 . 9794 ( Peak ) 0 . 9796 ( Inflection ) 0 . 9393 ( Remote ) Unit Cell dimensions ( Å ) a = 84 . 19 b = 84 . 19 c = 239 . 6 α = 90 . 0 β = 90 . 0 γ = 90 . 0a = 84 . 24 b = 84 . 24 c = 239 . 7 α = 90 . 0 β = 90 . 0 γ = 90 . 0a = 84 . 22 b = 84 . 22 c = 239 . 7 α = 90 . 0 β = 90 . 0 γ = 90 . 0Resolution ( Å ) 68 . 9–2 . 368 . 9–2 . 368 . 9–2 . 3Completeness ( overall/inner/outer shell ) 99 . 8/99 . 9/99 . 799 . 7/99 . 9/99 . 799 . 8/99 . 7/99 . 7Rmerge ( overall/inner/outer shell ) 0 . 130/0 . 066/0 . 9420 . 122/0 . 048/1 . 0010 . 131/0 . 052/0 . 973Rpim ( overall/inner/outer shell ) 0 . 066/0 . 036/0 . 4670 . 062/0 . 026/0 . 4970 . 066/0 . 028/0 . 486Mean I/sd ( I ) ( overall/inner/outer shell ) 7 . 4/18 . 8/1 . 77 . 8/21 . 8/1 . 77 . 2/20 . 2/1 . 6Multiplicity ( overall/inner/outer shell ) 4 . 8/4 . 1/4 . 94 . 7/4 . 1/4 . 94 . 7/4 . 0/4 . 8Se sites found/expected14/12 To ascertain the role of the coiled-coil domain of L . major SAS-6 , we tried to crystallize constructs that included both the N-terminal domain and parts of the coiled-coiled domain , but initially failed to obtain diffraction-grade crystals . However , by introducing the F257E mutation to strongly weaken head-to-head dimerization we managed to crystallize construct Lm SAS-697–320 F257E that contained the N-terminal head domain and the first seven heptad-repeats of the coiled-coil domain . We subsequently solved its X-ray structure to a resolution of 2 . 9 Å ( Table 1; Figure 1E ) . The asymmetric unit of the crystal contained four molecules that were highly similar to each other and superposed with an rmsd of 1 . 17 ± 0 . 50 Å in secondary structure matching with 165 ± 16 selected pairs . The crystal structure revealed that the L . major SAS-6 coiled-coil domain is a parallel dimer and packs via conserved interactions against the N-terminal head-domains as seen in other SAS-6 homologues ( Figure 1E , Figure 1—figure supplement 1D ) . Thus , our structural analyses demonstrate that L . major SAS-6 is highly similar to other SAS-6 homologues . Our structural analysis also revealed the presence of an alternative arrangement of the head-to-head homo-dimer of L . major SAS-6 . In the Lm SAS-697–274 crystal , the A–A homo-dimer ( formed by chain A and symmetry-related chain A ) shares the features of the B–C homo-dimer described above . However , in the A–A homo-dimer , the Y215 sidechain is flipped into its own hydrophobic pocket resulting in the displacement from this pocket of residue F257 of its homo-dimer partner ( Figure 1C ) probably weakening the interaction . Intriguingly , we found a similar arrangement in the Lm SAS-697–320 F257E crystal . The F257E mutation abolishes head-to-head dimerization in solution ( Figure 1D ) , yet , in the crystal , due to the high protein and precipitant concentrations , some of the SAS-6 molecules present are found nevertheless in head-to-head dimers and form a curved octamer . In this octamer residues Y215 are also swung into their own hydrophobic pockets , while E257 point away from them ( Figure 1—figure supplement 2 ) . These data suggest that the closed Y215 conformation corresponds to a low-affinity dimerization state of the head domains and might therefore constitute a potential regulatory mechanism of SAS-6 oligomerization . In solution , the presence of a dimerization-impaired state could also explain the relatively low affinity of head-to-head dimerization of wild-type L . major SAS-6 apparent in analytical ultracentrifugation ( ∼600 μM compared to 50–100 μM observed for other species ) since other SAS-6 homologues do not have aromatic residues at the equivalent position of Y215 that could play such a role . To test whether the Y215 closed conformation significantly compromises dimerization in solution , we mutated Y215 in Lm SAS-697–274 to lysine that is unable to block the hydrophobic pocket in a similar way . Subsequently , we subjected the purified protein to analytical ultracentrifugation ( Figure 1D ) . The measured KD of head-to-head dimerization of the Y215K mutant was ∼200 μM and therefore approximately threefold lower than for the corresponding wild-type protein . We conclude that Y215 acts to partially inhibit head-to-head dimerization of L . major SAS-6 . The presence of curved SAS-6 octamers in the crystal of the Lm SAS-697–320 F257E mutant that is strongly impaired in its ability to form head-to-head dimers suggests that wild-type versions of SAS-6 would adopt even larger assemblies . The low affinity of head-to-head dimerization together with the concomitant sample heterogeneity makes EM studies of the resulting assemblies technically challenging . To overcome these limitations we tried to crystallize L . major SAS-6 constructs with both dimerization interfaces intact . Obtaining crystals that diffracted well enough to solve their structure proved difficult . However , with the wild-type construct of Lm SAS-697–320 , we finally succeeded to find a crystal form of the space-group H3 that diffracted to a resolution of ∼3 . 5 Å along the l-axis with anisotropy limiting the resolution along the h-k plane to ∼4 . 2 Å . Using the structure of the Lm SAS-697–274 B–C homo-dimer as a search model , a molecular replacement solution could be found that allowed the subsequent placement of the coiled-coil part present in the construct ( Figure 2 ) . The unit cell of the Lm SAS-697–320 crystal form contained three 9-fold symmetric SAS-6 rings with three SAS-6 dimers constituting the ASU . These three dimers were highly similar to each other and overlayed with an rmsd of 0 . 88 ± 0 . 29 Å in secondary structure matching with 357 ± 14 selected pairs . No electron density was seen inside the SAS-6 rings . The inner diameters of the SAS-6 rings are ∼19 nm and correspond well to the diameters of cartwheel hubs observed in vivo ( Lacomble et al . , 2009; Guichard et al . , 2010 , 2012 ) . In the crystal , SAS-6 rings are stacked onto each other . Neighbouring rings interact through inter-digitation of their coiled-coil domains in an antiparallel way ( Figure 2—figure supplement 1 ) . We confirmed our structural model by calculating a phased anomalous map using the refined phases based on our model and the amplitudes of an isomorphic dataset collected from crystals of the selenomethionine derivative ( Figure 2B; Table 3 ) . Peaks in this map correlated with the positions of methionines in our model . Thus , under appropriate conditions , L . major SAS-6 can adopt ninefold symmetric rings that are highly similar to cartwheels observed in vivo . 10 . 7554/eLife . 01812 . 008Figure 2 . L . major SAS-697–320 crystallizes as a ninefold symmetric ring with dimensions similar to those of centriolar cartwheels observed in vivo . ( A ) Ribbon presentation of the L . major SAS-697–320 structure . Shown is the ring assembly present in the unit cell of the L . major SAS-697–320 crystal . Protein chains are colored alternatingly in green and red to allow easier comparison with Figure 1 . Top: side-view , bottom: face-on view of the L . major SAS-697–320 ring structure . The nonagon in the center of the ring indicates the ( quasi- ) ninefold symmetry axis . The ASU of the crystal contained six SAS-6 monomers that are labelled from A–F . No clear electron density could be seen for the distal part of the coiled-coil of the A–B dimer , probably due to the lack of stabilizing crystal packing interactions compared to the C–D and E–F dimer ( Figure 2—figure supplement 1B ) . ( B ) Detailed view of the region boxed in ( A ) . Shown in blue sticks are the methionine-side chains of the SAS-697–320 model . In magenta , iso-mesh representation of the phased anomalous difference map at a contour level of σ = 5 showing the selenium positions in the crystallized selenomethionine derivate of L . major SAS-697–320 . DOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 00810 . 7554/eLife . 01812 . 009Figure 2—figure supplement 1 . The crystal packing interactions observed in the wild-type L . major SAS-697–320 crystal . ( A ) Stacking of the L . major SAS-697–320 rings in the crystal . Rings are stacked directly onto each other through interactions of the N-terminal domains of SAS-6 . Left: face-on view , right: turned by 90° around the x-axis . ( B ) Lateral interactions of the L . major SAS-697–320 rings in the crystal . Top: face-on view , the arrow indicates the view direction shown at the bottom as a close-up . The coiled-coil stalks interact in an antiparallel way with each other . Their termini touch the head domains of adjacent rings . DOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 00910 . 7554/eLife . 01812 . 010Figure 2—figure supplement 2 . The interfaces critical for ring formation in the wild-type L . major SAS-697–320 crystal are similar in orientation to those observed in the SAS-697–320 F257E and the SAS-697–274 crystals . ( A ) Overlay of the wild-type L . major SAS-697–320 ring structure ( green ) with the two dimers present in the ASU of the L . major SAS-697–320 F257E crystal ( A–B dimer in magenta , C–D dimer in red ) . The SAS-697–320 F257E A–B/C–D dimer superposed to wild-type L . major SAS-697–320 with an rmsd of 0 . 91 Å/1 . 62 Å in secondary structure matching with 330/319 selected pairs respectively . ( B ) Overlay of the wild-type L . major SAS-697–320 ring structure ( green ) with the B–C dimer present in the ASU of the L . major SAS-697–274 crystal ( blue ) . Secondary structure matching of these two structures resulted in an rmsd of 0 . 77 Å ( 283 selected pairs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 01010 . 7554/eLife . 01812 . 011Table 3 . SeMet L . major SAS-697−320 WT dataset analysisDOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 011BeamlineESRF BM14Space groupH3Wavelength ( Å ) 0 . 97872 ( Peak ) Unit Cell dimensions ( Å ) a = 481 . 7 b = 481 . 7 c = 42 . 9 α = 90 . 0 β = 90 . 0 γ = 120 . 0Resolution ( Å ) 48 . 2–4 . 0/5 . 4 ( anisotropy ) Completeness ( overall/inner/outer shell ) 100 . 0/98 . 8/100 . 0Rmerge ( overall/inner/outer shell ) 0 . 194/0 . 054/2 . 118Rpim ( overall/inner/outer shell ) 0 . 091/0 . 026/0 . 987Mean I/sd ( I ) ( overall/inner/outer shell ) 3 . 8/13 . 7/0 . 9Multiplicity ( overall/inner/outer shell ) 5 . 6/5 . 5/5 . 6 To find out whether we could inhibit SAS-6 oligomerization , we conducted a small-scale fragment screen using a custom library of halogenated fragments ( HEFLib ) ( Wilcken et al . , 2012 ) . First , pools of compounds were screened for their ability to bind to and thereby cause a shift perturbation in the {1H , 15N}-HSQC NMR spectrum of 15N labelled Lm SAS-6 . To avoid ambiguities in the interpretation of shift perturbations that could stem from the partial presence of oligomers , we used the Lm SAS-697–274 F257E mutant for these binding studies that is monomeric in solution ( Figure 1D ) . One dimensional spin-echo experiments provided a crude estimate of 18 ms for the backbone amide 1H T2 relaxation time constants , consistent with the molecular mass of the construct ( ∼20 kDa ) ( Anglister et al . , 1993 ) . To determine the putative interaction sites of binding candidates , we assigned the backbone resonances of Lm SAS-697–274 F257E using 13C , 15N double-labelled protein and mapped the HSQC chemical shift perturbations onto the crystal structure of wild-type Lm SAS-697–274 . Strong perturbations in chemical shifts are most consistent with compound PK9119 ( ( 5-bromo-7-ethyl-1H-indol-3-ylmethyl ) -dimethyl-amine , Figure 3A ) binding adjacent to the head-to-head dimerization interface of Lm SAS-697–274 ( Figure 3B , Figure 3—figure supplement 1A , B ) . Smaller , but significant shift changes indicate that binding may alter this interface by affecting the conformation of the helix-turn-helix motif that constitutes a part of it . We also examined aromatic side chain 1H resonances of the Phe and Tyr residues using Cβ–Hδ correlation maps , which enabled us to identify two side-chains ( F199 , F212 ) that are perturbed on binding of PK9119 ( Figure 3B , Figure 3—figure supplement 1C ) . These two side-chains cluster together around the helix-turn-helix motif . HSQC chemical shift titration experiments with PK9119 and L . major SAS-697–274 F257E suggest a millimolar binding affinity; the low solubility of PK9119 in aqueous solutions and lack of a reference compound for competition binding assays makes more accurate KD determinations technically challenging . 10 . 7554/eLife . 01812 . 012Figure 3 . A small chemical compound can inhibit SAS-6 oligomerization . ( A ) The chemical structure of compound PK9119 as a structure formula ( top ) or three-dimensional model ( bottom ) . ( B ) Heat map of the chemical shift perturbations in the {1H-15N}-HSQC spectrum of 15N-labelled L . major SAS-697–274 F257E in the presence of 2 mM PK9119 . Data are plotted onto the crystal structure of wild-type L . major SAS-697–274 . Higher shift perturbation is depicted in warmer colors , whilst prolines ( not observable in the HSQC ) are colored grey , and unassigned/untraced residues are colored white . The magenta F257 is from the homo-dimer partner and is inserted into the hydrophobic pocket of the dimerization interface . Note that the chemical shift perturbations cluster close to this pocket . Side-chains are drawn for F199 and F212 that showed robust perturbations in ( HB ) CB ( CGCD ) HD correlation spectra in the presence of 1 mM PK9119 ( Figure 3—figure supplement 1C ) . ( C ) SEC-MALS chromatogram of L . major SAS-697–424 showing the refractive index signal with the derived molar masses indicated by the thicker horizontal lines . L . major SAS-697–424 displayed a distribution of masses from that of the dimer up to >200 kDa consistent with a concentration driven self-association equilibrium . In the presence of 1 mM PK9119 the maximal mass was almost halved . The F257E mutant displayed a constant mass of 71 kDa in the absence and presence of PK9119 consistent with the mass of a dimer of L . major SAS-697–424 . All samples were injected on SEC-MALS at 25 mg/ml ( 675 μM in monomer ) . Due to dilution during SEC the peak concentrations achieved were a factor of ∼10 lower than this ( ∼68 μM in monomer ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 01210 . 7554/eLife . 01812 . 013Figure 3—figure supplement 1 . Chemical shift perturbation of SAS-6 by PK9119 . ( A ) {1H-15N}-HSQC overlay of 15N-labelled L . major SAS-697–274 F257E in the presence or absence of 1 mM or 2 mM PK9119 . ( B ) Graphical representation of the chemical shift perturbations from the data in panel A , 2 mM PK9119 . Perturbation is quantified as a weighted combination Δδ1H + ( Δδ15N/5 ) . The blue line indicates that the trajectory of the S170 correlation is not quantified , as its peak could not be traced in the presence of PK9119 . ( C ) Overlay of ( HB ) CB ( CGCD ) HD spectra for 13C/15N labelled protein in the presence or absence of 1 mM PK9119 , showing Cβ–Hδ correlations for Phe and Tyr side chain resonances . Cβ resonance frequencies of three unperturbed correlations ( F184 , F185 and Y230 ) , indicated by a horizontal line , are unresolved to within the 0 . 2 ppm/point resolution of the Cβ assignment spectra , and are not assigned . DOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 01310 . 7554/eLife . 01812 . 014Figure 3—figure supplement 2 . PK9119 affects the oligomerization of zebrafish SAS-6 . SEC-MALS chromatogram of Danio rerio SAS-61–326 showing the refractive index signal with the derived molar masses indicated by the thicker horizontal lines . Dr SAS-61–326 displayed a distribution of masses from that of the dimer up to ∼250 kDa consistent with a concentration driven self-association equilibrium . In the presence of 1 mM PK9119 , the maximal observed mass was reduced to ∼160 kDa . The F131D mutant displayed a constant mass of 75 kDa in the absence and presence of PK9119 consistent with the mass of a Dr SAS-61–326 dimer . All samples were injected on SEC-MALS at 10 mg/ml ( 263 μM in monomer ) . Due to dilution during SEC , the peak concentrations were a factor of ∼10 lower than this ( ∼26 μM in monomer ) . The largest species seen for Dr SAS-61–326 are bigger than seen for L . major SAS-697–424 ( Figure 3C ) , despite a threefold lower concentration in monomer . This observation is consistent with the weaker head-to-head dimerization KD for L . major SAS-6 seen by analytical ultracentrifugation when compared with measurements on D . rerio SAS-6 ( van Breugel et al . , 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01812 . 014 To determine whether PK9119 affects oligomerization of L . major SAS-6 , we subjected wild-type and F257E mutant protein to size-exclusion chromatography—multi-angle light scattering ( SEC-MALS ) in the presence or absence of 1 mM PK9119 ( Figure 3C ) . Since we were unable to make full length L . major SAS-6 recombinantly , we used a L . major SAS-6 construct that contained the N-terminal domain and approximately half of its coiled-coil domain ( Lm SAS-697–424 ) for this assay . Similar to the previous findings ( Gopalakrishnan et al . , 2010; Kitagawa et al . , 2011; van Breugel et al . , 2011 ) , we did not find evidence for a stable ring-fraction of SAS-6 in solution , but found a complex equilibrium of SAS-6 oligomers ( ranging up to approximately SAS-6 hexamers ) for the wild-type construct in the absence of PK9119 , while the F257E mutant was a stable dimer under these conditions , in agreement with stable dimer formation through its coiled-coil domain . When the runs were repeated in the presence of PK9119 , we saw a clear shift in the elution volume for the wild-type but not the F257E SAS-6 construct towards smaller molecular weights and a decrease in the analysed mass from MALS , demonstrating that PK9119 partially affects head-to-head dimerization . We also repeated this experiment with the equivalent constructs of zebrafish SAS-6 ( Dr SAS-61–326 , wild-type and F131D ) under similar conditions . The results showed that zebrafish SAS-6 oligomerization was also affected by the presence of PK9119 , although to a lesser extent than L . major SAS-6 oligomerization ( Figure 3—figure supplement 2 ) . Thus , PK9119 appears to be a general inhibitor of SAS-6 oligomerization in vitro with some preference for the L . major variant , and represents a starting point for further screening of chemical analogs or fragment evolution . SAS-6 organizes the rotationally ninefold symmetric cartwheel ( Nakazawa et al . , 2007; Kitagawa et al . , 2011; van Breugel et al . , 2011 ) , an early assembly intermediate of centrioles that participates in establishing their symmetry and diameter ( Nakazawa et al . , 2007; Brito et al . , 2012; Gonczy , 2012 ) . Based on high-resolution structures of SAS-6 fragments and in silico modeling , detailed models of how SAS-6 self-associates to organize these cartwheels have been proposed ( Nakazawa et al . , 2007; Kitagawa et al . , 2011; van Breugel et al . , 2011 ) . However , these models have so far not been confirmed; EM studies with recombinant SAS-6 constructs were either of too low resolution ( Kitagawa et al . , 2011 ) or showed assemblies that were not ninefold symmetric ( Gopalakrishnan et al . , 2010; van Breugel et al . , 2011 ) . Furthermore , several studies found no evidence for complete ring-formation by recombinant SAS-6 in solution ( Gopalakrishnan et al . , 2010; Kitagawa et al . , 2011; van Breugel et al . , 2011 ) . Thus , it has been unclear if and under what conditions SAS-6 is sufficient to assemble ninefold symmetric cartwheels and thereby assist in dictating centriole symmetry . We were unable to detect cartwheel formation by recombinant L . major SAS-6 in solution , confirming these previous findings . However , we demonstrate that under high protein and precipitant concentrations , a L . major SAS-6 construct crystallized as ninefold symmetric rings with a diameter very similar to cartwheel hubs in vivo; this provides the first unambiguous experimental evidence that SAS-6 is able to form ninefold symmetric cartwheels on its own . Our results suggest that SAS-6 needs to be highly concentrated locally in order to be able to form cartwheels . Intriguingly , in vivo , SAS-6 is indeed highly enriched at the site of centriole formation ( Kleylein-Sohn et al . , 2007; Strnad et al . , 2007; Dammermann et al . , 2008; Blachon et al . , 2009; Sonnen et al . , 2012; Lettman et al . , 2013 ) . However , in vivo , SAS-6 assembly occurs in the context of other essential centriole duplication factors , some of which can be found in a complex with SAS-6 ( Stevens et al . , 2010; Tang et al . , 2011; Lin et al . , 2013 ) . It is therefore likely that additional centriole proteins assist SAS-6 assembly , especially in light of the fact that several studies ( Gopalakrishnan et al . , 2010; Kitagawa et al . , 2011; van Breugel et al . , 2011 ) , and our study presented here , have been unable to demonstrate efficient cartwheel assembly by SAS-6 in solution . Why does SAS-6 not form cartwheels efficiently ? The interaction interfaces that are critical for ring formation ( i . e . , the head-to-head dimer interface and the head-domain–coiled-coil interface ) are both relatively small ( Figures 1 , 2; Kitagawa et al . , 2011; van Breugel et al . , 2011 ) . In solution , SAS-6 oligomers are therefore likely to show some ‘wobble’ that would make ring closure inefficient and assembled rings unstable . However , in our ring structure , both of these critical interfaces are similar in their orientations to those observed in the crystals of the Lm SAS-697–320 F257E and the Lm SAS-697–274 constructs that did not crystallize as rings ( Figure 2—figure supplement 2 ) . Thus , although clearly not occurring efficiently in solution , our data suggest that ninefold symmetric ring formation might correspond to a weakly favored SAS-6 conformation that needs to be stabilized . In our crystal structure , rings are stabilized by crystal packing interactions and , in vivo , probably by SAS-6 interacting proteins . This model could explain why SAS-6 is required for the faithful establishment of centriole symmetry , while not being the sole determinant of their ninefold symmetry in vivo ( Nakazawa et al . , 2007 ) . In basal bodies , cartwheels are stacked onto each other with a vertical distance between cartwheel hubs of ∼8 nm ( Guichard et al . , 2012 ) . Although we observed a similar stacking in our cartwheel hub structure ( Figure 2—figure supplement 1 ) , the corresponding distance was only ∼4 nm . Furthermore , although a comparable packing was previously observed in a crystal of the N-terminal head domain of zebrafish SAS-6 ( van Breugel et al . , 2011 ) , the underlying packing interactions are based on a non-conserved interface and are therefore unlikely to be of relevance in vivo . Recent EM tomograms of basal bodies from Trichonympha ( Guichard et al . , 2012 , 2013 ) rather suggest that ring stacking in vivo is based on parallel interactions between the SAS-6 coiled-coil stalks and vertical interactions of SAS-6 associated components at the periphery of centriolar cartwheels . Interestingly , we discovered a dimerization-impaired state of Leishmania major SAS-6 that could provide a potential regulatory mechanism for its assembly . In this state , residue Y215 blocks the hydrophobic pocket into which F257 from its homo-dimer partner binds . Mutating Y215 to lysine resulted in an apparent threefold increase in dimerization affinity as measured by analytical ultracentrifugation . If open and closed Y215 conformations are in equilibrium with each other , the dimerization affinity of the closed Y215 state in solution would be even lower than the measured value , since , in this case , ultracentrifugation analyses the association of a mixture of open and closed states . We speculate that a release of this closed SAS-6 state through either a regulatory protein or a direct phosphorylation of Y215 could provide a simple mechanism to trigger SAS-6 assembly locally by increasing the concentration of assembly efficient SAS-6 . Should this mechanism indeed be used in vivo ( which is currently unknown ) , it would probably be confined to the Trypanosomatids as other SAS-6 homologues appear not to have tyrosine/aromatic residues in the position equivalent to Y215 that could play an equivalent role . Finally , as a proof-of-principle we show that oligomerization of SAS-6 can be inhibited by the small molecule PK9119 in vitro . The evolutionary conservation of the hydrophobic pocket involved in dimerization together with PK9119’s relatively broad activity in inhibiting both L . major and ( to a lesser extent ) D . rerio SAS-6 could suggest that PK9119 targets this pocket . However , our NMR binding studies are most consistent with PK9119 binding adjacent to the head-to-head dimerization interface and thereby altering its structure subtly ( Figure 3B ) . Confirming this notion , when we modeled the SAS-6–PK9119 complex with the HADDOCK software package ( available at http://haddock . science . uu . nl/services/HADDOCK/haddock . php ) ( de Vries et al . , 2010; Wassenaar et al . , 2012 ) , using high ambiguity restraints derived from the observed chemical shift perturbations in HSQC spectra ( 6 restraints >0 . 2 ppm ) , we found that of the 200 docked structures 98% occupied a single cluster in which PK9119 was bound to helix α1 , but not on the side lining the hydrophobic pocket , but on its opposite side , facing away from this pocket ( data not shown ) . Clearly , an elucidation of a high-resolution structure of PK9119 bound to its target would be important to understand how exactly PK9119 functions in inhibiting L . major SAS-6 . The binding affinity of our compound is currently low ( in the mM range ) and it does not show strong species specificity . To establish whether PK9119 has any potential to be improved in these critical aspects , it will be essential to systematically explore chemical modifications of its central indole scaffold . Regardless of these limitations , our demonstration that SAS-6 can be inhibited in vitro is a first , small step towards the goal of developing SAS-6 inhibitors that also could be used in vivo , for example as a cell-biological tool . All L . major constructs were made synthetically as codon-optimized genes ( IDT , Coralville , Iowa ) . Zebrafish SAS-6 constructs were described earlier ( van Breugel et al . , 2011 ) . All constructs were N-terminally His-tagged . Proteins were expressed in E . coli BL21 Rosetta and purified using standard methods via NiNTA ( Qiagen , Hilden , Germany ) chromatography , proteolytic tag cleavage , size-exclusion chromatography and ion-exchange chromatography . The L . major SAS-697–274 and SAS-697–320 selenomethionine derivatives were purified in the same way , but expression was in M9 medium supplemented with 2 mM MgSO4 , 0 . 4% ( wt/vol ) glucose , 25 µg/ml FeSO4 . 7H2O , 40 µg/ml amino acid mix ( excluding Methionine ) , 1 µg/ml riboflavin , 1 µg/ml niacinamide , 0 . 1 µg/ml pyridoxine monohydrochloride , 1 µg/ml thiamine and 40 µg/ml seleno-L-methionine . All purified L . major constructs included the extra sequence GP , zebrafish SAS-6 GPH at their N-termini from the cloning/protease cleavage site . SeMet L . major SAS-697–274 crystals were obtained using the sitting drop method with a reservoir solution of 100 mM bisTris pH 5 . 1 , 200 mM MgCl2 , 20% ( wt/vol ) PEG-3350 at 16°C . Drops were set up using 100 nl protein solution and 100 nl of reservoir solution . After half a day , the crystals were mounted in 100 mM bisTris pH 5 . 1 , 200 mM MgCl2 , 10% ( wt/vol ) PEG-3350 , 25% ( wt/vol ) Glycerol and flash-frozen in liquid nitrogen . Native L . major SAS-697–274 was crystallized at 16°C using the sitting drop method with 200 nl of the protein solution and 200 nl of the reservoir solution ( 100 mM bisTris pH 5 . 1 , 200 mM MgCl2 , 20% [wt/vol] PEG-3350 ) . The crystals were mounted in 100 mM bisTris pH 5 . 1 , 200 mM MgCl2 , 10% ( wt/vol ) PEG-3350 , 25% ( wt/vol ) glycerol after 1 day and flash-frozen in liquid nitrogen . L . major SAS-697–320 F257E was crystallized in sitting drops in 100 mM NaCitrate pH 5 . 85 , 21% ( wt/vol ) PEG-3000 at 16°C using 1 μl of protein solution and 1 μl of the reservoir solution . The crystals were mounted after 4–5 days in 100 mM NaCitrate pH 5 . 85 , 21% ( wt/vol ) PEG-3000 and increasing amounts of PEG-400 to a final concentration of 20% ( wt/vol ) before flash-freezing them in liquid nitrogen . L . major SAS-697–320 WT crystals were obtained using the sitting drop method with a reservoir solution of 100 mM bisTris pH 6 . 23 , 100 mM NaAcetate ( not pH adjusted ) , 26 . 5–27% ( wt/vol ) PEG-400 at 16°C . Drops were set up using 1 . 8 μl of protein solution and 1 . 8 μl of the reservoir solution . After 3 days , the crystals were mounted in 100 mM bisTris pH 6 . 23 , 100 mM NaAcetate ( not pH adjusted ) , 28% ( wt/vol ) PEG-400 . SeMet L . major SAS-697–320 WT crystals were obtained using the sitting drop method with a reservoir solution of 100 mM bisTris pH 6 . 23 , 100 mM NaAcetate ( not pH adjusted ) , 25 . 5% ( wt/vol ) PEG-400 at 16°C . Drops were set up using 1 . 5 μl of protein solution and 1 . 5 μl of the reservoir solution . After 3 days , the crystals were mounted in 100 mM bisTris pH 6 . 23 , 100 mM NaAcetate ( not pH adjusted ) , 30% ( wt/vol ) PEG-400 . The protein concentrations of the crystallized constructs were determined by the Bradford assay with BSA as a standard and were: 80 . 3 mg/ml ( SeMet L . major SAS-697–274 ) , 29 . 3 mg/ml ( L . major SAS-697–274 ) , 94 . 7 mg/ml ( L . major SAS-697–320 F257E ) , 50 . 3 mg/ml SeMet L . major SAS-697–320 and 50 . 1 mg/ml ( L . major SAS-697–320 ) . Data sets were integrated and scaled using MOSFLM ( Leslie and Powell , 2007 ) ( SeMet and native L . major SAS-697–274 , L . major SAS-697–320 ) or XDS ( Kabsch , 2010 ) ( L . major SAS-697–320 F257E and SeMet L . major SAS-697–320 ) . Data sets were scaled using SCALA or AIMLESS ( Evans , 2006; Evans and Murshudov , 2013 ) . The L . major SAS-697–274 structure was solved from the corresponding 3-wavelength SeMet dataset by MAD using the SHELX CDE pipeline in HKL2MAP ( Pape and Schneider , 2004 ) , resulting in clear electron density into which an initial model was built using BUCANNEER ( Cowtan , 2006 , 2008 ) and manual building . REFMAC ( Murshudov et al . , 2011 ) was used to refine the model against the native data set with manual building done in Coot ( Emsley and Cowtan , 2004 ) . L . major SAS-697–320 WT and F257E were solved by molecular replacement in Phaser ( McCoy et al . , 2007 ) using the L . major SAS-697–274 structure as a search model ( SAS-697–274 monomer ( F257E ) or the BC-dimer ( WT ) . The models were subsequently further built in Coot ( Emsley and Cowtan , 2004 ) and refined in REFMAC ( Murshudov et al . , 2011 ) and Phenix . refine ( Afonine et al . , 2005 ) using NCS and ( for WT L . major SAS-697–320 ) TLS refinement with separate TLS groups for the globular N-terminal and the coiled-coil domains and also using as a reference model restraint the B chain of L . major SAS-697–274 ( residue 130–271 ) . Refinement yielded clear density for the missing coiled-coil part of these constructs . Equilibrium sedimentation experiments were performed on an Optima XL-I analytical ultracentrifuge ( Beckmann , Brea , California ) using An50Ti rotors . Sample volumes of 110 µl with protein concentrations of 100 , 200 , and 400 μM were loaded in 12 mm 6-sector cells and centrifuged at 11300 , 17000 , and 21200 rpm until equilibrium was reached at 4°C . At each speed , comparison of several scans was used to judge whether or not equilibrium had been reached . Buffer conditions were 50 mM Tris , 100 mM NaCl , pH 8 . 0 . The solvent density and viscosity ( ρ = 1 . 00557 g/ml and η = 1 . 6056 mPa⋅s ) were calculated using Sednterp ( Dr Thomas Laue , University of New Hampshire , Sednterp server available at: http://sednterp . unh . edu . Desktop version can be downloaded from: http://bitcwiki . sr . unh . edu/index . php/Downloads ) . Data were processed and analysed using UltraSpin software ( available at: http://www . mrc-lmb . cam . ac . uk/dbv/ultraspin2/ ) and SEDPHAT ( Schuck , 2003 ) . Small molecules were screened for binding to approximately 40 µM 15N-labelled protein in aqueous phosphate buffer ( 25 mM Phosphate , 150 mM NaCl , 2 mM DTT , pH 7 . 2 ) and up to 2 mM ligand concentration , with a total of 5% ( vol/vol ) DMSO-d6 . {1H-15N}-fast-HSQC spectra ( Mori et al . , 1995 ) were recorded using a Bruker Avance spectrometer operating at 800 MHz 1H frequency , with a 5 mm cryogenic inverse probe and sample temperature of 298 K . The digital resolution of the processed data was 3 . 2 and 5 . 3 Hz/point in f2 and f1 , respectively . Backbone resonance assignments were obtained from HNCACB , CBCA ( CO ) NH , HN ( CA ) CO , HNCO and HNCANH spectra at 600 MHz 1H frequency , acquired using unmodified Bruker pulse programs and a protein concentration of 400 µM . Aromatic sidechain resonances were assigned from a ( HB ) CB ( CGCD ) HD spectrum . Data were processed using TopSpin version 3 ( commercially available from Bruker , Billerica , Massachusetts , details available at http://www . bruker . com/products/mr/nmr/nmr-software/software/topspin/ ) and analysed using Sparky ( Goddard & Kneller , UCSF , San Francisco , available at http://www . cgl . ucsf . edu/home/sparky/ ) . Models of the complex between SAS-6 and PK9119 were generated by submitting the crystal structure coordinates of a L . major SAS-697–274 monomer to the WeNMR server ( available at: https://www . wenmr . eu ) ( de Vries et al . , 2010; Wassenaar et al . , 2012 ) , using default HADDOCK parameters and CNS topology parameters for PK9119 based on the PRODRG predictions provided on the HADDOCK server ( available at http://haddock . science . uu . nl/services/HADDOCK/haddock . php ) . The mass in solution of L . major SAS-697–424 , wild-type and F257E mutant , Dr SAS-61−326 , wild-type and F131D mutant , was determined by SEC-MALS measurements using a Wyatt Heleos II 18 angle light scattering instrument coupled to a Wyatt Optilab rEX online refractive index detector . Detector 12 in the Heleos instrument was replaced with Wyatt’s QELS detector for dynamic light scattering measurement . Protein samples ( 100 μl ) were resolved on a Superdex S-200 10/300 analytical gel filtration column ( GE Healthcare , Little Chalfont , UK ) running at 0 . 5 ml/min in 50 mM bisTris , 100 mM NaCl , pH 7 . 0 buffer , containing 0 . 126% ( vol/vol ) DMSO and ±1 mM chemical compound PK9119 ( ( 5-bromo-7-ethyl-1H-indol-3-ylmethyl ) -dimethyl-amine , Sigma-Aldrich , St . Louis , Missouri ) before passing through the light scattering and refractive index detectors in a standard SEC-MALS format . Buffers were filtered through a 0 . 22-μm filter before usage to remove any PK9119 precipitates . Protein concentration was determined from the excess differential refractive index based on 0 . 186 RI increment for 1 g/ml protein solution . The concentration and the observed scattered intensity at each point in the chromatograms were used to calculate the absolute molecular mass from the intercept of the Debye plot using Zimm’s model as implemented in Wyatt’s ASTRA software ( commercially available from Wyatt technology , Santa Barbara , California , details at: http://www . wyatt . com/products/software/astra . html ) .	Many cells have tiny hair-like structures called cilia on their surface that are important for communicating with other cells and for detecting changes in the cell’s surroundings . Some cilia also beat to move fluids across the cell surface—for example , to move mucus out of the lungs—or act as flagella that undergo rapid whip-like movements to propel cells along . Cilia are formed when a small cylindrical structure in the cell called a centriole docks against the cell membrane and subsequently grows out . However , many of the details of this process are poorly understood . One of the earliest events in centriole assembly is the formation of a central structure that looks like a cartwheel . This cartwheel acts as a scaffold onto which the rest of the centriole is then added . It has been proposed that a protein called SAS-6 can build this cartwheel just by interacting with itself . However , this has so far not been shown clearly . Now , using a technique called X-ray crystallography , van Breugel et al . directly confirm this hypothesis . This is significant because it demonstrates that the simple self interaction of a protein could lie at the heart of building a complex structure like a centriole . The single-celled human parasites that spread diseases such as Leishmaniasis , Chagas disease , and sleeping sickness rely on flagella to move around and interact with their surroundings . If SAS-6 cannot assemble into the cartwheel structure , flagella cannot form correctly , potentially stopping the parasites . By screening a library of small molecules , van Breugel et al . found one that partially disrupted the interactions of SAS-6 with itself in the test tube . This small molecule interacted only very weakly with SAS-6 and was not specific for SAS-6 from the disease-causing organism . These unfavourable properties therefore make this compound of no immediate use . However , this result nevertheless shows that small molecules can impair SAS-6 function at least in the test tube and that the development of a more efficient inhibitor might therefore be possible .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	Structure of the SAS-6 cartwheel hub from Leishmania major
Deep sequencing analyses have shown that a large fraction of genomes is transcribed , but the significance of this transcription is much debated . Here , we characterize the phylogenetic turnover of poly-adenylated transcripts in a comprehensive sampling of taxa of the mouse ( genus Mus ) , spanning a phylogenetic distance of 10 Myr . Using deep RNA sequencing we find that at a given sequencing depth transcriptome coverage becomes saturated within a taxon , but keeps extending when compared between taxa , even at this very shallow phylogenetic level . Our data show a high turnover of transcriptional states between taxa and that no major transcript-free islands exist across evolutionary time . This suggests that the entire genome can be transcribed into poly-adenylated RNA when viewed at an evolutionary time scale . We conclude that any part of the non-coding genome can potentially become subject to evolutionary functionalization via de novo gene evolution within relatively short evolutionary time spans . Genome-wide surveys have provided evidence for 'pervasive transcription' , i . e . , much larger portions of the genome are transcribed than would have been predicted from annotated exons ( Clark et al . , 2011; Hangauer et al . , 2013; Kellis et al . , 2014 ) . Most are expected to be non-coding RNAs ( lncRNAs ) of which some have been shown to be functional . However , the general conservation level of these additional transcripts tends to be low , which raises the question of their evolutionary turnover dynamics ( Kutter et al . , 2012; Kapusta and Feschotte , 2014 ) . They are currently receiving additional attention , since they could be a source for de novo gene formation via a proto-gene stage ( Carvunis et al . , 2012; Ruiz-Orera et al . , 2014; Neme and Tautz , 2014 ) . It has been shown that de novo gene emergence shows particularly high rates in the youngest lineages ( Tautz and Domazet-Loso , 2011 ) , indicating that there is high turnover of such transcripts and genes between closely related species . Indeed , comparative studies of de novo genes between Drosophila species ( Palmieri et al . , 2014 ) and within Drosophila populations ( Zhao et al . , 2014 ) have confirmed this . A number of possibilities have been discussed by which new transcripts are generated in previously non-coding regions , including single mutational events , stabilization of bi-directional transcription and insertion of transposable elements with promotor activity ( Brosius , 2005; Gotea et al . , 2013; Neme and Tautz , 2013; Wu and Sharp , 2013; Sundaram et al . , 2014; Ruiz-Orera et al . , 2015 ) . Detailed analyses of specific cases of emergence of a de novo gene have shown that single step mutations can be sufficient to generate a stable transcript in a region that was previously not transcribed and translated ( Heinen et al . , 2009; Knowles and McLysaght , 2009 ) . The unequivocal identification of de novo transcript emergence can only be made in a comparison between very closely related evolutionary lineages , where orthologous genomic regions can be fully aligned , even for the neutrally evolving parts of the genome ( Tautz et al . , 2013 ) . While the available genome and transcriptome data for mammals and insects are sufficient to screen for specific cases of de novo transcript emergence , they are still too far apart of each other to allow a comprehensive genome-wide assessment . Our analysis here is therefore based on a new dataset that reflects a very shallow divergence time-frame for relatives of the house mouse ( Mus musculus ) . We selected populations , subspecies and species with increasing phylogenetic distance to the Mus musculus reference sequence ( Keane et al . , 2011 ) . This reference was derived from an inbred strain of the subspecies Mus musculus domesticus and we use samples from three wild type populations of M . m . domesticus as the most closely related taxa , separated from each other by about 3 , 000–10 , 000 years . Further , we use samples from the related subspecies M . m . musculus and M . m . castaneus , which are separated since 0 . 3–0 . 5 million years . The other samples are recognized separate species with increasing evolutionary distances ( Figure 1 ) . We call this set of populations , subspecies and species collectively 'taxa' in the following . Altogether they span 10 million years of divergence , which corresponds to an average of 6% nucleotide difference for the most distant comparisons . 10 . 7554/eLife . 09977 . 003Figure 1 . Phylogenetic relationships and time estimates for the taxa used in the study . New genome sequences were generated for taxa with * . A common genome was constructed across all taxa ( Figure 1—figure supplement 1 ) based on a mapping algorithm that is not affected by the sequence divergence between the samples ( Appendix 1 ) . Figure 1—figure supplement 2 shows the intersection of genome coverage between the named species . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 00310 . 7554/eLife . 09977 . 004Figure 1—figure supplement 1 . Scheme for the establishment of the 'common genome' using genomic reads and the mouse reference genome . The common genome represents the portion of the reference which is present and detectable across all species . The genome sequencing , processing and sequence analysis were done in the same way as for transcriptomes , effectively removing possible biases derived from sequencing and mapping . Note that the assignment of the common genome fraction was done after mapping all genomic and transcriptomic reads to the reference , i . e . the mapping process was not affected by a reduced mapping target . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 00410 . 7554/eLife . 09977 . 005Figure 1—figure supplement 2 . Venn diagrams of representation of the common genome , derived from 200bp windows covered in genomic reads in species with more than one million years divergence to the reference . Windows covered by all four species are used as the common genome ( shown as the intersection of all species ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 005 We obtained genome sequence reads for all taxa and mapped them to the mouse reference genome , using an algorithm that was specifically designed to deal efficiently with problems that occur in cross-mapping between diverged genomes ( Sedlazeck et al . , 2013; see Appendix 1 for validation ) . All regions that could be unequivocally mapped for all taxa were then used for further analysis . We refer to this as the 'common genome' which allows comparisons on those regions of the genomes which have not been gained or lost along the phylogeny , i . e . , are common across all taxa ( Figure 1—figure supplement 1 ) . It represents 71 . 7% of the total reference genome length ( Figure 1—figure supplement 2 ) . Hence , we are nominally not analyzing about a third of the total genome length , but this corresponds to the highly repetitive parts for which unique and reliable mapping of transcriptomic reads would not be possible . Also , changes in transcription derived from gain or loss of genomic regions do not contribute to the patterns described below . We chose three tissues for transcriptome sequencing , including testis , brain and liver . Previous studies had shown that testis and brain harbor the largest diversity of transcripts ( Necsulea and Kaessmann , 2014 ) . We sequenced only the poly-A+ fraction of the RNA , i . e . , our focus is on coding and non-coding exons in processed RNA . We use non-overlapping sliding windows of 200nt to assay for presence or absence of reads within the windows and express overall coverage as the fraction of windows showing transcription ( see methods for details ) . We use only uniquely mapping reads , implying that we neglect the contributions and dynamics at repetitive loci . We display three thresholds of window coverage , the minimum being coverage by at least a single read , while the higher ones represent at least 10 and 100 reads respectively . The first serves as a very inclusive metric of low-level transcription , with the drawback of potentially including noise into the analysis , due to stochasticity in sampling , while the others represent thresholds for more abundant transcripts that are unlikely to be affected by sampling noise . Among the three tissues analyzed , liver has the lowest overall read coverage while brain and testis have similar overall levels ( Figure 2A–C ) . Combining the data from all three tissues or triplicating the read depth for one tissue ( brain ) increases the overall coverage in a similar way ( Figure 2D , E ) . 10 . 7554/eLife . 09977 . 006Figure 2 . Transcriptome coverage of the common genome per taxon . ( A–C ) Liver , brain and testis , respectively , sequenced at approximately the same depth . ( D ) Combination of samples from A–D . ( E ) Additional sequencing of brain samples at 3x depth , compared to B . ( F ) Combination of all samples , including additional brain sequencing . Three coverage levels are represented by colors from light blue to dark blue: window coverage with at least 1 , 10 and 100 reads . Taxon abbreviations as summarized in Figure 1 , with closest to the reference genome to the left of each panel and most divergent one to the right . Note that the slight rise in low read coverage for the distant taxa could partially be due to slightly more mismapping of reads at this phylogenetic distance ( see Appendix 1 for simulation of mapping efficiency ) , but is also affected by a larger fraction of singleton reads ( compare Figure 4—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 006 Figure 2F shows the total coverage across all tissues and all sequencing runs , which amounts to an average of 50 . 0 ± 2 . 5% per taxon . Hence , for each tissue , as well as in this combined set , we observe a very similar coverage in all taxa , with only a slight increase in the low expressed fraction for the most distant comparisons ( see also legend Figure 2 ) . This more or less stable pattern across phylogenetic time could either be due to the same regions being transcribed in all taxa , or a more or less constant rate of turnover of gain and loss of transcription between taxa . To test these alternatives , we have asked which part of the transcribed window coverage is shared between the taxa and which is specific to the taxa . For this , we consider three classes: i ) windows that are found in a single taxon only , ii ) windows that are found in 2–9 taxa , i . e . more than one but not in all and iii ) windows shared among all taxa ( Figure 3; Figure 3—figure supplement 1 shows an extended version where class ii ) is separated into each individual group ) . However , such an analysis could potentially be subject to a sampling problem , i . e . not finding a transcript in a taxon does not necessarily imply true absence , but could also be due to failure of sampling . This would be particularly problematic for singleton reads , since the probability of falsely not detecting one in a second sample that expresses it at the same level is about 37% . However , given that we ask whether it is detected in any of the other 9 taxa , the probability of falsely not detecting it if it exists across all of them becomes small ( 0 . 01% ) ( see also further analysis on singletons below ) . 10 . 7554/eLife . 09977 . 007Figure 3 . Distribution of shared and non-shared windows with transcripts for each taxon , based on the aggregate dataset across all three tissues . Three classes are represented: i ) windows that are found in a single taxon only , ii ) windows found in 2–9 taxa and iii ) windows shared among all 10 taxa ( from left to right in each panel ) . Windows with transcripts were first classified as belonging to one of the three classes , independent of their coverage , and were then assigned to the coverage classes represented by the blue shading ( from light blue to dark blue: window coverage with at least 1 , 10 and 100 reads ) . Taxon names as summarized in Figure 1 . Figure 3—figure supplement 1 shows an extended version where class ii ) is separated into each individual group . Relative enrichment of annotated genes in the conserved class is shown in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 00710 . 7554/eLife . 09977 . 008Figure 3—figure supplement 1 . Distribution of shared transcripts according to the number of taxa shared , based on the aggregate dataset across all three tissues . Windows with transcripts were first classified as belonging to each of the sharing categories ( from 1 to 10 ) , independent of their coverage , and were then assigned to the coverage classes represented by the blue shading ( from light blue to dark blue: window coverage with at least 1 , 10 and 100 transcripts ) . Taxon names as summarized in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 00810 . 7554/eLife . 09977 . 009Figure 3—figure supplement 2 . Windows transcribed across most species ( 9 or more ) are strongly enriched in genes known from the reference genome , while windows transcribed in some taxa ( 8 or less ) are strongly depleted from known genes . The effect is most evident for protein-coding genes , but still present for non-coding genes . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 009 Between 1 and 7% of transcribed windows are unique to one taxon only , with the more distant taxa showing the higher percentages ( Figure 3 ) . Most of these taxon-specific transcripts are lowly expressed ( <10 reads per window ) , but the more distant taxa ( MAT and APO in Figure 3I , J ) show also some more highly expressed ones . We find a total of 6566 windows with read counts >50 that occur in a single taxon only , mostly in the long branches leading to MAT ( 1638 windows ) and APO ( 4485 windows ) , but some also between the most closely related taxa ( 43 windows for DOM , including populations; 38 windows for MUS , including populations ) . Approximately 18% of windows show transcripts shared across all taxa . These include most of the very highly expressed ones ( >100 reads per window ) , but also a fraction of the low expressed ones ( Figure 3 ) . They are also enriched in annotated genes , especially in exons of protein coding genes , but also in non-coding genes ( Figure 3—figure supplement 2 ) . The class ii ) windows ( sharing between 2 and 9 taxa in Figure 3 ) represents the genes showing more or less turnover between taxa , with more turnover the more distant they are of each other ( Figure 3—figure supplement 1 ) . This class constitutes cumulatively the largest fraction ( between 26 and 33% of whole genome coverage - Figure 3 ) , supporting the notion of a fast turnover of most of the transcribed regions between taxa . The taxon-specific turnover of transcripts is also reflected in a distance tree of shared coverage . Taxa that are phylogenetically closer to each other share more transcripts , i . e . the tree topology mimics that of a phylogenetic tree based on molecular sequence divergence ( Figure 4A , B ) . This implies that the turnover of the transcripts is not random , but time dependent . However , the relative branch lengths are much extended for the more closely related taxa compared to the molecular distances , implying that there is a particularly high turnover between them . 10 . 7554/eLife . 09977 . 010Figure 4 . Distance tree comparisons based on molecular and transcriptome sharing data . ( A ) Molecular phylogeny based on whole mitochondrial genome sequences as a measure of molecular divergence ( black lines represent the branch lengths , dashed lines serve to highlight short branches ) . ( B ) Tree based on shared transcriptome coverage of the genome , using correlations of presence and absence of transcription of the common genome . All nodes have bootstrap support values of 70% or more ( n = 1000 ) . ( C ) Tree based on shared transcriptome coverage of singleton reads only from subsampling of the extended brain transcriptomes . Left is the consensus tree with the variance component between samples depicted as triangles , right is the same tree , but only for the branch fraction that is robust to sampling variance . Taxon names as summarized in Figure 1 . Figure 4—figure supplement 1 shows the fraction of singletons in dependence of each sample in each taxon , Figure 4—figure supplement 2 in dependence of read depth . Figure 4—figure supplement 3 shows an extended version of the analysis shown in 4C for higher coverage levels . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 01010 . 7554/eLife . 09977 . 011Figure 4—figure supplement 1 . Fraction of windows with singletons ( one paired read ) of the common genome per taxon . ( A-C ) Liver , brain and testis , respectively , sequenced at approximately the same depth . ( D ) Combination of samples from A–D . ( E ) Additional sequencing of brain samples at 3x depth , compared to B . ( F ) Combination of all samples , including additional brain sequencing . Light gray indicates singletons observed in each individual sample/taxon combination . Dark gray indicates singletons across the whole experiment , i . e . not re-detected in any other tissue or taxon . Taxon abbreviations as summarized in Figure 1 , with closest to the reference genome to the left of each panel and most divergent one to the right . Note that the rise in singleton number for the distant taxa can be ascribed to the longer branch length , i . e . absence of closely related taxa in which the singleton could have been re-detected . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 01110 . 7554/eLife . 09977 . 012Figure 4—figure supplement 2 . Reduction of singletons in dependence of aggregate sequencing depth . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 01210 . 7554/eLife . 09977 . 013Figure 4—figure supplement 3 . Trees based on shared transcriptome coverage of the genome , using binary correlations . We used the deep sequenced brain samples to estimate the proportion of sampling artifacts in terminal branches , and effectively subtracted the proportion of artifacts to obtain reliable phylogenetic signals . Each brain sample was split in three completely independent samples of 100 million reads . Top: Trees constructed using: regions covered only with one read in each taxon , regions covered by 1 and 5 reads ( very low expression ) , regions covered by any reads , regions above 10 reads ( mid expression ) and regions above 100 reads ( high expression ) . The percentage shown indicates the average level of sampling artifacts for each threshold , derived from the length of the terminal branches not found in all replicates of each taxon , i . e . the uncorrelated portion across samples of the same origin . These numbers are highest for the lowly expressed regions , and are lowest for the highly expressed regions , and are more or less constant within comparisons . Once subtracted , the phylogenetic signal remains robust . Taxon names as summarized in Figure 1 . The figure part with the 1 read fraction corresponds to Figure 4C . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 013 To assess in how much this could be due a sampling variance problem at low expression levels , we have separately analyzed the transcripts that are represented by single reads only , since these should be most sensitive towards sampling problems . Depending on read depth and tissue , they constitute about 2–12% of the common windows when assessed on a per sample basis ( Figure 4—figure supplement 1 ) . However , most of these singletons in a given sample were re-detected in another tissue or another taxon ( Figure 4—figure supplement 1 ) , such that less than 2% are present in a given taxon ( Figure 4—figure supplement 1 ) and less than 7% cumulatively throughout the whole dataset ( Figure 4—figure supplement 2 ) . We used the extended brain sample reads , split them into three non-overlapping sets of about 100 Mill reads for each taxon and constructed trees out of these sets using only the singleton reads . This is the equivalent of repeating the same experiment three times . We find indeed differences in the resulting trees , i . e . there is a measureable sampling variance . By constructing a consensus tree , we can partition the data into a variable and a common component . We find that 88% of the branch length is influenced by sampling variance , while the remaining 12% still recover the expected topology ( Figure 4C ) . When we use a read coverage of 1–5 for the same analysis , we find that 52% of the branch length are subject to sampling variance and for all reads combined it is 35% ( Figure 4—figure supplement 3 ) . Hence , at the 100 Mill read level , we have a noticeable effect of sampling variance , but this does not erase the underlying signal . Also , the analysis in Figure 4B is based on 600 Mill reads per taxon , where sampling variance is expected to be further lowered . The high dynamics of transcriptional turnover between taxa raises the question whether all parts of the genome might be accessible to transcription at some point in evolutionary time . To explore this possibility , we used a rarefaction approach to simulate the addition of one taxon at a time and used the curve to predict the behavior of adding more taxa than the ones in the present study . We compared this approach to a curve of increasing depth of sequencing , by taking subsets at 10% intervals to understand whether depth or taxonomic diversity have different behavior in this respect . We assume that in each species only a subset of the genome is transcribed , therefore the increase in depth of sequencing would saturate at some point below 100% . Conversely , if each taxon is transcribing slightly different portions of the genome due to a steady turnover , increasing the total number of sampled taxa should increase the saturation more than the increase that could be achieved by sequencing depth . This is indeed what we find . The addition of taxa indeed leads to a further increase in transcriptomic coverage , with a generalized linear model best describing the data as increasing in a logarithmic fashion ( Figure 5A ) . In contrast , we observe an asymptotic behavior of the curve for increasing depth of sequencing , with apparent saturation reached at 84 . 1% , close to the 83 . 2% that we have already achieved ( Figure 5B ) . 10 . 7554/eLife . 09977 . 014Figure 5 . Rarefaction , subsampling and saturation patterns using all available samples and reads . ( A ) Sequencing depth saturation as estimated from an increase in the number of taxa . ( B ) Sequencing depth saturation as estimated from increasing read number . Blue dots indicate increases per sub-sampled sequence fraction or taxon added from our dataset . Gray dotted line indicates the predicted behavior from the indicated regression , and gray area shows the prediction after doubling the current sampling either by additional taxa ( A ) or in sequencing effort ( B ) . Each analysis was tested for logarithmic and asymptotic models . Best fit was selected from ΔBIC , with Bayes factor shown and qualitative degree of support shown . Standard deviations are shown as black lines in A , and are too small to display in B ( note that due to the sampling scheme for this analysis , the values above 50% are not statistically independent and that the 100% value constitutes a single data point without variance measure ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 014 Combined with the previous results , this allows two major conclusions . First , random transcriptional noise ( technical or biological ) or deficiencies in sampling low level transcripts should not be major factors in our analysis , since saturation with sequencing depth would not be possible under a singleton dominated regime . Furthermore , low level transcripts ( including singletons ) have detectable biological signal ( Figure 4C ) . Second , the data are consistent with the above outlined ideas that the evolutionary turnover leads to steady – and almost unlimited – transcriptional exploration of the genome , when summed over multiple parallel evolutionary lineages and taxa . The above overall statistical consideration would still allow for the possibility of the existence of a few scattered genomic islands that are not accessible to transcription because of structural reasons ( so-called transcriptional deserts – Montavon and Duboule , 2012 ) or heterochromatically packed because they are not encoding genes required in the respective tissues . Hence , we analyzed also the size distribution of transcript-free genomic regions in our dataset . We find that the maximum observed length of non-transcribed regions is 6 kb ( Figure 6 ) , suggesting that apparent transcriptional deserts in one taxon are readily accessible to transcription in other taxa , at least for the non-repetitive windows of the genome that are analyzed here . 10 . 7554/eLife . 09977 . 015Figure 6 . Comparative analysis of lengths of regions transcribed or not transcribed across all data ( including deeper brain sequencing ) in all samples . Size distribution of regions not covered in any transcript ( green ) versus size distribution of regions with at least one transcript ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 015 Various studies have shown that many more regions of the genome are transcribed than are annotated as exons ( Ponting and Belgard , 2010; Kapranov and St . Laurent , 2012 ) . The significance of this additional transcription has been largely unclear and it has even been considered as noise , either biological or technical . Here we were able to trace the turnover of these extra transcripts . Our data suggest that many have sufficient stability to reflect a phylogenetic distance distribution that mimics the phylogeny of the taxa . Hence , they should not simply be considered as noise . Rather , their lifetime should be sufficient to expose them to evolutionary testing and in this way they become a substrate for de novo evolution of genes . On the other hand , they appear to have only a limited lifetime in case they do not acquire a function , i . e . there is also high turnover of the transcribed regions between taxa . This turnover has as a consequence that within a timespan of a few million years practically the whole genome is covered by transcription at some point in time , i . e . no major transcript-free islands exist . We have here sampled only three tissues . If more tissues and more life stages were sampled , one would expect an even higher coverage of the genome within a given taxon . Such deep analyses have been done in the ENCODE projects ( http://www . genome . gov/10005107 ) and they have confirmed pervasive transcription ( Clark et al . , 2011; Hangauer et al . , 2013; Kellis et al . , 2014 ) at the single-taxon level . Still , we expect that the turnover of transcribed regions between taxa would also apply to the other tissues and stages , i . e . evolutionary testing of new transcripts would relate to all tissues and stages . This turnover is contrasted by the set of conserved genes across taxa , for which even the expression levels may be maintained across larger evolutionary distances ( Pervouchine et al . , 2015 ) . We see a particularly large number of lineage-specific transcripts among the most closely related taxa . This becomes most evident in the distance tree in Figure 4B where the branch length of the three populations of M . m . domesticus , which have separated only a few thousand years ago , are almost as long as those of the sister species M . spretus that has separated almost 2 Mill . years ago . Although this is partially influenced by sampling variance of low expressed transcripts ( Figure 4C ) , this suggests that at the very short evolutionary distances ( thousands of years ) there is an even higher turnover of transcripts than at the longer time frames ( millions of years ) . Such a pattern of unequal rates suggests that weak selection could act against many newly arising transcripts , such that they can exist for a short time at a population scale , but not over an extended time . Hence , we expect that the presence of such transcripts will be polymorphic at the population level , similar as it has been shown in Drosophila ( Zhao et al . , 2014 ) . We have done a preliminary analysis of transcriptional variance between four individuals of each of the taxa and find this expectation fulfilled , but a more extensive study is required to obtain reliable data at this level . We expect that a fraction of new transcripts interacts with other genes and cellular processes , either via providing a positive function or via being slightly deleterious . Our data do not allow at present to speculate on how large this 'functional' fraction would be , but this could become subject to future experimental studies . It is also as yet open whether the transcripts exert their functions as RNAs or via translation products . The analysis of ribosome profiling data has shown that many RNAs that were initially classified as non-coding can be associated to ribosomes , i . e . are likely translated ( Wilson and Masel , 2011; Carvunis et al . , 2012; Ruiz-Orera et al . , 2014 ) . On the other hand , when tracing the origin of de novo genes , one finds frequently that they act first as RNA and acquire open reading frames only at a later stage ( Cai et al . , 2008; Kapranov and St . Laurent , 2012; Reinhardt et al . , 2013 - see discussion in Schlötterer , 2015 ) . For some of the de novo evolved genes in Drosophila it has been shown that they have assumed essential functions for the organism , such that knockouts of them are lethal ( Chen et al . , 2010 ) . Global analyses of new gene emergence trends suggest that the de novo evolution process has been active throughout the evolutionary history ( Neme and Tautz , 2013 ) . Hence , the possibility of a transition from new transcript emergence over acquisition of reading frames towards assuming essential genetic functions is well documented . The idea that many de novo transcripts are slightly deleterious is concordant with the fact that various cellular processes maintain a balance between RNA transcription and degradation ( Houseley and Tollervey , 2009; Jensen et al . , 2013 ) . In yeast and mammals it has been shown that several molecular pathways exist that degrade excess transcripts , in particular the ones from bidirectional promoter activity ( Jensen et al . , 2013; Wu and Sharp , 2013 ) . Hence , the fact that many of the transcripts found by deep sequencing occur only at low levels does not necessarily imply a low level of transcription , but could alternatively be due to fast targeting by a degradation machinery . 10 . 7554/eLife . 09977 . 016Table 1 . Genome sequencing and read mapping information relative to the C57Bl/6 reference strain ( GRCm38 . 3/mm10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 016SpeciesUniquely mapping reads ( MAPQ >25 ) Mean coverage depth ( window based ) Reference coverage ( % windows ) Total sequence divergenceAccession ReadsAccession BAMsApodemus uralensis4 . 46E+0840x78 . 23%5 . 60%ERS942341ERS946059Mus mattheyi5 . 58E+0852x77 . 19%4 . 50%ERS942343ERS946060Mus spretus7 . 71E+0852x93 . 91%1 . 70%ERS946096Mus spicilegus6 . 16E+0857x84 . 39%1 . 60%ERS942342ERS946061 The percentage of divergence was estimated from mappings using NextGenMap ( Sedlazeck et al . , 2013 ) . Only uniquely mapping reads were considered and mapping quality greater than 25 . Variation was estimated from the alignments using samtools mpileup ( Li et al . , 2009 ) . Divergence was calculated as number of changes divided by the genome size . ** Corresponds to study accession PRJEB11535 . All other accessions deposited under studies PRJEB11513 and PRJEB11533 . 10 . 7554/eLife . 09977 . 017Table 2 . Transcriptome reads from each sample sequenced , mapped and normalized . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 017Taxon CodeTissueLanesQC-passed readsMapped reads ( % total ) Normalized subset ( %total ) ( % mapped ) Accession ReadsAccession BAMsDOMCBBrain0 . 33x1 . 30E+081 . 26E+0896%9 . 15E+0770%73%ERS946023ERS942305DOMCBLiver0 . 33x1 . 41E+081 . 17E+0883%9 . 07E+0764%77%ERS946025ERS942306DOMCBTestis0 . 33x1 . 26E+081 . 22E+0896%1 . 19E+0894%98%ERS946026ERS942307DOMMCBrain0 . 33x1 . 17E+081 . 13E+0896%9 . 15E+0778%81%ERS946027ERS942309DOMMCLiver0 . 33x1 . 34E+081 . 09E+0881%9 . 07E+0768%84%ERS946029ERS942310DOMMCTestis0 . 33x1 . 42E+081 . 37E+0896%1 . 19E+0883%87%ERS946030ERS942311DOMAHBrain0 . 33x9 . 49E+079 . 15E+0796%9 . 15E+0796%100%ERS946019ERS942301DOMAHLiver0 . 33x1 . 16E+081 . 02E+0888%9 . 07E+0778%89%ERS946021ERS942302DOMAHTestis0 . 33x1 . 61E+081 . 55E+0896%1 . 19E+0874%77%ERS946022ERS942303MUSKHBrain0 . 33x1 . 33E+081 . 28E+0896%9 . 15E+0769%72%ERS946035ERS942313MUSKHLiver0 . 33x1 . 03E+089 . 07E+0788%9 . 07E+0788%100%ERS946037ERS942314MUSKHTestis0 . 33x1 . 36E+081 . 31E+0896%1 . 19E+0887%91%ERS946038ERS942315MUSVIBrain0 . 33x1 . 23E+081 . 19E+0896%9 . 15E+0774%77%ERS946031ERS942317MUSVILiver0 . 33x1 . 23E+089 . 47E+0777%9 . 07E+0774%96%ERS946033ERS942318MUSVITestis0 . 33x1 . 32E+081 . 27E+0896%1 . 19E+0890%93%ERS946034ERS942319CASBrain0 . 33x1 . 21E+081 . 16E+0896%9 . 15E+0776%79%ERS946039ERS942321CASLiver0 . 33x1 . 23E+081 . 01E+0882%9 . 07E+0774%90%ERS946041ERS942322CASTestis0 . 33x1 . 23E+081 . 19E+0896%1 . 19E+0896%100%ERS946042ERS942323SPIBrain0 . 33x1 . 34E+081 . 29E+0896%9 . 15E+0768%71%ERS946043ERS942325SPILiver0 . 33x1 . 05E+089 . 82E+0793%9 . 07E+0786%92%ERS946045ERS942326SPITestis0 . 33x1 . 44E+081 . 38E+0896%1 . 19E+0883%86%ERS946046ERS942327SPRBrain0 . 33x1 . 09E+081 . 05E+0896%9 . 15E+0784%87%ERS946047ERS942329SPRLiver0 . 33x1 . 35E+081 . 20E+0889%9 . 07E+0767%76%ERS946049ERS942330SPRTestis0 . 33x1 . 34E+081 . 29E+0896%1 . 19E+0888%92%ERS946050ERS942331MATBrain0 . 33x1 . 12E+081 . 04E+0893%9 . 15E+0782%88%ERS946051ERS942333MATLiver0 . 33x1 . 23E+081 . 12E+0891%9 . 07E+0774%81%ERS946053ERS942334MATTestis0 . 33x1 . 32E+081 . 23E+0893%1 . 19E+0890%97%ERS946054ERS942335APOBrain0 . 33x1 . 36E+081 . 18E+0887%9 . 15E+0767%78%ERS946055ERS942337APOLiver0 . 33x1 . 13E+081 . 00E+0889%9 . 07E+0780%91%ERS946057ERS942338APOTestis0 . 33x1 . 38E+081 . 20E+0887%1 . 19E+0886%99%ERS946058ERS942339All accessions deposited under studies PRJEB11533 and PRJEB11513** . Our results provide an evolutionary dynamics perspective where emergence , functionalization and decay of gene functions should be seen as an evolutionary life cycle of genes ( Neme and Tautz , 2014 ) . De novo gene birth should no longer be considered as the result of unlikely circumstances , but rather as an inherent property of the transcriptional apparatus and thus a mechanism for testing and revealing hidden adaptive potential in genomes ( Brosius , 2005; Masel and Siegal , 2009 ) . Within this evolutionary perspective , any non-genic part of the genome has the possibility to become useful at some time . 10 . 7554/eLife . 09977 . 018Table 3 . Additional sequencing effort , focused only on brain samples . Reads sequenced , mapped and normalized . DOI: http://dx . doi . org/10 . 7554/eLife . 09977 . 018Taxon CodeTissueLanesQC-passed readsMapped reads ( % total ) Normalized subset ( % total ) ( % mapped ) Accession ReadsAccession BAMsDOMCBBrain1x3 . 89E+083 . 76E+0897%3 . 19E+0882%85%ERS946024ERS942308DOMMCBrain1x3 . 76E+083 . 64E+0897%3 . 19E+0885%88%ERS946028ERS942312DOMAHBrain1x3 . 46E+083 . 35E+0897%3 . 19E+0892%95%ERS946020ERS942304MUSKHBrain1x4 . 64E+084 . 49E+0897%3 . 19E+0869%71%ERS946036ERS942316MUSVIBrain1x4 . 13E+084 . 00E+0897%3 . 19E+0877%80%ERS946032ERS942320CASBrain1x4 . 35E+084 . 21E+0897%3 . 19E+0873%76%ERS946040ERS942324SPIBrain1x4 . 31E+084 . 16E+0897%3 . 19E+0874%77%ERS946044ERS942328SPRBrain1x3 . 87E+083 . 73E+0896%3 . 19E+0882%85%ERS946048ERS942332MATBrain1x3 . 62E+083 . 40E+0894%3 . 19E+0888%94%ERS946052ERS942336APOBrain1x4 . 33E+083 . 77E+0887%3 . 19E+0874%84%ERS946056ERS942340All accessions deposited under studies PRJEB11533* and PRJEB11513** . The youngest divergence point sampled , at about 3 , 000 years , corresponds to the split between two European populations of Mus musculus domesticus ( Cucchi et al . , 2005 ) one from France ( Massif Central = DOMMC ) and one from Germany ( Cologne-Bonn area = DOMCB ) ( Ihle et al . , 2006 ) . These European populations in turn have diverged from an ancestral M . m . domesticus population in Iran ( Ahvaz = DOMAH ) about 12 , 000 years ago ( Hardouin et al . , 2015 ) . The European M . m . domesticus are also the closest relatives of the reference genome , the C57BL/6J strain Didion and de Villena , 2013 ) . We included two populations of Mus musculus musculus; one from Austria ( Vienna = MUSVI ) and one from Kazakhstan ( Almaty = MUSKH ) . These two populations are supposed to have a longer divergence between then the European M . m . domesticus populations , but a more accurate estimate is currently not available . We set the divergence for analyses at around 10 , 000 years as an approximate estimate . M . m . domesticus has diverged from M . m . musculus and Mus musculus castaneus about 0 . 4 to 0 . 5 million years ago , with a subsequent divergence , not long after , between M . m . musculus and M . m . castaneus ( Suzuki et al . , 2013 ) . We included M . m . castaneus ( CAS ) from Taiwan as a representative of the subspecies . To account for longer divergence times , we included Mus spicilegus ( SPI; estimated divergence of 1 . 2 million years ) ; Mus spretus ( SPR; estimated divergence of 1 . 7 million years ) ( Suzuki et al . , 2013 ) ; Mus matteyii ( MAT; subgenus Nannomys ) , the North African miniature mouse ( estimated divergence of 6 . 6 million years ) ( Catzeflis and Denys , 1992; Lecompte et al . , 2008 ) , and Apodemus uralensis , the Ural field mouse ( APO; estimated divergence of 10 . 6 million years ) ( Lecompte et al . , 2008 ) . The population-level samples ( M . m . domesticus and M . m . musculus ) included are maintained under outbreeding schemes , which allows for natural polymorphisms to be present in the samples . All other non-population samples are kept as more or less inbred stock , and therefore fewer polymorphisms are expected . All mice were obtained from the mouse collection at the Max Planck Institute for Evolutionary Biology , following standard rearing techniques which ensure a homogeneous environment for all animals . Mice were maintained and handled in accordance to FELASA guidelines and German animal welfare law ( Tierschutzgesetz § 11 , permit from Veterinäramt Kreis Plön: 1401–144/PLÖ-004697 ) . A total of 60 mice were sampled , as follows: Eight male individuals from each population-level sample ( outbreds ) , Iran ( DOMAH ) , France ( DOMMC ) , and Germany ( DOMCB ) of Mus musculus domesticus , and Austria ( MUSVI ) and Kazakhstan ( MUSKH ) of Mus musculus musculus . Four male individuals from the remaining taxa ( partially inbred ) : Mus musculus castaneus ( CAS ) , Mus spretus ( SPR ) , Mus spicilegus ( SPI ) , Mus mattheyi ( MAT ) and Apodemus uralensis ( APO ) . Mice were sacrificed by CO2 asphyxiation followed immediately by cervical dislocation . Mice were dissected and tissues were snap-frozen within 5 min post-mortem . The tissues collected were liver ( ventral view: front right lobe ) , both testis and whole brain including brain stem . One individual from each of M . spicilegus , M . spretus , M . mattheyi , and Apodemus uralensis were selected for genome sequencing . DNA was extracted from liver samples . DNA extraction was performed using a standard salt extraction protocol . Tagged libraries were prepared using the Genomic DNA Sample preparation kit from Illumina , following the manufacturers’ instructions . After library preparation , the samples were run in IlluminaHiSeq 2000 at a depth of approximately 2 . 6 lanes per genome . Library insert size is ~190bases and paired-end reads were 100 bases long . Library preparation and sequencing was performed at the Cologne Center for Genomics . Sequencing read statistics are provided in Table 1 . Data are available under the study accessions PRJEB11513 , PRJEB11533 and PRJEB11535 , from the European Nucleotide Archive ( http://www . ebi . ac . uk/ena/ ) . The sampled tissues of each taxon were used for RNA extraction with the RNAeasy Mini Kit ( QIAGEN ) and RNA was pooled at equimolar concentrations . RNA quality was measured with the Agilent RNA Nano Kit , for the individual samples and pools . Samples with RIN values above 7 . 5 were used for sequencing . Library preparation was done using the Illumina TruSeq library preparation , with mRNA purification ( poly-A+ selection ) , following manufacturers’ instructions . Sequencing was done in Illumina HiSeq , 2000 sequencer . Libraries for each group were tagged , pooled and sequenced in a single lane , corresponding to approximately one third of a HiSeq2000 lane . Library insert size is ~190bases and paired-end reads were 100 bases long . Additional sequencing of the brain samples was performed to identify potential limitations in depth of sequencing . For this , each brain library was sequenced on a full lane of a HiSeq2000 . All library preparation and sequencing was done at the Cologne Center for Genomics ( CCG ) . Sequencing read statistics are provided in Tables 2 and 3 . Data are available under the study accessions PRJEB11513 and PRJEB11533 , from the European Nucleotide Archive ( http://www . ebi . ac . uk/ena ) . All raw data files were trimmed for adaptors and quality using Trimmomatic ( Lohse et al . , 2012 ) . The quality trimming was performed basewise , removing bases below quality score of 20 ( Q20 ) , and keeping reads whose average quality was of at least Q30 . Reads whose trimmed length was shorter than 60 bases were excluded from further analyses , and pairs missing one member because of poor quality were also removed from any further analyses . The reconstruction of transcriptomes using high-throughput sequencing data is not trivial when comparing information across different species to a single reference genome . This is due to the fact that most of the tools designed for such tasks do not work in a phylogenetically aware context . For this reason , any approximation which deals with fractional data ( i . e . any high-throughput sequencing setup available to this date ) is limited by the detection abilities of the software of choice and by the quality of the reference ( transcriptome and genome ) . Given the high quality state of the mouse genome repositories , we decided to take a reference-based approach , in which all analyses are centered in the reference genome of the C57BL/6 laboratory strain of Mus musculus domesticus , which enables direct comparisons across all species based on the annotations of the C57BL/6 laboratory strain . Transcriptome and genome sequencing reads were aligned against the mm10 version of the mouse reference genome ( Waterston et al . , 2002 ) from UCSC ( Karolchik et al . , 2014 ) using NextGenMap which performs extremely well with divergences of over 10% compared to other standard mapping software ( Sedlazeck et al . , 2013 ) , as confirmed by our own simulations ( Appendix 1 ) . The program was run under default settings , except for --strata 1 and --silent-clip . The first option enforces uniquely mapping reads and the second drops the unmapped portion of the reads , to avoid inflating coverage statistics . This is particularly relevant around exon-intron boundaries , where exonic reads are forced into intronic regions unless this option is set . We produced normalized versions of the alignments per tissue . This was achieved by counting the total amount of uniquely mapped reads in each taxon for each tissue , and sampling without replacement a fraction of each file which would result in the roughly the same absolute number of uniquely mapped reads for all samples of the same tissue ( summarized in Table 2 and Table 3 ) . We performed coverage statistics on 200 bp windows , to minimize problems derived from the fractional nature of the data , in which a few nucleotides could be absent from a sequenced fragment due to the preparation of the samples , low quality towards read ends , or a few mismatches during mapping . Coverage statistics were computed from normalized alignment files with the featureCounts program from the Subreads suite ( Liao et al . , 2014 ) . In order to avoid counting reads twice if they would span two windows ( which would be the case for most reads ) , we assigned reads to the window where more than half of the read was present . Genomic reads were used as a metric of empiric mapability for the coverage statistics , i . e . to identify which regions can be reliably detected . For this , we removed from the mapping results against the reference genome ( see above ) all regions that were not mapped across the phylogeny based on the genomic reads from the taxa more than 1 Mill years apart . The remaining portion we call the ‘common genome’ in all analyses . It is important to highlight that this is not the same as synteny , since we did not perform any co-linearity analyses between fragments , but rather represent the mere presence in the species , in any possible order . The common genome serves to limit mapping artifacts , since the reads observed in each window must not only be uniquely mapping , but also be present and detectable in all the genomes considered . We report coverage only from windows in the common genome for several reasons . First , we want to compare changes in transcription in regions which are present across all taxa , so the region must be present at the genome level . Second , the observation of transcriptome coverage on a region of the reference genome without genomic coverage from the respective taxon could represent mapping artifacts . Thus by enforcing coverage on both levels , and in all taxa at the genomic level , we reduce mapping artifacts and errors . Third , we assume that the transcriptional properties of the common genome should be general enough that they represent the properties of each of the genomes of the taxa under study . Summary data for coverage of all genomes and transcriptomes are available under the Dryad accession associated with this manuscript ( doi:10 . 5061/dryad . 8jb83 ) . We performed genome-wide correlations of coverage to infer distance between the taxa under study . Correlations of two types were initially used: rank-based ( spearman correlation ) and binary ( phi correlation ) . From correlation matrices , we constructed Manhattan distance matrices and from those we further constructed neighbor-joining trees to describe the proximity between any two taxa based on shared transcriptome information . We focus mostly on the presence or absence of transcriptional coverage . For this reason , we used only the binary correlations in the figures . In this representation , closely related organisms have more shared transcriptomic coverage than distantly related organisms . Analyses were performed in R , using the function dist ( ) from the stats package and nj ( ) from the ape package ( Paradis et al . , 2004 ) . Additionally , whole mitochondrial genomes were obtained for each taxon as consensus sequences from mapped reads using samtools mpileup ( Li et al . , 2009 ) . The sequences were aligned with MUSCLE ( Edgar , 2004 ) , and a NJ tree was constructed with the dist . dna ( ) and nj ( ) functions from the ape package Paradis et al . , 2004 ) . All trees were tested with 1000 bootstraps with the boot . phylo ( ) function from the ape package . Reported nodes have a support of 70% or greater . The extensive sequencing of brain samples were used to obtain estimates of how sampling might affect the terminal branch lengths of trees based on low coverage regions . For this , we split the alignments into three non-overlapping sets of 100 million reads per taxon , such that each set would contain sets of independent observations . Paired-read relationships were maintained , so that pairs of the same fragments would be in the same set . From this , we obtained trees as mentioned before , and the portions of the branches of each taxon which were shared across sets were considered as robust to sampling biases , while the discordant portions between samples were considered to be due to sampling variance . Summary data from subsampled sets are available under the Dryad accession associated with this manuscript ( doi:10 . 5061/dryad . 8jb83 ) . Transcriptome experiments tend to be limited by the depth of sequencing , with highly expressed genes being relatively easy to sample , and rare transcripts becoming increasingly difficult to find . Given the large amount of data generated , we investigated whether our data show signals of coverage saturation from subsets of the data of different sizes . The total experiment , comprising ten taxa , corresponds to 6 . 4 x 109 reads ( or 6 . 4 billion reads ) . We subsampled ( samtools view -s ) portions of mapped reads for each taxon , ranging between 10% to 100% , at 10% intervals . The observation of coverage saturation in this case would indicate that our sequencing efforts likely cover most of the transcribed regions of the common genome . Summary data are available under the Dryad accession associated with this manuscript ( doi:10 . 5061/dryad . 8jb83 ) . In parallel , we estimated the individual and combined contribution of each taxon to the transcriptomic coverage of the common genome . Not all samples have the same phylogenetic distance to each other ( some species have more representatives than others ) . To account for this we generated one hundred arrays of the ten taxa with random order , and recorded the coverage after the addition of each taxon in each array . The observation of coverage saturation in this setup would indicate that taxonomic sampling is sufficient to cover most of the potentially transcribed regions of the common genome . In order to estimate whether our data continued to increase or approached saturation , we tested two alternative models: a generalized linear model with logarithmic behavior ( ever increasing ) or a self-starting nonlinear regression model ( saturating ) . The best fit was decided based on the minimum BIC value between the two models , and an estimate of the Bayes factor was computed from the difference of BIC values and support was obtained from standard criteria ( Kass and Raftery , 1995 ) . Analyses were performed in R , using the functions glm ( ) , nls ( ) , SSasymp ( ) , and BIC ( ) from the stats package ( R Core Team , 2014 ) . Transcribed and non-transcribed windows of the common genome were defined by the continuous presence or absence of transcriptomic coverage from mapping information of each taxon and tissue . Neighboring transcribed regions across species were combined to obtain stretches of transcriptionally active common genome . Annotations of Mus musculus from Ensembl v81 ( Cunningham et al . , 2015 ) were used to infer the relative contribution of known genes to the observed transcription across species . We partitioned the sets between genes , exons , and introns , and those were further partitioned between protein- coding and non-coding genes . To determine if the overlaps are significantly different from a random distribution of the features along the genome , we randomized 1000 times each of the annotated intervals ( genes , exons , introns , and subsets of coding and non-coding ) along the genome using shuffleBed from the bedtools suite ( Quinlan and Hall , 2010 ) , and compared the overlap to various transcribed regions ( single taxa , less than 9 taxa , more than 8 taxa , 10 taxa , and transcribed in any taxon ) . Multiple testing corrections were performed and significant comparisons are reported at 5% FDR . Furthermore , since we assume that most annotations fall within transcribed regions in any species , we used the total transcriptomic coverage across all taxa to calculate potential discrepancies in the shuffling method . The ratios of expected and observed coverage of total transcription across taxa for a given feature were calculated to define the range of ratios for which comparisons were also non-significant , i . e . , where we could not rule out method bias .	Traditionally , the genome – the sum total of DNA within a cell – was thought to be divided into genes and ‘non-coding’ regions . Genes are copied , or “transcribed” , into molecules called RNA that perform essential tasks in the cell . The roles of the non-coding regions were often less clear , although it has since become apparent that some are also transcribed and generate low levels of RNA molecules . However , many debate how significant this transcription is to living organisms . Neme and Tautz have now used a technique called deep RNA sequencing to study the RNA molecules produced in several different species and types of mice whose last common ancestor lived 10 million years ago . Different species produced RNA molecules from different portions – both genes and non-coding regions – of their genomes . Comparing these RNA sequences suggests that changes to the regions that are transcribed occur relatively quickly for a large portion of the genome . Furthermore , there have been no significant areas of the common ancestor’s genome that have not been transcribed at some point in at least one of its descendent species . This therefore suggests that over a relatively short evolutionary period , any part of the genome can acquire the ability to be transcribed and potentially form a new gene . The next challenge is to find out how often these transcribed non-coding parts of the genome show important biochemical activities , and how they find their way into becoming new genes .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2016	Fast turnover of genome transcription across evolutionary time exposes entire non-coding DNA to de novo gene emergence
To better understand how organisms make decisions on the basis of temporally varying multi-sensory input , we identified computations made by Drosophila larvae responding to visual and optogenetically induced fictive olfactory stimuli . We modeled the larva's navigational decision to initiate turns as the output of a Linear-Nonlinear-Poisson cascade . We used reverse-correlation to fit parameters to this model; the parameterized model predicted larvae's responses to novel stimulus patterns . For multi-modal inputs , we found that larvae linearly combine olfactory and visual signals upstream of the decision to turn . We verified this prediction by measuring larvae's responses to coordinated changes in odor and light . We studied other navigational decisions and found that larvae integrated odor and light according to the same rule in all cases . These results suggest that photo-taxis and odor-taxis are mediated by a shared computational pathway . Many small organisms navigate their environments by decoding temporal variations in receptor activity to bias motor output decisions ( Berg and Brown , 1972; Sawin et al . , 1994; Pierce-Shimomura et al . , 1999; Ryu and Samuel , 2002; Fishilevich et al . , 2005; Louis et al . , 2008; Luo et al . , 2008 , 2010; Albrecht and Bargmann , 2011; Gomez-Marin et al . , 2011; Lockery , 2011; Gershow et al . , 2012; Kane et al . , 2013 ) . The dynamics of these organisms' decision-making are intimately linked to the properties of the underlying chemical or neural substrates ( Segall et al . , 1986; Korobkova et al . , 2004; Miller et al . , 2005; Chronis et al . , 2007; Clark et al . , 2007; Emonet and Cluzel , 2008; Suzuki et al . , 2008; Asahina et al . , 2009; Shimizu et al . , 2010; Bretscher et al . , 2011; Busch et al . , 2012; Kato et al . , 2014; Luo et al . , 2014 ) . In their natural environments , animals are confronted with variable and frequently conflicting inputs arriving via multiple sensory pathways . How these simple organisms respond to multi-modal stimuli and how their neural circuits integrate and prioritize multi-sensory information remains unknown . We sought to decode the computations underlying the Drosophila larva's response to visual and olfactory cues , when presented individually or in combination . The second instar larva uses a similar strategy for navigating environments with spatially varying odor concentrations , light levels , and temperatures . Larvae move forward in a series of relatively straight runs interspersed with reorienting turns , and increase the frequency and magnitude of their turns in response to unfavorable changes in the stimulus ( Sawin et al . , 1994; Busto et al . , 1999; Scantlebury et al . , 2007; Louis et al . , 2008; Luo et al . , 2010; Gomez-Marin et al . , 2011; Gershow et al . , 2012; Kane et al . , 2013; Klein et al . , 2014 ) . During turns , larvae survey the local environment using side-to-side head sweeps to determine the direction of the next run ( Sawin et al . , 1994; Luo et al . , 2010; Gomez-Marin et al . , 2011; Gershow et al . , 2012; Kane et al . , 2013; Klein et al . , 2014 ) . That the larva uses the same strategy to respond to a variety of stimuli suggests navigation may be mediated by a single circuit that combines input from the various sensory organs . Alternately , this apparent commonality might result from the larva's limited repertoire of motor outputs with which to implement a navigational response . Perhaps , independent circuits mediate the ‘decision’ portion of each navigational algorithm and only converge at the motor output level ( Frye and Dickinson , 2004 ) . We sought a computational model that could describe the transformation from sensory input to navigational decisions for uni- and multi-modal stimuli and that could differentiate between shared and parallel navigational circuits . The Linear-Nonlinear-Poisson ( LNP ) model ( Chichilnisky , 2001; Dayan , 2001; Ringach and Shapley , 2004; Bialek and van Steveninck , 2005; Schwartz et al . , 2006; Kim et al . , 2011 ) is widely used to relate time-varying input to stochastic output . In this model , decisions ( e . g . , to initiate a turn , Figure 1A , B ) are generated according to a Poisson process whose underlying rate at time t is:r ( t ) =f[ ( A∗S ) ( t ) ]; ( A∗S ) ( t ) =∫0∞A ( τ ) S ( t−τ ) dτ , where S ( t ) is the input signal , A is a linear filter , and f is a static nonlinear function ( Figure 1B ) . Thus for uni-modal inputs , we aimed to develop models of the form:rO ( t ) =f ( xO ( t ) ) ( odor ) , rL ( t ) =g ( xL ( t ) ) ( light ) , where xO ( t ) = ( AO∗SO ) ( t ) and xL ( t ) = ( AL∗SL ) ( t ) are the outputs of the linear filters for odor and for light , respectively . For multi-modal input , we sought either a model in which turns are initiated independently by separate circuitsrO−L ( t ) =f ( xO ( t ) ) +g ( xL ( t ) ) ( independent pathways ) , or one in which odor and light information are combined more generallyrO−L ( t ) =h ( xO ( t ) , xL ( t ) ) ( nonlinear integration ) . 10 . 7554/eLife . 06229 . 003Figure 1 . Identifying computations underlying the decision to initiate a turn . ( A ) Computation: on the basis of sensory input ( light or odor in this work ) the larva decides whether or not to end a run and begin a turn . ( B ) LNP model of the computation: Sensory input is processed by a linear filter to produce an intermediate signal . The rate at which the larva initiates turns is a nonlinear function of this signal . Turns are initiated stochastically according to a Poisson process with this underlying turn rate . ( C ) Reverse-correlation to determine LNP parameters: ( 1 ) We presented groups of larvae with either blue or red light with randomly varying intensity derivatives . Blue light provided a visual stimulus , while red light activated CsChrimson expressed in sensory neurons . In multi-sensory experiments , uncorrelated red light and blue light signals were presented simultaneously . Larvae were observed under infrared illumination , and their behaviors were analyzed with machine vision software . ( 2 ) We calculated the ‘turn-triggered average’ ( TTA ) or the reverse-correlation between turn initiation and stimulus by averaging the stimulus that preceded each moment any larva initiated a turn . The TTA approximates the convolution kernel for the linear response of the LNP model . ( 3 ) Using the TTA as a convolution kernel and the known input signal , we computed the intermediate filtered signal . ( 4 ) Using the inferred filtered signal and the observed times at which turns were initiated , we found the nonlinear rate function by dividing the distribution of filtered signal values at the time of turn initiation ( ‘turn-triggered-ensemble’ ) by the distribution of all filtered signal values . Illustrations adapted from ( Kane et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 003 We began by exploring larvae's responses to uni-modal stimuli . Drosophila larvae avoid light and carbon dioxide and are attracted to Ethyl Acetate ( EtAc ) . The sensory input pathways are well-characterized for each of these stimuli . The larva's navigational response to light is mediated primarily by four photoreceptor neurons in each of two primitive eye-spots ( Hassan et al . , 2005; Sprecher and Desplan , 2008; Keene et al . , 2011; Keene and Sprecher , 2012; Kane et al . , 2013 ) . A single pair of Gr21a expressing receptor neurons mediates the larva's CO2 response ( Python and Stocker , 2002; Faucher , 2006; Jones et al . , 2007; Kwon et al . , 2007 ) . Or42a and Or42b are the primary EtAc olfactory receptors ( Kreher et al . , 2005 , 2008; Asahina et al . , 2009 ) , and larvae are capable of decoding odor gradients on the basis of only Or42a or Or42b receptor neurons ( Louis et al . , 2008; Asahina et al . , 2009 ) . Larvae initiate turns in response to increasing light intensities ( Sawin et al . , 1994; Hassan et al . , 2005; Scantlebury et al . , 2007; Kane et al . , 2013 ) and carbon dioxide concentrations ( Gershow et al . , 2012 ) and decreasing EtAc concentrations ( Gomez-Marin et al . , 2011; Gershow et al . , 2012 ) . To investigate the larva's decision to turn in response to visual cues , we presented 448 nm blue light to wild-type larvae . To probe olfactory computations , we expressed UAS-CsChrimson ( Klapoetke et al . , 2014 ) , a red light activable cation channel , in Gr21a , Or42a , and Or42b receptor neuron pairs . We used 655 nm red light ( outside the sensitive range of the larva's visual pigments [Salcedo et al . , 1999] ) to activate these neurons while presenting a constant dim blue light background ( Klapoetke et al . , 2014 ) to mask any visual response to the red light . We presented groups of larvae with fluctuating levels of red or blue light and analyzed their resulting behaviors using machine vision software ( Gershow et al . , 2012 ) to identify each navigational decision ( Figure 1C-1 ) . We determined the parameters of the LNP model by measuring the reverse-correlation ( Chichilnisky , 2001; Dayan , 2001; Westwick et al . , 2003; Ringach and Shapley , 2004; Bialek and van Steveninck , 2005; Schwartz et al . , 2006; Kim et al . , 2011; Klein et al . , 2014 , Theobald et al . , 2010 ) between the stimulus and evoked behaviors ( Figure 1C , Video 1 ) . 10 . 7554/eLife . 06229 . 015Video 1 . Calculating the turn-triggered average . Left panel: annotated video image of individual larva . Thin white line: larva's path ( past and future ) . Gold dots: markers along midline of animal , used to determine posture . Upper left corner: time ( since experiment start ) and behavioral state . Right panels: top: light derivative ( AU ) vs . time; current time is indicated by dashed cyan line . Middle panels: speed and body bend angle , metrics used to determine behavioral state , vs . time . Shading indicates behavioral state ( blue = run , white = turn; within turns , red = rejected head sweep , green = accepted head sweep ) . Current time is indicated by cyan dot . Bottom panel: turn-triggered average light ( TTA ) intensity derivative ( AU ) , calculated based on turns preceding the one shown . Animation: The time preceding and following individual turns is featured . At the moment a larva initiates a turn , we ‘grab’ the sequence of light intensity derivatives and add it to a running average ( shown at the bottom ) . As we include more turns in the average ( number of included turns is indicated by ‘turn #’ above the left panel—note logarithmic spacing of turn #s ) , we build up a ‘TTA’ that approximates the linear filter in the LNP model . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 015 For light , odor , and carbon dioxide , the derivative of stimulus intensity is more salient to larvae than the stimulus value itself ( Gomez-Marin et al . , 2011; Gershow et al . , 2012; Kane et al . , 2013; Gomez-Marin and Louis , 2014 ) . Typically in reverse-correlation experiments , rapidly changing uncorrelated stimulus intensities are provided ( e . g . by choosing random binary values , random normally distributed values , or values from an M-sequence ) . Because of regression to the mean , these sequences have derivatives that are anticorrelated on time scales longer than the update period . Thus , if we provided uncorrelated stimulus intensities and larvae turned in response to an increase in light , we would expect an average decrease in light after turns . This would complicate analyses of decisions , like whether to accept or reject a head sweep , the larva makes following a turn . We therefore chose a stimulus optimized for analysis of the larva's response to derivatives , a Brownian random walk , whose derivatives ( on all time scales ) are independent identically distributed Gaussian variables . First , we computed the turn-triggered average ( TTA ) of the light intensity derivatives ( Figure 2A ) . We found: wild-type larvae turn in response to an increase in blue light but were unresponsive to red light masked by dim blue light ( even when fed all-trans-retinal ) ; larvae expressing CsChrimson in EtAc sensing neurons ( Or42a>CsChrimson , Or42b>CsChrimson ) turned in response to a decrease in red light; and larvae expressing CsChrimson in the CO2 receptor ( Gr21a>CsChrimson ) turned in response to an increase in red light . These results match the larva's strategies for navigating static gradients of natural stimuli; turning in response to increases in light , decreases in EtAc , and increases in CO2 . In all cases ( excepting the wild-type red-light control ) , larvae considered changes over the previous ∼2 . 5 s when deciding to turn , and derivatives were maximally salient about 0 . 6–0 . 8 s before a turn was actually initiated . 10 . 7554/eLife . 06229 . 004Figure 2 . Unimodal reverse-correlation experiments . Top row , Berlin wild-type larvae were stimulated with blue ( λpeak = 448 nm; max intensity = 74 μW/cm2 ) light . All other rows , larvae of indicated genotype were stimulated with red light ( λpeak = 655 nm; max intensity = 911 μW/cm2 ) while constant dim blue light ( intensity = 3 . 7 μW/cm2 ) served as a visual mask . Column A: Turn triggered average . Average stimulus preceding ( and following ) each turn initiation . Turns are initiated at time 0 ( indicated with dashed line ) . The black line is the smoothed TTA used as the linear filter . Column B: Measured turn rates as a function of calculated filter output . Line and shaded region represent mean turn rate and standard error due to counting statistics . Black line is the nonlinear turn rate modeled as a ratio-of-Gaussians ( Pillow and Simoncelli , 2006 ) . Column C: Step responses predicted by LNP model . Square waves of light with period 20 s and duty cycle 50% were presented to larvae . The LNP model was used to predict the resulting turn rates . Top graphs: light level vs time in cycle . A favorable change happens at t = 0 and an unfavorable change at t = 10 s . Bottom graphs: measured and predicted turn rates vs time in cycle . Black line and shaded region represent mean turn rate and standard error due to counting statistics . The cyan line is the exact prediction of the model using the parameters found from the corresponding reverse-correlation experiments . ( A , B ) The stimulus and analysis were cyclic , so the time range from −2 to 0 s is identical to that from 18–20 s . Column D: Size-sorted turn-triggered average . As in A , but turns were sorted into large ( heading change during turn > rms heading change ) and small turns . Displayed averages are lowpassed with a Gaussian filter ( σ = 0 . 5 s ) to clarify the long time-scale features . Column E: Head-sweep triggered average ( for first head-sweep of turn ) . Average stimulus surrounding accepted ( teal ) and rejected ( red ) head-sweeps . Head sweeps were initiated at time t = 0 and concluded at a variable time in the future . The mean head-sweep duration ( 1 . 25 s ) is indicated by the shaded region . See Table 1 for number of experiments , animals , and so on . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 00410 . 7554/eLife . 06229 . 005Figure 2—figure supplement 1 . LNP model parameters are stable for duration of 20 min experiments . Experiments of Figure 2 analyzed separately using data only from the first 10 min of experiment ( teal ) or only from the second 10 min of experiment ( purple ) or from entire 20 min data set ( black ) . Column A: Turn triggered average . As in Figure 2A . The same convolution kernel is recovered from all three data sets . Column B: Measured turn-rates as a function of calculated filter output . As in Figure 2B . The turn rates vary across the three data sets mainly at high values of the filter output ( stimulus conditions most likely to lead to turning ) . Column C: LNP model fits can predict response to white noise signals . Data from the first 10 min of the experiments were used to find LNP model parameters . The parameterized models were then used to predict the turn rate at each time point during the second 10 min of experiments . The measured turn rate during the second 10 min is plotted as a function of the model predictions . At high turn rates , for visual and fictive attractive odor stimuli , the turn rate is lower in the final 10 min than predicted by fits to the first 10 min , suggesting modest adaptation . For fictive CO2 , the measured turn rate is higher than predicted , suggesting modest sensitization . Error bars represent the uncertainty due to counting statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 005 For each set of reverse-correlation experiments , we used the TTA as the convolution kernel to calculate the linear filter outputs and computed the turn rate as a nonlinear function of the filter output ( Figure 2B ) . To test the predictive power of the LNP model , we presented larvae with light intensity square waves . We compared the resulting turn rates to those predicted by our LNP model fits ( Figure 2C ) . The cyan lines in Figure 2C show the exact predictions of the model , with no free parameters . We found that in all cases , except for Gr21a>CsChrimson , the step response has a longer duration than predicted by the LNP model , and there is an unpredicted sustained decrease in turn rate following a favorable change ( at t = 0 ) , but given the inherent variability of behavior and the simplicity of the LNP model , the reverse-correlation experiments predicted the temporal course of the larva's response to step changes surprisingly well . We were concerned that over the course of the 20 min behavioral experiments , continuous exposure to red light might exhaust the ability of CsChrimson to excite neural activity or that larvae might adapt to the stimulus presentation and cease responding . To test whether this was the case , we separately analyzed the first and second 10 min of the experiments ( Figure 2—figure supplement 1 ) . In all cases , we recovered the same TTA ( Figure 2—figure supplement 1A ) using data from only the first 10 min , only the second 10 min , or the entire data set . We found that for visual and attractive odor inputs , the slope of the nonlinear function was steeper ( Figure 2—figure supplement 1B ) in the first 10 min than in the second 10 , indicating larvae were slightly less responsive to stimulus changes in the latter half of the experiments . For fictive CO2 , on the other hand , we found that larvae were actually more responsive in the second 10 min , representing sensitization rather than adaptation . In all cases , the results for the two halves of the experiments were similar enough to justify using all 20 min of data in our analyses . To test the self-consistency of our model , we used the parameters extracted from the first 10 min of experiments to predict the larva's turning rate in the second 10 min ( Figure 2—figure supplement 1C ) . We found excellent agreement between predictions and measurements at lower turn rates . At higher turn rates , we found that in the second 10 min , larvae turned less than predicted for light and attractive odor cues and more than predicted for CO2 cues , again reflecting modest adaptation and sensitization . Next , we explored how larvae bias the size and directions of turns . Larvae make larger turns when subjected to unfavorably changing conditions ( Luo et al . , 2010; Gershow et al . , 2012; Kane et al . , 2013 ) , so we calculated the TTA for large and small turns separately ( Figure 2D ) . For approximately 10–15 s prior to the start of large turns , there was an average gradual increase in blue light levels for visual experiments , and an average gradual decrease in red light levels for Or42a>CsChrimson and Or42b>CsChrimson experiments . Over the same time period , for small turns , there was a slight average decrease in blue light and increase in red light . Interestingly , immediately prior to the initiation of a turn , the average change in stimulus was the same for large and small turns . Thus , larvae consider the change in light intensity and attractive odor concentration over the previous 10–15 s when deciding the size of their turns , but the size of the stimulus change that leads the larva to actually initiate a turn appears not to influence turn size . In contrast , for Gr21a>CsChrimson larvae , the size of a turn was determined by the magnitude of the increase immediately preceding a turn . After initiating turns , larvae use head-sweeps as probes to find a favorable direction for the next run ( Luo et al . , 2010; Gomez-Marin et al . , 2011; Gershow et al . , 2012; Kane et al . , 2013; Gomez-Marin and Louis , 2014; Klein et al . , 2014 ) . Head-sweeps in favorable directions are more likely to be accepted , beginning a new run in that direction . Head-sweeps in unfavorable directions are more likely to be rejected , resulting in one or more additional head-sweeps . To characterize the larva's decision to accept or reject a head-sweep , we measured the head-sweep triggered average , aligned to the start of a head-sweep , separately for rejected and accepted head-sweeps ( Figure 2E ) . All head-sweep triggered averages show a large change immediately before the start of the head-sweep; this is the stimulus change that triggered the larva's decision to turn . We found that during accepted head-sweeps , average blue light levels and Gr21a activity decreased and Or42a and Or42b activity increased , all favorable changes . During rejected head-sweeps , the reverse was true: average blue light and Gr21a activity levels increased and Or42a and Or42b activity decreased . Surprisingly , for Gr21a>CsChrimson larvae , larger increases in activity prior to the head-sweep start led to an increased probability of rejecting the head-sweep . Our uni-modal experiments showed that CsChrimson induced activity in olfactory neurons evokes navigational behaviors and that reverse correlation can be used to identify both visual and olfactory computations . We also found that activity in CO2 receptor neurons is interpreted according to different rules than for light or attractive odors . We next asked how the larva integrates visual and olfactory information when making navigational decisions . We carried out reverse-correlation experiments with simultaneous uncorrelated light and attractive odor stimuli , using dim blue light to activate the visual system and intense red light to activate CsChrimson expressed in Or42a receptor neurons . We found the TTA for both signals ( Figure 3A ) , and we applied the resulting filters to our input signals to find xO ( t ) and xL ( t ) , the outputs of the linear odor and light filters , at each point in time . We scaled the filters so that xO ( t ) and xL ( t ) had unit variance in the stimulus ensemble ( Pillow and Simoncelli , 2006 ) . To determine whether the larva's turning decisions result from independent olfactory and visual pathways , we examined the statistics of the turn-triggered ensemble ( Figure 3B ) . In our white noise experiments , xO ( t ) and xL ( t ) are independent Gaussian variables with mean 0 , so it can be shown ( Bialek and van Steveninck , 2005 ) E[xOxL\|turn]∝E[∂2rO−L ( t ) ∂xO∂xL] . 10 . 7554/eLife . 06229 . 006Figure 3 . Multi-modal reverse-correlation experiments suggest attractive odor and light signals are combined linearly and early . Or42a>CsChrimson larvae were presented with independently varying Brownian light intensities . Reverse-correlation analysis was carried out as in Figure 1 . ( A ) TTA . Average change in red ( fictive odor ) and blue light intensities preceding turns . ( B ) Turn triggered ensemble . Top: 2D density histogram of calculated odor and light filter outputs at initiation of each turn . Bottom: 1D density histograms of filter outputs ( xO , xL ) and their linear combinations ( u , v ) . DKL ( P ( x\|turn ) \|\|P ( x ) ) is the Kullback-Leibler divergence from the turn-triggered distribution to the distribution of x at all times . Larger values indicate that x carries more information about the decision to turn . ( C , D ) Predicted turn triggered ensemble according to , ( C ) independent pathways model and ( D ) early linear combination model . Top panel: predicted density . Bottom panel: difference between predicted density and measured density . DKL ( data\|\|model ) is the Kullback-Leibler divergence of the model from the data; smaller values indicate a better match . *** = P ( Independent pathways model ) /P ( Early linear combination model ) < 0 . 00001; Aikake Information Criterion Test . ( E ) Coordinate rotation described in text and used in bottom panel of B . Orthogonal coordinates ( u , v ) are rotated 33° relative to ( xO , xL ) . Rotation is shown overlaid on turn-triggered probability density ( B ) . See Table 1 for number of experiments , animals , and so on . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 00610 . 7554/eLife . 06229 . 007Figure 3—figure supplement 1 . Graphical explanation of the independent pathways model . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 00710 . 7554/eLife . 06229 . 008Figure 3—figure supplement 2 . Graphical explanation of the early linear combination model . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 00810 . 7554/eLife . 06229 . 009Figure 3—figure supplement 3 . Visual and fictive olfactory stimuli do not cross-talk . Larvae were presented with same red and blue Brownian light stimuli as in Figure 3 . TTA of red and blue stimuli are shown on same axes as in Figure 3A . ( A ) Reproduced from Figure 3A: Larvae expressing CsChrimson in Or42a receptor neurons turn in response to increasing blue light and decreasing red light ( fictive odor ) . ( B ) Genetically blinded larvae expressing CsChrimson in Or42a receptor neurons turn in response to decreasing red light ( fictive odor ) but are unresponsive to blue light . ( C ) Wild-type larvae not expressing CsChrimson turn in response to increasing blue light but are unresponsive to red light . See Table 1 for number of experiments , animals , and so on . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 009 In general , this value will be nonzero , but in the independent pathways model ( Figure 3C , Figure 3—figure supplement 1 ) , the mixed partial derivative is identically 0 , soE[xOxL\|turn]=0 ( independent pathways ) . In fact , we found <xOxL\|turn>=0 . 23 , disfavoring the independent pathways hypothesis . If larvae integrate odor and light information when making turning decisions , what form does this integration take ? A potentially favorable ( Ma et al . , 2006; Angelaki et al . , 2009 ) approach would be for the larva to use a simple linear combination of uni-modal filter outputs as the basis for downstream multi-modal processing ( Figure 3—figure supplement 2 ) . In this case , the turn rate would be given byrO−L=h ( cos ( θ ) xO ( t ) +sin ( θ ) xL ( t ) ) ( early linear combination ) . Here , θ is a constant reflecting the relative importance of each filter output to the computation ( θ = 0 would mean larvae respond only to odor and θ = 90° would mean larvae respond only to light ) . We used both the independent pathways and early linear combination models to fit the turn-triggered probability density of ( xO , xL ) and found better agreement between the data and the early linear combination prediction ( Figure 3D ) than the independent pathways prediction ( Figure 3C ) . In the early linear combination model , the turn rate is a function of a 1-dimensional stimulus vector—cos ( θ ) xO ( t ) +sin ( θ ) xL ( t ) —so a rotation of odor-light convolution space ( Figure 3E ) ( u ( t ) v ( t ) ) = ( cos ( θ ) sin ( θ ) −sin ( θ ) cos ( θ ) ) ( xO ( t ) xL ( t ) ) , should produce a single parameter , u , that carries as much information about turn decisions as the pair ( xO , xL ) together , and an orthogonal parameter , v , that carries no information at all . The Kullback-Leibler divergence between the stimulus and turn-triggered distributions describes the amount of information stimulus parameters carry about the decision to initiate a turn ( Pillow and Simoncelli , 2006 ) ; we calculated this divergence for the pair ( xO , xL ) and for xO , xL , u , and v individually ( Figure 3B ) and found that u alone is nearly as informative as both xO and xL and that v contributes very little to the decision to turn , further supporting the early linear combination model . Can the observed summation of visual and odor inputs be explained by blue light activation of the CsChrimson channel ? To probe for cross-talk between visual and olfactory channels , we repeated the multi-sensory experiments ( again presenting both red and blue stimuli simultaneously ) using genetically blinded larvae expressing CsChrimson in Or42a neurons and using wild-type larvae not expressing CsChrimson ( Figure 3—figure supplement 3 ) . Blind larvae responded only to red light ( Figure 3—figure supplement 3B ) while larvae not expressing CsChrimson responded only to blue light ( Figure 3—figure supplement 3C ) . To further compare the independent pathways and early linear combination models , we measured larvae's responses to simultaneous step changes in Or42a receptor neuron activity and blue light ( Figure 4 ) . We presented larvae with all possible combinations of favorable , neutral , and unfavorable steps of red light ( fictive odor ) and blue light ( visual cue ) . We used the kernels calculated from the reverse-correlation experiments and fit the nonlinear functions parameterizing the early linear combination and independent pathways model to the observed turn rates . We found that despite having fewer free fit parameters , the early linear combination model ( magenta line , Figure 4 ) better predicted the response to conflicting and aligned multi-sensory input than the independent pathways model ( cyan line , Figure 4 ) . 10 . 7554/eLife . 06229 . 010Figure 4 . Multi-modal step responses support early linear combination of odor and light signals . Turn rates vs time for Or42a>CsChrimson larvae responding to coordinated increases and decreases of red and blue light . All steps occur at t = 0 . Left column: no change in fictive odor , center column: red light increases at t = 0 right column: red light decreases at t = 0 . Top row: no change in visual stimulus , center row: blue light increases at t = 0 , bottom row: blue light decreases at t = 0 . For instance , in panel iii blue light increased at time 0 , while red light remained constant; in iv , both red and blue light increased at time 0; and in v , blue light increased and red light decreased at time 0 . Black line and shaded region represent mean turn rate and standard error due to counting statistics . Cyan line is the best-fit ( maximum likelihood estimate , 6 parameter fit ) prediction of the independent pathways model . Magenta line is the best-fit ( maximum likelihood estimate , 4 parameter fit ) prediction of the early linear combination model . Note that the time axis is the same for each subplot , but the turn rate axis varies . *** = P ( Independent pathways model ) /P ( Early linear combination model ) < 0 . 00001; Aikake Information Criterion Test , measured for the entire data set . See Table 1 for number of experiments , animals , and so on . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 010 For changes in only light or only odor , larvae increase their turn rates significantly more for unfavorable changes ( panels ii and iii ) than they decrease their turning in response to favorable changes ( i and vi ) . Thus the independent pathways model , which sums separate rates for light and odor changes , predicts that in response to conflicting favorable and unfavorable changes ( iv and viii ) , the larvae will still significantly increase their turning . Instead , conflicting changes in odor and light cancel each other out , as predicted by the early linear combination model . That larvae turn in response to increases in blue light ( iii , v ) but decreases in red light ( ii , v ) further confirms that the two light sources are activating different sensory pathways . For instance , if blue light were primarily activating CsChrimson , we would expect that decreasing blue and red light together ( viii ) would provoke more turning than decreasing red light alone ( ii ) , but in fact increasing blue light while decreasing red light ( v ) provokes the largest increase in turning . We wondered whether the larva might use the same linear combination of light and odor signals to make other navigational decisions , like whether to accept or reject head-sweeps . Although we previously determined a maximally informative combination of filtered odor and light inputs , the convolution kernels have similar shapes , so we might reasonably use the same rules to combine the raw input stimuli . ( μ ( t ) ν ( t ) ) = ( cos ( θ ) sin ( θ ) −sin ( θ ) cos ( θ ) ) ( O ( t ) L ( t ) ) ;O ( t ) =−dIreddt/Ir0;L ( t ) =dIbluedt/Ib0 . Here , Ir0 and Ib0 are the normalization factors required to make the convolution kernels have unit variance , and θ was determined as a parameter of the early linear combination model ( Figure 3D ) . We carried out reverse-correlation analysis on these rotated coordinates . We found: the turn triggered average of ν ( t ) was nearly 0 ( Figure 5A ) ; the size-sorted TTA of ν ( t ) was nearly 0 and the same for both small and large turns ( Figure 5B ) ; and the head-sweep triggered average of ν ( t ) was nearly 0 and the same for accepted and rejected head-sweeps ( Figure 5C ) . Thus , larvae used a single linear combination of odor and light—μ ( t ) —to determine whether to turn , how large of a turn to make , and whether to reject or accept head-sweeps , strongly suggesting odor and light inputs are combined at early stages of the navigational circuitry . 10 . 7554/eLife . 06229 . 011Figure 5 . All navigational decisions appear to be based on a single linear combination of odor and light inputs . ( A–C ) Reverse correlation in rotated coordinate system . μ , ν are linear combinations of the raw input stimuli according to the same scaling as used to combine filtered odor and light signals in Figure 3 . ( A ) Turn-triggered averages . Average change in μ , ν prior to start of a turn . ( B ) Size-sorted turn-triggered averages . Displayed averages are lowpassed with a Gaussian filter ( σ = 0 . 5 s ) to clarify the long time-scale features . ( C ) Head-sweep-triggered average ( for first head-sweep of turn ) . Shaded region indicates mean head-sweep duration ( 1 . 25 s ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 011 Finally , we asked whether the larva might shift its attention between stimulus inputs . For instance , if a larva initiates a turn due to an increase in light , might it prioritize light changes over odor changes in deciding whether or not to accept the turn's first head-sweep ? To test this , we re-examined the head-sweep acceptances and rejections from the multi-stimulus white noise experiments , sorting them based on the favorability of light and odor changes preceding the turn start ( Figure 6A ) . For instance , in quadrant II , larvae turned following a recent favorable odor change and unfavorable light change . If an unfavorable light change caused otherwise neutral larvae to attend more strongly to light , we would expect that following a quadrant II turn , larvae would be more attentive to light changes than average . Alternately , if some larvae in the population were already attending to light more strongly than odor , we would expect them to be over-represented in the group of larvae turning in response to unfavorable changes in light coupled with favorable changes in odor . Thus , we would again expect that following a quadrant II turn , larvae would be attending more strongly to light than in the average population . Similarly , following quadrant IV turns , we would expect larvae to be attending more strongly to odor . 10 . 7554/eLife . 06229 . 012Figure 6 . Probing for attentional shifts during multi-modal noise experiments . ( A ) Turn-triggered ensemble ( duplicated from Figure 3B ) , with quadrants highlighted . Color scale the same as in 3B . Quadrants I–IV indicate which stimulus or stimuli likely provoked the larva to turn ( I—both odor and light stimulated turning; II—light , but not odor , stimulated turning; III—neither odor nor light were changing unfavorably; IV—odor , but not light , stimulated turning ) . Each turn was assigned to one of these quadrants based on the filtered signal values ( xO , xL ) at the time the turn started . ( B , C ) Difference in intensity changes during accepted and rejected head sweeps ( first head-sweep of turn only ) . Difference in mean rate of change in intensity over mean head sweep duration ( 1 . 25 s , shaded region in 5C ) between rejected and accepted head-sweeps . Error bars are ±1 s . e . m . ( B ) Changes in odor and light intensities . ( C ) Changes in rotated coordinate system . See Table 1 for number of experiments , animals , and so on . DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 012 To measure the importance of light and odor changes during head sweeps , we subtracted the average derivative of light level during accepted head sweeps from the average derivative of light level during rejected head sweeps ( Figure 6B ) . Regardless of which stimulus likely triggered the decision to turn , for both light and odor , we found the same differences between accepted and rejected head-sweeps . Thus , we found no evidence that the larva modulated the relative importance of light or odor changes when deciding whether to accept or reject a head-sweep . Indeed , for all combinations of light and odor changes preceding a turn , the same linear combination of odor and light inputs appeared to be equally salient in deciding whether to accept or reject a head-sweep ( Figure 6C ) . Here , we demonstrated the power of reverse-correlation analysis of larvae's behavioral responses to white-noise visual and fictive olfactory stimuli to decode the computations underlying the Drosophila larva's navigation of natural environments . We showed that this analysis could be used to decode the rules by which the larva integrates signals from distinct sensory organs . Larvae appear to use a single linear combination of odor and light inputs to make all navigational decisions , suggesting these signals are combined at early stages of the navigational circuitry . In this work , we used optogenetics to explore how perturbations in the activities of identified neurons are interpreted behaviorally . We expressed CsChrimson in specific neurons to relate patterns of activity in these neurons to decisions regulating the frequency , size , and direction of turns ( Figure 2 ) . Using model parameters extracted from reverse-correlation experiments , we were able to predict how larvae would respond to novel perturbations of these neurons' activities ( Figure 2C ) . We explored how activity in one particular neuron type modulated the larva's responses to a natural light stimulus ( Figures 3 , 5 , 6 ) and predicted how the larva's natural response to blue light steps ( Figure 4-iii ) would be altered by simultaneous perturbation of this neuron ( Figure 4-iv , v ) . Here we addressed sensory neurons , but our approach can be used generally to identify computations carried out on activities of interneurons , to determine whether activity in a neuron is interpreted as attractive or aversive , to measure how that activity combines with other sources of information to produce decisions , and to find neurons most responsible for making navigational decisions ( Koulakov et al . , 2005 ) . The following strains were used: Canton-S and Berlin wild type ( gift of Justin Blau ) , w1118;;20XUAS-CsChrimson-mVenus ( Bloomington Stock #55136 , gift of Vivek Jayaraman and Julie Simpson , Janelia Research Campus ) , w;;Gr21a-Gal4 ( Bloomington stock #23890 ) , w;;Or42a-Gal4 ( Bloomington stock #9969 ) , w*;;Or42b-Gal4 ( Bloomington stock #9972 ) , GMR-hid/CyO , P{sevRas1 . V12} ( Bloomington stock #5771 ) . 40 virgin female UAS-CsChrimson flies were crossed with 20 males of the selected Gal4 line . F1 progeny of both sexes were used for experiments . Flies were placed in 60 mm embryo-collection cages ( 59–100 , Genesee Scientific , San Diego , CA ) and allowed to lay eggs for 3 hr at 25°C on enriched food media ( ‘Nutri-Fly German Food’ , Genesee Scientific ) . For all experiments except the Berlin response to blue light ( Figure 2 , top row and Figure 2—figure supplement 1 , top row ) , the food was supplemented with 0 . 1 mM all-trans-retinal ( ATR , R2500 , Sigma Aldrich , St . Louis , MO ) , and cages were kept in the dark during egg laying . When eggs were not being collected for experiments , flies were kept either on plain food or agar ( neither containing ATR ) . Petri dishes containing eggs and larvae were kept at 25°C ( ATR+ plates were wrapped in foil ) for 48–60 hr . Second instar larvae were separated from the food using 30% sucrose solution and washed in deionized water . Larval stage was verified by size and spiracle morphology . Preparations for experiments were carried out in a dark room , under dim red ( for photo-taxis experiments ) or blue ( for CsChrimson experiments ) illumination . Prior to beginning experiments , larvae were dark adapted on a clean 2 . 5% agar surface for a minimum of 10 min . Approximately 30–50 larvae were transferred with a wet paintbrush to a 23 cm square dish ( Corning BioAssay Dish #431111 , Fisher Scientific , Pittsburgh , PA ) , containing 2 . 5% ( wt/vol ) bacteriological grade agar ( Apex , cat#20-274 , Genesee Scientific ) and 0 . 75% ( wt/vol ) activated charcoal ( DARCO G-60 , Fisher Scientific ) . The charcoal darkened the agar and improved contrast in our dark-field imaging setup . The plate was placed in a darkened enclosure and larvae were observed under strobed 850 nm infrared illumination ( ODL-300-850 , Smart Vision Lights , Muskegon , MI ) using a 14 fps 5 MP rolling shutter CMOS camera ( Basler acA2500-14gm , Graftek Imaging , Austin , TX ) in global-reset-release mode and an 18 mm c-mount lens ( 54-857 , Edmund Optics , Barrington , NJ ) equipped with an IR-pass filter ( Hoya R-72 , Edmund Optics ) . The experiments of Figure 3—figure supplement 3B were recorded using a 4 MP global shutter CMOS camera ( Basler acA2040-90umNIR , Graftek Imaging ) operating at 20 fps and a 35 mm focal length lens ( Fujinon CF35HA-1 , B&H Photo , New York , NY ) . A microcontroller ( Teensy++ 2 . 0 , PJRC , Sherwood , OR ) coordinated the infrared strobe and control of the stimulus light source , so stimulus presentation and images could be aligned to within the width of the strobe window ( 2–5 ms ) . Videos were recorded using custom software written in LABVIEW and analyzed using the MAGAT analyzer software package ( Gershow et al . , 2012 ) . Further analysis was carried out using custom MATLAB scripts . Table 1 gives the number of experiments , animals , turns , and head sweeps analyzed for each experimental condition . Software is available at https://github . com/GershowLab . 10 . 7554/eLife . 06229 . 013Table 1 . Numbers of experiments , animals , turns , and head sweeps for all figuresDOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 013Genotype#expts#animalshours#turnsrms turn size#large turns#small turns#accepted head sweeps#rejected head sweepsUni-modal reverse-correlation experiments ( Figure 2A , B , D , E , Figure 2—figure supplement 1 ) Berlin615052 . 6659476 . 42462413241392455 Canton-S7334117 . 7882466 . 03086573855703254 Or42a>CsChrimson518061 . 1697168 . 92531444041222849 Or42b>CsChrimson624664 . 4956573 . 33480608562153350 Gr21a>CsChrimson522754 . 2876075 . 13392536854243336Uni-modal step experiments ( Figure 2C ) Berlin49534 . 73674––––– Or42a>CsChrimson210736 . 63905––––– Or42b>CsChrimson29921 . 52599––––– Gr21a>CsChrimson211122 . 12384–––––Multi-modal reverse-correlation experiments ( Figure 3 , 5 , 6 , Figure 3—figure supplement 3 ) Or42a>CsChrimson1260813621 , 07566 . 6722513 , 85012 , 7958280 quadrant I–––10 , 363–––62164147 quadrant II–––3301–––20861215 quadrant III–––1684–––1088596 quadrant IV–––5727–––34052322 GMR-Hid , Or42a>CsChrimson312128 . 94842––––– Canton-S316639 . 93412–––––Multi-modal step experiments ( Figure 4 ) Or42a>CsChrimson5250507859–––––#expts: Number of 20 min experiments . For reverse-correlation experiments , each experiment presented a different stimulus sequence with the same statistical properties; for step experiments , the same stimulus pattern was presented in each experiment . #animals: Approximate number of animals , taken by finding the maximum number of animals tracked in a 30-s window during each experiment . #hours: total observation time in units of larva-hours . Observing 3 larvae for 20 min each would equal 1 larva-hour . #turns: total number of turns observed and used in analysis . rms turn size: root mean square turn size in degrees ( defined as angular difference in run heading immediately before and after a turn ) for the set of experiments . #large/small turns: number of turns with angular changes larger/smaller than the rms turn size . #accepted head sweeps: number of times the first head sweep of a turn was accepted , ending in a new run . #rejected head sweeps: number of times the first head sweep of a turn was rejected , leading to another head sweep . 10 . 7554/eLife . 06229 . 014Table 2 . Kullback-Leibler divergences for Figure 3DOI: http://dx . doi . org/10 . 7554/eLife . 06229 . 014KL divergencek-NNmodel data as normally distributedSzegö-PSD methodFigure 3B: DKL ( ( P ( xo , xl\|turn ) \|\|P ( xo , xL ) ) 0 . 3510 . 325–Figure 3B: DKL ( ( P ( xo\|turn ) \|\|P ( xo ) ) 0 . 2360 . 2230 . 235Figure 3B: DKL ( ( P ( xL\|turn ) \|\|P ( xL ) ) 0 . 1030 . 1000 . 104Figure 3B: DKL ( ( P ( u\|turn ) \|\|P ( u ) ) 0 . 3340 . 3240 . 333Figure 3B: DKL ( ( P ( v\|turn ) \|\|P ( v ) ) 0 . 0060 . 00020 . 003Figure 3C: DKL ( data\|\|model ) 0 . 0620 . 035–Figure 3D: DKL ( data\|\|model ) 0 . 0300 . 007–KL divergence: the divergence to be calculated . k-NN: divergence calculated using the k-nearest neighbors algorithm . This value is displayed in Figure 3 . model data as normal distributed: the distributions are modeled as Gaussians , whose divergence is calculated analytically . Szegö-PSD method: divergence between 1D distributions calculated by an alternate method . We built a custom circuit board ( Advanced Circuits , Colorado ) containing 66 deep red high brightness LEDs ( Philips Lumileds , LXM3-PD01 , 655 nm central wavelength ) and 12 royal blue high brightness LEDs ( LXML-PR01-0500 , 447 . 5 nm central wavelength ) evenly distributed over ∼25 cm × 25 cm . The LEDs were driven at constant current by a switch-mode LED driver circuit ( based on LT3518 , linear technology ) operating at a switching frequency of 2 MHz . The on-current was set by interchangeable feedback resistors and could be modulated separately for red and blue LEDs . The intensity of the red and blue LEDs was controlled separately by pulse-width-modulation . Illumination was provided from above the larvae; the LED circuit board was at the same height as the recording camera ( ∼50 cm above the behavioral arena ) . For multi-sensory experiments , the maximum red light intensity ( 911 μW/cm2 ) was 300 times greater than the maximum blue light intensity ( 3 μW/cm2 ) . CsChrimson is slightly more sensitive to 655 nm than 448 nm light , so the blue light signal perturbed olfactory receptor neuron activity by less than 0 . 3% of the red-light signal's perturbation . We calibrated the optical power of the LEDs using a photodiode power meter ( S121C , Thorlabs , New Jersey ) set to the central wavelength of the LEDs . We measured the uniformity of the stimulus light sources by imaging a Lambertian projector screen ( Dalite 41466 , Cousin's Video , Ohio ) placed in the plane of the experimental arena under stimulus LED illumination . Stimulus protocols were generated with MATLAB and stored on an SD card for use by the microcontroller . Light intensity was modulated using pulse-width-modulation with a frequency of ∼112 Hz ( constrained to update exactly 8 eight times per camera frame ) . We chose a Brownian random walk , whose derivatives on all time scales are independent identically distributed Gaussian variables , to analyze the larva's response to derivatives of stimulus intensity . Light levels were specified by values between 0 ( off ) and 255 ( maximum intensity ) . Sequences of light levels corresponding to a random walk with reflecting boundary conditions were generated according to these rules:I0=127 , Ij=−Ij if Ij<0 , Ij=510−Ij if Ij>255 , Ij+1=Ij+N ( 0 , σ ) , where N ( 0 , σ ) was a Gaussian random variable with mean 0 and variance σ2 . For the experiments described in this work σ = 3 . At an update rate of 112 Hz , this represents a diffusion constant of 504 ( light levels ) 2/s . After the sequence of light levels was generated , the levels were rounded to the nearest integer value before being transferred to the microcontroller for use in experiments . Sequences were not reused within an experimental group but might be reused between groups . For multi-modal experiments , independent sequences were used for each stimulus . For step response experiments of Figure 2C , a square wave with a period of 20 s and duty cycle 50% ( 10 s high , 10 low ) was presented . The low and high intensities were symmetrically distributed about the mean light intensity of the reverse-correlation experiments . For the coordinated step response experiments of Figure 4 , we presented steps of red and blue light intensity . Every 10 s , each signal either increased from low to high , decreased from high to low , or remained constant . The sequence of steps was chosen so that all combinations ( except for both levels remaining constant ) were presented . For each stimulus , the low and high intensities were symmetrically distributed about the mean light intensity of the reverse-correlation experiments . As described previously ( Gershow et al . , 2012; Kane et al . , 2013 ) , videos of behaving larvae were recorded using LabView software into a compressed image format ( mmf ) that discards the stationary background . These videos were processed using computer vision software ( written in C++ using the openCV library ) to find the position and posture ( head , tail , midpoint , and midline ) of each larva and to assemble these into tracks , each following the movement of a single larva through time . These tracks were analyzed by Matlab software to identify behaviors , especially runs , turns , and head sweeps . The sequence of light intensities presented to the larvae was stored with the video recordings and used for reverse-correlation analysis . In Figures 3 , 4 , we fit the nonlinear turn rate to the observed data using the ratio-of-Gaussians function with r¯ , σ , and μ as fit parameters . The probability of observing at least one turn in an interval ∆t given an underlying turn rate r is 1 − e−r∆t; in the limit of short ∆t , this reduces to r∆t . The probability of not observing a turn is e−r∆t . Therefore given a model of the turn rate , the probability of observing a particular experimental outcome is given bylog ( P ( data\|model ) ) =∑turnlog ( r ( x ) Δt ) −∑no turnr ( x ) Δt , where x is the filtered signal , r ( x ) is the turn rate predicted by the model , and ∆t is the sampling rate . ∑no turn is the sum over all points when larvae were in runs and thus capable of initiating turns . We used the MATLAB function fmincon to find the parameters that maximized this log-likelihood . For Figures 3C , 4 ( cyan line ) , the separate pathways model rate function was given byr ( x ) =rO¯e− ( xO−μO ) 22σO2σOe−x22+rL¯e− ( xL−μL ) 22σL2σLe−x22 , with fit parameters rO¯ , σO , μO , rL¯ , σL , and μL . For Figures 3D , 4 ( magenta line ) , the early linear combination model rate function was given byr ( x ) =r¯e− ( u−μ ) 22σ2σe−u22;u=cos ( θ ) xO+sin ( θ ) xL , with fit parameters r¯ , σ , μ , and θ . We estimated the KL divergences in Figure 3 using the k-nearest neighbors algorithm ( Wang et al . , 2009 ) as implemented in MATLAB by ( Szabó , 2014 ) . As a check , for Figure 3B , we also analytically computed the KL divergence between Gaussian distributions with the same means and ( co- ) variances as the sampled turn-triggered and stimulus ensembles ( Pillow and Simoncelli , 2006 ) , and for Figure 3C , D , we numerically calculated the KL divergence between the model predictions and a multivariate Gaussian with the same mean and covariance as the measured turn-triggered density . For the univariate distributions of 3B , we also estimated the divergence using an alternate method based on Szegö's theorem ( Ramırez et al . , 2009 ) as implemented by ( Szabó , 2014 ) . These estimates are shown in Table 2 . None of the conclusions of this paper depend on the method of estimation used .	Living organisms can sense cues from their surroundings and respond in appropriate ways . For example , animals will often move towards the smell of food or away from potential threats , such as predators . However , it is not fully understood how an animal's nervous system is set up to allow sensory information to control how the animal navigates its environment . It is also not clear how animals ‘decide’ what to do when they receive conflicting information from different senses . Optogenetics is a technique that allows neuroscientists to control the activities of individual nerve cells simply by shining light on to them . Fruit fly larvae have a simple but well-studied nervous system , and they are nearly transparent , so scientists can use optogenetics to activate nerve cells in freely moving larvae . Fruit fly larvae move in a series of forward ‘runs’ and direction-changing ‘turns’ and use sensory cues to decide when to turn , how large of a turn to make , and whether to turn left or right . Gepner , Mihovilovic Skanata et al . used optogenetics to stimulate different combinations of sensory nerve cells in larvae , while tracking the larvae's movements to discover exactly what information they used to make these decisions . An independent study by Hernandez-Nunez et al . also used a similar approach . Fruit fly larvae are attracted towards scents from rotting fruit and are repelled by light—in particular , larvae are most sensitive to blue light but cannot detect red light . Therefore , Gepner , Mihovilovic Skanata et al . could expose the larvae to blue light to activate light-sensing nerve cells as normal , and use red light to activate odor-sensing nerve cells via optogenetics . These experiments showed that larvae changed direction more often when the level of blue light was increased or when the level of red light ( which simulated the detection of odors from rotting fruits ) was decreased . Analysis of the data from these experiments revealed that larvae essentially assign negative values to the blue light and positive values to the ‘odor-mimicking’ red light . The larvae then use the sum of these two values to dictate their next move . This suggests that navigation in response to both light and odors is supported by the same pathways in a larva's nervous system . The approach of using optogenetics in combination with quantitative analysis , as used in these two independent studies , is now opening the door to a more complete understanding of the connections between the activities of sensory nerve cells and perception and behavior .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Computations underlying Drosophila photo-taxis, odor-taxis, and multi-sensory integration
Despite their fundamental importance for body size regulation , the mechanisms that stop growth are poorly understood . In Drosophila melanogaster , growth ceases in response to a peak of the molting hormone ecdysone that coincides with a nutrition-dependent checkpoint , critical weight . Previous studies indicate that insulin/insulin-like growth factor signaling ( IIS ) /Target of Rapamycin ( TOR ) signaling in the prothoracic glands ( PGs ) regulates ecdysone biosynthesis and critical weight . Here we elucidate a mechanism through which this occurs . We show that Forkhead Box class O ( FoxO ) , a negative regulator of IIS/TOR , directly interacts with Ultraspiracle ( Usp ) , part of the ecdysone receptor . While overexpressing FoxO in the PGs delays ecdysone biosynthesis and critical weight , disrupting FoxO–Usp binding reduces these delays . Further , feeding ecdysone to larvae eliminates the effects of critical weight . Thus , nutrition controls ecdysone biosynthesis partially via FoxO–Usp prior to critical weight , ensuring that growth only stops once larvae have achieved a target nutritional status . Environmental conditions mould the developmental programs of many organisms to produce dramatic differences in body size and shape , in developmental time and in pigmentation patterns ( Beldade et al . , 2011 ) . In insects , environmental cues often mediate their effects by regulating the timing and amount of hormone biosynthesis at specific points in development ( Koyama et al . , 2013 ) . These changes in hormone production have been associated with a wide variety of environmentally induced changes in morphology , including the dramatic reshaping of the body in honeybee castes and seasonal wing pattern polyphenisms in butterflies ( Beldade et al . , 2011; Koyama et al . , 2013 ) . Understanding the molecular underpinnings through which environmental conditions modify hormone production would provide valuable insight into our understanding of developmental plasticity . Larvae of the fruit fly , Drosophila melanogaster , provide a tractable model to address this question . Drosophila larvae regulate their body size and developmental timing in response to nutritional conditions , similar to many other animals ( Shingleton , 2011 ) . Early in the third ( final ) larval instar ( L3 ) , a small peak of the steroid hormone ecdysone has been proposed to induce a developmental transition known as critical weight ( Mirth and Riddiford , 2007; Mirth and Shingleton , 2012 ) . The critical weight ecdysone peak responds to both environmental cues and internal developmental processes . Environmental cues including nutrition , temperature and oxygen levels affect the timing of the critical weight ecdysone peak ( Caldwell et al . , 2005; Colombani et al . , 2005; Mirth et al . , 2005; Callier et al . , 2013; Ghosh et al . , 2013 ) . In addition , the neuropeptide important for inducing all ecdysone peaks , prothoracicotropic hormone ( PTTH ) , stimulates ecdysone biosynthesis at critical weight ( McBrayer et al . , 2007; Ou et al . , 2011 ) . The combination of environmental and developmental regulation of this ecdysone peak ensures that developmental timing can be altered with changes in environmental conditions ( Gibbens et al . , 2011; Mirth and Shingleton , 2012 ) . Critical weight itself determines the duration of the growth period , and therefore final body size , in response to environmental conditions including nutrition ( Beadle et al . , 1938; Nijhout and Williams , 1974b; Mirth et al . , 2005; Shingleton et al . , 2005; Stieper et al . , 2008 ) . Before larvae reach critical weight , starvation delays the onset of metamorphosis . After critical weight , larvae initiate metamorphosis without any developmental delay even when starved . Feeding ecdysone to larvae with genetically-induced delays in critical weight rescues the timing of the onset of metamorphosis ( Stieper et al . , 2008; Parker and Shingleton , 2011 ) . Nutrition regulates size in organisms ranging from flies to humans via the insulin/insulin-like growth factor signaling ( IIS ) /Target of Rapamycin ( TOR ) signaling pathway ( Grewal , 2009 ) . The IIS/TOR pathway controls body size by regulating growth rate , and also by regulating the timing of critical weight to determine the duration of the growth period ( Nijhout , 2003 ) . At critical weight , the IIS/TOR pathway acts directly on the glands that synthesize ecdysone , the prothoracic glands ( PGs ) , to alter the timing of the expression of several cytochrome P450 ( CYP450 ) genes necessary for ecdysone biosynthesis ( Colombani et al . , 2005; Layalle et al . , 2008 ) . Increasing IIS/TOR activity in the PGs causes precocious ecdysone biosynthesis , precocious critical weight transitions , precocious metamorphosis and dramatic reductions in body size ( Caldwell et al . , 2005; Colombani et al . , 2005; Mirth et al . , 2005; Layalle et al . , 2008 ) . Reducing IIS/TOR activity in the PGs induces the opposite effects . However , the mechanisms through which the IIS/TOR pathway mediates these effects have been unclear . Under well-fed conditions , insulin-like peptides ( ILPs ) are secreted into the insect blood or hemolymph ( Masumura et al . , 2000; Ikeya et al . , 2002; Rulifson et al . , 2002; Geminard et al . , 2009 ) . By binding to the Insulin Receptor ( InR ) , ILPs activate IIS/TOR signaling in the target tissues ( Brogiolo et al . , 2001; Ikeya et al . , 2002 ) . Activating IIS/TOR signaling regulates a series of phosphokinases , including Akt . Akt , in turn , phosphorylates a negative regulator of growth , Forkhead Box class O transcription factor ( FoxO ) , displacing it from the nucleus to the cytoplasm ( Junger et al . , 2003 ) . In starved larvae , FoxO localizes in the nucleus where it acts on its targets , such as 4E-binding protein ( 4E-BP , also known as Thor ) , to suppress cell growth and division ( Puig et al . , 2003 ) . In mammalian cells , FoxO binds to several nuclear hormone receptors ( NHRs ) , such as constitutive androstane receptor ( CAR ) and pregnane X receptor ( PXR ) , to regulate CYP450 expression ( Kodama et al . , 2004 ) . The functional ecdysone receptor is composed of two NHRs , Ecdysone Receptor ( EcR ) and Ultraspiracle ( Usp ) . Since many CYP450 enzymes are involved in ecdysone biosynthesis ( Gilbert et al . , 2002 ) , this led us to the hypothesis that the effect of IIS/TOR signaling on ecdysone biosynthesis is mediated by the interaction between FoxO and either EcR or Usp . Here , we provide definitive evidence that critical weight results from the small nutrition-sensitive ecdysone peak early in the L3 . Further , we report that IIS/TOR regulates the timing of ecdysone biosynthesis at critical weight via a novel mechanism , the direct association of FoxO and Usp . With these findings , we have constructed a detailed model of the molecular mechanisms underlying environmentally-sensitive ecdysone biosynthesis during critical weight , an event that ultimately determines the duration of the growth period and accordingly final body size . Previous studies have shown that activating IIS/TOR signaling in the PGs induces early critical weight transitions , precocious ecdysone biosynthesis at wandering , and precocious metamorphosis ( Caldwell et al . , 2005; Colombani et al . , 2005; Mirth et al . , 2005; Layalle et al . , 2008 ) . This has led authors to propose that the small pulse of ecdysone early in the L3 ( Warren et al . , 2006 ) is nutrition-sensitive and induces critical weight in Drosophila ( Mirth and Riddiford , 2007; Koyama et al . , 2013; Mirth et al . , 2014 ) . However , these studies have not measured ecdysone concentrations with sufficient resolution early in the instar to show that ecdysone biosynthesis was delayed in starved pre-critical weight larvae . Therefore , we first examined whether this early ecdysone peak is delayed in starved larvae . In accordance with our hypothesis , we found that the small ecdysone peak that occurs around 10 hr after L3 ecdysis ( AL3E ) in well-fed larvae is suppressed in starved larvae , at least until 18 hr AL3E ( Figure 1A ) . Thus , the timing of this early peak is indeed sensitive to nutrition . 10 . 7554/eLife . 03091 . 003Figure 1 . Nutrition regulates the timing of the critical weight ecdysone peak and exogenous ecdysone eliminates developmental delays in pre-critical weight larvae . ( A ) Nutrition is necessary to induce a small ecdysone peak at the early L3 . We used 30–38 w[1118] larvae for each sample and three biologically independent samples for each time point . Each point indicates the mean ecdysone concentration ± SEM . Points sharing the same letter indicate the mean concentration at the time ±2 hr are statistically indistinguishable from one another; points that differ in letters are significantly different ( p < 0 . 05 ) . The arrowhead along the x axes indicates the age at which w[1118] larvae reached critical weight from Figure 1B . ( B ) Exogenous ecdysone administration throughout the L3 eliminates developmental delay in starved , pre-critical weight w[1118] larvae . The larvae were continuously fed a fly medium containing 0 . 15 mg/g 20E or transferred at given time points on to a starvation medium ( 1% agar ) containing the same concentration of 20E . Inset shows the weight ±95% confidence intervals at which larvae reach critical weight . The age and size at which larvae reach critical weight was determined using breakpoint analysis and means and ±95% confidence intervals were calculated from 1000 bootstrap datasets . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 00310 . 7554/eLife . 03091 . 004Figure 1—figure supplement 1 . Ecdysone administration reduced body size . ( A ) Feeding larvae with 20E-supplemented fly medium reduces body size in w[1118] animals . The numbers indicate p-values by ANOVA and pairwise t tests . ( B ) Continuously fed w[1118] larvae show linear growth curve during their feeding period . Each point indicates the mean weight ± S . D . N = 12–16 . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 004 Next we reasoned that if this early peak of ecdysone induced critical weight , feeding ecdysone to starved , pre-critical weight larvae should eliminate the delay in their development . To determine when wild type larvae reach critical weight , we starved carefully staged larvae of defined age classes on non-nutritive agar and measured the time it takes for them to reach pupariation from the onset of starvation . A hallmark of critical weight is that before it is attained starvation delays the onset of metamorphosis ( Beadle et al . , 1938; Nijhout and Williams , 1974b; Mirth et al . , 2005; Shingleton et al . , 2005; Stieper et al . , 2008 ) , whereas after critical weight larvae metamorphose early when starved . We estimate the age at critical weight using breakpoint analysis , which fits a bi-segmental linear regression to the relationship between age at starvation and time to pupariation , and calculates the age at critical weight as the inflection point where this relationship changes ( Stieper et al . , 2008; Ghosh et al . , 2013; Testa et al . , 2013 ) . We then use the linear relationship between larval weight and larval age to convert the age at which larvae reach critical weight to the size at which larvae reach critical weight ( Figure 1—figure supplement 1B ) . Finally , we repeated the analysis on 1000 bootstrap datasets to generate 95% confidence intervals for the age and size of larvae when they reach critical weight . Data and scripts for the analysis of size and age at critical weight , including the growth rate data , for all genotypes and treatments are available from the Dryad Digital Repository: http://dx . doi . org/10 . 5061/dryad . 75940 ( Koyama et al . , 2014 ) . Wild type larvae reached critical weight at 8 . 66 hr AL3E ( Figure 1B , Supplementary file 1 ) , correlating with the time when well-fed , wild type larvae show a peak of ecdysone ( Figure 1A ) . When we added the active form of ecdysone , 20-hydroxyecdysone ( 20E ) , to the medium even the youngest larvae no longer delayed their onset of metamorphosis when starved ( Figure 1B ) . Instead , larvae starved on 20E-supplemented agar between the ages 0–8 hr AL3E pupariated 32 hr after the onset of starvation ( Supplementary file 1 ) . Finally , larvae fed 20E-supplemented fly medium throughout the L3 were more than 25% smaller than control larvae ( Figure 1—figure supplement 1A ) . These results demonstrate that this early peak of ecdysone is nutrition sensitive and that it induces critical weight . We next sought to understand how nutrition regulated the timing of the critical weight ecdysone peak . We hypothesized that IIS/TOR signaling controlled the timing of this ecdysone peak , and therefore critical weight , via FoxO . We reasoned that if FoxO was involved in regulating ecdysone biosynthesis , FoxO would be present in the PG nuclei immediately after the molt to the L3 and would become progressively excluded from the nucleus as the larvae fed and approached critical weight . We found that FoxO was localized primarily in the nuclei of the PG cells of newly ecdysed L3 larvae ( 0 hr AL3E ) ( Figure 2A ) . As the larvae fed , FoxO was gradually transported out of the nuclei into the cytoplasm . At 5 hr AL3E , FoxO appeared evenly dispersed inside the PG cells ( Figure 2B ) . By 10 hr AL3E , immediately after critical weight ( Figure 1B ) , FoxO was mostly localized in cytoplasm of fed larvae ( Figure 2C , D ) . Thus , FoxO appears to be progressively transported out of the nucleus as larvae approached critical weight . 10 . 7554/eLife . 03091 . 005Figure 2 . FoxO co-localizes with Usp in the PGs of pre-critical weight larvae and FoxO binds to Usp . ( A–E ) FoxO progressively moved out of the nuclei and into the cytoplasm of the PG cells in response to nutrition . PGs from w[1118] larvae at the onset of the L3 ( A ) , fed for 5 ( B ) , 10 ( C ) and 15 hr ( D ) or starved for 15 hr ( E ) were immunostained for FoxO , Usp and phalloidin . The scale bar is 10 µm . ( F ) GST-pulldown shows that FoxO binds to Usp but not to EcR . ( G ) FoxO associates with Usp before larvae reach critical weight but does not affect EcR–Usp association . Newly molted w[1118] larvae ( 0–5 hr AL3E ) were either protein-starved ( St ) on 20% sucrose solution or fed on a standard fly medium ( Fed ) for additional 24 hr , and then the anterior halves of larvae without the fat body and salivary glands were used for protein extraction . We also examined pre-critical weight FoxO mutant ( FoxO Δ94/Df ( 3R ) Exel8159 ) larvae ( 0–5 hr ) as a negative control . Precipitation was performed using the anti-Usp antibody . ( H ) Usp but not EcR associates with FoxO in co-immunoprecipitation assays using anti-Usp and anti-EcR antibodies . No AB indicates the no-antibody control . Protein extracts were prepared as in ( G ) . ( I ) Presence of 20E neither changes FoxO–Usp binding properties nor induces FoxO–EcR association in a GST-pulldown assay . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 005 For FoxO to regulate ecdysone biosynthesis in a nutrition-dependent manner , we would expect that it would remain in the nucleus in starved , pre-critical weight larvae . In larvae starved from 0–15 hr AL3E , FoxO remained in the nuclei of the PG cells ( Figure 2E ) . In contrast , FoxO was found primarily in the cytoplasm in fed controls . Taken together , the localization of FoxO suggests that it could be involved in regulating ecdysone biosynthesis at critical weight . Since FoxO associates with NHRs to regulate CYP450 gene expression in mammalian cells , we hypothesized that FoxO could associate with either EcR or Usp to regulate the nutrition-sensitive ecdysone peak by regulating the expression of CYP450 ecdysone biosynthesis genes . Using GST-pulldown assays , we found that FoxO bound to Usp but not to EcR in vitro ( Figure 2F ) . Co-immunoprecipitation experiments using larval extracts showed that FoxO bound to Usp only in pre-critical weight or starved larvae , but not in well-fed post-critical weight larvae ( Figure 2G , H ) . FoxO neither bound to EcR , nor did it impede EcR/Usp binding in starved larvae ( Figure 2G , H ) . This suggests that FoxO could interact with Usp to regulate the critical weight ecdysone peak , and further that this interaction is unlikely to interfere with EcR/Usp function . Because in vertebrates FoxO-NHR interactions sometimes change in the presence of hormones ( Schuur et al . , 2001; Li et al . , 2003; Kodama et al . , 2004 ) , we tested whether 20E altered FoxO/Usp binding or induced FoxO/EcR binding . The presence of 20E neither changed the binding properties of the FoxO–Usp interaction nor induced a FoxO–EcR association ( Figure 2I ) . If FoxO/Usp interactions regulate ecdysone biosynthesis at critical weight , we would expect that altering the expression of FoxO or Usp in the PGs would change both the size and the age at which larvae reach critical weight . We used Phantom ( Phm ) -Gal4 , a Gal4 driver specific for the PG cells , to overexpress FoxO in the PGs . These larvae attained critical weight at larger sizes and 10 hr later than in controls ( Figure 3A ) . Overexpressing Usp in the PGs did not produce any significant difference in either the size or the age at which critical weight was achieved ( Figure 3B ) . Overexpressing both FoxO and Usp in the PGs resulted in larvae that reached critical weight more than 13 hr later and about 1 mg larger than control larvae ( Figure 3C ) . Further , the size of these larvae at critical weight was significantly larger than when either FoxO or Usp was overexpressed in the PGs alone ( Figure 3D ) . These data suggest that both FoxO and Usp regulate the timing of critical weight . 10 . 7554/eLife . 03091 . 006Figure 3 . Manipulating FoxO and/or Usp in the PGs changes the timing of critical weight . ( A–C ) Age at which animals are starved in relation to the time to pupariation from the onset of starvation for Phm>FoxO ( A ) , Phm>Usp ( B ) , Phm>FoxO , Usp ( C ) animals and their parental controls ( Phm>+ and no driver , ND ) . ( D ) Critical weight was compared when either or both FoxO and/or Usp were overexpressed in the PGs . ( E–G ) Age at which animals are starved in relation to the time to pupariation from the onset of starvation for Phm>dsFoxO ( E ) , Phm>dsUsp ( F ) , and Phm>dsFoxO , dsUsp ( G ) and their parental controls . ( H ) Critical weight was compared when either or both FoxO and/or Usp were knocked down in the PGs . Insets show the size at critical weight ±95% confidence intervals . The age at which larvae reached critical weight ±95% confidence intervals was determined by breakpoint analysis . Points or columns sharing the same letter indicate the groups that are statistically indistinguishable from one another; points or columns that differ in letters are significantly different ( Permutation Test , p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 00610 . 7554/eLife . 03091 . 007Figure 3—figure supplement 1 . Manipulating FoxO and/or Usp in the PGs changes the body size . ( A ) Overexpressing FoxO and/or Usp in the PGs increases body size . ( B ) Knocking down both FoxO and Usp in the PGs decreases body size . One and two asterisks indicate p < 0 . 05 and p < 0 . 01 , respectively , by ANOVA and pairwise t-tests . ( C ) Knocking down both FoxO and Usp in the PGs reduces body size while overexpression of both genes increases size of pharate adult females . From left to right , the pupae are Phm>dsFoxO , dsUsp , Phm>+ and Phm>FoxO , Usp . The scale bar is 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 007 In contrast , knocking down either FoxO or Usp alone in the PGs reduced the size but not the age at critical weight ( Figure 3E or Figure 3F , respectively ) . When we simultaneously knocked down both FoxO and Usp in the PGs , larvae reached critical weight significantly earlier at smaller sizes ( Figure 3G ) than knocking down either FoxO or Usp alone ( Figure 3H ) . These knock down experiments corroborate our results from our FoxO and Usp overexpression experiments and provide further evidence that both FoxO and Usp suppress ecdysone biosynthesis . Because critical weight is a key determinant of final body size , we also weighed pharate adults as a proxy of final adult size . Overexpressing either FoxO or Usp in the PGs significantly increased body size compared to parental controls ( Figure 3—figure supplement 1A ) . In addition , females that overexpressed both FoxO and Usp together in the PGs had significantly larger body sizes than those overexpressing either FoxO or Usp alone ( Figure 3—figure supplement 1A , ANOVA interaction term , p = 0 . 045 ) . Knocking down Usp resulted in a significant decrease in body size ( Figure 3—figure supplement 1B ) . Knocking down FoxO caused a slight , but significant decrease in body size in males but not in females . However , knocking down both Usp and FoxO in the PGs dramatically reduced body size ( Figure 3—figure supplement 1B , C ) . These results demonstrate that altering the size and timing of critical weight , by manipulating expression of FoxO and Usp , has definitive effects on final adult body size . Our data show that the earliest ecdysone peak in the L3 regulates critical weight and that FoxO and Usp alter the timing of this transition . To confirm that FoxO and Usp regulate critical weight by controlling the timing of ecdysone biosynthesis , we examined the expression of two CYP450 ecdysone biosynthetic genes , phm and disembodied ( dib ) , known to be sensitive to IIS/TOR signaling ( Colombani et al . , 2005; Layalle et al . , 2008 ) , in larvae with altered FoxO and Usp expression . In addition , we quantified the expression of an ecdysone response gene , e74B ( eip74ef isoform B ) , which tracks the early effects of ecdysone signaling ( Caldwell et al . , 2005; Colombani et al . , 2005; Layalle et al . , 2008 ) in these larvae . In the parental controls , both phm and dib increased in expression around 8 hr AL3E , shortly before the critical weight ecdysone peak ( Figure 4A , B , D , E ) . E74B expression peaks around 12 hr in parental controls , after the critical weight ecdysone peak ( Figure 4C , F ) . When both FoxO and Usp were overexpressed in the PGs , the increase in phm and dib expression was delayed ( Figure 4A , B ) and e74B expression remained low up to 20 hr AL3E ( Figure 4C ) . In contrast , when we knocked down FoxO and Usp , both phm and dib expression levels were high immediately after the molt to the L3 ( Figure 4D , E ) and e74B expression was nearly undetectable at ecdysis but increased rapidly thereafter ( Figure 4F ) . Taken together , these results suggest that alterations in FoxO and Usp affect the timing of ecdysone biosynthesis at critical weight . 10 . 7554/eLife . 03091 . 008Figure 4 . Altering FoxO and Usp expression also alters phm , dib and e74B expression . ( A–C ) Relative phm ( A ) , dib ( B ) and e74B ( C ) mRNA expression in Phm>FoxO , Usp animals were quantified by quantitative PCR . ( D–F ) Relative phm ( D ) , dib ( E ) and e74B ( F ) mRNA expression in Phm>dsFoxO , dsUsp animals were quantified by qPCR . We normalized the values using an internal control , RpL3 . Then , we standardized the expression level of each gene by fixing the values at 0 hr in Phm>+ animals as 1 in all figures . We used 4–6 larvae for each sample and three biologically independent samples for each time point . Each point indicates the relative mean expression ± SEM . Points sharing the same letter indicate the mean expression at the time ±2 hr are statistically indistinguishable from one another; points that differ in letters are significantly different ( p < 0 . 05 ) . Arrowheads along the x axes indicate the age at which each genotype reached critical weight from Figure 3A , C , G . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 008 Although our results suggest that both FoxO and Usp act in the PGs to regulate the timing of critical weight ecdysone peak , thereby mediating the timing of critical weight , they do not allow us to distinguish whether FoxO and Usp regulate ecdysone biosynthesis independently or together via the FoxO/Usp complex . To discern between these two possibilities , we developed a genetic tool to manipulate the FoxO–Usp interaction . First , we identified the Usp binding site in the FoxO protein using GST-pulldown assays . We created overlapping GST-tagged FoxO fragments and , using increasingly smaller overlapping fragments , we narrowed down the Usp binding region to a 35 amino acid region overlapping with 5 amino acids in the C-terminal end of the forkhead ( DNA binding ) domain ( Figure 5A ) . This motif is well conserved across arthropod species including ticks and water fleas , but is not conserved in FoxO proteins in other ecdysozoans or vertebrates ( Figure 5—figure supplement 1 ) . Interestingly , this Usp binding motif is different from the well-known ‘LXXLL’-type NHR binding motif identified in vertebrates ( Heery et al . , 1997 ) . Next , we identified eighteen candidate amino acids by comparing the crystal structure of mammalian FoxO3a to the Drosophila FoxO sequence ( Tsai et al . , 2007 ) and selecting residues that occupied positions permissive for protein–protein interactions . We mutated each of these to alanine . At least 4 of the 18 amino acid residues appeared to be involved in FoxO–Usp binding ( residues W172 , N175 , R202 and K204 ) . When we introduced these single point mutations into the full length FoxO protein , they showed only mild reductions in FoxO–Usp binding ( Figure 5A ) . We then tested four double–mutant combinations ( W172-R202 , W172-K204 , N175-R202 , and N175-K204 ) all of which were sufficient to dramatically reduce FoxO–Usp interactions ( Figure 5A ) . Two of these double–mutant combinations partially reduced the FoxO activity ( W172-R202 and W172-K204 ) ( Figure 5—figure supplement 2C , D ) , as determined by the expression of known FoxO targets InR and 4E-BP ( Puig et al . , 2003 ) . Because our aim was to disrupt FoxO–Usp binding , but not FoxO function , these were excluded from further analyses . The remaining two double–mutant combinations ( N175-R202 or FoxO NR , and N175-K204 or FoxO NK ) showed normal translocation to the nucleus ( Figure 5—figure supplement 2A , B ) and did not affect FoxO's ability to regulate InR and 4E-BP promoter activities ( Figure 5—figure supplement 2C , D , respectively ) . 10 . 7554/eLife . 03091 . 009Figure 5 . The Usp binding site in FoxO protein was identified and FoxO NK mutation showed reduced binding affinity to Usp . ( A ) Point mutations were induced in the FoxO protein at site of the amino acids indicated in bold . Point mutations indicated in red showed reduced binding affinity to Usp . For a loading control , we used Coomassie Brilliant Blue staining to detect GST-FoxO fusion protein . ( B ) UAS FoxO and UAS FoxO NK transgenes show similar expression levels . We overexpressed either FoxO or FoxO NK using C765-Gal4 . The wing discs were dissected from early white prepupae . We used C765>+ as a parental control , and Histone H3 as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 00910 . 7554/eLife . 03091 . 010Figure 5—figure supplement 1 . Amino acid sequence alignments of the Usp binding motif across arthropods and with non-arthropods . All FoxO sequence information except for Daphnia pulex FoxO was obtained from the NCBI and aligned using the ClustalW2 . The NCBI Reference numbers are: XP_001662969 . 1 ( Aedes aegypti ) , XP_001122804 . 2 ( Apis mellifera ) , HE648216 . 1 ( Blattella germanica ) , JQ081294 . 1 ( Bombyx mori ) , NP_996204 . 1 ( Drosophila melanogaster ) , XP_002433432 . 1 ( Ixodes scapularis ) , XP_001607658 . 2 ( Nasonia vitripennis ) , EEZ98556 . 1 ( Tribolium castaneum ) , NP_001021597 . 1 ( Caenorhabditis elegans , Daf-16 ) and NP_062713 . 2 ( Mus musculus , FoxO1 ) . Daphnia pulex FoxO sequence was obtained from Grigoriev et al . , 2012 . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 01010 . 7554/eLife . 03091 . 011Figure 5—figure supplement 2 . FoxO NK does not change the Usp-independent function of FoxO . ( A and B ) FoxO NK protein translocated to the cytoplasm in the presence of insulin . In all conditions , Dmel cells were transfected with 0 . 4 µg of plasmid . 66 hr after transfection , cells were split into two groups on cover glasses and one was treated with 10 µg/ml bovine insulin for additional 6 hr . These cells were then fixed and processed for immunocytochemistry against HA-tag followed by DAPI and phalloidin staining ( A ) . The scale bar is 10 µm . The HA-tagged FoxO signal intensity in nucleus and entire cell was quantified using ImageJ ( B ) . N = 27–41 . Values indicate mean % ± SEM . Columns sharing the same letters indicate the groups that are statistically indistinguishable from one another; columns with different letters are significantly different ( p < 0 . 05 ) . ( C and D ) FoxO NK activates FoxO target genes in luciferase assays . FoxO NK activates both the InR ( C ) and 4E-BP ( D ) promoters ( N = 4 ) . We used the ampr construct to transfect an equal amount of plasmid in all treatments . Values indicate Luciferase activity/s/mg protein ± SEM . Columns sharing the same letters indicate the groups that are statistically indistinguishable from one another; columns with different letters are significantly different ( ANOVA , p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 01110 . 7554/eLife . 03091 . 012Figure 5—figure supplement 3 . FoxO NK shows Usp-independent FoxO activity in transgenic flies . Adult wing size was quantified in the animals in which transgenes were overexpressed using either C765- ( A ) or MS1096- ( B ) Gal4 . Right wings from females were mounted and photographed , and then wing area was measured by ImageJ . N = 12–16 for ( A ) and N = 16–22 for ( B ) . Values indicate mean area ( mm2 ) ± SEM . Adult eye size was quantified in the animals in which transgenes were overexpressed using either eyeless- ( Ey- ) ( C ) or GMR- ( D ) Gal4 . Left eyes of females were photographed , and then eye area was measured by ImageJ . N = 13–23 for ( C ) and N = 15–29 for ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 012 Finally , we tested whether FoxO NR and FoxO NK retained the ability to suppress tissue growth . We placed the wild type FoxO , FoxO NR and FoxO NK constructs under the control of a UAS promoter and inserted them into the fly genome , using targeted integration ( Groth et al . , 2004 ) to control for positional effects of the transgenes . All three constructs have the same genetic background and differ only in the two amino acids mutated to interfere with FoxO–Usp binding . We then drove expression of the FoxO , FoxO NR and FoxO NK in the wings , using the C765- and MS1096-Gal4 drivers , and in the eyes , using the eyeless ( Ey ) - and GMR-Gal4 drivers . For all drivers , overexpression of FoxO and FoxO NK reduced organ size to a similar degree ( Figure 5—figure supplement 3 ) . FoxO NR showed milder reductions in tissue size ( Figure 5—figure supplement 3 ) , therefore in subsequent experiments we used FoxO NK . We confirmed that both FoxO and FoxO NK expressed FoxO protein at the same level ( Figure 5B ) . Taken together , FoxO NK showed reduced FoxO–Usp affinity , but maintained Usp-independent FoxO function . To explore whether FoxO and Usp regulate critical weight independently or together as a complex , we drove expression of FoxO NK in the PGs of developing larvae and compared them with larvae expressing wild type FoxO in the PGs . Larvae that overexpressed FoxO NK in their PGs reached critical weight earlier and at smaller sizes than those that overexpressed wild type FoxO , albeit later than the parental controls ( Figure 6A ) . Thus , impeding FoxO–Usp binding reduced the delay in critical weight induced by FoxO overexpression . Similarly , pupae in which FoxO NK was overexpressed in the PGs were significantly smaller than pupae that overexpressed wild type FoxO , although they were still larger than pupae from the parental controls ( Figure 6—figure supplement 1A ) . 10 . 7554/eLife . 03091 . 013Figure 6 . Interfering FoxO–Usp association changes the timing of critical weight . ( A and B ) Age at which animals are starved in relation to the time to pupariation from the onset of starvation for Phm>FoxO and Phm>FoxO NK in the FoxO wild type background ( A ) , and P0206>FoxO and P0206>FoxO NK in the FoxO mutant background ( B ) and their parental controls . ( C and D ) Feeding ecdysone throughout the L3 eliminates developmental delay in P0206>+ ( C ) and in P0206>FoxO ( D ) , FoxO mutant larvae . The larvae were continuously fed 0 . 15 mg/g 20E as described in Figure 1 . Data for ND>FoxO and Phm>FoxO in A and for the non-20E-treated data in C and D were re-plotted from Figure 3A and Figure 6B , respectively . Insets show the size at critical weight ( mg ) ±95% confidence intervals . The age at which larvae reach critical weight ±95% confidence intervals was determined by breakpoint analysis . Points or columns sharing the same letters indicate the groups that are statistically indistinguishable from one another; points or columns that differ in letters are significantly different ( Permutation Test , p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 01310 . 7554/eLife . 03091 . 014Figure 6—figure supplement 1 . FoxO NK overexpression in the PGs reduced the body size phenotype . ( A and B ) FoxO NK overexpression in the PGs reduced the body size phenotype in FoxO wild type ( A ) and FoxO mutant ( B ) animals . ( C and D ) Feeding 20E reduces body size in P0206>+ ( C ) and in P0206>FoxO ( D ) , FoxO mutant animals . One and two asterisks indicate p < 0 . 05 and p < 0 . 01 , respectively , and the numbers indicate p-values by ANOVA and pairwise t tests . n . s . indicates no significance . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 01410 . 7554/eLife . 03091 . 015Figure 6—figure supplement 2 . The effects of overexpressing FoxO using P0206-Gal4 is due to the function of FoxO in the PGs and FoxO NK shows proper Usp-independent transcriptional activity . ( A ) Overexpressing FoxO in the oenocytes or corpora allata does not affect developmental timing . We used PromE ( 800 ) -Gal4 as an oenocyte specific driver and Aug21-Gal4 as a corpora allata specific driver . Values indicate average L3 duration ± SEM . Columns sharing the same letters indicate the groups that are statistically indistinguishable from one another; columns with different letters are significantly different ( ANOVA , p < 0 . 05 ) . N = 60–137 . ( B ) Overexpressing FoxO in the oenocytes or corpora allata does not affect body size . N = 27–45 . Two asterisks indicate p < 0 . 01 by ANOVA and pairwise t tests . n . s . indicates no significance . ( C ) Overexpressing Usp with FoxO NK did not show any additional delay of the timing of critical weight . ( D ) Overexpressing both Usp and FoxO NK in the ring gland ( using P0206-Gal4 ) of FoxO null animals did not significantly change female or male pupal weight when compared to overexpressing FoxO NK alone . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 015 Because FoxO alters developmental timing via its effects in other tissues , like the fat body ( Colombani et al . , 2005 ) , we eliminated effects of the endogenous gene by overexpressing FoxO in the PGs of FoxO null mutant larvae ( FoxO Δ94/Df ( 3R ) Exel8159 ) ( Slack et al . , 2011 ) . When we used the Phm-Gal4 driver to overexpress FoxO in the PGs of FoxO null animals , most larvae did not survive to the L3 . To circumvent this problem , we used P0206-Gal4 , which expresses Gal4 moderately in the PGs ( Mirth et al . , 2005 ) . Since P0206-Gal4 driver expresses Gal4 in other tissues such as the oenocytes and corpora allata , we tested the effects of overexpressing FoxO in these other tissues . To do this , we compared the duration of the L3 and final body size when overexpressing FoxO in the ring gland , oenocytes and corpora allata , using P0206-Gal4 , in the oenocytes alone , using PromE ( 800 ) -Gal4 ( Billeter et al . , 2009 ) , and in the corpora allata alone , using Aug21-Gal4 ( Mirth et al . , 2005 ) all in the FoxO null mutant background . We found that overexpressing FoxO in the oenocytes and corpora allata does not affect the duration of the L3 . In contrast , overexpressing FoxO in the ring glands using P0206-Gal4 prolonged the duration of the L3 compared to parental controls ( Figure 6—figure supplement 2A ) . Further , overexpressing FoxO in the oenocytes and corpora allata did not affect final body size , whereas overexpressing FoxO using P0206-Gal4 increased body size ( Figure 6—figure supplement 2B ) . Taken together , our data suggest that FoxO overexpression in the oenocytes and corpora allata had no measurable effect on growth rate or the duration of the L3 . Thus , we conclude that developmental delay and size increase induced by overexpressing FoxO using P0206-Gal4 is due to the functions of FoxO in the PGs . Similar to what we found in the FoxO wild type background , FoxO null larvae that overexpressed FoxO NK in their PGs reached critical weight earlier and at smaller sizes than larvae that overexpressed wild type FoxO in their PGs ( Figure 6B ) . Further , we confirmed that overexpressing both FoxO NK and Usp in the PGs of FoxO null mutant larvae did not alter the age and size at critical weight ( Figure 6—figure supplement 2C ) nor did it alter final body size when compared to overexpressing FoxO NK alone ( Figure 6—figure supplement 2D ) . These results suggest that Usp does not affect the timing of critical weight on its own and that critical weight is regulated , at least in part , by FoxO/Usp . We next tested whether exogenous ecdysone could rescue the delay in critical weight induced by the FoxO/Usp complex . To do this , we assessed critical weight in P0206>FoxO , FoxO null larvae and the P0206>+ , FoxO null parental controls on medium supplemented with 20E . We found that adding 20E to the medium altered the relationship between age at starvation and time to the onset of metamorphosis in both genotypes ( Figure 6D ) . In the parental controls , larvae starved before 12 hr AL3E on ecdysone-supplemented agar did not delay the onset of metamorphosis , pupariating 36 hr after starvation ( Figure 6C , Supplementary file 1 ) . When we starved P0206>FoxO , FoxO null larvae on 20E supplemented agar , time to pupariation increased with age of starvation from 0–12 hr AL3E , with larvae showing a maximum time to pupariation of 57 hr ( Supplementary file 1 ) at 12 hr AL3E , then decreased thereafter . This suggests that ( 1 ) 20E administration eliminated the strong delays in time to metamorphosis seen in pre-critical weight P0206>FoxO , FoxO null larvae and ( 2 ) FoxO overexpression in the PGs has additional stage-specific effects on the time to pupariation after the critical weight ecdysone peak and this effect is nutrition-sensitive . In addition to the effects we observed on developmental time , we found that both P0206>FoxO , FoxO null and P0206 , FoxO null parental control animals were significantly smaller when they were continuously fed 20E-supplemented normal fly medium when compared to animals reared on normal fly medium alone ( Figure 6—figure supplement 1C , D ) . These data demonstrate that altering the timing of critical weight , by exogenous ecdysone administration , impacts final adult size . The goal of this study was to uncover the molecular mechanism through which nutrition regulates ecdysone synthesis at critical weight . Our results show that FoxO and Usp interact to regulate critical weight and suggest that this interaction alters the timing of the ecdysone peak . To definitively test whether the FoxO/Usp complex regulates ecdysone biosynthesis at critical weight , we examined whether the FoxO/Usp complex altered the timing of phm and dib mRNA expression , the timing of e74B expression and finally the timing of the ecdysone peak itself . The expression of both phm and dib mRNA peaked shortly before critical weight in the parental controls ( Figure 7A , B and Figure 7—figure supplement 1A–C ) . However , P0206>FoxO , FoxO null larvae showed significant delays in this peak ( Figure 7B , D ) . Overexpressing FoxO NK in the PGs reduced the delays induced by FoxO overexpression ( Figure 7B , D ) . Similarly in the wild type background , phm and dib expression was upregulated significantly earlier when FoxO NK was expressed in the PGs than when FoxO was overexpressed in this tissue ( Figure 7—figure supplement 1A , B ) . 10 . 7554/eLife . 03091 . 016Figure 7 . The FoxO/Usp complex suppresses critical weight through inhibiting ecdysone biosynthesis in the PGs . ( A–C ) Relative phm ( A ) , dib ( B ) , and the ecdysone response gene e74B mRNA expression ( C ) in the FoxO mutant backgrounds were quantified by qPCR . We normalized the values by an internal control , ribosomal protein large subunit 3 ( RpL3 ) . Then , we standardized the expression level of each gene by fixing the values at 0 hr in P0206>+ animals as 1 . We used 5–6 larvae for each sample and three biologically independent samples for each time point . Each point indicates the relative mean expression ± SEM . Points sharing the same letters indicate the mean expression at the time ±2 hr are statistically indistinguishable from one another; points that differ in letters are significantly different ( p < 0 . 05 ) . ( D ) The FoxO/Usp complex suppresses ecdysone biosynthesis during critical weight period in larvae with FoxO mutant backgrounds . We used 32–46 larvae for each sample and three biologically independent samples for each time point . Each point indicates the mean ecdysone concentration ± SEM . Points sharing the same letter indicate the mean concentration at the time ±2 hr are statistically indistinguishable from one another; points that differ in letters are significantly different ( p < 0 . 05 ) . Arrowheads along the x axes indicate the age at which each genotype reached critical weight from Figure 6B . ( E ) Overexpressing FoxO or FoxO NK equally reduces the PG size of the FoxO null mutant larvae . The PGs were dissected at 24 hr AL3E and stained with phalloidin . After photographing , these areas were quantified using the ImageJ . Each bar indicates the mean area ± SEM . N = 7–10 . Columns sharing the same letter indicate the groups that are statistically indistinguishable from one another; columns that differ in letters are significantly different ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 01610 . 7554/eLife . 03091 . 017Figure 7—figure supplement 1 . The FoxO/Usp complex delays ecdysone synthesis and ecdysone response gene expression in the FoxO wild type background . Relative phm ( A ) , dib ( B ) and e74B ( C ) mRNA expression in Phm>FoxO and Phm>FoxO NK animals were quantified by qPCR . We normalized the values using an internal control , RpL3 . Then , we standardized the expression level of each gene by fixing the values at 0 hr in Phm>+ animals as 1 in all figures . We used 4–6 larvae for each sample and three biologically independent samples for each time point . Each point indicates the relative mean expression ± SEM . Points sharing the same letter indicate the mean expression at the time ±2 hr are statistically indistinguishable from one another; points that differ in letters are significantly different ( p < 0 . 05 ) . Arrowheads along the x axes indicate the age at which each genotype reached critical weight from Figure 3A , 6A . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 017 Even if we observe alterations in ecdysone biosynthesis gene expression , this does not necessarily mean that ecdysone biosynthesis is affected when we manipulate FoxO expression in the PGs . To assess if overexpressing FoxO in the PGs affected ecdysone signaling , we examined the expression of e74B . In the parental control larvae , e74B mRNA expression was up-regulated around 12 hr AL3E , shortly after critical weight ( Figure 7C , Figure 7—figure supplement 1C ) . Overexpressing FoxO in the PGs delayed the up-regulation of e74B in both the FoxO mutant and wild type backgrounds ( Figure 7C , Figure 7—figure supplement 1C ) . Finally , interfering with FoxO/Usp complexes , by overexpressing FoxO NK , in the PGs reduced this delay in both the FoxO mutant and wild type backgrounds ( Figure 7C , Figure 7—figure supplement 1C ) . Thus , FoxO/Usp complex plays a role in regulating the dynamics of ecdysone signaling at critical weight . Finally , to show that the FoxO/Usp complex regulates ecdysone biosynthesis at critical weight , we measured ecdysone concentrations in larvae that expressed either FoxO or FoxO NK in their PGs from 6 hr AL3E until the nutrition-dependent critical weight ecdysone peak . Overexpressing FoxO in the PGs of FoxO mutant larvae induced a significant delay in the critical weight ecdysone peak ( Figure 7D ) . In addition , the maximum concentration of this peak was approximately 50% lower than the critical weight ecdysone peak in parental controls . The critical weight ecdysone peak occurred significantly earlier in P0206>FoxO NK larvae than in P0206>FoxO larvae . This difference in the timing of the critical weight ecdysone peak was not due to differences in the effects between FoxO and FoxO NK on PG size . Overexpressing either FoxO or FoxO NK induced indistinguishable reductions in PG size ( Figure 7E ) . Taken together our results show that FoxO acts to control the timing of ecdysone biosynthesis via the FoxO/Usp complex , but also via Usp-independent mechanisms . Increasing IIS/TOR activity in the PGs induces precocious critical weight and reducing its activity in the PGs prolongs this transition ( Caldwell et al . , 2005; Colombani et al . , 2005; Mirth et al . , 2005; Layalle et al . , 2008 ) . Because IIS/TOR signaling positively regulates the expression of CYP450 ecdysone biosynthetic genes , phm and dib ( Colombani et al . , 2005; Layalle et al . , 2008 ) , we previously hypothesized that IIS/TOR exerted these effects by regulating the timing of the small peak of ecdysone that coincides with critical weight ( Mirth and Riddiford , 2007; Mirth and Shingleton , 2012 ) . Our data both tested this hypothesis and identified a novel interaction between the IIS/TOR and ecdysone signaling pathways . We have found that interactions between FoxO and Usp regulate ecdysone biosynthesis , critical weight and body size . This allows us to propose a model for nutrition-sensitive ecdysone biosynthesis during critical weight ( Figure 8 ) . During the molt to the L3 , larvae undergo a period of starvation while they expel their mouthparts ( Park et al . , 2002 , 2003 ) . As a consequence , IIS/TOR signaling activity in the PGs is reduced and FoxO remains in the nucleus and forms a complex with Usp . The FoxO/Usp complex suppresses ecdysone biosynthesis at least in part by repressing transcription of phm and dib , although we do not know whether this repression is direct . Once larvae start feeding , increasing IIS/TOR activity in the PGs results in the phosphorylation of FoxO , causing the dissociation FoxO/Usp complexes as FoxO moves out of the nucleus . This progressive dissociation of FoxO/Usp complexes results in a gradual rise in ecdysone biosynthesis . Once ecdysone reaches a threshold , it triggers critical weight . Afterwards , the time to metamorphosis is set and can no longer be delayed by starvation . For other ecdysone peaks , a negative feedback loop induced by ecdysone signaling itself down-regulates ecdysone biosynthesis ( Sakurai and Williams , 1989; Takaki and Sakurai , 2003; Moeller et al . , 2013 ) . We expect that negative feedback by ecdysone results in the decline in ecdysone biosynthesis after the critical weight peak . 10 . 7554/eLife . 03091 . 018Figure 8 . Proposed model: Nutrition regulates ecdysone biosynthesis during critical weight through FoxO/Usp . At the onset of the L3 ( left ) , IIS/TOR signaling is reduced in the PG cells and the FoxO/Usp complex suppresses ecdysone biosynthesis either directly , as drawn , or indirectly . As the larvae feed , FoxO becomes phosphorylated and transported out of the nucleus , thereby dissociating FoxO/Usp complexes . As a result , ecdysone biosynthesis becomes derepressed ( upper right ) . After critical weight , ecdysone reduces its own biosynthesis through a negative-feedback loop . In starved conditions , the IIS/TOR signaling activity in the PGs remains low , thereby unphosphorylated FoxO remains inside of nuclei forming complexes with Usp ( lower right ) . This inhibits ecdysone biosynthetic gene expression , thereby repressing ecdysone biosynthesis and delaying metamorphosis . FoxO on its own or with an unknown partner ( s ) may also regulate ecdysone biosynthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 03091 . 018 In contrast , when larvae are starved before attaining critical weight , FoxO remains in the nucleus . In these larvae , the FoxO/Usp complex suppresses ecdysone biosynthesis and delays critical weight . Consequentially , the onset of metamorphosis is delayed . This work uncovers a mechanism that allows IIS/TOR signaling to control ecdysone biosynthesis , providing an elegant means for nutrition to regulate body size . Although the ecdysone peak at critical weight is environmentally-sensitive , many other peaks that occur throughout the larval period show less plasticity in response to environmental cues . Ecdysone biosynthesis is also regulated by a developmental neuropeptide , prothoracicotropic hormone ( PTTH ) . Several extrinsic and intrinsic stimuli affect PTTH secretion , such as photoperiod , oxygen concentrations , signals released from damaged imaginal discs , and the sesquiterpenoid ‘status quo’ hormone juvenile hormone ( Truman , 1972; Nijhout and Williams , 1974a , 1974b; McBrayer et al . , 2007; Halme et al . , 2010; Callier and Nijhout , 2011; Colombani et al . , 2012; Garelli et al . , 2012; Mirth and Shingleton , 2012 ) . Activating downstream targets of PTTH signaling in the PGs accelerates the onset of metamorphosis ( Caldwell et al . , 2005 ) and ablating the PTTH-producing cells delays critical weight ( McBrayer et al . , 2007; Rewitz et al . , 2009 ) . Further , without PTTH the ecdysone peak that stimulates wandering behavior , where the larvae emerge from the food to begin metamorphosis , is dramatically delayed ( McBrayer et al . , 2007; Rewitz et al . , 2009; Gibbens et al . , 2011 ) . Thus in contrast to IIS/TOR signaling whose major effects are to control the critical weight ecdysone peak , PTTH regulates all ecdysone peaks . Why particular ecdysone peaks are more sensitive to IIS/TOR signaling is unclear , however understanding the mechanisms underlying this differential sensitivity may be key to understanding developmental plasticity . FoxO also regulates the critical weight ecdysone peak independently of Usp; overexpressing FoxO NK in the PGs still induces delays in ecdysone biosynthesis and critical weight , even if these delays are more moderate than those induced by wild type FoxO . Thus , our data suggest that FoxO plays additional roles in regulating ecdysone biosynthesis , either on its own or through interaction with other binding partners . The effects of starvation on ecdysone biosynthesis do not appear to be the same for all stages of development . Even though starvation causes a delay in development before attaining critical weight , once they reach critical weight , starvation induces moderate acceleration in the time to metamorphosis ( Mirth et al . , 2005; Stieper et al . , 2008 ) . This suggests that reducing IIS/TOR signaling induces a mild acceleration of ecdysone biosynthesis at later stages of the L3 development . How IIS/TOR activity regulates ecdysone biosynthesis differently depending on the stage of development is unclear , but it may result from interaction of alternate FoxO binding partners . Our findings have broad implications for our understanding of the mechanisms of size regulation and the development of other environmentally-sensitive traits . In other insects , traits such as seasonal wing morphs in butterflies ( Koch and Bückmann , 1987; Koch et al . , 1996; Oostra et al . , 2011 ) or horn length in male dung beetles ( Emlen and Nijhout , 1999 ) arise from differential regulation of ecdysone biosynthesis ( Koyama et al . , 2013 ) . Horn length in dung beetles is highly nutrition-dependent , with small , poorly-fed males bearing small horns and large , well-fed males having disproportionately larger horns ( Emlen , 1994 , 1997 ) . Small-horned males have a characteristic peak of ecdysone in their final instar absent in their well-fed , larger conspecifics ( Emlen and Nijhout , 1999 , 2001 ) . Our data propose a mechanism through which nutrition , via FoxO–Usp interactions , might regulate this peak ( Koyama et al . , 2013 ) . FoxO is also known to form complexes with many vertebrate NHRs , including thyroid hormone ( Zhao et al . , 2001 ) , androgen ( Li et al . , 2003; Fan et al . , 2007 ) and estrogen receptors ( Schuur et al . , 2001 ) . The steroid sex hormones , such as testosterone and estrogen , are important for initiating puberty and the development of adult characters in humans . In girls , reaching a body mass of 48 kg determines the timing of first menses ( Frisch and Revelle , 1970; Freedman et al . , 2003; Gluckman and Hanson , 2006; Ahmed et al . , 2009 ) . Obese girls reach this mass faster , resulting in earlier onset of puberty ( Freedman et al . , 2003; Gluckman and Hanson , 2006; Ahmed et al . , 2009 ) possibly due to higher levels of insulin signaling ( Codner and Cassorla , 2009; Lee et al . , 2011; Lombardo et al . , 2009; von Berghes et al . , 2011 ) . These findings suggest that IIS/TOR activity regulates the production of the steroid sex hormones to regulate developmental timing in vertebrates . Furthermore , two mammalian NHRs , CAR and PXR , associate with FoxO1 to regulate the expression of the CYP450 enzymes ( Kodama et al . , 2004 ) . The similarity in the roles of FoxO/NHR complexes between mammals and insects provides a testable model that FoxO-NHR complexes regulate environmentally-sensitive development in a wide range of organisms . Wild type FoxO and FoxO NK were amplified by RT-PCR using cDNA made from whole body extract of post-feeding ( wandering ) w[1118] larvae . After sequencing , the constructs were inserted into pUAST attB vector using EcoRI and KpnI whose recognition sites are included on the primers , then integrated on the second chromosome by site-directed insertion using the phiC31 integrase and an attP landing site carrying recipient line , y[1] w[1118]; PBac{y[+]-attP-9A} VK00018 ( Bloomington Drosophila Stock Center #9736 ) ( Groth et al . , 2004 ) . w; UAS Usp 26A3 line was a gift from Dr Michael O'Connor ( University of Minnesota ) . We used FoxO Δ94 , a gift from Dr Linda Partridge ( University College London ) , with the deficiency line , w[1118]; Df ( 3R ) Exel8159/TM6B , Tb[1] ( #7976; Bloomington ) as our FoxO null mutant . We obtained the PromE ( 800 ) -Gal4 ( also known as Oe-Gal4 ) line from Dr Carlos Ribeiro ( Champalimaud Centre for the Unknown ) . For Usp and FoxO knock down experiments , we used y[1] v[1]; P{y[+t7 . 7] v[+t1 . 8] = TRiP . JF02546}attP2 ( #27258; Bloomington ) as UAS double-stranded ( ds ) Usp and Vienna Drosophila RNAi Center 107786 as UAS dsFoxO . Egg collections were performed on normal food plates and larvae were reared at controlled densities without additional yeast ( about 200 eggs/60 mm diameter normal fly medium plate ) . Newly molted L3 larvae were collected every 2 hr . Collected larvae were raised in a normal cornmeal/molasses medium without additional yeast until the appropriate time point . For starvation treatments , we used 1% non-nutritional agar . To determine the duration of the L3 , pupariation time was observed every 2 hr until all treated larvae pupariated or died . We defined pupariation as cessation of movement with evaginated spiracles . All treatments were performed at 25°C under constant light to avoid the effect of circadian rhythm on PTTH secretion . To analyze critical weight , we used a breakpoint analyses as previously described ( Stieper et al . , 2008; Ghosh et al . , 2013; Testa et al . , 2013 ) . We constructed growth curves by weighing larvae across a range of defined ages . We then starved larvae of different age classes on non-nutritive agar media and measured the time it took for them to reach pupariation from the onset of starvation , checking for pupariation every 2 hr . We converted the time at which larvae reached critical weight to size using linear regression models from the growth curves . For ecdysone feeding experiments , we added 0 . 15 mg of 20E ( SciTech Chemicals , Dejvice-Hanspaulka , Czech Republic ) to 1 g of normal fly medium or starvation medium ( 1% non-nutritive agar ) . After 20E was added , the media were well mixed and spun down a day before use . To measure time to pupariation from the onset of starvation , larvae were collected as above in 2 hr intervals from the molt . They were then transferred to 20E-supplemented fly medium until they reached the desired age for transfer to 20E-supplemented agar . As a proxy for adult body size , we individually weighed pharate adults . Pharate adults , which were about 6–14 hr before eclosion , were collected from vials , carefully cleaned off using distilled water and a paint brush , and then dried for 15 min on paper towels . Once dry , pharate adults were individually weighed on an ultra-microbalance ( Sartorius , SE2 ) . We observed the presence or absence of male specific sex combs through pupal cases under a stereoscope to distinguish males from females . Concentrations of 20E were quantified using a 20-Hydroxyecdysone EIA kit ( Cayman Chemicals ) . Carefully staged larvae were washed in distilled water twice , briefly dried on paper towels , weighed and flash-frozen on dry ice . Larvae were preserved in three-times their volume of ice cold methanol and kept at −80°C until use . Ecdysone extraction was performed as previously described ( Mirth et al . , 2005 ) . Concentrations of 20E were quantified according to the manufacturer's instructions . Entire coding regions of Drosophila foxo , ecr-A , ecr-B1 and usp cDNAs were isolated by RT-PCR using cDNA made from w[1118] wandering larvae . For ecr-A , ecr-B1 and usp RT-PCR , forward primers were designed for gene specific sequences with a Flag-tag sequence on the 5′-end and reverse primers were designed for gene specific sequences including native stop codons . To create point mutation constructs , we designed primers containing point mutation ( s ) and performed standard site-directed mutagenesis methods with minor modifications . GST-tagged protein was purified by Glutathione Sepharose 4B ( GE Healthcare ) . Flag-tagged protein was detected by the anti-Flag M2 monoclonal antibody ( 1:1000 , Sigma ) . For co-immunoprecipitation assays , we used 500 µg of larval protein or cell extract , the AB11 ( anti-Usp monoclonal antibody ) [gifts from Drs Sho Sakurai ( Kanazawa University ) and Lynn M Riddiford ( Janelia Farm Research Campus , HHMI ) ] and DDA2 . 7 ( anti-EcR monoclonal antibody , Developmental Studies Hybridoma Bank ) . For western blots , the antibodies we used were: anti-Usp ( 1:1000 , AB11 ) , anti-EcR ( 1:5000 , DDA2 . 7 ) , anti-FoxO ( 1:1000 ) ( Puig et al . , 2003 ) and anti-Histone H3 ( 1:1000 , Cell Signaling ) . Immunocytochemistry was performed using standard methods as described previously ( Mirth et al . , 2009 ) . The antibodies we used were: anti-Usp ( 1:100 , AB11 ) , anti-FoxO ( 1:1000 ) and anti-HA ( 1:100 , Covance ) . For nuclei and actin staining , we used DAPI ( Invitrogen ) and Phalloidin ( Sigma ) , respectively . Total RNA was extracted from entire larval bodies using TRIzol ( Invitrogen ) . After DNase treatment , total RNA concentration was quantified and 1 µg total RNA was converted to cDNA using oligo dT primers and reverse transcriptase . qPCR was performed using SYBR Green PCR Master Mix ( Applied Biosystems ) and ABI 7900HT ( Applied Biosystems ) . Primers are listed in Supplementary file 2 . The Dmel cell line was used for all cell culture experiments . Cells were cultured in the Express Five SMF medium ( Gibco ) without any serum , insulin or additives , unless mentioned . Transfection was performed using FuGENE HD Transfection Reagent ( Roche ) , according to the manufacturer's instructions . For insulin treatment , transfected cells were re-suspended 66 hr after transfection , and split into two groups . 10 µg/ml bovine insulin ( Sigma ) was added into the medium of one of these groups . Cells were kept for additional 6 hr at 25°C . Luciferase assays were performed using the Luciferase Assay System ( Promega ) , according to the manufacturer's instructions . To transfect equal amount of plasmid between all treatments , we used bacterial ampicillin resistance gene ( ampr ) . InR- and 4E-BP-luciferase constructs were made according to previous study ( Puig et al . , 2003 ) . Briefly , we designed restriction enzyme site-attached primers ( Supplementary file 3 ) and amplified these promoter regions by PCR using w[1118] genomic DNA . After sequencing , we digested these fragments by NotI and BamHI and inserted into modified pAc5-V5-His B vector ( Invitrogen ) . Data for pharate adult weight for males and females , critical weight , growth rates , qPCR and ecdysone quantifications are deposited in Dryad ( doi:10 . 5061/dryad . 75940 ) ( Koyama et al . , 2014 ) . In addition , we have uploaded the scripts to generate the breakpoint plots , calculate critical size from the growth curves and to perform the permutation tests .	The development of many of an animal's traits and characteristics are sensitive to the conditions of its environment . The conditions experienced early on in life , in particular , can alter how an animal grows and develops . The availability of food during development , for example , affects the final body size of many animals—including the larvae of fruit flies , which must reach a so-called ‘critical weight’ before they can change into adults . A hormone called ecdysone controls when a larva turns into an adult insect . This hormone's levels peak when a fruit fly larva has reached its critical weight , and this peak triggers a cascade of events that ultimately will cause the larva to stop growing and to change into an adult fly . A signaling pathway involving insulin is also known to help regulate body size in response to nutrition . But it was unclear how the timing , and size , of the ecdysone peak is altered to match the larva's diet—and how the insulin-related pathway exerts its effect on body size . Now , Koyama et al . have found a molecular link between this signaling pathway and the ecdysone hormone that can explain how nutrition can regulate the growth of the flies . First , Koyama et al . found that the ecdysone peak is delayed when young fruit fly larvae were starved before they had reached the critical weight . On the other hand , feeding these starved larvae the ecdysone hormone prevented this delay; and many of these larvae developed into underweight adults that were about three-quarters the size of a typical fully-grown adult . The ecdysone hormone is made by cells within certain glands in the larvae . Koyama et al . also found that a protein called FoxO , which inhibits the insulin-related pathway , is transported out of the nuclei of these gland cells when growing larvae gain weight . But when larvae that had not reached their critical weight were starved , the FoxO protein was kept within the cell nuclei . Koyama et al . found that , in the nucleus , the FoxO protein blocks the production of ecdysone by interacting with a protein that forms part of the ecdysone receptor . This protein complex delays the expression of genes that are involved in making the hormone . When larvae were fed , the FoxO protein began to leave the nucleus , which allowed ecdysone production to resume . Future studies could now test whether FoxO controls other traits that are affected by the insect's diet ( such as lifespan ) , and if growth-related hormones in other animals are affected by a similar mechanism .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2014	Nutritional control of body size through FoxO-Ultraspiracle mediated ecdysone biosynthesis
We use the myotendinous junction of Drosophila flight muscles to explore why many integrin associated proteins ( IAPs ) are needed and how their function is coordinated . These muscles revealed new functions for IAPs not required for viability: Focal Adhesion Kinase ( FAK ) , RSU1 , tensin and vinculin . Genetic interactions demonstrated a balance between positive and negative activities , with vinculin and tensin positively regulating adhesion , while FAK inhibits elevation of integrin activity by tensin , and RSU1 keeps PINCH activity in check . The molecular composition of myofibril termini resolves into 4 distinct layers , one of which is built by a mechanotransduction cascade: vinculin facilitates mechanical opening of filamin , which works with the Arp2/3 activator WASH to build an actin-rich layer positioned between integrins and the first sarcomere . Thus , integration of IAP activity is needed to build the complex architecture of the myotendinous junction , linking the membrane anchor to the sarcomere . Cell adhesion to the extracellular matrix ( ECM ) is essential for the development and homeostasis of multiple cell types and tissues ( Winograd-Katz et al . , 2014 ) . Adhesion to the ECM is primarily mediated by integrins , α/β heterodimeric transmembrane receptors . The medical importance of integrin adhesion is exemplified by the range of human diseases resulting from its loss or misregulation ( Winograd-Katz et al . , 2014 ) . The extracellular domains of both integrin subunits binds ECM ligands , whereas it is primarily the intracellular domain of the β subunit that recruits intracellular integrin-associated proteins ( IAPs ) . As β subunit cytoplasmic tails lack enzymatic activity and almost all are short ( ~47 residues ) it is the IAPs that mediate integrin signalling and interaction with the actin cytoskeleton ( Campbell and Humphries , 2011 ) . Over 150 different proteins have been implicated in integrin adhesion , and 60 are consistently found in adhesions ( Horton et al . , 2015 ) . We aim to discover why so many proteins are required for the seemingly simple task of linking the ECM to the actin cytoskeleton . Two features of integrin adhesion sites may help explain the complexity of the machinery required: their complex architecture , and their ability to respond to mechanical forces . Super-resolution microscopy of IAPs revealed that focal adhesions in fibroblasts have a complex nano-architecture with distinct layers in the Z-axis: a membrane proximal integrin signalling layer , a force transduction layer and an actin regulatory layer that connects to the actin stress fiber ( Kanchanawong et al . , 2010 ) . Most IAPs localise to specific layers , but talin spans two layers with its N-terminus in the integrin signalling layer and C-terminus in the force transduction layer . It is not yet known whether a similar nano-architecture is present in stable integrin adhesions within tissues . In Drosophila embryonic muscles talin is orientated similarly to fibroblasts , whereas in the developing wing both ends are close to the membrane ( Klapholz et al . , 2015 ) . This suggests diverse architectures of integrin adhesion sites , and an additional rationale for the large IAP number . The maturation and stability of focal adhesions in vertebrate cells in culture requires force from actin polymerisation and/or acto-myosin contraction ( Sun et al . , 2016 ) . Building integrin-mediated attachment sites in muscles during embryogenesis has two phases: the initial attachment phase that does not require myosin-mediated contractions , followed by an increase in IAP levels and stoichiometry changes , as muscles begin contracting ( Bulgakova et al . , 2017 ) . Such a force-responsive mechanism may ensure adhesion strength is balanced with contractile force . The recruitment of vinculin by talin is a valuable paradigm for how adhesion can be strengthened in response to force , with force-induced stretching of talin unfolding protein domains to reveal binding sites for vinculin , and bound vinculin providing new links to actin ( Klapholz and Brown , 2017 ) . A valuable way to assess the contribution of each IAP is to compare the defects that occur when they are genetically removed . The reduced number of paralogs has made this comparison easier in Drosophila . Focusing on integrin function at muscle attachment sites , analogous to myotendinous junctions , all IAPs are present but the consequence of removing IAPs ranges from complete loss of integrin adhesion to no apparent defect ( Bulgakova et al . , 2012 ) . The loss of some IAPs could be tolerated either because they contribute a minor function not necessary in the laboratory , or due to redundancy between IAPs . To discover how these apparently minor IAPs contribute to integrin adhesion , and test their redundancy , we utilized an integrin adhesion that resists the strenuous activity of flight . Indirect flight muscles ( IFMs ) are the major muscles in the adult thorax that power flight ( Gunage et al . , 2017 ) . IFMs are more similar to vertebrate muscles than most Drosophila muscles , because IFMs are fibrillar as opposed to tubular muscles such as the adult leg muscle . Fibrillar muscles are made up of myofibers , which in turn contain hundreds of myofibrils . The myofibril sarcomeres are of standard structure , comprising a regular repeating structure of overlapping actin and myosin filaments . Actin filament barbed ends are anchored at Z-lines with their pointed ends towards the center of the sarcomere , while bipolar myosin filaments overlap with actin filaments in the center of the sarcomere and are anchored at M-lines . In the IFMs , sarcomeres form simultaneously along the length of each myofibril during pupal development ( Weitkunat et al . , 2014 ) and then elongate by growth from the actin pointed ends , and widen by circumferential addition ( Mardahl-Dumesnil and Fowler , 2001; Shwartz et al . , 2016 ) . At the ends of each myofibril are modified terminal Z-lines ( MTZ ) , which contain sites of integrin adhesion to the epidermal tendon cell . The myofibril and tendon cell membranes form finger-like protrusions that interdigitate with one another ( hereafter referred to as interdigitations ) . The attachment sites in IFMs appear by electron microscopy more complex and larger than those in larval muscles , with ~1 µm separating plasma membrane from first sarcomere versus 0 . 1–0 . 5 µm ( Reedy and Beall , 1993; Tepass and Hartenstein , 1994 ) . Furthermore , IFMs contract 200 times per second ( Fry et al . , 2003 ) , whereas larval muscles contract once every ~5 s during crawling ( Crisp et al . , 2008 ) . We use the IFMs to discover functions for four IAPs ( tensin , RSU1 , focal adhesion kinase ( FAK ) and vinculin ) that surprisingly are not needed for viability of the fly , despite all colocalizing at integrin adhesion sites throughout development , and being conserved in all metazoa ( Alatortsev et al . , 1997; Grabbe et al . , 2004; Kadrmas et al . , 2004; Torgler et al . , 2004; Maartens et al . , 2016 ) ( diagrammed in Figure 1—figure supplement 1A ) . We recently reported a new phenotype in flies lacking vinculin ( Maartens et al . , 2016 ) , consisting of defective actin organisation in IFMs , suggesting that IFMs might reveal functions for other IAPs . Here we show that indeed IFMs reveals a new phenotype for each IAP examined , and genetic interactions reveal how they work together by balancing positive and negative control . Tensin is able to bind integrins ( Haynie , 2014 ) , and contributes to focal adhesion maturation ( Rainero et al . , 2015 ) and integrin inside-out activation ( Georgiadou et al . , 2017 ) . In IFMs we also find tensin elevates integrin activity , and discover this is negatively regulated by FAK , a cytoplasmic tyrosine kinase ( Kleinschmidt and Schlaepfer , 2017 ) , thus providing a new mechanism for its inhibition of integrin adhesions ( Ilić et al . , 1995 ) . Like FAK , we discovered that RSU1 has a negative regulatory role in IFMs . RSU1 is a component of the IPP complex , containing integrin-linked-kinase , PINCH and parvin , via its binding to PINCH , and the function of this complex is not well understood ( Xu et al . , 2016 ) . Here we show that instead of promoting IPP function , as previously thought ( Kadrmas et al . , 2004; Dougherty et al . , 2005 ) , RSU1 inhibits PINCH activity . By examining the localization of different IAPs and actin-binding proteins , we show that the MTZ is composed of 4 distinct layers . One of these layers , an actin rich region between the actin layer adjacent to the membrane and the start of the first sarcomere , requires vinculin , the mechanosensitive actin-binding protein filamin ( Razinia et al . , 2012 ) and the nucleation promoting factor WASH ( Alekhina et al . , 2017 ) . Finally , we find that the mechanosensitive protein vinculin has dual functions at IFM adhesions . As mentioned , vinculin recruitment by stretched talin is a paradigm for mechanotransduction , yet we find that vinculin , once opened , can also function independently of talin , building an actin zone by promoting mechanical opening of filamin . To understand how IAPs contribute to integrin-mediated adhesion we sought to identify defects that occur in their absence . An unexpected finding is that some highly conserved IAPs can be genetically removed without impairing viability or fertility . Thus , we have been searching for biological processes that fail in the absence of this group of IAPs . We recently discovered that whereas the genetic removal of vinculin does not impair viability or fertility ( Alatortsev et al . , 1997 ) , it does cause a fully penetrant phenotype in the highly active IFMs in the adult ( Maartens et al . , 2016 ) . This prompted us to examine the IFMs in flies lacking three additional IAPs that are also present at integrin adhesion sites throughout development yet can be removed without impairing larval muscle function or viability: tensin , RSU1 and FAK ( Grabbe et al . , 2004; Kadrmas et al . , 2004; Torgler et al . , 2004 ) ( although not required for viability , loss of RSU1 and tensin causes wing blisters ) . We analysed the development of the myofibrils in each muscle by staining for filamentous actin at different stages of development and imaging by confocal microscopy ( Figure 1A ) . No gross defects in muscle morphology were observed , but higher magnification revealed defects in myofibril morphology . To identify which phenotypes we should focus on , we first quantified aspects of myofibrils of flies that are 1 day old ( after eclosion [AE] ) : myofibril width; sarcomere length; and the length of myofibril attachment sites , which in IFMs are called modified terminal Z-lines ( MTZ ) ( Reedy and Beall , 1993 ) ( Figure 1B–D ) . Myofibril width and sarcomere length were normal in IFMs lacking FAK or tensin , whereas IFMs lacking RSU1 had a reduced myofibril width and increased sarcomere length . Loss of each of these IAPs caused defects in MTZ length and unique alterations to MTZ structure ( Figure 1A ) , so we focused our analysis on the MTZ . In wild type , the MTZ contains a brightly-staining block of actin , and the plasma membrane at the attachment site is interdigitated with the tendon cell , giving a wavy appearance ( Reedy and Beall , 1993 ) . The MTZ extends ~1 µm away from the plasma membrane to the Z-line at the start of the first sarcomere . In the absence of FAK the MTZ became modestly elongated , to 1 . 25 µm ( Figure 1A and E ) . Myofibrils lacking RSU1 had an even more elongated MTZ , with a clear elongation of the interdigitations . Loss of tensin resulted in elongated and irregularly shaped MTZs , and approximately 20% of myofibrils were detached ( Figure 1A and E ) . We confirmed that these defects were due to loss of these three IAPs , rather than any other mutation on the mutant chromosomes , as a transgenic copy of each gene rescued the defects ( Figure 1—figure supplement 1B and D ) . In addition , these phenotypes result from loss of the IAP in the muscle since driving RNAi targeting these IAPs specifically in the muscles phenocopied the mutant phenotypes ( Figure 1—figure supplement 1B and D ) . We then used transmission electron microscopy ( TEM ) to examine the MTZ defects in more detail ( Figure 1B ) . During the development of the muscles the attachment site becomes progressively more interdigitated with the tendon cell ( Reedy and Beall , 1993 ) , and this process appears defective in the absence of RSU1 and tensin . We see clear evidence of detachment of the muscle in the tensin mutant , but not in RSU1 , showing that in the latter the more elongated MTZ does not reflect partial detachment . In the FAK mutant the electron dense layer adjacent to the plasma membrane , which we infer is the site of concentrated IAPs , was more continuous than in wild type ( in wild type MTZs , this layer is reduced at the tips of interdigitations ) , thus revealing a role for FAK in keeping integrin adhesion structures as discrete contacts . To discover when the defects first appeared , we next examined the MTZs at different stages of development . During pupal stages , the MTZs in all mutants appeared normal ( Figure 1A and Figure 1—figure supplement 1D ) . However , immediately after eclosion , and prior to flight , all mutants showed an MTZ phenotype ( Figure 1A and Figure 1—figure supplement 1D ) . This suggests that FAK , RSU1 and tensin are required either to build or strengthen the attachment so that it can withstand tension during pupal stages ( Weitkunat et al . , 2014 ) . When we compared the phenotypes right after eclosion to 1 day later or 14 days later , we observed that the FAK mutant phenotype did not change , whereas the defects caused by absence of RSU1 or tensin became worse with age , presumably due to muscle activity . These defects were not strong enough to prevent flight ( Figure 1—figure supplement 1C ) . Fewer flies lacking tensin and RSU1 flew than wildtype flies , but this defect was not reproduced by muscle-specific knockdown of tensin or RSU1 , suggesting that reduced flight is more likely caused by the wing blisters in these mutants . As loss of tensin and RSU1 caused strong phenotypes , we hypothesised that their effects may be due to mislocalisation of integrins or other IAPs . We examined the distribution of integrins and other IAPs , using endogenous genes tagged with fluorescent proteins , all of which are fully functional ( see Key Resources Table ) , but did not find any dramatic loss of those examined: βPS integrin , talin , PINCH , paxillin , GIT or vinculin ( Figure 2 , Figure 2—figure supplement 1 ) . However , paxillin was elevated to 250% in the RSU1 mutant and to 150% in the tensin mutant . The IPP components ILK and PINCH were affected differently by the absence of RSU1; ILK was reduced 50% whereas PINCH was unaffected ( Figure 2L ) . In addition , vinculin distribution showed the extreme stretching of the MTZ in the absence of tensin ( Figure 2F ) . Loss of tensin did not alter RSU1 levels or distribution ( Figure 2H ) , nor did loss of RSU1 alter tensin ( Figure 2K ) . To summarize , we have discovered that removing each IAP caused a unique and relatively mild defect in myofibril attachment , none of which resulted in the loss of recruitment of another IAP . To test for compensatory activities between these proteins we examined the consequences of removing two at a time . Flies lacking RSU1 and tensin died at pupal stages , however in these pupae the MTZs appeared normal ( Figure 3—figure supplement 1A ) , suggesting the lethality is caused by defects in another tissue . Therefore , we employed RNAi to specifically knockdown tensin in muscles , using the muscle specific driver mef2-Gal4 . Loss of RSU1 did not worsen the phenotype of tensin knockdown ( Figure 3A and Figure 3—figure supplement 1C ) , demonstrating that these two proteins do not compensate for each other . Combining removal of FAK and RSU1 resulted in the unexpected rescue of the RSU1 phenotype ( Figure 3A ) . This supports a role for FAK in downregulating integrin function ( Ilić et al . , 1995 ) , and the expansion of integrin adhesions ( Figure 1B ) . To confirm that increased integrin activity explains the rescue , we combined RSU1 loss with an integrin β subunit mutant with increased activity . We used the allele mys[b28] , a mutation in the hybrid domain ( V423E ) that strongly elevates affinity for ligand ( Kendall et al . , 2011 ) , and this also rescued the phenotype ( Figure 3A and Figure 3—figure supplement 1C ) . Activated integrin alone did not cause any detectable changes ( Figure 3A and Figure 3—figure supplement 1C ) , allowing us to infer that the MTZ elongation caused by FAK loss is not due to elevating integrin , but rather loss of a positive contribution by FAK to MTZ formation . Consistent with this , activated integrin rescued MTZ elongation caused by the absence of FAK ( Figure 3A and Figure 3—figure supplement 1C ) . This mutant combination also shows that ‘double’ elevation of integrin activity did not cause detectable defects . Thus , in IFMs , FAK has both positive and negative activities . Activated integrin also partially rescued tensin loss , but for this IAP removing FAK did not rescue ( Figure 3A and Figure 3—figure supplement 1C ) . This suggests that tensin normally elevates integrin activity , and this is inhibited by FAK . This predicts that elevating FAK would inhibit tensin further , producing a phenotype similar to loss of tensin . Using the act88F-Gal4 driver to overexpress Fak-GFP , we observed a similar phenotype to loss of tensin ( Figure 3A and Figure 3—figure supplement 1C ) . Of note , this is different than embryonic muscles , where FAK overexpression also causes muscle detachment , but loss of tensin does not ( Grabbe et al . , 2004; Torgler et al . , 2004 ) , and therefore demonstrates that FAK achieves negative regulation of integrins by different mechanisms in larval muscles and IFMs . Furthermore , the finding that activating integrin only partially rescues loss of tensin suggests that tensin does more than activate integrin; consistent with this , neither activated integrin nor loss of FAK rescued the wing blister phenotype of tensin mutants ( data not shown ) . We also examined overlap in function between these IAPs and talin , which is the single IAP to date that is required for all integrin adhesion in Drosophila ( Klapholz and Brown , 2017 ) . Talin reduction ( by removing one gene copy ) substantially enhanced the defects caused by tensin loss , but had no effect on FAK or RSU1 ( Figure 3A and Figure 3—figure supplement 1C ) . Thus , full levels of talin are especially important in the absence of tensin , and the overlap in function between talin and tensin is consistent with both increasing integrin activity ( Georgiadou et al . , 2017; Klapholz and Brown , 2017 ) . We next explored the importance of RSU1 binding to PINCH for its function . Because loss of PINCH is embryonic lethal ( Clark et al . , 2003 ) and because the RSU1 mutant phenotype cannot be explained by a loss of PINCH recruitment ( Figure 2L ) , we instead overexpressed PINCH , anticipating that this may rescue IFM defects since elevating PINCH levels partially rescues hypercontraction mutants in larval and flight muscle ( Pronovost et al . , 2013 ) . Unexpectedly , driving high levels of PINCH alone with the Gal4 system was sufficient to cause defects closely resembling the loss of RSU1 , and combining overexpression of PINCH with loss of RSU1 enhanced this phenotype ( Figure 3A and Figure 3—figure supplement 1C ) . Conversely , reduction of PINCH ( by removing one gene copy ) partially rescued the loss of RSU1 ( Figure 3A and Figure 3—figure supplement 1C ) . This indicates that the function of RSU1 is to inhibit PINCH activity , rather than aid its function . This led to a model ( Figure 3B ) where free PINCH has an important activity that must be carefully regulated and is suppressed by RSU1 binding or integrin activity ( since activated integrin or loss of FAK rescues loss of RSU1 ) . The wing blisters caused by the absence of RSU1 are also rescued by activating integrin or removing FAK ( data not shown ) , suggesting that this phenotype is also caused by loss of PINCH inhibition . To further understand this PINCH activity we performed structure-function analysis using constructs which lacked either the RSU1-binding LIM5 domain and adjacent LIM4 domain ( PINCHΔ4–5 ) or the ILK-binding LIM1 domain ( PINCHΔ1 ) ( Figure 3—figure supplement 1 ) . PINCHΔ4–5 localized indistinguishably to full length PINCH and caused a similar , though weaker , phenotype , ruling out a model where RSU1 inhibits PINCH activity by competing off another protein that binds to LIM5 and is necessary for PINCH activity . PINCHΔ1 no longer caused a phenotype , nor was it tightly localized , showing that ILK-binding is essential for PINCH recruitment and activity , consistent with results in the larval muscles ( Zervas et al . , 2011 ) . Overexpression of ILK did not cause a phenotype ( Figure 3—figure supplement 1 ) , suggesting that it is PINCH that is limiting , and supporting the importance of tightly regulating PINCH activity . Taken together these findings have allowed us to propose a model for how these IAPs contribute to integrin function ( Figure 3B ) . Both talin and tensin contribute to the activity of integrin , with tensin being inhibited by FAK to generate discrete integrin adhesive structures . RSU1 and integrin keep PINCH levels at the correct level; in the absence of RSU1 increased PINCH activity results in fewer , longer interdigitations . These findings highlight the tight balance in IAP activity required to make a wild type adhesive contact . With this improved understanding of the role of IAPs , we examined how vinculin fits into this picture . The defects caused by FAK , RSU1 and tensin are all in the MTZ adhesion structure , close to the membrane . In contrast , loss of vinculin results in disruption to the actin organization up to 25 µm from the muscle ends , with a lack of a clear distinction between the MTZ and the terminal sarcomeres , although overall muscle morphology appeared normal ( Figure 4A ) . This defect was rescued by a transgenic copy of vinculin or muscle muscle-specific expression of vinculin , confirming that the defect was caused by loss of vinculin ( Figure 4—figure supplement 1A and B ) . The vinculin defect arose earlier than those caused by loss of FAK , tensin or RSU1 , as an elongation of the terminal actin block was detectable in pupal stages ( Figure 4A 48 hr APF and Figure 4—figure supplement 1B ) . The extent of actin disruption progressively increased as adult flies aged from 0 to 14 days ( Figure 4A and Figure 4—figure supplement 1B ) . This was quantified by measuring the distance between muscle terminus and the first M-line , rather than MTZ length , because of the loss of a distinct edge of the MTZ . Analysis of the vinculin mutant by TEM ( Figure 4D ) showed defects in addition to the expansion of the MTZ . The even layer of electron dense material at the membrane appears disrupted on the myofibril side of the attachment , but not the tendon cell side . In some areas it is thinner , whereas other regions contain round electron-dense structures , which are also seen throughout the MTZ . One day old flies lacking vinculin were able to fly normally but after 14 days , only 20% were able to fly ( Figure 4B ) . However , muscle specific knockdown of vinculin did not reduce flying ability , yet caused comparable IFM actin defects , indicating that loss of flight in flies lacking vinculin is not due to the IFM defects . A role for vinculin in organizing actin distant from the membrane is consistent with the difference between the distribution of vinculin and other IAPs . Whereas the distribution of the other IAPs looked identical to integrin , vinculin extended up to 1 µm away from the muscle ends ( Figure 2D ) , but not in the tendon cell , consistent with the greater importance of vinculin in maintaining an even layer electron dense material on the myofibril side of the attachment . We confirmed this with increased resolution microscopy , using structured illumination microscopy ( SIM ) ( Figure 5 ) . Whereas tensin and RSU1 colocalized with integrin , vinculin was separate from integrin ( Figure 5C–G ) . GFP on the N-terminus of talin also colocalised with integrin , whereas a more C-terminal insertion of GFP into talin extended 0 . 5–1 µm from the membrane ( Figure 5B ) . Thus , it is feasible for the distal accumulation of vinculin to occur by binding stretched talin . This fits with the stretching of talin in fibroblasts of ~500 nm ( Margadant et al . , 2011 ) , and additional stretching of vinculin . With further experimental perturbation , we were able to divide the defects caused by loss of vinculin into two components , one perturbing adhesion , similar to the defects caused by FAK , RSU1 and tensin , and the second disrupting actin organization . The division was achieved by eliminating the binding of vinculin head to talin , either by deletion of the head domain , or by competitive inhibition through overexpression of a single vinculin-binding-site helix from talin ( Maartens et al . , 2016 ) . Both caused aberrant attachment , but little disruption to actin in adjacent sarcomeres ( Figure 4C and Figure 4—figure supplement 1B ) . In addition , TEM imaging revealed that vinculin tail did not rescue the disruption to electron dense material at the membrane , nor the appearance of aggregates away from the membrane ( Figure 4D ) . Thus , vinculin binding to talin contributes to integrin-mediated adhesion , and surprisingly the tail domain could function on its own to keep actin organized . To our knowledge this is the first demonstration that the tail domain can function independently , and it is also significant that the tail alone gets recruited to the MTZ ( Figure 4C ) . It was difficult to discern how the extended actin defect arose in the absence of vinculin , as it was not clear which actin structure became disorganized , whether: ( 1 ) MTZ actin expands , pushing the first sarcomere away; ( 2 ) MTZ actin mixes with sarcomeric actin; or ( 3 ) overall actin organization near muscle ends deteriorates . To resolve between these possibilities , we examined actin-binding proteins and components of muscle sarcomeres , which revealed three major insights . First , the distribution of the M-line component Unc-89/obscurin ruled out model 1 of MTZ pushing the first sarcomere away , as M-lines were visible within the region of disrupted actin ( Figure 6H ) . Second , using super-resolution microscopy ( Figure 5 ) and standard confocal microscopy ( Figure 5—figure supplement 1 ) we were able to divide the MTZ into four distinct regions based on their composition , which appear analogous to the different regions discovered by super-resolution in the Z-axis of focal adhesions ( Kanchanawong et al . , 2010 ) . The four regions are: ( 1 ) the adhesion structure at the membrane , analogous to the integrin signalling layer ( ISL ) in focal adhesions , which contains most IAPs ( N-terminus of talin , paxillin , GIT , ILK , PINCH , RSU1 , FAK and tensin ) ; ( 2 ) the next layer , analogous to the force transduction layer ( FTL ) , containing actin , the C-terminus of talin and vinculin; ( 3 ) an unexpected novel , actin-rich region , which appears analogous to the actin regulatory layer , but instead of containing Ena/VASP and zyxin , contains filamin , Arp3 , and the Arp2/3 activator WASH , as well as the C-terminus of talin and vinculin . Because of these differences , we term this novel layer the muscle actin regulatory layer or MARL; and ( 4 ) the first Z-line with high levels of ZASP and α-actinin , analagous to the stress fiber attached to the focal adhesion ( Figure 5A and B ) . The segregation of these proteins is not absolute , as ZASP and α-actinin were present in both the FTL and MARL , and in addition components from the MARL were detected at low levels in all Z-lines ( Figure 5—figure supplement 1 ) . The third key finding is that the MARL is expanded in the absence of vinculin ( e . g . filamin , ZASP ) , whereas the ISL ( e . g . paxillin ) was not substantially altered ( Figures 6B , D , F , J and Figure 6—figure supplement 1B ) . In addition , filamin levels were reduced to 40% in the absence of vinculin ( Figure 6J ) . Thus , these findings favour explanation 2 , where it is the MARL that expands into the sarcomeres , and therefore that vinculin has a role in the formation and/or stability of the MARL , consistent with the function of the tail domain in this region . The MARL is distinguished by two actin-binding proteins associated with branched actin networks , Arp2/3 and filamin . Actin within the FTL is likely to link the MARL to the membrane , whereas the first Z-line assembles at the outer edge of the MARL , where it anchors the parallel actin filaments of the sarcomere . This highlights the ability of actin-binding proteins to associate with discrete actin structures in adjacent parts of the cell . Vertebrate muscles contain filamin-C at the myotendinous junction ( van der Ven et al . , 2000 ) , suggesting they have a comparable structure . Our next goal was to discover how vinculin contributes to the formation and function of the MARL . The concentration of Arp2/3 , its nucleation promoting factor WASH , and filamin in the MARL suggested that the actin in this structure may form a cross-linked network , and we sought to determine whether these proteins contribute to MARL actin assembly . Flies lacking filamin are viable , albeit female sterile due to defects in the actin-based ring canals during oogenesis ( Robinson et al . , 1997 ) . Loss of filamin ( using a deletion of the C-terminal half [Huelsmann et al . , 2016] ) did not alter overall muscle shape , but caused reduced MARL size in pupal stages and just after eclosion ( Figure 7A and Figure 7—figure supplement 1B and D ) , which converted to MARL expansion 1 day after eclosion and thereafter ( Figure 7A and Figure 7—figure supplement 1B ) , resembling the vinculin mutant phenotype . The reduction in the MARL is even more dramatic when visualized by TEM ( Figure 7B ) . A flight defect was observed in flies lacking filamin overall and when specifically knocked down in muscles ( Figure 7C ) , consistent with previous observations ( González-Morales et al . , 2017 ) . Muscle-specific knockdown of Arp2/3 components is lethal and results in severe muscle defects ( Schnorrer et al . , 2010 ) . However , flies lacking the Arp2/3 nucleation promotion factor WASH are viable ( Nagel et al . , 2017 ) , and we found they had no gross changes to muscle morphology , had a smaller MARL during pupal stages , and a progressively elongated MARL during adult stages , similar to loss of filamin ( Figure 7A and Figure 7—figure supplement 1B and D ) . The F-actin staining in the MARL was reduced ( Figure 7A ) in the absence of WASH , as were levels of filamin ( Figure 7—figure supplement 1E and F ) . TEM confirmed that the MARL was less dense in MTZs lacking WASH and showed defects in the morphology of the interdigitations ( Figure 7B ) . Muscle-specific RNAi of filamin and WASH confirmed that this phenotype is due to loss of these proteins in the muscles ( Figure 7—figure supplement 1A and B ) . In contrast to filamin , flies lacking WASH flew normally ( Figure 7C ) , suggesting it is the defects in sarcomere structure ( González-Morales et al . , 2017 ) rather than defects at the MTZ that impair flight in the absence of filamin . We were intrigued by the observation that two proteins that aid MARL formation , vinculin and filamin , are mechanosensitive ( Rognoni et al . , 2012; Atherton et al . , 2016; Huelsmann et al . , 2016 ) . We therefore wondered whether mechanical signaling might be required to build the MARL . To test the role of the mechanosensing region of filamin , we examined the IFM defects caused by mutations that either delete it ( ∆MSR ) , make it harder to open ( closed ) , or make it easier to open ( open ) ( Huelsmann et al . , 2016 ) . The ∆MSR and closed mutant had similar defects as deletion of the C-terminus ( Figure 7D ) , although they were not as strong , indicating that mechanosensing contributes the majority , but not all , of filamin activity in forming the MARL . The open filamin did not cause any MTZ defects , but we observed a gain of function phenotype , consisting of the formation of bright , ectopic bands of actin , usually associated with a Z-line ( Figure 7D ) . Similar ectopic actin bands , termed ‘zebra bodies’ , occur with missense mutations in the troponin gene ( Nongthomba et al . , 2007 ) or knockdown of muscle components ( Schnorrer et al . , 2010 ) . Overexpression of wild type vinculin or vinculin tail also produced zebra bodies , usually at the second-most terminal Z-line ( Figure 7—figure supplement 1G and Figure 4B ) . Overexpression of WASH was sufficient to enlarge the MARL , with an apparent increase in actin levels ( Figure 8A and Figure 7—figure supplement 1B ) . We further manipulated MARL proteins to determine how they worked together . None of the double mutant combinations of vinculin , filamin and WASH showed enhancement or suppression of the expected additive phenotype ( Figure 8A and Figure 7—figure supplement 1B and C ) . As these are all null alleles , this shows they are all required in the same pathway to make the MARL . It is also of interest that the reduced MARL in the absence of WASH does not result in a reduction in the expanded actin caused by loss of vinculin or filamin . Open filamin partially rescued the loss of vinculin ( Figure 8A and Figure 7—figure supplement 1C ) , showing that a key function of vinculin is to open filamin . In contrast , open filamin did not rescue the absence of WASH , suggesting that WASH is downstream of filamin . Overexpression of WASH did not rescue the loss of vinculin or filamin and instead made the loss of filamin phenotype worse ( Figure 8A and Figure 7—figure supplement B and C ) ; reciprocally , filamin does not appear to be required for overexpressed WASH to expand the MARL , judging by the elevated actin staining at the termini of the disrupted actin . Thus , the aberrant MARL that forms in the absence of filamin appears expanded by additional WASH . Consistent with filamin being downstream of vinculin , vinculin tail-induced zebra bodies contained filamin , and did not form when only closed filamin was present ( Figure 7—figure supplement 1G ) , whereas open filamin still made zebra bodies in the absence of vinculin ( Figure 7—figure supplement 1G ) . The genetic interactions suggested that vinculin tail and filamin may interact directly or indirectly . To test this , we targeted a portion of filamin to the surface of the mitochondria with an ActA peptide ( Bubeck et al . , 1997 ) to test whether it could recruit vinculin tail within IFMs . We tagged the short version of filamin ( filamin90 , containing the C-terminal 9 filamin repeats , including the 6 that form the MSR and dimerization repeat ) , both in open and closed form , with RFP and the ActA mitochondrial targeting sequence . When expressed in the IFMs , the majority of the fusion protein formed clusters on the mitochondrial surface , but some was still recruited to the MARL ( Figure 8B ) . All of the clusters formed by open filamin90 recruited large amounts of actin , whereas only a fraction of the closed filamin90 clusters did so ( Figure 8B ) . Both were able to recruit vinculin tail ( Figure 8C ) . Furthermore , the recruitment of these fusion proteins to the MARL required vinculin ( Figure 8E ) . Overexpression of WASH resulted enlarged the clusters formed by both closed and open filamin 90 ( Figure 8D and not shown ) , with the filamin surrounding a sphere of actin and WASH-GFP . Thus , the C-terminus of filamin is able to recruit vinculin , actin and WASH to an ectopic location . We can put these results together into a speculative model ( Figure 8F ) whereby formation of the MARL involves a series of mechanotransduction events: the first step is the mechanical stretching of talin , which will stabilize the open conformation of vinculin and expose the actin-binding tail; in the second step , the vinculin tail anchors the C-terminus of filamin to actin filaments , thus generating three actin contact points for the filamin dimer and permitting it to sense stretch in a more dense actin meshwork; in the third step , forces on the actin meshwork linked by filamin and vinculin leads to opening up of the filamin MSR , which then activates WASH , promoting Arp2/3-mediated formation of new actin branches , thus expanding the MARL . The MTZ revealed both positive and inhibitory actions of FAK , with the latter consistent with the role of FAK in adhesion disassembly ( Ilić et al . , 1995 ) . Both loss of FAK and activated integrin supressed the phenotypes caused by loss of RSU1 or vinculin , but only activated integrin alleviated the defects caused by the absence of tensin , suggesting that FAK inhibition requires tensin activity , and in turn , tensin elevates integrin activity . This fits with the recent discovery that tensin contributes to the inside-out activation of integrins via talin ( Georgiadou et al . , 2017 ) . FAK and tensin thus form a balanced cassette that we imagine responds to upstream signals to regulate integrin activity . Further work is needed to discover how tensin increases integrin activity , how this is inhibited by FAK , and what signals control this regulatory cassette . One model would have tensin activating integrin by direct binding to the β subunit cytoplasmic tail , and FAK inhibition by phosphorylation of tensin , but an alternative is that they have antagonistic roles in integrin recycling ( e . g . Rainero et al . , 2015 ) . RSU1 is part of the complex containing ILK , PINCH and Parvin ( IPP complex ) , and binds the 5th LIM domain of PINCH ( Kadrmas et al . , 2004 ) . Loss of RSU1 causes milder phenotypes than loss of ILK , PINCH or parvin , and these phenotypes have previously been interpreted as a partial loss of IPP activity . Our findings indicate that the phenotypes observed in the absence of RSU1 are due to too much PINCH activity , and therefore the role of RSU1 is to keep PINCH activity in check . This suggests that PINCH is perhaps the key player of the IPP complex , and is recruited to adhesions by integrin via ILK , and kept in check by integrin and RSU1 . The importance of regulating active PINCH levels is consistent with the dosage sensitivity of PINCH: reducing PINCH partially rescues the dorsal closure defect in embryos lacking the MAPK Misshapen ( Kadrmas et al . , 2004 ) , and elevating PINCH rescues hypercontraction caused by loss of Myosin II phosphatase ( Pronovost et al . , 2013 ) . Reducing the interaction of PINCH with ILK had unexpectedly no phenotype , but in combination with the loss of RSU1 becomes lethal ( Elias et al . , 2012 ) ; the lethality can now be interpreted as being caused by too much PINCH activity , rather than too little . Excess ‘free’ PINCH results in elongated membrane interdigitations and elevated paxillin levels . This suggests that PINCH has an important role at the cell cortex , consistent with cortical proteins in the PINCH interactome ( Karaköse et al . , 2015 ) . Too much parvin activity also causes lethality , which is suppressed by elevating ILK levels ( Chountala et al . , 2012 ) . Thus , it is increasingly clear that the functions of IPP components need to be tightly controlled . We gained some insight into how RSU1 inhibits PINCH activity by demonstrating that ∆LIM4 , 5 PINCH still caused longer interdigitations . This rules out RSU1 blocking the binding of another protein from binding LIM5 , and suggests instead that RSU1 bound to LIM5 must be inhibiting the activity of LIM1-3 . Vinculin has a dual function in the MTZ: its head domain promotes FTL stability via binding talin , and its tail promotes MARL formation . Our analysis of the vinculin mutant by electron microscopy showed a phenotype within the electron dense layer close to the membrane that we presume corresponds to the integrin signalling layer . It suggests that vinculin may mediate interactions between IAPs that aid in keeping this as an even layer . The fact that the disruption to this layer is only evident on the muscle side of the interaction raises the question of how similar the integrin junctions are on the two sides of this cell-cell interaction via an intervening ECM . Many other sites of integrin-mediated adhesion in Drosophila involve integrins on both sides of the interaction ( Bökel and Brown , 2002 ) and by electron microscopy the electron dense material looks similar on the two sides ( Reedy and Beall , 1993 ) , and we would expect that both sides need to resist the same forces . Even with structured illumination microscopy we cannot resolve the two sides of the membrane , but our results show that the C-terminus of talin and vinculin are not pulled away from the membrane in the adult tendon cells . This suggests either that vinculin has a different role in the tendon cell , with a different configuration , as we observed for talin in the pupal wing ( Klapholz et al . , 2015 ) , or it is absent . The vinculin tail function in MARL formation does not require that vinculin is bound to talin , but we suspect that in the wild type it is talin-binding that converts vinculin into an open conformation , permitting the tail to trigger MARL formation with filamin , as outlined in our working model in Figure 8F . A key function of vinculin tail in the MARL is to aid the mechanical opening of the filamin mechanosensitive region . We presented evidence suggesting this is achieved by the vinculin tail anchoring the C-terminus to actin , but further work is required to determine if there is direct binding between the two proteins . Similarly , our results indicate that the Arp2/3 nucleation promoting factor WASH is part of the same pathway as filamin and acts downstream of it , but the connection between the two has yet to be resolved . This new function for WASH is distinct from its best characterized role regulating actin on intracellular vesicles during endosomal sorting and recycling ( Nagel et al . , 2017 ) , but WASH also has additional roles in the nucleus and the oocyte cortex ( Verboon et al . , 2018 , 2015 ) , showing that it is a versatile protein . Given the myofibril defects seen with loss of RSU1 , tensin , vinculin and filamin it might be expected that mutations in genes encoding these IAPs might be implicated in muscle disease . Indeed , mutations in integrin α7 , talin and ILK are associated with muscular myopathies in humans and mice ( Winograd-Katz et al . , 2014 ) . Mutations in the genes encoding RSU1 , tensin and vinculin have not been linked to muscle myopathies , but mutations in filamin are linked to myofibrillar myopathies ( Vorgerd et al . , 2005 ) . However , given the subtlety of these defects in Drosophila , one might predict that mutations in genes encoding these IAPs are associated with subtle defects in humans such as reduced sporting performance or susceptibility to muscle injury . We are not aware of any mutations in genes encoding these IAPs being related to athletic performance or injury susceptibility ( Maffulli et al . , 2013 ) , but we expect that these IAPs would be good candidates for further study in this area . One way that these IAPs may contribute to athletic performance is by building a muscle shock absorber , the MARL , which protects the myofibrils from contraction-induced damage . The concept of muscle shock absorbers is well established since tendons perform this function ( Roberts and Azizi , 2010 ) . The presence of filamin , Arp3 , vinculin and α-actinin in the MARL suggests that the MARL contains branched and bundled actin filaments . Branched actin networks have been shown to be viscoelastic ( Blanchoin et al . , 2014 ) and actin crosslinkers such as filamin have been shown to reduce viscosity and increase elasticity of actin networks ( Koenderink et al . , 2009 ) . Further study into the functional nature of the MARL should increase our understanding of athletic performance and injury susceptibility . Flies were grown and maintained on food consisting of the following ingredients for 20 litres of food: 150 g agar , 1100 g glucose , 700 g wheat flour , 1000 g dry yeast , 500 ml nipagin 10% , 80 ml proprionic acid and 200 ml penicillin/streptomycin . Animals of both sexes were used for this study with the exception of vinculin-; mef2 >vinculinTail RFP where only males were used . Expression of UAS::vinculin-RFP and UAS::vinculinTail-GFP ( Maartens et al . , 2016 ) were expressed in muscles with P{Gal4-Mef2 . R}3 ( Bloomington Drosophila stock centre , 54591 ) . UAS::vinculinHead-RFP ( Maartens et al . , 2016 ) was expressed specifically in the IFMs using Gal4-act88f ( Bryantsev et al . , 2012 ) , since expression with Gal4-mef2 is lethal . Flight assays were performed as previously described ( Weitkunat and Schnorrer , 2014 ) . Briefly , flies were dropped into plastic tube ( 1 meter long and 15 cm wide ) coated with a layer of oil , with water at the bottom . Flightless flies were counted as those that fall directly in the water and flying flies were counted as those that stick to the sides of the tube . 30 flies per condition were tested and percentage of flightless flies was calculated . To ensure we had a null allele , a new ics mutant allele was generated by excision of the P-element insertion P{GT1}icsBG02577 , which is inserted 129 bp downstream of the translational start site . Two hundred single males of the genotype w; P{GT1}icsBG02577/CyO; mus309N1/mus309D2 Sb P{∆2 , 3}99B were crossed in individual vials to w; Sco/CyO; Dr/TM6 virgin females . From the progeny of each cross , three white-eyed excision males were individually crossed to w; Gla/CyO virgin females . Deletions were identified by PCR and characterized by sequencing . The allele ics2 is a 1775 bp deletion which deletes 781 bp upstream and 994 bp downstream of the P-element insertion site , deleting the first 165 of 283 RSU1 residues . βPS-mcherry was generated by CRISPR . A plasmid containing homology arms of mys ( Drosophila gene encoding βPS ) for the generation of βPS-GFP ( Klapholz et al . , 2015 ) , was used as a template to generate a homology plasmid for the generation of βPS-mcherry . An 8502 bp fragment containing 4465 pb of the C-terminal coding region of mys , GFP and 3894 bp of downstream non-coding sequence was cloned into pBluescript II KS ( addgene ) using Not1 and Xba1 . GFP was excised by digestion with Hind3 and Sac1 and replaced with mcherry . A guide RNA targeting the C-terminus of βPS was cloned into the pCFD3 plasmid as previously described ( Port et al . , 2014 ) . The sequence of the oligonucleotide used to generate the guide RNA were 5’-GTCGCTTCAAGAACCCCATGTATG-3’ and 5’-AAACCATACATGGGGTTCTTGAAG-3’ . The homology plasmid and the guide RNA plasmid were co-injected into y1 P{nos-cas9 , w+} M{3xP3-RFP . attP}ZH-2A w* embryos . Injected adults were crossed to FM7 and the progeny were screened at larval stages by fluorescence microscopy for the presence of βPS-mcherry . BACR13D24 ( Berkeley Drosophila genome project ) containing the genomic region of Drosophila zyxin was used as a template . A 9865 bp fragment containing the zyxin genomic region was excised from the BAC using Spe1 and Not1 and cloned into pBluescript II KS ( addgene ) . A 1601bp region 5’ to the Spe1 site was PCR amplified from Drosophila genomic DNA and cloned into TOPO TA ( Thermo Fisher Scientific ) . The forward primer added an Mlu1 site and had the sequence 5’-ACGCGTGGGAATAGCAAACGCCACA-3’ , while the reverse primer was 5’-TGTGTACTTGCGCATTCACA-3’ . pBluescript II KS containing the GFP sequence was used a template for GFP . A 738 bp product containing GFP was PCR amplified and cloned into TOPO TA . An Aat2 site was added with the forward primer 5’-GACGTCAGAACATTCGAGCTCATCGATGAGTAAAGGA-3’ and an Fsp1 site was added with the reverse primer 5’ TGCGCAATAAATAAAATGAGCACTCAATTTATTTGTATAGTTCATCCATGC-3’ . GFP was cloned into the 3’ of the zyxin genomic sequence at Fsp1 and Aat2 sites . The resulting 10 , 603 bp zyxin-GFP sequence was excised using Spe1 and Not1 , and the 1601 bp 5’ region was excised from TOPO TA using Mlu1 and Spe1 , and both were cloned into Mlu1 and Not1 sites of the vector pWhiteAttPRabbit ( Brown , N and Klapholz , B unpublished ) . The vector contains attB sites and the white gene . The constructs were inserted into position 51 on chromosome II . A wild type ics genomic rescue construct tagged with GFP , RSU1-GFP , was prepared by subcloning a 12 , 888 bp AatII to NotI genomic DNA fragment containing 6853 bp upstream of the translational start site of the ics locus into the targeted P-element transformation vector pWhiteAttPRabbit . To fuse GFP to the COOH terminus of RSU1 , XhoI and HindIII sites were inserted between the last codon and the stop codon of the ics gene , and at the NH2 and COOH termini of GFP . A short linker amino acid sequence was introduced at the NH2 terminus of GFP so that the fusion protein junction corresponds to RAYSSSSMSK , with the linker underlined . The constructs were inserted into position 51 on chromosome II and recombined with ics2 . Using the UAS-PINCH-GFP construct described in Zervas et al . , 2011 we deleted LIM domains 4 and 5 ( residues 198–317 of isoform Stck-PA ) and generated transgenic lines by P-element mediated tranformation . Gibson cloning was used to clone a 2756 bp genomic fragment containing filamin90 and RFPmito into pUASP-attB ( Drosophila Genomics Resource center ) . RFPmito was amplified from MrRFPmito plasmid ( Maartens et al . , 2016 ) using primers 5’-ccagatcgatgtcTCGAGCTCCAGCATGGTG-3’ and 5’- ttaacgttaacgttcgaggtcgactctagaACTAGTTGTTCTTGCGCAGTTG-3’ and filamin90 was amplified from pGE-attB-filaminopen-GFP or pGE-attB-filaminclosed-GFP ( Huelsmann et al . , 2016 ) using 5’- gtacccgcccggggatcagatccgcggccgcGGCCGCAAAATGACTACG-3’ and 5’- tgctggagctcgaGACATCGATCTGGAATGG-3’ . Fragments were incubated with pUASP-attB linearize with Xba1 and Not1 . IFMs were dissected from pupae and adult flies as previously described ( Weitkunat and Schnorrer , 2014 ) . Briefly pupal thoraces were dissected in PBS and fixed in 4% formaldehyde in PBS + 0 . 3% Triton X-100 ( PBST ) for 15 min . Adult thoraces were dissected in PBS and fixed in 4% formaldehyde in PBS for 1 hr . Thoraces were washed in PBST and dissected further as required . Thoraces were blocked in PBST +0 . 5% BSA and incubated in primary antibody overnight at 4°C . Primary antibodies used were rabbit anti-obscurin ( Burkart et al . , 2007 ) diluted 1:100 , rat anti-filamin C-terminus ( Sokol and Cooley , 1999 ) diluted 1:250 and rabbit anti-arp3 diluted 1:500 ( Stevenson et al . , 2002 ) . Thoraces were incubated with anti-rabbit-Alex Fluor 488 ( Abcam ) if needed and phalloidin conjugated with rhodamine ( Thermo Fisher Scientific ) , Dylight650 ( Cell Signaling Technology ) or Alex Fluor 488 ( Thermo Fisher Scientific ) . IFMs were further dissected and mounted in Vectashield ( VectorLabs ) . Confocal images were taken with an Olympus FV1000 confocal microscope using a 60×/1 . 35 NA objective and 3x zoom . A single stack of 5 z-sections spaced by 0 . 5 µm was imaged per individual . SIM images were taken with a DeltaVision OMX BLAZE V3 ( Applied Precision ) using an Olympus 10 × 0 . 4 NA air objective or 60 × 1 . 4 NA oil objective , 488 nm and 593 nm laser illumination and standard excitation and emission filter sets . Raw data was reconstructed using softWoRx 6 . 0 ( Applied Precision ) software . Images were processed using ImageJ ( NIH ) and Adobe Photoshop . Half thoraces were dissected from 1 day old flies in PBS and fixed in glutaraldehyde/2% formaldehyde in 0 . 05 M sodium cacodylate buffer pH 7 . 4 for 4 hr at 4°C . Thoraces were osmicated in 1% OsO4/1 . 5% potassium ferricyanide/0 . 05 M NaCAC pH 7 . 4 overnight at 4°C and then treated with 0 . 1% thiocarbohydrazide/DIW for 20 min . Thoraces were osmicated again in 2% OsO4 for 1 hr and finally embedded in Quetol resin . 70 nm sections were imaged on a Tecnai G2 80-200kv transmission electron microscope at 6500x magnification . Phenotypes were analysed manually using ImageJ . The line tool was used to measure MTZ length , myofibril width , sarcomere length and the distance from the myofibril end to the first M-line . 10–30 myofibrils per individual were measured from 10 individuals per genotype . Average MTZ length of distance to first M-line was calculated per individual . Statistical significance was determined by Mann Whitney U test ( p<0 . 05 ) using Prism software ( GraphPad ) . Fluorescence intensity plots for SIM images were using the plot profile function in ImageJ ( NIH ) . Profiles were processed and distance between peaks was calculated in Microsoft Excel . Fluorescence intensity of GFP tagged proteins at the MTZ were quantified using a Matlab script adapted from Bulgakova et al . ( 2017 ) . Briefly , all objects in each frame were detected by a series of dilation , hole filling and eroding . The resulting objects were filtered by their area , eccentricity and orientation ( more than 45° to long image axis ) to exclude all objects that did not represent MTZs . MTZ minor axis length and mean intensities of the resulting objects were collected from original non-modified confocal frames and averaged . Mean intensity was then multiplied by the mean axis length to correct for differences in the MTZ length in different mutants . Statistical significance was determined by Mann Whitney U test ( p<0 . 05 ) using Prism software ( GraphPad ) . Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact , Nick Brown ( nb117@cam . ac . uk ) .	Our body consists of many different types of cells that build our tissues and organs . To do so , cells need to be able to stick together . One family of proteins called integrins helps to keep cells connected . They sit across cell membranes and anchor cells to the networks of protein fibres outside cells that link and strengthen our organs , and also connect muscles to tendons . In fruit flies , the indirect flight muscle attach to the thorax of the insect , and create wing movements by ‘deforming’ the thorax . These flight muscles resemble the muscles of animals with a backbone , and consist of many different fibres . At the end of these fibres is a plaque of a protein important for muscle contraction , known as actin . Integrins attach to these actin plaques , allowing the ends of the muscle to anchor to the tendon . Integrins form complexes with so-called 'integrin-associated proteins' inside the cell , which regulate integrin . Integrins and integrin-associated proteins are essential for proper muscle development , but until now it was not fully understood how they interact with each other . Here , Green et al . explored the role of some of these proteins in the indirect flight muscles of fruit flies . This revealed that the connection between muscle and tendon is a balancing act . Some integrin associated proteins boost the attachment , whilst others block it . One protein , tensin , increased integrin attachment , whilst another , FAK , blocked tensin , decreasing attachment . Similarly , a protein called PINCH expands the attachment , whilst a protein called RSU1 reduced the activity of PINCH to the correct level . Moreover , the end point of the muscle fibres was discovered to be composed of four distinct layers , including a newly identified layer of actin , which is built by three other integrin-associated proteins . The flight muscles of fruit flies are similar in structure to the skeletal muscles that move our own limbs . An important next step is to discover whether these integrin-associated proteins work similarly in our muscles . A better understanding of how they work together could help with research into diseases of the muscles .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2018	Novel functions for integrin-associated proteins revealed by analysis of myofibril attachment in Drosophila
Homotypic or entotic cell-in-cell invasion is an integrin-independent process observed in carcinoma cells exposed during conditions of low adhesion such as in exudates of malignant disease . Although active cell-in-cell invasion depends on RhoA and actin , the precise mechanism as well as the underlying actin structures and assembly factors driving the process are unknown . Furthermore , whether specific cell surface receptors trigger entotic invasion in a signal-dependent fashion has not been investigated . In this study , we identify the G-protein-coupled LPA receptor 2 ( LPAR2 ) as a signal transducer specifically required for the actively invading cell during entosis . We find that G12/13 and PDZ-RhoGEF are required for entotic invasion , which is driven by blebbing and a uropod-like actin structure at the rear of the invading cell . Finally , we provide evidence for an involvement of the RhoA-regulated formin Dia1 for entosis downstream of LPAR2 . Thus , we delineate a signaling process that regulates actin dynamics during cell-in-cell invasion . Entosis has been described as a specialized form of homotypic cell-in-cell invasion in which one cell actively crawls into another ( Overholtzer et al . , 2007 ) . Frequently , this occurs between tumor cells such as breast , cervical , or colon carcinoma cells and can be triggered by matrix detachment ( Overholtzer et al . , 2007 ) , suggesting that loss of integrin-mediated adhesion may promote cell-in-cell invasion . This is further supported by the fact that homotypic cell-in-cell structures can be regularly found when tumor cells are released into fluid exudates such as ascites or during pleural carcinosis ( Overholtzer and Brugge , 2008 ) . Although the consequence of entotic invasion is not well understood , the process may contribute to tumor progression by inducing aneuploidy in human cancers ( Krajcovic et al . , 2011 ) . The ultimate outcome of an entotic event also depends on the fate of the invaded cell , which can remain viable or even divide inside or escape from the host cell or undergo vacuolar degradation ( Florey et al . , 2010; Krajcovic et al . , 2011 ) . It was previously shown that for a cell to invade into a neighboring cell Rho-dependent signaling and actin are required ( Overholtzer et al . , 2007 ) . However , potential extracellular ligands or cell surface receptors involved in this migratory process are entirely unknown . Furthermore , what type of actin structures and which actin polymerization factor triggers active cell-in-cell invasion in a signal-regulated fashion remained unclear . In this study , we investigated actin-mediated entotic invasion and delineate a signaling pathway downstream of the LPAR2 that ultimately targets the formin mDia1 for polarized actin assembly at the rear of the invading cell to drive cell-in-cell invasion . To monitor actin assembly during life cell-in-cell invasion over time , we generated MCF10A cells expressing either mCherry- or GFP-LifeAct . Red and green LifeAct-cell populations were mixed and plated on top of polyHEMA-coated coverslips to prevent matrix adhesion and to induce entotic incidences . Under these conditions , cell-in-cell invasion was confirmed to require ROCK as assessed using the ROCK-inhibitor Y-27632 ( Video 1 ) ( Overholtzer et al . , 2007 ) . Interestingly , imaging LifeAct-expressing cells over time , we consistently observed that specifically the actively invading cell displayed extensive blebbing early on during invasion followed by the formation of an actin-rich uropod-like structure at the rear of the invading cell ( Figure 1A; Video 2 ) . Plasma membrane blebbing was highly dynamic under these cell culture conditions with a bleb cycle lasting about 2 min ( Figure 1B; Video 3 ) and the total number of blebs ranged from 60 to 100 blebs per cell depending on the MCF10A cell size . Notably , blebbing is a frequently observed phenomenon during amoeboid or rounded cancer cell invasion through 3-dimensional collagen requiring ROCK-dependent contractility ( Sahai and Marshall , 2003; Kitzing et al . , 2007 ) . The presence of the polarized actin-rich cup at the rear of the entosing cell could be confirmed using phalloidin staining to visualize endogenous actin filaments ( Figure 1C ) or by confocal microscopy of LifeAct-expressing MCF10A cells ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 02786 . 003Video 1 . ROCK activity is required for entosis . Comparison between control and Y27632 ( 5 μM ) -treated MCF10A cells cultured on polyHEMA demonstrating the requirement of ROCK for cell blebbing and cell-in-cell invasion . Time ( min ) is indicated in the upper right corner . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 00310 . 7554/eLife . 02786 . 004Figure 1 . Actin dynamics during entotic invasion and stimulation of entosis by LPA . ( A and A′ ) MCF10A cells expressing LifeAct-mCherry ( red ) or LifeAct-GFP ( green ) were monitored over time ( Video 2 ) as indicated to visualize actin polymerization during cell-in-cell invasion . Note the specific blebbing activity of the invading cell and the actin-rich structure at the cell rear ( green ) . Differential interference contrast ( DIC ) is added for each frame . ( B ) Bleb-dynamics were analyzed from eight different live cells expressing LifeAct-GFP , ± SD . ( C ) Entotic MCF10A cells labeled for F-actin using phalloidin ( red ) and nuclei using DAPI ( blue ) . Scale bar 5 μm . ( D ) Increasing concentrations of LPA stimulate entosis in MCF10A cells under serum-free conditions . ( n = 4 ± SD ) . ( E ) Effects of adding the LPAR1 , 2 and 3 receptor blocker Ki16425 on LPA-induced entosis in MCF10A cells ( n = 3 ± SEM analyzed by one way ANOVA followed by Dunnett's post-tests compared with LPA-induced group ) . ( F ) Effects of adding the LPAR1 , 2 and 3 receptor blocker Ki16425 on entosis in MCF10A cells after addition of 5% horse serum ( n = 4 ± SEM analyzed by one way ANOVA followed by Dunnett's post-tests compared with serum-induced group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 00410 . 7554/eLife . 02786 . 005Figure 1—figure supplement 1 . Formation of an actin-rich uorpod-like structure during entotic invasion . MCF10A cells cultured on poly-Hema expressing LifeAct-mCherry ( red ) or LifeAct-GFP ( green ) were monitored over time as indicated to visualize actin polymerization during cell-in-cell invasion . Each time frame represents a confocal scan using a LSM 700 ( Zeiss ) . Differential interference contrast ( DIC ) is added for each frame . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 00510 . 7554/eLife . 02786 . 006Video 2 . Actin dynamics during entotic invasion . MCF10A cells expressing LifeAct-mCherry ( red ) or LifeAct-GFP ( green ) were monitored over time as indicated ( upper right corner ) to visualize actin dynamics during cell-in-cell invasion . Video corresponds to Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 00610 . 7554/eLife . 02786 . 007Video 3 . mDia1 is required for entotic blebbing . MCF10A cells cultured on polyHEMA expressing LifeAct-mCherry ( red ) or LifeAct-GFP ( green ) silenced for mDia1 or control respectively were monitored over time as indicated to visualize actin dynamics during blebbing . Video corresponds to Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 007 We noticed that high fetal calf serum ( FCS ) concentrations enhance entosis ( not shown ) . A ligand known to be present at micromolar concentrations in FCS is lysophosphatidic acid ( LPA ) . Thus , we speculated that LPA might trigger entosis . Indeed , under serum-free conditions addition of LPA efficiently stimulated entotic events of MCF10A cells in a concentration-dependent manner already at nanomolar concentrations ( Figure 1D ) . This ligand mediated cell-in-cell invasion was dependent on cell surface receptor activity since the LPA-receptor 1 , 2 and 3 inhibitor Ki16425 completely blocked LPA-stimulated entosis ( Figure 1E ) . Comparable results were obtained when entosis was triggered by serum ( Figure 1F ) . These data identify the soluble extracellular ligand and serum component LPA as a mediator of cell-in-cell invasion . LPA transduces its multiple cellular effects via binding to specific LPA-receptors , which belong to the large superfamily of G-protein-coupled receptors ( GPCRs ) . As there are several different LPA-receptors present in human tissues ( Choi et al . , 2010 ) , we set out to identify the receptor responsible for entotic invasion using an siRNA approach . Interestingly , silencing of LPAR2 resulted in a robust and significant reduction of entotic events , while LPAR5 suppression moderately affected entosis ( Figure 2A ) . 10 . 7554/eLife . 02786 . 008Figure 2 . LPAR2 triggers invasive motility during entosis . ( A ) MCF10A cells treated with indicated siRNAs for 48 hr were analyzed for entosis ( n = 3 ± SD analyzed by one way ANOVA followed by Dunnett's post-tests compared with siMOCK group ) . ( B ) MCF10A cells stably expressing mCherry-H2B or GFP-H2B were treated with indicated siRNAs before equal cell numbers were mixed and plated to analyze entotic invasion . ( C ) HEK293 cells were transfected with LPAR2 cDNA before analyzation for entosis ( n = 3 ± SD , p<0 . 05 , t test ) . ( D ) Immunolabeling of endogenous LPAR2 ( red ) and nuclei ( DAPI ) of MCF10A cells fixed at different stages during entosis as indicated . Scale bar 5 μm . ( E ) Immunolabeling of transfected Flag-tagged LPAR2 ( green ) , F-actin ( phalloidin , red ) , and nuclei ( DAPI ) of invading HEK293 cells undergoing entosis with or without 5 min addition of 100 nM latrunculin B ( LatB ) before fixation . Arrows indicate disassembled F-actin . Scale bar 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 008 To investigate whether LPAR2 is specifically required for the actively invading cell and not for the host cell or both , we applied a two-color entosis assay by stably expressing either GFP- or mCherry-H2B and treated each cell population with siRNA against LPAR2 . One phenotypic hallmark characterizing the host cell from the invading cell during cell-in-cell invasion is the typically half-moon-shaped nucleus ( Figure 1C; Overholtzer and Brugge , 2008 ) . Examination of entotic events using confocal fluorescence microscopy revealed that only cells silenced for LPAR2 failed to actively invade into another , while LPAR2 suppression did not inhibit the host cell during this process ( Figure 2B ) . Notably , transient expression of LPAR2 in HEK293 cells significantly triggered entotic invasion ( Figure 2C ) , suggesting that disease-associated overexpression or upregulation of LPAR2 as observed in various human cancers ( Goetzl et al . , 1999; Kitayama et al . , 2004; Yun et al . , 2005; Wang et al . , 2007 ) may be instrumental for entosis . Next , we assessed the endogenous localization of LPAR2 in entotic cells using immunofluorescence microscopy . Staining of cells with anti-LPAR2 antibodies showed a cortical signal that was distinctively increased at the rear of the invading cell in particular during more progressed phase of entotic invasion ( Figure 2D ) , which could be confirmed on transiently expressed Flag-LPAR2 ( Figure 2E ) , suggesting that LPAR2-signaling occurs in a defined and more polarized manner . Flag-LPAR2 polarization to the trailing cell rear was independent of downstream actin organization as assessed by addition of latrunculin B , which completely perturbed the cortical actin cytoskeleton ( Figure 2E , lower panel ) . These results establish the LPAR2 as a signal transducer at the cell surface for cell-in-cell invasion . LPAR2 can initiate intracellular signaling via coupling to multiple Gα subunits from the Gi , Gq , and G12/13 family of heterotrimeric G-proteins ( Choi et al . , 2010 ) . Silencing various Gα subunits by siRNA revealed that only suppression of Gα12/13 effectively and significantly blocked entosis ( Figure 3A ) . Consistently , LPAR2-triggered entotic invasion specifically required Gα12/13 , but not Gα11 or Gαq ( Figure 3B ) , clearly demonstrating that LPAR2 signals through G12/13 heterotrimeric G-proteins to promote homotypic cell-in-cell invasion . Furthermore , expression of Gα12 or of a constitutively active mutant Gα12Q/L robustly induced entotic events in the absence of LPA , and this effect was further increased upon addition of 2 μM LPA ( Figure 3C ) . Thus , a canonical LPAR2/ Gα12/13 module critically mediates entosis . 10 . 7554/eLife . 02786 . 009Figure 3 . Gα12/13 and PDZ-RhoGEF are required for entosis . ( A ) MCF10A cells treated with indicated siRNAs for 48 hr were analyzed for relative entosis rates ( n = 5 ± SD analyzed by one way ANOVA followed by Dunnett's post-tests compared with siMOCK group ) . ( B ) HEK293 cells expressing Flag-LPAR2 were treated with indicated siRNAs for 48 hr before analyzing entosis rate ( n = 3 ± SD analyzed by one way ANOVA followed by Dunnett's post-tests compared with Flag-LPAR2-expressing siMOCK group ) . ( C ) HEK293 cells expressing indicated proteins were analyzed for entosis in lipid-depleted medium with or without ( w/o ) the addition of LPA as indicated . ( n = 3 ± SD analyzed by two way ANOVA followed by Bonferroni post-tests ) . ( D ) MCF10A cells treated with indicated siRNAs for 48 hr were analyzed for entosis ( n = 3 ± SD analyzed by one way ANOVA followed by Dunnett's post-tests compared with siMOCK group ) . ( E ) Localization of GFP-PDZ-RhoGEF ( green ) , DAPI ( blue ) , and LifeAct-mCherry ( red ) expressed in MCF-7 cells was analyzed by confocal microscopy . Bright-field image merged with DAPI and LifeAct is shown to reveal the cell-in-cell structure ( left panel ) . Note the accumulation of PDZ-RhoGEF at the actin-rich uropod-like structure of the invading cell . Scale bar 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 00910 . 7554/eLife . 02786 . 010Figure 3—figure supplement 1 . Myosin II activity is present at the actin-rich cup at the invading cell rear . Immunolabeling of endogenous phospho-MLC2 ( red ) and phalloidin staining of F-actin ( green ) of a MCF10A cell undergoing entosis . Nuclei are labeled by DAPI ( blue ) . Scale bar 5 μm . Arrows point at p-MLC enrichment at the trailing cell rear . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 01010 . 7554/eLife . 02786 . 011Figure 3—figure supplement 2 . Analysis of siRNA treatments . ( A ) RT-PCR quantifications for indicated siRNA treatments in MCF10A cells . ( B ) Western blot analysis of MCF10 cell lysates of cell treated with the indicated siRNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 011 Gα12/13 proteins have been shown to directly relay receptor signal informations by binding to the RGS-domain containing RhoGEFs p115-rhoGEF , LARG , or PDZ-RhoGEF ( Fukuhara et al . , 2001 ) for activation of the small GTPase RhoA ( Fukuhara et al . , 2001 ) . Therefore , we used siRNA to suppress each of the three RhoGEFs in cells . Interestingly , siRNA-mediated knock-down of PDZ-RhoGEF specifically inhibited entotic events ( Figure 3D ) . Furthermore , analyzing ectopically expressed GFP-PDZ-RhoGEF during cell-in-cell invasion revealed a strikingly polarized distribution to the invading cell rear where it strongly colocalized with F-actin as determined by LifeAct ( Figure 3E ) . Similar observations were made by staining for phosphorylated myosin-light-chain II ( pMLC2 ) ( Figure 3—figure supplement 1 ) , a downstream target of the Rho-ROCK pathway , in agreement with previous findings ( Wan et al . , 2012 ) . These data argue that PDZ-RhoGEF promotes localized actin assembly at the cell rear during entotic invasion . Our LifeAct or non-transfected cell analysis showed that during entosis the invading but not the receiving cells display rigorous membrane blebbing ( Figure 1A; Video 2 ) reminiscent of bleb-associated cancer cell invasion ( Fackler and Grosse , 2008 ) . We have shown previously that bleb-associated cancer cell invasion through collagen matrices requires the activity of the Diaphanous formin mDia1 downstream of RhoA ( Kitzing et al . , 2007 ) . We therefore hypothesized that the actin nucleation factor mDia1 may also be involved in entotic invasion . Interestingly , we found endogenous mDia1 to be enriched at the cell rear of the invading cell ( Figure 4A ) , suggesting that mDia1 function spatially controls entosis . Indeed , ectopically expressed mDia1-GFP was localized to the actin-rich cup formed in HEK293 cells upon LPAR2 transfection to trigger entotic invasion ( Figure 4B ) . This mDia1 localization is also in good agreement with our data showing that the RhoA activator PDZ-RhoGEF is strongly accumulated at the actin-rich cell rear ( Figure 3E ) . 10 . 7554/eLife . 02786 . 012Figure 4 . The formin mDia1 mediates cell-in-cell invasion downstream of LPAR2 . ( A ) Immunolabeling of endogenous mDia1 ( green ) and phalloidin staining of F-actin ( red ) of a MCF10A cell undergoing entosis . Nuclei are labeled by DAPI ( blue ) . Scale bar 5 μm . ( B ) Visualization of mDia1-GFP ( green ) and mCherry-LifeAct ( red ) localization at the invading cell rear in fixed and non-permeabilized HEK293 cells co-transfected with LPAR2 to trigger cell-in-cell invasion events . Merged image including bright-field and DAPI ( blue ) is shown in the right panel . Scale bar 5 μm . ( C ) Immunolabeling of endogenous Ezrin ( green ) and F-actin ( red ) of control and mDia1 siRNA-treated MCF10A cells . ( D ) MCF10A cell population after incomplete siRNA treatment against mDia1 showing mDia1 knockdown of the upper two cells ( red only ) and endogenous mDia1 detection of the lower three cells were labeled for mDia1 ( green ) and F-actin ( red ) . Note the presence of mDia1 on cellular blebs , while the two upper mDia1-negative cells fail to bleb . 2 frames are shown from a confocal z-scan using a LSM 700 ( Zeiss ) . ( E ) MCF10A cells treated with indicated siRNAs were analyzed for the number of blebbing cells ( n = 3 ± SD , p<0 . 007 , t test ) . ( F ) MCF10A cells pretreated for 40 min with 20 μM of the LPAR inhibitor Ki16425 before analysis of the number of blebbing cells ( n = 3 ± SD , p<0 . 001 , t test ) . ( G ) MCF10A cells expressing LifeAct-GFP ( green ) or LifeAct-mCherry ( red ) silenced for control or mDia1 respectively . White arrowheads in the first frame indicate red ( siDia1 ) and green ( siMOCK ) cell in contact with a host cell . Red arrowhead indicates addition of 100 nM Latrunculin B ( LatB ) at time frame 104 min . ( H ) MCF10A cells treated with indicated siRNAs for 48 hr were analyzed for entosis ( n = 3 ± SD analyzed by one way ANOVA followed by Dunnett's post-tests compared with siMOCK group ) . ( I ) HEK293 cells expressing Flag-LPAR2 to trigger cell-in-cell invasion events were treated with indicated siRNAs for 48 hr before analyzing entosis rates ( n = 3 ± SD analyzed by One way ANOVA followed by Dunnett's post-tests compared with Flag-LPAR2 expressing siMOCK group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 01210 . 7554/eLife . 02786 . 013Figure 4—figure supplement 1 . mDia1 is required for blebbing . ( A ) MCF10A cells cultured on polyHEMA expressing LifeAct-mCherry ( red ) or LifeAct-GFP ( green ) silenced for mDia1 or control respectively were monitored over time as indicated to visualize actin polymerization during blebbing . Each time frame represents a confocal scan using a LSM 700 ( Zeiss ) . Differential interference contrast ( DIC ) is added for each frame . Note that mDia1 silenced cells fail to display any bleb activity ( red ) . ( A′ ) Western analysis for mDia1 siRNA treatment . ( B ) Immunolabeling of endogenous Ezrin ( green ) and phalloidin staining of F-actin ( red ) of a MCF10A cell undergoing entosis . Nuclei are labeled by DAPI ( blue ) . Scale bar 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 013 Analyzing the cells silenced for mDia1 , we found that mDia1 was required for Ezrin-positive bleb formation as well as blebbing ( Figure 4C , D , E , Figure 4—figure supplement 1; Video 3 ) , while mDia1 was localized to cellular blebs in control silenced cells ( Figure 4D ) . Similarly , blebbing was highly sensitive to LPAR inhibition with Ki16425 ( Figure 4F; Video 4 and 5 ) . Importantly , mDia1 knockdown MCF10A cells were strongly impaired to undergo entotic invasion as compared to control siRNA-treated cells ( Figure 4G , H; Video 6 ) . Under these conditions , when latrunculin B was added just before completion of entosis , we observed the rapid dispersion of the trailing actin-rich cup leading to failure of cell-in-cell invasion ( Figure 4G; Video 6 ) , pointing towards a crucial role for polarized actin assembly during final stages of entosis . Suppression of mDia1 by siRNA treatment also strongly and specifically inhibited LPAR2-triggered entosis ( Figure 4I ) in HEK293 cells , showing that mDia1 is an essential factor that acts downstream of LPAR2 during cell-in-cell invasion . 10 . 7554/eLife . 02786 . 014Video 4 . Actin dynamics during blebbing . MCF10A cells expressing LifeAct-mCherry ( red ) were monitored over time as indicated ( upper right corner ) to visualize blebbing cells . Video corresponds to quantifications in Figure 4F as control to Video 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 01410 . 7554/eLife . 02786 . 015Video 5 . Effects of LPAR inhibition on cell blebbing . MCF10A cells expressing LifeAct-mCherry ( red ) were treated with the LPAR inhibitor Ki16425 and monitored over time to visualize effects on cell blebbing . Video corresponds to quantifications in Figure 4F . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 01510 . 7554/eLife . 02786 . 016Video 6 . mDia1 is required for cell-in-cell invasion . MCF10A cells expressing LifeAct-GFP ( green ) or LifeAct-mCherry ( red ) silenced for control or mDia1 respectively were monitored over time ( indicated in each pabel ) for entosis . Latrunculin B ( 100 nM ) was added before full completion of cell-in-cell invasion at time point 104 ( min ) to dissolve the trailing actin-rich uropod-like structure . Video corresponds to Figure 4G . DOI: http://dx . doi . org/10 . 7554/eLife . 02786 . 016 Homotypic cell-in-cell structures have been reported in metastatic carcinoma cells harvested from exudates or urine samples ( Overholtzer and Brugge , 2008 ) , corresponding to the non-adhesive experimental culture conditions on hydrogel . It is tempting to speculate that such active invasive and complex process may result in some survival advantage or even represent an escape mechanism for carcinoma cells , although often the inner cell undergoes a cell death process involving components of the autophagy machinery ( Florey et al . , 2011 ) . In this study , we report on a cell surface receptor pathway that facilitates active invasion to produce a cell-in-cell structure . Interestingly , some of these components such as RhoA and mDia1 have been shown to function during rounded cancer cell invasion , which is similarly accompanied by cell blebbing ( Sanz-Moreno and Marshall , 2010 ) , although this processes can still depend on integrin-based matrix adhesions . Our findings suggest that entotic invasion , although independent of integrins , resembles at least in some aspects that of amoeboid and bleb-dependent motility . Indeed , actin filaments also coincide at the Ezrin-rich uropod in amoeboid blebbing thereby pushing cells through collagen-I ( Lorentzen et al . , 2011 ) and ezrin is further an essential component for non-apoptotic blebbing ( Charras et al . , 2006 ) . It seems reasonably that Ezrin is potentially also required for entosis as we observed strong Ezrin localization at the invading cell uropod ( Figure 4—figure supplement 1B ) , however , its precise role and regulation in this process remains a future task for investigations . The relevance of entosis for tumor progression in vivo is currently unclear . Nevertheless , our data uncover LPA and LPAR2 as important drivers of entotic invasion , both of which are also important factors during cancer metastasis , suggesting that entosis may be a phenomenon associated with advanced malignancy . All cell lines were obtained from American Type Culture Collection . MCF10A cells were cultured as described ( Debnath et al . , 2003 ) . MCF7 and HEK293 cells were cultured in Dulbecco's modified Eagle's medium ( DMEM ) plus 10% heat inactivated FBS . Cell Dissociation Buffer , enzyme-free , PBS-based ( Life Technologies ) . Oligonucleotides of small interfering RNA ( siRNA ) for LPA receptors , G protein alpha subunits , and RhoGEFs were synthesized by QIAGEN . Oligonucleotides of siRNA for mDia1 were purchased from IBA GmbH . LPA ( 1-Oleoyl Lysophosphatidic Acid ) and EDG family inhibitor Ki16425 were purchased from Cayman Chemical Company . Poly ( 2-hydroxyethyl methacrylate ) ( PolyHEMA ) was purchased from Polysciences Inc . Antibodies against EDG4 were from Assay Biotechnology; LPAR receptors , Ezrin and LARG from Santa Cruz Biotechnology; PDZ-RhoGEF from IMGENEX; pMLC2 from Sigma and mDia1 from BD Biosciences . pCMV6-XL5 LPAR2 expression vector were purchased from OriGene ( SC117226 ) . pWPXL-based lentiviral expression vectors for H2B , LPAR2 , Gα12 , and Gα12Q/L were generated using standard PCR-based procedures or in the case of LifeAct-GFP were a kind gift from Oliver Fackler . Delipidized FCS was from Bio&SELL e . K . MCF10A or MCF7 cells were transiently transfected with 10–50 nM siRNA oligonucleotides by using INTERFERin ( Polyplus Transfection Inc ) and knockdown was quantified by qPCR or confirmed by Western analysis ( Figure 3—figure supplement 2 ) . The following FlexiTube siRNA ( QIAGEN ) were used: GNAI1 SI00032256; GNAI2 SI02780505; GNAI3 SI00088942; GNAQ SI02780512; GNA11 SI0265947; GNA12 SI00096558; GNA13 SI00089761; GNA14 SI00062321; ARHGEF1 SI00302680; ARHGEF11 SI00108129; ARHGEF12 SI04352278; AKAP13 SI02224173; EDG1 SI00376229; EDG4 SI00067494; EDG7 SI00097545; GRP23 SI00075292; GPR92 SI00126231; P2RY5 SI00081116; The following siRNAs from IBA GmbH were used: mDia1 . 1 #97082N/97083N; mDia1 . 2 #97273/274N . For general quantitative assessments monolayer cells were trypsinized ( or subjected to Cell Dissociation Buffer in case of HEK293 ) to obtain a single-cell suspension before plating on Ultra Low Cluster Plate ( costar cat . 3473 ) at densities of 300 . 000–400 . 000 cells per well . Cells processed for immunostaining were fixed in suspension by adding 1:1 vol/vol 8% formalin for 10 min . Cells were then rehydrated in PBS and centrifuged on to 12-mm cover slips using a Cytospin Cytofuge12 at 1500 rpm for 4 min by high acceleration . Fixed samples were washed in PBS and PBST and blocked in Blocking Buffer ( PBS , 0 . 2% Triton X-100 , 5% Goat Serum , 0 . 1% BSA , 0 . 4% Glycerol ) before antibody addition . Nuclei were stained with DAPI and F-actin using Alexa-647-phalloidin or Alexa-488-phalloidin . Time-lapse microscopy was performed on dishes coated with PolyHEMA as described ( Overholtzer et al . , 2007 ) . Images were obtained using a Nikon Eclipse-Ti equipped with Perfect Focus under humidified conditions at 37°C ( Tokai hit stage top incubator ) using a Nikon 40x oil objective . Confocal microscopy was performed using a 63x objective on a ZEISS LSM-700 . For all entosis quantifications more than 600 cells were counted per coverslip .	Entosis is the invasion of one cell by another and can be observed in aggressive cancers . Although the invading cell is usually killed , the surviving cell is sometimes left with the wrong number of chromosomes . This suggests that entosis may help cancer to progress because cells with an abnormal number of chromosomes are common in cancers . For entosis to occur , the invading cell must be released from the tissue that surrounds it , so it can move towards and attach to the cell it is about to invade . Very little is currently known about the cellular and molecular events that enable these processes to occur . Purvanov et al . studied entosis in cells grown in the laboratory and observed that invading cells produce bulges and projections at their rear end for invasion . These projections contain a protein called mDia1 . This protein is involved in controlling the growth of the cytoskeleton—the structure that helps cells to both maintain their shape and to move . Adding the signaling molecule lysophosphatidic acid , which is present in human serum , increased the likelihood that cells would invade others . From this , Purvanov et al . established the identities of the proteins involved in transmitting the lysophosphatidic acid signal that controls mDia1 activity during entosis . Changes to this signaling pathway have been associated with cancer and how it spreads between different organs and its involvement in entosis lends further support to the notion that there may be a link between cell-in-cell invasion and the advancement of cancer .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	G-protein-coupled receptor signaling and polarized actin dynamics drive cell-in-cell invasion
MAF1 represses Pol III-mediated transcription by interfering with TFIIIB and Pol III . Herein , we found that MAF1 knockdown induced CDKN1A transcription and chromatin looping concurrently with Pol III recruitment . Simultaneous knockdown of MAF1 with Pol III or BRF1 ( subunit of TFIIIB ) diminished the activation and looping effect , which indicates that recruiting Pol III was required for activation of Pol II-mediated transcription and chromatin looping . Chromatin-immunoprecipitation analysis after MAF1 knockdown indicated enhanced binding of Pol III and BRF1 , as well as of CFP1 , p300 , and PCAF , which are factors that mediate active histone marks , along with the binding of TATA binding protein ( TBP ) and POLR2E to the CDKN1A promoter . Simultaneous knockdown with Pol III abolished these regulatory events . Similar results were obtained for GDF15 . Our results reveal a novel mechanism by which MAF1 and Pol III regulate the activity of a protein-coding gene transcribed by Pol II . Transcription by RNA polymerase III ( Pol III ) is regulated by MAF1 , which is a highly conserved protein in eukaryotes ( Pluta et al . , 2001; Reina et al . , 2006 ) . MAF1 represses Pol III transcription through association with BRF1 , a subunit of initiation factor TFIIIB , which prevents attachment of TFIIIB onto DNA . This interaction also inhibits Pol III from binding to BRF1 , which in turn prevents recruitment of Pol III to Pol III promoters . Furthermore , MAF1 also inhibits Pol III transcription through direct binding with Pol III , which interferes with the recruitment of Pol III to the assembled TFIIIB/DNA complexes ( Desai et al . , 2005; Vannini et al . , 2010 ) . In addition , association of MAF1 with Pol III-transcribed genes has been detected genome-wide concomitant with an increase in occupation during repression; this indicates that direct interaction of MAF1 with Pol III genes is also an important attribute of repression ( Roberts et al . , 2006 ) . MAF1 has also been proposed to have the potential to repress Pol II-mediated transcription via repression of TBP transcription due to binding of MAF1 to the Elk-1 site on the TBP promoter ( Johnson et al . , 2007 ) . Thus , to investigate the potential regulatory role of MAF1 in Pol II genes , we carried out MAF1 knockdown coupled with microarray analysis . Microarray analysis showed that 124 genes were upregulated and 170 genes were downregulated more than twofold after MAF1 knockdown . Ingenuity Pathway Analysis ( IPA ) indicated that most of these genes are related to cell proliferation . Among them , CDKN1A ( also known as p21 ) was significantly upregulated and the mechanism of induced transcription of this gene after MAF1 knockdown was further investigated . CDKN1A is a cyclin-dependent kinase inhibitor that inhibits cell cycle progression through interaction with cyclins and cyclin-dependent kinases ( CDKs ) . As a member of the Cip and Kip family of CDK inhibitors , CDKN1A mediates p53-dependent cell-cycle arrest at the G1 phase by inhibiting the activity of CDK2 and CDK1 ( also known as CDC2 ) . In addition , CDKN1A also inhibits the activity of proliferating cell nuclear antigen and blocks DNA synthesis and repair as well as cell-cycle progression . As a result , CDKN1A can regulate many cellular processes , such as proliferation , differentiation , apoptosis , metastasis , cell survival , and stem cell renewal . Expression of CDKN1A can be regulated at the transcriptional level by oncogenes and tumor suppressor proteins that bind various transcription factors to specific elements in response to a variety of intracellular and extracellular signals ( Abbas and Dutta , 2009; Warfel and El-Deiry , 2013 ) . In this study , we showed that MAF1 can bind to the CDKN1A promoter to repress its transcription . Enhanced binding of Pol III after MAF1 knockdown induced CDKN1A transcription and chromatin looping by recruiting common Pol II and Pol III transcription factors as well as binding of TBP , p300 , CFP1 , and PCAF , along with increase in histone modifications associated with gene activation . Simultaneous knockdown of Pol III and MAF1 abolished both promoter looping and activation of CDKN1A transcription , which indicates that Pol III actively participated in regulation of Pol II genes . Similar results were observed in another cell proliferation-related gene , GDF15 . These observations reveal a new type of gene regulation in which binding of MAF1 regulates Pol III-mediated transcriptional activation and chromatin looping of Pol II genes . To examine whether MAF1 has the potential to repress Pol II-transcribed genes , we first examined the knockdown effect of MAF1 by quantitative RT-PCR ( qRT-PCR ) and immunoblot using multiple siRNAs ( Figure 1A , B ) . The siRNA with the strongest knockdown effect was used to perform expression analysis using microarray . 124 Pol II-transcribed genes were upregulated more than twofold after MAF1 knockdown . Among them , CDKN1A was significantly upregulated , resulting in the downregulation of positive cell cycle regulators . Consistent with expression data , flow cytometry analysis showed that MAF1 knockdown arrested cells at the G1 phase ( Figure 1C ) . We carried out qRT-PCR to confirm whether CDKN1A expression was upregulated by MAF1 knockdown . Efficiency of MAF1 knockdown was verified by the strong upregulation of two products of Pol III , pretRNATyr and pretRNALeu ( Reina et al . , 2006 ) ( Figure 1D ) . Consistent with microarray analysis , qRT-PCR and immunoblot analysis showed that CDKN1A expression was upregulated about 10-fold after MAF1 knockdown ( Figure 1D , E ) . GAPDH , ACTB , and TAF5 , genes that were not affected by MAF1 knockdown in the microarray , were chosen as the control for qRT-PCR . Expression of these genes was not affected by MAF1 knockdown ( Figure 1D ) . 10 . 7554/eLife . 06283 . 003Figure 1 . MAF1 knockdown strongly upregulates CDKN1A expression and arrests MCF-7 cells at the G0/G1 phase . Analysis of MAF1 expression after MAF1 knockdown using three different siRNAs in MCF-7 cells by quantitative RT-PCR ( A ) and immunoblot analysis ( B ) . The immunoblot results were quantified ( left panel ) using α-tubulin as a loading control on a representative gel ( right panel ) . ( C ) MAF1 knockdown arrested the MCF-7 cell cycle at the G0/G1 phase . At 72 hr after knockdown , cells were stained with propidium iodide and subjected to cell cycle analysis by flow cytometry ( top panel ) . The quantification results show that MAF1 knockdown increased cells arrested at the G0/G1 phase by 16 . 4% ± 1 . 76% ( bottom panel ) . ( D ) Quantitative RT-PCR of genes in MCF-7 cells subjected to siRNA knockdown for 72 hr . CDKN1A expression was upregulated 10-fold , and upregulation was abolished by double knockdown of MAF1 and POLR3A . Relative expression normalized to GAPDH is displayed . ( E ) Immunoblot analysis of CDKN1A expression after MAF1 knockdown in MCF-7 cells . The results were quantified ( left panel ) using α-tubulin as a loading control on a representative gel ( right panel ) . All data shown represent mean ± SD , n ≥ 3 , p < 0 . 01 , p < 0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 00310 . 7554/eLife . 06283 . 004Figure 1—figure supplement 1 . MAF1 knockdown upregulates CDKN1A and GDF15 expression in HCT116p53+/+ ( wild-type ) , HCT116p53−/− ( p53-null ) , MCF-10A , and MDA-MB-231 cell lines . ( A ) Immunoblot analysis of p53 expression in wild-type and p53-null HCT116 . Quantitative RT-PCR of genes in HCT116 wild-type ( B ) and HCT116p53−/− ( C ) cells subjected to siRNA knockdown for 72 hr . ( D ) Immunoblot analysis of MAF1 , CDKN1A , and p53 expression in p53-null HCT116 subjected to MAF1 knockdown . Quantitative RT-PCR of genes in MCF-10A ( E ) and MDA-MB-231 ( F ) cells subjected to siRNA knockdown for 72 hr . Expression of CDKN1A and GDF15 was upregulated independent of p53 after MAF1 knockdown . POLR3A expression analysis after POLR3A knockdown using three different siRNAs in MCF-7 cells by quantitative RT-PCR ( G ) and immunoblot analysis ( H ) . The results were quantified ( left panel ) using α-tubulin as a loading control on a representative gel ( right panel ) . Relative expression normalized to GAPDH is displayed for all quantitative RT-PCR . All data shown represent mean ± SD , n = 3 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 004 Because CDKN1A is a downstream target of p53 ( Allen et al . , 2014 ) , we further performed MAF1 knockdown in HCT116p53+/+ ( wild-type ) and HCT116p53−/− ( p53-null ) cell lines to analyze whether the induced CDKN1A expression is dependent on p53 . The absence of p53 in HCT11653−/− was confirmed by immunoblot ( Figure 1—figure supplement 1A ) . CDKN1A expression was induced after MAF1 knockdown in both wild-type and p53-null HCT116 , which indicates that the activation is independent of p53 ( Figure 1—figure supplement 1B , C ) . Immunoblot analysis also showed that CDKN1A protein level was upregulated after MAF1 knockdown in p53-null HCT116 ( Figure 1—figure supplement 1D ) . CDKN1A activation by MAF1 knockdown was also found in a non-tumorigenic cell line ( MCF-10A ) and a p53 mutant breast cancer cell line ( MDA-MB-231 ) ( Figure 1—figure supplement 1E , F ) . Together , these results demonstrate that MAF1 can regulate CDKN1A expression in a variety of cell types independent of p53 . CDKN1A activation after MAF1 knockdown could be due either to interference of binding of transcription factors to the CDKN1A promoter by MAF1 or to the active recruitment or activation of Pol III after MAF1 knockdown . To determine which of these two mechanisms are involved in this process , we carried out simultaneous knockdown of both MAF1 and Pol III . The former mechanism would not be affected by the double knockdown , whereas the latter would be . The effect of Pol III knockdown was analyzed by qRT-PCR and immunoblot using multiple siRNA sequences ( Figure 1—figure supplement 1G , H ) . Simultaneous knockdown of Pol III and MAF1 indeed abolished the induction of CDKN1A expression by knockdown of only MAF1 ( Figure 1D and Figure 1—figure supplement 1B , C , E , F ) in five different cell lines . A control experiment using two Pol III genes , pretRNATyr and pretRNALeu , confirmed the efficiency of Pol III knockdown ( Figure 1D and Figure 1—figure supplement 1B , C , E , F ) . Knockdown of Pol III alone did not significantly affect CDKN1A expression ( Figure 1D ) . This result indicates that Pol III plays a critical role in activation of CDKN1A expression by MAF1 knockdown . Because mRNA levels can be affected by transcription , post-transcriptional processing as well as RNA turnover rate , upregulation of gene expression after MAF1 knockdown could be due to post-transcriptional mechanisms other than transcription activation . To demonstrate that the induced CDKN1A expression indeed occurs at the transcriptional level , we analyzed the rate of nascent transcription by conducting a nuclear run-on experiment after MAF1 knockdown or simultaneous knockdown of MAF1 and Pol III . The run-on nascent RNA was labeled with biotin , affinity purified , and analyzed by RT-PCR . A negative control without biotin labeling was used . Consistent with qRT-PCR analysis , nascent transcription of CDKN1A was indeed induced after MAF1 knockdown , whereas transcription of ACTB and TAF5 was not affected ( Figure 2A , B ) . Simultaneous knockdown of MAF1 and Pol III diminished the induced nascent RNA transcription of CDKN1A by knockdown of MAF1 alone ( Figure 2A , B ) . These results indicate that Pol III is required for the induction of CDKN1A upon MAF1 knockdown . 10 . 7554/eLife . 06283 . 005Figure 2 . MAF1 knockdown upregulates CDKN1A at the transcriptional level . ( A ) For run-on assay , MCF-7 cells were subjected to siRNA knockdown of MAF1 ( KD MAF ) or simultaneous knockdown of MAF1 and Pol III for 72 hr ( KD P/M ) . Nuclei were prepared , and a run-on reaction was performed . Run-on biotin-labeled newly transcribed RNA ( Run-on ) was affinity purified and subjected to RT-PCR ( left panel ) . Input indicates total RNA before affinity purification , and a negative control was performed by omitting biotinylated nucleotides and subjected to RT-PCR ( right panel ) . ( B ) The run-on results were quantified , and the data shown represent mean ± SD , n = 3 , p < 0 . 05 , *p < 0 . 01 ( t-test ) . ( C ) Schematic diagram of the CDKN1A promoter , including locations of exon 1 ( black rectangle ) , SINE ( MIR3 ) , CpG island ( green rectangle ) , guanine-cytosine ( GC ) skew , and R-loop foot-printing region ( blue rectangle ) . ( D ) Each vertical black line indicates the position of a cytosine on the sense DNA strand . ( E ) Analysis of R-loop foot-printing was performed by native sodium bisulfite treatment followed by PCR amplification and cloning . A total of at least 10 clones were obtained for each knockdown condition ( knockdown control , ‘KD Ctrl’; knockdown MAF1 , ‘KD MAF1’; and simultaneous knockdown of MAF1 and Pol III , ‘KD MAF1/Pol III’ ) . Each vertical red line represents a converted cytosine to thymine in the sense direction ( CDKN1A mRNA ) for the knockdown control , knockdown MAF1 , and simultaneous knockdown of MAF1 and Pol III . Percentage indicates how many clones at a particular cytosine were converted . Knockdown MAF1 extended the length of R-loop formation in CDKN1A , whereas simultaneous knockdown of MAF1 and Pol III abolished the extension . This indicates that regulation of CDKN1A expression by MAF1 and Pol III occurs at the transcriptional level . Background conversion ( approximately 5% of cytosine ) may be seen because of DNA breathing during the prolonged incubation at 37°C in our data and data produced by others ( Yu et al . , 2003 ) . ( F ) Schematic diagram of ACTB , including locations of exons , CpG island , GC skew , and R-loop foot-printing region . ( G ) Each vertical black line indicates the position of a cytosine on the sense DNA strand . ( H ) Each vertical red line represents a converted cytosine to thymine in the sense direction ( ACTB mRNA ) for knockdown control and knockdown MAF1 . Knockdown MAF1 did not affect the length of R-loop in ACTB , which correlates with the expression data from Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 005 Recent evidence indicated that R-loop formation positively correlates with active transcription in human cells by maintaining the unmethylated state at promoters with skewed guanine-cytosine ( GC ) content ( Ginno et al . , 2012 ) . The high GC skew of the CDKN1A promoter prompted us to test whether the R-loop was present in this region during the activation of transcription by MAF1 knockdown ( Figure 2C ) . We performed R-loop foot-printing by native sodium bisulfite treatment , which converts cytosine to uracil only on the single-stranded DNA ( Yu et al . , 2003 ) . MAF1 knockdown resulted in the formation of the extended R-loop in the gene body of CDKN1A , which indicates active transcription , whereas simultaneous knockdown of Pol III and MAF1 inhibited R-loop formation ( Figure 2C–E ) . The R-looping of the control gene ACTB , an active housekeeping gene with high GC skew and expression , was not affected by MAF1 knockdown ( Figure 2F–H ) . The results of R-loop formation and nuclear run-on described above further confirmed that the upregulation of CDKN1A expression by MAF1 knockdown and recruitment of Pol III occurred at the transcriptional level . Expression analysis indicated that CDKN1A expression is strongly upregulated after MAF1 knockdown , and simultaneous knockdown of MAF1 and Pol III diminished the induced expression . These results indicate that removal of MAF1 may induce recruitment of Pol II and Pol III to activate transcription . To examine this possibility , chromatin-immunoprecipitation ( ChIP ) analysis followed by quantitative PCR ( qPCR ) was performed under various knockdown conditions . Efficiency of MAF1 or Pol III knockdown as well as simultaneous knockdown of MAF1 and Pol III was verified by the binding of Pol III to pretRNAArg and pretRNALeu genes . The results showed enhanced binding of Pol III at pretRNAArg and pretRNALeu after MAF1 knockdown , and the binding was diminished after double knockdown ( Figure 3A , B ) . Examination of CDKN1A gene in the UCSC Genome Database shows that there are two transcription start sites , NM_001220777 ( long form ) and NM_001220778 ( short form ) , which are 2 . 25 kb apart ( Figure 3—figure supplement 1A ) . Although the expression of both forms was induced after MAF1 knockdown , the short form had higher expression level and stronger promoter activity in the MCF-7 cell line ( Figure 3—figure supplement 1B–D ) . ChIP analysis showed that MAF1 was associated with both transcription start site regions ( Figure 3C , D ) . Furthermore , MAF1 knockdown resulted in the depletion of this regulatory factor with concomitant increase in the binding of both Pol II and Pol III polymerases to both transcription start site regions ( Figure 3E , F ) . 10 . 7554/eLife . 06283 . 006Figure 3 . MAF1 knockdown enhanced binding of Pol III and Pol II at the CDKN1A promoter . ChIP was performed in MCF-7 cells subjected to siRNA knockdown for 72 hr . DNA isolated from immunoprecipitated chromatin was subjected to qPCR and calculated as indicated in the ‘Materials and methods’ . Significant binding of Pol III was detected at two tRNA genes , tRNAArg ( A ) and tRNALeu ( B ) , after MAF1 knockdown ( KD MAF1 ) . The enhance binding of Pol III was diminished when there was simultaneous knockdown of MAF1 and Pol III ( KD M/Pol III ) . ( C ) Diagram of the CDKN1A promoter , including locations of exon 1 ( long form: L-Ex1 , and short form: Ex1 ) , SINEs ( AluSx and MIR3 ) , and ChIP–qPCR amplicons ( p21-L , p1 , p2 , and p3 ) . ( D ) Binding of MAF1 was detected at the CDKN1A promoter , which diminished after MAF1 knockdown . ( E ) Enhanced binding of Pol III was detected at the CDKN1A promoter after MAF1 knockdown . ( F ) MAF1 knockdown indicates enhanced binding of Serine-5-phosphorylated Pol II , which was abolished when there was simultaneous knockdown of Pol III and MAF1 . ( G ) Enhanced binding of BRF1 was detected at the CDKN1A promoter after MAF1 knockdown . ( H ) Top panel: diagram of the ACTB promoter , including locations of each exon ( Ex1 to Ex4 ) and ChIP–qPCR amplicons ( p1 , p2 , and p3 ) . Bottom panel: neither MAF1 nor Pol III was detected at the ACTB promoter . Only binding of Pol II was detected at the ACTB promoter . ( I ) Top panel: diagram of the TAF5 promoter , including locations of exon 1 and ChIP–qPCR amplicons ( p1 , p2 , and p3 ) . Bottom panel: neither MAF1 nor Pol III was detected at the TAF5 promoter . Only binding of Pol II was detected at the TAF5 promoter . All data shown represent the mean ± s . e . m . , n ≥ 3 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 00610 . 7554/eLife . 06283 . 007Figure 3—figure supplement 1 . Expression and promoter activity of CDKN1A transcripts in the MCF-7 cell line . ( A ) Representative diagram of NM_001220777 ( CDKN1A-L , long form ) and NM_001220778 ( CDKN1A-S , short form ) . ( B ) cDNA samples used in Figure 1C were used to analyze CDKN1A transcript expression . Expression of CDKN1A-S was measured relative to that of CDKN1A-L . ( C ) Expression of both CDKN1A transcripts was upregulated after MAF1 knockdown , and the upregulation was abolished by simultaneous knockdown of MAF1 and Pol III . A relative expression normalized over GAPDH is displayed . ( D ) CDKN1A-L and CDKN1A-S promoter regions were constructed and cloned into a pGL3-basic reporter plasmid , as indicated in the ‘Materials and methods’ . Luciferase reporter assays were performed in MCF-7 cells , and the results were normalized with those for β-galactosidase . Promoter activity was measured relative to CDKN1A-L . All data shown represent mean ± SD , n ≥ 3 , p < 0 . 05 , p < 0 . 01 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 007 Consistent with the expression data , there was significant increase in the binding of active , Serine-5-phosphorylated Pol II at the CDKN1A promoter after MAF1 knockdown , which indicates that the gene was in the active transcription state . The binding of active Pol II was abolished after simultaneous knockdown of Pol III and MAF1 ( Figure 3F ) . Simultaneous knockdown of MAF1 and BRF1 , a subunit of TFIIIB that associates with Pol III and is required for binding of Pol III to the DNA template , also abolished the enhanced binding of Pol III and Pol II after MAF1 knockdown ( Figure 3E , F ) . ChIP analysis also indicated induced binding of BRF1 after MAF1 knockdown , whereas the binding was diminished under simultaneous knockdown of either Pol III or BRF1 with MAF1 ( Figure 3G ) . These results therefore support the mechanism that recruitment of Pol III to the promoter after MAF1 knockdown enhances CDKN1A expression . Expression of ACTB and TAF5 was not affected because MAF1 and Pol III did not bind to their promoters ( Figure 3H , I ) and their expression was not affected by MAF1 knockdown; therefore , they were used as negative controls . The transcription of these two control genes was not affected by single or double knockdown of Pol III and/or MAF1 ( Figure 1D ) . Because ChIP data revealed that MAF1 may directly bind to the promoter , we searched for potential binding sites in the CDKN1A promoter region . We noticed a MIR3 element ( a SINE with Pol III promoter ) that could be transcribed by Pol III and therefore might represent the target-binding site of MAF1 . To test this possibility , we used an in vitro DNA binding reaction ( Britten , 1996; Toth and Biggin , 2000 ) to examine whether purified MAF1 protein could bind to the cloned CDKN1A promoter . Consistent with ChIP , in vitro DNA binding assay showed that purified MAF1 protein could indeed bind to the CDKN1A promoter that contained the MIR3 element ( Figure 4A , B ) , but the binding was abolished when the MIR3 repeat element was deleted , which indicates specificity of MAF1 binding to the MIR3 element ( Figure 4A , B ) . To further show that the Pol III promoter of the MIR3 element is responsible for the binding , an in vitro binding assay was carried out with DNA in which the MIR3 DNA sequence from the CDKN1A promoter had a deleted or mutated A-box sequence ( Pol III promoter element ) . In vitro DNA-protein binding assays performed by the above method or colorimetric assay ( Abcam , ab117139 ) both showed that the DNA with deleted or mutated A-box sequences exhibited significantly lower binding of MAF1 , which indicates the specificity of MAF1 for the Pol III promoter element ( Figure 4A–C ) . Moreover , consistent with ChIP data , MAF1 did not bind in vitro to the ACTB promoter that did not contain a SINE or sequences that would resemble the Pol III promoter element ( Figure 4A , B ) . To the best of our knowledge , this is the first demonstration of direct binding of MAF1 to a specific DNA sequence . 10 . 7554/eLife . 06283 . 008Figure 4 . In vitro binding and transcription assays demonstrate MAF1-regulated Pol III-mediated activation of Pol II-regulated genes . ( A ) Diagrams of Pol II promoters ( CDKN1A , ACTB , RPPH1 , and GDF15 ) with locations of exon 1 , SINEs ( red ) , and constructed DNA template ( green arrow ) for the in vitro MAF1 binding assay . ( B ) An in vitro DNA binding assay was performed as described in the ‘Materials and methods’ . In brief , DNA template , MAF1 protein ( His-tagged ) , and Anti-6× His tag antibody were added to the binding reaction ( Protein + Ab ) . A negative control was performed by substituting IgG antibody for Anti-6× His tag antibody ( Protein + IgG ) or with only the Anti-6× His tag antibody for the MAF1 protein ( Ab only ) . DNA isolated from the immunoprecipitated protein–DNA complex was subjected to qPCR . Deletion of a SINE in the CDKN1A template as well as deletion or mutation of the Pol III A-box element in the CDKN1A and GDF15 template depleted MAF1 binding . Binding of MAF1 to RPPH1 or ACTB promoters was not detected . Data shown are the mean ± SD , n ≥ 3 , p < 0 . 01 , *p < 0 . 001 ( t-test ) . ( C ) An in vitro DNA–protein binding assay was performed using a colorimetric assay kit ( ab117139 ) . The assayed DNA template ‘p21’ ( DNA template with a Pol III A-box element obtained from CDKN1A ) was labeled with biotin ( a probe ) . Purified MAF1 protein ( His tag ) ( 80R-1955 , Fitzgerald ) was used for the binding assay . Different competitors ( described below ) were added to the mixture to demonstrate the specificity of binding of MAF1 at the Pol III promoter element . Competitors: ‘self’ indicates the same DNA template without the biotin label , ‘GDF’ indicates the non-labeled DNA template that contained the Pol III promoter element obtained from the GDF15 promoter , and ‘Mut’ indicates the Pol III A-box element was mutated in the DNA template . A blank control was performed without the addition of protein , and the degree of enrichment was calculated by subtracting the value of the blank control . MAF1 directly bound to the Pol III promoter element , but the mutant form did not . Data shown are the mean ± SD , n = 3 , p < 0 . 001 ( t-test ) . ( D ) In vitro transcription assays were performed on CDKN1A and TAF5 using the HeLaScribeR Nuclear Extract in vitro Transcription System ( Promega ) , as indicated in the ‘Materials and methods’ . Inhibition of Pol II transcription was performed by addition of α-amantin during in vitro transcription of CDKN1A and TAF5 . The MAF1 protein was pre-incubated with template DNA before addition of nuclear extract to enable binding of MAF1 to the template DNA . ( E ) Different antibodies , as indicated , were pre-incubated with nuclear extract before adding template DNA to perform in vitro transcription to deplete the target protein of interest . For the control , no antibody was added prior to in vitro transcription . In vitro transcription performed on Pol III-transcribed RPPH1 and Pol II-transcribed TAF5 served as controls . In vitro transcription performed on CDKN1A and GDF15 revealed that removal of MAF1 promoted transcription , whereas A-box-deleted GDF15 , denoted as ‘GDF15 ( Del ) ’ , did not . The degree of enrichment of all performed in vitro transcription was calculated relative to the ratio of signals obtained from the input RNA after subtraction of the negative control ( no biotin labeling ) . All data shown represent the mean ± s . e . m . , n ≥ 3 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 008 Although the several types of evidence discussed above strongly support the regulation of CDKN1A by recruitment of Pol III to the promoter , the effect observed in vivo nevertheless could be due to some other indirect effect . To directly demonstrate the enhancement of Pol II transcription by removing MAF1 and recruiting Pol III , we carried out in vitro transcription using commercial HeLa cell nuclear extract . Constructed DNA templates of the genes analyzed are described in the ‘Materials and methods’ . The in vitro , newly transcribed RNA was labeled with biotin , and the products were affinity purified . The nature of the affinity-purified nascent RNA was then analyzed by qRT-PCR . Negative control without biotin labeling was used . First , we showed that the in vitro transcription was indeed mediated by Pol II by inhibition of transcription either by α-amanitin treatment ( Figure 4D ) or by depletion of Pol II in the extract using anti-Pol II antibody ( Figure 4E ) . When purified MAF1 protein was pre-incubated with DNA template prior to the addition of nuclear extract , CDKN1A transcription was repressed with respect to the control ( Figure 4D ) . This result is consistent with in vivo expression analysis and the in vitro binding assay . Together these results suggest that MAF1 protein serves as repressor of CDKN1A transcription . As a control , pre-incubation with MAF1 protein did not affect TAF5 transcription ( Figure 4D ) because MAF1 did not bind to this DNA template in vivo ( Figure 3I ) . When nuclear extract was pre-incubated with an anti-MAF1 antibody to deplete MAF1 during in vitro transcription , CDKN1A transcription was significantly upregulated compared with that of the control , which was pre-incubated with IgG or no antibody ( Figure 4E ) . Simultaneous depletion of Pol III and MAF1 by pre-incubation nuclear extract with Pol III and MAF1 antibodies abolished the enhancement of transcription after depletion of MAF1 alone ( Figure 4E ) . Specificity of the Pol III antibody was demonstrated by the inhibitory effect of anti-Pol III antibody on in vitro transcription of the Pol III-transcribed RPPH1 promoter in a pSUPER plasmid . Addition of the anti-MAF1 antibody did not induce RPPH1 transcription ( Figure 4E ) because binding of MAF1 was not detected in this gene by the in vitro MAF1 binding assay ( Figure 4B ) . These experiments recapitulated the in vivo transcription regulation of CDKN1A by MAF1 and Pol III . Taken together with the in vivo and in vitro nuclear run-on expression analyses , these results unambiguously demonstrate that MAF1 can serve as a repressor of the CDKN1A promoter , and that recruiting Pol III after MAF1 depletion is crucial for the activation of CDKN1A transcription . The data above indicate that binding of Pol III to the CDKN1A promoter after MAF1 knockdown is crucial for enhanced transcription . This indicates that Pol III may help recruit the regulatory factors necessary for efficient Pol II transcription . To test this hypothesis , we carried out ChIP analysis to examine the Pol III-dependent recruitment of transcription activators after MAF1 removal . ChIP analysis showed that MAF1 knockdown resulted in significantly enhanced levels of active histone modifications , including H3K4me3 , H3K9Ace , and H3K27Ace , in the 5′ regions of CDKN1A ( Figure 5A , B ) . The enhanced active histone marks after MAF1 knockdown were abolished under simultaneous knockdown of Pol III and MAF1 ( Figure 5B ) . The histone repression marker , H3K27me3 , was detected at the 5′ regions and decreased after MAF1 knockdown , but the level was restored under simultaneous knockdown of Pol III and MAF1 ( Figure 5B ) . 10 . 7554/eLife . 06283 . 009Figure 5 . MAF1 knockdown induces Pol II initiation , active histone marks ( H3K4me3 , H3K9Ace , and H3K27Ace ) , and binding of CFP1 , p300 , PCAF , TBP , and POLR2E at the CDKN1A promoter . ( A ) Diagram of the CDKN1A promoter , including locations of exon 1 ( Ex1 ) , SINEs ( AluSx and MIR3 ) , and ChIP qPCR amplicons ( p21-L , p1 , p2 , and p3 ) . ( B ) Knockdown coupled with ChIP assays with antibodies for H3K27me3 , H3K4me3 , H3K27Ace , and H3K9Ace were performed in MCF-7 cells subjected to siRNA knockdown for 72 hr . DNA isolated from immunoprecipitated chromatin was subjected to qPCR and calculated as described in the ‘Materials and methods’ . Knockdown MAF1 ( KD MAF1 ) enhanced active histone marks H3K4me3 , H3K27Ace , and H3K9Ace , whereas simultaneous knockdown of Pol III and MAF1 ( KD M/Pol III ) abolished the enhanced histone marks . ChIP with anti-CFP1 ( IP: CFP1 ) ( C ) , anti-p300 ( IP: p300 ) ( D ) , anti-PCAF ( IP: PCAF ) , ( E ) anti-TBP ( IP: TBP ) ( F ) , and anti-POLR2E ( IP: POLR2E ) ( G ) antibodies were performed as described in ( B ) . Knockdown MAF1 ( KD MAF1 ) enhanced binding of CFP1 , p300 , PCAF , TBP , and POLR2E , whereas simultaneous knockdown of Pol III and MAF1 ( KD M/Pol III ) abolished the enhanced binding . All data shown are the mean ± s . e . m . , n ≥ 3 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 00910 . 7554/eLife . 06283 . 010Figure 5—figure supplement 1 . Enhanced gene expression by MAF1 knockdown is abolished by simultaneous knockdown of MAF1 with TBP or CFP1 . ( A ) Quantitative RT-PCR of genes in MCF-7 cells subjected to siRNA knockdown of MAF1 , CFP1 , or simultaneous knockdown of both for 72 hr . CDKN1A expression was upregulated after MAF1 knockdown , and the upregulation was abolished by simultaneous knockdown of MAF1 and CFP1 . ( B ) Quantitative RT-PCR of genes in MCF-7 cells subjected to siRNA knockdown of MAF1 , TBP , or simultaneously knockdown of knockdown of MAF1 and TBP for 72 hr . Expression of CDKN1A and TBP was upregulated after MAF1 knockdown , and the upregulation was abolished by simultaneous knockdown of MAF1 and TBP . Relative expression normalized to 18S is displayed . All data shown represent the mean ± SD , n ≥ 3 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 010 Because H3K4 methylation is catalyzed by the SET1/MLL family of histone methyltransferases in humans ( Shilatifard , 2012 ) , we performed knockdown assays to investigate which methyltransferase is responsible for H3K4me3 modification after MAF1 knockdown . Previously , CFP1 ( SET1C-specific subunit ) and p300 were shown to act cooperatively to regulate H3K4me3 modification and CDKN1A transcription ( Tang et al . , 2013 ) . Indeed , the MAF1 knockdown-induced transcription of CDKN1A was downregulated after CFP1 ( CXXC1 ) knockdown ( Figure 5—figure supplement 1A ) . p300 and PCAF have been shown to regulate K27 and K9 acetylation , respectively , of the CDKN1A promoter ( Love et al . , 2012 ) . Thus , we next analyzed whether CFP1 , p300 , and PCAF could bind to the CDKN1A promoter after MAF1 knockdown . As expected , the removal of MAF1 by knockdown induced binding of CFP1 , p300 , and PCAF to the CDKN1A promoter ( Figure 5C–E ) . Furthermore , simultaneous knockdown of MAF1 and Pol III abolished the induced binding of these factors along with active histone marks , which indicates that Pol III is required to recruit these factors to the CDKN1A promoter for histone modifications ( Figure 5C–E ) . Because TBP is an important factor required in both Pol II ( part of TFIIB ) and Pol III ( part of TFIIIB ) transcription ( Zhao et al . , 2003 ) , we next determined whether binding of TBP was enhanced after MAF1 knockdown . Indeed , enhanced binding of TBP was observed at the CDKN1A promoter after MAF1 knockdown , and the binding was abolished when there was simultaneous knockdown of Pol III and MAF1 ( Figure 5F ) . Simultaneous knockdown of MAF1 and TBP also abolished the enhanced expression of CDKN1A by knockdown of only MAF1 ( Figure 5—figure supplement 1B ) , which indicates that TBP is important for CDKN1A transcription activation . We also observed enhanced binding of a common subunit of all three RNA polymerases , that is , POLR2E ( RPB5 ) , after MAF1 knockdown , and this enhanced binding was also abolished when there was simultaneous knockdown of Pol III and MAF1 ( Figure 5G ) , indicating interplay between Pol II and Pol III polymerases . Because binding of Pol III and Pol II was detected at the 5′ flanking regions of both long and short CDKN1A forms in the UCSC Genome Database after MAF1 knockdown , we performed 3C analysis to investigate whether chromatin looping occurs between these two regions . Chromatin looping was detected between the regions after MAF1 knockdown , but not in adjacent regions ( Figure 6A , B ) . Moreover , simultaneous knockdown of MAF1 with either Pol III or BRF1 ( a subunit of TFIIIB ) disrupted the looping formation ( Figure 6C ) . These results demonstrate that Pol III is required for induced chromatin looping after MAF1 knockdown . 10 . 7554/eLife . 06283 . 011Figure 6 . Pol III is required for chromatin looping at the CDKN1A promoter after MAF1 knockdown . ( A ) Schematic diagram of CDKN1A with the orientation of 3C primers ( arrows: 5r , 4r , 3r , 2r , and 2f ) and location of exon 1 ( long form: L-Ex1; short form: Ex1 ) . ( B ) MCF-7 cells were subjected to siRNA knockdown of MAF1 ( KD MAF1 ) for 72 hr . 3C assay was performed as indicated in the ‘Materials and methods’ , and DNA was subjected to PCR . Chromatin looping was detected after MAF1 knockdown from 2r to 2f ( top panel ) and are shown by a representative gel ( bottom panel ) . ( C ) The induced chromatin looping after MAF1 knockdown was diminished when either Pol III ( KD M/Pol III ) or BRF1 ( KD M/BRF1 ) underwent simultaneous knockdown with MAF1 ( top panel ) and are shown by a representative gel ( bottom panel ) . All data shown represent the mean ± s . e . m . , n ≥ 3 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 011 The above results demonstrate that MAF1 knockdown can activate CDKN1A expression by recruiting Pol III and Pol II along with histone-modifying factors . To demonstrate that this type of mechanism also regulates expression of other Pol II genes , we performed expression analysis of GDF15 , which is another cell proliferation-related gene that is upregulated after MAF1 knockdown as found by microarray analysis . As expected , qRT-PCR analysis showed that GDF15 expression was strongly upregulated after MAF1 knockdown , and simultaneous knockdown of MAF1 with Pol III diminished the induced expression ( Figure 1D ) . ChIP analysis also indicated binding of MAF1 at the 5′ flanking region of GDF15 ( Figure 7A , B ) . 10 . 7554/eLife . 06283 . 012Figure 7 . Pol III is required for chromatin looping at the GDF15 promoter after MAF1 knockdown . ( A ) Schematic diagram of GDF15 with ChIP–qPCR amplicons ( AluSx , 3C , MIR , p1 , p2 , and p3 ) , the orientation of 3C primers ( arrows: 3C-3r , 3C-2r , and 3C-1f ) , and locations of exons ( Ex1 and Ex2 ) . ( B ) ChIP with anti-MAF1 antibody ( IP: MAF1 ) was performed in MCF-7 cells subjected to siRNA knockdown of MAF1 ( KD MAF1 ) or simultaneous knockdown of MAF1 and Pol III ( KD M/Pol III ) for 72 hr . Binding of MAF1 was detected at the GDF15 promoter , which diminished after MAF1 knockdown . ( C ) A 3C assay was performed as indicated in the ‘Materials and methods’ , and DNA was subjected to PCR . Chromatin looping was detected after MAF1 knockdown from 3C-3r to 3C-1f ( top panel ) and is shown by a representative gel ( bottom panel ) . ( D ) The induced chromatin looping after MAF1 knockdown ( KD MAF1 ) was diminished when MAF1 underwent simultaneous knockdown with either Pol III ( KD M/Pol III ) or BRF1 ( KD M/BRF1 ) ( top panel ) and is shown by a representative gel ( bottom panel ) . ( E ) ChIP with anti-Pol III antibody ( IP: Pol III ) or anti-Pol II antibody ( IP: Pol II ) was performed in MCF-7 cells subjected to siRNA knockdown . Enhanced binding of Pol III was detected at the GDF15 promoter after MAF1 knockdown , which was abolished when there was simultaneous knockdown of Pol III and MAF1 ( KD M/Pol III ) . ( F ) MAF1 knockdown indicates enhanced binding of serine 5-phosphorylated Pol II , which was abolished when there was simultaneous knockdown of Pol III and MAF1 . All data shown represent the mean ± s . e . m . , n ≥ 3 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 012 We also employed an in vitro transcription assay using HeLa cell nuclear extract to demonstrate the importance of the Pol III promoter element in regulation of GDF15 transcription after MAF1 knockdown . When nuclear extract was pre-incubated with anti-MAF1 antibody to deplete MAF1 during in vitro transcription , GDF15 transcription was significantly upregulated compared with the control with pre-incubation with IgG or no antibody ( Figure 4E ) . We also noticed an MIR repeat element in the 5′ flanking region of GDF15 . As in the case of CDKN1A , deletion of the Pol III A-box element associated with this repeat also abolished the enhancement of GDF15 in vitro transcription when anti-MAF1 antibody was added to the extract ( Figure 4E ) , indicating that this element mediated MAF1 binding . Indeed , the in vitro binding assay using purified MAF1 protein indicated that MAF1 did not bind to this deletion mutant but did bind to the wild-type sequence ( Figure 4B ) . A 3C assay was performed to further investigate whether MAF1 knockdown induces chromatin looping . The analysis indicated that there was chromatin looping between a promoter region and a region that is 12-kb upstream of the GDF15 promoter after MAF1 knockdown ( Figure 7C ) . Furthermore , the looping was abolished under simultaneous knockdown of MAF1 with either Pol III or BRF1 ( Figure 7D ) . Similar to the CDKN1A results , induced binding of Pol III and Pol II to the GDF15 promoter was observed after MAF1 knockdown , whereas the binding was diminished after simultaneous knockdown of Pol III and MAF1 ( Figure 7E , F ) . These results show that transcription of GDF15 , like CDKN1A , is upregulated after MAF1 knockdown by recruiting Pol III , and Pol III is required for chromatin looping at the GDF15 promoter . To demonstrate the role of the Pol III promoter element in Pol III-mediated activation of the Pol II gene , we analyzed the effect of deletion of the Pol III promoter element using a reporter assay . The promoter regions of CDKN1A , GDF15 , and TAF5 , as indicated in the ‘Materials and methods’ , were cloned into the reporter plasmid pGL3 . Promoter-driven luciferase activities of CDKN1A and GDF15 were upregulated , whereas TAF5 promoter-driven expression was not affected after MAF1 knockdown ( Figure 8A ) . Simultaneous knockdown of both Pol III and MAF1 abolished the upregulation caused by knockdown of MAF1 alone ( Figure 8A ) . These results thus recapitulated the results of in vivo endogenous gene analysis . 10 . 7554/eLife . 06283 . 013Figure 8 . Demonstration of MAF1- and Pol III-mediated transcription regulation using a reporter gene assay . ( A ) Promoter regions of Pol II genes were constructed and cloned into pGL3-basic reporter plasmids , as indicated in the ‘Materials and methods’ . Luciferase reporter assays were performed in MCF-7 cells subjected to siRNA knockdown of MAF1 or simultaneous knockdown of Pol III and MAF1 . Results are normalized with β-galactosidase and presented relative to knockdown control cells transfected with pGL3-basic . ( B ) The consensus sequence of the A-box ( −447 to −437 ) in the GDF15 promoter ( −889 to +110 ) was either deleted or mutated . Reporter assays were performed in MCF-7 cells subjected to siRNA knockdown of MAF1 ( KD MAF1 ) . All data shown represent the mean ± s . e . m . , n ≥ 3 , p < 0 . 05 , p < 0 . 01 , *p < 0 . 001 ( t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06283 . 013 Because the majority of SINEs are transcribed by the type II internal Pol III promoter that contains an A-box and B-box ( Okada and Ohshima , 1995 ) , our model indicates that mutation of the Pol III promoter element in the promoter-associated SINE should abolish the enhancement of reporter expression after MAF1 knockdown . To test this possibility , we chose the GDF15 promoter for analysis because it was shown ( Ichikawa et al . , 2008 ) that deletion of the −465 to −429 sequence , which contains the Pol III promoter element , did not affect promoter activity . We mutated the A-box ( −447 to −437 ) in the GDF15 promoter ( −889 to +110 ) in the reporter and examined the effect of mutation on reporter expression . Under regular cell culture conditions , no significant change in GDF15 promoter activity was observed in our results when the A-box was mutated or deleted , in consistent with the results of Ichikawa et al . ( 2008 ) . However , deletion of A-box in the GDF15 promoter diminished the upregulation of the reporter after MAF1 knockdown ( Figure 8B ) . These results further demonstrate that MAF1 represses CDKN1A and GDF15 promoter activity by binding to the Pol III promoter element . Moreover , recruitment of Pol III after MAF1 depletion is crucial for transcription activation of these genes . In this research , we showed that MAF1 bound to promoter-associated SINEs associated with type II Pol III promoters and that depletion of MAF1 enhanced transcription activity and chromatin looping by the recruitment of Pol III along with active Pol II and factors associated with these promoters . Both in vivo gene expression and R-looping analysis as well as in vitro transcription using the HeLa nuclear extract and in vitro binding using purified MAF1 protein revealed that MAF1 represses CDKN1A and GDF15 promoter activity by binding to the SINE repeats within their promoters . This result strongly indicates a novel transcription regulatory mechanism whereby MAF1 also acts as a specific repressor of some Pol II genes by binding to promoter-associated SINEs . Binding specificity was demonstrated by an in vitro DNA binding assay with purified MAF1 to wild-type CDKN1A and GDF15 promoters and the lack of binding to promoters with mutations in the SINE . To the best of our knowledge , this is the first time that MAF1 has been shown to bind to specific DNA sequences . We also demonstrated that recruiting Pol III to SINEs in the 5′ flanking region is required to promote Pol II gene transcription , epigenetic modifications , and chromatin looping after MAF1 knockdown . Recruitment of Pol III , positive regulatory factors , and common transcription factors ( TBP and POLR2E ) of Pol II and Pol III demonstrate a novel mechanism of activating Pol II genes through Pol III-mediated activation mechanism . Gene expression induced by chromatin remodeling through SINEs has also been described in neuronal genes that undergo acetylation of distal promoter SINEs by p300 to translocate genes to transcription factories ( Crepaldi et al . , 2013 ) . In our result , MAF1 knockdown promoted recruitment of p300 , which has been shown to promote acetylation of histone K27 and active transcription by Pol II . However , how extensively this mechanism regulates genes is currently unknown . Histone H3K4me has been suggested to serve as a hallmark of enhancer ( Herz et al . , 2012 ) . Examination of ENCODE database revealed that this epigenetic mark was presented in both 5′ flanking regions of CDKN1A . Indeed the identified p53-binding site located between 1 . 4 kb and 2 . 3 kb upstream of CDKN1A has been identified as the enhancer region ( Melo et al . , 2013; Leveille et al . , 2015 ) . However , the SINE with the MAF1-binding site located at 2 . 65 kb upstream of short promoter exerted no enhancer activity in the luciferase assay prior to or after MAF1 knockdown ( data not shown ) . The chromatin looping we observed for CDKN1A and GDF15 after MAF1 knockdown may be mediated through the proximity and interaction between the sets of transcriptional factors recruited in the two 5′ flanking regions after chromatin remodeling as proposed by Crepaldi et al . ( 2013 ) . Because the background expression of Pol II genes slightly decreased when Pol III was knockdown , Pol III may be able to regulate the expression of some minor alleles without MAF1 being bound to the promoter . However , it is difficult to unequivocally validate this possibility without specific technology that can efficiently separate different alleles in cells . Human genome analysis indicated that 71% of genes contained SINEs in their promoter regions . Microarray analysis showed that 124 genes were upregulated after MAF1 knockdown . Of these , 76% contained SINEs within the promoter region , which indicates the regulation potential of these genes by MAF1 and Pol III . However , microarray analysis of steady-state mRNA level alone is not sufficient to show whether these genes are directly regulated by MAF1 at the transcriptional level or as the result of downstream secondary effects . Further analysis using in vitro transcription , reporter gene analysis , and nuclear run-on would be required to unambiguously establish how general Pol II genes are regulated by MAF1 and Pol III . In vitro transcription using HeLa cell extracts with depletion or addition of specific transcription regulators could provide very strong support of the involvement of specific factors in transcription regulation . Close proximity of Pol III genes to Pol II genes has been observed genome-wide ( Oler et al . , 2010 ) . Active Pol III-transcribed genes and non-coding RNAs often associate with Pol II transcription start sites . Pol II and its associated epigenetic marks are also present at active Pol III-transcribed genes ( Barski et al . , 2010; Raha et al . , 2010; Canella et al . , 2012 ) . This shows that there is common epigenetic regulation between these two types of transcription units , and the polymerases may work with one another to regulate gene expression . Indeed , cross-talk between Pol III and Pol II transcription factors , such as TFIIS in Pol III transcription , has also been reported in yeast and mice ( Ghavi-Helm et al . , 2008; Carriere et al . , 2011 ) . The core Pol III transcription factor TFIIIC can also directly regulate transcription from a Pol II promoter ( Kleinschmidt et al . , 2011 ) . Binding of TFIIIC to SINE promoters has been shown to mediate the relocation and transcription of neuronal genes ( Crepaldi et al . , 2013 ) . RPPH1 , to which BRF2 , Pol III , GTF2B , and Pol II bind , can be transcribed by either Pol II or Pol III ( James Faresse et al . , 2012 ) . Our results indicate that Pol III and Pol II association may have functional relevance for genome functional organization , because simultaneous knockdown of both Pol III and MAF1 diminished the induced active transcription caused by knockdown of only MAF1 . Furthermore , enhanced binding of TBP to TFIIIB and TFIIB can lead to formation of Pol III and Pol II complexes to initiate transcription ( Zhao et al . , 2003 ) . We propose that this type of SINE-associated-Pol II promoter architecture may introduce an additional layer of control in gene expression . Recently , MAF1 was shown to be a negative regulator of transcription of all three polymerases , Pol I , Pol II , and Pol III , through mediating TBP expression ( Johnson et al . , 2007 ) . Johnson et al . showed that MAF1 binds to the Elk-1-binding site of the TBP promoter to prevent the binding of Elk-1 . Indeed , there is a SINE with an A-box ( −10 to +1 ) and B-box ( −105 to −95 ) that encompass the Elk-1-binding site of the TBP promoter . In our analysis , MAF1 knockdown only slightly increased TBP expression ( 1 . 6-fold ) compared with the results reported by Johnson et al . ( twofolds ) . This may be due to the already high expression of TBP in cell lines , and we did not detect binding of MAF1 to this active promoter . Based on our results , we propose the following mechanism of control of Pol II gene transcription by MAF1 and Pol III: Before removal of MAF1 from SINEs , Pol II is in a transcriptionally engaged but paused state , where TBP/TFIIB is pre-assembled and remains at the promoter ( Guenther et al . , 2007; Kwak et al . , 2013; Venters and Pugh , 2013 ) . Removal of MAF1 by knockdown then leads to recruitment of TFIIIB through enhanced binding of TBP; the shared surface of TBP then directs both Pol II and Pol III binding through association with TFIIB and TFIIIB , respectively . Further recruitment of active regulatory factors would then induce transcription by Pol II and Pol III . This model is consistent with those proposed by previous studies , which were based on a component of TBP-associated complexes , p300 , interacting with SET1C-coupled histone modifications to activate CDKN1A transcription ( Abraham et al . , 1993; Tang et al . , 2013 ) . Moreover , our model is also supported by a previous study on the relocation of inducible neuronal genes to transcription factors that involve acetylation of distal promoter SINEs by p300 ( Crepaldi et al . , 2013 ) . MCF-7 , MCF-10A , and MDA-MB-231 cell lines were originally obtained from ATCC ( Manassas , VA ) , and cultured in RPMI , HuMEC and DMEM medium ( Invitrogen; Waltham , MA ) , respectively . HCT-116p53+/+ ( wild-type ) and HCT116p53−/− ( p53-null ) cell lines were originated from Bert Vogelstein ( John Hopkins University ) and cultured in McCoy's 5A medium ( Bunz et al . , 1998 ) . Each medium was supplemented with 10% of fetal bovine serum and incubated in a humidified 37°C incubator with 5% CO2 . Knockdown assay was performed using siRNA obtained from MISSION RNA ( Sigma-Aldrich; St . Louis , MO ) . Inhibition of expression of MAF1 ( [#1] SASI_Hs01_00135954 , [#2] SASI_Hs01_00135956 and [#3] SASI_Hs01_00135958 ) , Pol III ( POLR3A ) ( [#1] SASI_Hs01_00046568 , [#2] SASI_Hs01_00046571 and [#3] SASI_Hs01_00046572 ) , BRF1 ( SASI_Hs01_00131187 ) , CFP1 ( SASI_Hs02_00322879 ) , and TBP ( SASI_Hs01_00122768 ) was achieved by transfection with Lipofectamine RNAiMax ( Invitrogen ) according to the manufacturer's protocol for 72 hr . MISSION siRNA Universal Negative Control ( Sigma ) was used as knockdown control . Cells were transfected in serum-free medium . After 8 hr , the siRNA containing medium was replaced with complete medium . Cells were lysed at 4°C in RIPA lysis buffer ( 50 mm Tris-HCl , pH 7 . 2 , 150 mm NaCl , 5 mm EDTA , 1% [wt/vol] NP-40 , 1% [wt/vol] SDS and protease and phosphatase inhibitor mixtures [Roche Applied Science; Penzberg , Germany] ) . The lysates were cleared by centrifugation ( 15 , 000×g for 15 min ) , resolved on a 10% SDS-polyacrylamide gel , and transferred onto a nitrocellulose membrane . The antibody dilutions used were rabbit anti-POLR3A ( 1:1000; ab96328 , Abcam; Cambridge , England ) , rabbit anti-MAF1 ( 1:1000; GTX106776 , Acris; Herford , Germany ) , rabbit anti-CDKN1A ( 1:1000; ab18209 , Abcam ) , and mouse anti-tubulin ( 1:10 , 000; ab7291 , Abcam ) . Cells were grown to 85% confluence in 6 cm tissue culture dish . Each 6 cm dish was washed with 1× phosphate buffered saline ( PBS ) for three times . Total RNA was extracted using TRIreagent ( Invitrogen ) protocol . The integrity of the RNA extract was checked by 1 . 2% ( wt/vol ) agarose gel electrophoresis and the concentration of RNA was estimated by ultraviolet spectrophotometry . Affymetrix microarray was performed using Human U133 plus 2 . 0 ( Affymetrix; Santa Clara , CA ) . Details of the methods for RNA quality , sample labeling , hybridization , and expression analysis were according to the manual of Affymetrix Microarray Kit . All Affymetrix data are MIAME compliant and that the raw data have been deposited in a MIAME compliant database , GEO . The microarray data were deposited at the NCBI GEO website ( GEO accession number GSE42239 ) . Reverse transcription was performed by using superScript III RNase H- Reverse Transcriptase ( Invitrogen ) and random hexamer according to the manufacturer's protocol . Quantitative PCR was performed using KAPA SYBR FAST ( KK4603 , KAPA Biosystems; Wilmington , MA ) on ABI StepOnePlus Real-Time PCR System ( Invitrogen ) . All reactions were performed in triplicate with KAPA SYBR FAST plus 10 μM of both the forward and reverse primer according to the manufacturer's recommended thermo cycling conditions , and then subjected to melting curve analysis . The calculated quantity of the target gene for each sample was divided by the average sample quantity of the housekeeping genes , glyceraldehydes-3-phosphate dehydrogenase ( GAPDH ) or 18S to obtain the relative gene expression . MCF-7 knockdown cells were collected by trypsinization and washed twice with ice-cold PBS . The cells were resuspended in 0 . 3 ml of PBS and fixed by slowly adding 3 ml of 70% cold ethanol . Cells were fixed at −20°C for 1 hr . The fixed cells were washed with ice-cold PBS and rehydrated for 15 min . After centrifuging at 200×g for 5 min , cells were resuspended in 0 . 1 mg/ml of propidium iodide and 0 . 6% of Triton X-100 in 500 μl of PBS . Then add 500 μl of 2 mg/ml of RNase A and incubate in the dark for 45 min . Data were collected using a FACScan flow cytometry system ( BD; Franklin lakes , NJ ) . Nuclear run-on reactions were performed by supplying biotin-16-UTP to nuclei , and labeled transcripts were bound to streptavidin-coated magnetic beads as described by Patrone et al . ( 2000 ) with minor modifications . Nuclei were prepared from MCF-7 cells by resuspension in Nonidet P-40 lysis buffer ( 10 mM HEPES , pH 7 . 3 , 10 mM NaCl , 3 mM MgCl2 , 150 mM sucrose , and 0 . 5% Nonidet P-40 ) . Nuclei were isolated , and the pellets were resuspended in 1 ml of glycerol buffer ( 50 mM Tris-Cl , pH 8 . 3 , 40% glycerol , 5 mM MgCl2 , and 0 . 1 mM EDTA ) . 1 ml of transcription buffer ( 20 mM Tris-Cl , pH 8 . 0 , 200 mM KCl , 5 mM MgCl2 , 4 mM dithiothreitol , 4 mM each of ATP , GTP , and CTP , 200 mM sucrose , and 20% glycerol ) was added in the nuclei along with 10 μl of biotin-16-UTP or UTP for run-on reaction or negative control , respectively ( Roche ) . After incubation at 29°C for 30 min , the reaction was terminated by the addition of 12 μl of 250 mM CaCl2 , and 12 μl of RNase-free DNase I and incubated at 29°C for 10 min . To purify RNA , a TRIreagent extraction , phenol-chloroform extraction , and isopropanol ( Sigma ) precipitation were then performed . A small aliquot ( 5 μl from a total of 50 μl ) was saved as input control . Dynabeads M-280 streptavidin ( Invitrogen ) were mixed with an equal volume of the isolated RNA samples for 20 min at 42°C for 20 min and 2 hr at room temperature . After washing with 15% formamide and 2× SSC , the beads were resuspended in 45 μl of nuclease-free water . Reverse transcription was performed by using superScript III RNase H—Reverse Transcriptase ( Invitrogen ) . Total cDNA was then synthesized by means of random hexamer primed reverse transcription of captured molecules . The gel pictures were quantified with ImageJ ( provided by NIH: http://imagej . nih . gov/ij/ ) . The purified run-on products where normalized with internal control ( GAPDH ) to obtain the relative transcription levels for each gene . Knockdown assay was performed using siRNA obtained from MISSION RNA ( Sigma ) . Inhibition of expression of Pol III ( SASI_Hs01_00046568 ) and MAF1 ( SASI_Hs01_00135954 ) was achieved by transfection with Lipofectamine RNAiMax ( Invitrogen ) according to the manufacturer's protocol for 72 hr . DNA purification and single-stranded R loop foot-printing were carried out as previously described with slight modifications ( Yu et al . , 2003 ) . 500 ng of purified genomic DNA was bisulfite converted by adding CT Conversion Reagent from the EZ DNA Methylation-Gold Kit ( Zymo Research; Irvine , CA ) at 37°C for 16 hr in the dark . PCR amplified region for cloning is shown in Figure 2A , F as foot-printing region . The PCR product was gel eluted and ligated to sequencing vector yT&A ( Sigma ) . Approximately , 20 individual clones were sequenced for all PCR products , and the sequencing data were analyzed and aligned to CDKN1A or ACTB genomic sequence . The sequence of the beginning and end of each clone is trimmed due to low quality of sequencing . A background conversion ( approximately 5% of cytosine ) may be seen possibly due to DNA breathing during the prolonged incubation at 37°C in our data and others ( Yu et al . , 2003 ) . Approximately , 1–2 clones showed both cytosine to thymine and guanine to adenine conversions , which is known as ‘mosaic molecules’ ( Yu et al . , 2003 ) . ChIP assay was performed according to the manufacturer's protocol ( Upstate Biotechnology , Inc . ; Lake Placid , NY ) with slight modifications . Human MCF-7 cells were fixed with 1% of formaldehyde at room temperature for 10 min . The cells were lysed and the chromatin was sonicated to 200–500 bp fragments by Bioruptor sonicator ( cycle condition of 25 s on and 25 s off in a total of 25 min at highest output ) . Chromatin was immunoprecipitated by using Pol III ( ab96328 , Abcam ) , Pol II ( ab5131 , Abcam ) , MAF1 ( GTX106776 , Acris ) , H3K4me3 ( 04–745 , Millipore; Billerica , MA ) , H3K27me3 ( ABE44 , Millipore ) , TBP ( ab28175 , Abcam ) , H3K9Ace ( 06–942 , Millipore ) , H3K27Ace ( 07–360 , Millipore ) , CFP1 ( ABE211 , Millipore ) , p300 ( 05–257 , Millipore ) , POLR2E ( ab180151 , Abcam ) BRF1 ( ab74221 , Abcam ) or IgG ( ab46540 , Abcam ) antibody , with 10 μg/ml of BSA and 50 μl of Dynabeads Protein A and G ( Invitrogen ) for overnight at 4°C . The beads were washed once with each washing buffer , including low salt immune complex wash buffer , high salt immune complex wash buffer , and LiCl immune complex wash buffer , and twice with 1× TE buffer . Precipitates were eluted with 1% of SDS and 100 mM of NaHCO3 . Proteinase K was added to the samples , and rotated at 65°C for 2 hr followed by 95°C for 10 min and cooled down to room temperature . RNase A was added and samples were incubated at 37°C for 1 hr . After genomic DNA extraction , qPCR was performed . The degree of enrichment is calculated relative to the ratio of signals obtained in the input DNA fraction subtracting IgG-immunoprecipitated DNA . 3C assay was performed according to ( Dekker et al . , 2002 ) with some modifications . MCF-7 cells were fixed in 2% formaldehyde for 10 min at room temperature and quench with 0 . 125 M glycine . After centrifugation for 15 min at 3500 rpm , the cells were suspended in lysis buffer ( 10 mM Tris-HCl pH 8 . 0 , 10 mM NaCl , 0 . 2% Nonidet P-40 and 1:500 Complete protease inhibitor cocktail; Roche ) for 90 min on ice . Next , the nuclei were pelleted by centrifugation for 15 min at 2500 rpm , resuspended in 500 μl of 1× NEB buffer 4 plus 0 . 3% SDS and incubated at 37°C for 1 hr . After the addition of Triton-X to a final concentration of 1 . 8% to sequester the SDS , the mixture was incubated at 37°C for 1 hr , which was followed by the addition of 800 U of PstI and incubation at 37°C overnight to digest the chromatin . The reaction was terminated by adding SDS to a final volume of 1 . 6% and then the solution heated to 65°C for 20 min . Ligation of DNA in situ was carried out using 0 . 5–2 . 0 ng/μl of chromatin in 800 μl of ligation buffer ( NEB; Ipswich , MA ) plus 1% Triton-X and 30 Weiss Units of T4 ligase ( NEB ) for 4 hr at 16°C . After reversing of the crosslinks with proteinase K digestion at 65°C overnight , the DNA was purified by phenol-chloroform extraction and ethanol precipitation . The ligation products were detected by PCR using primers located near Pst1 cutting sites . The PCR products were purified from an agarose gel , cloned and sequenced . CDKN1A ( with or without MIR3 ) , ACTB , GDF15 ( including deleted or mutated A-box ) and RPPH1 template DNA was obtained by PCR followed by gel elution ( Qiagen; Venlo , Netherlands ) according to the manufacturer's protocol . The deletion of MIR3 was performed as described by PCR-mediated deletion and checked by sequencing ( Lee et al . , 2004 ) . The purified DNA was further used for in vitro DNA binding reactions as described previously with slight modifications ( Britten , 1996; Toth and Biggin , 2000 ) . The in vitro protein-DNA binding assay coupled with immunoprecipitation was performed as following: 20 ng of DNA template , 400 ng of MAF1 protein ( His tag ) ( 80R-1955 , Cantor Fitzgerald; New York , NY ) , 400 ng of Anti-6X His tag antibody ( ab18184 , Abcam ) , protease inhibitor ( 539134 , Calbiochem; La Jolla , CA ) , and 200 ng of BSA was added into 50 μl of binding buffer ( 20 mM HEPES [pH7 . 6] , 150 mM NaCl , 0 . 25 mM EDTA , 10% glycerol , 0 . 2% NP40 , and 1 mM DTT ) . A negative control was performed by substituting IgG antibody for Anti-6× His tag antibody ( Protein + IgG ) or with only the Anti-6× His tag antibody for the MAF1 protein ( Ab only ) . The mixture was rotated at 4°C for 10 min and on ice for 30 min . 10 μl of Dynabeads Protein G ( Invitrogen ) was added to the mixture and rotated at 4°C for 10 min and on ice for 30 min . The immunoprecipitated DNA-protein complexes were then washed twice with washing buffer ( 20 mM Tris [pH 7 . 5] , 0 . 25 mM EDTA , 10% glycerol , and 0 . 2% NP40 ) and once with TE buffer by each rotating at 4°C for 5 min . Elution was performed with 1% of SDS and 0 . 1 M of NAHCO3 . Input DNA was prepared as 1 ng of template DNA ( 5% of 20 ng ) . Proteinase K was added to the samples , and rotated at 65°C for 2 hr followed by 95°C for 10 min and cooled down to room temperature . DNA isolated from immunoprecipitated protein-DNA complex was subjected to qPCR . The degree of enrichment is calculated relative to the ratio of signals obtained in the input DNA fraction . Biotin-labeled ( labeled at 5′ ) and non-labeled CDKN1A ( 5′-AATCAACAACTTTGTATACTTAAGTTCAGTGGACCTCAATTTCCTCATCTGTGAAATAAA-3′ ) as well as mutated A-box template DNA ( 5′-AATCAACAACTTTGTATA CTTCCCATCCCAAAACCTCAATTTCCTCATCTGTGAAATAAA-3′ ) was obtained by oligo synthesis from Genomics ( Houston , TX ) . The oligos were annealed and used for in vitro DNA-protein binding assay by the DNA-Protein Binding Assay Kit ( Colorimetric ) provided by Abcam ( ab117139 ) . The assay was performed according to manufacturer's protocol . In brief , 40 ng of biotin-labeled DNA template and 500 ng of MAF1 protein ( His tag ) ( 80R-1955 , Fitzgerald ) was used for the binding assay . For competition assay , 200 ng of competitor DNA was added to the mixture . Anti-6× His tag antibody ( ab18184 , Abcam ) and Goat anti-Mouse IgG2b heavy chain ( HRP ) antibody ( ab97250 , Abcam ) were prepared and added according to manufacturer's protocol . Blank control was performed without the addition of protein as specified by the kit and the degree of enrichment is calculated by subtracting with blank control . The upstream promoter regions of CDKN1A ( −864 to +41 of NM_001220778 for short form and −1249 to +92 of NM_001220777 for long form ) , GDF15 ( −889 to +110 ) , and TAF5 ( −998 to +157 ) genes were cloned into the pGL3-basic Reporter Vector ( Promega; Fitchburg , WI ) . Knockdown assay was performed as mentioned above for 24 hr before MCF-7 cells were transfected with the plasmids using Lipofectamine LTX ( Invitrogen ) , along with a plasmid expressing β-galactosidase for normalization . The plasmids were transfected for 48 hr , and the cells were lysed and luciferase assay was conducted using the Luciferase Assay System ( Promega ) using a fluorimetric plate reader . In vitro transcription was performed by using HeLaScribeR Nuclear Extract in vitro Transcription System ( Promega Cat . #E3110 ) according to the manufacturer's protocol with slight modifications . Template DNA was prepared by linearizing the constructed promoter region of CDKN1A , GDF15 , and TAF5 as used in Luciferase assay , as well as RPPH1 promoter as used in in vitro MAF1 binding assay . In vitro transcription was performed by incubation with linear form of constructed promoter region with nuclear extract , transcription buffer , magnesium ion , GTP , CTP , ATP , biotin-16-UTP , RNase inhibitor , and 30 μg of yeast tRNA . Negative control was performed by incubation with non-biotin labeled NTPs . 0 . 2 μg of α-amanitin was added during in vitro transcription for inhibition of Pol II transcription . 3 μg of MAF1 protein ( His tag ) ( 80R-1955 , Fitzgerald ) used in in vitro MAF1 binding assay was pre-incubated with template DNA before adding nuclear extract to enable binding of MAF1 to the template DNA . Anti-Pol III ( ab96328 , Abcam ) , anti-Pol II ( ab5131 , Abcam ) , anti-MAF1 ( GTX106776 , Acris ) , or anti-IgG ( ab46540 , Abcam ) antibody was pre-incubated with nuclear extract for 15 min to deplete the target protein of interest . After incubation at 30°C for 1 hr , the reaction was terminated by the addition of 175 μl of HeLa Extract Stop Solution ( Promega Cat . #E3110 ) . A TRIreagent extraction , phenol-chloroform extraction , and isopropanol ( Sigma ) precipitation were then performed to purify RNA . A small aliquot ( 2 μl from a total of 22 μl ) was saved as ‘total nuclear RNA’ for each condition . The biotinylated RNA was isolated using streptavidin-coated magnetic beads as described in Run-on assay . Reverse transcription was performed by using superScript III RNase H- Reverse Transcriptase ( Invitrogen ) . Total cDNA was then synthesized by means of random hexamer primed reverse transcription of captured molecules .	An organism's genetic material is made of segments of DNA called genes , which contain instructions to make proteins . First , copies of the DNA are made using another molecule called ribonucleic acid ( RNA ) in a process known as transcription . Then the RNA is used as a template to make a protein . During transcription , enzymes called RNA polymerases move along the DNA to produce the RNA copies . When a cell is actively growing it needs large quantities of new proteins to be made , and so the level of transcription is higher . However , if a cell experiences stress caused by adverse environmental conditions ( e . g . , high temperatures ) , it can conserve resources by shutting down transcription . For example , one RNA polymerase—called Pol III—makes RNA copies with the help of a protein called BRF1 and several other proteins . However , when a cell is under stress , another protein called MAF1 can interfere with transcription by binding to BRF1 , which prevents it from interacting with Pol III . Previous work has suggested that MAF1 can also inhibit the activity of another RNA polymerase called Pol II , but it was not clear how this could work . Lee et al . studied the effect of MAF1 on transcription in human cells . The experiments show that MAF1 blocks the transcription of many genes that are transcribed by Pol II , including one called CDKN1A . CDKN1A is involved in regulating many important processes , including the growth of cells and cell death . Cells that produced lower amounts of MAF1 had higher levels of CDKN1A transcription , and several proteins—including Pol II , Pol III and BRF1—were more able to bind to this gene . However , this effect was not observed in cells that also produced lower levels of Pol III or BRF1 , suggesting that Pol III is needed for Pol II to be able to transcribe CDKN1A . Taken together , Lee et al . 's findings suggest that MAF1 inhibits the transcription of CDKN1A , and possibly other genes transcribed by Pol II , by regulating the activity of Pol III . Further research is needed to understand the details of how this works .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2015	MAF1 represses CDKN1A through a Pol III-dependent mechanism
Animal–animal recognition within , and across species , is essential for predator avoidance and social interactions . Despite its essential role in orchestrating responses to animal cues , basic principles of information processing by the vomeronasal system are still unknown . The medial amygdala ( MeA ) occupies a central position in the vomeronasal pathway , upstream of hypothalamic centers dedicated to defensive and social responses . We have characterized sensory responses in the mouse MeA and uncovered emergent properties that shed new light onto the transformation of vomeronasal information into sex- and species-specific responses . In particular , we show that the MeA displays a degree of stimulus selectivity and a striking sexually dimorphic sensory representation that are not observed in the upstream relay of the accessory olfactory bulb ( AOB ) . Furthermore , our results demonstrate that the development of sexually dimorphic circuits in the MeA requires steroid signaling near the time of puberty to organize the functional representation of sensory stimuli . Throughout the animal kingdom , dedicated signals allow individuals to identify and distinguish members of their own species: songbirds attract mates during bouts of singing , cichlid fish communicate dominance status with dramatic changes in body coloration , and rodents signal their social status through the emission of chemical cues ( Brainard and Doupe , 2002; Tinbergen , 1951; Maruska and Fernald , 2011; Dulac and Kimchi , 2008 ) . In turn , the execution of instinctive behaviors such as mating , parenting , territorial defense , and predator avoidance is triggered and modulated by sensory signals within a given physiological context , including the sex , endocrine , or developmental status of an individual ( Tinbergen , 1951; Insel and Fernald , 2004 ) . However , the brain circuits that organize species- and sex-specific instinctive behaviors in response to animal cues are still poorly characterized . Chemosensory communication in rodents offers a unique opportunity to explore the organizational and functional principles of social behavior circuits . Impairment in either vomeronasal or olfactory signaling results in dramatic deficits in predator avoidance and social communication ( Bean and Wysocki , 1989; Stowers et al . , 2002; Mandiyan et al . , 2005; Yoon et al . , 2005; Kimchi et al . , 2007; Kobayakawa et al . , 2007; Isogai et al . , 2011 ) , demonstrating that chemical cues are essential mediators of animal–animal recognition in mice and that the main olfactory system and the vomeronasal pathway work in concert to guide behavioral responses to chemosensory input ( Dulac and Wagner , 2006; Dulac and Kimchi 2008 ) . The main olfactory epithelium ( MOE ) and vomeronasal organ ( VNO ) are distinguished by the types of chemical cues that each detects: volatile compounds are primarily detected by the MOE while non-volatile signals are preferentially processed by the VNO . At the circuit level , neurons in the MOE project to the main olfactory bulb ( MOB ) , which in turn projects axons to the lateral amygdala , primary olfactory cortex , and ultimately to areas involved in the cognitive processing of odorants ( Sosulski et al . , 2011 ) . In addition , neurons expressing the neuropeptide GnRH , a master regulator of reproduction in vertebrates , receive major inputs from primary olfactory areas ( Boehm et al . , 2005; Yoon et al . , 2005 ) . In contrast , vomeronasal circuits largely bypass cognitive centers , as sensory neurons of the VNO send axons to the accessory olfactory bulb ( AOB ) , which in turn projects to nuclei of the medial amygdala ( MeA ) , the bed nucleus of the stria terminalis , and the hypothalamus ( Petrovich et al . , 2001 ) , areas involved in controlling innate behaviors and neuroendocrine changes . Recent studies have begun to explore the basic principles of vomeronasal information coding that lead to specific social and defensive behavioral responses . The identification of key components of VNO signal transduction , such as the TRPC2 ion channel ( Liman et al . , 1999 ) , and the members of the V1R and V2R families of vomeronasal receptors ( VRs ) ( Dulac and Axel , 1995; Herrada and Dulac , 1997; Matsunami and Buck , 1997; Ryba and Tirindelli , 1997; Dulac and Torello , 2003 ) , has established the molecular foundation of vomeronasal sensing . In vitro recording and imaging of VNO neuronal activity after exposure to various stimuli , as well as in vivo characterization of the vomeronasal receptor response profile to a wide range of animal cues , have revealed that distinct populations of VNO receptors identify ethologically relevant chemosignals with high specificity and that signal detection in the VNO plays an essential role in distinguishing behaviorally relevant sensory information ( Holy et al . , 2000; Leinders-Zufall et al . , 2000; Nodari et al . , 2008; Leinders-Zufall et al . , 2009; Haga et al . , 2010; Isogai et al . , 2011; Turaga and Holy , 2012 ) . In contrast to the apparent specificity of VNO responses for distinct animal cues , a large proportion of output neurons in the AOB , the primary targets of VNO sensory projections , were shown to respond to multiple classes of stimuli ( Luo et al . , 2003; Hendrickson et al . , 2008; Ben-Shaul et al . , 2010 ) . In particular , many AOB mitral/tufted cells respond with similar strength to cues such as female and predator odors , which are clearly associated with opposing behavioral outcomes ( Ben-Shaul et al . , 2010 ) . The complex responses of AOB mitral/tufted cells , together with their distinctive multi-branched dendritic trees , suggest that they integrate sensory information across distinct vomeronasal receptors ( Wagner et al . , 2006 ) . Ultimately , downstream vomeronasal centers that control behavioral outputs must be able to interpret and disambiguate the relevant information from these broad patterns of AOB activity . Several lines of evidence suggest that the MeA plays a central role in the vomeronasal–sensorimotor transformation that leads to specific behavioral responses . The MeA is the primary recipient of AOB inputs , it projects to distinct nuclei of the hypothalamus involved in social and defensive responses ( Petrovich et al . , 2001; Choi et al . , 2005; Lin et al . , 2011 ) and disruptions in MeA signaling cause profound deficits in social and predator recognition ( Ferguson et al . , 2001; Li et al . , 2004 ) . However , despite this central position in the vomeronasal sensory pathway , little is known about how neuronal activity in the MeA participates in the transformation of AOB inputs into behaviorally relevant signals . We have investigated the characteristics of MeA neuronal activity in both male and female mice in response to conspecific and heterospecific chemosignals . Our results uncovered several emergent features in the MeA representation of olfactory information that are not present in either VNO or AOB responses . These features suggest how key sensory parameters relevant for sex-specific , social and defensive behaviors are extracted from the complex activity pattern of the AOB . Our findings provide significant insights into the neural processing of social cues and open new avenues for further understanding the emergence of sex-specific behaviors . To investigate how the MeA responds to chemosensory cues , we combined multisite extracellular recordings with sensory stimulation of the VNO in anesthetized mice ( Figure 1A ) . Linear electrode arrays ( Neuronexus ) were positioned dorsal to the MeA based on stereotaxic coordinates ( 1 . 7-2 . 0 mm lateral and caudal from bregma; Franklin and Paxinos , 2007 ) and advanced until the electrode tip reached the ventral brain surface . This targeting strategy was intentionally chosen to focus on the posterior MeA ( MeApd and MeApv ) , allowing simultaneous sampling of single and multi-unit activity from 32 evenly spaced sites distributed between MeA subnuclei with distinct functions . MeA units responsive to sensory stimuli were localized to the ventral surface of the brain and extended approximately 1 mm dorsally , consistent with the anatomy of the posterior MeA . The locations of all recording sites were confirmed through postmortem histology of dye-labeled electrode tracks ( Figure 1B ) . 10 . 7554/eLife . 02743 . 003Figure 1 . Experimental system for recording MeA sensory responses . ( A ) Vomeronasal and olfactory structures are shown in yellow and blue , respectively . Multichannel electrophysiological probes are stereotaxically positioned in the MeA to continuously record neural responses to sensory stimulation . VNO stimulus presentation ( orange arrow ) is achieved by placing nonvolatile stimuli in the nostril followed by electrical stimulation of the sympathetic nerve trunk ( SNT ) to activate the VNO pump and permit access of stimuli into the VNO . VNO stimuli are washed out through the NPD . MOE stimulation is achieved by controlling airflow of volatile stimuli into the nostril ( blue arrow ) , which access the MOE , and are eliminated through a tracheotomy . ( B ) Diagram illustrating a coronal section through the posterior MeA with red dots indicating the expected dorsal–ventral distribution of recording sites . A single fluorescent electrode tract accurately targeted to MeA is shown in the inset . ( C ) Timecourse of VNO and MOE stimulation trials ( top ) . Electrophysiological signals recorded from a single MeA electrode during four successive trials reveal a well-isolated unit responding only to female stimuli following electrical stimulation of SNT ( stimulation artifacts are evident at 20 s ) . ( D ) Percentage of MeA units responding to VNO vs MOE stimulation with chance rates indicated by a horizontal dashed line . ( E ) Sagittal section of whole brain showing DAPI staining and the site of MeA FluoroGold iontophoresis . ( F ) Dense retrograde labeling of AOB projection neurons . ( G ) Fraction of AOB ( 99 . 8% ) and MOB ( 0 . 2% ) neurons that are retrogradely labeled by FluoroGold iontophoresis in the MeA . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 00310 . 7554/eLife . 02743 . 004Figure 1—figure supplement 1 . Clustering and analysis of multichannel electrophysiological recordings . ( A ) Spike waveforms of four MeA units as recorded on two channels ( channels on which signals from these units were not detected are not shown for simplicity ) . ( B ) Projections of spike waveforms on the first Principal components for each of the two channels shown in A . Even in this partial representation ( projections on additional Principal Components are not shown ) four clusters are clearly evident ( same color scheme as in A ) . Initial clusters were generated by the KlustaKwik program and then adjusted manually . ( C ) Histograms showing the distribution of interspike intervals within each cluster . Units a and d were classified as ‘multi-units’ as they did not show a clear refractory period . Units b and c showed negligible refractory period violations ( <4 ms ) and were accordingly classified as ‘single units’ . ( D ) The average spike waveform for all well-isolated MeA units . The lighter lines indicate STD . ( E ) Distribution of the spike maximum to spike minimum ratio for MeA units . ( F ) Distribution of spike widths ( distance between the minimum and maximum ) for all well-isolated MeA units . ( G ) Scatterplot showing the relationship between spike width and the min/max ratio . ( H ) Spike waveforms for four AOB units ( as recorded on two channels ) . ( I ) Principal component projections for the four clusters depicted in H . ( J ) ISI histograms for the units shown in H and I . Unit b was classified as ‘multi-unit’ due to significant violations of the refractory period . The three other units ( a , c , d ) recorded by these channels were classified as single units . ( K ) Average spike waveform for all well-isolated AOB units , lighter lines indicate STD . ( L ) Distribution of spike maximum to spike minimum ratio for AOB units . ( M ) Distribution of spike widths for all well-isolated AOB units . ( N ) Scatterplot showing the relationship between spike width and the min/max ratio for AOB units . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 00410 . 7554/eLife . 02743 . 005Figure 1—figure supplement 2 . Baseline electrophysiological characteristics of MeA responses . ( A ) Firing rates ( spikes/second ) for the 20 s prior to VNO stimulation are plotted against the firing rate during the 40 s after stimulus application . Each point indicates an isolated single unit with the most significant ( p<0 . 01 , Nonparametric ANOVA ) stimulus induced response plotted for the post stimulus firing rate . Nearly all units lie above the line of unity ( black diagonal ) indicating that VNO stimulation positively drives MeA activity . ( B ) Single MeA units were grouped based on the pre stimulus firing rate , and the average response selectivity is plotted for units within each of these groups . A clear correlation was observed with less active neurons responding more specifically ( black line; R = −0 . 31: p<0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 00510 . 7554/eLife . 02743 . 006Figure 1—figure supplement 3 . MOE-driven responses in the MOB and PLCO . To ensure that the MOE stimulation used in our anesthetized preparation was capable of driving responses in olfactory areas , we recorded from the MOB and PLCO . The top three rows are single units recorded from the MOB and the bottom row is a unit recorded from olfactory amygdala . Each column indicates the presentation of a different stimulus . For each panel , raster plots indicating the timing of individual action potentials elicited by multiple presentations of the same stimulus ( shaded region ) . Histograms of the mean response and standard error for the same data are shown below each raster plot . Time zero indicates alignment to the start of stimulus presentation . All units responded significantly to at least one stimulus ( p<0 . 01; Nonparametric ANOVA ) . Gray boxes indicate that a particular stimulus was not presented during that unit's recording session . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 006 Single units were distinguished from multi-unit activity based on spike shape , separation of principal component projections , and the presence or absence of a refractory period between successive spikes ( Figure 1—figure supplement 1; Harris et al . , 2000; Hazan et al . , 2006 ) . In most experiments , ∼2–6 recording sites were located dorsal to the MeA and were easily distinguished from the units in the MeA , which remain quiet in the absence of vomeronasal stimulation . Multiple MeA units were recorded simultaneously during each recording session ( 17 . 5 ± 11 . 7 units/session ) with ∼40% ( 6 . 6 ± 6 . 4 units/session ) identified as single units . In this study , 197 units ( 82 single units; 115 multi units ) recorded from the MeA of 55 adult BALB/c mice ( 10–12 weeks old; sexually naive ) responded to sensory stimuli ( p<0 . 01 , non-parametric ANOVA ) . Baseline firing rates of single MeA units were typically very low ( 0 . 21 ± 0 . 02 Hz; Figure 1—figure supplement 2 ) . Our relatively unbiased multi-electrode sampling approach allowed identification of many neurons that did not display any spiking activity prior to stimulus presentation . Upon VNO stimulation , MeA units responded in stimulus-locked and stimulus-specific manner ( Figure 1C ) that resembled sensory responses observed in the AOB in both latency and duration ( Ben-Shaul et al . , 2010 ) . MeA units typically increased their firing rate after a preferred stimulus was presented reaching a peak firing rate ( 8 . 1 ± 0 . 31 Hz ) ∼5 s after stimulus presentation . Units with significant responses were identified by comparing the spike rates prior to stimulus presentation to spike rates following stimulus presentation using a non-parametric ANOVA performed at the significance level of p≤0 . 01 ( see ‘Materials and methods’ ) . The MeA has been described as receiving main olfactory projections in addition to vomeronasal input ( Kang et al . , 2009; Martinez-Marcos , 2009 ) . This prompted us to investigate the respective contributions of MOE and VNO stimulations to MeA responses . MOE stimulation elicited clear olfactory-evoked neuronal activity in the MOB and the posterolateral cortical amygdala ( PLCO ) in response to volatile urinary cues and odorants ( Figure 1—figure supplement 3 ) . However , while we observed robust VNO-mediated responses in the MeA , we did not detect stimulus-specific responses for MOE-delivered stimuli ( units responsive to MOE stimuli at p≤0 . 01: 1 . 1% , 95% CI: 0 . 8–1 . 6%; Figure 1D ) . Rather , in agreement with Samuelsen and Meredith ( 2009 ) and Miyamichi et al . ( 2011 ) , we found little evidence that distinct volatile stimuli elicit specific sensory responses in the MeA . Therefore , the vomeronasal system emerged as the dominant sensory input to the MeA units we recorded . We further investigated MOB to MeA connectivity by iontophoretically injecting FluoroGold into the MeA at the same stereotaxic location used for electrophysiology . All injection sites ( 6 animals ) encompassed most of the posterior MeA , with more diffuse labeling extending both anterior and posterior ( Figure 1E ) . Rare neurons were found retrogradely labeled in the posterior MOB , consistent with previous reports ( Kang et al . , 2011 ) . However , we found that FluoroGold retrogradely labeled ∼100 times more AOB than MOB neurons , clearly demonstrating the dominance of AOB inputs , especially in light of the much larger number of MOB compared to AOB neurons ( Figure 1F , G; see Miyamichi et al . , 2011 ) . Therefore , functional and anatomical evidence confirms that the MeA , as targeted in these experiments , receives input predominantly from the AOB ( Pro-sistiaga et al . , 2007 ) . Modulatory influences are important for a variety of sensory pathways ( King and Palmer , 1985; Sherman and Guillery , 1998; Bergan and Knudsen , 2009 ) and any main olfactory inputs to the MeA cells recorded here are likely subtler than our experiments were designed to detect . MeA stimuli-evoked responses typically followed electrical stimulation of the sympathetic nerve trunk , which induces uptake of stimuli into the VNO lumen . Some evoked responses appeared after stimulus application but before stimulation of the sympathetic nerve trunk suggesting that stimuli can gain access to the VNO lumen in our anesthetized preparation , likely due to occasional spontaneous VNO pump activation . In these cases , aside from the onset time of response , response properties of individual units were indistinguishable from responses following nerve trunk stimulation . Based on these observations , our analysis includes the full epoch in which sensory stimuli activate the VNO . Vomeronasal sensory neurons detect diverse non-volatile chemical cues found in the urine , tears , saliva , and sweat of other animals ( Lin da et al . , 2006; He et al . , 2008; Haga et al . , 2010 ) . While all of these stimulus sources are likely to convey important sensory information about the sex , age , and behavioral state of other animals , we initially focused on urine stimuli from predators , female mice , and male mice as these readily available stimuli elicited vigorous responses from single MeA units . Unless otherwise noted , predator stimuli were a mixture of bobcat , fox , and rat urine diluted 1:100 in Ringer's solution; female and male stimuli were a mixture of urine from BALB/c , C57 , and CBA strains diluted 1:100 in Ringer's solution . Sensory responses of MeA units typically exhibited high specificity for one stimulus or a subset of the tested stimuli ( Figure 2 ) , with a smaller number of MeA units responding broadly to multiple stimuli from both conspecific and non-conspecific sources ( Figure 2 , bottom four rows ) . 10 . 7554/eLife . 02743 . 007Figure 2 . MeA sensory responses to VNO stimuli . ( Left ) Each row shows the responses elicited in a single MeA unit by four different VNO stimuli , with each stimulus presented 5 times . The order of stimulus presentation was randomized during the experiment , but is shown grouped by stimulus for clarity . ( Right ) Histograms showing the mean response and standard error ( shaded region ) for each unit . Responses were aligned to the onset of stimulus presentation . All significant responses ( p<0 . 01; Nonparametric ANOVA ) are indicated by an asterisk in the top right corner of the average histogram plots . Response magnitudes for each unit were normalized to the maximum response for all stimuli . Colored bars ( top and bottom ) indicate the 40 s epoch following stimulus presentation that was considered for all analyses . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 00710 . 7554/eLife . 02743 . 008Figure 2—figure supplement 1 . MeA units responsive to different stimulus categories are spatially segregated . ( A ) A single electrophysiological probe was positioned in the MeA of an adult male mouse ( left ) and 13 MeA units were simultaneously recorded on 8 evenly spaced recording sites ( right: rows ) . Dashed lines indicate the recording sites/depth . The average responses to four stimuli ( columns ) are shown for each unit ( asterisks indicate significant responses; bold asterisks indicate unit classification as ‘predator’ or ‘conspecific’ ) . ( B ) The dorsoventral distance between pairs of neurons with different stimulus classifications ( either ‘conspecific’ or ‘predator’ ) is shown for all simultaneously recorded pairs . Positive values indicate the conspecific unit was dorsal to the predator unit . Red , blue , and shaded purple curves represent data obtained from female , male , and all animals respectively with each curve skewed towards conspecific units being located dorsally ( p<0 . 00001 for all distributions: signrank test; horizontal dashed line indicates the mean for all animals: 160 , 95% CI = 123 to 189 µm ) . ( C ) The dorsoventral distance is plotted against the difference in stimulus specificity for all unit pairs . Regression analysis indicates a weak relationship between the dorsoventral distance and the difference in stimulus specificity for unit pairs ( R2 = 0 . 033 , F = 8 . 6 , p=0 . 003; black line; light gray points omitted due to high leverage ) . Therefore , the dorsoventral topography is largely a result of quantitative biases in the numbers of ‘conspecific’ and ‘predator’ neurons located dorsal vs ventral , while only 3% of the variance is explained by a smooth transition from ‘predator’ selective neurons ( ventral ) to ‘conspecific’ selective neurons ( dorsal ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 00810 . 7554/eLife . 02743 . 009Figure 2—figure supplement 2 . AOB sensory responses to VNO stimuli . Each row ( A-D ) shows data obtained from a single well-isolated AOB unit , and each column indicates the stimulus presented ( female , male , predator , Ringer's ) . For each panel , raster plots ( top , shaded region ) show the timing of spikes that occurred during each of 5–7 trials . Histograms of the mean spiking response are shown below each raster plot with standard error ( shaded color region ) . ( E ) Sagittal section through the olfactory bulb . DAPI staining is shown in blue and the electrodes were coated with DiI ( red ) prior to insertion . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 009 One of the best-known functional features of the MeA , demonstrated via immediate early gene studies , is the segregation into dorsal and ventral processing subdivisions . The posterior dorsal MeA , or MeApd , is essential for reproductive and social behaviors while the ventral MeA , or MeApv , is essential for defensive behaviors ( Fernandez-Fewell and Meredith , 1994; Choi et al . , 2005; Wu et al . , 2009 ) . We stereotaxically targeted the posterior MeA such that the array of recording sites would span both the MeApv and MeApd . Consistent with previous findings , we found that a clear topographical segregation of sensory responses is apparent in the real time response properties of MeA units . Neurons that responded most strongly to 'conspecific' stimuli were typically located dorsal to neurons that responded most strongly to 'predator' stimuli , and this was true for data recorded from male or female animals ( Figure 2—figure supplement 1 ) . However , we also observed MeA neurons that responded significantly to multiple stimuli with varying response strengths ( Figure 2—figure supplement 1A ) , indicating that individual neurons distributed throughout the MeA have access to information about both conspecific and defensive stimuli . Units recorded in the AOB , the sole target of VNO output and the primary input to the MeA , display a wide range of response selectivity with some units responding similarly to seemingly contradictory sensory signals ( Hendrickson et al . , 2008; Ben-Shaul et al . , 2010 ) . This finding prompted us to compare the level of specificity of sensory responses in the MeA and AOB . The AOB external cellular layer was targeted by making a small craniotomy immediately rostral to the rhinal sinus and advancing electrodes into the AOB at a 30° angle . Consistent with all previous electrophysiology in the MOB ( Bhalla and Bower , 1997; Kay and Laurent , 1999; Rinberg et al . , 2006; Doucette et al . , 2011 ) and AOB ( Ben-Shaul et al . , 2010 ) , showing that granule cells are virtually undetectable through extracellular recordings , units responsive to VNO stimuli were only detected from electrodes located in the mitral cell layer . Similarly , systematic post-mortem histology from experiments yielding isolated units invariably confirmed accurate targeting to the AOB mitral cell layer ( Figure 2—figure supplement 2 ) . Thus , we can assume with a high level of confidence that we chiefly recorded from AOB projection neurons and not interneurons . Male mouse , female mouse , and predator stimuli each elicited robust responses in the MeA and AOB . Using a threshold for statistical significance of p<0 . 01 , the fractions of units showing sensory responses are 6 . 3 , 5 . 1 , and 6 . 9% for female , male , and predator stimuli respectively for MeA units , as compared to 16 . 6 , 11 . 4 , and 21 . 3% for AOB units ( p<0 . 01 , non-parametric ANOVA ) . Thus , the percentage of units responding to each stimulus is dramatically lower in the MeA than it is in the AOB ( Figure 3A ) . Given the extremely low activity of MeA units in the absence of sensory stimuli , even small responses can reach high statistical significance ( see Figure 2: units 8 and 15 ) . To further compare the stimulus selectivity of individual neurons , we calculated a selectivity index that ranges linearly from 0 ( all stimuli elicit equal responses: center of triangles in panels C , D ) to 1 ( only one stimulus elicits a response: vertices of triangles in panels C , D ) for units recorded from the AOB ( mean 0 . 38; 95% CI: 0 . 36 to 0 . 41; see Supplementary Information ) and from the MeA ( mean 0 . 54; 95% CI: 0 . 50 to 0 . 58 ) . Comparison of the response distributions between the AOB and the MeA reveals a clear and significant shift to values with higher selectivity in the MeA ( Figure 3B; p<0 . 0001; permutation test ) . 10 . 7554/eLife . 02743 . 010Figure 3 . Decreased frequency and increased specificity of sensory responses in the MeA compared to the AOB . ( A ) The percentage of single AOB units ( dashed curves ) and MeA units ( solid curves ) exhibiting statistically significant responses to male ( blue ) , female ( red ) , and predator ( green ) vomeronasal stimuli as the threshold for inclusion was varied from p<0 to p<0 . 2 ( abscissa ) . The diagonal gray line indicates the predicted false positive rate . ( B ) Distribution of response selectivity ( ‘Materials and methods’ ) showing a shift towards higher specificity in MeA ( solid line ) as compared to the AOB ( dashed line ) . ( C ) Selectivity of sensory responses for units recorded in the adult AOB ( 197 units from male and female animals ) . ( D ) Selectivity of sensory responses for units recorded in the adult MeA ( 274 units from male and female animals ) . Each point represents the response profile of an individual unit , with at least one significant response , to male , predator , and/or female stimuli . Points located near a vertex ( more frequent in the MeA ) represent units that respond most strongly for the stimulus indicated at that vertex whereas points at the center ( more frequent in the AOB ) represent units that respond similarly to all stimuli . Insets ( C and D ) show correlation between responses for each pair of stimuli . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 01010 . 7554/eLife . 02743 . 011Figure 3—figure supplement 1 . Leverage analysis for stimulus response correlations . ( Top row ) Pseudocolor plots of correlation values with all data points included . ( Middle row ) Scatter plot of stimulus 1 ( abscissa ) against stimulus 2 ( ordinate ) for all comparisons . High leverage points ( lev >2*Rn/N; where Rn = # regression parameters and N = the size of the data set ) are shown in red . ( Bottom row ) Pseudocolor plots of correlation values with high leverage data points omitted . Each column represents a different data set ( adult AOB , adult MeA , juvenile MeA ) . The similarity between correlations with or without high leverage points included ( A to G; B to H; C to I ) indicates that the correlation is not driven by only a few neurons , but rather is a property of the population . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 01110 . 7554/eLife . 02743 . 012Figure 3—figure supplement 2 . Principal components analysis of MeA categorization data . ( A top ) Principal component scores for principal components 1–5 . ( A bottom ) For each matrix: rows indicate the normalized response of a single MeA unit to ten different stimuli drawn from three behaviorally relevant categories male ( balbC M , C57 M , CBA M ) , female ( balbC F , C57 F , CBA F ) , predator ( fox , rat , bobcat ) , and a Ringer's control . Columns indicate the stimulus presented , and vertical white lines indicate category boundaries . Each matrix shows the same data set sorted by the five largest principle components . Together , these components account for over 90% of the variance in sensory responses observed in our recordings . ( B ) Pairwise correlation of MeA responses , for all responding units , elicited by VNO stimulation . Positive correlations between stimulus-induced responses are indicated by red and anti-correlations are indicated by blue . Within-category correlations are generally positive and between-category correlations are generally negative . ( C ) Multidimensional scaling analysis shows that stimuli from a single category populate a region of the response space that is non-overlapping with stimuli drawn from a different category . Lines connecting individual data points indicate ethologically defined stimulus categories . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 012 The emergence of selectivity in the MeA is further shown by plotting the responses of all units on a triangular axis for which the position of each point indicates the relative magnitude of the responses elicited by each of three stimuli ( Figure 3C: AOB , Figure 3D: MeA ) . The vertices represent exclusive responses to one of the stimuli . Comparison of the spatial distribution of individual neurons between the AOB and the MeA indicates that , whereas AOB units tend to populate the central region of the triangular plot , MeA units more evenly span the entire area with an increased density at the vertices . Dividing the triangular response space into four equal quadrants ( dashed lines Figure 3C , D ) showed that 44 . 4% of AOB units vs 21 . 6% of MeA units reside in the central quadrant associated with less specific responses . Thus , flow of information from the AOB to the MeA is associated with a sharpening of stimulus-driven responses . The increased selectivity of sensory responses in the MeA is similarly reflected by a marked decorrelation of responses evoked by different sensory stimuli across the MeA population ( AOB: Figure 3C inset; MeA: Figure 3D inset; Figure 3—figure supplement 1 ) . The sharpened response pattern observed in the MeA ( Figure 3 ) suggests that MeA activity could represent higher-level categorical information . We measured responses in male mice to stimuli from three ethologically defined classes: female ( estrus: BALB/c , CBA , and C57 ) , male ( BALB/c , CBA , and C57 ) , and defensive cues ( fox , bobcat , and rat urine ) . A principal component analysis performed on responses evoked by these stimuli revealed that the first PC reflected the distinction between ‘female’ and ‘predator’ categories and accounted for 41% of the total response variance ( Figure 3—figure supplement 2 ) . The second PC distinguished between ‘female’ vs ‘male’ stimuli and accounted for 19% of the total variance . Therefore , most of the variation of sensory responses in the MeA of male mice is devoted to the distinction between ‘female’ from ‘predator’ and 'male' categories respectively . However , individual sensory responses in the MeA rarely form sharply defined stimulus categories , but rather , maintain enough information to distinguish between individual elements of each category ( Figure 3—figure supplement 2 ) . Sexually dimorphic behaviors of male and female animals to conspecific stimuli represent a dramatic and reproducible example of individual variability: male stimuli elicit territorial responses from another male mouse , but reproductive displays from a receptive female mouse . Sex-specific behaviors imply the existence of sexually dimorphic sensory processing; however , little is known about in vivo functional differences between the sexes at the single neuron level . To assess the role of the MeA for sex-specific computation within the vomeronasal pathway , we compared the activity of MeA units recorded from male vs female animals in response to urinary stimuli from male and female conspecifics ( Figure 4 ) . This analysis revealed a striking sexual dimorphism in the stimulus selectivity of MeA units ( Figure 4B ) , such that MeA responses are dramatically more frequent to opposite-sex stimuli . We quantified the sex preference of individual neurons using an index that ranges from −1 to 1 , reflecting responses exclusive to male or female stimuli , respectively . Distributions of responses from both male and female MeA neurons appear strikingly skewed away from zero , toward stimuli representing the opposite sex ( Figure 4E; p<0 . 0001 permutation test ) . Indeed , 82 . 7% of male MeA units respond more strongly to female stimuli and 83 . 3% of female MeA units respond stronger to male stimuli . Thus , for both sexes , there is an unmistakable overrepresentation of responses to stimuli from the opposite sex . In addition , the ratio of predator to conspecific responsive units was significantly higher in the MeA of female vs male mice ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 02743 . 013Figure 4 . Sexual dimorphism of adult MeA responses . ( A ) Responses of AOB neurons to vomeronasal stimuli in adult male ( 210 units ) and female ( 64 units ) mice . ( B ) Responses of MeA neurons to vomeronasal stimuli in adult male ( 106 units ) and female ( 91 units ) mice . ( C ) Responses of MeA neurons to vomeronasal stimuli in juvenile male ( 37 units ) and female ( 50 units ) mice . Units shown in panels A–C are the same data shown in Figure 3C , D but classified according to the sex of the animal recorded . Blue circles indicate units recorded from male mice and red squares indicate data recorded from female mice . ( D–F ) Sex-specificity ( ‘Materials and methods’ ) histograms are shown for all units recorded from male ( blue ) and female ( red ) animals in the adult AOB ( D ) , adult MeA ( E ) and juvenile MeA ( F ) . Horizontal lines ( above ) indicate the mean and 95% confidence interval ( bootstrap CI ) of the mean for each distribution . Data collected from males vs females were only different in the adult MeA ( AOB: p=0 . 26 adult MeA: p<0 . 00001; juvenile MeA: p=0 . 18; permutation tests ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 01310 . 7554/eLife . 02743 . 014Figure 4—figure supplement 1 . Sexual dimorphism in the dominance of predator versus conspecific responses . A predator/conspecific index was calculated as: ( Rpred-Rconspecific ) / ( Rpred + Rconspecific ) , where RP is the response to predator stimuli and RC is the stronger of the responses to male or female stimuli . Histograms show predator/conspecific index for units recorded from male ( blue ) and female ( red ) animals in the adult AOB ( A ) , adult MeA ( B ) , and juvenile MeA ( C ) . Horizontal lines above histograms indicate the bootstrapped 95% confidence interval for the mean . Only data from the MeA of adult males and females exhibited a clear difference ( AOB: p=0 . 35; adult MeA: p<0 . 001; juvenile MeA: p=0 . 42; permutation tests ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 014 The sexually dimorphic responses observed in the MeA could be generated by specific neuronal processing within the MeA or inherited from sexually dimorphic response patterns at the AOB . We compared the responses of AOB units recorded in both male and female mice , and , in contrast to the MeA sensory representation , found little evidence for sexual dimorphism ( Figure 4A , D ) . Rather , the responses of AOB units recorded from both male and female animals share a moderate bias in responses for female stimuli ( Figure 4D ) . Thus , the profound sexual dimorphism of sensory representations described here is an emergent property of the MeA . Circuits underlying sexually dimorphic behaviors are organized perinatally by the early action of steroid hormones , and activated when an animal reaches reproductive maturity . This developmental schema , known as the organizational/activational hypothesis of sexual dimorphism in the brain ( Phoenix et al . , 1959; McCarthy , 2008 ) , proposes the establishment of sex-specific circuit features during a perinatal ‘organizational’ stage that are subsequently made functional during a peripubertal ‘activational’ stage . To understand when the MeA circuitry acquires its sex-specific functional features , we recorded responses to sensory stimuli in the MeA of juvenile male and female mice at 18–21 days , which is roughly 1 week prior to puberty onset ( 28 days; Nathan et al . , 2006 ) . We found that , while single units recorded from the MeA of juvenile mice responded to VNO stimuli , they were far less selective for male , female , and predator cues ( Figure 4C , F ) than those in the adult MeA ( Figure 4B , E ) , and instead showed closer similarity with the response selectivity identified in the adult AOB ( Figure 4A ) . Thus , 34 . 7% of juvenile units were in the central quadrant associated with less specific responses , as opposed to 21 . 6% in the adult MeA . The selectivity of MeA responses to opposite sex cues is absent in juvenile animals , and was not significantly different between males and females ( Figure 4F; p=0 . 06 , permutation test ) , a feature also shared with the adult AOB and strikingly different from the adult MeA . Thus , MeA responses in juvenile mice are not sexually dimorphic , even though the perinatal organizational phase of MeA sexual development , during which steroid-dependent cell death generates sexually dimorphic MeA cell numbers , is essentially complete ( Wu et al . , 2009 ) , and appear instead highly reminiscent to the responses observed in the AOB . Much of testosterone's influence on sexually dimorphic patterning of the brain is achieved after its conversion to estrogen by aromatase and activation of estrogen receptors . Therefore , aromatase expression in the brain is a key factor in the development of sexually dimorphic behaviors and brain circuits . Although sparse in the rodent brain , the MeApd is a primary site of aromatase expression , and sexual dimorphisms in MeA regional volume , cell number , and neural connectivity are mediated by aromatase-dependent sex-steroid signaling ( see Cooke et al . , 1999; McCarthy , 2008 ) . The dorsal bias in aromatase expression parallels a dorsal bias of sexually dimorphic sensory responses we observed . We therefore reasoned that the sexually dimorphic functional differences observed here may be impaired in mice lacking aromatase function , and investigated sensory responses in the MeA of adult aromatase knockout mice ( ArKO−/−; Fisher et al . , 1998 ) . As predicted , MeA units from ArKO−/− males frequently responded to multiple stimuli; whereas , MeA units from age matched heterozygous littermates ( ArKO+/− ) were similar to those observed in wild-type males ( Figure 5A ) . Indeed , the sex specificity of neurons recorded in adult ArKO−/− males is largely symmetric around 0 , and presents a pattern of responses intermediate to those observed in intact juvenile and adult male mice ( Figure 5B ) . In contrast , adult heterozygous littermates ( ArKO+/− ) displayed a bias towards responses to female stimuli similar to that observed in wild-type males ( Figure 5A , B ) . Like juvenile animals , most ( 42 . 1% ) of MeA units recorded from adult ArKO−/− males reside in the central quadrant associated with less specific responses , compared to 19 . 5% of units recorded from ArKO+/− . Similarly , response patterns to different stimuli are much more correlated in the MeA of ArKO−/− males ( Figure 5C , bottom ) than in ArKO+/− littermates ( Figure 5C , top ) or wild-type males ( Figure 3D ) , reflecting a loss of discriminability at the individual unit level . These results suggest that estrogen signaling ( via conversion of testosterone by aromatase ) is essential for generating the pattern of sensory responses observed in the adult male MeA . The fact that adult ArKO−/− males are also different from juvenile males indicates that other processes , such as direct testosterone signaling , may play an important role in developing the adult MeA sensory representation . 10 . 7554/eLife . 02743 . 015Figure 5 . Importance of aromatase signaling for the development of sexually dimorphic MeA responses . ( A ) Population summary of MeA responses to male , female , and predator stimuli recorded from adult male ArKO mice ( light blue circles ) or heterozygous male littermates ( dark blue diamonds ) . All plotted units responded significantly to at least one stimulus . ( B ) Sex-specificity histograms for units recorded from ArKO males ( blue fill ) , heterozygous male littermates ( dark blue; no fill ) , and wild-type juvenile males ( light blue; no fill ) . Horizontal lines indicate the mean and 95% confidence interval of the mean of each distribution . ( C ) Matrices of correlation for the population responses between pairs of sensory stimuli for heterozygous males ( top ) and ArKO males ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 015 To determine if perinatal hormone exposure is sufficient to establish a masculine pattern of MeA neural function , we injected newborn female pups with estradiol–benzoate at three postnatal days ( p0 , p7 , and p14 ) , which has been shown to induce a male-like aromatase expression pattern and a subset of male behaviors in adult female mice ( MacLusky and Naftolin , 1981; Wu et al . , 2009 ) . We then recorded sensory evoked MeA responses in these females when they reached adulthood ( 10–14 weeks ) . Neurons responding most strongly to same-sex stimuli are nearly absent from the MeA of untreated adult females . However , in estrogen treated females , 51 . 2% of MeA neurons respond more strongly to female than to male stimuli , with 26 . 7% responding to female stimuli at least 3 times stronger than to male stimuli ( Figure 6A , B ) . This novel subset of neurons ( female responsive in female animals ) is located more dorsally relative to predator-responsive neurons , as is typical of female responsive neurons in males ( Figure 6C ) . In addition , estrogen treatment resulted in a clear reduction of the number of neurons selective for male stimuli ( Figure 6B ) . 10 . 7554/eLife . 02743 . 016Figure 6 . Estrogen influences the development of sexually dimorphic MeA responses . ( A ) Population summary of MeA responses to male , female , and predator stimuli recorded from adult untreated female mice ( red squares ) or estrogen treated adult females ( dark red triangles ) . ( B ) Sex-specificity histograms for units recorded from estrogen-treated adult females ( dark red fill ) , adult untreated females ( red; no fill ) , and untreated adult males ( blue; no fill ) for comparison . Estrogen treatment reduced responses in adult female mice to male stimuli ( leftward ) and increased the frequency of responses to same-sex stimuli ( rightward ) . Horizontal lines indicate the bootstrapped 95% confidence interval for the mean of each distribution . ( C ) Comparison of the dorsal-ventral locations of male vs predator responsive units ( blue ) and female vs predator responsive units ( red; see Figure 2—figure supplement 1 ) in the MeA of estrogen treated females . Both male responsive and female responsive units are dorsal to predator responsive units . DOI: http://dx . doi . org/10 . 7554/eLife . 02743 . 016 To fully understand how patterned activation of vomeronasal receptors triggers specific behavioral and endocrine responses , one must determine how representations of sensory stimuli are transformed through successive stages of vomeronasal processing . Sensory cues are initially detected by VNO sensory neurons , each expressing a single VR ( Leinders-Zufall et al . , 2000; Haga et al . , 2010; Isogai et al . , 2011 ) , which then project to the AOB . While some AOB units display clear selectivity for distinct VNO stimuli , AOB units typically respond to multiple sensory stimuli , and many AOB units respond to multiple stimuli associated with conflicting behavioral outcomes such as female conspecifics and predators ( Ben-Shaul et al . , 2010 ) . Nearly all AOB output neurons project to the MeA , however , the convergence of input in the MeA lead to a different pattern of sensory responses , such that the fraction of MeA neurons responding to any given stimulus is smaller , and the specificity of MeA responses elicited by sensory stimuli is greater than observed in the AOB . One interpretation of our data is that the low stimulus specificity observed in AOB responses represents a recording bias towards local inhibitory interneurons . This , however , seems extremely unlikely because our recordings targeted the mitral cell layer , and because ample evidence from olfactory bulb electrophysiology indicates that granule cells are virtually undetectable through extracellular recordings ( Bhalla and Bower , 1997; Kay and Laurent , 1999; Rinberg et al . , 2006; Ben-Shaul et al . , 2010; Doucette et al . , 2011 ) . Instead , the complex AOB representation we observe is likely generated by the integration of VR inputs by individual AOB mitral cells ( Wagner et al . , 2006 ) . The nature of the units recorded from in the MeA is presently unknown , and further identification by genetic methods is an important goal for future studies . A general feature of sensory systems is that neurons located at later stages of sensory processing are less active than those at earlier stages . Sparse coding at higher processing areas is well described in vision and olfaction ( Olshausen and Field , 1996; Quiroga et al . , 2005; Poo and Isaacson , 2009 ) , and is thought to reflect the construction of increasingly complex sensory representations . Similarly , we find that the fraction of neurons activated by any given stimulus is approximately threefold greater in the AOB compared to the MeA . While the function of unresponsive units remains unclear , a reasonable suggestion is that these units respond to stimuli that were not presented in these experiments . Alternatively , vomeronasal-induced behaviors are strongly influenced by behavioral state ( e . g . : dominant vs subordinate , estrus vs non-estrus , or adult vs juvenile ) , and it is possible that only a subset of MeA neurons are active in any given behavioral or physiological state . We also found that MeA neurons of adult mice respond to vomeronasal stimuli with significantly more specificity for behaviorally distinct classes of stimuli than AOB neurons , and the number of MeA neurons responding to behaviorally conflicting stimuli , although not absent , is lower than observed in the AOB . Thus , the complex responses typical of AOB neurons are interpreted by the MeA in a manner that promotes the emergence of sensory responses more closely associated with a specific behavioral output , and may help generate unambiguous and mutually exclusive behaviors ( e . g . : aggression , courtship , parenting , or predator defense ) . The greater MeA selectivity may result from the preferential contribution of the most selective AOB units to MeA responses , or from the transformation of AOB inputs by sensory processing within the MeA . The apparent functional disambiguation of AOB inputs by MeA units suggests a logic and specificity in AOB to MeA connectivity that strongly contrasts with the distributive pattern of projection reported from the MOB to the piriform cortex ( Sosulski et al . , 2011 , Choi et al . , 2005 ) . As observed in the VNO and AOB , MeA responses display a relatively long latency ( several seconds ) and slow time-course ( tens of seconds ) that is likely a direct result of the slow dynamics of large non-volatile sensory stimuli that activate VRs ( Holy et al . , 2000; Hendrickson et al . , 2008; Ben-Shaul et al . , 2010; Shpak et al . , 2012 ) . Nearly half of the neurons in this study were completely silent until a preferred sensory stimulus was presented , and nearly all single-unit responses identified in this study consisted of stimulus-evoked increases in neural activity . This is consistent with c-Fos induction observed in large subsets of MeA neurons following exposure of animals to reproductive and defensive stimuli ( Choi et al . , 2005 ) and the strong monosynaptic excitatory input from AOB fibers on MeA neurons ( Bian et al . , 2008; Niimi et al . , 2012 ) . Since the MeA receives a broad range of modulatory inputs from other brain regions and is likely to be affected by the animal's behavioral state , it is possible that the MeA functions differently in awake vs anesthetized animals . Sensory processing in the main olfactory system is clearly altered by anesthesia level and behavioral state ( Rinberg et al . , 2006; Cazakoff et al . , 2014 ) , and relating the findings from the current study to data from behaving animals will be an important step for future studies . The most extreme behavioral differences within a species are found in the way male vs female animals respond to the same sensory stimuli . In some species , sexually dimorphic processing of sensory cues is already prominent in the sensory epithelium . For example , male and female moths ( Manduca sexta ) and silkworms ( Bombyx mori ) express an array of olfactory receptors that are not expressed by the opposite sex ( Hansson et al . , 1989; Nakagawa et al . , 2005; Große-Wilde et al . , 2010 ) . Thus , moths of either sex have access to channels of sensory information associated with sex-specific behaviors that are completely absent in the opposite sex ( Schneiderman et al . , 1986 ) . The roundworm C . elegans displays a variation on this theme such that sexually dimorphic activity of primary sensory neurons ( ASI ) during development shapes a sexually dimorphic downstream circuit that then contributes to guiding sexually dimorphic behaviors in the adult ( White and Jorgensen , 2012 ) . In other species , stimuli are similarly detected by sensory epithelia of both sexes , but are differentially interpreted by distinct circuits in the male vs female brain . For example , the Drosophila pheromone 11-cis Vaccenyl acetate ( cVA ) elicits sexually dimorphic behaviors , but the sensory neurons that detect cVA are anatomically and functionally indistinguishable in males and females ( Kurtovic et al . , 2007; Datta et al . , 2008 ) . Instead , the influence of cVA on sexually dimorphic behaviors is achieved through differential projections of cVA input to the sexually dimorphic network of fru expressing neurons in the central nervous system ( Datta et al . , 2008; Ruta et al . , 2010 ) . Even more centrally based , the capacity for male , but not female , finches to sing is attributed to circuit differences in premotor nuclei ( Nottebohm and Arnold , 1976; Konishi and Akutagawa , 1985; Kirn , 2010 ) . Thus , the pervasive existence of sexually dimorphic behaviors underscores their evolutionary value , but the strategies employed to achieve sexual dimorphism differ widely between species . One of the most striking features of VNO-mediated behaviors is their sexually dimorphic expression: a male mouse elicits territorial aggression in another male mouse , but courtship in a receptive female mouse . Similarly , pups trigger infanticide in virgin males but parental behavior in virgin females . One therefore expects that sex-specific differences in the behavioral significance of certain cues be reflected in the male- and female-specific patterns of neuronal activity within the vomeronasal pathway . However , although sexually dimorphic anatomical features have indeed long been identified at every stage of the vomeronasal pathway , including in the VNO , AOB , amygdala , and hypothalamic areas ( Guillamón and Segovia , 1997 ) , there is little evidence for substantial differences in sensory responses of the male and female VNO or AOB ( Nodari et al . , 2008; Herrada and Dulac , 1997; Kang et al . , 2009; Ben-Shaul et al . , 2010; Haga et al . , 2010 ) . Thus , downstream circuits are likely to contribute significantly to generate , or amplify , sexually dimorphic sensory processing . Previous studies show that silencing the VNO , genetically or surgically , brings about a set of behaviors that are typically seen in the opposite sex ( Stowers et al . , 2002; Kimchi et al . , 2007 ) . These findings demonstrate the essential contribution of the vomeronasal system to the control of social behaviors , and show that effector circuits for behaviors of both sexes exist in the brain , and are activated or repressed by vomeronasal activity in a sex-specific manner ( Kimchi et al . , 2007 ) . The transformation of the vomeronasal sensory representation in the MeA precisely fulfills the characteristics required of such a behavioral switch . Unlike the AOB , in which sensory responses are similar in males and females , over 80% of the recorded MeA neurons respond preferentially to stimuli from the opposite sex . Accordingly , an entire category of ‘same-sex’ responses is decimated in the MeA representation . Same-sex information is not entirely lost: indeed , the lack of male to male aggression in TRPC2−/− mice demonstrates that the representation of male stimuli , in male mice , is present in wild-type animals , depends on vomeronasal signaling , and is capable of driving robust behaviors ( Stowers et al . , 2002 ) . Same sex-information may therefore either rely on the minimal representation found in the MeA , or be processed preferentially by other secondary vomeronasal relays . Puberty marks the transitional period during which animals develop the secondary sex traits characteristic of adult animals , and begin to display adult-typical social behaviors . It is generally believed that testosterone accounts for most , if not all , of development of the known sexually dimorphic structures in the mammalian brain ( Morris et al . , 2004 ) . The manner by which testosterone shapes the nervous system , however , depends on the specific neural circuit , such that cell death , synapse formation , synapse elimination , neurogenesis , and changes to neuron morphology all play important roles ( Cooke et al . , 1999; Ibanez et al . , 2001; Morris et al . , 2004; Zhang et al . , 2008; Forger & de Vries , 2010 ) . The influence of sex-steroids on the brain is strictly controlled according to the developmental stage , with two phases described as critical . During an initial phase , which occurs near birth , neural structures are thought to be differentially ‘organized’ by exposure to testosterone in males , but not in females ( Phoenix et al . , 1959; McCarthy , 2008 ) . This initial phase is thought to create clear structural and anatomical differences within sexually dimorphic regions . During a second phase , which occurs during puberty , sexually dimorphic circuits are ‘activated’ such that they begin functioning as mature circuits capable of effecting sexually dimorphic behaviors ( Phoenix et al . , 1959 ) . The MeA of mice is morphologically sexually dimorphic such that males have a larger posterodorsal MeA ( MeApd ) than females ( Cooke et al . , 1999; Morris et al . , 2004 ) , a difference primarily attributed to differential cell death at birth ( Wu et al . , 2009 ) . Our recordings from the MeA of mice in the week prior to puberty onset demonstrate that MeA units had not developed the sexually dimorphic response patterns characteristic of adult animals , although these recordings were performed after sex-specific cell death is largely complete ( Wu et al . , 2009; see ; Forger and de Vries , 2010 ) . Therefore , the organizational phase of steroid-dependent MeA cell death likely provides a template for future sexually dimorphic sensory responses , but is not in itself sufficient to endow the MeA with adult function that might , for example , require further elimination of neurons with the wrong sex cue specificity . Importantly , juvenile MeA neurons are not silent but rather respond vigorously to vomeronasal stimuli in a manner very similar to AOB neurons in adult animals , demonstrating that the MeA is already an actively functioning circuit . Taken together , our results indicate that a second phase of circuit organization is required to shape sex-specific responses in adults . Our recordings from animals with manipulated hormone levels and hormone signaling show that sexually dimorphic patterns of MeA function in adult mice are estrogen dependent . Previous experiments studying an independently made ArKO mouse line reported significant impairment in social recognition ( Bakker et al . , 2002 ) , which may result from the failure of MeA neurons in ArKO animals to maturate adult patterns of MeA responses . In some species of mammals , the influence of sex-steroids on MeA anatomy persists well into adulthood ( Cooke et al . , 1999 ) supporting our conclusion that additional hormone-dependent circuit changes occur after the perinatal organizational phase . Moreover , a multi-stage organizational model of sex-specific behavioral circuits is supported by elegant experiments in hamsters demonstrating profound rearrangements of circuits underlying social behaviors during puberty ( Zehr et al . , 2006 ) . Several models could account for the late development of sexually dimorphic responses observed in the MeA . Death of MeA neurons during development is important for sculpting a sexually dimorphic circuit . However , we now know that the functional properties of MeA units can be similar even though the number of cells in the male vs female MeA is different . Therefore , other processes that build on these early differences must exist , and different mechanisms for achieving proper functional sexual dimorphism in the MeA can be proposed , each providing unique and testable predictions on how neural circuits in the adult MeA develop . Synaptic input from the AOB to the MeA may be sculpted in adult mice to disproportionately favor AOB input carrying opposite sex information to the MeA . Alternatively , or in addition to , maturation of inhibition within the MeA at puberty may act to silence , or specifically shunt less robust MeA responses . This latter mechanism could act in a manner similar to the control of critical period plasticity within the cortex through a time- and steroid-dependent maturation of inhibitory connectivity ( Hensch , 2005 ) . The emergence of specific topographically segregated , and sexually dimorphic MeA responses indicate that the anatomical and functional processes by which response selectivity is achieved in the MeA do not act randomly . Rather , specific circuit mechanisms yet to be uncovered must act together to allow features of vomeronasal information to be extracted from the overall AOB representation based on intrinsic MeA properties , as well as , the sex , age , and physiological status of the individual animal . Remarkably , the differences in MeA activity observed in males vs females , juveniles vs adults , or mutants vs wild-type animals correlate tightly with distinct patterns of social behaviors from each of these groups . This suggests that the age- and sex-specific transformations in MeA sensory processing uncovered by our study are likely to underlie fundamental changes in social behavior throughout development . Mice ( adult , litters of juvenile mice , and pregnant females for estrogen treatments of newborn pups ) were purchased from Charles River Laboratories ( Wilmington , MA ) . ArKO male mice were kindly provided by Dr Evan Simpson ( Fisher et al . , 1998 ) and bred in house . All experiments were performed in strict compliance with the National Institute of Health and Harvard University . Mice were anesthetized with 100 mg/kg ketamine and 10 mg/kg xylazine and the skin overlying the throat was cut with dissecting scissors . The trachea was exposed by moving overlying gland tissue and separating the right and left sternohyoid muscles . A tracheal incision was made using fine scissors and a 15-mm length polyethylene tube ( I . D . 0 . 76 mm , O . D . 1 . 22 mm , Franklin Lakes , NJ ) was inserted in the caudal end of the cut trachea and held in place with Vetbond glue ( 3 M , St . Paul , MN ) . This tracheotomy was subsequently used to maintain anesthesia ( 0 . 5–2% isoflurane in pure oxygen ) for the duration of the electrophysiological experiments . The rostral end of the trachea was sealed closed to prevent efflux of fluid from the cervical cavity to the nasal cavity and the VNO , or incubated and used to control airflow across the MOE . The sympathetic nerve trunk was gently isolated from connective tissue and enclosed with a stimulating cuff electrode using the carotid artery for structural support . All incisions were then closed with Vetbond , and the mouse was placed in a custom built stereotaxic apparatus . To allow for cleaning the nasal cavity and the VNO between different stimulus presentations ( see below ) , a plastic tube was inserted below the mouth to allow suction of fluid through the nasopalatine duct . Fluid flow was regulated with a computer controlled solenoid valve ( Takasago , Japan ) connected to a vacuum line . Surgical silk sutures ( 6/0 , CP Medical , Portland , OR ) were inserted into the cheek skin to gently pull them laterally to prevent occlusion of the naso-palatine duct during suction . In experiments requiring stimulation of the MOE , the mouse was tracheotomized and delivery of volatile stimuli was achieved through computer controlled inhalation . The rostral part of the tracheotomy incision was connected to a vacuum line through a polyethylene tube ( I . D . 0 . 76 mm , O . D . 1 . 22 mm ) . Airflow through the nasal cavity via the tube was gated by a computer controlled solenoid valve ( Takasago , Japan ) and regulated by a flow controller ( Omega , Stamford , CT ) . Urine was collected from adult female and male mice of the BALB/c , CBA ( Charles River Laboratories; Wilmington , MA ) or C57Bl6 ( Jackson Laboratories , Bar Harbor , Maine ) strains and immediately placed in liquid nitrogen , for subsequent storage in −80°C . Predator urine samples were obtained from PredatorPee ( Lincoln , Maine ) . Male , female or predator urine samples comprised samples pooled from all three strains , or species , except when they were tested individually ( Figure 3—figure supplement 2 ) . All sensory stimuli ( olfactory and vomeronasal ) were presented 5 or more times in a pseudorandomized order during each experiment . VNO Stimuli were applied by placing 1 µl of stimulus ( 1:100 dilution ) directly into the nostril . After a delay of 20 s , a stepped square-wave stimulation train ( duration: 1 . 6s , current: ±100 µA , frequency: 30 Hz ) , was applied through the sympathetic nerve cuff electrode to facilitate VNO pumping and stimulus entry to the VNO lumen . Following a second delay of 40 s , a solenoid controlling suction to the nasopalatine duct was opened and 1–2 ml of Ringer's solution was passed through the nostril and out the nasopalatine duct to cleanse the VNO . This cleansing procedure lasted 30 s and was accompanied by sympathetic trunk stimulation to facilitate stimulus elimination from the VNO lumen . Volatile stimuli were presented using computer-controlled airflow past the main olfactory epithelium and out the rostral tracheotomy . Volatile stimuli were introduced by passing the air stream through a microfiber filter ( Whatman: 6823-1327 ) containing 30 µl of stimuli ( undiluted male , female , and predator urine; 1/10 and 1/100 dilutions in mineral oil of 2-heptanone , isoamyl acetate , and acetate phenone ) . For AOB recordings , we used a 4-shank configuration with 8 recording channels per shank ( NeuroNexus Technologies: a4x8-5 mm 50-200-413 ) . A craniotomy was opened immediately rostral of the rhinal sinus , the dura was removed around the penetration site , and electrophysiological probes were advanced into the AOB at an angle of ∼30° using a hydraulic micromanipulator ( Siskiyou , Oregon , US ) . Identification of the AOB external cellular layer was performed as described in Ben-Shaul et al . ( 2010 ) . As previously reported for the MOB ( Bhalla and Bower , 1997; Kay and Laurent , 1999; Rinberg et al . , 2006; Doucette et al . , 2011 ) , large well-isolated spikes were found exclusively when the electrode tracks traversed the external cellular layer of the AOB . In contrast , negligible neural activity was observed from electrode tracks located in the granule cell layer . These results suggest that AOB projection neurons ( mitral and tufted cells ) represent the majority of neurons recorded in these experiments . A subset of the AOB data reported in this manuscript were originally reported in Ben-Shaul et al . ( 2010 ) , but have been extensively reanalyzed . MeA recordings were made with a 2-shank configuration with 16 recording channels per shank ( NeuroNexus Technologies: a2x16-10 mm 100-500-413 ) . Briefly , a craniotomy was opened dorsal to the MeA based on stereotaxic coordinates ( 1 . 7–2 . 0 mm lateral from midline and 1 . 7–2 . 0 mm caudal from bregma ) . Electrophysiological probes were positioned using a stereotaxic manipulator ( MP-285; Sutter instruments , Novato , CA ) and advanced 5–6 mm from dorsal surface of the brain until a minute deflection in the electrode shaft indicated that the electrode tip had reached the ventral surface . Accurate targeting was associated with a progression of distinct electrophysiological features as the electrode probe advanced through the brain . For example , large amplitude spikes were abundant as the electrodes penetrated the hippocampus . Ventral to the hippocampus , there was a period of relative quiescence , followed by large oscillatory spikes immediately dorsal to the MeA . The spontaneous neuronal activity in the MeA was typically low and isolated action potentials were distinctly smaller in magnitude than those encountered dorsal to the MeA—we used these distinctive characteristics to improve accurate targeting during the experiment . Indeed , post-mortem histology confirmed that the majority of our recordings traversed the entire dorsoventral extent of the MeA . Electrophysiology probes were coated with one of four fluorescent dyes ( DiI , DiO , DiD , Invitrogen , Carlsbad , CA; FluoroGold , Fluorochrome , LLC ) . At the end of each experiment , the brain was extracted , fixed , and 50-μm coronal sections were made for histological analysis of the electrode tract location to confirm accurate stereotaxic targeting to either the AOB external cellular layer or posterior MeA ( Figure 1B ) . In nearly all cases , electrode tracts were accurately targeted to either the posterior MeA or the mitral cell layer of the AOB . Data was excluded for the few recording sessions with poor targeting . 32 recording channels were band-pass filtered ( 300–5000 Hz ) and continuously sampled at 25 kHz using an RZ2 processor , PZ2 preamplifier , and two RA16CH head-stage amplifiers ( TDT , Alachua , FL ) . Custom MATLAB ( Mathworks , Natick , MA ) programs were used to extract 3 . 5 ms spike waveforms from the continuous data . The number of recording channels on which a given spike was detectable depended on the inter site distance ( more overlap for 50 µm , less overlap for 100 µm ) , and the recording target ( typically more for the AOB that for the MeA ) . Once a spike was detected on any given channel , waveforms were extracted for all electrode channels in close proximity ( a total of 8 channels per group ) . Thus , each single spike was described by its voltage waveforms on 8 contiguous channels and all of these signals were used for spike sorting . Specifically , on each recording session , we calculated the 1st and 2nd principal components ( PCs ) for each channel . Then , each spike was described by its projections ( PC loadings ) on the first two PCs for each of the 8 channels in its group . Thus , the shape of each spike was described by a set of 16 PC loading values . These PC loading values were input to KlustaKwik ( Harris et al . , 2000 ) for automatic classification . Since the algorithm is designed to ‘over cluster’ , clusters were then manually verified and adjusted using Klusters ( Hazan et al . , 2006 ) . Spike clusters were evaluated by consideration of their spike shapes , projections on principal component space ( calculated independently for each recording session ) and autocorrelation functions . Classification of a spike cluster as representing a ‘single unit’ required that the cluster displayed a distinct spike shape and was fully separated from both the origin ( noise ) and other clusters ( multi-unit ) with respect to at least one principal component projection . We also verified that the interspike interval histogram for a given ‘single unit’ cluster demonstrated a clear refractory period ( Figure 1—figure supplement 1 ) . A total of 3543 units were recorded from adult BALB/c mice for all experiments . 1938 units were recorded from the MeA of adult BALB/c mice of which 197 responded to VNO stimulation ( 106 from male animals; 91 from female animals ) . 1605 units were recorded from the AOB of adult BALB/c mice of which 274 responded to VNO stimulation ( 210 from male animals; 64 from female animals ) . 517 units were recorded from the MeA of juvenile BALB/c mice of which 87 responded to VNO stimulation ( 37 from male animals; 50 from female animals ) . 290 units were recorded from the MeA of adult BALB/c female mice perinatally treated with estrogen , with 48 responding to VNO stimulation . 737 units were recorded from the MeA of ArKO−/− and ArKO+/− mice of which 99 responded to VNO stimulation ( 41 from ArKO+/−; 58 from ArKO−/− ) . All statistical analyses were performed with custom Matlab code . Significant responses were identified by comparing each unit's spiking rate during the pre-stimulation ( 20 s prior to stimulation ) and post-stimulation ( 40 s after stimulation ) epochs using a non-parametric ANOVA performed at the significance level of p≤0 . 01 . This post-stimulus window ensured inclusion of responses occurring prior to and after electrical stimulation of the sympathetic nerve . Response magnitude was quantified as the change in average firing rate during the 40 s following stimulus presentation relative to the firing rate during the 20 s prior to stimulus presentation . Spike rates for histograms were binned into 2 s epochs , and then averaged across repeated stimulus presentations . Unless otherwise noted , statistics comparing two populations of units ( e . g . : responses from male vs female animals ) were performed using a nonparametric permutation test ( Efron and Tibshirani , 1993 ) as observed distributions were not typically normally distributed . The precise geometry of the multichannel electrophysiology probes allowed us to determine the relative dorsal–ventral location of MeA neurons with high accuracy . First , single units with statistically significant responses to at least one stimulus were classified in a winner-take-all manner as ‘conspecific responsive’ or ‘predator responsive’ based on the magnitude of stimulus-driven responses . Next , the physical distance between the recording sites for each pair of units ( between the two categories ) was calculated . For Figure 2—figure supplement 1B , positive distances indicate that the conspecific unit was dorsal to the predator unit . Then , we calculated a ‘predator/conspecific contrast ratio’ , ( respcon – resppred ) / ( respcon + resppred ) , for each unit that ranges from −1 ( entirely predator responsive ) to 1 ( entirely conspecific responsive ) . The predator/conspecific contrast ratio values were then subtracted for each predator vs conspecific unit comparison , in a manner identical to the distance analysis described above , to determine if there was a correlation between the location of a unit and the specificity of response tuning ( Figure 2—figure supplement 1C ) . Pearson's correlation values ( insets: Figure 3C , D; Figure 6C ) were calculated , for all responsive MeA units , based on the array of responses to stimulus ‘A’ vs stimulus ‘B’ . To ensure that the observed correlations/regressions were not spurious results generated by a few outliers , we performed a partial leverage analysis . Analyses were then recalculated with high leverage points removed ( Figure 2—figure supplement 2 ) , and in no cases did the exclusion of high leverage points significantly alter the final result . A non-metric multidimensional scaling was used to reduce the dimensionality of the stimulus category distance matrix , so that the distance between each stimulus could be approximated in two dimensions based on Kruskal's normalized stress criterion ( Figure 3—figure supplement 2C ) . Since this algorithm can vary when given a random seed , we performed the analysis 100 times , and the most common result ( ∼55% of cases ) was shown . No cases were observed in any of these repetitions in which the triangles formed by different categories intersected or overlapped . Triangle plots ( Figure 3C , D ) of sensory responses to male female and predator stimuli were generated in three steps . First , the pre-stimulus firing rate was subtracted from the post-stimulus firing rate of each unit in order to account for differences in baseline firing rates . The absolute value of the baseline subtracted response was then taken to account for responses that consisted of a reduction in activity as compared to the pre-stimulus epoch . As can be seen in Figure 1—figure supplement 2A , such cases are an extreme minority in our data as units typically were inactive before a stimulus was presented and respond to stimuli with an increase in activity . Second , baseline adjusted firing rates were normalized by dividing by the summed responses to all stimuli such that:respMale+respFemale+respPred=1 . At this point , each unit can be plotted on an equilateral triangle with vertices at ( 0 , 0 , 1 ) , ( 0 , 1 , 0 ) , and ( 1 , 0 , 0 ) , where the three axes represent the normalized responses to male , female and predator stimuli . Third , the plane segment defined by the intersections of this plane with the three Cartesian axes was then rotated for ease of visualization . The selectivity index ( Figure 3B ) was calculated as the normalized distance from the center of this triangle such that data with the highest selectivity ( located at vertices of the triangles ) had a selectivity of 1 and the least selective units ( center of triangles ) had a selectivity of 0 . The calculation for each individual unit was as follows:Response selectivity= ( p−x¯ ) 2+ ( m−x¯ ) 2+ ( f−x¯ ) 2d Where p , m , and f are the responses to predator , male , and female stimuli respectively . The value corresponding to equal responses to all stimuli is x¯ , and d is the maximum possible value of ( p−x¯ ) 2+ ( m−x¯ ) 2+ ( f−x¯ ) 2 for standardization . In our case , x¯=13 and d = 23 . During a single surgical session , a small craniotomy was made dorsal to the MeA and a glass electrode ( 1 mm borosilicate; 20-25 µm outer tip diameter ) filled with the retrograde neuronal tracer Fluoro-Gold ( Fluorochrome , LLC; 10% in water ) was lowered into the MeA . Prior to insertion into the brain , a retaining current of −2 μA was applied and the outer surface of the glass electrode was thoroughly washed , to minimize unintentional labeling along the electrode path dorsal to the injection site . At the injection site ( 1 . 80 mm lateral , 1 . 8 mm posterior , and 5 . 5 mm ventral from Bregma ) the polarity of current was switched and +5 μA , constant current was applied for 10 min . Following each injection , a −2 μA retaining current was reapplied for 10 min prior to and during the retraction of the electrode . The scalp was then sutured and a small amount of Vetbond was applied to close the wound . Animals were euthanized following a 7-day post-operative survival period , and the brains were fixed in 4% paraformaldehyde in saline . Serial vibratome sections ( 50 μm thickness ) of the entire brain and olfactory bulbs were cut , counterstained with NeuroTrace Green ( Invitrogen ) , and mounted on gelatin-coated slides . Fluoro-Gold labeling was visualized and analyzed using a fluorescence microscope with UV filters . All histological sections containing either AOB or MOB were counted and included in the data analysis . Estrogen treatments consisted of , 5 µg of 17β Estradiol-Benzoate ( Sigma ) dissolved in 0 . 1 ml sesame oil and injected subcutaneously to female pups on days P1 , P8 , and P15 ( Wu et al . , 2009 ) . Mice were subsequently tested between the ages of 10 and 12 weeks .	Many animals emit and detect chemicals known as pheromones to communicate with other members of their own species . Animals also rely on chemical signals from other species to warn them , for example , that a predator is nearby . Many of these chemical signals—which are present in sweat , tears , urine , and saliva—are detected by a structure called the vomeronasal organ , which is located at the base of the nasal cavity . When this organ detects a particular chemical signal , it broadcasts this information to a network of brain regions that generates an appropriate behavioral response . Two structures within this network , the accessory olfactory bulb and the medial amygdala , play an important role in modifying this signal before it reaches its final destination—a region of the brain called the hypothalamus . Activation of the hypothalamus by the signal triggers changes in the animal's behavior . Although the anatomical details of this pathway have been widely studied , it is not clear how information is actually transmitted along it . Now , Bergan et al . have provided insights into this process by recording signals in the brains of anesthetized mice exposed to specific stimuli . Whereas neurons in the accessory olfactory bulb responded similarly in male and female mice , those in the medial amygdala showed a preference for female urine in male mice , and a preference for male urine in the case of females . This is the first direct demonstration of differences in sensory processing in the brains of male and female mammals . These differences are thought to result from the actions of sex hormones , particularly estrogen , on brain circuits during development . Consistent with this , neurons in the medial amygdala of male mice with reduced levels of estrogen showed a reduced preference for female urine compared to control males . Similarly , female mice that had been previously exposed to high levels of estrogen as pups showed a reduced preference for male urine compared to controls . In addition to increasing understanding of how chemical signals—including pheromones—influence the responses of rodents to other animals , the work of Bergan et al . has provided clues to the neural mechanisms that underlie sex-specific differences in behaviors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Sex-specific processing of social cues in the medial amygdala
Kainate receptors ( KARs ) are a subfamily of glutamate receptors mediating excitatory synaptic transmission and Neto proteins are recently identified auxiliary subunits for KARs . However , the roles of Neto proteins in the synaptic trafficking of KAR GluK1 are poorly understood . Here , using the hippocampal CA1 pyramidal neuron as a null background system we find that surface expression of GluK1 receptor itself is very limited and is not targeted to excitatory synapses . Both Neto1 and Neto2 profoundly increase GluK1 surface expression and also drive GluK1 to synapses . However , the regulation GluK1 synaptic targeting by Neto proteins is independent of their role in promoting surface trafficking . Interestingly , GluK1 is excluded from synapses expressing AMPA receptors and is selectively incorporated into silent synapses . Neto2 , but not Neto1 , slows GluK1 deactivation , whereas Neto1 speeds GluK1 desensitization and Neto2 slows desensitization . These results establish critical roles for Neto auxiliary subunits controlling KARs properties and synaptic incorporation . Most excitatory synaptic transmission in the brain is mediated by glutamate acting on AMPA and NMDA subtypes of glutamate receptors . However , there is a third subtype of ionotropic glutamate receptor termed kainate receptor ( KAR ) comprising GluK1-5 subunits . These receptors are unusual in that they are expressed at only a subset of glutamatergic synapses ( Contractor et al . , 2011; Jane et al . , 2009; Lerma and Marques , 2013 ) . The most studied synaptic KARs are those expressed at hippocampal CA3 mossy fiber synapses ( Nicoll and Schmitz , 2005 ) . These receptors are expressed postsynaptically and generate a slow EPSC . They are also expressed presynaptically and contribute to the profound frequency facilitation , a hallmark of these synapses . In the CA1 region of the hippocampus , KARs are expressed postsynaptically at excitatory synapses in interneurons ( Cossart et al . , 1998; Frerking et al . , 1998 ) . However , no detectable synaptic KAR EPSCs have been recorded from CA1 pyramidal neurons ( Bureau et al . , 1999; Castillo et al . , 1997; Granger et al . , 2013 ) , despite the fact that functional KARs are expressed on these neurons ( Bureau et al . , 1999; Ruano et al . , 1995 ) . What might determine whether an excitatory synapse expresses KARs ? Recently , auxiliary subunits of KARs , referred to as Neto1 and Neto2 , have been identified ( Copits and Swanson , 2012; Straub and Tomita , 2012; Zhang et al . , 2009 ) . These neurophilin tolloid-like proteins are single pass transmembrane CUB ( complement C1r/C1s , Uegf and Bmp1 ) domain-containing proteins . Both Neto1 and Neto2 are known to alter the kinetics of KARs ( Copits et al . , 2011; Straub et al . , 2011; Zhang et al . , 2009 ) . More specifically Neto2 slows deactivation and desensitization of GluK2 receptors ( Zhang et al . , 2009 ) . Neto1 slows deactivation and desensitization of GluK2/5 and deletion of Neto1 in mice speeds the decay of the KAR-mediated hippocampal mossy fiber EPSC ( Straub et al . , 2011; Tang et al . , 2011; Wyeth et al . , 2014 ) . Thus Neto1 can largely explain the biophysical mismatch between heterologously expressed KARs and endogenously expressed KARs . However , the study of the interaction between GluK1 receptor and Neto proteins is limited . It has been reported that Neto1 speeds GluK1 desensitization , whereas Neto2 slows it ( Copits et al . , 2011 ) , but deactivation was not examined . Although the primary role of Neto proteins appears to be the modulation of KAR function , their role in receptor trafficking is less clear . Neto2 has no effect on the surface expression of GluK2 in oocytes ( Zhang et al . , 2009 ) , although it has been reported to enhance surface expression of GluK1 in HEK cells and cultured neurons ( Copits et al . , 2011 ) . The knock-out of Neto1 in mice does not alter the neuronal surface expression or synaptic localization of GluK2/5 ( Straub et al . , 2011 ) , although other studies reported a decrease in PSD expression of GluK2 when Neto1 was knocked-out ( Tang et al . , 2011; Wyeth et al . , 2014 ) . Finally , it has been reported that Neto2 can target GluK1 to synapses of primary cultured neurons ( Copits et al . , 2011; Palacios-Filardo et al . , 2014 ) . However , it remains controversial whether Neto1 and Neto2 are required for the surface and synaptic expression of GluK1 receptor . If so , it remains unclear whether the bases for these two trafficking steps are the same or not . The lack of endogenous expression of GluK1 in CA1 pyramidal neuron provides a null background in which to study the rules governing GluK1 function . Indeed , recent studies have shown that expression of GluK1 and Neto2 results in the appearance of KAR synaptic currents ( Copits et al . , 2011; Granger et al . , 2013; Palacios-Filardo et al . , 2014 ) . Therefore , we have selected the CA1 neuron as a model to study the roles of Neto1 and Neto2 in the surface and synaptic trafficking and kinetics of the GluK1 receptor . CA1 pyramidal neurons express functional kainate receptors ( Bureau et al . , 1999 ) . However , no detectable synaptic KAR-mediated responses can be detected ( Bureau et al . , 1999; Castillo et al . , 1997; Granger et al . , 2013 ) . We wondered if the lack of synaptic responses might be due to a limited expression of the auxiliary subunit Neto1 or Neto2 ( Ng et al . , 2009; Palacios-Filardo et al . , 2014 ) . We therefore expressed these proteins exogenously in CA1 neurons of cultured rat hippocampal slices through biolistic transfection and measured the synaptic responses by dual whole-cell recordings . Neither Neto1 ( Figure 1A1 ) nor Neto2 ( Figure 1B1 ) had any effect on the size of the synaptic response recorded at −70 mV or the NMDA receptor ( NMDAR ) response recorded at +40 mV ( Figure 1A2 and B2 ) . Moreover , overexpression had no effect on paired-pulse ratio , a measure of presynaptic release probability ( Neto1 vs control: 1 . 29 ± 0 . 11 vs 1 . 23 ± 0 . 08 , p>0 . 05; Neto2 vs control: 1 . 56 ± 0 . 08 vs 1 . 46 ± 0 . 1 , p>0 . 05 ) . One possibility is that KARs were recruited to the synapse , but that they replaced synaptic AMPA receptors ( AMPARs ) . To test this possibility we applied the AMPAR selective antagonist GYKI53655 . The antagonist completely blocked the responses both in Neto1 ( Figure 1A1 ) and Neto2 ( Figure 1B1 ) expressing neurons , suggesting that Neto proteins cannot promote incorporation of the endogenous KARs into synapses . It should be noted that in dissociated neuronal cultures expression of Neto1 or Neto2 generated infrequent KAR mediated synaptic responses in a small fraction of cells ( Palacios-Filardo et al . , 2014 ) . Perhaps the lack of synaptic KARs is due to the limited expression of these receptors in these neurons . We thus expressed GluK1 , but this did not affect the size of the response recorded at −70 mV ( Figure 1C1 ) or the NMDAR response ( Figure 1C2 ) , as well as the paired-pulse ratio ( GluK1 vs control: 1 . 89 ± 0 . 17 vs 1 . 89 ± 0 . 22 , p>0 . 05 ) . Furthermore , GYKI53655 fully blocked the EPSCs indicating that functional KARs were not recruited to the synapse . We next expressed GluK1 together with Neto1 and in this case there was a large increase in the size of synaptic response recorded at −70 mV and GYKI53655 only partially blocked the response ( Figure 2A ) . We selected a concentration of 100 μM GYKI53655 to ensure that all AMPARs were blocked ( Bleakman et al . , 1996 ) . This concentration , however , will block approximately 20% of KAR mediated responses ( Bleakman et al . , 1996 ) and thus the currents remaining in GYKI53655 underestimate the actual contribution of GluK1 receptors to synaptic transmission . These experiments were repeated by expressing GluK1 along with Neto2 . As reported previously ( Granger et al . , 2013 ) the synaptic response was greatly increased and GYKI53655 only partially blocked the response ( Figure 2B ) . Although presynaptic KARs are known to regulate glutamate release at mossy fibers , sparse expression of GluK1 receptors in CA1 neurons has no effect on presynaptic release probability as there is no significant change of paired-pulse ratio ( GluK1/Neto1 vs control: 1 . 46 ± 0 . 18 vs 1 . 7 ± 0 . 26 , p>0 . 05; GluK1/Neto2 vs control: 1 . 36 ± 0 . 12 vs 1 . 52 ± 0 . 16 , p>0 . 05 ) . These findings and those in Figure 1 are summarized in Figure 2C , showing that synaptic KAR responses are only observed when GluK1 is expressed along with either Neto1 or Neto2 . To determine if the synaptic delivery of KARs is depended on synaptic activity , we incubated the cultured slices in NBQX and AP5 to inhibit AMPARs and NMDARs activation during the expression of KARs . We then compared the evoked synaptic responses between experimental and control neurons . However , the receptor antagonists did not prevent the synaptic incorporation of either GluK1/Neto1 or GluK1/Neto2 ( Figure 2—figure supplement 1 ) , suggesting that synaptic activity is not required for Neto-dependent GluK1 synaptic trafficking . 10 . 7554/eLife . 11682 . 003Figure 1 . Individual overexpression of Neto1 , Neto2 or GluK1 has no effect on synaptic transmission . Rat hippocampal slice cultures were biolistically transfected with Neto1 ( A , n=12 ) , Neto2 ( B , n=11 ) or GluK1 ( C , n=22 ) . Simultaneous dual whole-cell recordings from a transfected CA1 pyramidal neuron ( green trace ) and a neighboring wild type one ( black trace ) were performed . The evoked EPSCs ( eEPSCs ) were measured at −70 mV and +40 mV ( the current amplitudes were measured 100 ms after stimulation ) . Open and filled circles represent amplitudes for single pairs and mean ± SEM , respectively . Insets show sample current traces from control ( black ) and experimental ( green ) cells . The scale bars for representative eEPSC trace were 25 pA and 25 ms . Bar graphs show normalized eEPSC amplitudes ( mean ± SEM ) of −70 mV ( A1 , 82 . 24 ± 14 . 64% control , p > 0 . 05; B1 , 77 . 9 ± 12 . 9% control , p > 0 . 05 and C1 , 116 . 58 ± 15 . 53% control , p > 0 . 05 ) and +40 mV ( A2 , 81 . 74 ± 8 . 42% control , p > 0 . 05; B2 , 78 . 53 ± 14 . 35% control , p > 0 . 05 and C2 , 101 . 8 ± 12 . 06% control , p > 0 . 05 ) presented in scatter plots . All the statistical analyses are compared to respective control neurons with two-tailed Wilcoxon signed-rank sum test . The eEPSC amplitudes measured at −70 mV after GYKI53655 ( 100 μM ) wash-in in A1 , B1 and C1 were also normalized according to respective pretreated control neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 00310 . 7554/eLife . 11682 . 004Figure 2 . Neto1 and Neto2 promote GluK1 receptor synaptic targeting . ( A ) Scatter plots show eEPSC amplitudes of control and GluK1/Neto1-cotransfected CA1 neurons in rat hippocampal slice cultures measured at −70 mV in the absence or presence of GYKI53655 . Filled circles show mean ± SEM . Insets show sample current traces from control ( black ) and GluK1/Neto1-expressing ( green ) cells . The scale bars for representative eEPSC trace were 50 pA and 25 ms . Bar graph show normalized eEPSC amplitudes ( mean ± SEM ) of pretreated ( n=19 , 470 . 65 ± 85 . 6% control , *** p < 0 . 0005 ) and GYKI53655 treated ( n=7 , 150 . 72 ± 51 . 8% control pretreatment , * p < 0 . 05 ) cells . ( B ) Scatter plots show eEPSC amplitudes of control and GluK1/Neto2-cotransfected CA1 neurons in rat hippocampal slice cultures measured at −70 mV in the absence or presence of GYKI53655 . Filled circles show mean ± SEM . Insets show sample current traces from control ( black ) and GluK1/Neto1-expressing ( green ) cells . The scale bars for representative eEPSC trace were 50 pA and 25 ms . Bar graph show normalized eEPSC amplitudes ( mean ± SEM ) of pretreated ( n=17 , 689 . 52 ± 195 . 16% control , *** p < 0 . 0005 ) and GYKI53655 treated ( n=7 , 317 . 63 ± 83 . 12% control pretreatment , * p < 0 . 05 ) cells . ( C ) Summary of the normalized evoked EPSC amplitudes at −70 mV as percent of respective control ± SEM for each indicated transfection . All the statistical analyses are compared to respective control neurons with two-tailed Wilcoxon signed-rank sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 00410 . 7554/eLife . 11682 . 005Figure 2—figure supplement 1 . Neto1 and Neto2 regulation of GluK1 synaptic expression is independent of synaptic activity . Scatter plots show eEPSC amplitudes of control and GluK1/Neto1 ( A ) or GluK1/Neto2 ( B ) transfected neurons in rat hippocampal slice . Slices were treated with 25 μM NBQX and 100 μM AP5 during the incubation culture and measured at −70 mV . Filled circles show mean ± SEM . Insets show sample current traces from control ( black ) and experimental ( green ) cells . The scale bars for representative eEPSC trace were 50 pA and 25 ms . Bar graph show normalized eEPSC amplitudes ( mean ± SEM ) ( A: GluK1/Neto1 , n=6 , 493 . 41 ± 206 . 95% control , * p < 0 . 05; B: GluK1/Neto2 , n=7 , 734 . 43 ± 321 . 83% control , * p < 0 . 05 ) presented in scatter plots . All the statistical analyses are compared to respective control neurons with two-tailed Wilcoxon signed-rank sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 00510 . 7554/eLife . 11682 . 006Figure 2—figure supplement 2 . NMDAR EPSCs are increased in GluK1/Neto1 and GluK1/Neto2 expressing neurons . Scatter plots show eEPSC amplitudes of control and GluK1/Neto1 or GluK1/Neto2 transfected neurons in rat hippocampal slice cultures measured at +40 mV ( 100 ms after stimulation ) . Filled circles show mean ± SEM . Insets show sample current traces from control ( black ) and experimental ( green ) cells . The scale bars for representative eEPSC trace were 50 pA and 25 ms . Bar graphs show normalized eEPSC amplitudes ( mean ± SEM ) ( A: GluK1/Neto1 , n=7 , 162 . 15 ± 39 . 44% control , * p < 0 . 05; B: GluK1/Neto2 , n=10 , 134 . 28 ± 16 . 85% control , ** p < 0 . 01 ) presented in scatter plots . ( C ) Summary of the normalized evoked NMDAR EPSC amplitudes at +40 mV as percent of respective control ± SEM for each transfection . ( D ) Scatter plots show peak eEPSC amplitudes of control and GluK1/Neto1 transfected neurons in rat hippocampal slice cultures measured at +40 mV in the presence of NBQX ( 25 μM ) . Bar graphs show normalized eEPSC amplitudes ( mean ± SEM ) ( n=7 , 178 . 73 ± 34 . 72% control , * p < 0 . 05 ) presented in scatter plots . All the statistical analyses are compared to respective control neurons two-tailed Wilcoxon signed-rank sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 006 Interestingly , when GluK1 was expressed along with Neto1 or Neto2 there was a significant increase in the size of the NMDAR EPSCs ( Figure 2—figure supplement 2A–C ) . We further confirm the increased NMDAR-mediated synaptic response with recordings done in the presence of NBQX to block AMPAR and KAR EPSCs ( Figure 2—figure supplement 2D ) . These results raise the possibility that GluK1 together with Netos has a synaptogenic effect . We , therefore , filled neurons with Alexa Fluor 568 dye and analyzed the density of dendritic spines as a proxy for the density of excitatory synapses ( Figure 3A and B ) . However , we found no difference in spine density in neurons expressing GluK1/Neto1 or GluK1/Neto2 compared to control . 10 . 7554/eLife . 11682 . 007Figure 3 . GluK1 synaptic expression has no effect on spinogenesis and does not replace endogenous synaptic AMPA receptors . ( A ) Sample images of primary apical dendrites from control ( upper ) and GluK1/Neto1 overexpressed ( lower ) neurons imaged using super-resolution structured illumination microscopy ( SIM ) . Bar graph in right shows average spine density ( control , n = 8 , 0 . 56 ± 0 . 06/μm; GluK1/Neto1 , n = 9 , 0 . 5 ± 0 . 06/μm; p > 0 . 05 ) . Scale bar: 5 μm . ( B ) Sample images of primary apical dendrites from control ( upper ) and GluK1/Neto2 overexpressed ( lower ) neurons imaged using SIM . Bar graph in right shows average spine density ( control , n = 8 , 0 . 41 ± 0 . 03/μm; GluK1/Neto2 , n = 7 , 0 . 49 ± 0 . 03/μm; p > 0 . 05 ) . Scale bar: 5 μm . All the statistical analyses are compared to respective control neurons with unpaired two-tailed t test . ( C and D ) Scatter plots show eEPSC amplitudes of control and GluK1/Neto1 ( C ) or GluK1/Neto2 ( D ) cotransfected neurons in rat hippocampal slice cultures measured at −70 mV in the presence of the GluK1 antagonist ACET ( 1 μM ) . Filled circles show mean ± SEM . Insets show sample current traces from control ( black ) and experimental ( green ) cells . The scale bars for representative eEPSC trace were 25 pA and 25 ms . Bar graph show normalized eEPSC amplitudes ( mean ± SEM ) ( A: n=12 , 112 . 33 ± 15 . 36% control , p > 0 . 05; B: n=9 , 89 . 7 ± 15 . 38% control , p > 0 . 05 ) presented in scatter plots . All the statistical analyses are compared to respective control neurons with two-tailed Wilcoxon signed-rank sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 007 How are KARs incorporated into synapses ? Do they displace synaptic AMPARs or do they add additional receptors to the already activated synapse ? To address these questions we expressed GluK1 with either Neto1 ( Figure 3C ) or Neto2 ( Figure 3D ) and recorded synaptic responses in the presence of the GluK1 selective antagonist ACET . In this case there was no significant difference in the AMPAR-mediated responses between control and experimental neurons , indicating that AMPARs are not displaced by the synaptic expression of KARs . There are two possible explanations for the results in Figure 2 and Figure 3 . Either the expressed KARs populate synapses that already express AMPARs ( Figure 4—figure supplement 1A ) or they are excluded from AMPAR expressing synapses and selectively populate silent synapses , i . e . those that do not express AMPARs ( Figure 4—figure supplement 1B ) . In the former situation one would expect the size of quantal events to be larger , whereas in the latter case one might expect to see primarily a change in frequency . To test these predictions we replaced Ca2+ with Sr2+ , which desynchronizes the induced transmitter release ( Oliet et al . , 1996 ) . We then simultaneously recorded from a control cell and an experimental one expressing GluK1 and Neto1 ( Figure 4A ) or Neto2 ( Figure 5A ) to examine the amplitude and frequency of asynchronous EPSCs ( aEPSCs ) . In cells expressing GluK1 and Neto1 , we observed no change in quantal size ( Figure 4A2 and Figure 4—figure supplement 2A1 ) , but a large increase in frequency ( Figure 4A3 and Figure 4—figure supplement 2A2 ) . We observed the same results when expressing GluK1 and Neto2 ( Figure 5A1-A3 and Figure 4—figure supplement 2C1-C2 ) . These results tell us that expression of KARs results in synapse unsilencing . The finding that the AMPAR EPSC in GluK1 expressing cells is not reduced in ACET indicates that KARs do not cause a net loss of synaptic AMPARs ( Figure 3C and D ) . However , there remains a possibility that synaptic KARs are expressed at all synapses , displacing a portion of synaptic AMPARs to previously silent synapses ( Figure 4—figure supplement 1C ) . In this scenario each synapse would contain a mixture of AMPARs and KARs , and selectively inhibiting KARs activity would be expected to reduce the size of aEPSCs . To test this idea , we repeated the asynchronous electrophysiological recordings in the presence of the GluK1 antagonist ACET . In the presence of ACET the increase in aEPSC frequency in the GluK1/Neto1-expressing neurons ( Figure 4A3 ) was no longer observed ( Figure 4B2-C2 and Figure 4—figure supplement 2B2 ) . Importantly we saw no reduction in the size of aEPSCs ( Figure 4C1 and Figure 4—figure supplement 2B1 ) . We found the same results with GluK1/Neto2 expressing neurons ( Figure 5B2-B3 , Figure 4—figure supplement 2D1-2D2 ) . The lack of change in aEPSC size in ACET ( Figure 4C1 , Figure 5B2 , Figure 4—figure supplement 2B1 and Figure 4—figure supplement 2D1 ) , indicates that synaptic GluK1 and AMPARs do not co-localize at the same synapses . 10 . 7554/eLife . 11682 . 008Figure 4 . Neto1 specifically targets GluK1 receptors to silent synapse . ( A1 ) Representative sample traces of asynchronous EPSCs ( aEPSCs ) simultaneously recorded in the presence of Sr2 + from control ( black ) and GluK1/Neto1-coexpressed ( green ) neurons . The first 50 ms following stimulation was excluded from analysis . The scale bars for single representative aEPSC traces were 10 pA and 10 ms . ( A2 ) aEPSC amplitude is not significantly changed in GluK1/Neto1-expressing neurons ( n=15 , control: 13 . 08 ± 0 . 89 pA , GluK1/Neto1: 13 . 96 ± 0 . 89 pA , p > 0 . 05 ) . Plot shows single pairs ( open circles ) and mean ± SEM ( filled circles ) . ( A3 ) aEPSC frequency is significantly increased in neurons expressing GluK1 and Neto1 ( n=15 , control: 1 . 65 ± 0 . 38 Hz , GluK1/Neto1: 4 . 01 ± 0 . 79 Hz , * p < 0 . 0001 ) . Plot shows single pairs ( open circles ) and mean ± SEM ( filled circles ) . ( B1 and B2 ) Representative sample traces of aEPSCs recorded in the presence of Sr2+ from control ( black ) and GluK1/Neto1-coexpressed ( green ) neurons before ( B1 , left ) and after ( B2 , right ) ACET treatment . The first 50 ms following stimulation was excluded from analysis . The scale bars for single representative aEPSC trace were 10 pA and 10 ms . ( C1 ) Plot shows single pairs ( open circles ) and mean ± SEM ( filled circles ) of aEPSC amplitude from control and GluK1/Neto1 transfected neurons ( n=15 , control: 12 . 41 ± 0 . 82 pA , GluK1/Neto1: 12 . 38 ± 0 . 83 pA , p > 0 . 05 ) recorded in the presence of ACET . ( C2 ) The aEPSC frequency in neurons expressing GluK1 and Neto1 is not significantly different from control ones in the presence of ACET ( n=15 , control: 1 . 21 ± 0 . 22 Hz , GluK1/Neto1: 1 . 46 ± 0 . 26 Hz , p > 0 . 05 ) . Plot shows single pairs ( open circles ) and mean ± SEM ( filled circles ) . ( D1 and D2 ) Representative sample traces of aEPSCs recorded in the presence of Sr2 from control ( black ) and GluK1/Neto1-coexpressed ( green ) neurons before ( D1 , left ) and after ( D2 , right ) GYKI53655 ( 30 μM ) treatment . The first 50 ms following stimulation was excluded from analysis . The scale bars for single representative aEPSC trace were 10 pA and 10 ms . ( D3-D4 ) Plots show single paired ( open circles ) and mean ± SEM ( filled circles ) of aEPSC amplitude ( D3 ) and frequency ( D4 ) from control and GluK1/Neto1-cotransfected neurons before ( black and green , n=5; amplitude: control: 14 . 81 ± 0 . 85 pA , GluK1/Neto1: 14 . 3 ± 1 . 13 pA , p > 0 . 05; frequency: control: 1 . 66 ± 0 . 19 Hz , GluK1/Neto1: 2 . 88 ± 0 . 54 Hz , p < 0 . 05 ) and after 30 μM GYKI53655 treatment ( blue and red , amplitude: GluK1/Neto1: 11 . 71 ± 1 . 66 pA; frequency: GluK1/Neto1: 0 . 7 ± 0 . 15 Hz ) . All the statistical analyses are compared to respective control neurons with two-tailed Wilcoxon signed-rank sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 00810 . 7554/eLife . 11682 . 009Figure 4—figure supplement 1 . The models of Neto proteins-regulated synaptic GluK1 receptors localization . Endogenous AMPARs are blue and GluK1 receptors are green . ( A ) AMPARs and GluK1 receptors are localized at the same synapses . ( B ) AMPARs and GluK1 receptors are localized at different synapses . ( C ) Endogenous AMPARs are redistributed by synaptic GluK1 receptors and some of the synapses are populated with both AMPARs and GluK1 receptors . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 00910 . 7554/eLife . 11682 . 010Figure 4—figure supplement 2 . GluK1 receptors specifically traffic to silent synapses in the presence of Neto1 or Neto2 . ( A1 and A2 ) Cumulative distribution plots of aEPSC amplitude and frequency from control ( black ) and GluK1/Neto1-coexpressed ( green ) neurons . Cumulative distribution functions show no irregularities . ( B1 and B2 ) Cumulative distribution plots of aEPSC amplitude and frequency from control ( black ) and GluK1/Neto1-coexpressed ( green ) neurons in the presence of ACET . Cumulative distribution functions show no irregularities . ( C1 and C2 ) Cumulative distribution plots of aEPSC amplitude and frequency from control ( black ) and GluK1/Neto2-coexpressed ( green ) neurons . Cumulative distribution functions show no irregularities . ( D1 and D2 ) Cumulative distribution plots of aEPSC amplitude and frequency from control ( black ) and GluK1/Neto2-coexpressed ( green ) neurons in the presence of ACET . Cumulative distribution functions show no irregularities . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 01010 . 7554/eLife . 11682 . 011Figure 5 . Neto2 specifically targets GluK1 receptors to silent synapse . ( A1 ) Representative sample traces of aEPSCsfrom control ( black ) and GluK1/Neto2-coexpressed ( green ) neurons . The first 50 ms following stimulation was excluded from analysis . The scale bars for single representative aEPSC trace were 10 pA and 10 ms . ( A2 ) aEPSC amplitude is not significantly changed in GluK1/Neto2 expressed neurons ( n=15 , control: 10 . 56 ± 0 . 89 pA , GluK1/Neto2: 11 . 44 ± 0 . 67 pA , p > 0 . 05 ) . Plot shows single pairs ( open circles ) and mean ± SEM ( filled circles ) . ( A3 ) aEPSC frequency is significantly increased in neurons expressing GluK1 and Neto2 ( n=15 , control: 1 . 32 ± 0 . 17 Hz , GluK1/Neto2: 2 . 45 ± 0 . 26 Hz , * p < 0 . 0005 ) . Plot shows single pairs ( open circles ) and mean ± SEM ( filled circles ) . ( B1 ) Representative sample traces of aEPSCs simultaneously recorded in the presence of Sr2+ and ACET from control ( black ) and GluK1/Neto2-coexpressed ( green ) neurons . The first 50 ms following stimulation was excluded from analysis . The scale bars for single representative aEPSC trace were 10 pA and 10 ms . ( B2 ) Plot shows single pairs ( open circles ) and mean ± SEM ( filled circles ) of aEPSC amplitude from control and GluK1/Neto2 transfected neurons ( n=12 , control: 12 . 03 ± 0 . 74 pA , GluK1/Neto2: 11 . 1 ± 1 . 19 pA , p > 0 . 05 ) . ( B3 ) The aEPSC frequency in neurons expressing GluK1 and Neto2 is not significantly different from control ones in the presence of ACET ( n=12 , control: 1 . 42 ± 0 . 33 Hz , GluK1/Neto1: 1 . 48 ± 0 . 25 Hz , p > 0 . 05 ) . Plot shows single pairs ( open circles ) and mean ± SEM ( filled circles ) . All the statistical analyses are compared to respective control neurons with two-tailed Wilcoxon signed-rank sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 011 We also examined the effect of GYKI53655 on aEPSCs and found that at a relative low concentration ( 30 μM ) the aEPSCs from control cells were totally blocked , while a large reduction in aEPSC frequency and minimal reduction in aEPSC amplitude from GluK1/Neto1-transfected cells ( Figure 4D1-D4 ) were observed . Taken together , these results suggest that KARs are excluded from synapses that are already populated with AMPARs . Rather they appear to selectively populate synapses that lack AMPARs , i . e . silent synapses . Perhaps even more intriguing is that the average size of GluK1-mediated quantal events is the same as the AMPAR-mediated events . This implies that the average single channel conductance and number of receptors at GluK1 synapses is the same as that for AMPAR expressing synapses , or more likely , that there is some type of homeostatic process that governs the number of synaptic KARs . Our above results indicate that synaptic trafficking of GluK1 receptors is dependent on Neto proteins . We next examined which region ( s ) in Neto proteins are responsible for targeting GluK1 to the synapse . As Neto1 and Neto2 are single transmembrane proteins , we first tested the involvement of their intracellular domains . Deletion of the entire C-tail ( Δ161 ) of Neto1 prevents the targeting of GluK1 receptors to the synapse ( Figure 6A , B and H ) whereas deletion of the last 4 amino acids ( Δ4 ) , a putative PDZ binding motif , has no significant effect ( Figure 6F and H ) . Deletion of the last 20 amino acids ( Δ20 ) of Neto1 has a substantial effect on synaptic GluK1 currents ( Figure 6E and H ) , as did a larger deletion of 41 amino acids ( Δ41 ) ( Figure 6D and H ) . These truncation experiments suggest that the last 20 amino acids of Neto1 are critical for the synaptic incorporation of functional GluK1 receptors . It has been reported that the AMPAR auxiliary subunit stargazin can be phosphorylated in the intracellular C-tail , which regulates its interaction with negatively charged lipid bilayers and therefore synaptic AMPAR activity ( Sumioka et al . , 2010; Tomita et al . , 2005 ) . We therefore mutated three serines and a tyrosine in this region to alanines simultaneously ( S3Y/A ) and found that the Neto1S3Y/A mutant disrupted GluK1 synaptic expression to the same extent as deleting the last 20 amino acids ( Figure 6G and H ) . 10 . 7554/eLife . 11682 . 012Figure 6 . Neto1-mediated GluK1 synaptic trafficking is dependent on the critical serine and tyrosine residues in the intracellular region . ( A ) Amino acid sequence of the Neto1 C-tail . The truncation mutants generated are indicated by arrows . The Neto1S3Y/A is a mutant in which the highlighted three serines and one tyrosine residues within the last 20 amino acids are mutated to alanines . ( B–G ) Scatter plots of eEPSCs at −70 mV for GluK1 co-expressed with various Neto1 mutants . Open circles are individual pairs and filled are mean ± SEM . Insets show sample current traces from control ( black ) and experimental ( green ) cells . The scale bars for representative eEPSC trace were 50 pA and 25 ms . ( H ) Normalized evoked EPSC amplitudes at −70 mV as percent of respective control ± SEM for each transfection ( GluK1/Neto1: n=19 , 470 . 65 ± 85 . 59% control; GluK1/Neto1Δ161: n=18 , 87 . 7 ± 8 . 67% control , * p < 0 . 0001; GluK1/Neto1Δ81: n=18 , 97 . 8 ± 13 . 1% control , * p < 0 . 0001; GluK1/Neto1Δ41: n=24 , 179 . 2 ± 39 . 0% control , p < 0 . 005; GluK1/Neto1Δ20: n=14 , 178 . 1 ± 25 . 5% control , p < 0 . 005; GluK1/Neto1Δ4: n=23 , 365 . 62 ± 80 . 22% control , p > 0 . 05; GluK1/Neto1S3Y/A: n=17 , 189 . 48 ± 40 . 26% control , * p < 0 . 05 ) . All the statistical analyses are tested with the group co-overexpressing GluK1 and wildtype Neto1 using Mann-Whitney U-test . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 012 We next looked for the domain ( s ) in Neto2 that are critical for GluK1 synaptic trafficking . Deletion of the entire C-terminal domain ( Δ148 ) fully diminished GluK1 receptor synaptic targeting by Neto2 ( Figure 7A , B and I ) whereas deleting the last 4 amino acids ( Δ4 ) has no effect on the function of Neto2 ( Figure 7F and I ) . Moreover , deletions of 115 amino acids ( Δ115 ) ( Figure 7C and I ) and 86 amino acids ( Figure 7D and I ) also significantly impaired the function of Neto2 . These results indicate that the last 86 amino acids of Neto2 are critical for the incorporation of synaptic GluK1 receptors . To narrow down the critical region of Neto2 , we carried out further deletions within the last 86 amino acids . Of particular importance is the region between 86 and 74 amino acids ( Δ86-74 ) as its deletion significantly impaired Neto2-regulated GluK1 synaptic targeting ( Figure 7G and I ) , whereas deletion of last 74 amino acids ( Δ74 ) has no significant effect on GluK1-mediated synaptic response ( Figure 7E and I ) . Furthermore , mutation of the serines in this region as well as the threonine just next to this region to alanines ( S4T/A ) also impaired GluK1 synaptic expression to the same extent as this critical deletion mutant ( Δ86-74 ) ( Figure 7H and I ) . 10 . 7554/eLife . 11682 . 013Figure 7 . Neto2-mediated GluK1 synaptic trafficking is dependent on the critical serine and threonine residues in the intracellular region . ( A ) Amino acid sequence of the Neto2 C-tail . The truncation mutants generated are indicated by arrows . The Neto2S4T/A is a mutant in which the highlighted four serines within region 86-74 and one threonine just next to this region are mutated to alanines . ( B–H ) Scatter plots of eEPSCs at −70 mV for GluK1 co-expressed with various Neto2 mutants . Open circles are individual pairs and filled are mean ± SEM . Insets show sample current traces from control ( black ) and experimental ( green ) cells . The scale bars for representative eEPSC trace were 50 pA and 25 ms . ( I ) Normalized evoked EPSC amplitudes at −70 mV as percent of respective control ± SEM for each transfection ( GluK1/Neto2: n=17 , 473 . 69 ± 65 . 08% control; GluK1/Neto2Δ148: n=19 , 109 . 96 ± 13 . 27% control , * p < 0 . 0001; GluK1/Neto2Δ115: n=25 , 148 . 51 ± 25 . 59% control , * p < 0 . 0005; GluK1/Neto2Δ86: n=24 , 163 . 72 ± 18 . 08% control , * p < 0 . 0005; GluK1/Neto2Δ74: n=15 , 535 . 85 ± 129 . 35% control , p > 0 . 05; GluK1/Neto2Δ4: n=16 , 567 . 38 ± 134 . 55% control , p > 0 . 05; GluK1/Neto2Δ86–74: n=21 , 219 . 09 ± 35 . 39% control , p < 0 . 01; GluK1/Neto2S4T/A: n=18 , 229 . 98 ± 32 . 99% control , p < 0 . 01 ) . All the statistical analyses are tested with the group co-overexpressing GluK1 and wildtype Neto1 using Mann-Whitney U-test . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 013 The synaptic delivery of KARs involves at least two steps . The receptors first have to be properly assembled and delivered to the surface , followed by targeting of the surface receptors to the synapse . As it has been reported that Neto2 increases GluK1 surface expression ( Copits et al . , 2011 ) , Neto proteins may simply increase the pool of surface receptors to such an extent that the receptors passively populate synapses . To test this possibility , we measured surface GluK1 expression electrophysiologically by pulling outside-out membrane patches from the soma , with the goal of determining whether Neto mutants that had impaired KAR synaptic localization were simply unable to increase surface KAR trafficking . Since KAR currents desensitized rapidly , glutamate was applied using ultra-fast perfusion . All currents were recorded in the presence of GYKI53655 ( 100 μM ) to block AMPAR-mediated response . In wild type patches from CA1 neurons , we were unable to detect any glutamate-evoked current ( Figure 8A ) . This contrasts to KAR-mediated currents recorded in a whole cell recording configuration with bath application of agonist ( Bureau et al . , 1999 ) . Presumably the low density of these receptors accounts for the lack of current in outside-out patches . In neurons expressing only GluK1 we saw small , but significant , glutamate evoked currents whereas in patches from neurons co-expressing either Neto1 or Neto2 with GluK1 very large currents were recorded ( Figure 8A ) . We then examined the Neto mutants that greatly impaired synaptic responses and looked for the Neto mutants to modulate KAR surface expression . If there were an additional targeting role for Netos we would expect some of these mutants , which impaired GluK1-mediated synaptic responses , to generate extrasynaptic KAR currents similar in magnitude to that recorded when wild type Netos were expressed with GluK1 . Indeed , both Neto1S3Y/A and Neto2S4T/A mutants generated currents in outside-out patches of similar size to those generated by wild type Neto1 and Neto2 ( Figure 8A ) . These results provide strong evidence that there is , in fact , a role for these auxiliary proteins in targeting surface GluK1 receptors to synapses . 10 . 7554/eLife . 11682 . 014Figure 8 . Neto1 and Neto2 increase GluK1 receptor surface trafficking and biophysical properties . ( A ) Bar graphs show the amplitude of GluK1 currents ( mean ± SEM ) from outside-out patches pulled from wild type and transfected CA1 neurons with indicated plasmids and exposed to 1 or 100 ms applications of 10 mM glutamate and 100 μM GYKI53655 ( WT , n=7 , 8 . 57 ± 2 . 51 pA , * p < 0 . 0005; GluK1: n=10 , 81 . 65 ± 11 . 26 pA; GluK1/Neto1: n=9 , 1231 . 94 ± 242 . 92 pA , * p < 0 . 0001; GluK1/Neto1S3Y/A: n=7 , 967 . 14 ± 138 . 30 pA , * p < 0 . 0005; GluK1/Neto2: n=10 , 1022 . 84 ± 241 . 81 pA , * p < 0 . 0005; GluK1/Neto2S4T/A: n=9 , 1035 . 22 ± 115 . 00 pA , * p < 0 . 0001 ) . All the statistical analyses are compared to GluK1 single overexpression using Mann-Whitney U-test . Sample traces and scale bar are shown to the right . ( B ) DIV 10 neurons were transfected with HA-GluK1 and Neto1 or Neto2 , as indicated . At DIV 13 , cells were stained for surface GluK1 and the intensity of surface GluK1 was quantitated ( 3 dendrites per neuron ) using Metamorph analysis software . Scale bar , 20 μm . Images at the bottom of each panel are higher magnification from the enclosed region . Scale bar , 5 μm . ( C ) Bar graph shows the surface expression of GluK1 ( mean ± SEM ) from three independent experiments ( GluK1: n=34; GluK1/Neto1: n=33; GluK1/Neto2: n=34 ) . An unpaired two-tailed t-test was used to determine the significance of the data: *** p < 0 . 0001 . ( D and E ) Bar graphs show mean ± SEM GluK1 deactivation ( d , GluK1: n=10 , 3 . 7 ± 0 . 35 ms; GluK1/Neto1: n=7 , 3 . 43 ± 0 . 42 ms , p > 0 . 05; GluK1/Neto2: n=10 , 5 . 33 ± 0 . 46 ms , * p < 0 . 05 ) and desensitization ( e , GluK1: n=6 , 12 . 42 ± 2 . 26 ms; GluK1/Neto1: n=8 , 4 . 93 ± 0 . 59 ms , * p < 0 . 05; GluK1/Neto2: n=7 , 27 . 49 ± 3 . 26 ms , ** p < 0 . 005 ) from outside-out patches pulled from indicated transfection CA1 neurons and exposed to 1 or 100 ms applications of 10 mM glutamate and 100 μM GYKI53655 , respectively . All the statistical analyses are compared to GluK1 single overexpression using Mann-Whitney U-test . Sample traces are shown to the right and are peak-normalized . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 01410 . 7554/eLife . 11682 . 015Figure 8—figure supplement 1 . GluK1 receptor is localized at synapse . ( A ) Electrophysiogical recordings of whole-cell puffing with 1 mM Glutamate ( left ) or 1 mM kainic acid ( right ) from wildtype ( upper ) or HA-GluK1/Neto1 coexpressed ( lower ) CA1 pyramidal neurons . ( B ) Scatter plots show eEPSC amplitudes measured at −70 mV of control and HA-GluK1/Neto1 or Myc-GluK1/Neto1 transfected neurons in rat hippocampal slice . Filled circles show mean ± SEM . Insets show sample current traces from control ( black ) and experimental ( green ) cells . The scale bars for representative eEPSC trace were 50 pA and 25 ms . Bar graph show normalized eEPSC amplitudes ( mean ± SEM ) ( HA-GluK1/Neto1 , n=7 , 192 . 41 ± 99 . 67% control , p > 0 . 05; Myc-GluK1/Neto1 , n=8 , 78 . 89 ± 14 . 04% control , p > 0 . 05 ) presented in scatter plots . ( C ) Scatter plots show eEPSC amplitudes measured at –70 mV of control and GluK1/Myc-Neto2 transfected neurons in rat hippocampal slice . Filled circles show mean ± SEM . Insets show sample current traces from control ( black ) and experimental ( green ) cells . The scale bars for representative eEPSC trace were 100 pA and 25 ms . Bar graph show normalized eEPSC amplitudes ( mean ± SEM ) ( n=10 , 238 . 94 ± 55 . 43% control , * p < 0 . 05 ) presented in scatter plots . ( D ) DIV 10 neurons were transfected with Myc-Neto2 alone or together with GluK1 as indicated . At DIV 13 , cells were stained for surface and intracellular Neto2 as well as presynaptic marker VGLUT1 using Metamorph analysis software . Scale bar , 20 μm . Images at the bottom of each panel are higher magnification from the enclosed region . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11682 . 015 It has been shown that Neto2 has no effect on the surface delivery of the GluK2 receptors but instead causes a large increase in GluK2-evoked current by changing the gating properties of the receptors ( Zhang et al . , 2009 ) . Thus it was important to determine if Netos actually increase the surface expression of GluK1 . We therefore transfected dissociated hippocampal neurons with an HA-tagged GluK1 receptor alone or together with Netos and examined surface expression of GluK1 . In the absence of Netos , the surface labeling was very weak , although the neuron clearly expressed GluK1 ( Figure 8B ) . However , in the presence of either Neto1 or Neto2 , GluK1 was abundantly expressed on the surface ( Figure 8B and C ) . Consistently , GluK1-specific current could be observed through whole-cell puffing with glutamate or kainic acid in the present of Neto1 auxiliary subunit ( Figure 8—figure supplement 1A ) . Together , these results indicate that both Neto1 and Neto2 can drive the robust surface expression of GluK1 . However , the tagged-GluK1 receptors cannot traffic to synapse as the synaptic responses were not increased even coexpressing HA-GluK1 or Myc-GluK1 together with Neto1 protein ( Figure 8—figure supplement 1B ) . To test whether the GluK1 receptor indeed localized at the synapse , we used a tagged-Neto2 mutant which promote GluK1 synaptic expression efficiently ( Figure 8—figure supplement 1C ) and found that the surface expressed GluK1/Neto2 was partially colocalized with presynaptic marker VGLUT1 ( Figure 8—figure supplement 1D ) . Together with the above electrophysiological findings that: ( 1 ) the aEPSCs from the experimental cells were not significantly reduced by GluK1 specific inhibitor ACET ( Figure 4A–C ) ; ( 2 ) the critical mutants Neto1S3Y/A and Neto2S4T/A impair GluK1 synaptic expression while maintain its surface expression ( Figure 6G , 7H and 8A ) , it strongly supports the notion that the GluK1 receptors are indeed localized at the synapses and mediated synaptic transmission . In addition to measuring the size of the peak currents , we also measured the rate of deactivation and desensitization , parameters that might be affected by Neto proteins . Neto1 had no effect on the deactivation of GluK1-mediated currents , but Neto2 did slow deactivation ( Figure 8D ) . By contrast , Neto1 enhanced the rate of desensitization , whereas Neto2 slowed the rate of desensitization ( Figure 8E ) , in agreement with previous results ( Copits et al . , 2011 ) . Expression of GluK1 by itself results in very little KAR surface currents in CA1 pyramidal cells . By contrast co-expression of GluK1 with either Neto1 or Neto2 generates currents approaching a nA in outside out patches . This effect cannot be explained by changes in desensitization , because Neto1 actually increases desensitization and yet generates current of similar magnitude to Neto2 , which slows desensitization . In addition , we used ultra fast application to avoid this possibility . An increase in single channel conductance and/or open probability could contribute to the enhanced currents , but are highly unlikely to account for the massive currents recorded with both Neto1 and Neto2 . Finally , the surface staining of GluK1 was unequivocally enhanced when Neto proteins are coexpressed . All these results demonstrate the critical role of Netos in the delivery of GluK1 to the surface . Neto2 has no effect on the surface expression of GluK2 in oocytes ( Zhang et al . , 2009 ) . The closest comparison to the present result are those of Copits et al . , Copits et al . ( 2011 ) who found that Neto2 , and to a much lesser degree Neto1 , enhanced peak GluK1-mediated currents in HEK cells and Neto2 , but not Neto1 , enhanced surface staining for GluK1 in neurons . This difference between our results and those of Copits et al . , Copits et al . ( 2011 ) might be because different isoforms of GluK1 were used in this and our studies . Sequence comparison of the two isoforms suggests that the intracellular C-tail of GluK1 might be involved in Neto1-regulated surface trafficking . And in agreement with that study ( Copits et al . , 2011 ) , we found that Neto1 enhanced the rate of GluK1 desensitization while Neto2 greatly slowed the rate of desensitization . The mechanism by which Netos modulate delivery is unclear . It could indicate that Neto proteins serve as chaperones and are required for the proper folding and maturation of the KARs , analogous to the role of TARPs in AMPAR trafficking ( Jackson and Nicoll , 2011 ) . Alternatively or additionally , Neto proteins could play a more direct role in delivering mature receptors to the surface . Although CA1 pyramidal cells express functional GluK2 surface receptors ( Bureau et al . , 1999 ) , synaptic KARs are absent from excitatory synapses ( Bureau et al . , 1999; Castillo et al . , 1997; Granger et al . , 2013 ) . This absence is not due to the lack of either Neto1 or Neto2 because expression of these proteins did not lead to the appearance of synaptic KARs . Thus either the level of KAR expression is insufficient for synaptic targeting or some other critical protein is missing from CA1 pyramidal cells . Expression of GluK1 also failed to generate synaptic KAR currents , although it was expressed on the surface , albeit at low levels . With either Neto1 or Neto2 , GluK1 generated large synaptic currents . This was accompanied with large expression of receptors on the cell surface . There are two possibilities to account for the presence of KARs at synapses . First , the density of the receptors on the surface could be so high that they simply flood the synapse , without any specific targeting signal . Second , their presence at synapses requires a separate targeting mechanism . We believe the latter is the case . We found that critical mutants of Neto1 and Neto2 in their C-terminal domains , which severely limited the synaptic accumulation of GluK1 receptors , had little or no effect on GluK1 surface expression . Expression of GluK1 and Neto caused a large increase in aEPSC frequency , but no change in aEPSC amplitude . Furthermore , the GluK1 antagonist ACET , blocked the increase in frequency but had no effect on aEPSCs amplitude . And comparing to the wild type Neto proteins , the critical mutants Neto1S3Y/A and Neto2S4T/A promote GluK1 surface trafficking to the same extent but both impair its synaptic expression . Moreover , the Neto2/GluK1 receptors are partially colocalized with presynpatic marker VGLUT1 . All these findings suggest that the GluK1 receptors are indeed localized at the synapse and the synaptic GluK1 responses are not due to the spread of glutamate from the synapse . Moreover , these results indicate that KARs and AMPARs do not co-localize at the same synapse , either by adding additional receptors to the synapse or by redistributing synaptic AMPARs . Instead , they appear to selectively populate previously silent synapses , i . e . synapses with NMDARs but no AMPARs . This model raises two sets of intriguing questions . First , what accounts for the fact that on average KAR synapses generate aEPSCs identical in size to AMPAR synapses ? This observation suggests a homeostatic process , although the synaptic expression of KARs occurs in the absence of synaptic activity . Second , we know that during LTP an individual synapse , which contains AMPARs before LTP , can accumulate additional AMPARs during LTP ( Harvey and Svoboda , 2007; Lee et al . , 2009; Matsuzaki et al . , 2004; Oliet et al . , 1996 ) . We also know that expressed GluK1 receptors at CA1 synapses on an AMPAR null background exhibit normal levels of LTP ( Granger et al . , 2013 ) . These findings raise a number of interesting questions . Does the genetic removal of AMPARs from the synapse now allow the synapse to accept KARs ? Although KARs are excluded from synapses that express AMPARs , can LTP drive KARs into AMPAR containing synapses ? Previous studies have reported that Neto1 is involved in synaptic NMDAR function although the findings were inconsistent . In one study it was found that Neto1 is critical for NMDAR subunit NR2A synaptic expression in CA1 neuron ( Ng et al . , 2009 ) , but another study showed that NR2B but not NR2A synaptic expression is increased in CA3 neurons of Neto1 knock-out mice ( Wyeth et al . , 2014 ) . However , here we found neither Neto1 nor Neto2 itself has any effect on NMDAR EPSC . Both GluK1/Neto1 and GluK1/Neto2 coexpression increased the size of the NMDAR EPSC , although this increase was much more modest than the EPSC observed at −70 mV . This could result from a modest synaptogenic effect . Although we did not observe an increase in spine density , the modest effects might be difficult to see with our imaging . Alternatively synapses could be added to the shaft and thus not visible in our spine density quantification . We sought to define the critical domain ( s ) of Neto1 and Neto2 required for the synaptic trafficking of GluK1 . For Neto1 the critical region is the last 20 amino acids . Except for the PDZ ligand domain , which is not required , there is no obvious homology to known protein-protein interaction domains . There are putative phosphorylation sites in this region and their mutation disrupts trafficking . For Neto2 the critical region was located in the middle of the C-terminal domain and could be narrowed down to a 12 amino acid stretch . Again there is no obvious homology of this region to known protein-protein binding motifs . There are putative phosphorylation sites in this region , which when mutated disrupt synaptic trafficking of KARs . It will be of interest in future studies to determine the potential roles of phosphorylation of Neto1 and Neto2 and the involved kinase ( s ) in GluK1 synaptic trafficking . In summary , this study has characterized the properties of Neto1 and Neto2 in controlling GluK1 receptor synaptic trafficking in hippocampal neurons . We have selected an excitatory synapse that normally does not express KARs , in order to determine the minimal requirements that govern the insertion of KARs into excitatory synapses . Our results demonstrate that Neto auxiliary proteins have two functionally distinct roles in the biology of the GluK1 type of KAR: First , they are essential for the delivery of receptors to the surface and for their targeting to the synapse . Second , they modify the gating kinetics of GluK1 . These properties are reminiscent of those of TARPs , which perform remarkably similar roles in the biology of AMPARs . It will be interesting to see how many of the properties we describe at CA1 synapses are held in common with excitatory synapses that normally express KARs , e . g . hippocampal mossy fiber synapses . The cDNAs of rat GluK1 ( gift from Dr . Stephen F . Heinemann ) , mouse Neto1 ( purchased from Open Biosystems ) and rat Neto2 ( gift from Dr . Susumu Tomita ) as well as the Neto1 and Neto2 mutants were subcloned into pCAGGS vector for biolistic transfection . Organotypic hippocampal slice cultures were made as previously described ( Schnell et al . , 2002 ) . Slices from P6-P8 rats were biolistically transfected with indicated plasmids together with FUGW-EGFP plasmid as a tracer on DIV 2 and then on DIV 8 dual whole-cell recordings in area CA1 were done by simultaneously recording responses from a fluorescent transfected neuron and neighboring untransfected control neuron . Pyramidal neurons were identified by morphology and location . Series resistance was monitored on-line , and recordings in which series increased to >30 MOhm or varied by >50% between neurons were discarded . Dual whole-cell recordings measuring evoked EPSCs used artificial cerebrospinal fluid ( ACSF ) bubbled with 95% O2/5% CO2 consisting of ( in mM ) 119 NaCl , 2 . 5 KCl , 4 CaCl2 , 4 MgSO4 , 1 NaH2PO4 , 26 . 2 NaHCO3 , 11 Glucose . 100 μM picrotoxin was added to block inhibitory currents and 4 μM 2-Chloroadenosine was used to control epileptiform activity . Intracellular solution contained ( in mM ) 135 CsMeSO3 , 8 NaCl , 10 HEPES , 0 . 3 EGTA , 5 QX314-Cl , 4 MgATP , 0 . 3 Na3GTP , 0 . 1 spermine . A bipolar stimulation electrode was placed in stratum radiatum , and responses were evoked at 0 . 2 Hz . Peak AMPAR and GluK1 currents were recorded at −70 mV , and NMDAR current amplitudes 100 ms following the stimulus were recorded at +40 mV . Paired-pulse ratio was determined by delivering two stimuli 40 ms apart and dividing the peak response to stimulus 2 by the peak response to stimulus 1 . All these data were analyzed off-line with custom software ( IGOR Pro , free download from following site: https://www . wavemetrics . com/order/order_igordownloads . htm ) . For Sr2+-evoked asynchronous EPSC recording , the ACSF was the same as above with the equimolar substitution of SrCl2 for CaCl2 . 100 μM picrotoxin was also included but without 2-Chloroadenosine . Stimulation was increased from 0 . 2 Hz to 2 Hz to optimize the frequency of Sr2+-evoked responses ( Oliet et al . , 1996 ) . Sr2+-evoked aEPSCs were analyzed off-line with custom IGOR PRO software , and in all cases at least 100 quantal events were used . For fast application , somatic out-side out patches were excised from wild type or transfected CA1 pyramidal neurons using 3–5 MΩ pipettes . The fast responses to glutamate were recorded at −70 mV . Glutamate pulses of 1 or 100 ms were applied to patches by a theta-glass pipette every 10–20 s using a piezoelectric controller ( Siskiyou ) ( Shi et al . , 2009 ) . Glutamate ( 10 mM ) was dissolved in the HEPES ACSF consisting of ( in mM ) NaCl 140 , KCl 5 , MgCl2 1 . 4 , CaCl2 1 , EGTA 5 , HEPES 10 , NaH2PO4 1 , D-glucose 10 , with pH adjusted to 7 . 4 , with the addition of 100 μM D-APV , 0 . 5 M tetrodotoxin and 100 μM GYKI53655 to isolate GluK-mediated currents . The control barrel contained the same HEPES ACSF with all the inhibitors and 1 mM sucrose but except glutamate . The open-tip response had a switch on and off time of less than 200 μs . Responses were collected with a Multiclamp 700A amplifier ( Axon Instruments ) , filtered at 2 kHz , and digitized at 10 kHz . Slice cultures were maintained and transfected as described above and on DIV 8 a transfected CA1 pyramidal neuron and a wild type one were patched simultaneously and filled with Alexa Fluor 568 dyes through the patch pipette for about 15–20 min . After filling , slices were fixed in 4% PFA/4% sucrose in PBS for 30 min at room temperature , followed by washing at least three times with PBS . Then slices were mounted and imaged by using super-resolution microscopy ( N-SIM Microscope System , Nikon ) . The experimental cells were identified by GFP fluoresces . Images along the stretch of CA1 pyramidal neuron primary apical dendrite from 100 μm to 200 μm from the cell body were acquired with a 100x oil objective in 3D-SIM mode using supplied SIM grating ( 3D EX V-R 100x/1 . 49 ) and processed and reconstructed using supplied software ( NIS-Elements , Nikon ) . Spine density analysis was performed manually on individual sections using ImageJ . For determining surface expression , an N-terminal HA tag was inserted after the signal peptide in GluK1 . DIV 10 rat hippocampal cultures were transfected with HA-GluK1 alone or together with Neto1 or Neto2 . At DIV 13 , surface GluK1 ( green ) was labeled with a rabbit HA antibody ( Abcam , Cat . No . ab9110 ) at room temperature for 10 min , followed by Alexa-488 secondary antibody ( Life technologies , A11034 ) . The images were captured as Z-stacks using a 63X oil immersion objective of LSM 510 Meta Zeiss confocal microscope . A projection image was created using different optical sections ( 0 . 35 μm ) and is presented . To determine changes in surface expression , the amount of surface GluK1 divided by the area of the ROI was calculated from 3 dendritic regions per neuron using Metamorph . The data presented is mean ± SEM from three independent experiments . Significance of evoked dual whole-cell recordings and aEPSC frequency compared to controls was determined using the two-tailed Wilcoxon signed-rank sum test . For all experiments involving un-paired data , including all outside-out patch data , a Mann-Whitney U-test with Bonferonni correction for multiple comparisons was used . Paired-pulse ratios and spine densities were analyzed with unpaired t test . Data analysis was carried out in Igor Pro ( Wavemetrics ) , Excel ( Microsoft ) , and GraphPad Prism ( GraphPad Software ) .	Information is transmitted in the brain by cells called neurons . To communicate with neighboring cells , neurons release chemicals called neurotransmitters across a structure called a synapse that forms a junction between the cells . The neurotransmitters bind to receptors on the surface of the receiving neuron , and depending on the type of neurotransmitter released , make that neuron either more or less likely to signal to its neighbors . Excitatory neurotransmitters make neurons more likely to signal , and glutamate is the most common excitatory neurotransmitter in the brain . There are several different types of receptor that can bind to glutamate , one of which – the kainate receptor – is found at relatively few synapses . These synapses include some in the hippocampus , a region of the brain that is important for memory . Researchers have recently identified two auxiliary proteins , called Neto1 and Neto2 , that interact with kainate receptors and appear to affect how strongly the kainate receptors respond when glutamate binds to them . However , the effect of the Neto proteins on one particular subunit of the kainate receptors – called GluK1 – had not been investigated in depth . CA1 pyramidal neurons are a group of neurons in the hippocampus that are able to produce kainate receptors , but these receptors are not found in CA1 pyramidal neuron synapses . Sheng et al . have now studied CA1 pyramidal neurons from rats , and found that these cells produce a limited amount of GluK1 on their surfaces . However , when GluK1 is expressed together with Neto1 or Neto2 , GluK1 receptors appear on the cell surface . Through an independent mechanism Neto proteins also promote the targeting of surface GluK1 to the synapse . Unexpectedly , GluK1 was excluded from synapses that contain another type of glutamate receptor called AMPA receptors . By measuring the effect of Neto1 and Neto2 on the behavior of GluK1 , Sheng et al . found that these proteins modified how the receptor responded to prolonged exposure to glutamate . Specifically , Neto1 increased how quickly GluK1 became desensitized to glutamate , while Neto2 decreased the rate of desensitization . This study demonstrates that Neto proteins play critical roles in controlling the location and biophysical properties of kainate receptors . It will be of interest to see how the present findings apply to other excitatory synapses in the brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Neto auxiliary proteins control both the trafficking and biophysical properties of the kainate receptor GluK1
Coupling of endoplasmic reticulum ( ER ) stress to dimerisation-dependent activation of the UPR transducer IRE1 is incompletely understood . Whilst the luminal co-chaperone ERdj4 promotes a complex between the Hsp70 BiP and IRE1’s stress-sensing luminal domain ( IRE1LD ) that favours the latter’s monomeric inactive state and loss of ERdj4 de-represses IRE1 , evidence linking these cellular and in vitro observations is presently lacking . We report that enforced loading of endogenous BiP onto endogenous IRE1α repressed UPR signalling in CHO cells and deletions in the IRE1α locus that de-repressed the UPR in cells , encode flexible regions of IRE1LD that mediated BiP-induced monomerisation in vitro . Changes in the hydrogen exchange mass spectrometry profile of IRE1LD induced by ERdj4 and BiP confirmed monomerisation and were consistent with active destabilisation of the IRE1LD dimer . Together , these observations support a competition model whereby waning ER stress passively partitions ERdj4 and BiP to IRE1LD to initiate active repression of UPR signalling . In eukaryotes , the endoplasmic reticulum ( ER ) is the central organelle for the synthesis of proteins destined for secretion and membrane insertion . The ER lumen harbours a specialised protein folding and processing machinery that constitutes the protein folding capacity of the ER . To ensure that the environment for productive protein maturation is maintained , both folding capacity and the inward flux of newly synthesised proteins are regulated by a pervasive negative feedback signalling pathway , the unfolded protein response ( UPR ) ( Kozutsumi et al . , 1988; Cox et al . , 1997 ) . In mammalian cells , this pathway involves three known signaling branches each directed by a unique signal transducer resident in the ER membrane , IRE1 , PERK and ATF6 . An imbalance between folding load and capacity ( ER stress ) activates these sensors initiating a rectifying transcriptional and translational response to defend protein-folding homeostasis in the compartment ( reviewed in Walter and Ron , 2011 ) . While details of downstream events and their physiological significance are relatively well characterised ( reviewed in Wang and Kaufman , 2016 ) , the molecular mechanisms of the earliest events in UPR activation remain incompletely understood . IRE1 , conserved in all eukaryotes and therefore the best-studied UPR transducer ( Cox et al . , 1993; Mori et al . , 1993 ) , detects ER stress via its luminal domain ( IRE1LD ) , initiating dimerisation-dependent autophosphorylation of its cytosolic domain ( Shamu and Walter , 1996 ) . The subsequent allosteric activation of the cytosolic endoribonuclease domain ( Lee et al . , 2008 ) leads to unconventional splicing of the mRNA encoding the XBP1/HAC1 transcription factor ( Cox and Walter , 1996; Yoshida et al . , 2001; Calfon et al . , 2002 ) , thereby promoting translation of an effector that drives a conserved gene-expression program . Two models have been put forth to describe how IRE1LD senses ER stress . A direct binding model posits that unfolded proteins act as ligands stabilising IRE1’s dimeric/oligomeric state thereby promoting its activation . This model is supported by the crystal structure of the core luminal domain from S . cerevisae IRE1 , showing an IRE1LD dimer interface traversed by a groove with architectural similarity to the major histocompatibility peptide-binding complexes ( MHCs ) ( Credle et al . , 2005 ) . Peptide ligands of the yeast IRE1LD have been identified and their addition to dilute solutions of yeast IRE1LD enhances the population of higher order species , although a clear shift from monomers to dimers was not readily observable ( Gardner and Walter , 2011 ) . The luminal domain of the broadly expressed alpha isoform of human IRE1 ( hIRE1αLD ) also crystallises as a dimer , with an overall architecture similar to the yeast protein , however , barring conformational changes , the MHC-like groove is too narrow to accommodate a peptide ( Zhou et al . , 2006 ) . Recently , peptides have been identified that bind hIRE1LD and affect its oligomeric state , as assessed by analytical ultracentrifugation ( AUC ) . Moreover , nuclear magnetic resonance ( NMR ) reported on peptide-induced structural rearrangements within the hIRE1αLD that also affected residues near the MHC-like groove . Hence , it has been proposed that the structure of Zhou et al . ( 2006 ) represents a ‘closed’ conformation of the peptide-binding groove that can shift towards an ‘open’ state to allow peptide binding ( Karagöz et al . , 2017 ) . However , a co-crystal structure of the ligand-bound yeast or human IRE1LD is not available and it remains unclear if and how peptide ligands affect hIRE1LD dimerisation , the first crucial step of its activation . An alternative hypothesis posits that IRE1 is repressed by interacting with a major component of the ER folding machinery , the heat-shock protein ( Hsp70 ) chaperone BiP . It is proposed that upon stress , unfolded proteins accumulate and compete for BiP interaction , thereby kinetically disrupting the inhibitory IRE1-BiP complex . This chaperone inhibition model draws parallels between the regulation of the UPR and its cytosolic counterpart , the heat-shock response , in which chaperones associate with the transcription factor Hsf1 , in eukaryotes , and σ32 , in bacteria , to interfere with their activity ( Abravaya et al . , 1992; Shi et al . , 1998; Tomoyasu et al . , 1998 ) . This model is supported by an inverse correlation between ER stress-induced IRE1 activity and the amount of ER-localised BiP recovered in complex with it ( Bertolotti et al . , 2000; Okamura et al . , 2000; Oikawa et al . , 2009 ) . Further molecular insight into the chaperone inhibition mechanism was gained recently by the discovery of ERdj4 as an ER-localised J-domain protein that selectively represses IRE1 activity in vivo and loads BiP onto the IRE1LD , thereby promoting monomerisation in vitro ( Amin-Wetzel et al . , 2017 ) . Whilst other modes of BiP binding to the IRE1LD have been proposed ( Carrara et al . , 2015; Kopp et al . , 2018 ) the aforementioned observations suggest a mechanism in which BiP engages the IRE1LD as an Hsp70 substrate: ATP-bound BiP initially interacts with the IRE1LD with high kon and high koff rates and only captures IRE1LD as a substrate ( in the ADP bound state , with low koff rates ) after ERdj4 co-chaperone-instructed ATP hydrolysis . This model draws on the conventional view whereby J-domain proteins act as adaptors that enable efficient substrate recognition via their divergent targeting domains and subsequent binding of Hsp70s , promoted by their conserved J-domain that stimulates Hsp70’s ATPase activity ( reviewed in Kampinga and Craig , 2010 ) . J-domain co-chaperones act in concert with nucleotide exchange factors ( NEFs , reviewed in Behnke et al . , 2015 ) to accelerate Hsp70s’ cycles of substrate binding and release , resulting in substrate-selective ultra-affinity ( Misselwitz et al . , 1998; De Los Rios and Barducci , 2014 ) , which is the basis for the assembly of Hsp70-substrate complexes . Whilst ERdj4’s repressive action on IRE1 signalling in cells and its ability to promote a complex between IRE1LD and BiP that favours the former’s monomeric state in vitro fit the chaperone inhibition model , they remain correlative findings and may be causally unrelated . For example , it is possible that ERdj4’s repressive action in cells arises from its role in eliminating IRE1LD activating ligands and not from catalysing the repressed , monomeric IRE1LD-BiP complex observed in vitro . Here , in support of the chaperone inhibition model , we report that enforced targeting of endogenous BiP to endogenously-expressed IRE1LD represses UPR signalling in cells , thereby establishing that BiP can directly repress IRE1 in vivo and that features of the IRE1LD that specify its repression in cells also specify its ability to undergo actively-driven monomerisation by ERdj4 and BiP in vitro . An inverse correlation between ER stress-induced IRE1 activity and the amount of BiP recovered in complex with it has been previously observed ( Bertolotti et al . , 2000; Okamura et al . , 2000; Oikawa et al . , 2009 ) but a causal link between BiP binding and IRE1 activity status had never been conclusively established . To assess the effect of BiP binding on the activity of IRE1 in vivo , we modified the endogenous Ern1 locus to encode an ER targeted J-IRE1 fusion protein consisting of IRE1α’s endogenous signal peptide , an N-terminally fused J-domain ( derived from ERdj4 ) followed by the endogenous IRE1α coding sequence ( Figure 1—figure supplement 1 ) . The alpha isoform accounts for all measurable activity in CHO cells and is referred to as IRE1 hereafter . By employing this fusion protein , we expected to stimulate BiP’s ATPase activity in close proximity to the IRE1LD thereby promoting formation of an IRE1-BiP complex . As control , a point mutant ERdj4 J-domain was used that had the histidine of the highly conserved HPD motif replaced by glutamine ( JQPD ) compromising the stimulation of BiP‘s ATPase activity ( Wall et al . , 1994 ) . The glycine-phenylalanine-rich ( G/F ) region of ERdj4 was included as a flexible linker , to allow the J-domain to explore the entire surface of IRE1LD . We deemed that low level expression of endogenous IRE1 ( and hence J-IRE1 ) would minimise IRE1-independent effects of this chimeric J-domain protein on the ER folding environment , effects that could not be excluded as having contributed to the previously-noted repressive effect of ERdj4 over-expression on the UPR ( Amin-Wetzel et al . , 2017 ) . Using an Ern1 null cell line with a genomic deletion encompassing the IRE1LD-encoding exons 2–12 ( ΔIRE1 , previously described in Kono et al . , 2017 ) , we reconstituted the endogenous locus with either wild-type IRE1 , J-IRE1 or JQPD-IRE1 fusion . Additionally , the cell lines stably expressed XBP1s::Turquoise and CHOP::GFP reporters that are controlled by the IRE1 and PERK UPR branches , respectively . Flow cytometry analysis showed that reconstitution of the locus with wild-type IRE1 rescues the non-responsive XBP1s::Turquoise phenotype of the ΔIRE1 cells towards stress induced by tunicamycin ( Figure 1A ) . In comparison , cells expressing the J-IRE1 fusion showed low XBP1::Turquoise reporter levels , indicating repressed IRE1 activity , even under stress . Repression was dependent on the integrity of the J-domain as ΔIRE1 cells reconstituted with the mutant JQPD-IRE1 acquired nearly wild-type stress responsiveness . The J-IRE1 protein was not otherwise compromised , as it was still able to respond to the ER stressor SubA , a protease that inactivates BiP by cleaving its interdomain linker ( Paton et al . , 2006 ) ( Figure 1B ) . These findings are consistent with BiP serving as a direct trans-acting factor to specify repression mediated by a J-domain presented in cis to the IRE1LD . A role for the cis-active J-domain in recruiting BiP to the IRE1LD is supported by immunoprecipitation ( IP ) of endogenous IRE1 prepared from the cells described above . More BiP was recovered in complex with the J-IRE1 chimera compared to the wild-type IRE1 whilst the mutant JQPD-IRE1 fusion associated with a similar amount of BiP as the wild-type ( Figure 1C ) , which is in accordance to their similar phenotype detected by flow cytometry . To further validate these in vivo observations , we reconstituted the system in vitro using recombinant proteins purified from bacteria . First , pull down of either C-terminally biotinylated IRE1LD ( IRE1LD-bio ) or J-IRE1LD ( J-IRE1LD-bio ) was performed . We assessed the formation of a BiP-IRE1LD-bio complex on SDS-PAGE after recovery on immobilised streptavidin ( Figure 1D ) . Whilst BiP recovery in complex with IRE1LD-bio was dependent on the presence of both ERdj4 and ATP in the binding assay , complex formation of BiP and J-IRE1LD-bio required only ATP . Next , we tested how BiP binding affected J-IRE1LD’s oligomeric status in vitro using a Förster resonance energy transfer ( FRET ) -based assay to continuously monitor the monomer-dimer equilibrium ( as described previously , Amin-Wetzel et al . , 2017 ) . A donor IRE1LD labelled with Oregon Green ( OG ) was pre-equilibrated either with an IRE1LD or J-IRE1LD acceptor molecule labelled with TAMRA ( TMR ) . As previously observed , BiP , ERdj4 , and ATP were all required to monomerise the IRE1LD homodimer as reflected in the time-dependent increase in donor fluorescence until a kinetically maintained pseudo steady state was reached ( Figure 1E ) . In contrast , heterodimeric FRET pairs containing the J-IRE1LD fusion and IRE1LD were monomerised by BiP in an ATP-dependent manner , but did not require ERdj4 in trans . The nucleotide-dependent , BiP-induced monomerisation of the J-IRE1LD containing heterodimer occurred with an approximately four-fold higher initial velocity and a higher plateau in the pseudo steady state of the reaction . Taken together these findings suggest that the fused J-domain enables efficient formation of the IRE1LD-BiP complex , thereby promoting monomerisation , which leads to repression of IRE1 activity . To examine the role of peptides in regulating the monomer-dimer equilibrium of IRE1LD and hence its activity , we turned to a 12-mer peptide ( MPZ-N ) derived from myelin protein zero . MPZ-N is the best studied ligand for mammalian IRE1LD and was recently proposed to directly interact with the peptide-binding groove thereby influencing IRE1LD‘s oligomeric status ( Karagöz et al . , 2017 ) . When introduced into the FRET-based assay , MPZ-N had no measurable effect on donor fluorescence . However , as the optical readout of this assay is sensitive mostly to monomerisation ( as reflected in an increase in donor fluorescence , Figure 1E ) it would be a relatively insensitive measure of MPZ-N peptide driven dimerisation . Therefore , we sought different assays to report on the ability of the MPZ-N peptide to promote IRE1LD dimers . The distribution of IRE1LD between monomers and dimers can be tracked by size exclusion chromatography ( SEC ) , as evidenced by the concentration-dependence of the peak elution time of IRE1LD and two dimerisation-compromised mutants: a previously characterised W125A variant ( Zhou et al . , 2006 ) and a new , more severe P108A variant ( Figure 2—figure supplement 1A ) . Both mutations are predicted to decrease hydrophobic interactions across the dimer interface ( Figure 2—figure supplement 1B ) . Addition of MPZ-N peptide ( at concentrations exceeding the reported K1/2 max for binding of 16 µM , Karagöz et al . , 2017 ) did not affect the peak elution time of IRE1LD , itself introduced into the assay at 500 nM , near the Kd for IRE1LD dimerisation ( Zhou et al . , 2006 ) ( Figure 2A and B ) . To confirm these observations , we made use of an alternative assay reporting on IRE1LD’s dimerisation status . To this end , we employed a modified IRE1LD Q105C that forms a disulphide across the dimer interface , creating a covalently stabilised dimer when placed in oxidising conditions ( Figure 2—figure supplement 1C ) and the aforementioned dimerisation-compromised versions of the IRE1LD ( W125A and P108A ) . Differential scanning fluorimetry ( DSF ) revealed that the melting temperature ( Tm ) of disulphide-linked IRE1LD Q105C SS was ~10 °C higher than the Tm the wild-type protein , a Tm difference that was effaced by reduction of the dimer-stabilising disulphide ( Figure 2—figure supplement 1D ) . By contrast , the IRE1LD monomeric variants exhibited a Tm 5–10 °C lower than the wild-type . These observations established a correlation between the monomer-dimer equilibrium and the Tm of the protein consistent with dimerisation-mediated stabilisation of the IRE1LD . A ligand , stabilising the IRE1LD dimer , is predicted to increase the Tm , however , addition of MPZ-N peptide had no effect on the Tm of IRE1LD ( Figure 2—figure supplement 1D , the significance of the lowering of Tm observed at the highest concentrations of peptide remains to be determined ) . To gain insight into the mode of MPZ-N binding to IRE1LD we made further use of the disulphide-linked IRE1LD Q105C SS . The crystallised IRE1LD Q105C SS dimer proved identical in structure to the wild-type protein ( root-mean squared deviation ( RMSD ) of 0 . 46 Å over 227 Cα atoms ) except for the presence of a conspicuous density corresponding to a C105-C105 trans-protomer disulphide , thereby locking the proposed binding groove in the ‘closed’ conformation ( Figure 2C , Figure 2—figure supplement 1E and Table 1 ) . Nonetheless , a fluorescence polarisation binding assay , using FAM-labelled MPZ-N , showed that binding to the IRE1LD was not compromised by the disulphide ( Figure 2D ) , leading us to conclude that MPZ-N does not obligatorily bind within the proposed MHC-like groove of the IRE1LD . This conclusion is also consistent with the paramagnetic relaxation enhancement ( PRE ) experiments with IRE1LD and an MPZ-proxyl-labelled peptide ( Karagöz et al . , 2017 ) , which present a distance constraint of 10 Å between Ile186 of the IRE1LD and the labelled Cys5 of the peptide . Figure 2—figure supplement 2 shows that the extended peptide is free to explore the entire surface of one face of the IRE1LD and may therefore bind in locations other than the MHC-like groove , without violating this distance constraint . Given the evidence for BiP’s role in IRE1 repression , we tried to identify regions in IRE1LD that might be important for such regulation . BiP , as an Hsp70 chaperone , typically interacts with unfolded or flexible regions in its client proteins ( Rüdiger et al . , 1997 ) and we held that this might also be the case for its interaction with the IRE1LD . Therefore , we sought clues to map these flexible regions by collecting data on the structural dynamics of IRE1LD in solution as evaluated by hydrogen-1H/2H-exchange experiments in combination with mass spectrometry ( HX-MS ) . IRE1LD was pre-equilibrated for 30 min at 30°C followed by an exchange reaction in deuterium oxide ( D2O ) buffer for 30 and 300 s . Subsequent analysis of deuteron incorporation was performed as described previously ( Hentze and Mayer , 2013 ) . Information on peptic peptides covering 85% of the IRE1LD sequence was obtained ( Table 2 ) . The extracted percentage of exchange ( %ex ) for each peptic peptide contained information about the thermodynamic stability of structural elements , the hydrogen bonding and solvent accessibility of backbone amide hydrogens ( Figure 3A left panel ) . Projection of these values onto the crystal structure showed that regions in the hydrophobic core exhibited significant protection from exchange ( low %ex ) , whereas surface exposed areas were more dynamic ( high %ex ) ( Figure 3A right panel ) . This method identified the region encompassing residues 303–378 as being especially flexible , a conclusion consistent with the observation that though it was present in the constructs used for crystallisation , residues 308–357 were resolved in neither the crystal structures of wild-type IRE1LD ( Zhou et al . , 2006 ) ( Figure 3A right panel dotted line ) nor the disulphide-linked IRE1LD Q105C SS variant here . Similar characteristics apply to residues 379–444 , covering the so-called tail region that connects the structured core of the IRE1LD with the transmembrane domain ( Figure 3A right panel dotted line and Figure 3—figure supplement 1A ) . Moreover , the latter residues overlap with a region of IRE1LD implicated in its basal repression in an overexpression cell-based assay ( Oikawa et al . , 2007; Oikawa et al . , 2009 ) . To probe the putative loop ( residues 308–357 ) and the tail region ( residues 390–444 ) for their importance in maintaining the repressed state of IRE1 in vivo , we devised a CRISPR-Cas9 mutagenesis strategy ( Figure 3B ) . By targeting only unstructured regions within IRE1LD , we hoped to preserve the integrity of the core structure whilst favouring mutations that might de-repress IRE1 activity . After introducing a set of guide RNAs targeting the region of interest together with the Cas9 endonuclease into cells , error prone non-homologous end joining ( NHEJ ) resulted in a series of mutations , ranging from small in-frame indels at a single guide site to larger ones spanning two guide sites . The IRE1 reporter was used to select rare clones exhibiting a de-repressed IRE1 phenotype ( XBP1s::Turquoise bright ) . The CHOP::GFP reporter was used to exclude clones exhibiting a general perturbation of ER protein homeostasis . Iterative rounds of fluorescence-activated cell sorting ( FACS ) enriched the XBP1s::Turquoise bright population . Clones that had acquired IRE1-independent XBP1s::Turquoise reporter expression were purged based on their unresponsiveness to the IRE1 inhibitor 4µ8c ( Cross et al . , 2012 ) . The Ern1 locus of individual clones with deregulated IRE1 activity found in the final pool was sequenced ( Figure 3C left panel ) . As expected , all the putative deregulating deletions/mutations maintained the frame of the IRE1 coding sequence ( Figure 3C right panel and Figure 3—figure supplement 1B ) . These observations suggested that deletions of unstructured regions of IRE1LD could deregulate IRE1 activity . To confirm the suggested role of deletions of the loop and the tail region of IRE1LD in deregulating its activity in cells , we reconstituted the endogenous Ern1 locus of the ∆IRE1 cell line with the most extensive IRE1 deletion variants identified above: the Δloop ( missing residues 313–338 ) , the Δtail ( missing residues 391–444 ) or both ( ΔΔ ) . The reconstituted alleles de-repressed IRE1 activity , as indicated by the elevated basal XBP1s::Turquoise signal ( Figure 4A and Figure 4—figure supplement 1A ) . The IRE1 ΔΔ double deletion had the strongest deregulated phenotype under basal conditions . Like the shorter deletions , the IRE1 ΔΔ double deletion nonetheless retained some responsiveness to stress , albeit with a narrowed dynamic range ( Figure 4A , compare untreated to tunicamycin-treated samples ) . To establish if the deregulating deletion affected the association of the IRE1LD with BiP , we compared the amount of BiP that co-immunoprecipitated with the endogenously expressed wild-type or IRE1 ΔΔ ( Figure 4B ) . Despite variation in the total BiP signal intensity in the three independent repeats ( Figure 4B , lower panel ) , paired analysis revealed that significantly less BiP was associated with the IRE1 ΔΔ mutant . The same was observed in a transient transfection system in which IRE1’s cytosolic effector domains were replaced with glutathione S-transferase ( GST ) . Compared to the wild-type IRE1LD-GST bait , the amount of BiP recovered by glutathione affinity chromatography in association with the variants was significantly lower in context of the single deletions and even lower in case of the double-deletion IRE1LD ΔΔ-GST ( Figure 4C ) . Together , the observations described above confirm a role for the flexible regions of the IRE1LD in maintaining IRE1 in a repressed state in vivo and suggest that such repression may reflect a role for these flexible regions in specifying BiP binding . To follow up on this suggestion , Bio-Layer Interferometry ( BLI ) was used to compare BiP’s association with the biotinylated wild-type or double-deleted IRE1LD ∆∆ immobilised on the sensor . Immersing the sensor into a solution containing ERdj4 , BiP and ATP gave rise to an association curve , that was reproducibly attenuated when IRE1LD ΔΔ was bound as a ligand compared to the wild-type IRE1LD ( Figure 5—figure supplement 1A , left traces ) . A similar qualitative defect in BiP binding to IRE1LD ΔΔ was also observed when the full-length ERdj4 was replaced by its isolated J-domain ( that lacks the regions required for specific targeting to the IRE1LD ) ( Figure 5—figure supplement 1A , right traces ) . This last observation suggested that the defect in J-domain-mediated BiP binding to IRE1LD ΔΔ , had a component that was independent of recruitment of ERdj4 to IRE1LD by the former’s targeting domain and implied that the deleted region of IRE1LD had a role in specifying BiP association . This was explored further using J-IRE1LD fusion proteins as BLI ligands to enforce ATP hydrolysis by BiP in proximity to the wild-type or double-deleted IRE1LD ( independent of the role these flexible regions of IRE1LD might have in J-domain co-chaperone recruitment ) . Immersing a BLI sensor loaded either with J-IRE1LD or J-IRE1LD ΔΔ into a solution containing BiP and ATP revealed a reproducible defect of BiP association to J-IRE1LD ΔΔ ( Figure 5A left panel ) . No association was observed in presence of the substrate binding-deficient BiPV461F mutant . The dissociation in presence of ATP remained similar for both wild-type and J-IRE1LD ΔΔ ( Figure 5A right panel ) , as expected of a process limited by BiP’s rate of nucleotide exchange . The measurements above report on BiP’s interaction with the IRE1LD in the context of J-domain-mediated , ATP hydrolysis-driven ultra-affinity ( Misselwitz et al . , 1998; De Los Rios and Barducci , 2014 ) . To examine the role of IRE1LD’s flexible regions in its affinity for BiP-ADP ( an interaction that reports on a segment of the ultra-affinity cycle ) we combined BiP with C-terminally biotinylated IRE1LD ( either IRE1LD-bio or IRE1LD ∆∆-bio ) in presence of ADP and absence of J-domain protein . Given the slow association of BiP-ADP with substrates and the slow dissociation of BiP oligomers a lengthy equilibration ( 16 hr ) was allowed . Almost three-fold less BiP was recovered in complex with IRE1LD ∆∆-bio than with IRE1LD-bio ( Figure 5B ) . BiP association was concentration-dependent , destabilised by ATP and was not observed with BiPV461F . Coupling of BiP’s two domains was dispensable for this interaction with IRE1LD-bio , as it was also observed with the domain-uncoupled BiPADDA ( Preissler et al . , 2015a ) . Together , these observations point to a role for the flexible regions of IRE1LD in specifying BiP association as a conventional substrate of this Hsp70 . An additional role for the flexible regions in ERdj4 recruitment was not evident within the sensitivity of the tools available to us , and therefore remains unexcluded . BiP binding in vitro promotes dissociation of the IRE1LD dimer ( Amin-Wetzel et al . , 2017 and Figure 1E ) . Therefore , we employed the same FRET-based assay to determine if impaired BiP binding affected monomerisation of IRE1LD ΔΔ -containing dimers . Wild-type fluorescent donor-labelled IRE1LD was allowed to dimerise with acceptor-labelled IRE1LD or IRE1LD ΔΔ and the rate at which BiP , ERdj4 and ATP promoted dissociation of these dimers was measured by following the increase in donor fluorescence over time . The initial velocity of BiP-mediated monomerisation of the IRE1LD ΔΔ containing heterodimers was considerably slower than monomerisation of wild-type homodimers ( Figure 5C ) . In BLI experiments , BiP association to monomeric IRE1LD P108A was faster than to the enforced dimeric IRE1LD Q105C SS ( Figure 5—figure supplement 1B ) , raising the concern that both diminished BiP binding to the IRE1LD ΔΔ observed in BLI ( Figure 5A ) and the slower monomerisation of the IRE1LD ΔΔ containing FRET pair ( Figure 5C ) might reflect intrinsically enhanced stability of the IRE1LD ΔΔ-containing dimers . However , SEC of the purified proteins performed over a range of protein concentrations reported on similar affinities of the wild-type and IRE1LD ΔΔ dimers ( Figure 5—figure supplement 1C and D ) yielding K1/2 max values in the same order of magnitude as the KD of dimerisation measured by AUC ( Zhou et al . , 2006 ) . Together , these observations suggest that diminished BiP binding to IRE1LD ΔΔ resulted in an impairment of BiP-driven IRE1LD monomerisation . To complement the kinetic observations pointing to impaired BiP-driven monomerisation of IRE1LD ΔΔ with structural correlations , HX-MS was performed . To establish the HX-MS signature of monomerisation , deuteron incorporation was compared between wild-type and dimerisation-defective IRE1LD W125A or IRE1LD P108A mutants . This reported on monomerisation-induced deprotection of several peptic peptides from the IRE1LD ( Figure 6A ) . Projecting these areas onto the crystal structure revealed that monomerisation affected HX at the dimer interface but also in parts further away ( Figure 6B ) . IRE1LD monomerisation thus induced structural rearrangements across the protein . Moreover , the difference plot in HX reported on a gradation between both mutant variants , as IRE1LD P108A had an enhanced signature of monomerisation compared to IRE1LD W125A , matching the hierarchy of dimer instability observed by SEC and DSF analysis ( Figure 2—figure supplement 1A and D , respectively ) . A similar deprotection signature , affecting most of the peptic peptides that are exposed upon monomerisation , was observed when IRE1LD was incubated with BiP and ERdj4 in presence of ATP ( Figure 6C upper row , box 1: residues 77–128 and box 2: residues 280–302 ) . Monomerisation was dependent on the integrity of all components of the reaction , as neither the substrate binding BiPV461F mutant nor the ERdj4QPD supported the pattern of deprotection observed in the monomeric versions of IRE1LD ( the significance of the protection afforded by BiPV461F and ERdj4QPD to some peptides is presently unknown ) . IRE1LD ΔΔ exhibited delayed monomerisation in presence of BiP , ERdj4 and ATP: IRE1LD ΔΔ’s signature of monomerisation was absent after 30 s incubation in D2O ( Figure 6C lower row ) and was faint even after an exchange reaction of 300 s ( Figure 6—figure supplement 1B and C lower row ) . In the absence of BiP , ERdj4 and ATP the difference plot comparing deuteron incorporation into IRE1LD and IRE1LD ΔΔ was negligible ( Figure 6—figure supplement 1D ) providing independent confirmation of the SEC measurements pointing to similar stability of the wild-type and IRE1LD ΔΔ mutant dimer ( Figure 5—figure supplement 1C and D ) . Thus , HX-MS provided an orthogonal assay to the FRET-based measurement , reporting on BiP-mediated monomerisation of IRE1LD and a kinetic defect in this process brought about by deletion of flexible regions in the luminal domain that enforce IRE1’s repressed state in cells . Close inspection of the HX-MS data revealed that some of the peptides ( e . g . peptides 636 . 3802+ and 655 . 273+ corresponding to residues 96–106 and 297–302 , respectively ) exhibited clear bimodal isotope distribution . This characteristic is a signature for the EX1 exchange regime , indicative of the presence of two discrete subpopulations of molecules: a more folded and therefore low exchanging subpopulation and a more open , high exchanging subpopulation ( Figure 7—figure supplement 1A and Materials and methods section ) . The contribution of low and high exchanging subpopulations to each isotope peak was determined by fitting the isotope peak maxima versus m/z data points ( Figure 7A ) to a two Gaussian distribution model ( Hentze et al . , 2016 ) . From the fit parameters the fraction of each isotope peak that belongs to the low and high exchanging subpopulation was calculated [Figure 7—figure supplement 1A ( blue and red parts of the bars ) and 1B] . Comparison with the unexchanged and the 100% control samples revealed that the low exchanging subpopulation was largely protected from HX , whereas the high exchanging subpopulation had almost all amide protons exchanged for deuterons . Moreover , the low exchanging subpopulation converted into the high exchanging subpopulation with time ( Figure 7—figure supplement 1A , compare 30 and 300 s incubation in D2O ) . Interestingly , the degree of conversion into the high exchanging subpopulation was more pronounced for the IRE1LD P108A monomeric mutant than for wild-type IRE1LD and essentially complete after 300 s ( Figure 7—figure supplement 1A left panel ) . SEC analysis of IRE1LD P108A showed that at 5 µM ( the concentration at which the protein was diluted into D2O ) it is mostly monomeric ( Figure 2—figure supplement 1A ) . Hence , these data suggest that the conversion from the low exchanging subpopulation to the high exchanging subpopulation was a feature of the monomeric state . HX is a quasi-irreversible reaction: Once a molecule has transiently assumed a high exchanging conformation ( and undergone the exchange ) the signature of having transited through a high exchanging conformation remains even if the protein is in a conformational equilibrium ( and individual molecules transit back to the low exchanging conformation ) . Thus , HX-MS detects the transition to the high exchange endpoint . The observation that for wild-type IRE1LD the transition from the low ( blue ) to the high ( red ) exchanging population occurred with much slower kinetics than for IRE1LD P108A ( Figure 7—figure supplement 1A , compare left panel , monomeric IRE1LD P108A with the right panel , wild-type IRE1LD ) suggests that a higher proportion of IRE1LD monomers increased the transition rate , whereas the presence of IRE1LD dimers leads to a reduction of the rate constant . Hence , the extracted transition rate ktrans reports on IRE1LD’s monomer-dimer equilibrium during the reaction . Next , we compared the ktrans of peptic peptide 655 . 273+ from wild-type IRE1LD in presence and absence of BiP , ERdj4 and ATP . Due to pre-incubation of the reactions , the three-protein system already had a higher proportion of monomeric IRE1LD at the point of dilution into D2O ( reflected in a greater proportion of the high mass population at the earliest measurement ) . Nevertheless , an accelerated time-dependent increase in the proportion of monomeric IRE1LD was observed in the BiP-treated sample , indicating an increase in ktrans ( Figure 7A and B ) . Acceleration of ktrans was also observed with peptide 636 . 3802+ in presence of BiP , ERdj4 and ATP ( Figure 7C and Figure 7—figure supplement 1C ) . Because it is affected by peptide-specific flexibility , ktrans itself is not a direct measure of the first order dissociation rate of the IRE1LD dimer ( its koff ) , however , the difference observed in ktrans for any individual peptide measured under two conditions mainly reports on differences in IRE1LD dimer dissociation . Therefore , these findings imply that BiP-induced IRE1LD monomerisation has a component arising from active destabilisation of the dimer . The notion that a chaperone machinery with an Hsp70 , such as BiP , as its terminal effector might negatively regulate activity of an upstream UPR transducer , such as IRE1 , has the appeal of simplicity: Hsp70’s can potently affect the structure and function of their clients . The level of free BiP is kept low by inactivating oligomerisation and AMPylation and is further limited by client titration ( Preissler and Ron , 2019 ) . Therefore , the availability of a BiP-dependent machinery to serve as an active repressor of IRE1 is a plausible inverse measure of the level of ER stress . For years , the inverse relationship between the recovery of BiP in complex with IRE1 and exposure of cells to conditions causing ER stress has provided the only experimental support for this chaperone repression model ( Bertolotti et al . , 2000; Okamura et al . , 2000; Oikawa et al . , 2009 ) . The recent establishment of an ATP- and co-chaperone-dependent system in which BiP promotes a pool of monomeric , inactive-state IRE1LD further supports the model by revealing BiP’s potential to affect a major change in IRE1’s activity in vitro ( Amin-Wetzel et al . , 2017 ) . Here , we provide much needed further support for the chaperone repression model by demonstrating that directing endogenous BiP to bind endogenous IRE1LD as a substrate also attenuates signalling in cells , thus revealing BiP’s potential as a direct IRE1 repressor in vivo . A structure-based targeted approach identified regions of IRE1LD that impart a repressed state in vivo . The same regions proved important for ATP and co-chaperone-dependent BiP-mediated conversion of active-state IRE1LD dimers to inactive-state monomers in vitro and their presence accelerated the formation of an ATP and co-chaperone-dependent complex with BiP in vitro . Monomerisation was observed in both a FRET-based assay , involving labelled molecules of IRE1LD , and in an HX-MS assay with intact molecules , thus establishing a firm correlation between the determinants of IRE1 that regulate its function in vivo and those that specify its regulation in vitro by a BiP-led machinery ( Figure 8 ) . Correlation between factors involved in BiP regulation of IRE1 in vitro and UPR activity in vivo have been previously noted: Deregulated AMPylation of BiP activates IRE1 in cells ( Preissler et al . , 2015b ) and BiP AMPylation in vitro blocks IRE1LD monomerisation ( Amin-Wetzel et al . , 2017 ) . ERdj4 acts in concert with BiP to monomerise IRE1LDin vitro and loss of ERdj4 from cells de-represses IRE1 in vivo ( Amin-Wetzel et al . , 2017 ) . However , genetic lesions in trans-acting ER-localised factors also have the potential to broadly alter the state of the ER and thereby unleash processes that affect IRE1 independently ( of any direct interaction with BiP ) . Indirect effects are less likely a consequence when IRE1LD is modified in cis . Therefore , whilst it is impossible to rule out contributions from factors other than the BiP machinery to the deregulation of IRE1 that arises from deletion of the unstructured regions of its luminal domain , attenuation of BiP-mediated IRE1 repression in cells emerges as a parsimonious unifying explanation for the findings presented here . It is further notable that there is nothing in our observations to speak against the possibility that extended regions of unfolded ER proteins serve as activating ligands of IRE1 by binding across the IRE1LD dimer interface and stabilising it ( Karagöz et al . , 2019 ) . IRE1 signalling is triggered by an imbalance between unfolded proteins and BiP . The latter results in more potential ligands for IRE1 and fewer molecules of its client-free ATP bound BiP repressor ( Bakunts et al . , 2017; Vitale et al . , 2019 ) . Thus , the two proposed mechanisms for IRE1 activation , could well co-exist . However , our findings do raise questions regarding the strength of the experimental evidence supporting the current ideas how unfolded proteins may serve as activating ligands of IRE1 . The evidence rests prominently on the activity of a peptide , MPZ-N , nominated as a model activating ligand of IRE1LD ( Karagöz et al . , 2017 ) . Our findings do not support the notion that this peptide specifically engages the MHC-like groove traversing the dimer interface , as a disulphide , crystallographically proven to lie across this groove , thereby locking the helices in a ‘closed’ conformation , had no effect on the binding of MPZ-N to IRE1LD . Furthermore , MPZ-N binding to IRE1LD did not stabilise it thermodynamically , whereas the aforementioned disulphide , which mimics the proposed dimer-stabilising effect of a bound peptide , increased the melting temperature IRE1LD by 10 °C . Nor did MPZ-N promote a shift in the monomer-dimer equilibrium of IRE1LD as assessed by SEC . These concerns , along with the lack of crystallographic data supporting engagement of the groove by ligands , suggest the need for further experiments to test the role of unfolded proteins as direct IRE1 activators . HX-MS analysis revealed neither an ERdj4-dependent nor BiP and ATP-dependent protection within IRE1LD to suggest their binding site . It has been proposed that ERdj4’s bacterial homolog , DnaJ , exploits mostly side chain interactions to bind clients , with a strong preference for aromatic residues ( Rüdiger et al . , 2001 ) . Such interactions , were they to serve as the basis for IRE1LD recognition by ERdj4 , would be only visible to HX-MS if they stabilised the underlying secondary structure . Given the conventional mode of BiP action on IRE1LD ( ATP and co-chaperone-dependent and abolished by the BiPV461F mutation ) , one would expect protection of 4–5 hydrogen amides by IRE1LD engagement in the chaperone’s substrate-binding domain . However , partial occupancy of multiple sites may have diluted any HX-MS signature of BiP binding . This is supported by the observation that at the concentrations of ERdj4 and BiP used in the HX-MS assay , the FRET assay reported peak fluorescence of only ~40% of the unquenched donor ( at the kinetically driven pseudo steady state plateau of the reaction , Figure 1E ) . Thus , the lack of a clear ATP- and ERdj4-dependent BiP binding signature is consistent with the dynamic nature of BiP’s interaction with IRE1LD . Interestingly , we observed an ATP-independent protection against deuteron incorporation within IRE1LD that was also evident in presence of the substrate binding-defective BiPV461F . This might reflect a non-conventional interaction of BiP’s nucleotide binding domain with IRE1LD , as proposed by the Ali lab ( Kopp et al . , 2018; Kopp et al . , 2019 ) . However , this protection is not correlated to the activity-state of IRE1LD and its significance thus remains to be established . Mechanistically , BiP’s interaction with IRE1LD shares features with other situations in which Hsp70s bind to native clients thereby regulating their activity: DnaJ-directed , DnaK-mediated destabilisation of E . coli σ32 ( Rodriguez et al . , 2008 ) , functional regulation of the glucocorticoid receptor ( Kirschke et al . , 2014 ) , regulation of the activity of the tumour suppressor p53 ( Boysen et al . , 2019; Dahiya et al . , 2019 ) , Hsf1 regulated heat shock gene expression ( Abravaya et al . , 1992 ) and Hsc70-mediated destabilisation of clathrin coats ( Sousa et al . , 2016 ) . All these have in common client destabilisation and likely initiate at unstructured regions of the substrate . Thus , it seems reasonable to suggest that an important aspect of BiP’s ability to affect the disposition of IRE1LD’s monomer-dimer equilibrium arises from its interaction with the flexible regions identified here . Bimodal analysis of the HX-MS data suggested that ERdj4-directed BiP binding can accelerate dimer disassembly . This is consistent with the ability of the IRE1LD dimer to serve as a ligand for ERdj4 and BiP ( here and Amin-Wetzel et al . , 2017 ) . While BiP binding to and stabilisation of IRE1LD monomers may also contribute to shifting the monomer-dimer equilibrium towards the former , the HX-MS experiment suggests an ( additional ) active role for BiP in dimer destabilisation . This may arise from a BiP-binding induced bias of the ensemble of IRE1LD dimers towards conformers preferentially populated in the monomer . A similar mechanism of conformational selection has been proposed for DnaK-mediated destabilisation of E . coli σ32 ( Rodriguez et al . , 2008 ) . Such ‘allosteric’ action is consistent with the observation that monomerisation has effects on IRE1LD structure that are far removed from the dimer interface . Alternatively , BiP binding may destabilise the IRE1LD dimer by entropic pulling ( De Los Rios et al . , 2006 ) , as has been suggested in Hsc70-mediated destabilisation of clathrin coats ( Sousa et al . , 2016 ) ( Figure 8 ) . The latter mechanism would be further favoured by assembly of BiP oligomers on the surface of the IRE1LD , a possibility consistent with the >1:1 stoichiometry of BiP:IRE1LD complexes observed in some experiments ( Amin-Wetzel et al . , 2017 ) ( although the latter may also reflect multiple BiP binding sites ) . As shown here , flexible regions of IRE1LD contribute measurably to its repression in cells and to BiP-driven monomerisation in vitro . This observation is consistent with the idea that these regions serve as initiation points for BiP binding to promote dimer disassembly via entropic pulling , allosterically induced conformational changes or both . Considerable redundancy seems built into the process , as the deregulated IRE1∆∆ allele retained a measure of stress responsiveness in cells and the IRE1LD ∆∆ dimer was still slowly undone in a BiP-dependent process in vitro . Such redundancy has been observed previously: in both yeast and human , IRE1 deletion of the tail region connecting the structured core of IRE1LD to the transmembrane domain partially deregulated IRE1 , whilst retaining partial responsiveness to ER stress ( Oikawa et al . , 2007; Oikawa et al . , 2009 ) . Redundancy in the structural features of the IRE1LD dimer that render it a substrate for BiP-dependent disassembly and the non-equilibrium kinetic nature for BiP’s action could serve as the basis for a smoothly graded response to variation in the levels of ER stress . The parental strains for the CRISPR-Cas9-mediated homologous recombination approaches were the previously described ΔLD15 dual CHOP::GFP and XBP1s::Turquoise UPR reporter Chinese Hamster Ovary CHO-K1 cell lines ( Kono et al . , 2017 ) and have been authenticated as CHO-K1 using the criteria of successful targeting of essential genes using a species-specific CRISPR whole genome library , and sequencing of the wild-type or mutant alleles of the genes studied that confirmed the sequence reported for the corresponding genome . The cell lines have tested negative for mycoplasma contamination using a commercial kit ( MycoAlert ( TM ) Mycoplasma Detection Kit , Lonza ) . None of the cell lines is on the list of commonly misidentified cell lines maintained by the International Cell Line Authentication Committee . The CRISPR-Cas9-mediated mutagenesis strategy and the transient transfection of GST-tagged IRE1LD was performed with CHO-K1 S21 dual UPR reporter cells ( Sekine et al . , 2016 ) . Cells were cultured in Ham’s nutrient mixture F12 ( Sigma ) . All cell media was supplemented with 10% ( v/v ) serum ( FetalClone-2 , Hyclone ) , 2 mM L-glutamine ( Sigma ) , 100 U/ml penicillin and 100 μg/ml streptomycin ( Sigma ) . Cells were grown in tissue culture dishes or multi-well plates ( Corning ) at 37°C and 5% CO2 . Tunicamycin ( Melford ) treatment was at 2 . 5 μg/ml for 16 hr , 2-Deoxyglucose ( 2DG ) ( Sigma ) treatment at 4 mM for 16 hr and 4μ8c ( Cross et al . , 2012 ) treatment at 10 μM for 7 days . The drugs were mixed with pre-warmed culture medium and immediately added to the cells by medium exchange . Cells were transfected using Lipofectamine LTX ( Life Technologies ) transfection reagent with reduced serum medium Opti-MEM ( Life Technologies ) following the manufacturer’s instructions . To analyse the effect of IRE1 variants expressed from the endogenous Ern1 locus on the UPR ( Figure 1A and B , Figure 4—figure supplement 1A and B ) , flow cytometry was performed . Cells were washed once in PBS and collected in PBS containing 4 mM EDTA . Single-cell fluorescent signals ( 20 , 000/sample ) were analysed by dual-channel flow cytometry with an LSRFortessa cell analyser ( BD Biosciences ) . FACS was performed on either a Beckman Coulter MoFlo or a BD FACSMelody cell sorter . Cells were washed once in PBS and then incubated 5 min in PBS supplemented with 0 . 5% BSA and 4 mM EDTA before sorting into fresh media . CHOP::GFP fluorescence was detected with excitation laser at 488 nm , filter 530/30 nm; XBP1s::Turquoise fluorescence with excitation laser 405 nm , filter 450/50 nm and mCherry fluorescence with excitation laser 561 , filter 610/20 . To generate clonal cell lines stably expressing a version of IRE1 the transfected cells were treated with 2-Deoxyglucose to gate for cells showing high CHOP::GFP XBP1s::Turquoise fluorescence . Cas9 guides were either manually designed following standard guidelines ( Ran et al . , 2013 ) or taken from the CRISPy database ( URL: http://staff . biosustain . dtu . dk/laeb/crispy/ , ( Ronda et al . , 2014 ) . Cells were transfected with the Cas9 and guide constructs and grown for seven days before they were analysed by flow cytometry or FACS . For the in vivo mutagenesis strategy ( Figure 3B and C and Figure 3—figure supplement 1B ) , a series of guides that tiled the two regions of interest , set A covering the putative loop ( residues 308–362 ) and set B covering the tail ( residues 368–444 ) was designed . Set A and set B guide-Cas9 encoding plasmids were transfected singly or in different pairwise combinations into IRE1 wild-type expressing cells ( CHO-K1 S21 CHOP::GFP , XBP1s::Turquoise dual reporter cell line ) and pooled to create population 0 ( Figure 3C ) . Rare de-repressing IRE1 mutants were enriched from the mutagenised population by iterative rounds of FACS ( populations 1 and 2 ) followed by a selection against clones that had acquired IRE1-independent XBP1s::Turquoise reporter expression , as assessed by their unresponsiveness to the IRE1 inhibitor 4µ8c . Genomic DNA was extracted from final clones , PCR used to amplify the loci of interest and the resultant products were sequenced . The genomic DNA was extracted from cells by incubation in Proteinase K solution ( 100 mM Tris-HCl pH 8 . 5 , 5 mM EDTA , 200 mM NaCl , 0 . 25% SDS , 0 . 2 mg/ml Proteinase K ) overnight at 50 °C . Next , Proteinase K was heat inactivated at 98 °C for 20 min before the supernatant was collected and used as a template in PCR reactions before sequencing . To facilitate the interpretation of the sequencing data , the changes in size of alleles modified by Cas9 was determined by capillary electrophoresis on a 3730xl DNA analyser ( Applied Biosystems ) . For that , sample preparation was performed with one of the oligonucleotides having a 5’ 6-carboxyfluorescein ( FAM ) flurophore in the PCR reaction . The activity of IRE1 variants was analysed by introducing them into the endogenous Ern1 locus of CHO-K1 CHOP::GFP and XBP1s::Turquoise dual UPR reporter cells using a Ern1 null cell line ( ∆IRE1 as described in Kono et al . , 2017 ) . Cells were transfected with a Cas9-CRISPR guide construct targeting the Ern1 locus ( UK1903 ) together with the respective repair templates ( UK2425 for chimeric J-IRE1 , UK2426 for JQPD-IRE1 , UK1968 for wild-type IRE1 , UK2384 for IRE1 ∆loop , UK2385 for IRE1 ∆tail , UK2386 for IRE1 ∆∆ ) and grown for 7 days before further analysis . Cells that successfully repaired the IRE1 locus were selected by FACS by gating for cells exhibiting increased XBP1s::Turquoise fluorescence after 2-deoxyglucose treatment . Cells transfected with J-IRE1 as repair template were additionally transiently transfected with a plasmid encoding SubA wild-type , mutant or an empty vector ( UK1452 , UK1459 , UK1314 respectively ) before FACS . Data shown in Figure 4A and Figure 4—figure supplement 1A was acquired after transient transfection using a mixed population of cells and data shown in Figures 1A , B , C and 4B and Figure 4—figure supplement 1B with clonal cell lines . Cell lysis was performed as described previously ( Amin-Wetzel et al . , 2017 ) . All reagents were kept on ice throughout . Cells were washed in PBS , removed from the culture dish in PBS + 1 mM EDTA with a cell scraper and then pelleted at 370 × g for 5 min at 4°C . Cells were incubated in lysis buffer ( 1% Triton X-100 , 150 mM NaCl , 20 mM HEPES-KOH pH 7 . 5 , 10% glycerol , 1 mM phenylmethylsulphonyl fluoride ( PMSF ) , 4 μg/m Aprotinin , 2 μg/ml Pepstatin A and 2 μM Leupeptin ) for 5 min . Next , the samples were clarified at 21 , 130 g for 10 min at 4°C . The supernatant was transferred to a fresh tube and protein concentration measured with BioRad protein assay reagent ( Bio-Rad ) . For BiP co-IP experiments , non-specific binding of BiP to protein-A sepharose beads was decreased by digitonin treatment ( Le Gall et al . , 2004 ) to remove non-membrane associated BiP from cells prior to lysis . After pelleting , cells were washed in HNC buffer ( 50 mM HEPES-KOH pH 7 . 5 , 150 mM NaCl , 2 mM CaCl2 ) and then incubated in HNC + 0 . 1% ( w/v ) digitonin ( Calbiochem ) for 10 min . Cells were then washed in HNE buffer ( 50 mM HEPES-KOH pH 7 . 5 , 150 mM NaCl , 1 mM EGTA ) before proceeding to lysis using lysis buffer supplemented with 10 mM MgCl2 , 6 mg/ml glucose and 50 U/ml Hexokinase ( H4502 Sigma ) to deplete ATP and stabilise BiP-substrate interactions . To analyse the amount of BiP co-immunoprecipitated with endogenously expressed IRE1 variants ( Figures 1C and 4B ) or transiently transfected IRE1LD-GST variants ( Figure 4C ) , Protein A sepharose 4B beads ( Zymed Invitrogen ) or Glutathione ( GSH ) Sepharose 4B beads ( GE Healthcare ) were equilibrated in lysis buffer . Next , 20 μl beads per sample and anti-IRE1α were added to lysates and left rotating for 1 hr at 4°C . The beads were then washed in lysis buffer and residual liquid removed using a syringe . The protein from the beads was eluted in SDS sample buffer containing 20 mM DTT . Anti-mouse IRE1α serum ( NY200 ) was used for IP and immunoblot detection of endogenous IRE1α ( Bertolotti et al . , 2000 ) . An anti-hamster BiP antibody was used for immunoblot detection of endogenous BiP ( Avezov et al . , 2013 ) . Anti-GST serum was used for immunoblot detection of GST fusion proteins ( Ron and Habener , 1992 ) . Samples were run on standard polyacrylamide Tris-HCl gels and transferred to Immobilon-P PVDF membrane ( Pore size 0 . 45 μm , Sigma ) . Membranes were then blocked in 5% ( w/v ) dried skimmed milk in PBS , washed in TBS with 0 . 1% Tween-20 and exposed to various primary antibodies/antisera followed by incubation with IRDye fluorescently labelled secondary antibodies . Imaging was carried out with using a LICOR CLx Odyssey infrared imager . Coomassie-staining was carried out with Instant Blue ( Expedeon ) . Signal quantitation from SDS-PAGE gels or from immunoblots was carried out using the ImageJ software ( NIH ) . Initial crystals were obtained by screening commercial crystallisation plates via 200 nl protein ( 16 mg/ml ) plus 200 nl well solution in 96-well sitting drop plates at 20°C . The best diffraction dataset was collected from a crystal grown in 9% MPD , 0 . 1 M HEPES-KOH pH 7 . 5 microseeded ( D'Arcy et al . , 2007 ) from diluted initial crystals in 20% MPD , 0 . 1 M HEPES-KOH pH 7 . 5 . Crystals were briefly soaked into 9% MPD , 0 . 1 M HEPES-KOH pH 7 . 5 , 25% ( v/v ) glycerol and cryocooled in liquid nitrogen . Diffraction data was collected at beamline I04-1 in the Diamond Synchrotron Light Source ( DLS ) and processed by the XIA2 pipeline ( Winter , 2010 ) implementing Dials ( Winter et al . , 2018 ) for indexing and integration , Pointless for space group determination , and Aimless for scaling and merging ( Evans , 2011 ) . The structure was solved by searching the published IRE1LD core structure ( PDB 2HZ6 ) using Phaser ( McCoy et al . , 2007 ) . One molecule of IRE1 was found in one asymmetric unit , but the electron density around Cys105 and the SG-SG bond length suggested that Cys105 formed a disulphide bond with the symmetric Cys105 . Further refinement was performed iteratively using COOT ( Emsley et al . , 2010 ) and refmac5 ( Winn et al . , 2001 ) ( Table 1 ) in CCP4i2 interface ( Potterton et al . , 2018 ) and phenix . refine ( Liebschner et al . , 2019 ) . MolProbity ( Chen et al . , 2010 ) was consulted throughout the refinement process . Molecular graphics were generated with PyMOL Molecular Graphics System , Educational-use-only version 4 . 5 Schrodinger , LLC ) .	Cells produce many protein molecules . These are made of chains of building blocks called amino acids that then fold into three-dimensional shapes . Specialist proteins known as chaperones assist this folding process . For example , the chaperone BiP helps other proteins fold in a compartment within the cell called the endoplasmic reticulum . To match the supply of chaperones to the demand of unfolded proteins , cells have stress receptors , such as IRE1 in the endoplasmic reticulum . IRE1 responds to changing levels of unfolded proteins by generating signals that tell cells whether they need more chaperones . Previous studies in a test tube suggest that when levels of unfolded proteins are low , BiP represses IRE1 signalling . However , when the levels of unfolded proteins increase , the unfolded proteins compete with IRE1 for BiP , releasing the brake BiP imposes on IRE1 signalling . It remained unclear if BiP regulates IRE1 in the same way in living cells . To address this question , Amin-Wetzel , Neidhardt et al . studied IRE1 signalling in mammalian cells grown in the laboratory . The experiments revealed that cells containing a modified version of IRE1 to which BiP binds more strongly had less IRE1 signalling . On the other hand , cells containing versions of IRE1 that BiP binds less well had more active IRE1 signalling . These findings suggest that in cells , as in the test tube , unfolded proteins and IRE1 compete for BiP binding . This relationship comprises a simple mechanism allowing cells to sense and respond to the burden of unfolded proteins in their endoplasmic reticulum . Over time , the amount of unfolded proteins in the cell likely contributes to the development of aging-related diseases such as adult-onset diabetes . A better understanding of how cells handle unfolded proteins may lead to more effective treatments for these diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2019	Unstructured regions in IRE1α specify BiP-mediated destabilisation of the luminal domain dimer and repression of the UPR
A significant body of research in cognitive neuroscience is aimed at understanding how object concepts are represented in the human brain . However , it remains unknown whether and where the visual and abstract conceptual features that define an object concept are integrated . We addressed this issue by comparing the neural pattern similarities among object-evoked fMRI responses with behavior-based models that independently captured the visual and conceptual similarities among these stimuli . Our results revealed evidence for distinctive coding of visual features in lateral occipital cortex , and conceptual features in the temporal pole and parahippocampal cortex . By contrast , we found evidence for integrative coding of visual and conceptual object features in perirhinal cortex . The neuroanatomical specificity of this effect was highlighted by results from a searchlight analysis . Taken together , our findings suggest that perirhinal cortex uniquely supports the representation of fully specified object concepts through the integration of their visual and conceptual features . Semantic memory imbues the world with meaning and shapes our understanding of the relationships among object concepts . Many neurocognitive models of semantic memory incorporate the notion that object concepts are represented in a feature-based manner ( Rosch and Mervis , 1975; Tyler and Moss , 2001; Rogers and McClelland , 2004 ) . On this view , our understanding of the concept ‘hairdryer’ is thought to reflect knowledge of observable perceptual properties ( e . g . visual form ) and abstract conceptual features ( e . g . ‘used to style hair’ ) . Importantly , there is not always a one-to-one correspondence between how something looks and what it is; a hairdryer and a comb are conceptually similar despite being visually distinct , whereas a hairdryer and a gun are conceptually distinct despite being visually similar . Thus , a fully-specified representation of an object concept ( i . e . one that can be distinguished from any and all other concepts ) , requires integration of its perceptual and conceptual features . Neuroimaging research suggests that object features are coded in the modality-specific cortical regions that supported their processing at the time of acquisition ( Thompson-Schill , 2003 ) . For example , knowledge about the visual form of an object concept is thought to be coded in occipito-temporal visual processing regions ( Martin and Chao , 2001 ) . However , neurocognitive models of semantic memory differ with respect to how distributed feature representations relate to fully specified object concepts . On one view , these representations are thought to emerge through interactions among modality-specific cortical areas ( Kiefer and Pulvermüller , 2012; Martin , 2016 ) . Within a competing class of theories , they are thought to reflect the integration of modality-specific features in trans-modal convergence zones ( Damasio , 1989; Rogers et al . , 2004; Binder and Desai , 2011 ) , such as the anterior temporal lobes ( ATL ) ( Patterson et al . , 2007; Tranel , 2009; Ralph et al . , 2017 ) . The dominant view of the ATL as a semantic hub was initially shaped by neuropsychological investigations in individuals with semantic dementia ( SD ) ( Patterson et al . , 2007 ) . Behaviorally , SD is characterized by the progressive loss of conceptual knowledge across all receptive and expressive modalities ( Warrington , 1975; Hodges et al . , 1992 ) . At the level of neuropathology , SD is associated with extensive atrophy of the ATL , with the earliest and most pronounced volume loss in the left temporal pole ( Mummery et al . , 2000; Galton et al . , 2001 ) . Most important from a theoretical perspective , patients with SD tend to confuse conceptually similar objects that are visually distinct ( e . g . hairdryer – comb ) , but not visually similar objects that are conceptually distinct ( e . g . , hairdryer – gun ) , indicating that the temporal pole expresses conceptual similarity structure ( Graham et al . , 1994; see Peelen and Caramazza , 2012; Chadwick et al . , 2016 , for related neuroimaging evidence ) . Taken together , these findings suggest that the temporal pole supports multi-modal integration of abstract conceptual , but not perceptual , features . Notably , however , a considerable body of research indicates that the temporal pole may not be the only ATL structure that supports feature-based integration . The representational-hierarchical model of object coding emphasizes a role for perirhinal cortex ( PRC ) , located in the medial ATL , in feature integration that is distinct from that of the temporal pole ( Murray and Bussey , 1999 ) . Namely , within this framework PRC is thought to support the integration of conceptual and perceptual features . In line with this view , object representations in PRC have been described in terms of conceptual feature conjunctions in studies of semantic memory ( Moss et al . , 2005; Bruffaerts et al . , 2013; Clarke and Tyler , 2014; 2015; Wright et al . , 2015 ) , and visual feature conjunctions in studies of visual processing ( Barense et al . , 2005; 2007; 2012; Lee et al . , 2005; Devlin and Price , 2007; Murray et al . , 2007; O'Neil et al . , 2009; Graham et al . , 2010 ) . However , it is difficult to synthesize results from these parallel lines of research , in part , because conceptual and perceptual features tend to vary concomitantly across stimuli ( Mur , 2014 ) . For example , demonstrating that ‘horse’ and ‘donkey’ are represented with greater neural pattern similarity in PRC than are ‘horse’ and ‘dolphin’ may reflect differences in conceptual or perceptual relatedness . Thus , although the representational-hierarchical account was initially formalized nearly two decades ago ( Murray and Bussey , 1999 ) , direct evidence of integration across conceptual and perceptual features remains elusive . In the current study , we used fMRI to identify where in the brain visual and conceptual object features are stored , and to determine whether and where they are integrated at the level of fully specified object representations . To this end , we first generated behavior-based models that captured the visual and conceptual similarities among a set of object concepts , ensuring that these dimensions were not confounded across stimuli ( Figure 1 ) . Next , participants were scanned using task contexts that biased attention to either the conceptual or visual features of these well-characterized object concepts ( Figure 2 ) . We then used representational similarity analysis ( RSA ) ( Kriegeskorte and Kievit , 2013 ) , implemented using ROI- and searchlight-based approaches , to determine where the brain-based similarity structure among object-evoked multi-voxel activity patterns could be predicted by the similarity structure in the behavior-based visual and conceptual similarity models . We predicted that lateral occipital cortex ( LOC ) , an occipito-temporal region that has been implicated in the processing of visual form ( Grill-Spector et al . , 1999; Kourtzi and Kanwisher , 2001; Milner and Goodale , 2006 ) , would represent stored visual object features in a visual similarity code . Based on the neurocognitive models of semantic memory reviewed , we predicted that the temporal pole would represent stored conceptual object features in a conceptual similarity code ( Patterson et al . , 2007; Ralph et al . , 2017 ) . We also predicted conceptual similarity coding in parahippocampal cortex , which has been linked to the representation of the contextually-based co-occurrence of objects ( Bar , 2004; Aminoff et al . , 2013 ) . Critically , objects that are regularly encountered in the same context ( e . g . ‘comb’ and ‘hairdryer’ in a barbershop ) often share many conceptual features ( e . g . ‘used to style hair’ ) . Thus , to the extent that shared conceptual features directly shape contextual meaning , object-evoked responses in parahippocampal cortex may express conceptual similarity structure . Returning to the primary objective of the study , we predicted that PRC would uniquely represent the visual and conceptual features that define fully-specified object concepts in an integrated similarity code . Using a data-driven approach , we first generated behavior-based models that captured the visual and conceptual similarities among 40 targeted object concepts ( Figure 1 ) . Notably , our visual similarity model and conceptual similarity model were derived from behavioral judgments provided by two independent groups of participants . For the purpose of constructing the visual similarity model , the first group of participants ( N = 1185 ) provided pairwise comparative similarity judgments between object concepts ( Figure 1A ) . Specifically , a pair of words was presented on each trial and participants were asked to rate the visual similarity between the object concepts to which they referred using a 5-point Likert scale . Similarity ratings for each pair of object concepts were averaged across participants , normalized , and expressed within a representational dissimilarity matrix ( RDM ) . We refer to this RDM as the behavior-based visual RDM . For the purpose of constructing the conceptual similarity model , a second group of participants ( N = 1600 ) completed an online feature-generation task ( McRae et al . , 2005; Taylor et al . , 2012 ) ( Figure 1B ) . Each participant was asked to generate a list of conceptual features that characterize one object concept ( e . g . hairdryer: ‘used to style hair’ , ‘found in salons’ , ‘electrically powered’ , ‘blows hot air’; comb: ‘used to style hair’ , ‘found in salons’ , ‘has teeth’ , ‘made of plastic’ ) . Conceptual similarity between all pairs of object concepts was quantified as the cosine angle between the corresponding pairs of feature vectors . With this approach , high cosine similarity between object concepts reflects high conceptual similarity . Cosine similarity values were then expressed within an RDM , which we refer to as the behavior-based conceptual RDM . We next performed a second-level RSA to quantify the relationship between our behavior-based visual RDM and behavior-based conceptual RDM . This comparison is denoted by the gray arrow between behavior-based RDMs in Figure 1 . Critically , this analysis revealed that the model RDMs were not significantly correlated with one another ( Kendall’s tau-a = 0 . 01 , p=0 . 10 ) , indicating that differences in visual and conceptual features were not confounded across object concepts . In other words , ensuring that these different types of features varied independently across stimuli ( e . g . hairdryer – gun; hairdryer – comb ) , rather than concomitantly ( e . g . horse – donkey; horse – dolphin ) , allowed us to isolate the separate influence of visual and conceptual features on the representational structure of object concepts in the brain . In this example , a hairdryer and a gun are visually similar but conceptually dissimilar , whereas a hairdryer and a comb are visually dissimilar but conceptually similar . We next sought to compare our behavior-based RDMs with a corpus-based model of conceptual similarity . To this end , we implemented a word2vec language model , which mapped 3 million words to 300 feature vectors in a high-dimensional space ( Mikolov et al . , 2013 ) . The model was trained using ~100 billion words from a Google News dataset . From this model , we calculated the cosine similarity between feature vectors for all pairs of words in our stimulus set . These data were expressed in a 40 × 40 word2vec RDM ( Figure 1—source data 1 contains the word2vec RDM ) . Importantly , the word2vec RDM was significantly correlated with our behavior-based conceptual RDM ( Kendall’s tau-a = 0 . 11 , SE = 0 . 0141 , p<0 . 00001 ) , suggesting that both models captured the conceptual similarity structure among the object concepts . However , the word2vec RDM was also significantly correlated with our behavior-based visual RDM ( Kendall’s tau-a = 0 . 04 , SE = 0 . 0130 , p<0 . 001 ) . This result suggests that , in line with our objectives , the behavior-based conceptual RDM captured semantic similarity selectively defined as conceptual object features , whereas the word2vec RDM may have captured a broader definition of semantic similarity , that is , one that includes both visual semantics and abstract conceptual features . Consistent with this view , gun and hairdryer were conceptually unrelated in our behavior-based conceptual RDM ( cosine = 0 ) , whereas the word2vec RDM suggested modest conceptual similarity ( cosine = 0 . 16 ) . Although this difference is likely determined by multiple factors , it is important to note that gun and hairdryer had a relatively high visual similarity index in our behaviour-based visual RDM ( normalized mean rating = 0 . 58 ) . These data highlight a theoretically important distinction between our behaviorally derived conceptual feature-based statistics and corpus-based estimates of semantic similarity . Specifically , the former allow for distinctions between visual and conceptual object features , whereas corpus-based models may not . We used fMRI to estimate the representational structure of our 40 object concepts from neural activity patterns in an independent group of participants ( Figure 2 ) . Given our specific interest in understanding pre-existing representations of object concepts rather than bottom-up perceptual processing , all stimuli were presented as words . This approach ensured that conceptual and visual features were extracted from pre-existing representations of object concepts . That is to say , both conceptual and visual features were arbitrarily related to the physical input ( i . e . the orthography of the word ) . By contrast , when pictures are used as stimuli , visual features are accessible from the pictorial cue , whereas conceptual features require abstraction from the cue . Functional brain data were acquired over eight experimental runs , each of which consisted of two blocks of stimulus presentation . All 40 object concepts were presented sequentially within each block , for a total of 16 repetitions per concept . On each trial , participants were asked to make a ‘yes/no’ property verification judgment in relation to a block-specific verification probe . Half of the blocks were associated with verification probes that encouraged processing of visual features ( e . g . ‘is the object angular ? ” ) , and the other half were associated with verification probes that encouraged processing of conceptual features ( e . g . ‘is the object a tool ? ” ) . Each run consisted of one visual feature verification block and one conceptual feature verification block , with order counterbalanced across runs . With this experimental design , we were able to characterize neural responses to object concepts across two task contexts: a visual task context ( Figure 2A ) and a conceptual task context ( Figure 2B ) . Behavioral performance on the scanned property verification task indicated that participants interpreted the object concepts and property verification probes with a high degree of consistency ( Figure 3 ) . Specifically , all participants ( i . e . 16/16 ) provided the same yes/no response to the property verification task on 88 . 4% of all trials . Agreement was highest for the ‘living’ verification probe ( 96 . 8% ) and lowest for the ‘non-tool’ verification probe ( 73 . 2% ) . Moreover , the proportion of trials on which all participants provided the same response did not differ between the visual feature verification task context ( mean = 87 . 3% collapsed across all eight visual probes ) and the conceptual feature verification task context ( mean = 89 . 5% collapsed across all eight conceptual probes ) ( z = 0 . 19 , p=0 . 85 ) . Response latencies were also comparable across the visual feature verification task context ( mean = 1361 ms , SD = 303 ) and the conceptual feature verification task context ( mean = 1376 ms , SD = 315 ) ( t ( 15 ) =1 . 00 , p=0 . 33 , 95% CI [−49 . 09 , 17 . 71 ) . We next quantified pairwise similarities between object-evoked multi-voxel activity patterns using a first-level RSA ( Figure 2 ) . For the purpose of conducting ROI-based RSA , we focused on multi-voxel activity patterns obtained in PRC , the temporal pole , parahippocampal cortex , and LOC . ROIs from a representative participant are presented in Figure 4 . These ROIs were selected a priori based on empirical evidence linking their respective functional characteristics to visual object processing , conceptual object processing , or both . Our primary focus was on PRC , which has been linked to integrative coding of visual object features and conceptual object features across parallel lines of research ( Barense et al . , 2005; 2007; 2012; Lee et al . , 2005; O'Neil et al . , 2009; Bruffaerts et al . , 2013; Clarke and Tyler , 2014; 2015; Wright et al . , 2015; Erez et al . , 2016 ) . The temporal pole has primarily been linked to processing of conceptual object properties ( Mummery et al . , 2000; Galton et al . , 2001; Patterson et al . , 2007; Pobric et al . , 2007; Lambon Ralph et al . , 2009; Peelen and Caramazza , 2012; Chadwick et al . , 2016 ) . Parahippocampal cortex has been implicated in the conceptual processing of contextual associations , including representing the co-occurrence of objects , although its functional contributions remain less well defined than the temporal pole ( Bar and Aminoff , 2003; Aminoff et al . , 2013; Ranganath and Ritchey , 2012 ) . Lastly , LOC , which is a functionally defined region in occipito-temporal cortex , has been revealed to play a critical role in processing visual form ( Grill-Spector et al . , 1999; Kourtzi and Kanwisher , 2001; Milner and Goodale , 2006 ) . Because we did not have any a priori predictions regarding hemispheric differences , estimates of neural pattern similarities between object concepts were derived from multi-voxel activity collapsed across ROIs in the left and right hemisphere . Object-specific multi-voxel activity patterns were estimated in each run using general linear models fit to data from the visual and conceptual task contexts , separately . Mean object-specific responses were then calculated for each task context by averaging across runs . Linear correlation distances ( Pearson’s r ) were calculated between all pairs of object-specific multi-voxel activity patterns within each task context and expressed in participant-specific brain-based visual task RDMs and brain-based conceptual task RDMs . The brain-based visual task RDMs captured the neural pattern similarities obtained between all object concepts in the visual task context ( i . e . while participants made visual feature verification judgments ) ( Figure 2A ) , and the brain-based conceptual task RDMs captured the neural pattern similarities obtained between all object concepts in the conceptual task context ( i . e . while participants made conceptual feature verification judgments ) ( Figure 2B ) . We implemented second-level RSA to compare behavior-based visual and conceptual RDMs with the brain-based visual and conceptual task RDMs ( these comparisons are denoted by the solid vertical and diagonal arrows in Figure 5 ) . All RDMs were compared in each ROI using a ranked correlation coefficient ( Kendall’s tau-a ) as a similarity index ( Nili et al . , 2014 ) . Inferential statistical analyses were performed using a one-sided Wilcoxon signed-rank test , with participants as a random factor . A Bonferroni correction was applied to adjust for multiple comparisons ( 4 ROIs x 2 behavior-based RDMs x 2 brain-based RDMs = 16 comparisons , yielding a critical alpha of . 003 ) . With this approach , we revealed that object concepts are represented in three distinct similarity codes that differed across ROIs: a visual similarity code , a conceptual similarity code , and an integrative code . Results from our ROI-based RSA analyses are shown in Figure 6 and discussed in turn below . Consistent with its well-established role in the processing of visual form , patterns of activity within LOC reflected the visual similarity of the object concepts ( Figure 6A ) . Specifically , the brain-based visual task RDMs obtained across participants in LOC were significantly correlated with the behavior-based visual RDM ( Kendall’s tau-a = 0 . 045 , p<0 . 002 ) , but not the behavior-based conceptual RDM ( Kendall’s tau-a = −0 . 006 , p=0 . 72 ) . In other words , activity patterns in LOC expressed a visual similarity structure when participants were asked to make explicit judgments about the visual features that characterized object concepts ( e . g . whether an object is angular in form ) . By contrast , the brain-based conceptual task RDMs obtained across participants in LOC were not significantly correlated with either the behavior-based visual RDM ( Kendall’s tau-a = 0 . 006 , p=0 . 13 ) or the behavior-based conceptual RDM ( Kendall’s tau-a = 0 . 003 , p=0 . 65 ) . That is to say , activity patterns in LOC expressed neither visual nor conceptual similarity structure when participants made judgments that pertained to conceptual object features ( e . g . whether an object is naturally occurring ) . Considered together , these results suggest that LOC represented perceptual information about object concepts in a task-dependent visual similarity code . Specifically , when task demands biased attention toward visual features , signals in LOC generalized across visually related object concepts even when they are conceptually distinct ( e . g . hairdryer – gun ) . Patterns of activity obtained in parahippocampal cortex , which has previously been associated with the processing of semantically-based contextual associations ( Bar and Aminoff , 2003; Aminoff et al . , 2013 ) , reflected the conceptual similarity of the object concepts ( Figure 6B ) . First , the brain-based visual task RDMs obtained across participants in parahippocampal cortex were not significantly correlated with either the behavior-based visual RDM ( Kendall’s tau-a = 0 . 005 , p=0 . 26 ) or the behavior-based conceptual RDM ( Kendall’s tau-a = 0 . 009 , p=0 . 26 ) . In other words , activity patterns in parahippocampal cortex expressed neither visual nor conceptual similarity structure when participants made judgments that pertained to conceptual object features ( e . g . whether an object is symmetrical ) . Second , the brain-based conceptual task RDMs obtained across participants in parahippocampal cortex were not significantly related to the behavior-based visual RDM ( Kendall’s tau-a = −0 . 008 , p=0 . 55 ) , but they were correlated with the behavior-based conceptual RDM ( Kendall’s tau-a = 0 . 046 , p<0 . 002 ) . Thus , activity patterns in parahippocampal cortex expressed a conceptual similarity structure when participants were asked to make explicit judgments about the conceptual features that characterized object concepts ( e . g . whether an object is a tool ) . Put another way , conceptual information was represented in parahippocampal cortex in a task-dependent manner that generalized across conceptually related object concepts even when they were visually distinct ( e . g . hairdryer – comb ) . In line with theoretical frameworks that have characterized the temporal pole as a semantic hub ( Patterson et al . , 2007; Tranel , 2009 ) , patterns of activity within this specific ATL structure reflected the conceptual similarity of the object concepts ( Figure 6D ) . Specifically , whereas the brain-based visual task RDMs obtained across participants in the temporal pole were not significantly correlated with the behavior-based visual RDM ( Kendall’s tau-a = 0 . 006 , p=0 . 25 ) , they were correlated with the behavior-based conceptual RDM ( Kendall’s tau-a = 0 . 035 , p<0 . 001 ) . In other words , activity patterns in the temporal pole expressed a conceptual similarity structure when participants were asked to make explicit judgments about the visual features that characterized object concepts ( e . g . whether an object is elongated ) . Similarly , whereas the brain-based conceptual task RDMs obtained across participants in the temporal pole were not significantly correlated with the behavior-based visual RDM ( Kendall’s tau-a = 0 . 0005 , p=0 . 47 ) , they were correlated with the behavior-based conceptual RDM ( Kendall’s tau-a = 0 . 05 , p<0 . 0001 ) . Thus , activity patterns in the temporal pole expressed a conceptual similarity structure when participants were asked to make explicit judgments about either the visual or conceptual features that characterized object concepts ( e . g . whether an object is dark in color , or whether an object is pleasant ) . In other words , conceptual information was represented in the temporal pole in a task-invariant manner that generalized across conceptually related object concepts even when they were visually distinct ( e . g . hairdryer – comb ) . Results obtained in PRC support the notion that this structure integrates visual and conceptual object features ( 6C ) , as first theorized in the representational-hierarchical model of object representation ( Murray and Bussey , 1999 ) . Namely , we revealed that the brain-based visual task RDMs obtained across participants in PRC were significantly correlated with both the behavior-based visual RDM ( Kendall’s tau-a = 0 . 052 , p<0 . 0001 ) , and the behavior-based conceptual RDM ( Kendall’s tau-a = 0 . 036 , p<0 . 0003 ) . Similarly , the brain-based conceptual task RDMs obtained across participants were also correlated with both the behavior-based visual RDM ( Kendall’s tau-a = 0 . 035 , p<0 . 002 ) , and the behavior-based conceptual RDM ( Kendall’s tau-a = 0 . 057 , p<0 . 0001 ) . In other words , activity patterns in PRC expressed both visual and conceptual similarity structure when participants were asked to make explicit judgments about the visual features that characterized object concepts ( e . g . whether an object is round ) and when participants were asked to make explicit judgments about the conceptual features that characterized object concepts ( e . g . whether an object is manufactured ) . Numerically , patterns of activity in PRC showed more similarity to the behavior-based visual RDM than to the behavior-based conceptual RDM in the visual task context , and vice versa in the conceptual task context . Therefore , we performed a 2 [behavior-based RDMs] x 2 [brain-based task RDMs] repeated measures ANOVA to formally test for an interaction between behavior-based model and fMRI task context . For this purpose , all Kendall’s tau-a values were transformed to Pearson’s r co-efficients ( r = sin ( ½ π tau-a ) , Walker , 2003 ) , which were then Fisher-z transformed . The task x model interaction neared , but did not reach , significance ( F ( 1 , 15 ) = 3 . 48 , p=0 . 082 ) . In sum , these findings indicate that PRC simultaneously expressed both conceptual and visual similarity structure , and did so regardless of whether participants were asked to make targeted assessments of conceptual or visual features . In other words , activity patterns in PRC captured the conceptual similarity between hairdryer and comb , as well as the visual similarity between hairdryer and gun , and did so irrespective of task context . Critically , these results were obtained despite the fact that the brain-based RDMs were orthogonal to one another ( i . e . not significantly correlated ) . Considered together , these results suggest that , of the a priori ROIs considered , PRC represents object concepts at the highest level of specificity through integration of visual and conceptual features . For the purpose of comparison , we next examined similarities between the word2vec RDM and the brain-based RDMs using the same procedures described in the previous section . Results are presented in Figure 6—figure supplement 1 . These analyses revealed significant positive correlations between the word2vec RDM and the brain-based conceptual task RDMs in parahippocampal cortex ( Kendall’s tau-a = 0 . 05 , p<0 . 01 ) , PRC ( Kendall’s tau-a = 0 . 035 , p<0 . 01 ) , and the temporal pole ( Kendall’s tau-a = 0 . 029 , p<0 . 01 ) . The word2vec RDM was also significantly correlated with the brain-based visual task RDMs in PRC ( Kendall’s tau-a = 0 . 025 , p<0 . 05 ) and the temporal pole ( Kendall’s tau-a = 0 . 027 , p<0 . 05 ) . Notably , this pattern of results was identical to that obtained using the behavior-based conceptual RDMs in parahippocampal cortex , PRC , and the temporal pole . Interestingly , however , the word2vec RDM was also significantly correlated with the brain-based visual task RDMs in LOC ( Kendall’s tau-a = 0 . 028 , p<0 . 05 ) . This result is consistent with the observation that the word2vec RDM was significantly correlated with our behavior-based visual RDM , and further suggests that corpus-based models of semantic memory likely capture similarities between object concepts at the level of abstract conceptual properties and visual semantics . Having examined the relationships between behavior-based RDMs and brain-based RDMs , we next sought to directly characterize the relationships between brain-based conceptual and visual RDMs within each ROI ( these comparisons are denoted by the dashed horizontal arrow in the bottom of Figure 5 ) . These analyses were conducted using the same methodological procedures used to compare behavior-based RDMs with brain-based RDMs in the previous section . A Bonferroni correction was applied to adjust for multiple comparisons ( 16 brain-based comparisons , yielding a critical alpha of . 003 ) . Using second-level RSAs , we asked whether the brain-based visual task RDMs and brain-based conceptual task RDMs had a common similarity structure within a given ROI . Results are plotted in Figure 7A . Importantly , we found a significant positive correlation in PRC ( Kendall’s tau-a = 0 . 063 , p<0 . 0001 ) , and a trend toward a significant correlation in the temporal pole ( Kendall’s tau-a = 0 . 032 , p=0 . 012 ) . Conversely , brain-based visual and conceptual task RDMs were not significantly correlated in either parahippocampal cortex ( Kendall’s tau-a = 0 . 008 , p=0 . 12 ) , or LOC ( Kendall’s tau-a = −0 . 008 , p=0 . 92 ) . These results suggest that object concepts were represented similarly within PRC , and to a lesser extent within the temporal pole , regardless of whether they were encountered in a visual or conceptual task context . We next conducted second-level RSAs to quantify representational similarities between the brain-based visual task RDMs obtained across different ROIs . In other words , we asked whether activity in different ROIs ( e . g . PRC and LOC ) reflected similar representational distinctions across object concepts within the visual task context . Results are plotted in Figure 7B . Interestingly , these analyses did not reveal any significant results between any of our ROIs ( all Kendall’s tau-a <0 . 01 , all p>0 . 07 ) . These findings indicate that PRC and LOC , two regions that expressed a visual similarity code , represented different aspects of the visual object features . Finally , we quantified the representational similarities between the brain-based conceptual task RDMs obtained across different ROIs . In other words , we asked whether activity in different ROIs ( e . g . PRC and the temporal pole ) reflected similar representational distinctions across object concepts within the conceptual task context . Results are plotted in Figure 7C . This set of analyses did not reveal any significant results between any of our ROIs ( all Kendall’s tau-a <0 . 016 , all p>0 . 012 ) . These findings indicate that the three regions that expressed a conceptual similarity code ( i . e . , PRC , parahippocampal cortex , and temporal pole ) , represented different aspects of the conceptual object features . The RSAs reported thus far have quantified relationships among behavior-based and brain-based RDMs that reflected similarities between different object concepts ( e . g . between ‘hairdryer’ and ‘comb’ ) . We next quantified within-object similarities ( e . g . between ‘hairdryer’ and ‘hairdryer’ ) across visual and conceptual task contexts ( e . g . ‘is it living ? ’ or ‘is it angular ? ” ) using first-level RSAs . Specifically , we calculated one dissimilarity value ( 1 – Pearson’s r ) between the mean multi-voxel activity patterns evoked by a given object concept across different task contexts . These 40 within-object dissimilarity values were expressed along the diagonal of an RDM for each ROI in each participant , separately ( Figure 8A ) . We next calculated mean within-object dissimilarity by averaging across the diagonal of each RDM for the purpose of performing statistical inference . Results are presented in Figure 8B . Within-object similarity did not differ from zero in either LOC ( Pearson’s r = 0 . 007 , p=0 . 20 ) or parahippocampal cortex ( Pearson’s r = −0 . 008 , p=0 . 87 ) , suggesting that a given object concept was represented differently across the visual and conceptual task contexts in these ROIs . These findings are consistent with the task-dependent nature of the similarity codes we observed in these regions ( Figure 6A and B ) . Conversely , within-object similarity was significantly greater than zero in the temporal pole ( Pearson’s r = 0 . 34 , p<0 . 05 , Bonferroni corrected for four comparisons ) , indicating that this structure represents a given object concept similarly across different task contexts . This observation is consistent with results from the previous section which revealed that the similarities between object concepts in the temporal pole are preserved across task contexts ( Figure 7A ) . These findings reflected the fact that the same conceptual object information ( e . g . ‘used to style hair’ and ‘found in salons’ ) was carried in multi-voxel activity patterns obtained in each task context ( Figure 6D ) . Within-object similarity was also significantly greater than zero in PRC ( Pearson’s r = 0 . 41 , p<0 . 01 , Bonferroni corrected for four comparisons ) , again indicating that a given object concept was represented similarly across different task contexts . This finding dovetails with our result from the previous section which revealed that the similarities between object concepts in PRC were preserved across task contexts ( Figure 7A ) . When considered together , we interpret this pattern of results in PRC as further evidence of integrative coding , reflecting the fact that this structure carried the same conceptual ( e . g . ‘used to style hair’ and ‘found in salons’ ) and visual ( e . g . visually similar to a gun ) object information in both task contexts ( Figure 6C ) . Decades of research has been aimed at understanding how object concepts are represented in the brain ( Warrington , 1975; Hodges et al . , 1992; Martin et al . , 1995; Murray and Bussey , 1999; Chen et al . , 2017 ) , yet the fundamental question of whether and where their visual and conceptual features are integrated remains unanswered . Progress toward this end has been hindered by the fact that these features tend to vary concomitantly across object concepts . Here , we used a data-driven approach to systematically select a set of object concepts in which visual and conceptual features varied independently ( e . g . hairdryer – comb , which are conceptually similar but visually distinct; hairdryer – gun , which are visually similar but conceptually distinct ) . Using RSA of fMRI data , we revealed novel evidence of task-dependent visual similarity coding in LOC , task-dependent conceptual similarity coding in parahippocampal cortex , task-invariant coding in the temporal pole , and task-invariant integrative coding in PRC . Several aspects of our data provide novel support for the notion that PRC uniquely represents the visual and conceptual features that define fully specified object concepts in an integrated similarity code . First , this was the only region of the brain in which both visual and conceptual object coding was revealed . Moreover , these effects were observed regardless of whether fMRI task demands biased attention toward visual or conceptual object features . These results are particularly striking given the fact that they were revealed using a behavior-based visual similarity model and a behavior-based conceptual similarity model that were orthogonal to one another . In other words , the degree of similarity between multi-voxel activity patterns obtained while participants made conceptual judgments , such as whether a ‘hairdryer’ is man-made or a ‘gun’ is pleasant , was captured by the degree of visual similarity between these object concepts . Likewise , the degree of similarity between multi-voxel activity patterns obtained while participants made visual judgments , such as whether a ‘hairdryer’ is angular or a ‘comb’ is elongated , was captured by the degree of conceptual similarity between these object concepts . In both cases , PRC carried information about pre-existing representations of object features that were neither required to perform the immediate task at hand , nor correlated with the features that did in fact have task-relevant diagnostic value . Moreover , we also found that the brain-based visual task RDMs and brain-based conceptual task RDMs were correlated with one another across task contexts in PRC . That is to say , the similarity between ‘hairdryer’ and ‘gun’ was comparable regardless of whether task demands biased attention toward visual or conceptual features . Likewise , we also revealed that PRC also represented a given object concept similarly across task contexts , that is , ‘hairdryer’ evoked a pattern of activation that was comparable across task contexts . When considered together , these results suggest that , at the level of PRC , it may not be possible to fully disentangle conceptual and perceptual information . An important but challenging objective for future research will be to determine whether this pattern of results can be replicated at the level of individual neurons . What is the behavioral relevance of fully specified object representations in which visual and conceptual features are integrated ? It has previously been suggested that such representations allow for discrimination among stimuli with extensive feature overlap , such as exemplars from the same category ( Murray and Bussey , 1999; Noppeney et al . , 2007; Graham et al . , 2010; Clarke and Tyler , 2015 ) . In line with this view , individuals with medial ATL lesions that include PRC typically have more pronounced conceptual impairments related to living than non-living things ( Warrington and Shallice , 1984; Moss et al . , 1997; Bozeat et al . , 2003 ) , and more striking perceptual impairments for objects that are visually similar as compared to visually distinct ( Barense et al . , 2007 , Barense et al . , 2010; Lee et al . , 2006 ) . Functional MRI studies in neurologically healthy individuals have also demonstrated increased PRC engagement for living as compared to non-living objects ( Moss et al . , 2005 ) , for known as compared to novel faces ( Barense et al . , 2011; Peterson et al . , 2012 ) , and for faces or conceptually meaningless stimuli with high feature overlap as compared to low feature overlap ( O'Neil et al . , 2009; Barense et al . , 2012 ) . In a related manner , fully specified object representations in PRC have also been implicated in long-term memory judgments . For example , PRC has been linked to explicit recognition memory judgments when previously studied and novel items are from the same stimulus category ( Martin et al . , 2013; 2016; 2018 ) , and when subjects make judgments about their lifetime of experience with a given object concept ( Duke et al . , 2017 ) . Common among these task demands is the requirement to discriminate among highly similar stimuli . In such scenarios , a fully specified representation that reflects the integration of perceptual and conceptual features necessarily enables more fine-grained distinctions than a purely perceptual or conceptual representation . This study also has significant implications for prominent neurocognitive models of semantic memory in which the ATL is characterized as a semantic hub ( Rogers et al . , 2006; Patterson et al . , 2007; Tranel , 2009 ) . On this view , the bilateral ATLs are thought to constitute a trans-modal convergence zone that abstracts conceptual information from the co-occurrence of features otherwise represented in a distributed manner across modality-specific cortical nodes . Consistent with this idea , we have shown that a behavior-based conceptual similarity model predicted the similarity structure of neural activity patterns in the temporal pole , irrespective of task context . Specifically , neural activity patterns associated with conceptually similar object concepts that are visually distinct ( e . g . hairdryer – comb ) were more comparable than were conceptually dissimilar concepts that are visually similar ( e . g . hairdryer – gun ) , even when task demands required a critical assessment of visual features . This observation , together with results obtained in PRC , demonstrates a representational distinction between these ATL structures , a conclusion that dovetails with recent evidence indicating that this region is not functionally homogeneous ( Binney et al . , 2010; Murphy et al . , 2017 ) . Ultimately , this outcome suggests that some ATL sub-regions play a prominent role in task-invariant extraction of conceptual object properties ( e . g . temporal pole ) , whereas others appear to make differential contributions to the task-invariant integration of perceptual and conceptual features ( e . g . PRC ) ( Ralph et al . , 2017; Chen et al . , 2017 ) . Convergent evidence from studies of functional and structural connectivity in humans , non-human primates , and rodents have revealed that PRC is connected to the temporal pole , parahippocampal cortex , LOC , and nearly all other unimodal and polymodal sensory regions in neocortex ( Suzuki and Amaral , 1994; Burwell and Amaral , 1998; Kahn et al . , 2008; McLelland et al . , 2014; Suzuki and Naya , 2014; Wang et al . , 2016; Zhuo et al . , 2016 ) . Importantly , our results have linked LOC to the representation of visual object features , and the temporal pole and parahippocampal cortex to the representation of conceptual object features . Thus , PRC has the connectivity properties that make it well suited to be a trans-modal convergence zone capable of integrating object features that are both visual and conceptual in nature . An interesting challenge for future research will be to determine how differentially attending to specific types of object features shapes functional connectivity profiles between these regions . Although speculative , results from the current study suggest that attention may modulate information both within and between the ROIs examined . First , we see visual similarity coding in LOC only when task demands biased attention to visual object features , and conceptual similarity coding in parahippocampal cortex only when task demands biased attention to conceptual object features . Second , we saw a trend toward an interaction between behavior-based models and fMRI task context in PRC , such that visual similarity coding was more pronounced in the visual task context than was conceptual similarity coding , and vice versa . Thus , attending to specific types of features did not merely manifest as univariate gain modulation . Rather , attention appeared to modulated multi-voxel activity patterns . Another novel aspect of our findings is that parahippocampal cortex exhibited conceptual similarity coding in the conceptual task context . Interestingly , it has been suggested that this structure broadly contributes to cognition by processing contextual associations , including the co-occurrence of objects within a context ( Bar , 2004; Aminoff et al . , 2013 ) . Critically , objects that regularly co-occur in the same context ( e . g . ‘comb’ and ‘hairdryer’ in a barbershop ) often share many conceptual features ( e . g . functional properties such as ‘used to style hair’ ) , but do not necessarily share many visual features . Thus , object-evoked responses in parahippocampal cortex may express feature-based conceptual similarity structure because objects with many shared conceptual features bring to mind an associated context , whereas objects that are visually similar but conceptually distinct do not ( e . g . hairdryer and gun ) . We note , however , that the current study was not designed to test-specific hypotheses about the contextual co-occurrence of objects , or how co-occurrence relates to conceptual feature statistics . Ultimately , a mechanistic account of object-based coding in PHC will require further research using a carefully selected stimulus set in which the strength of contextual associations ( i . e . co-occurrence ) between object concepts is not confounded with conceptual features . In summary , this study sheds new light on our understanding of how object concepts are represented in the brain . Specifically , we revealed that PRC represented object concepts in a task-invariant , integrative similarity code that captured the visual and conceptual relatedness among stimuli . Most critically , this result was obtained despite systematically dissociating visual and conceptual features across object concepts . Moreover , the striking neuroanatomical specificity of this result suggests that PRC uniquely supports integration across these fundamentally different types of features . Ultimately , this pattern of results implicates PRC in the representation of fully-specified objects . As a starting point , we chained together a list of 80 object concepts in such a way that adjacent items in the list alternated between being conceptually similar but visually distinct and visually similar but conceptually distinct ( e . g . bullet – gun – hairdryer – comb; bullet and gun are conceptually but not visually similar , whereas gun and hairdryer are visually but not conceptually similar , and hairdryer and comb are conceptually but not visually similar , etc . ) . Our initial stimulus set was established using the authors’ subjective impressions . The visual and conceptual similarities between all pairs of object concepts were then quantified by human observers in the context of a visual similarity rating task and a conceptual feature generation task , respectively . Results from these behavioral tasks were then used to select 40 object concepts used throughout the current study . Participants who completed the visual similarity rating task were presented with 40 pairs of words and asked to rate visual similarity between the object concepts to which they referred ( Figure 1A ) . Responses were made using a 5-point scale ( very dissimilar , somewhat dissimilar , neutral , somewhat similar , very similar ) . Each participant was also presented with four catch trials on which an object concept was paired with itself . Across participants , 95 . 7% of catch trials were rated as being very similar . Data were excluded from 28 participants who did not rate all four catch trials as being at least ‘somewhat similar’ . Every pair of object concepts from the initial set of 80 object concepts ( 3160 ) was rated by 15 different participants . We next quantified conceptual similarities between object concepts based on responses obtained in a conceptual feature generation task ( Figure 1B ) , following task instructions previously described by McRae et al . , 2005 . Each participant was presented with one object concept and asked to produce a list of up to 15 different types of descriptive features , including functional properties ( e . g . what it is used for , where it is used , and when it is used ) , physical properties ( e . g . how it looks , sounds , smells , feels , and tastes ) , and other facts about it , such as the category to which it belongs or other encyclopedic facts ( e . g . where it is from ) . One example object and its corresponding features from a normative database were presented as an example ( McRae et al . , 2005 ) . Interpretation and organization of written responses were guided by criteria described by McRae et al . , 2005 . Features were obtained from 20 different participants for each object concept . Data were excluded from 33 participants who failed to list any features . A total of 4851 unique features were produced across all 80 object concepts and participants . Features listed by fewer than 4 out of 20 participants were considered to be unreliable and discarded for the purpose of all subsequent analyses , leaving 723 unique features . This exclusion criterion is proportionally comparable to that used by McRae et al . , 2005 . On average , each of the 80 object concepts was associated with 10 . 6 features . We used a data-driven approach to select a subset of 40 object concepts from the initial 80-item set . These 40 object concepts are reflected in the behavior-based visual and conceptual RDMs , and were used as stimuli in our fMRI experiment . Specifically , we first ensured that each object concept was visually similar , but conceptually dissimilar , to at least one other item ( e . g . hairdryer – gun ) , and conceptually similar , but visually dissimilar , to at least one different item ( e . g . hairdryer – comb ) . Second , in an effort to ensure that visual and conceptual features varied independently across object concepts , stimuli were selected such that the corresponding behavior-based visual and conceptual similarity models were not correlated with one another . During scanning , participants completed a feature verification task that required a yes/no judgment indicating whether a given feature was applicable to a specific object concept on a trial-by-trial basis . We systematically varied the feature verification probes in a manner that established a visual feature verification task context and conceptual feature verification task context . Verification probes comprising the visual task context were selected to encourage processing of the visual semantic features that characterize each object concept ( i . e . shape , color , and surface detail ) . To this end , eight specific probes were used: shape [ ( angular , rounded ) , ( elongated , symmetrical ) ] , color ( light , dark ) , and surface ( smooth , rough ) . Notably , all features are associated with two opposing probes ( e . g . angular and rounded; natural and manufactured ) to ensure that participants made an equal number of ‘yes’ and ‘no’ responses . Verification probes comprising the conceptual feature verification task context were selected to encourage processing of the abstract conceptual features that characterize each object concept ( i . e . animacy , origin , function , and affective associations ) . To this end , eight specific verification probes were used: ( living , non-living ) , ( manufactured , natural ) , ( tool , non-tool ) , ( pleasant , unpleasant ) . The primary experimental task was evenly divided over eight runs of functional data acquisition . Each run lasted 7 m 56 s and was evenly divided into two blocks , each of which corresponded to either a visual verification task context or a conceptual feature verification task context . The order of task blocks was counter-balanced across participants . Each block was associated with a different feature verification probe , with the first and second block in each run separated by 12 s of rest . Blocks began with an 8 s presentation of a feature verification probe that was to be referenced for all intra-block trials . With this design , each object concept was repeated 16 times: eight repetitions across the visual feature verification task context and eight repetitions across the conceptual feature verification task context . Behavioral responses were recorded using an MR-compatible keypad . Stimuli were centrally presented for 2 s and each trial was separated by a jittered period of baseline fixation that ranged 2–6 s . Trial order and jitter interval were optimized for each run using the OptSeq2 algorithm ( http://surfer . nmr . mgh . harvard . edu/optseq/ ) , with unique sequences and timing across counterbalanced versions of the experiment . Stimulus presentation and timing was controlled by E-Prime 2 . 0 ( Psychology Software Tools , Pittsburgh , PA ) . Following completion of the main experimental task , each participant completed an independent functional localizer scan that was subsequently used to identify LOC . Participants viewed objects , scrambled objects , words , scrambled words , faces , and scenes in separate 24 s blocks ( 12 functional volumes ) . Within each block , 32 images were presented for 400 ms each with a 350 ms ISI . There were four groups of six blocks , with each group separated by a 12 s fixation period , and each block corresponding to a different stimulus category . Block order ( i . e . stimulus category ) was counterbalanced across groups . All stimuli were presented in the context of a 1-back task to ensure that participants remained engaged throughout the entire scan . Presentation of images within blocks was pseudo-random with 1-back repetition occurring 1–2 times per block . We performed RSA in four a priori defined ROIs . The temporal pole , PRC , and parahippocampal cortex were manually defined in both the left and right hemisphere on each participant’s high-resolution anatomical image according to established MR-based protocols ( Pruessner et al . , 2002 ) , with adjustment of posterior border of parahippocampal cortex using anatomical landmarks described by Frankó et al . ( 2014 ) . Lateral occipital cortex was defined as the set of contiguous voxels located along the lateral extent of the occipital lobe that responded more strongly to intact than scrambled objects ( p<0 . 001 , uncorrected; Malach et al . , 1995 ) . Scanning was performed using a 3 . 0 T Siemens MAGNETOM Trio MRI scanner at the Rotman Research Institute at Baycrest Hospital using a 32-channel receiver head coil . Each scanning session began with the acquisition of a whole-brain high-resolution magnetization-prepared rapid gradient-echo T1-weighted structural image ( repetition time = 2 s , echo time = 2 . 63 ms , flip angle = 9° , field of view = 25 . 6 cm2 , 160 oblique axial slices , 192 × 256 matrix , slice thickness = 1 mm ) . During each of eight functional scanning runs comprising the main experimental task , a total of 238 T2-weighted echo-planar images were acquired using a two-shot gradient echo sequence ( 200 × 200 mm field of view with a 64 × 64 matrix size ) , resulting in an in-plane resolution of 3 . 1 × 3 . 1 mm for each of 40 2 mm axial slices that were acquired in an interleaved manner along the axis of the hippocampus . The inter-slice gap was 0 . 5 mm; repetition time = 2 s; echo time = 30 ms; flip angle = 78° ) . These parameters yielded coverage of the majority of cortex , excluding only the most superior aspects of the frontal and parietal lobes . During a single functional localizer scan , a total of 360 T2-weighted echo-planar images were acquired using the same parameters reported for the main experimental task . Lastly , a B0 field map was collected following completion of the functional localizer scan Preprocessing and GLM analyses were performed in FSL5 ( Smith et al . , 2004 ) . Representational similarity analyses were performed using CoSMoMVPA ( http://www . cosmomvpa . org/; Oosterhof et al . , 2016 ) . Images were initially skull-stripped using a brain extraction tool ( BET , Smith , 2002 ) to remove non-brain tissue from the image . Data were then corrected for slice-acquisition time , high-pass temporally filtered ( using a 50-s period cut-off for event-related runs , and a 128 s period cut-off for the blocked localizer run ) , and motion corrected ( MCFLIRT , Jenkinson et al . , 2002 ) . Functional runs were registered to each participant’s high-resolution MPRAGE image using FLIRT boundary-based registration with B0-fieldmap correction . The resulting unsmoothed data were analyzed using first-level FEAT ( v6 . 00; fsl . fmrib . ox . ac . uk/fsl/fslwiki ) in each participant’s native anatomical space . Parameter estimates of BOLD response amplitude were computed using FILM , with a general linear model that included temporal autocorrelation correction and six motion parameters as nuisance covariates . Each trial ( i . e . object concept ) was modeled with a delta function corresponding to the stimulus presentation onset and then convolved with a double-gamma hemodynamic response function . Separate response-amplitude ( β ) images were created for each object concept ( n = 40 ) , in each run ( n = 8 ) , in each property verification task context ( n = 2 ) . Obtained β images were converted into t-statistic maps; previous research has demonstrated a modest advantage for t-maps over β images in the context of multi-voxel pattern analysis ( Misaki et al . , 2010 ) . In a final step , we created mean object-specific t-maps by averaging across runs . These data were used for all subsequent similarity analyses .	Our ability to interact with the world depends in large part on our understanding of objects . But objects that look similar , such as a hairdryer and a gun , may do different things , while objects that look different , such as tape and glue , may have similar roles . The fact that we can effortlessly distinguish between such objects suggests that the brain combines information about an object’s visual and abstract properties . Nevertheless , brain imaging experiments show that thinking about what an object looks like activates different brain regions to thinking about abstract knowledge . For example , thinking about an object’s appearance activates areas that support vision , whereas thinking about how to use that object activates regions that control movement . So how does the brain combine these different kinds of information ? Martin et al . asked healthy volunteers to answer questions about objects while lying inside a brain scanner . Questions about appearance ( such as “is a hairdryer angular ? ” ) activated different regions of the brain to questions about abstract knowledge ( “is a hairdryer manmade ? ” ) . But both types of question also activated a region of the brain called the perirhinal cortex . When volunteers responded to either type of question , the activity in their perirhinal cortex signaled both the physical appearance of the object as well as its abstract properties , even though both types of information were not necessary for the task . This suggests that information in the perirhinal cortex reflects combinations of multiple features of objects . These findings provide insights into a neurodegenerative disorder called semantic dementia . Patients with semantic dementia lose their general knowledge about the world . This leads to difficulties interacting with everyday objects . Patients may try to use a fork to comb their hair , for example . Notably , the perirhinal cortex is a brain region that is usually damaged in semantic dementia . Loss of combined information about the visual and abstract properties of objects may lie at the core of the observed impairments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Integrative and distinctive coding of visual and conceptual object features in the ventral visual stream
Kinase inhibitors are effective cancer therapies , but tumors frequently develop resistance . Current strategies to circumvent resistance target the same or parallel pathways . We report here that targeting a completely different process , autophagy , can overcome multiple BRAF inhibitor resistance mechanisms in brain tumors . BRAFV600Emutations occur in many pediatric brain tumors . We previously reported that these tumors are autophagy-dependent and a patient was successfully treated with the autophagy inhibitor chloroquine after failure of the BRAFV600E inhibitor vemurafenib , suggesting autophagy inhibition overcame the kinase inhibitor resistance . We tested this hypothesis in vemurafenib-resistant brain tumors . Genetic and pharmacological autophagy inhibition overcame molecularly distinct resistance mechanisms , inhibited tumor cell growth , and increased cell death . Patients with resistance had favorable clinical responses when chloroquine was added to vemurafenib . This provides a fundamentally different strategy to circumvent multiple mechanisms of kinase inhibitor resistance that could be rapidly tested in clinical trials in patients with BRAFV600E brain tumors . Signaling pathway-targeted therapies in cancer are greatly hampered by our inability to counteract the development of resistance . The RAF/MEK/ERK pathway is important in central nervous system tumors ( Gierke et al . , 2016; Mistry et al . , 2015 ) , and with BRAFV600E mutations in more than 50% of select tumors ( Penman et al . , 2015 ) there is great potential for the use of BRAFV600E inhibitors . Indeed , the first pediatric patient successfully treated with vemurafenib ( Rush et al . , 2013 ) was followed by similar case reports in brain tumor patients of all ages ( Bautista et al . , 2014; Skrypek et al . , 2014 ) , and clinical trials in children and adolescents are ongoing using both vemurafenib ( NCT01748149 ) and dabrafenib ( NCT01677741 ) . The initial excitement for BRAF inhibitors ( BRAFi ) in other tumors was tempered because the majority of patients who initially respond to RAF inhibition quickly develop resistance to therapy ( Hartsough et al . , 2014; Sun et al . , 2014 ) . This is a significant issue in brain tumors as well ( Levy et al . , 2014; Yao et al . , 2015 ) . There are multiple routes of acquired resistance to RAF inhibition ( Sun et al . , 2014; Rizos et al . , 2014 ) and circumventing these mechanisms usually involves either targeting the same pathway a different way or targeting a similar parallel pathway . A recent study of BRAFi resistance in colorectal cancer highlighted difficulties with this approach with a single tumor often harboring more than one mechanism of resistance . More importantly , when tumors became resistant to one combination of drugs , such as BRAF/MEK inhibition , there was cross-resistance to others such as BRAF/EGFR inhibition ( Ahronian et al . , 2015 ) . This concept is playing out in clinical trials as well . BRAF and MEK inhibition in BRAFV600E melanoma patients found a small increase in median progression free survival but failed after a short time . Further evidence found that patients who were treated with MEKi after they had developed BRAFi resistance had no objective clinical responses ( Kim et al . , 2013 ) . EGFR is another potential secondary target in melanoma , brain , and colorectal cancer . Although encouraging preclinical results have been obtained in these tumors ( Yao et al . , 2015; Corcoran et al . , 2012; Girotti et al . , 2013 ) , combined BRAF/EGFR inhibition similarly leads to incomplete and short-term responses in people ( Ahronian et al . , 2015; Pietrantonio et al . , 2016 ) . Autophagy inhibition is a potential method to reverse BRAFi resistance . Previous studies of kinase inhibitor resistance in adult BRAFWT gliomas with PTEN mutations resistant to phosphatidylinositol 3-kinase to AKT to mammalian target of rapamycin ( PI3K-AKT-mTOR ) pathway inhibitors found that autophagy inhibition improved response to dual PI3K-mTOR inhibitors ( Fan et al . , 2010 ) . Up-regulation of endoplasmic reticulum ( ER ) stress-induced autophagy after treatment with BRAFi has been shown in melanoma tumor biopsies and associated with the development of resistance to vemurafenib . Autophagy inhibition overcame the resistance through this mechanism in melanoma cell lines ( Ma et al . , 2014 ) . Previously , we reversed clinical and radiographic disease progression with the addition of the autophagy inhibitor chloroquine ( CQ ) in a patient with a BRAFV600E brainstem ganglioglioma who progressed while on vemurafenib ( Levy et al . , 2014 ) . This patient continued to experience disease regression on the combination of CQ plus vemurafenib for more than two and a half years , contrasting dramatically with her original response to vemurafenib that failed at 11 months ( Levy et al . , 2014 ) . These findings led us to hypothesize that autophagy inhibition provides a different way to circumvent BRAF inhibitor resistance in CNS tumors that avoids targeting the same or similar kinase pathways and might apply to multiple different mechanisms of kinase inhibitor resistance . Isogenic BRAFi resistant brain tumor cell lines ( 794R and AM38R ) were developed through chronic exposure to vemurafenib ( Figure 1A and quantification Figure 1B ) . Parental cells ( 794 and AM38 ) demonstrated a stable reduction the in ratio of pERK to ERK when treated with vemurafenib . In contrast , resistant cells recovered pERK:ERK ratios to almost baseline levels by 24 hr of drug exposure ( Figure 1C and quantification of pERK:ERK ratios Figure 1D ) . Unlike reports in vemurafenib-resistant melanoma cell lines ( Ma et al . , 2014 ) , neither basal nor drug-induced autophagy was increased in resistant cells ( 794R , AM38R , B76 ) compared to parental/sensitive cells ( 794 , AM38 , BT40 ) as determined by flow cytometry ( Figure 2A and B ) or by Western blot ( Figure 2C ) . Quantification of autophagic flux measured by Western blot ( Figure 2D ) demonstrated that 794 and 794R cells have a similar flux at all timepoints , while AM38R had a smaller accumulation of LC3II over six hours compared to AM38 parental cells . This may be in part to the AM38 parental cells that demonstrated a lower level of LC3II at baseline compared to AM38R cells . As flux is measured by comparison of all time-points to the time 0 baseline , the higher level of LC3II in AM38R cells at time 0 would reduce the final flux measurement at six hours . Taken together , these data demonstrated that development of resistance to BRAFi did not increase levels of autophagy in these cells . Additionally , autophagic flux was successfully blocked in both parental and resistant cells using 5 μM CQ ( Figure 2E ) , a dose that can be achieved clinically ( Augustijns et al . , 1992 ) . 10 . 7554/eLife . 19671 . 003Figure 1 . Brain tumor cell lines develop resistance to pharmacologic inhibition of BRAFV600E . ( A ) Comparison of parental ( P ) and isogenic resistant ( R ) cell line response long-term growth following BRAFi for 72 hr . Representative image shown . ( B ) Quantification of clonogenic growth shown in A . Two way ANOVA; mean ± s . e . m , n = 3 . p<0 . 05 . ( C ) Representative Western blot demonstrating decreased pERK suppression in resistant cells compared to parental cells following BRAFi . ( D ) Quantification of pERK:ERK ratios shown in C . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 00310 . 7554/eLife . 19671 . 004Figure 1—source data 1 . Quantification of long-term clonogenic growth assays in 794 and AM38 parental and resistant cells treated with increasing doses of vemurafenib . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 00410 . 7554/eLife . 19671 . 005Figure 2 . Parental and resistant BRAFV600E CNS tumor cells have similar levels of autophagy . ( A ) Representative histogram comparison of parental and resistant cell line autophagy . Cells with mCh-GFP-LC3 were exposed to either standard media or starvation EBSS media for 4 hr and analyzed by flow cytometry for the change in ratio of mCh to GFP signal as a measure of autophagic flux . ( B ) Quantification of basal and induced autophagy as measured in A ( mean ± s . e . m , n = 3 ) . There was no significant increase of autophagic flux in resistant over parental cell lines . ( C ) Representative westerns and ( D ) quantification of samples showing accumulation of LC3II in the presence of CQ as a measure of autophagic flux ( mean ± s . e . m , n = 3 ) . There was no significant increase of autophagic flux in resistant over parental cell lines . ( E ) Western blot showing inhibition of autophagy with IC50 CQ dose . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 00510 . 7554/eLife . 19671 . 006Figure 2—source data 1 . Quantification of autophagic flux by ( A ) Flow cytometry ) and ( D ) Western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 006 Resistant cells retained a dose dependent sensitivity to pharmacologic autophagy inhibition equal to their parental controls in long-term growth assays and approximately 50% growth inhibition was achieved using 5 μM CQ ( Figure 3A ) . Long-term growth assays demonstrated that the parental cells responded to both vemurafenib and CQ alone , and the combination resulted in an even greater reduction in cell growth as we previously reported ( Levy et al . , 2014 ) . In comparison , resistant cells showed little or no response to vemurafenib alone , but cell growth was dramatically reduced with the addition of CQ ( Figure 3B and C ) , indicating a synergistic effect of combined BRAF and autophagy inhibition in both the BRAFi-sensitive cells and their resistant derivatives . Calculated combination index ( CI ) values for these combinations confirmed a synergistic interaction between these drugs irrespective of whether the cells had become resistant to single agent BRAFi or not ( Table 1 ) . 10 . 7554/eLife . 19671 . 007Figure 3 . Pharmacologic inhibition of autophagy overcomes BRAFi resistance . ( A ) Long-term growth assay of parental and resistant cells in response to continuous autophagy inhibition ( mean ± s . e . m , n = 3 ) . ( B ) Representative and ( C ) quantified long-term growth of parental and resistant cells following continuous autophagy inhibition ( CQ ) , BRAF inhibition ( Vem ) , or combination therapy . Two way ANOVA; mean ± s . e . m , n = 3 . p<0 . 05 , # p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 00710 . 7554/eLife . 19671 . 008Figure 3—source data 1 . Quantification of long-term clonogenic growth assays in 794 and AM38 parental and resistant cells treated with ( A ) increasing doses of CQ and ( B–C ) vemurafenib , CQ , or a combination of the two drugs . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 00810 . 7554/eLife . 19671 . 009Table 1 . Combination index values for long-term growth assays in parental and resistant cells . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 009Cell line Vemurafenib 1 μM + CQ 5 μM 7940 . 41794R 0 . 74AM380 . 15AM38R0 . 85R= drug induced resistance; Value > 1 antagonistic , =1 additive , <1 synergistic . Clinical evidence has suggested that cells with BRAFi resistance sometimes develop cross-resistance to other inhibitors of this pathway , specifically MEK inhibition ( Kim et al . , 2013 ) . To test this in our resistant cells , we evaluated MEK inhibition with trametinib , which inhibits MEK1 and MEK2 . 794R cells demonstrated no cross-resistance shown by a significant decrease in cell growth similar to parental 794 cells treated with trametinib . Conversely , while AM38 parental cells treated with trametinib demonstrated a significant reduction in growth , AM38R cells had a minimal decrease in growth rates ( Figure 4A ) indicating that for AM38 cells , but not 794 cells , the acquisition of resistance to vemurafenib was associated with cross-resistance to MEK inhibition . However , as with the BRAFi , when autophagy inhibition with CQ was combined with trametinib in AM38R cells , a further decrease in cell growth was demonstrated in both short and long-term assays ( Figure 4B–D ) . These findings suggest that although the two cell lines had become resistant through different mechanisms , autophagy inhibition works similarly to reverse kinase inhibitor resistance in both cases . 10 . 7554/eLife . 19671 . 010Figure 4 . Autophagy inhibition overcomes cross-resistance to MEKi . ( A ) Percent growth over time in BRAFi resistant cell lines treated with 5 nM trametinib , a MEK inhibitor ( Tram ) . Growth measured by continuous Incucyte monitoring ( Two way ANOVA , mean ± s . e . m . , n = 3 , # p<0 . 001 ) . ( B ) Percent growth over time in AM38R cells treated with MEK inhibition ( Tram ) , autophagy inhibition ( CQ ) , or combination therapy . Growth measured by continuous Incucyte monitoring ( Two way ANOVA , mean ± s . e . m . , n = 3 , # p<0 . 001 ) . ( C ) Representative and ( D ) quantified long-term growth assay of parental and BRAFi resistant cells following continuous MEK inhibition ( Tram ) , autophagy inhibition ( CQ 5 μM ) , or combination therapy . One way ANOVA; mean ± s . e . m , n = 3 . p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 01010 . 7554/eLife . 19671 . 011Figure 4—source data 1 . Quantification of ( A ) % growth over time for 794 and AM38 parental and vemurafenib resistant cells treated with trametinib alone or ( B ) cells treated with trametinib , CQ or a combination of the two drugs . ( D ) Quantification of clonogenic growth assays of 794 and AM38 parental and vemurafenib resistant cells treated with either increasing doses of trametinib , or with trametinib , CQ or a combination of the two drugs . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 011 Autophagy-independent chemosensitization by CQ has been demonstrated previously ( Eng et al . , 2016; Maycotte et al . , 2012 ) . Therefore , genetic inhibition of autophagy was also performed to test if the responses seen were related to inhibition of the autophagic pathway or were due to another effect of CQ . Knockdowns of either ATG5 or ATG7 ( two essential regulators of canonical autophagy ) had a profound effect on the growth of both parental and resistant cell lines with a dramatic decrease in growth velocity . However , under conditions where there was measureable growth in the presence of RNAi , the addition of vemurafenib to resistant cells along with ATG5 or ATG7 knockdown resulted in a further decrease in growth rate , under conditions where there was measurable growth in the presence of RNAi . ( Figure 5A–B ) . Confirmation that ATG5 and ATG7 knockdown inhibited autophagy was shown by a decrease in the autophagy marker LC3II ( Figure 5C–D ) . Combination therapy with RNAi and vemurafenib also resulted in a decreased number of viable cells compared to pharmacologic BRAF inhibition or genetic autophagy inhibition alone ( Figure 5E–F ) . These data suggest that re-sensitization to the BRAFV600E inhibitor was due to autophagy inhibition and not another effect of CQ . More importantly , both the trametinib sensitive 794R and trametinib resistant AM38R cells responded equally well to genetic interference with autophagy . 10 . 7554/eLife . 19671 . 012Figure 5 . Genetic inhibition of autophagy overcomes BRAFi resistance . ( A–B ) Percent growth over time in resistant cell lines with control non-targeted ( NT ) RNAi compared to autophagy inhibition through RNAi against ( A ) ATG5 or ( B ) ATG7 , required autophagy proteins . Growth measured by continuous Incucyte monitoring ( mean ± s . e . m ( n = 3 ) ( C–D ) Representative westerns showing effectiveness of ( C ) ATG5 and ( D ) ATG7 RNAi and resultant decrease of LC3II . ( E–F ) Percent viable cells , by Cell Titer-Glo ( compared to control NT ) following 72 hr of vemurafenib ( Vem ) drug therapy with and without RNAi of essential autophagy proteins ATG5 and ATG7 . One way ANOVA; mean ± s . e . m ( n = 3 ) . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 01210 . 7554/eLife . 19671 . 013Figure 5—source data 1 . Incucyte timecourse and endpoint survival data . ( A and B ) Quantification of % growth over time for 794R and AM38R cells treated with RNAi to ATG5 #1 , ATG5#2 , ATG7#1 and ATG7#2 with and without vemurafenib . ( E ) Percent viable cells following RNAi to ATG5 #1 , ATG5#2 , ATG7#1 and ATG7#2 with and without vemurafenib . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 013 To test if autophagy inhibition can overcome drug resistance acquired during clinical treatment , we tested slice cultures of BRAFV600E and BRAFWT CNS tumors from patients who in some cases had been treated with vemurafenib . This strategy allows monitoring of treatment effects in tumor cells with a supporting microenvironment ex vivo . Patient #1 was diagnosed with epithelioid glioblastoma and had rapid recurrence following standard therapy including temazolamide and radiation . At relapse , a BRAFV600E mutation was identified , and there was a successful tumor control on vemurafenib for approximately 2 years . While on vemurafenib , a rapidly growing metastatic second recurrence developed . Following confirmation that the recurrent disease continued to harbor BRAFV600E , the largest lesions were treated with focal radiation while the metastatic disease required a chemotherapeutic approach . Tissue slices were collected at a biopsy of recurrence and evaluated for protein level changes , LDH release , and EdU incorporation following BRAFi with vemurafenib and autophagy inhibition with CQ . Western blotting demonstrated up-regulation of the pERK pathway following treatment with vemurafenib ( Figure 6A ) , consistent with known resistance mechanisms that cause paradoxical activation of the pathway through other RAF isoforms or amplification of COT ( Hatzivassiliou et al . , 2010; Johannessen et al . , 2010; Luke and Hodi , 2012 ) , but notably , also mechanistically different than when either of the BRAFV600E cell lines acquired resistance ( Figure 1C–D ) . CQ treatment caused accumulation of LC3II within the tumor slices compared to DMSO controls , indicating successful autophagy inhibition ( Figure 6A ) . Combination treatment with autophagy and BRAF inhibition resulted in significantly greater cytotoxicity than vemurafenib or CQ treatment alone as measured by LDH release ( Figure 6B ) . This was associated with reduced tumor cell growth measured by EdU incorporation ( Figure 6C ) . 10 . 7554/eLife . 19671 . 014Figure 6 . Autophagy inhibition improves clinically acquired BRAFi resistance . ( A ) Ex vivo slice culture of Patient #1 tumor showing up-regulation of pERK:ERK and inhibition of autophagic flux as indicated by LC3II accumulation by Western blot with quantification of triplicate samples , mean ± s . e . m , n = 3 . ( B ) Cumulative LDH release and ( C ) EdU incorporation as a measure of cytotoxicity and decreased cell proliferation in Patient #1 treated slice culture samples; One way ANOVA; mean ± s . e . m . p<0 . 05 . ( D ) In vitro cell line derived from Patient #1 showing retained response to pharmacologic inhibition of autophagy with decreasing viability and contrasting increase in LDH release with increasing doses of chloroquine ( CQ ) . ( E ) LDH release as a measure of cytotoxicity in Patient #1 cell line treated for 72 hr as indicated; vemurafenib ( Vem ) at 1 or 2 μM , CQ at 10 or 20 μM; Unpaired two-tailed Student’s t-test; mean ± s . e . m , n = 3 . ( F ) Long-term growth assay of Patient #1 cell line following autophagy inhibition ( CQ ) , BRAFi ( Vem ) , or combination therapy . Quantified collated data for triplicate experiments . Unpaired two-tailed Student’s t-test; mean ± s . e . m , n = 3 . p<0 . 05 . ( G ) Western blot analysis of pAKT , AKT , pS6 , pMEK , MEK , pERK , ERK , LC3I and LC3II in Patient #1 slice culture samples . Actin included as loading control . ( H ) Quantification of slice culture samples showing accumulation of LC3II in the presence of CQ as a measure of autophagic flux ( mean ± s . e . m , n = 3 ) . There was no significant difference of autophagic flux between the treatment groups . ( I ) Quantified densitometry ratios of phosphorylated proteins to total proteins shown in ( G ) for AKT , MEK , and ERK . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 01410 . 7554/eLife . 19671 . 015Figure 6—source data 1 . Western quantifications , LDH and survival data . ( A ) Densitometry quantification of Western blotting of slice culture samples . ( B ) Normalized LDH measures of Patient #1 slice culture samples . ( C ) EdU quantification by flow cytometry of Patient #1 slice culture samples . ( D ) LDH and cell viability of Patient #1 cell line treated with increasing doses of CQ . ( F ) Quantification of long-term clonogenic growth assays in Patient #1 cell line treated with vemurafenib , CQ or a combination of the two drugs . ( H ) Quantification of autophagy flux in Patient #1 slice culture samples . ( I ) Quantification of phosphorylated to total protein for AKT , MEK and ERK in Patient #1 slice culture samples . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 01510 . 7554/eLife . 19671 . 016Figure 6—figure supplement 1 . Caspase 3/7 activation occurs in the presence of BRAF and autophagy inhibition in cells with acquired BRAFi resistance . Patient #1 primary cells were treated with BRAFi ( Vem 1 μM ) , autophagy inhibition ( CQ 10 μM ) or a combination of the two drugs . Cell growth and caspase 3/7 activation was monitored every four hours using Incucyte monitoring for 72 hr . The area under the curve ( AUC ) for caspase 3/7 activation was normalized to the AUC of cell numbers over time ( mean ± s . e . m . , n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 016 A primary in vitro cell culture developed from Patient #1 also demonstrated sensitivity to pharmacologic inhibition of autophagy ( Figure 6D ) , consistent with our data in established cell lines ( Levy et al . , 2014 ) and suggesting that acquisition of resistance during clinical treatment with the BRAF inhibitor did not alter autophagy-dependency . Moreover , combination therapy demonstrated significantly higher cytotoxicity ( Figure 6E ) that was associated with an increase in caspase 3/7 activity in the combination treated cells ( Figure 6—figure supplement 1 ) . In a long-term growth assay , combination therapy also resulted in a reduction of cell growth compared to single drug treatments ( Figure 6F ) . Further pathway analysis of Patient #1 treated slice cultures was performed to evaluate the association between autophagy inhibition and the AKT/mTOR signaling pathway . Treatment with vemurafenib resulted in small decrease in pAKT and pS6 with an associated small increase in LC3II ( Figure 6A and G ) but no significant increase of autophagic flux in cells treated with vemurafenib ( Figure 6H ) . This is consistent with our previously published data showing vemurafenib did not have a significant effect on the autophagic flux in parental BRAFV600E cell lines ( Levy et al . , 2014 ) . As has also been previously reported ( Spears et al . , 2016 ) , treatment with CQ resulted in increase phosphorylation of AKT . Combination therapy resulted in pAKT levels similar to DMSO control . Phospho MEK expression was also increased in all samples treated with vemurafenib , although the pMEK:MEK ratio was lower in samples treated in combination with CQ , primarily due to an increase in total MEK in these samples ( Figure 6G and I ) . Importantly , in all treatments with vemurafenib , regardless of the presence of CQ , pERK was upregulated compared to DMSO control ( Figure 6A , G and I ) . Together , these data suggest that patient #1’s tumor acquired resistance to vemurafenib through a mechanism leading to ‘paradoxical’ RAF pathway activation by the drug , but that subsequent combination treatment with autophagy inhibition and vemurafenib could overcome this resistance . An evaluation of additional patient samples with verified BRAF mutations allowed assessment of the effectiveness of this approach in other brain tumor types . Combination treatment of ex vivo tumor from Patient #2 with a BRAFV600E positive pleomorphic xanthoastrocytoma resulted in significantly greater LDH release than vemurafenib treatment alone ( Figure 7A ) . In contrast , two patients ( Patients #3 and #4 ) with WT BRAF exhibited no increase in LDH release in any of the treatment conditions ( Figure 7B ) showing that as with our previous studies in cell lines , primary tumor samples with WT BRAF display no significant sensitivity to autophagy inhibition ( Levy et al . , 2014; Levy and Thorburn , 2012 ) . In Patient #2 , treatment with vemurafenib resulted in an increase in Edu incorporation while a reduction of Edu incorporation was seen in combination treated cells ( Figure 7C ) . An additional primary culture from Patient #5 , an adult with a BRAFV600E positive glioblastoma who had not received vemurafenib therapy , demonstrated inherent BRAFi resistance . However , as with Patients #1 and #2 , sensitivity to pharmacologic inhibition of autophagy was seen ( Figure 7D ) . Moreover , combination therapy with both vermurafenib and CQ again demonstrated significantly higher cytotoxicity compared to single drug treatment ( Figure 7E ) , and was associated with an increase in caspase 3/7 activation ( Figure 7—figure supplement 1 ) . Both short and long term growth assays demonstrated decreased cell growth when autophagy was inhibited with or without vemurafenib ( Figure 7F and G ) . Increased tumor growth was not seen in Patient #5 vemurafenib treated cells . The increase in Edu incorporation seen in Patient #2 was not related to paradoxical RAF pathway activation , as was seen in Patient #1 ( Figure 6A ) , although there was also not a substantial reduction in pERK overall . In contrast , Patient #5’ tumor cells did demonstrate a reduction in pERK signaling with exposure to vemurafenib ( Figure 7H–I ) . Of note across all these patients , pERK status did not correlate with resistance to vemurafenib . Rather , inhibition of autophagy resulted in re-sensitization in all three primary patient samples . Patient #5’s tumor contained additional mutations in PTEN as well as a TP53 mutation ( Table 2 ) . Mutations in PTEN are known to confer BRAFi resistance ( Paraiso et al . , 2011 ) , which could explain the inherent resistance found in this tumor . An evaluation of PTEN downstream effectors in Patient #5 cells found no significant effect of autophagy inhibition on pAKT , p21 or pS6 ( Figure 7J ) . 10 . 7554/eLife . 19671 . 017Figure 7 . Autophagy inhibition is effective in a variety of BRAFV600E tumor models . ( A ) Slice culture evaluation showing cytotoxicity as measured by LDH release in Patient #2 V600E mutant tumor . ( B ) No significant cytotoxicity as measured by LDH release is seen in Patients #3 and #4 with wild type ( WT ) BRAF tumors . ( C ) Decrease in EdU incorporation in Patient #2 V600E mutant tumor with combination BRAF ( Vem ) and autophagy ( CQ ) inhibition . ( D ) Cell line derived from Patient #5 with V600E mutant tumor showing retained response to pharmacologic inhibition of autophagy with decreasing viability and contrasting increase in LDH release with increasing doses of CQ . ( E ) LDH release in Patient #5 V600E mutant tumor cells treated with vemurafenib ( Vem ) at 1 or 2 μM , CQ at 10 or 20 μM autophagy inhibition ( CQ ) , BRAFi ( Vem ) , or combination therapy for 72 hr as indicated . Unpaired two-tailed Student’s t-test; mean ± s . e . m , n = 3 . p<0 . 05 . ( F ) Short-term ( five day ) growth assay demonstrating percent growth of Patient #5 cell line following autophagy inhibition ( CQ ) , BRAFi ( Vem ) , or combination therapy . ( G ) Representative long-term ( fourteen day ) clonogenic assay and quantified collated data for cells treated with combination drug therapy as indicated; Vem at 1 or 2 μM , CQ at 10 μM; Unpaired two-tailed Student’s t-test; mean ± s . e . m . # p<0 . 001 , n = 3 . ( H ) Representative Western blot and ( I ) quantification demonstrating pERK response in resistant primary patient samples following BRAFi ( Vem ) . ( J ) Western blot of PTEN downstream effectors in Patient #5 V600E mutant tumor cells , known to carry a PTEN mutation . No significant protein changes with BRAFi ( Vem ) , autophagy inhibition ( CQ ) , or combination therapy . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 01710 . 7554/eLife . 19671 . 018Figure 7—source data 1 . Western quantifications , LDH and survival data . ( A ) Normalized LDH release of Patient #2 slice culture samples . ( C ) EdU quantification by flow cytometry of slice culture samples . ( D ) LDH and cell viability of Patient #5 cell line treated with increasing doses of CQ . ( E ) Normalized LDH release of Patient #5 cell line treated with vemurafenib , CQ , or a combination of the two drugs . ( G ) Quantification of long-term clonogenic growth assays in Patient #5 cell line treated with vemurafenib , CQ , or a combination of the two drugs . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 01810 . 7554/eLife . 19671 . 019Figure 7—figure supplement 1 . Caspase 3/7 activation occurs in the presence of BRAF and autophagy inhibition in cells with inherent BRAFi resistance . Patient #5 primary cells were treated with BRAFi ( Vem 1 μM ) , autophagy inhibition ( CQ 10 μM ) or a combination of the two drugs . Cell growth and caspase 3/7 activation was monitored every four hours using Incucyte monitoring for 72 hr . The area under the curve ( AUC ) for caspase 3/7 activation was normalized to the AUC of cell numbers over time . mean ± s . e . m . , n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 01910 . 7554/eLife . 19671 . 020Table 2 . Mutation analysis of studied samples . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 020Sample BRAF status Additional mutations identified Patient 1 Sensitive BRAF c . 1799T>A; p . V600ENone Patient 1 Resistant BRAF c . 1799T>A; p . V600ENone Patient 5 Resistant BRAF c . 1799T>A; p . V600EPTEN c . 74T>C; p . L25STP53 c . 743G>A; p . R248Q794 BRAF c . 1799T>A; p . V600ENone 794R BRAF c . 1799T>A; p . V600ENone AM38 BRAF c . 1799T>A; p . V600ENone AM38R BRAF c . 1799T>A; p . V600ENone Multiple BRAFi resistance mechanisms have been described . This includes mutations in PTEN as shown above , as well as RAS mutations and activation of receptor tyrosine kinase signaling ( Luke and Hodi , 2012 ) . Feedback activation of EGFR has specifically been suggested as an escape pathway in BRAFV600E CNS tumor cells ( Yao et al . , 2015 ) . The above data suggest that autophagy inhibition is able to overcome BRAFi resistance in different tumor types , and that distinct mechanisms of resistance can be similarly targeted by this approach . To test this isogenic BRAF mutant cells carrying specific mutations known to confer vemurafenib resistance through distinct mechanisms were created . RAS activation through both KRAS ( Ahronian et al . , 2015 ) and NRAS ( Luke and Hodi , 2012 ) have been reported to result in BRAFi resistance . Using constitutively active mutants KRASG12V and NRASQ61K , we evaluated induction of resistance and the ability to reverse this resistance by pharmacologic autophagy inhibition with CQ . 794 and AM38 cell lines expressing either KRASWT or a non-target ( NT ) construct retained sensitivity to both increasing doses of vemurafenib and combination therapy with CQ ( Figure 8A–D ) , similar to that in parental cells ( Figure 3B ) . In contrast , both KRASG12V and NRASQ61K cells displayed the expected resistance to increasing doses of vemurafenib alone ( Figure 8A–D ) . Combination vemurafenib and CQ therapy in the RAS resistant lines resulted in a significantly increased response compared to either drug alone ( Figure 8A–D ) . Calculated CI values in RAS driven resistant cells showed synergy between vemurafenib and CQ ( Table 3 ) . Both the KRASG12V and NRASQ61K cells demonstrated increased pERK activity compared to WT and NT controls , indicating the expected up-regulation of the RAF-MEK-ERK pathway ( Figure 8E ) . 10 . 7554/eLife . 19671 . 021Figure 8 . Autophagy inhibition overcomes molecularly distinct mechanisms of BRAFi resistance . ( A ) to ( D ) Representative long-term clonogenic assays ( A and C ) and quantified collated data ( B and D ) for cells treated with combination drug therapy as indicated; Vem with an increasing dose of 1 , 2 or 3 μM , CQ at 5 μM; or a combination of Vem 1 μM and CQ 5 μM . Two way ANOVA; mean ± s . e . m . # p<0 . 001 , n = 3 . ( E ) Representative Westerns showing increased pERK expression in cells with KRASG12V and NRASQ61K compared to NT or KRASWT . ( F ) Percent growth at 140 hr in AM38 ( parental ) , AM38R ( resistant ) and AM38 NRASQ61K ( resistant ) cell lines treated with autophagy inhibition through RNAi against ATG5 , ATG7 or a combination of RNAi and vemurafenib . Growth measured by continuous IncuCyte monitoring . mean ± s . e . m , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 02110 . 7554/eLife . 19671 . 022Figure 8—source data 1 . Long term growth assay quantifications and incucyte timecourse data . ( B and D ) Quantification of long-term clonogenic growth assays in for 794R and AM38R cells with and without inserted mechanisms of resistance treated with increasing doses of vemurafenib and vemurafenib , CQ , or a combination of the two drugs . ( F ) Quantification of % growth over time for AM38 , AM38R and AM38 NRASQ61K cells treated with RNAi to ATG5 #1 , ATG5#2 , ATG7#1 and ATG7#2 with and without vemurafenib . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 02210 . 7554/eLife . 19671 . 023Figure 8—figure supplement 1 . Autophagy inhibition overcomes molecularly distinct mechanisms of BRAFi resistance . ( A–B ) Percent growth over time in AM38 parental , AM38R ( resistant ) and AM38 NRASQ61K ( resistant ) cell lines treated with control non-targeted ( NT ) RNAi , autophagy inhibition through RNAi against ATG5 ( required autophagy protein ) or a combination of RNAi and vemurafenib . Growth measured by continuous IncuCyte monitoring . mean ± s . e . m , n=3 . ( C ) Percent growth over time in AM38 parental , AM38R ( resistant ) and AM38 NRASQ61K ( resistant ) cell lines treated with control non-targeted ( NT ) RNAi , autophagy inhibition through RNAi against ATG7 . Growth measured by continuous IncuCyte monitoring . mean ± s . e . m , n=3 . ( D–E ) Western blot analysis demonstrating level of knockdown in NT , shATG5 and shATG7 treated cells . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 02310 . 7554/eLife . 19671 . 024Figure 8—figure supplement 1—source data 1 . Full image of ATG7 Western with associated actin blot for control to demonstrate shATG5 bands cut out of image . All ATG7 bands shown were run and developed on the same blot . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 02410 . 7554/eLife . 19671 . 025Table 3 . Combination index values for long-term growth assays in RAS driven resistant cells . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 025Cell line Vemurafenib 1 μM + CQ 5 μM 794 KRASV120 . 61794 NRAS61K0 . 73AM38KRASV120 . 54AM38NRAS61K0 . 01R= drug induced resistance; Value > 1 antagonistic , =1 additive , <1 synergistic . Because genetic autophagy inhibition is effective in reducing tumor cell growth alone and when combined with BRAFi ( Figure 5 and [Levy et al . , 2014] ) , we next tested if genetic inhibition of autophagy had a similar effect when a specific , molecularly-defined resistance mechanism was modeled . ATG5 inhibition had a profound effect on cell growth in ( sensitive ) AM38 cells as well as ( resistant ) AM38R and AM38 NRASQ61K cells . Due to the profound reduction in growth of cells with ATG knockdown alone ( Figure 8—figure supplement 1 A and B ) , endpoint growth percentages for all shATGs utilized were compared ( Figure 8F ) . Where cells were able to grow with shATG5 , the addition of vemurafenib resulted in a further reduction in growth ( Figure 8F and Figure 8—figure supplement 1A and B ) . This was also seen when a separate autophagy regulator , ATG7 , was silenced ( Figure 8F and Figure 8—figure supplement 1C ) . Moreover , efficient knockdown of ATG7 alone resulted in near immeasurable growth ( Figure 8—figure supplement 1C ) such that additional growth inhibition with the addition of vemurafenib was difficult to detect . Representative Westerns demonstrate shRNA silencing of ATG5 and ATG7 ( Figure 8—figure supplement D and E ) . Feedback activation of EGFR , represents another mechanistically-distinct resistance mechanism that can provide an escape pathway in BRAFV600E CNS tumor cells ( Yao et al . , 2015 ) . Therefore , we also developed cell lines with EGFR overexpression ( EGFRoe ) and evaluated their response to autophagy inhibition . Compared to parental cells , 794 and AM38 with EGFRoe demonstrated a faster growth velocity and reduced response to vemurafenib ( Figure 9A ) . When growth of EGFRoe cells was assessed for response to single drug and combination therapy , combination therapy resulted in a significantly slower growth velocity ( Figure 9B ) . Representative end-point images demonstrate the reduced number of cells seen with combination therapy ( Figure 9—figure supplement 1 ) . Analysis of the percent of viable cells also demonstrated a significant decrease in combination therapy compared to pharmacologic BRAF or autophagy inhibition alone ( Figure 9C ) . Western blotting analysis verified the increased pEGFR expression in EGFRoe cells ( Figure 9D ) . It is noted that EGFR was not significantly overexpressed in our drug induced polyclonal 794R or AM38R cells . This would suggest that EGFRoe was not the main driving resistance mechanism in those cells , although as a polyclonal population the resistant cells may contain multiple resistance mechanisms and a subpopulation may contain EGFRoe . 10 . 7554/eLife . 19671 . 026Figure 9 . Autophagy inhibition overcomes BRAFi resistance due to escape through EGFR . ( A ) Percent confluence over time in parental and EGFR overexpression ( EGFRoe ) cell lines treated with BRAFi ( Vem 1 μM ) , autophagy inhibition ( CQ 20 μM ) or a combination of the two . Growth measured by continuous IncuCyte monitoring ( mean ± s . e . m . , n = 3 ) . ( B ) Measure of percent viable cells ( compared to media control ) following 4 days of BRAFi ( Vem 1 μM ) , autophagy inhibition ( CQ 20 μM ) or a combination of the two . One way ANOVA; mean ± s . e . m ( n = 6 ) , p<0 . 05 . ( C ) Percent viable cells , by Cell Titer-Glo ( compared to control NT ) following four days of vemurafenib ( Vem ) drug therapy with and without CQ autophagy inhibition in EGFRoe resistant cells . One way ANOVA; mean ± s . e . m ( n = 3 ) . p<0 . 05 . ( D ) Western blot demonstrating EGFR and pEGFR overexpression in 794 and AM38 EGFRoe cells compared to parental and polyclonal resistant isogenic cells . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 02610 . 7554/eLife . 19671 . 027Figure 9—source data 1 . Incucyte timecourse and endpoint survival data . ( A–B ) Quantification of % growth over time for 794 and AM38 parental and EGFRoe cells treated with vemurafenib , CQ or a combination of the two drugs . ( C ) 794 and AM38 EGFRoe percent viable cells treated with vemurafenib , CQ or a combination of the two drugs . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 02710 . 7554/eLife . 19671 . 028Figure 9—figure supplement 1 . Combination BRAF and autophagy inhibition results in fewer cells with EGFR overexpression . Representative phase contrast images showing confluence of cells following four days of therapy as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 028 Based on our encouraging results reversing BRAFi resistance in different cell lines , and specifically in Patient #1’s ex vivo and in vitro samples , Patient #1 was treated following recurrence on vemurafenib alone with vemurafenib at standard dosing plus 250 mg daily of CQ during focal radiation of large primary lesions . Vemurafenib was continued and the CQ dose was increased to 500 mg daily following completion of radiation to treat leptomenengeal metastatic disease sites . The dosing schedule was modeled after previously published reports using CQ for autophagy inhibition in brain tumor patients ( Briceño et al . , 2003; Rojas-Puentes et al . , 2013; Sotelo et al . , 2006 ) . Consistent with the ex vivo and in vitro tumor modeling , addition of CQ to the continued vemurafenib treatment resulted in clinical improvement and decreased growth of metastatic tumor sites as shown by MRI ( Figure 10A–B ) . Robust LC3II accumulation in peripheral white blood cells following CQ therapy ( Figure 10C ) suggests effective autophagy inhibition at the doses of CQ used for this patient . In addition to these easily seen larger lesions , leptomeningeal enhancement throughout the subarachnoid CSF spaces improved with combination therapy as well . In stark contrast with the rapidly growing recurrent metastases , where significant growth was seen after just two weeks at the time of initial relapse , a continued favorable response to the combination therapy was maintained for 7 months . At this time , patient #1 had to stop therapy for unrelated medical issues . 10 . 7554/eLife . 19671 . 029Figure 10 . Autophagy inhibition decreases growth of metastatic glioblastoma in a patient resistant to BRAF inhibition . ( A ) Contrast-enhanced axial T1-weighted MR image demonstrates relapse of tumor over the right precentral gyrus ( Relapse , white box ) while receiving single drug BRAFi therapy . Significant response noted on the 6 month interval MRI following 6 months of combination autophagy and BRAF inhibition ( Response , white box ) . ( B ) Contrast-enhanced sagittal T1-weighted MR image illustrates tumor relapse over the right frontal lobe ( red and white boxes ) . Significant response was noted following 6 months of combination autophagy and BRAF inhibition in both radiated ( Response , red box ) and non-radiated ( Response , white box ) tumor . ( C ) Demonstration of clinical autophagy inhibition as measured by LC3II accumulation in white blood cells from Patient 1 . ( D ) Axial T2-weighted MR image demonstrates a left anterolateral medullary ganglioglioma ( red arrow ) at diagnosis and following 1 year of vemurafenib and vinblastine therapy . ( E ) Axial T2-weighted MR image demonstrates a progressive left anterolateral medullary ganglioglioma ( red arrow ) at relapse and stable tumor following vemurafenib and CQ therapy . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 029 This clinical response suggests that the addition of CQ to inhibit autophagy overcame vemurafenib resistance in this patient . This is consistent with our previously reported patient , who has had a durable clinical and radiographic response for over 2 ½ years with combined vemurafenib and CQ therapy following progression after 11 months of vemurafenib alone . Based on these promising clinical results , a third patient with a BRAFV600E brainstem ganglioglioma received CQ in addition to vemurafenib following clinical and radiologic disease progression on vemurafenib single drug therapy . Patient #6 had a complex medical situation , including extensive multi-focal dural ectasia throughout the CNS , resulting in a severe chronic pain disorder and central hypoventilation requiring non-invasive ventilation while sleeping . This patient developed an ill-defined expansion of the brainstem with a T2 bright mass of the left anterolateral medulla ( Figure 10D ) . A needle biopsy of this lesion was diagnostic for a BRAFV600E ganglioglioma . Initial chemotherapy with a combination of vemurafenib and vinblastine was initiated , as this had previously been shown as a successful combination in this type of tumor by our group ( Rush et al . , 2013 ) . Patient #6 completed one year of therapy with improvement of the brainstem lesion ( Figure 10D ) . It is important to note that evidence of shunt failure was present on this response scan with fourth ventricle enlargement . Due to the continued presence of tumor mass , she was maintained on vemurafenib single drug therapy for an additional six months when she presented with a worsening left facial ( CN VII ) palsy , increased difficulty swallowing , and balance deterioration . Associated with these worsening clinical symptoms on vemurafenib alone , progressive tumor growth was demonstrated on MRI ( Figure 10E ) . CQ ( 500 mg daily ) was added as a second agent and she continued on vemurafenib . In contrast with her response to vemurafenib alone , and consistent with the other patients where combined therapy overcame acquired resistance to vemurafenib , within four weeks of the addition of CQ Patient #6 showed clinical improvement with improved swallowing and CN VII nerve palsy . This is a similar response to our previous brainstem BRAFV600E ganglioglioma patient who also showed a rapid clinical improvement with the addition of CQ therapy ( Levy et al . , 2014 ) . This clinical improvement was maintained for two and a half months when , unfortunately , the patient developed further ventriculoperitoneal shunt failure requiring surgical intervention and her medications were discontinued to allow complete surgical healing . Within three weeks of stopping therapy , her swallowing difficulties recurred . An attempt was made to restart her medications , but complications from her swallowing difficulties resulted in an acute medical decline requiring intubation , and subsequently , she was unable to continue swallowing medications . An MRI at this time demonstrated an unchanged size of the brainstem mass ( Figure 10E ) . Given the patient’s inability to continue oral vemurafenib therapy and her multiple medical complications , the family elected to pursue palliative therapy only . Both Patient #1 and Patient #6 required a treatment interruption to allow for healing from unrelated medical issues . In both patients this resulted in significant clinical deterioration . For Patient #1 , during a six-month treatment interruption , she demonstrated progressive radiographic disease with associated clinical decline and was enrolled in a palliative care program ( Figure 11A ) . Once she had resolution of her additional medical issues , she requested to restart tumor directed therapy . Resistance to BRAF inhibition has been reported to be reversible following a period of treatment interruption in melanoma ( Seghers et al . , 2012 ) therefore Patient #1 was restarted on single agent vemurafenib . However a short interval scan done at four weeks demonstrated rapid tumor progression on vemurafenib alone ( Figure 11B ) indicating that her tumor retained its previously acquired resistance to the BRAF inhibitor . 10 . 7554/eLife . 19671 . 030Figure 11 . Autophagy inhibition decreases growth of metastatic glioblastoma in a patient resistant to BRAF inhibition . ( A ) Contrast-enhanced coronal T1-weighted MR image demonstrates relapse of tumor over the right precentral gyrus ( red arrow ) while on palliative care . ( B ) Contrast-enhanced coronal T1-weighted MR image demonstrates the progression of tumor ( red arrow ) following re-initiation of vemurafenib ( BRAFi ) single agent therapy . ( C ) Contrast-enhanced coronal T1-weighted MR image demonstrates the progression of tumor ( red arrow ) following trametinib ( MEK inhibition ) single agent therapy . ( D ) Contrast-enhanced coronal T1-weighted MR image demonstrates response of tumor with reduction in solid tumor mass ( red arrow ) following re-initiation of combination there with vemurafenib ( BRAFi ) and CQ ( autophagy inhibitor ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19671 . 030 Combination therapy targeting both BRAF and MEK has also been reported to be well tolerated and to result in significant progression-free survival in melanoma ( Flaherty et al . , 2012; Long et al . , 2014 ) . It has also been reported that combination BRAF inhibition and MEK inhibition prevents MAPK pathway reactivation and an improved reduction in pERK in glioma models ( Grossauer et al . , 2016 ) . A trial of combination vemurafenib plus trametinib was therefore initiated based on these data , but this combination was not well tolerated in this patient . Instead she was continued on single agent trametinib . After an initial stabilization lasting four weeks , Patient #1 had further clinical decline with an evolving left hemiparesis and the development of focal seizures . Imaging repeated two months after the addition of MEK inhibition demonstrated significant tumor growth ( Figure 11C ) . In a final attempt to control tumor growth , Patient #1 was restarted on combination vemurafenib and CQ . At an evaluation four weeks after re-initiation of the combination with autophagy inhibition therapy , Patient #1 reported an overall improvement . She had regained partial use of her left side including the ability to open/close her left hand and move her left arm at the elbow . She had also regained the ability to walk unassisted for short distances where she had become wheelchair bound . An MRI at this time demonstrated a measurable reduction in her main solid tumor mass ( Figure 11D ) . This patient therefore demonstrates the potential risks of cross-resistance and resistance to combination therapies targeting the same pathway . But also showed that despite a clearly highly resistant tumor , the addition of autophagy inhibition continued to have a clinical benefit . Targeted therapies such as kinase inhibitors that inhibit tumor-driving mutations are expanding in importance in cancer therapy with the continued identification of mutations across tumor types and the development of many selective inhibitors of these mutant enzymes . But with this potential comes difficulties in identifying which patients to treat and how to predict and counteract the development of resistance . A recent study by Ahronian et al . evaluated resistance in colorectal cancer and highlighted many of the difficulties in combatting resistance to BRAFi ( Ahronian et al . , 2015 ) . They found multiple resistance mechanisms across tumors and often more than one resistance mechanism in a particular patient . More importantly , they found that when cells became resistant to one combination of drugs , there was often cross-resistance to other potential combination therapies . This emphasizes the difficulty of targeting individual resistance mechanisms e . g . by adding an inhibitor of a downstream kinase such as MEK to circumvent resistance to BRAF inhibition . Tumor heterogeneity is also a significant concern with studies showing branched evolution in mutations and resistance mechanisms that can contribute to acquired resistance . Such heterogeneity can occur both temporally as well as geographically within the same patient ( Shi et al . , 2014 ) . The development of a therapeutic strategy that circumvents different molecularly distinct resistance mechanisms could potentially address these issues . Here we present evidence from cell lines with both experimentally acquired and different molecularly defined resistance mechanisms and from tumor samples from patients with clinically acquired and intrinsic BRAFi resistance , that pharmacological inhibition of autophagy with CQ can achieve this goal . In the cell lines where it was experimentally feasible , we could also demonstrate a similar effect with genetic inhibition of autophagy . Broad applicability is suggested because the various resistance mechanisms that could be overcome included examples that did and did not involve cross-resistance to MEK inhibition , mutational activation of both KRAS and NRAS , EGFR overexpression , paradoxical activation of the ERK pathway by vemurafenib , and a PTEN mutation . Taken together , our data suggest that regardless of the resistance mechanism to the BRAF inhibitor , autophagy inhibition was able to improve the response to BRAFi . We have seen rapid clinical responses ( in as little as six weeks ) in patients with both high and low-grade tumors with acquired BRAFi resistance . These first clinical responses have also been sustained . Patient #1 demonstrated control and reduction of her metastatic tumors for greater than seven months , whereas she had previously shown MRI-measurable tumor growth in as little as two weeks . Moreover , patient #1 has experienced clinical benefit when the autophagy inhibitor was added to vemurafenib despite having demonstrated subsequent acquisition of clinical resistance to MEK inhibition after the acquisition of BRAF resistance . Also , the first patient we reported ( Levy et al . , 2014 ) maintained sustained tumor regression on the combination of vemurafenib and CQ for more than 2 ½ years without significant clinical complications . Although these early clinical results are encouraging , our findings are in a limited number of patients and further clinical investigation is required to verify if this strategy of combining autophagy inhibition with BRAF inhibition provides a durable and widely applicable response in BRAFV600E tumors . In summary , pre-clinical and clinical experience invariably shows that tumor cells rapidly evolve ways around inhibition of mutated kinase pathways like the RAF pathway targeted here . However , based on our results , we hypothesize that by targeting an entirely different cellular process , i . e . autophagy , upon which these same tumor cells rely , it may be feasible to overcome such resistance and thus re-establish effective tumor control . Importantly , our data suggest that this strategy can work even when different resistance mechanisms apply . This can be done using CQ , which is an approved , safe , and inexpensive drug and , perhaps , other more potent autophagy inhibitors that are under development ( Egan et al . , 2015; Goodall et al . , 2014; Ronan et al . , 2014 ) . Importantly , in the context of BRAF mutant pediatric brain cancers where BRAF inhibition is already being tested , it should be feasible to quickly test this hypothesis in clinical trials . Experiments were designed to evaluate the hypothesis that autophagy inhibition provides a different way to circumvent BRAF inhibitor resistance in CNS tumors and might apply to multiple different mechanisms of kinase inhibitor resistance . The effect of autophagy inhibition to overcome resistance was initially evaluated in vitro in cell lines and then extended to include ex vivo studies of primary tumor samples . To ensure a complete evaluation of the effects on cell growth and death , both long and short-term growth assays were utilized as well as evaluation of LDH release and EdU incorporation as appropriate . Specificity to the autophagy pathway was evaluated with genetic inhibition studies . Final endpoints were defined prior to the start of each experiment . All in vitro experiments were completed with a minimum of three biologic replicates and where possible with triplicate technical replicates . Due to limitations in slice culture availability ( only primary biopsy samples available for analysis ) , ex vivo tumor experiments were limited to triplicate samples from the same biopsy sample . Details on replicates and statistical analysis are indicated in the figure legends . Primary patient samples were obtained from Children’s Hospital Colorado and collected in accordance with local and Federal human research protection guidelines and institutional review board regulations ( COMIRB 95–500 ) . Informed consent was obtained for all specimens collected . Statistical comparisons were completed using one and two way ANOVA nine and unpaired two-tailed Student’s t-test ( GraphPad Prism 6 . 0 , RRID: SCR_002798 ) as indicated in the figure legends . A P-value of less than 0 . 05 was considered statistically significant . Data shown are mean ± SEM except where indicated . Vemurafenib was obtained from LC Laboratories ( Woburn , MA ) . BT40 cells were derived from a primary patient sample and kindly provided as a gift from Dr . Peter Houghton ( Nationwide Children’s Hospital , Columbus , OH ) . The AM38 ( RRID:CVCL_1070 ) cell line was purchased from the Japan Health Sciences Foundation Health Science Research Resources Bank ( Osaka , Japan ) . 794 , Patient #1 and Patient #5 ( B76 ) cell lines were established from samples obtained during routine surgery at diagnosis or relapse . Cell line authentication was performed using short tandem repeat profiling and comparison with known cell line DNA profiles . Mycoplasma contamination testing was performed using a Lonza MycoAlert Mycoplasma Detection Kit ( Lonza Ltd . , Switzerland ) . Cell lines were maintained in media supplemented with 10–20% fetal bovine serum ( FBS ) ( Gibco , Carlsbad , CA ) , dependent on cell line requirements , and at 37°C in a humidified chamber of 5% CO2 . Cell line authentication was performed using short tandem repeat profiling and comparison to known cell line DNA profiles . Constructs utilized for inducing resistance were purchased from Addgene ( RRID: SCR_002037 ) as follows: pBABE-Puro-KRasG12V was a gift from Christopher Counter ( Addgene plasmid # 46746 ) , pBabe-Kras WT ( Addgene plasmid # 75282 ) and pBabe N-Ras 61K ( Addgene plasmid # 12543 ) was a gift from Channing Der . pBABE-puro human EGFR was constructed using SalI and SnaBI double digestion . Retrovirus particles were produced by cotransfecting GP2-293 cells ( Clontech Laboratories , Mountain View , CA , USA ) pBABE-puro human EGFR and Vesicular Stomatitis G protein ( VSVG ) using TransIT-LT1 ( Mirus ) . For cell line assays , cells were seeded at 1000–4000 cells per well , dependent on optimal conditions per line , in 96-well plates ( Corning , Corning , NY ) , and incubated overnight . Cells were treated with drug doses as indicated . For slice culture samples , media for each sample was collected at treatment days 0 , 2 , 4 , 6 , 8 and 10 . LDH release was quantitated using the Cytoscan-LDH Cytotoxicity Assay Kit ( G-Biosciences , St . Louis , MO ) according to manufacturer’s instructions . For short-term viability assays , cells were seeded at 1000 to 4000 cells , dependent on optimal conditions per line , in 96-well plates ( Corning , Corning , NY ) . RNAi cells were plated 48 hr after knockdown . Cells were treated as indicated . Viable cells were measured using the Cell Titer-Glo luminescent cell viability assay ( Promega , Madison , WI ) following the manufacturer's protocol . All experiments were performed three times in triplicate , and the proportion of cells per treatment group was normalized to control wells . For long-term viability assays , 750 cells were plated in 12-well plates ( Corning , Corning , NY ) and incubated overnight . Cells were treated as indicated . Fresh media or fresh media with drug was provided every three days until control wells had grown to approximately 80% confluence . Cells were fixed and stained using 0 . 4% crystal violet . Stained cells were solubilized in 33% acetic acid and absorbance was read at 540 nm . All experiments were performed three times in triplicate , and the proportion of cells was normalized to control wells . The combination index was calculated by the Chou-Talalay equation , which takes into account both the potency ( IC50 ) and shape of the dose-effect curve ( the M-value ) ( Chou and Talalay , 1984 ) . Combination index values less than 1 , equal to 1 , and more than one indicate synergism , additive effect , and antagonism respectively . A pLKO system ( Sigma-Aldrich , St . Louis , MO ) was utilized for RNAi of autophagy related proteins . TRC numbers for shRNAs used are: ATG7 #1 ( #7587 ) , ATG7 #2 ( #7584 ) , ATG5 #1 ( #151474 ) , ATG5 #2 ( 151963 ) , non-target ( SHC016 ) . Cells were transduced with lentivirus using 8 ug/mL polybrene and selected with the puromycin dose determined appropriate for each cell line . Level of targeted knockdown was determined by Western blot analysis . Cells were seeded at 1000 cells per well in a 96-well plate ( Costar , Corning , NY ) . Cells were cultured at 37° and 5% CO2 and monitored using an IncuCyte Zoom ( Essen BioScience , Ann Arbor , MI ) . Images were captured at 4 hr intervals from four separate regions per well using a 10x objective . Each experiment was done in triplicate and growth curves were created from percent confluence measurements or percent growth based on cell count per well . Cell lysates were harvested after treatments and time-points indicated using RIPA buffer ( Sigma , St . Louis , MO ) with phosphatase inhibitors ( Roche , Indianapolis , IN ) . Membranes were blocked in TBS-Tween 5% milk and probed with primary antibodies at manufacturer recommended concentrations . Primary antibodies used were: ATG 7 ( #8558S , RRID: AB_10831194 ) , ATG5 ( #12994S , RRID: AB_2630393 ) , p44/42 MAP kinase ( phosphorylated Erk1/2 ) ( #9101S , RRID: AB_331646 ) , p44/42 MAPK ( Erk1/2 ) ( #9102 , AM: 330744 ) ; EGFR ( #2232 , RRID: AB_331707 ) , pEGFR ( #4407 RRID: AB_331795 ) , p21 ( #2947 , RRID: AB_823586 ) , pAKT ( #4060 , RRID:AB_2315049 ) , AKT ( #4685 , AB_2225340 ) , pS6 ( #5364 , RRID:AB_10694233 ) ( Cell Signaling , Danvers , MA ) ; LC3 ( #NB100-2220 , RRID:AB_10003146 ) ( Novus Biologicals , Littleton , CO ) ; Anti-β-actin ( #12262 , RRID:AB_2566811 ) Cell Signaling , Danvers , MA ) was used as the protein loading control . MRI images were obtained using standard protocols on a Siemens 1 . 5T Avanto ( Munich , Germany ) scanner . Slice cultures from primary tumor samples were maintained on Millicell Culture Inserts ( Millipore , Billerica , MA ) according to manufacturer’s protocol . Briefly , three approximate 0 . 33 cm slices of primary tumor sample were placed onto a cell culture insert and maintained in specialized slice culture media ( Neurobasal A media containing B27 , glutamax , L-glutamine , HEPES and FGF ) . Slices were treated as indicated , and fresh media with drug , as appropriate , was changed every other day . Eight days after drug treatment EdU was added according to the Click-iT EdU Pacific Blue Flow Cytometry Assay Kit ( Life Technologies , Grand Island , NY ) . On day 10 of treatment , tumor slices were collected for protein ( Western blotting ) and EdU analysis ( flow cytometry ) . Flow data were acquired on a Gallios561 and analyzed using FlowJo . Media was collected on treatment day 0 , 2 , 4 , 6 , 8 , and 10 and stored at −80°C and analyzed together after day 10 for LDH as described above . Assays were performed in triplicate as tissue availability allowed . Cells constitutively expressing mCherry-GFP-LC3 were seeded at 2 × 105 in 60 mm plates and allowed to equilibrate overnight . Cells were exposed to either standard media , Earl’s Based Salt Solution ( EBSS ) starvation media ( Sigma , St . Louis , MO ) or vemurafenib as indicated for evaluation of induced autophagy . Flow data were acquired on a Gallios561 ( Beckman Coulter , RRID: SCR_008940 , Fort Collins , CO ) and analyzed using FlowJo V10 . 0 . 8 , RRID: SCR_008520 . Autophagic flux was determined by the ratio of mCherry:GFP . Library preparation was performed via the Illumina TruSight Tumor kit per the manufacturer’s instructions ( with minor modifications ) using 110–374 ng of DNA derived from frozen or FFPE tissue . This kit amplifies selected regions of 26 cancer-related genes . Libraries were sequenced on the Illumina MiSeq platform for a targeted depth of no less than 500x for any individual amplicon . A custom-built bioinformatics pipeline utilizing GSNAP for sequence alignment and FreeBayes for variant calling was employed for data analysis . All genomic regions were verified to be covered by at least 500 sequencing reads and identified variants were manually inspected using Integrative Genomics Viewer ( Broad Institute ) .	Cancers of the brain and spine are the second most common kind of tumor in children , after cancers of the blood and bone marrow . Yet brain and spine tumors kill more children than any other cancer , in part because many become resistant to treatment . Like in other cancers , cells in brain and spine tumors often use a process called autophagy to survive the treatments that are used to try and kill them . This process allows cells to recycle proteins and other things inside the cell and use them for energy when the cell is under stress . In 2014 , researchers reported that brain tumors carrying a mutation called BRAFV600E rely on autophagy to survive treatment with medications that target this mutation . These findings suggested that blocking autophagy might make the medications more effective against BRAFV600Emutant tumors and overcome the resistance . Now , Mulcahy Levy et al . – who include most of the researchers involved in the 2014 study – report that blocking autophagy does indeed overcome this kind of resistance in multiple types of tumor . The experiments made use of human brain tumor cells that can be grown in the laboratory and have been widely studied , as well as samples collected from patients . Mulcahy Levy et al . were able to block autophagy in the tumor cells by using genetic methods and , importantly , by using an approved and inexpensive drug that could be rapidly translated into clinical trials . Together these findings suggest that blocking autophagy in patients might be a safe and effective strategy to improve their response to existing therapies that target the BRAFV600E mutation . Future clinical trials are now needed to test more patients and verify if this treatment plan can be broadly effective in patients with these types of brain cancers .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2017	Autophagy inhibition overcomes multiple mechanisms of resistance to BRAF inhibition in brain tumors
The segmentation gene network in insects can produce equivalent phenotypic outputs despite differences in upstream regulatory inputs between species . We investigate the mechanistic basis of this phenomenon through a systems-level analysis of the gap gene network in the scuttle fly Megaselia abdita ( Phoridae ) . It combines quantification of gene expression at high spatio-temporal resolution with systematic knock-downs by RNA interference ( RNAi ) . Initiation and dynamics of gap gene expression differ markedly between M . abdita and Drosophila melanogaster , while the output of the system converges to equivalent patterns at the end of the blastoderm stage . Although the qualitative structure of the gap gene network is conserved , there are differences in the strength of regulatory interactions between species . We term such network rewiring ‘quantitative system drift’ . It provides a mechanistic explanation for the developmental hourglass model in the dipteran lineage . Quantitative system drift is likely to be a widespread mechanism for developmental evolution . An important question for evolutionary biology is how developmental processes can compensate for variable environmental , signalling , or regulatory inputs to create a constant phenotypic outcome ( Waddington , 1942 ) . Segment determination during early insect embryogenesis is a well-studied example of this phenomenon . The segmentation gene network produces very robust and conserved output patterns despite fast-evolving upstream inputs through maternal gradients and vastly different modes of segmentation dynamics in different insect taxa ( Sander , 1976; Davis and Patel , 2002 ) . This type of neutral network evolution—producing the same output based on different regulatory principles—is called developmental system drift or phenogenetic drift ( Weiss and Fullerton , 2000; True and Haag , 2001; Weiss , 2005; Haag , 2007; Pavlicev and Wagner , 2012 ) . It is believed to be a very widespread phenomenon and can be interpreted as phenotypically neutral evolution along so-called genotype ( meta- ) networks . Genotype networks consist of different regulatory network structures—connected by simple mutational steps—that produce the same patterning or phenotypic output ( Ciliberti et al . , 2007a , 2007b; Wagner and Lynch , 2008; Draghi et al . , 2010; Wagner , 2011 ) . In order to discover the mechanisms underlying developmental system drift , it is necessary to systematically and quantitatively investigate the structure and dynamics of evolving regulatory networks . In this study , we use the dipteran gap gene system—constituting the first regulatory layer of the segmentation gene network ( Foe and Alberts , 1983; Akam , 1987; Ingham , 1988; Jaeger , 2011 ) —as a model system to study developmental system drift . The gap gene system in Drosophila melanogaster ( family: Drosophilidae ) is one of the most thoroughly studied developmental gene regulatory networks today . The genetic and molecular mechanisms of gap gene regulation have been investigated extensively over the last few decades ( reviewed in Jaeger ( 2011 ) ) , and a number of mathematical models exist that faithfully reproduce gap gene expression dynamics in this species ( Reinitz et al . , 1995; Jaeger et al . , 2004a , 2004b; Perkins et al . , 2006; Ashyraliyev et al . , 2009; Manu et al . , 2009a , 2009b; Crombach et al . , 2012a , 2012b; Becker et al . , 2013 ) . In this study , we provide a brief summary of the most important regulatory principles that were revealed by this research . An overview of the structure of the gap gene network is given in Figure 1 ( grey inset ) . Gap genes receive regulatory inputs from maternal protein gradients formed by the gene products of bicoid ( bcd ) , hunchback ( hb ) , and caudal ( cad ) ( Figure 1 , top row of graphs ) ( reviewed in St Johnston and Nüsslein-Volhard ( 1992 ) ) . These gradients set up an initial asymmetry along the major or antero-posterior ( A–P ) axis of the embryo . Bcd and Cad activate the four trunk gap genes hb , Krüppel ( Kr ) , knirps ( kni ) , and giant ( gt ) , which become expressed in broad , overlapping domains ( Figure 1 , bottom row of graphs ) . As development proceeds , gap gene cross-repression refines the pattern ( Figure 1 , grey inset ) ( see Jaeger , 2011 for review ) . Domains of hb and kni as well as Kr and gt have mutually exclusive expression patterns and show strong mutual repression . This ‘alternating cushions’ mechanism maintains and sharpens the basic staggered arrangement of gap domains . The expression patterns of Kr/kni , kni/gt , and gt/hb overlap and show weaker repression with a posterior-to-anterior bias , which results in anterior shifts of each of these domains over time . Finally , trunk gap gene expression is repressed in the posterior pole region of the embryo by the terminal gap genes tailless ( tll ) and huckebein ( hkb ) . 10 . 7554/eLife . 04785 . 003Figure 1 . The evolution of the dipteran gap gene network . A simplified phylogenetic tree of the order Diptera indicates the relative position of M . abdita with regard to other species of flies , midges , and mosquitoes in which gap genes have been studied in some detail . M . abdita belongs to the brachyceran infra-order Cyclorrhapha ( marked in red ) ; paraphyletic nematoceran lineages are shown in black . Only cyclorrhaphan flies have a bcd gene . Other maternal gradients have been lost along various branches of the tree ( as indicated ) . For each species , we show an image of the adult ( top ) , as well as a schematic representation of the spatial arrangement of maternal gradients ( middle ) , and gap gene expression domains ( bottom ) . Y-axes show expression levels ( in normalised arbitrary units ) ; X-axes show position along the major axis of the embryo ( A: anterior , P: posterior ) . See key for colour coding . Solid colours indicate previously published expression patterns; faded colours represent previously unknown patterns reported in this study . Question marks indicate unknown maternal gradients or potentially missing gap domains . The inset ( grey background ) shows the gap gene network in D . melanogaster . Within the inset , background colour indicates major maternal regulatory inputs , boxes show the position of gap domains along the A–P axis; T-bars represent strong ( solid ) or weak ( dashed ) cross-repressive interactions among gap genes . Species ( families ) : D . melanogaster ( Drosophilidae ) : vinegar fly; E . balteatus ( Syrphidae ) : marmalade hoverfly; M . abdita ( Phoridae ) : scuttle fly; C . albipunctata ( Psychodidae ) : moth midge; A . gambiae ( Culicidae ) : malaria mosquito . All images are our own except A . gambiae image taken by Muhammad Mahdi Karim ( source: Wikimedia Commons ) . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 04785 . 003 Our understanding of the dipteran gap gene system outside the drosophilid family is much less well developed . This limitation applies in particular to nematoceran midges and mosquitoes ( Figure 1 , black phylogenetic branches on the right ) ( Jiménez-Guri et al . , 2013 ) . Qualitative ( Rohr et al . , 1999; García-Solache et al . , 2010; Jiménez-Guri et al . , 2014 ) and quantitative ( Crombach et al . , 2014; Janssens et al . , 2014 ) studies of segmentation gene expression in the moth midge Clogmia albipunctata ( family: Psychodidae ) indicate that posterior patterning—especially the onset of posterior hb expression—is delayed and that repression between Kr and gt may not be conserved in this species , which lacks a posterior gt domain . A similar posterior delay is observed in the malaria mosquito Anopheles gambiae ( Culicidae ) , where both posterior domains of hb and gt are present , but appear only after gastrulation in an A–P order which is reversed compared to that observed in D . melanogaster ( Goltsev et al . , 2004 ) . Unfortunately , several factors involved in gap gene regulation are still unknown in these nematoceran species , and the lack of functional evidence based on genetic perturbations limits the interpretation of experimental results and verification of network models . The situation is slightly better within the Cyclorrhapha , a sub-taxon of brachyceran or ‘higher’ flies ( Figure 1 , red phylogenetic branches on the left ) ( Wiegmann et al . , 2011; Jiménez-Guri et al . , 2013 ) . Apart from D . melanogaster , two other cyclorrhaphan species are emerging as experimentally tractable models . One of these is the marmalade hoverfly Episyrphus balteatus ( Syrphidae ) . In this species , the pattern and order of gap gene activation along the A–P axis is largely conserved compared to D . melanogaster , even though there is no maternal contribution to hb expression ( Figure 1 ) ( Lemke and Schmidt-Ott , 2009; Lemke et al . , 2010 ) . However , the precise dynamics of gap gene expression and the nature of gap gene cross-regulation have not yet been investigated in this species . The second emerging non-drosophilid model system is Megaselia abdita , a member of the early-branching cyclorrhaphan family Phoridae ( Figure 1 ) ( Wiegmann et al . , 2011; Rafiqi et al . , 2011a; Jiménez-Guri et al . , 2013 ) . In what follows , we will focus on this species . Gap gene expression in M . abdita is less well studied than in E . balteatus ( Figure 1 ) and little is known about gap gene regulation . Maternal transcripts of bcd and hb can be detected ( Stauber et al . , 1999 , 2000; Lemke et al . , 2008 ) , but no maternal contribution to cad expression is present in this species ( Stauber et al . , 2008 ) . Localised maternal bcd transcripts at the anterior pole extend further posterior , and bcd RNAi cuticle phenotypes affect more posterior segments than in D . melanogaster , indicating a broader role of Bcd in M . abdita ( Stauber et al . , 1999 , 2000 ) . In contrast , expression of maternal and zygotic hb appears very similar in both species ( Stauber et al . , 2000; Lemke et al . , 2008 ) . The zygotic anterior hb domain is severely depleted in embryos treated with bcd RNAi , suggesting activation of hb by Bcd as in D . melanogaster ( Lemke et al . , 2008 ) . Moreover , the regulatory role of hb seems to be similar as well , since cuticle phenotypes of hb RNAi-treated M . abdita embryos resemble hb hypomorphs in D . melanogaster ( Stauber et al . , 2000 ) . Finally , Kr expression has been mentioned to be similar in both species ( Rafiqi et al . , 2008 ) , although no data were shown since the study focused on extraembryonic tissues at later stages of development . The similarity of gap gene expression across cyclorrhaphan species—despite considerable variation in maternal gradients—provides an excellent opportunity to study developmental system drift in an experimentally tractable set of laboratory models . A comparative , network-level comparison of gap gene regulation would reveal how the network compensates for variable upstream inputs and inter-species differences in initial expression patterns . As a foundation for such an analysis we have previously sequenced the early embryonic transcriptome of M . abdita ( Jiménez-Guri et al . , 2013 ) , and have established a precise staging scheme for embryogenesis—with a particular focus on the blastoderm stage ( Wotton et al . , 2014 ) . In this paper , we present a comprehensive analysis of gap gene expression in M . abdita , based on quantitative data with high spatial and temporal resolution . We compare wild-type expression dynamics to an equivalent dataset from D . melanogaster ( Crombach et al . , 2012a ) . Our comparison reveals that gap gene expression differs significantly between the two species during early to mid blastoderm stage but converges to equivalent patterns before the onset of gastrulation . In particular , we find that the broader influence of Bcd in M . abdita causes gap domains to emerge at more posterior positions compared to D . melanogaster . In addition , the absence of maternal Cad causes a delay in the onset of anterior shifts of posterior gap domains in M . abdita . This delay is later compensated by stronger shifts compared to D . melanogaster . To investigate the regulatory changes underlying these altered dynamics , we use previously established RNAi protocols ( Stauber et al . , 2000; Lemke et al . , 2008 ) to systematically knock-down trunk and terminal gap genes . We then assay expression of the remaining trunk gap genes in each genetically perturbed background using the same high-resolution quantitative approach as for wild-type patterns . Our knock-down analysis reveals that qualitative regulatory structure—the type of interactions between gap genes—is strongly conserved between species . We do , however , detect changes significantly affecting the strength of gap gene cross-repression . Based on this , we suggest quantitative changes in regulatory mechanisms for the altered dynamics of gap gene expression in M . abdita . Our evidence suggests that such quantitative developmental system drift is a common feature of evolving developmental systems . We took images of M . abdita blastoderm-stage embryos stained for maternal co-ordinate and gap gene mRNAs as previously described ( Crombach et al . , 2012a ) . Embryos were classified by cleavage cycle ( C10–14A ) , and eight separate time classes within C14A ( T1–8 ) , according to the staging scheme in Wotton et al . ( 2014 ) . This results in a temporal resolution for our expression data of around 20 min during C13 and 7 min for each time class during C14A . Embryo images were processed as described in Crombach et al . ( 2012b ) to extract the position of expression domain boundaries ( Figure 2A ) . Table 1 and Supplementary file 1 summarise the number of embryos in our dataset . We then determined median boundary positions and expression variability for each gene at each time point across embryos , resulting in a quantitative , integrated spatio-temporal dataset for gap gene expression in M . abdita ( Figure 2B ) . Sample size and resolution of this dataset are comparable to our previously published gap gene mRNA expression data for D . melanogaster ( Table 1 , Supplementary file 1 ) ( Crombach et al . , 2012a ) and are much higher than those achieved for our quantitative dataset of gap gene expression in another non-model dipteran , the moth midge Clogmia albipunctata ( Crombach et al . , 2014 ) . All the expression patterns presented here are available online through the SuperFly database ( http://superfly . crg . eu ) ( Cicin-Sain et al . , 2015 ) . 10 . 7554/eLife . 04785 . 004Figure 2 . Data acquisition and processing of wild-type and RNAi knock-down embryos . As an example , we show kni mRNA expression in wild-type and hb RNAi-treated embryos . ( A ) Data acquisition: wild-type or RNAi-treated embryos were stained by single or double in situ hybridisation using an enzymatic ( colorimetric ) protocol as described in Crombach et al . ( 2012a ) . Embryo image shows a lateral view ( anterior to the left , dorsal up ) stained for kni mRNA ( purple ) . We extract boundary positions as described in Crombach et al . ( 2012b ) : first , we determine a 10% strip ( delimited by red lines ) along the midline of the dorso-ventral axis ( black line ) . After the extraction of the intensity profile within this strip ( magenta line in graph ) , we manually fit clamped splines to domain boundaries ( black lines with mid- and end-points indicated by circles ) . ( B ) Extracted boundaries from wild-type embryos are classified by gene and time class ( grey lines ) and used to calculate median boundary positions ( black line in upper panel ) . This yields an integrated spatio-temporal dataset of gene expression ( lower panel ) . Shaded area represents regions of active expression delimited by positions of half-maximal expression for median boundaries ( solid lines ) . Error bars represent 1 . 5x median absolute deviation ( MAD ) , which approximates one standard deviation . ( C ) RNAi expression data ( red ) are plotted against wild-type boundary positions ( grey ) . X-axes represent % A–P position ( where 0% is the anterior pole ) . Y-axes represent pixel intensity or relative mRNA concentration in arbitrary units , except for the lower panel in ( B ) , where the Y-axis represents time flowing downwards . C12/13: cleavage cycles 12/13; T1–8: time classes within C14A as defined in Wotton et al . ( 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04785 . 00410 . 7554/eLife . 04785 . 005Table 1 . mRNA expression datasets for M . abdita . and D . melanogasterDOI: http://dx . doi . org/10 . 7554/eLife . 04785 . 005Domainant hbant gtKrknipost gtpost hbBoundaryPAPAPAPAPAPC114---------------------C12222287444-2-4--11-11---C13811117714141125-318896161411-1-T1865981010737-151414107151311105-T2777771919748-138106519168981T3191413881212151018-1611159517151417116T441113109141413134-711121041817913107T5588871111984-121617161118171414108T6410107788101043914141715151514121114T7277131388662258811115586109T8898666675888121266667698This table shows the number of embryos used to calculate median positions for each expression boundary at each point in time ( M . abdita: white rows; D . melanogaster: grey rows ) . Ant: anterior , Post: posterior domain . A indicates anterior , P posterior boundary of a domain . Time classification as defined in Wotton et al . ( 2014 ) for M . abdita and in Surkova et al . ( 2008b ) for D . melanogaster: C11–13 correspond to cleavage cycles 11 to 13; T1–8 represent time classes subdividing C14A . Our M . abdita expression dataset consists of a total of 367 embryos ( 91 stained for hb , 83 for Kr , 87 for gt , and 106 for kni ) . An additional 115 embryos make up the dataset for maternal co-ordinate and terminal gap genes shown in Supplementary file 1 . The D . melanogaster gap gene dataset has been published previously ( Crombach et al . , 2012a ) . It is included here for comparison . Figure 3 shows the expression patterns of the trunk gap genes hb , Kr , kni , and gt in M . abdita and compares them to the equivalent patterns in D . melanogaster . We provide numerical values for boundary positions , domain widths , boundary shifts , and domain overlaps between M . abdita and D . melanogaster in Supplementary file 2A–D . Box 1 provides a brief discussion of the most salient expression features of maternal co-ordinate and gap genes in both species . More detailed descriptions and figures characterising gene expression patterns , as well as more detailed references to the Drosophila literature , are provided in Appendix I below . Tables with measured boundary positions can be found in Supplementary file 3 . 10 . 7554/eLife . 04785 . 006Figure 3 . Trunk gap gene expression in M . abdita . ( A ) Wild-type mRNA expression patterns for trunk gap genes hunchback ( hb ) , Krüppel ( Kr ) , giant ( gt ) , and knirps ( kni ) are shown in M . abdita embryos at selected time points ( cleavage cycle 12 , C12; and C14A , time classes T2/5/8 ) . Embryos are shown in lateral view; anterior is to the left , dorsal up . ( B ) Space-time plots of gap gene expression in M . abdita ( solid lines , filled areas ) and D . melanogaster ( dashed lines ) . hb is shown in yellow , Kr in green , gt in blue , kni in red . Lines represent positions of half-maximal expression for median boundaries; error bars for M . abdita data represent 1 . 5x median absolute deviation ( MAD ) , as in Figure 2B . Lightly coloured hb in T6–T8 represents down-regulation . Time progresses downward in ( A ) and ( B ) . C12/13: cleavage cycles 12/13; T1–8: time classes within C14A as defined in Wotton et al . ( 2014 ) for M . abdita and in Surkova et al . ( 2008b ) for D . melanogaster . DOI: http://dx . doi . org/10 . 7554/eLife . 04785 . 006 In summary , gap gene expression is qualitatively similar to M . abdita compared to D . melanogaster , yet shows measurable quantitative differences ( Figure 3A , B ) . On the one hand , the relative positioning and timing of expression domains with respect to each other is largely conserved . On the other hand , the dynamics of expression differ significantly between the species ( Figure 3B ) . In D . melanogaster , the posterior boundary of the anterior hb domain is stable , while this border shifts to the anterior in M . abdita , which has a stable posterior domain of anterior gt instead . Moreover , in the posterior region of the embryo , gap genes are initiated more posteriorly in M . abdita compared to D . melanogaster , and patterning is delayed with posterior gap domains retracting from the pole and shifting to the anterior at later stages in M . abdita . This delay is compensated by very rapid anterior shifts of posterior gap domains ( especially gt ) during the second half of the blastoderm stage . Finally , gap domains ( the central Kr domain in particular ) are wider in M . abdita than in D . melanogaster . This leads to greater overlaps between adjacent expression domains . The extended range of bcd in M . abdita ( Stauber et al . , 1999 , 2000 ) is consistent with the observed posterior displacement of initial gap domains during the early blastoderm stage , which resembles expression patterns in embryos with increased bcd dosage in D . melanogaster ( Driever and Nüsslein-Volhard , 1988; Driever et al . , 1989; Struhl et al . , 1989 ) . To test if the placement of gap domains depends on Bcd levels in M . abdita ( as in D . melanogaster ) , we stained embryos treated with bcd RNAi and processed them through the same pipeline as wild-type embryos to measure the position of gene expression boundaries ( Figure 2 and Figure 4 ) . Variability in the efficacy of gene knock-downs in different embryos yields a result similar to an allelic series in classical genetics , and thus allows us to explicitly assess anterior hb boundary sensitivity to Bcd levels . In Figure 4E , we plot individual gene expression boundaries from RNAi-treated embryos against the positions of the equivalent wild-type borders ( see also Figure 2C ) . We find that reduced levels of bcd in RNAi embryos result in an increasingly anterior position of the hb boundary in M . abdita ( Figure 4 ) , suggesting that the placement of gap domains is dependent on Bcd levels as in D . melanogaster . 10 . 7554/eLife . 04785 . 008Figure 4 . Gap domain boundary positioning is dependent on Bicoid levels in M . abdita . hb expression ( purple ) is shown in wild-type ( A ) and in bcd RNAi-treated embryos ( B–D ) . The position of the posterior boundary of the anterior hb domain moves anteriorly as Bcd levels are reduced by RNAi . ( E ) Summary graph comparing wild-type boundary positions ( grey ) to boundary positions affected by RNAi ( yellow lines ) . All embryos are at time class T4 . Embryo images show lateral views: anterior is to the left , dorsal is up . DOI: http://dx . doi . org/10 . 7554/eLife . 04785 . 008 The observed delay in posterior gap patterning could be linked to differences in cad expression between the two species: in particular , the absence of maternal cad and/or altered levels of zygotic cad gene products in M . abdita . We assess this possibility by creating germ-line clones ( GLC ) in D . melanogaster . The resulting embryos lack maternal , but retain one copy of zygotic , cad . They exhibit early initiation of posterior gt and kni expression within normal wild-type variability . However , as development proceeds during cleavage cycle 14A ( C14A ) these domains show substantial posterior displacement compared to the wild-type ( Figure 5 ) . This suggests that altered levels and timing of cad expression are indeed responsible for the delayed shifts of posterior gap domains in M . abdita . 10 . 7554/eLife . 04785 . 009Figure 5 . Absence of maternal cad delays posterior patterning in D . melanogaster . The expression of gt ( A–C ) and hb ( D–F ) is shown in wild-type embryos ( A , D ) and in cad germ-line clones ( cad GLC ) lacking only maternal cad ( B , E; purple stain: gt or kni , red stain: even-skipped ) . ( C , F ) Summary graphs comparing wild-type boundary positions ( grey ) to boundary positions in cad GLC ( coloured lines ) show posterior domains of gt ( C ) and abdominal domains of kni ( F ) that are displaced towards the posterior . Embryos are at time class T3 ( gt; A–C ) and T4 ( kni; D–F ) . Embryo images show lateral views: anterior is to the left , dorsal is up . DOI: http://dx . doi . org/10 . 7554/eLife . 04785 . 009 To investigate the role of gap–gap cross-regulatory interactions in the patterning of the M . abdita embryo , we systematically knocked down each trunk and terminal gap gene by embryonic RNAi . For the trunk gap genes , the resulting severe cuticle phenotypes are similar to those observed in the equivalent null mutants of D . melanogaster ( see Appendix II below for detailed descriptions ) . We stained RNAi-treated embryos for each of the trunk gap genes and processed them through the same pipeline as wild-type embryos to measure the position of gene expression boundaries ( Figure 2A , C ) ( Crombach et al . , 2012b ) . The resulting RNAi datasets are summarised in Table 2 and documented in more detail in Supplementary file 4 . We monitor the sensitivity of expression dynamics to specific knock-downs at a temporal resolution equivalent to our wild-type dataset . This high temporal resolution , in combination with a large sample size , enables us to detect subtle alterations in gene expression dynamics that would have been missed by traditional , qualitative RNAi analyses . We structure our discussion of interactions among gap genes in M . abdita according to the basic regulatory principles they contribute to in D . melanogaster ( reviewed in Jaeger , 2011 ) . 10 . 7554/eLife . 04785 . 010Table 2 . Overview of the RNAi dataset for M . abditaDOI: http://dx . doi . org/10 . 7554/eLife . 04785 . 010hb RNAigt RNAiKr RNAikni RNAitll RNAihkb RNAitll:hkb RNAihbn/a17/28 ( 60% ) 21/41 ( 51% ) 35/52 ( 67% ) 12/29 ( 41% ) 21/22 ( 95% ) 15/15 ( 100% ) gt28/35 ( 80% ) n/a13/20 ( 65% ) 24/30 ( 80% ) 24/30 ( 80% ) 10/10 ( 100% ) 35/36 ( 97% ) Kr24/53 ( 46% ) 7/17 ( 41% ) n/a26/40 ( 65% ) 7/18 ( 38% ) 10/10 ( 100% ) 12/21 ( 57% ) kni9/14 ( 64% ) 11/25 ( 44% ) 3/21 ( 14% ) n/a6/44 ( 14% ) 4/10 ( 40% ) 16/16 ( 100% ) Total10270821221215288A total of 637 RNAi-treated embryos were used for the analysis . This table shows the number of embryos that were stained for each of the trunk gap genes ( rows ) , in each RNAi-treated background ( columns ) , and the number of embryos that showed a knock-down phenotype ( see also percentages ) . The total number of embryos used for each knock-down experiment is shown in the bottom row . A more detailed breakdown of embryos per cleavage cycle and time class , including detailed plots of boundary positions , is provided in Supplementary file 4 . In this paper , we present a quantitative and systematic comparative analysis of the expression dynamics and regulatory structure of a developmental gene regulatory network—the gap gene system—responsible for A–P patterning in early dipteran embryos . Through the use of a medium-throughput data quantification pipeline ( Crombach et al . , 2012b ) and a recently developed detailed embryonic staging scheme ( Wotton et al . , 2014 ) , we are able to characterise and compare gap gene expression dynamics between M . abdita and D . melanogaster with unprecedented accuracy and spatio-temporal resolution ( Figure 3 ) . Our fine-grained analysis reveals that , on the one hand , expression dynamics differ significantly between the two species . For instance , the posterior boundary of the anterior hb domain shifts to the anterior over time in M . abdita , while it remains stationary throughout the blastoderm stage in D . melanogaster . On the other hand , the output of the system is strongly conserved across 180 million years of evolution . The position and relative arrangement of gap domains converge in both species towards the onset of gastrulation . Each of these domains is initially set up more posteriorly , but ‘catches up’ with its equivalent in D . melanogaster through more pronounced anterior boundary shifts in M . abdita ( Figure 9A ) . The timing of this convergence is delayed in the posterior of the embryo: posterior domain boundaries of posterior hb start to coincide at T8 ( when this border becomes fully resolved in M . abdita ) , those of gt around T6 , those of abdominal kni at T3 , those of central Kr at T2 , and those of anterior hb already around T1 ( although the latter diverge again temporarily at later stages ) ( Figure 9A ) . 10 . 7554/eLife . 04785 . 014Figure 9 . Comparison of posterior boundary positions and gene network structure between M . abdita and D . melanogaster . ( A ) This graph shows the position of the posterior boundaries of anterior gt ( blue ) and hb ( yellow ) , central Kr ( green ) , abdominal kni ( red ) , and posterior gt ( blue ) and hb ( yellow ) . Initial boundary positions are either more anterior ( dark grey ) or more posterior ( light grey background ) in M . abdita ( solid ) than in D . melanogaster ( dotted lines ) . Arrows with time classes T1–T8 indicate when boundary positions first converge between the two species . Time flows downward . ( B and C ) Gap gene network structure for D . melanogaster ( B ) and M . abdita ( C ) as reconstructed from our knock-down analysis , displayed using the same layout as the inset of Figure 1 . Red T-bar connectors indicate differences in interaction strength ( indicated by line width ) between M . abdita and D . melanogaster . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 04785 . 014 These differences in expression dynamics are a consequence of the altered distribution of maternal co-ordinate gene products in M . abdita . Posterior gap domains appear further posterior—due to the broadened distribution and influence of bcd—and shift to the anterior later than in D . melanogaster—due to the absence of maternal cad . This is later compensated by pronounced , faster domain shifts in the second half of C14A , an effect most clearly exhibited by Kr and gt ( Figure 9A ) . Correspondingly , our knock-down analysis of the gap gene network in M . abdita reveals a stronger posterior-to-anterior bias in cross-repression between neighbouring domains ( Figure 9; compare B to C ) . An equivalent posterior bias is responsible for the anterior shift in the posterior boundary of anterior hb . In this case , it is not only repression of hb by Kr , but also repression by Kni , which is markedly stronger than the corresponding reverse interactions ( Figure 9B , C ) . Taken together , this suggests that the more pronounced domain shifts of hb and posterior boundaries in M . abdita are caused by increased posterior cross-regulatory bias . Our analysis reveals how different developmental dynamics and regulatory interaction strengths in the gap gene network produce extremely similar patterns at the onset of gastrulation , despite significant differences in the initial phases of maternal co-ordinate and gap gene expression . This constitutes developmental system drift of a quantitative kind: instead of changes in the qualitative nature of interactions ( e . g . , a switch from activation to repression ) , we observe neutral or compensatory evolution through tuning of regulatory interaction strengths . While most published examples of developmental system drift are of the first , qualitative , kind , there is one previous study that proposed quantitative changes in signalling strength to account for the evolution of the vulval patterning system in Caenorhabditis elegans ( Hoyos et al . , 2011 ) . This analysis focuses on the relative contributions of different signalling modes—long-range gradient vs juxtacrine signalling relay—to downstream cellular patterning . In contrast , we have investigated the regulatory mechanisms of compensatory network evolution in full genetic detail , at the level of individual regulatory interactions . The precise molecular mechanism for such changes remains unknown in M . abdita at this point—and will be subject to future studies . It is likely to be based on altered numbers , affinities , or arrangements of transcription factor binding sites in cis-regulatory elements ( Levine , 2010; Wittkopp and Kalay , 2011 ) as observed in other dipteran lineages ( Ludwig and Kreitman , 1995; Bonneton et al . , 1997; Ludwig et al . , 1998 , 2000 , 2005; Hancock et al . , 1999; McGregor et al . , 2001; Shaw et al . , 2002; Gompel et al . , 2005; Hare et al . , 2008; Jeong et al . , 2008; Williams et al . , 2008; Peterson et al . , 2009; Bradley et al . , 2010; Frankel et al . , 2011; Wunderlich et al . , 2012; Paris et al . , 2013 ) . Although many such changes at the sequence level do not seem to significantly affect levels of gene expression , there are others that do ( Wunderlich et al . , 2012; Arnold et al . , 2014 ) . Here , we show that even these changes—that are not neutral at the level of cis-regulatory elements but lead to a quantitative difference in interaction strength—can lead to neutral drift if we consider their phenotypic effects at the systems- or network-level . Based on the observed high turnover rates for the evolution of binding sites in various evolutionary lineages ( McGregor et al . , 2001; Dermitzakis and Clark , 2002; Ludwig , 2002; Costas et al . , 2003; Dermitzakis et al . , 2003; Moses et al . , 2006; Yáñez-Cuna et al . , 2013 ) , we believe that such quantitative system drift is a common feature of network evolution . It supplies a mechanism for exploring alternative genotype spaces , increasing the range of mutationally accessible novel phenotypes ( Wagner , 2011; Jaeger and Monk , 2014 ) , and allows for dynamically different alternative regulatory processes to evolve in specific developmental contexts , such as that of segmentation ( Oates et al . , 2012; Richmond and Oates , 2012; Sarrazin et al . , 2012; Jaeger and Sharpe , 2014 ) . Based on these considerations , quantitative system drift is likely to be important , not only for neutral phenotypic evolution but also for the de novo evolution of biological form . Finally , quantitative system drift in early embryonic patterning networks , as reported here for the cyclorrhaphan gap gene system , provides an explanation for the developmental hourglass model ( Seidel , 1960; Sander , 1983; Slack et al . , 1993; Duboule , 1994; Raff , 1996 ) . This model predicts a minimum of morphological divergence during intermediate embryonic stages . It has been confirmed at the level of gene expression in three recent studies ( Domazet-Lošo and Tautz , 2010; Kalinka et al . , 2010; Quint et al . , 2012 ) . We observe the same phenomenon in our explicitly spatio-temporal context: maternal co-ordinate and early gap gene expression differ significantly between species , but later converge to a common pattern ( Figure 9A ) . Our knock-down analysis of the gap gene network in M . abdita provides causal regulatory explanations of how this can be achieved at the gene network-level . In this section , we describe the quantified wild-type mRNA expression patterns of M . abdita maternal co-ordinate and gap genes in detail and compare them to the published literature in D . melanogaster . In this section , we describe the gap gene RNAi cuticles of M . abdita and compare them to the published literature on gap gene mutants in D . melanogaster . In M . abdita , hb knock-down ( as previously published in Stauber et al . , 2000 ) results in the deletion of the thorax , with additional defects in abdominal segments A1–8 and A8 , and a reduced cephalopharyngeal skeleton , suggesting that the posterior gnathocephalon is also missing . This corresponds to a hypomorphic hb mutant phenotype in D . melanogaster ( Nüsslein-Volhard and Wieschaus , 1980; Jürgens et al . , 1984; Lehmann and Nüsslein-Volhard , 1987 ) . Other trunk gap gene knock-downs resemble D . melanogaster null mutants ( Figure 12 ) . M . abdita gt RNAi resulted in cuticles lacking abdominal segments A5–7 ( Figure 12B ) rather than just the denticle bands of abdominal segments A5–7 as in D . melanogaster ( i . e . , no naked cuticle is formed ) ( Wieschaus et al . , 1984a; Gergen and Wieschaus , 1986; Petschek et al . , 1987 ) . Additionally , head segments are affected , and A8 is slightly reduced with filzkörper not fully present . 10 . 7554/eLife . 04785 . 017Figure 12 . Cuticle phenotypes resulting from RNAi knock-down of gt , kni , and Kr in M . abdita . ( A ) A wild-type cuticle is shown for comparison to severe phenotypes in gt ( B ) , kni ( C ) , and Kr ( D ) RNAi-treated embryos . We tentatively assign segment identity to persisting abdominal segments based on their relative spatial order , the position of expression domains of the knocked-down genes in the blastoderm ( see Appendix I ) , and comparison to the corresponding cuticle phenotypes in D . melanogaster . T1–T3; thoracic segments , A1–A8; abdominal segments; solid line indicates fused segments . Cuticles are shown in lateral ( A , B ) , ventral ( C ) , or ventro-lateral ( D ) view . Anterior is to the left . hb RNAi knock-down phenotypes were previously published in Stauber et al . ( 2000 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04785 . 017 Following kni RNAi in M . abdita embryos , abdominal segments A1–7 merge ( Figure 12C ) . In the most severe knock-down phenotypes , the fused segments are marked by a single , irregular denticle field . Intermediate phenotypes may exhibit a few remaining discernible abdominal denticle belts . In some kni RNAi cuticles , more posterior markers as well as the posterior portion of the T3 denticle belt were distorted as well . Head segments appear unaffected . This severe kni RNAi phenotype is very similar to the null mutant kni phenotype in D . melanogaster ( Nüsslein-Volhard and Wieschaus , 1980; Jürgens et al . , 1984; Nauber et al . , 1988 ) . In severely affected M . abdita Kr knock-down cuticles , only the telson and A8 , plus one or two additional abdominal segments are formed ( putative A7 , and sometimes A6; Figure 12D ) . Other abdominal segments and all thoracic segments are either missing or fused , and head involution fails to occur . Head segments were strongly affected . Unlike in D . melanogaster , where A6 is duplicated as a mirror image in Kr mutants ( Nüsslein-Volhard and Wieschaus , 1980; Wieschaus et al . , 1984b ) , we cannot identify clear instances of local inversions or duplications in M . abdita . However , it is possible that we have missed these phenotypes , due to the less obvious polarity of the segmental denticle patterns in M . abdita compared to D . melanogaster .	Similar biological phenomena can result from different processes occurring in different organisms . For example , the early stages of how an insect develops from an egg can vary substantially between different species . Nonetheless , all insects have a body plan that develops in segments . The same outcome occurring as a result of different developmental steps is known as ‘system drift’ , but the mechanisms underlying this phenomenon are largely unknown . How the body segments of the fruit fly Drosophila develop has been extensively studied . First , a female fruit fly adds messenger RNA ( or mRNA ) molecules copied from a number of genes into her egg cells . These mRNA molecules are then used to produce proteins whose concentration varies along the length of each egg . These proteins in turn switch on so-called ‘gap genes’ in differing amounts in different locations throughout the fruit fly embryo . The activity of these genes goes on to define the position and extent of specific segments along the fruit fly's body . Like the fruit fly , the scuttle fly Megaselia abdita has a segmented body . However , mothers of this species deposit somewhat different protein gradients into their eggs . How the regulation of development differs in the scuttle fly to compensate for this change is unknown . Now , Wotton et al . have studied , in detail , how gap genes are regulated in this less well-understood fly species to understand the mechanisms responsible for a specific example of system drift . In the fruit fly , gap genes normally switch-off ( or reduce the expression of ) other gap genes within the same developing body segment , and Wotton et al . found that the same kind of interactions tended to occur in the scuttle fly . As such , the overall structure of the gap gene network was fairly similar between scuttle and fruit flies . There were , however , differences in the strength of these interactions in the two fly species . These quantitative differences result in a different way of making the same segmental pattern in the embryo . In this way , Wotton et al . show how tinkering with the strength of specific gene interactions can provide an explanation for system drift .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2015	Quantitative system drift compensates for altered maternal inputs to the gap gene network of the scuttle fly Megaselia abdita
Constraint-based modelling ( CBM ) is a powerful tool for the analysis of evolutionary trajectories . Evolution , especially evolution in the distant past , is not easily accessible to laboratory experimentation . Modelling can provide a window into evolutionary processes by allowing the examination of selective pressures which lead to particular optimal solutions in the model . To study the evolution of C4 photosynthesis from a ground state of C3 photosynthesis , we initially construct a C3 model . After duplication into two cells to reflect typical C4 leaf architecture , we allow the model to predict the optimal metabolic solution under various conditions . The model thus identifies resource limitation in conjunction with high photorespiratory flux as a selective pressure relevant to the evolution of C4 . It also predicts that light availability and distribution play a role in guiding the evolutionary choice of possible decarboxylation enzymes . The data shows evolutionary CBM in eukaryotes predicts molecular evolution with precision . Identifying specific evolutionary trajectories and modelling the outcome of adaptive strategies at the molecular levels is a major challenge in evolutionary systems biology ( Papp et al . , 2011 ) . The evolution of novel metabolic pathways from existing parts may be predicted using constraint-based modelling ( CBM ) ( Orth et al . , 2010 ) . In CBM , selective pressures are coded via the objective functions for which the model is optimised . The factors which constrain evolution are integrated into the models via changes in model inputs or outputs and via flux constraints . We hypothesised that the evolution of the agriculturally important trait of C4 photosynthesis is accessible to CBM . C4 photosynthesis evolved independently in at least 67 independent origins in the plant kingdom ( Scheben et al . , 2017 ) and it allows colonisation of marginal habitats ( Sage et al . , 2012 ) and high biomass production in annuals such as crops ( Sage , 2004; Edwards et al . , 2010 ) . The C4 cycle acts as a biochemical pump which enriches the CO2 concentration at the site of Rubisco to overcome a major limitation of carbon fixation ( Sage , 2004 ) . Enrichment is beneficial because Rubisco , the carbon fixation enzyme , can react productively with CO2 and form two molecules of 3-PGA , but it also reacts with O2 and produces 2-phosphoglycolate which requires detoxification by photorespiration ( Ogren and Bowes , 1971 ) . The ratio between both reactions is determined by the enzyme specificity towards CO2 , by the temperature , and the concentrations of both reactants , which in turn is modulated by stresses such as drought and pathogen load . Evolution of Rubisco itself is constrained since any increase in specificity is paid for by a reduction in speed ( Spreitzer and Salvucci , 2002 ) . Lower speeds most likely cause maladaptivity since Rubisco is a comparatively slow enzyme and can comprise up to 50% of the total leaf protein ( Ellis , 1979 ) . In the C4 cycle , phosphoenolpyruvate carboxylase affixes CO2 to a C3 acid , phosphoenolpyruvate ( PEP ) , forming a C4 acid , oxaloacetate ( OAA ) . After stabilisation of the resulting C4 acid by reduction to malate or transamination to aspartate , it is transferred to the site of Rubisco and decarboxylated by one of three possible decarboxylation enzymes , NADP-dependent malic enzyme ( NADP-ME ) , NAD-dependent malic enzyme ( NAD-ME ) , or PEP carboxykinase ( PEP-CK ) ( Hatch , 1987; Schlüter et al . , 2016b ) . Species such as corn ( Zea mays ) ( Pick et al . , 2011 ) and great millet ( Sorghum bicolor ) ( Döring et al . , 2016 ) use NADP-ME , species like common millet ( Panicum miliaceum ) ( Hatch , 1987 ) and African spinach ( Gynandropsis gynandra ) ( Feodorova et al . , 2010; Voznesenskaya et al . , 2007 ) use NAD-ME and species such as guinea grass ( Panicum maximum ) ( Bräutigam et al . , 2014 ) use mainly PEP-CK with the evolutionary constraints leading to one or the other enzyme unknown . Mixed forms are only known to occur between a malic enzyme and PEP-CK but not between both malic enzymes ( Wang et al . , 2014 ) . After decarboxylation , the C3 acid diffuses back to the site of phosphoenolpyruvate carboxylase ( PEPC ) and is recycled for another C4 cycle by pyruvate phosphate dikinase ( PPDK ) ( Hatch , 1987; Schlüter et al . , 2016b ) . All the enzymes involved in the C4 cycle are also present in C3 plants ( Aubry et al . , 2011 ) . In its most typical form , this C4 cycle is distributed between different cell types in a leaf in an arrangement called Kranz anatomy ( Haberlandt , 1904 ) . Initial carbon fixation by PEPC occurs in the mesophyll cell , the outer layer of photosynthetic tissue . The secondary fixation by Rubisco after decarboxylation occurs in an inner layer of photosynthetic tissue , the bundle sheath which in turn surrounds the veins . Both cells are connected by plasmodesmata which are pores with limited transfer specificity between cells . A model which may test possible carbon fixation pathways at the molecular level thus requires two cell architectures connected by transport processes ( Bräutigam and Weber , 2010 ) . CBM of genome-scale or close to it are well suited to study evolution ( summarised in Papp et al . , 2011 ) . Evolution of different metabolic modes from a ground state , the metabolism of Escherichia coli , such as glycerol usage ( Lewis et al . , 2010 ) or endosymbiotic metabolism ( Pál et al . , 2006 ) have been successfully predicted . Metabolic maps of eukaryotic metabolism are of higher complexity compared to bacteria since they require information about intracellular compartmentation and intracellular transport ( Duarte , 2004 ) and may require multicellular approaches . In plants , aspects of complex metabolic pathways , such as the energetics of CAM photosynthesis ( Cheung et al . , 2014 ) , and fluxes in C3 and C4 metabolism ( Boyle and Morgan , 2009; Gomes de Oliveira Dal’Molin et al . , 2011; de Oliveira Dal'Molin et al . , 2010; Arnold and Nikoloski , 2014; Saha et al . , 2011 ) have been elucidated with genome scale models . The C4 cycle is not predicted by these current C4 models unless the C4 cycle is forced by constraints ( Gomes de Oliveira Dal’Molin et al . , 2011; Mallmann et al . , 2014 ) . In the C4GEM model , the fluxes representing the C4 cycle are a priori constrained to the cell types ( Gomes de Oliveira Dal’Molin et al . , 2011 ) , and in the Mallmann model , the C4 fluxes are induced by activating flux through PEPC ( Mallmann et al . , 2014 ) . Models in which specific a priori constraints activated C4 were successfully used to study metabolism under conditions of photosynthesis , photorespiration , and respiration ( Saha et al . , 2011 ) and to study N-assimilation under varying conditions ( Simons et al . , 2013 ) . However , they are incapable of testing under which conditions the pathway may evolve . Schematic models suggest that the C4 cycle evolves from its ancestral metabolic state C3 photosynthesis along a sequence of stages ( summarised in Sage , 2004; Bräutigam and Gowik , 2016 ) . In the presence of tight vein spacing and of photosynthetically active bundle sheath cells ( i . e . Kranz anatomy ) , a key intermediate in which the process of photorespiration is divided between cell types is thought to evolve ( Monson , 1999; Sage et al . , 2012; Heckmann et al . , 2013; Bauwe , 2010 ) . The metabolic fluxes in this intermediate suggest an immediate path towards C4 photosynthesis ( Mallmann et al . , 2014; Bräutigam and Gowik , 2016 ) . ( Heckmann et al . , 2013 ) built a kinetic model in which the complex C4 cycle was represented by a single enzyme , PEPC . Assuming carbon assimilation as a proxy for fitness , the model showed that the evolution from a C3 progenitor species with Kranz-type anatomy towards C4 photosynthesis occurs in modular , individually adaptive steps on a Mount Fuji fitness landscape . It is frequently assumed that evolution of C4 photosynthesis requires water limitation ( Bräutigam and Gowik , 2016; Heckmann et al . , 2013; Mallmann et al . , 2014 ) . However , ecophysiological research showed that C4 can likely evolve in wet habitats ( Osborne and Freckleton , 2009; Lundgren and Christin , 2017 ) . CBM presents a possible avenue to study the evolution of C4 photosynthesis including its metabolic complexity in silico . In this study , we establish a generic two-celled , constraint-based model starting from the Arabidopsis core model ( Arnold and Nikoloski , 2014 ) . We test under which conditions and constraints C4 photosynthesis is predicted as the optimal solution . Finally , we test which constraints result in the prediction of the particular C4 modes with their different decarboxylation enzymes . In the process , we demonstrate that evolution is predictable at the molecular level in an eukaryotic system and define the selective pressures and limitations guiding the 'choice' of metabolic flux . Flux balance analysis requires five types of information , the metabolic map of the organism , the input , the output , a set of constraints ( i . e . limitations on input , directionality of reactions , forced flux through reactions ) , and optimisation criteria for the algorithm which approximate the selective pressures the metabolism evolved under . In this context , inputs define the resources that need to be taken up by the metabolic network to fulfil a particular metabolic function , which is related to the outputs , for example the synthesis of metabolites part of the biomass or other specific products . In CBM , the objective is most likely related to the in- and/or outputs . For reconstruction of the C3 metabolic map we curated the Arabidopsis core model ( Arnold and Nikoloski , 2014 ) manually ( Table 1 ) to represent the metabolism of a mesophyll cell in a mature photosynthetically active leaf of a C3 plant , further on called one-cell model ( provided in Figure 1—source data 1 ) . The Arabidopsis core model is a bottom-up-assembled , large-scale model relying solely on Arabidopsis-specific annotations and the inclusion of only manually curated reactions of the primary metabolism . The Arabidopsis core model is accurate with respect to mass and energy conservation , allowing optimal nutrient utilisation and biochemically sound predictions ( Arnold and Nikoloski , 2014 ) . For the inputs , we considered a photoautotrophic growth scenario with a fixed CO2 uptake of about 20 μmol/ ( m2s ) ( Lacher , 2003 ) . Light , sulphates , and phosphate are freely available . Due to the observation that nitrate is the main source ( 80% ) of nitrogen in leaves in many species ( Macduff and Bakken , 2003 ) , we set nitrate as the sole nitrogen source . If both ammonia and nitrate are allowed , the model will inevitably predict the physiologically incorrect sole use of ammonia since fewer reactions and less energy are required to convert it into glutamate , the universal amino group currency in plants . Water and oxygen can be freely exchanged with the environment in both directions . To compute the output , we assume a mature fully differentiated and photosynthetically active leaf , which is optimised for the synthesis and export of sucrose and amino acids to the phloem under minimal metabolic effort . Following the examples of models in bacteria , many plant models use a biomass function which assumes that the leaf is required to build itself ( de Oliveira Dal'Molin et al . , 2010; Arnold and Nikoloski , 2014; Saha et al . , 2011 ) using photoautotrophic that is ( Arnold and Nikoloski , 2014 ) or heterotrophic that is ( Cheung et al . , 2014 ) energy and molecule supply . In plants , however , leaves transition from a sink phase in which they build themselves from metabolites delivered by the phloem to a source phase in which they produce metabolites for other organs including sink leaves ( Turgeon , 1989 ) . The composition of Arabidopsis phloem exudate ( Wilkinson and Douglas , 2003 ) was used to constrain the relative proportions of the 18 amino acids and the ratio of sucrose : total amino acids ( 2 . 2 : 1 ) . To account for daily carbon storage as starch for export during the night , we assume that half of the assimilated carbon is stored in the one-cell model . We explicitly account for maintenance costs by the use of a generic ATPase and use the measured ATP costs for protein degradation and synthesis of a mature Arabidopsis leaf ( Li et al . , 2017 ) as a constraint . We initially assume a low photorespiratory flux according to the ambient CO2 and O2 partial pressures considering no heat , drought , salt or osmotic stress which may alter the ratio towards higher flux towards the oxygenation reaction . To develop a largely unconstrained model and detect possible errors in the metabolic map , we initially kept the model unconstrained with regard to fixed fluxes , flux ratios , and reaction directions . Different model iterations were run in ( re- ) design , simulate , validate cycles against known physiology with errors sequentially eliminated and a minimal set of constraints required for a C3 model recapitulating extant plant metabolism determined . After each change , the CBM predicted all fluxes which were output as a table and manually examined ( for example see Figure 1—source data 2 ) . The initial FBA resulted in carbon fixation by enzymes such as the malic enzymes which , in reality , are constrained by the kinetics of the enzymes towards decarboxylation . All decarboxylation reactions were made unidirectional towards decarboxylation to prevent erroneous carbon fixation in the flux distribution . The next iteration of FBA predicted loops through nitrate reductases which ultimately converted NADH to NADPH . We traced this loop to an error in the initial model , in which malate dehydrogenases in the cytosol and mitochondrion were NADP-dependent instead of NAD-dependent . After correction of the co-factor in the one-cell model , the loops through nitrate reductases were no longer observed . Another iteration predicted excessive flux through the mitochondrial membrane where multiple metabolites were exchanged and identified missing transport processes as the likely reason . Based on Linka and Weber ( 2010 ) , we added known fluxes across the mitochondrial and plastidic envelope membranes which remedied the excessive fluxes in the solution . The chloroplastic ADP/ATP carrier protein is constrained to zero flux since its mutant is only affected during the night but not if light is available ( Reiser et al . , 2004 ) . The obtained flux distribution still contained excessive fluxes through multiple transport proteins across internal membranes which ultimately transferred protons between the organelles and the cytosol . Since for most if not all transport proteins the precise protonation state of metabolites during transport is unknown and hence cannot be correctly integrated into the model , we allowed protons to appear and disappear as needed in all compartments . This provision precludes conclusions about the energetics of membrane transport . ATP generation occurred in a distorted way distributed across different organelles which were traced to the H+ consumption of the ATPases in mitochondria and chloroplasts . The stoichiometry was altered to to 3:1 ( chloroplast ) and 4:1 ( mitochondria ) ( Petersen et al . , 2012; Turina et al . , 2016 ) . We assume no flux for the chloroplastic NADPH dehydrogenase and plastoquinol oxidase because ( Josse et al . , 2000; Yamamoto et al . , 2011 ) have shown that their effect on photosynthesis is minor . In preparation for modelling the C4 cycle , we ensured that all reactions known to occur in C4 ( i . e . malate/pyruvate exchange , likely via DiT2 in maize [Weissmann et al . , 2016] , possibly promiscuous amino transferases [Duff et al . , 2012] ) are present in the one-cell model , since ( Aubry et al . , 2011 ) showed that all genes encoding enzymes and transporters underlying the C4 metabolism are already present in the genome of C3 plants . We integrated cyclic electron flow ( Shikanai , 2016 ) and alternative oxidases in the mitochondria ( Vishwakarma et al . , 2015 ) , since both have been hypothesised to be important during the evolution and/or execution of the C4 cycle . Models and analysis workflows provided as jupyter notebooks ( Thomas et al . , 2016 ) are available as supplementary material or can be accessed on GitHub https://github . com/ma-blaetke/CBM_C3_C4_Metabolism ( Blätke , 2019; copy archived at https://github . com/elifesciences-publications/CBM_C3_C4_Metabolism ) . The one-cell model comprises in total 413 metabolites and 572 reactions , whereof 139 are internal transporters , 90 are export and eight import reactions ( see also below ) , which are involved in 59 subsystems . Figure 1 provides an overview of the primary subsystems according to Arnold and Nikoloski ( 2014 ) . The one-cell model requires a photosynthetic photon flux density ( PPFD ) of 193 . 7 μmol/ ( m2s ) ( Table 2 ) . The one-cell model takes up the maximal amount of CO2 to produce the maximum amount of phloem sap , as well as 0 . 8 μmol/ ( m2s ) of NO3- and 18 . 2 μmol/ ( m2s ) of H2O . According to the assumed ratio of sucrose and amino acids in the phloem sap , the flux of sucrose predicted by the model is 0 . 5 μmol/ ( m2s ) and of amino acids 0 . 3 μmol/ ( m2s ) . The rate of oxygen supply by the network is 20 . 9 μmol/ ( m2s ) . Part of the complete flux table is displayed in Table 2; the full table is available , see Figure 1—source data 2 . The flux table of all reactions did not display circular fluxes , and the reactions were within expected physiological ranges ( Figure 1—source data 2 ) . The CO2 uptake rate and the phloem sap output have a positive linear relationship , see Figure 1—figure supplement 1 ( A ) . The same is true for the correlation of the PPFD and phloem sap output in the range of 100 μmol/ ( m2s ) –200 μmol/ ( m2s ) , see Figure 1—figure supplement 1 ( B ) . Above 200 μmol/ ( m2s ) , the CO2 uptake rate acts as a limiting factor restricting the increase of phloem sap production . If either the PPFD or the CO2 uptake rate is zero , the phloem sap cannot be produced , compare Figure 1—figure supplement 1 ( A ) and ( B ) . Most of the metabolic processes use ATP/ADP as main energy equivalent ( 60% ) , followed by NADP/NADPH ( 37 . 5% ) and NAD/NADH ( 2 . 4% ) , see Figure 1—figure supplement 2 ( D ) . Nearly all ATP is produced by the light reactions ( 97 . 2% ) and consumed by the reductive pentose phosphate cycle ( 94 . 1% ) , see Figure 1—figure supplement 2 ( A ) . The oxidative phosphorylation produces only ( 1% ) of ATP . In proportion , the maintenance cost for protein synthesis and degradation makeup 28% of the respiratory ATP produced by the oxidative phosphorylation ( Figure 1—figure supplement 2 ( E ) ) . Similarly , nearly all NADPH is produced by the light reaction ( 98 . 9% ) , which is consumed by the reductive pentose-phosphate cycle ( 98 . 3% ) as well ( Figure 1—figure supplement 2 ( B ) ) . The canonical glycolysis and photorespiration produce nearly equal amounts of NADH , 45% and 47 . 7% , significantly less NADH is produced through the pyruvate dehydrogenase activity 6 . 85% . Nitrate assimilation ( 45% ) , glutamate biosynthesis ( 47 . 7% ) , glyoxylate cycle ( 21 . 6% ) and alternative respiration ( 11 . 8% ) consume the produced NADH ( Figure 1—figure supplement 2 ( C ) ) . To rebuild the characteristic physiology of C4 leaves , we duplicated the one-cell model and connected the two network copies by bi-directional transport of cytosolic metabolites including amino acids , sugars , single phosphorylated sugars , mono-/di-/tri-carboxylic acids , glyceric acids , glycolate , glycerate , glyceraldehyde-3-phosphate , di-hydroxyacetone-phosphate and CO2 , see Materials and methods for details . Since CBM is limited to static model analysis , we introduced two Rubisco populations in the bundle sheath network to approximate CO2 concentration-dependent changes in the oxygenation : carboxylation ratio of Rubisco ( vR⁢B⁢O/vR⁢B⁢C ) itself . We kept the native constrained Rubisco population that is forced to undertake oxygenation reactions and added a CCM-dependent Rubisco population which can only carboxylate ribulose 1 , 5-bisphosphate . The CCM-dependent Rubisco population is only able to use CO2 produced by the bundle sheath network but not environmental CO2 released by the mesophyll . C4 plants have a higher CO2 consumption and thus , an increased CO2 uptake of 40 μmol/ ( m2s ) was allowed ( Leakey et al . , 2006 ) . All other constraints and the objective of the one-cell model are maintained in the two-cell model , see Figure 2 . Initially , we optimised for the classical objective function of minimal total flux through the metabolic network at different levels of photorespiration . These different levels of photorespiration integrate changes to external CO2 concentration and stomatal opening status which is governed by plant water status and biotic interactions . From the complete flux distribution , we extracted fluxes of PEPC and PPDK , the decarboxylation enzymes , Rubisco and metabolite transporter between the two cells to ascertain the presence of a C4 cycle , see Figure 3 and Figure 3—figure supplement 1 . At low photorespiratory levels , flux through PEPC is barely detectable ( Figure 3 ( A ) ) . If photorespiration increases to moderate levels , flux through PEPC can be predicted and increases to 40 μmol/ ( m2s ) , that is all CO2 is funnelled through PEPC , for high photorespiratory fluxes . Concomitant with flux through PEPC , the activity of the decarboxylation enzymes changes ( Figure 3 ( B ) ) . At low to intermediate levels of photorespiratory flux , glycine decarboxylase complex activity is predicted to shuttle CO2 to the bundle sheath at up to 4 . 7 μmol/ ( m2s ) . Decarboxylation of C4 acids is initially mostly mediated by PEP-CK and is largely taken over by NADP-ME at high fluxes through photorespiration . Flux through NAD-ME is very low under all photorespiration levels . The decarboxylation enzymes dictate flux through the different Rubiscos in the model ( Figure 3 ( C ) ) . At low photorespiratory flux , both the Rubiscos in mesophyll and bundle sheath are active . Only very little flux occurs through the CCM-dependent Rubisco , which is a result of the glycine decarboxylase ( Figure 3 ( B ) ) . With increasing photorespiratory flux , this flux through glycine decarboxylase increases ( Figure 3 ( B ) ) and therefore , total Rubisco activity exceeds the carbon intake flux ( Figure 3 ( C ) ) . Carbon fixation switches to the CCM-dependent Rubisco with increasing flux through PEPC ( Figure 3 ( A ) ) and the classic C4 cycle decarboxylation enzymes ( Figure 3 ( B ) ) . Flux through PPDK mostly reflects flux through PEPC ( Figure 3 ( D ) ) . The transport fluxes between the cells change with changing photosynthetic mode ( Figure 3 ( E and F ) ) . At low rates of photorespiration when PEPC is barely active , the only flux towards the bundle sheath is CO2 diffusion ( Figure 3 ( E ) ) with no fluxes towards the mesophyll ( Figure 3 ( F ) ) . In the intermediate phase glycolate and glycerate are predicted to be transported and a low-level C4 cycle dependent on the transport of aspartate , malate , PEP and alanine operates ( Figure 3 ( E ) and ( F ) ) . In case of high photorespiratory rates , the exchange between mesophyll and bundle sheath is mainly carried by malate and pyruvate ( Figure 3 ( E ) and ( F ) ) . Flux through PPDK ( Figure 3 ( D ) ) is lower than flux through PEPC ( Figure 3 ( A ) ) at the intermediate stage ( Figure 3 ( F ) ) . Evolution of C4 photosynthesis with NADP-ME as the major decarboxylation enzyme is predicted if the photorespiratory flux is high and model optimised for minimal total flux , in other words , resource limitation . Among the known independent evolutionary events leading to C4 photosynthesis , 20 are towards NAD-ME while 21 occurred towards NADP-ME ( Sage , 2004 ) . PEP-CK is dominant or at least co-dominant only in Panicum maximum ( Bräutigam et al . , 2014 ) , Alloteropsis semialata semialata ( Christin et al . , 2012 ) , and in the Chloridoideae ( Sage , 2004 ) . To analyse whether the predicted evolution of the C4 cycle is independent of a particular decarboxylation enzyme , we performed three separate experiments , where only one decarboxylation enzyme can be active at a time . The other decarboxylation enzymes were de-activated by constraining the reaction flux to zero resulting in three different predictions , one for each decarboxylation enzyme . The flux distributions obtained under the assumption of oxygenation : carboxylation ratio of 1 : 3 and minimisation of photorespiration as an additional objective predicts the emergence of a C4 cycle for each known decarboxylation enzyme . To visualise the possible C4 fluxes , the flux distribution for candidate C4 cycle enzymes was extracted from each of the three predictions and visualised as arc width and color ( Figure 4 ) . While the flux distribution in the mesophyll is identical for three predicted C4 cycles of the decarboxylation enzymes , it is diverse in the bundle sheath due to the different localisation of the decarboxylation and related transport processes , see Figure 4 . The flux distribution does not completely mimic the variation in transfer acids known from laboratory experiments ( Hatch , 1987 ) since all of the decarboxylation enzymes use the malate/pyruvate shuttle . In the case of NAD-ME and PEP-CK , the two-cell model also predicts a supplementary flux through the aspartate/alanine shuttle . We tested whether transfer acids other than malate and pyruvate are feasible and explored the near-optimal space . To this end , the model predictions are repeated , allowing deviation from the optimal solution and the changes recorded . Deviations from the optimal solution are visualised as error bars ( Figure 5 ) . Performing a flux variability analysis ( FVA ) and allowing the minimal total flux to differ by 1 . 5% , predicts that for most metabolites which are transferred between mesophyll and bundle sheath , the variability is similar for all three decarboxylation types . For the NAD-ME and PEP-CK types , changes in the near-optimal space were observed for the transfer acids malate , aspartate , pyruvate and alanine . Minor differences were present for triose phosphates and phosphoglycerates as well as for PEP . For the NADP-ME type , FVA identifies only minor variation ( Figure 5 ) . In the case of NAD-ME but not in the case of NADP-ME the activity of the malate/pyruvate shuttle can be taken over by the aspartate/alanine shuttle and partly taken over in case of PEP-CK , see Figure 5 . The aspartate/alanine shuttle is thus only a near-optimal solution when the model and by proxy evolutionary constraints are resource efficiency and minimal photorespiration . To analyse the effect of other conditions on the particular C4 state , we apply the minimisation of photorespiration as an additional objective to minimal total flux . Since NAD-ME and PEP-CK type plants use amino acids as transfer acids in nature , nitrogen availability has been tagged as a possible evolutionary constraint that selects for decarboxylation by NAD-ME or PEP-CK . When nitrate uptake was limiting , the optimal solution to the model predicted overall reduced flux towards the phloem output ( Figure 6—figure supplement 1 ) but reactions were predicted to occur in the same proportions as predicted for unlimited nitrate uptake . Flux through NADP-ME and supplementary flux through PEP-CK dropped proportionally , since restricting nitrogen limits the export of all metabolites from the system and reduced CO2 uptake is observed ( Figure 6—figure supplement 1 ) . Similarly , limiting water or CO2 uptake into the model resulted in overall reduced flux towards the phloem output ( Figure 6—figure supplement 1 ) but reactions were predicted to occur in the same proportions as predicted for unlimited uptake . Given that C4 plants sometimes optimise light availability to the bundle sheath ( Bellasio and Lundgren , 2016 ) we next explored light availability and light distribution . The model prediction is re-run with changes in the constraints , and the resulting tables of fluxes are queried for CO2 uptake and fluxes through the decarboxylation enzymes . In the experiment , we varied the total PPFD between 0 μmol/ ( m2s ) to 1000 μmol/ ( m2s ) and photon distribution in the range 0 . 1≤P⁢P⁢F⁢DB / P⁢P⁢F⁢DM≤2 , see Figure 6 . Under light limitation , if the total PPFD is lower than 400 μmol/ ( m2s ) , the CO2 uptake rate is reduced , leading to a decreased activity of the decarboxylation enzymes ( Figure 6 ( A ) ) . PEP-CK is used in the optimal solutions active under light-limiting conditions ( Figure 6 ( B ) ) . Under limiting light conditions , photon distribution with a higher proportion in the bundle sheath shifts decarboxylation towards NADP-ME but only to up to 26% . Under non-limiting conditions , the distribution of light availability determines the optimal decarboxylation enzyme . NADP-ME is the preferred decarboxylation enzyme with supplemental contributions by PEP-CK if light availability is near the threshold of 400 μmol/ ( m2s ) or if at least twice as many photons are absorbed by the mesophyll . Excess light availability and a higher proportion of photons reaching the bundle sheath leads to optimal solutions which favour PEP-CK as the decarboxylation enzyme . In the case of very high light availability and an abrupt shift towards the bundle sheath , NAD-ME becomes the optimal solution ( Figure 6 ( B ) ) . NAD-ME is the least favourable enzyme overall , only low activity is predicted under extreme light conditions , where the bundle sheath absorbs equal or more photons than the mesophyll ( Figure 6 ( B ) ) . PEP-CK complements the activity of NADP-ME and NAD-ME to 100% in many conditions , meaning the two-cell model also predicts the co-existence of PEP-CK/NADP-ME and PEP-CK/NAD-ME mode , while the flux distribution indicates no parallel use of NAD-ME and NADP-ME , compare Figure 6 ( B ) . Finally , we assumed that intercellular transport capacity for charged metabolites might be different between species . Assuming a fixed transport ratio between aspartate and malate ( Figure 6—figure supplement 1D ) introduces a shift in the C4 state . Higher proportions of malate exchange foster the use of NADP-ME ( Figure 6—figure supplement 1D ) . In contrast , higher portions of aspartate exchange foster the use of PEP-CK ( Figure 6—figure supplement 1D ) . To analyse evolution towards C4 photosynthesis based on C3 metabolism , a CBM of C3 metabolism is required ( Figure 1 ) . Design , simulation , validation cycles used current knowledge about plant biochemistry ( Heldt , 2015 ) to identify possible errors in the metabolic map required for modelling . Even after error correction ( Table 1 ) , a significant problem remained , namely excessive fluxes to balance protons in all compartments . This observation leads to the realisation that the biochemical knowledge about transport reactions does not extend to the protonation state of the substrates , which affects all eukaryotic CBM efforts . In plants , predominantly export and vacuolar transport reactions are directly or indirectly coupled with proton gradients to energise transport ( Bush , 1993; Neuhaus , 2007 ) . For chloroplasts and mitochondria , proton-coupled transport reactions have been described but may couple different metabolite transporters together rather than energising them ( Furumoto et al . , 2011 ) . Introducing proton sinks in all compartments solves the immediate modelling problem . However , intracellular transport reactions and their energetic costs are no longer correctly assessed by the model . Despite this band-aid fix which will be required for all eukaryotic constraint-based models which include proton-coupled transport reactions , the curated one-cell model correctly predicts energy usage and its distribution ( Figure 1—figure supplement 2 and Li et al . , 2017 ) . This indicates that in models which exclude vacuolar transport and energised export reactions , energy calculations remain likely within the correct order of magnitude . Overall , our one-cell model operates within parameters expected for a C3 plant: The predicted PPFD lies within the range of light intensities used for normal growth condition of Arabidopsis thaliana , which varies between 100 μmol/ ( m2s ) –200 μmol/ ( m2s ) , see Table 2 . The gross rate of O2 evolution for a PPFD of 200 μmol/ ( m2s ) is estimated to be 16 . 5 μmol/ ( m2s ) in the literature ( Sun et al . , 1999 ) , which is in close proximity to the predicted flux of the one-cell model , see Table 2 . For the amount of respiratory ATP that is used for maintenance , ( Li et al . , 2017 ) predicted an even lower proportion of energy 16% , see Figure 1—figure supplement 2 . The model’s flux map is in accordance with known C3 plant physiology ( Heldt , 2015 ) , and its input and output parameters match expected values ( Figure 2 ( B ) ) . The current model excludes specialised metabolism since the output function focuses solely on substances exported through the phloem in a mature leaf . If the model were to be used to study biotic interactions in the future , the addition of specialised metabolism in the metabolic map and a new output function would be required . Most evolutionary concepts about C4 photosynthesis assume that selective pressure drives pathway evolution due to photorespiration and carbon limitation ( Heckmann et al . , 2013 ) . Most extant C4 species occupy dry and arid niches ( Edwards et al . , 2010 ) , even more , the period of C4 plant evolution was accompanied with an increased oxygen concentration in the atmosphere ( Sage , 2004 ) . Therefore , it is frequently assumed that carbon limitation by excessive photorespiration drives the evolution of C4 photosynthesis . Yet , in most habitats plants are limited by nutrients other than carbon ( Agren et al . , 2012; Körner , 2015 ) . Ecophysiological analyses also show that C4 can evolve in non-arid habitats ( Liu and Osborne , 2015; Lundgren and Christin , 2017; Osborne and Freckleton , 2009 ) . To resolve this apparent contradiction , we tested whether resource limitation may also lead to the evolution of a C4 cycle . We optimised the model approximating resource limitation via an objective function for total minimal flux at different photorespiratory levels . Indeed , with increasing photorespiration , the optimisation for resource efficiency leads to the emergence of the C4 cycle as the optimal solution . Balancing the resource cost of photorespiration against the resource cost of the C4 cycle , the model predicts that N limitation may have facilitated C4 evolution given high levels of photorespiration . Other possible selective pressures such as biotic interactions can currently not be tested using the model since specialised metabolism is not included in the metabolic map or the output function . Extant C4 species have higher C : N ratios reflecting the N-savings the operational C4 cycle enables ( Sage et al . , 1987 ) . The photorespiratory pump using glycine decarboxylase based CO2 enrichment also emerges from the model , showing that C2 photosynthesis is also predicted under simple resource limitation . Indeed N-savings have been reported from C2 plants compared with their C3 sister lineages ( Schlüter et al . , 2016a ) . Simply minimising photorespiration as the objective function also yields C4 photosynthesis as the optimal solution . Hence , two alternatively or parallelly acting selective pressures towards C4 photosynthesis , limitation in C and/or N , are identified by the model . In both cases , the model correctly predicts the C4 cycle of carboxylation and decarboxylation and the C2 photorespiratory pump as observed in extant plants . The evolution of C4 photosynthesis in response to multiple selective pressures underscores its adaptive value and potential for agriculture . Intermediacy also evolves indicating that it , too , is likely an added value trait which could be pursued by breeding and engineering efforts . The optimal solutions for the metabolic flux patterns predict an intermediate stage in which CO2 transport via photorespiratory intermediates glycolate and glycerate ( Figure 3 ( E ) and ( F ) ) and decarboxylation by glycine decarboxylase complex ( Figure 3 ( B ) ) is essential . All of the models of C4 evolution ( Monson , 1999; Bauwe , 2010; Sage et al . , 2012; Heckmann et al . , 2013; Williams et al . , 2013 ) predict that the establishment of a photorespiratory CO2 pump is an essential intermediate step towards the C4 cycle . The photorespiratory CO2 pump , also known as C2 photosynthesis , relocates the photorespiratory CO2 release to the bundle sheath cells . Plants using the photorespiratory CO2 pump are often termed C3-C4 intermediates owing to their physiological properties ( Sage et al . , 2012 ) . Displaying the flux solution in Figure 3 on a metabolic map in Figure 3—figure supplement 1 clearly illustrates that increasing photorespiratory flux through Rubisco drives the two-cell metabolic model from C3 to C4 metabolism by passing the C3-C4 intermediate state . On the C3-C4 trajectory , the activity of Rubisco is shifted from the mesophyll to the bundle sheath , as well as from the constrained to the CCM-dependent Rubisco population as a consequence of the increased costs of photorespiration under increased pO2:pC⁢O2 ratio , see Equation 5 . The increase of the oxygenation rate in the photorespiration constraint drives the reprogramming of the metabolism to avoid oxygenation by establishing the C4 cycle . Therefore , our analysis recovers the evolutionary C3-C4 trajectory and confirms the emergence of a photorespiratory CO2 pump as an essential step during the C4 evolution also under optimisation for resources ( Heckmann et al . , 2013 ) . The model may also provide a reason for why some plant species have halted their evolution in this intermediary phase ( Scheben et al . , 2017 ) . Under the conditions of resource limitations and intermediate photorespiration , the model predicts intermediacy as the optimal solution . In a very narrow corridor of conditions , no further changes are required to reach optimality and the model thus predicts that a small number of species may remain intermediate . Since the model predicts C4 metabolism without specific constraints , different input and reaction constraints can be tested for their influence on the molecular nature of the C4 cycle . This approach may identify the selective pressure and boundaries limiting evolution . Initial optimisation without additional constraints or input limitations predict a C4 cycle based on decarboxylation by NADP-ME ( Figure 3 and Figure 3—figure supplement 1 ( A ) ) . This prediction recapitulates intuition; the NADP-ME based C4 cycle is considered the 'most straight forward' incarnation of C4 photosynthesis , it is always explained first in textbooks and is a major focus of research . The NADP-ME based cycle thus represents the stoichiometrically optimal solution when resource limitation or photorespiration are considered . Once NADP-ME is no longer available via constraint , PEP-CK and NAD-ME become optimal solutions albeit with a prediction of malate and pyruvate as the transfer acids ( Figure 6 ) . The FVA identified aspartate and alanine as slightly less optimal solutions ( Figure 5 ) . Since in vivo this slightly less optimal solution has evolved in all NAD-ME origins tested to date , kinetic rather than stoichiometric reasons suggest themselves for the use of aspartate and alanine ( Bräutigam et al . , 2018 ) . Since all extant C3 species and therefore also the ancestors of all C4 species contain all decarboxylation enzymes ( Aubry et al . , 2011 ) , it is unlikely that unavailability of an enzyme is the reason for the evolution of different decarboxylation enzymes in different origins ( Sage , 2004 ) . Stochastic processes during evolution , that is up-regulation of particular enzyme concentrations via changes in expression and therefore elements cis to the gene ( Bräutigam and Gowik , 2016 ) , may have played a role in determining which C4 cycle evolved . Alternatively , environmental determinants may have contributed to the evolution of different C4 cycles . Physiological experiments have pointed to a connection between nitrogen use efficiency and type of decarboxylation enzyme ( Pinto et al . , 2016 ) . Hence the variation in nitrogen input to the model was tested for their influence on optimal solutions with regard to decarboxylation enzymes . Input limitation of nitrogen , water as a metabolite , and CO2 limited the output of the system but did not change the optimal solution concerning decarboxylation Figure 6—figure supplement 1 making it an unlikely candidate as the cause . Differences in nitrogen use is possibly a consequence of decarboxylation type . In some grasses , light penetrable cells overlay the vascular bundle leading to different light availability ( summarised in Bellasio and Lundgren , 2016 and Karabourniotis et al . , 2000 ) and hence light availability and distribution were tested ( Figure 6 ( B ) ) . Changes in light input and distribution of light input between mesophyll and bundle sheath indeed altered the optimal solutions ( Figure 6 ( B ) ) . The changes in the solution can be traced to the energy status of the plant cells . For very high light intensities , the alternative oxidases in the mitochondria are used to dissipate the energy and hence a path towards NAD-ME is paved . Under light limitation , the C4 cycle requires high efficiency and hence PEP-CK which , at least in part allows energy conservation by using PEP rather than pyruvate as the returning C4 acid , is favoured . Interestingly , the sensitivity of different species towards environmental changes in light is influenced by the decarboxylation enzyme present ( Sonawane et al . , 2018 ) . NADP-ME species are less compromised compared to NAD-ME species by shade possibly reflecting an evolutionary remnant as NAD-ME is predicted to emerge only in high light conditions . PEP-CK is more energy efficient compared to malic enzyme based decarboxylation which requires PEP recycling by PPDK at the cost of two molecules of ATP ( Figure 3 ( D ) ) . Notably , two C4 plants known to rely on PEP-CK P . maximum and A . semialata ( African accessions ) are shade plants which grow in the understory ( Lundgren and Christin , 2017 ) . PEP-CK can be co-active with NADP-ME and NAD-ME ( Figure 6 ( B ) ) . This co-use of PEP-CK with a malic enzyme has been shown in C4 plants ( Pick et al . , 2011; Wingler et al . , 1999 ) and explained as an adaptation to different energy availability and changes in light conditions ( Pick et al . , 2011; Bellasio and Griffiths , 2014 ) . Dominant use of PEP-CK in the absence of malic enzyme activity as suggested ( Figure 3 ( B ) , Figure 3—figure supplement 1 and Figure 4 ) is rare in vivo ( Ueno and Sentoku , 2006 ) but observed in P . maximum and in A . semialata . While the model predictions are in line with ecological observations , we cannot exclude that kinetic constraints ( i . e . [Bräutigam et al . , 2018] ) may also explain why a stoichiometrically optimal solution such as the NADP-ME cycle is not favoured in nature where NADP-ME and NAD-ME species evolve in nearly equal proportions ( Sage , 2004 ) . CBM of photosynthetically active plant cells revealed a major knowledge gap impeding CBM , namely the unknown protonation state of most transport substrates during intracellular transport processes . When photoautotrophic metabolism was optimised in a single cell for minimal metabolic flux and therefore , optimal resource use , C3 photosynthetic metabolism was predicted as the optimal solution . Under low photorespiratory conditions , a two-celled model which contains a CCM-dependent Rubisco optimised for resource use , still predicts C3 photosynthesis . However , under medium to high photorespiratory conditions , a molecularly correct C4 cycle emerged as the optimal solution under resource limitation and photorespiration reduction as objective functions which points to resource limitation as an additional driver of C4 evolution . Light and light distribution was the environmental variable governing the choice of decarboxylation enzymes . Modelling compartmented eukaryotic cells correctly predicts the evolutionary trajectories leading to extant C4 photosynthetic plant species . Flux balance analysis ( FBA ) is a CBM approach ( Orth et al . , 2010 ) to investigate the steady-state behaviour of a metabolic network defined by its stoichiometric matrix S . By employing linear programming , FBA allows computing an optimised flux distribution that minimises and/or maximises the synthesis and/or consumption rate of one specific metabolite or a combination of various metabolites . Next to the steady-state assumption and stoichiometric matrix S , FBA relies on the definition of the reaction directionality and reversibility , denoted by the lower bound vm⁢i⁢n and upper bound vm⁢a⁢x , as well as the definition of an objective function z . The objective function z defines a flux distribution v , with respect to an objective c . ( 1 ) min/maxzFBA=cTvs . t . S⋅v=0vmin≤v≤vmax The degeneracy problem , the possible existence of alternate optimal solutions , is one of the major issues of constraint-based optimisation , such as FBA ( Mahadevan and Schilling , 2003 ) . To avoid this problem , we use the parsimonious version of FBA ( pFBA ) ( Lewis et al . , 2010 ) . This approach incorporates the flux parsimony as a constraint to find the solution with the minimum absolute flux value among the alternative optima , which is in agreement with the assumption that the cell is evolutionary optimised to allocate a minimum amount of resources to achieve its objective . ( 2 ) min/maxzp⁢F⁢B⁢A=∑\|vi\|s . t . S⋅v=0vm⁢i⁢n≤v≤vm⁢a⁢xcT⁢v=zF⁢B⁢A All FBA experiments in this study employ pFBA and are performed using the cobrapy module in a python 2 . 7 environment run on a personal computer ( macOS Sierra , 4 GHz Intel Core i7 , 32 GB 1867 MHz DDR3 ) . All FBA experiments are available as jupyter notebooks in the supplementary material and can also be accessed and executed from the GitHub repository https://github . com/ma-blaetke/CBM_C3_C4_Metabolism ( Blätke , 2019; copy archived at https://github . com/elifesciences-publications/CBM_C3_C4_Metabolism ) .	Virtually all plants use energy from sunlight to convert carbon dioxide and water into oxygen and sugars via a process called photosynthesis . This process has many steps that each rely on different enzymes to drive specific chemical reactions . Most plants use a pathway of enzymes that is referred to as C3 photosynthesis . Plants absorb carbon dioxide gas from the atmosphere . However , the levels of carbon dioxide in the atmosphere are very low , so this limits the amount of photosynthesis plants can perform . To overcome this problem , some plants have evolved a different type of photosynthesis – called C4 photosynthesis – with a mechanism that increases the levels of carbon dioxide in the cells . Today , plants that use C4 photosynthesis ( so-called ‘C4 plants’ ) typically grow faster than other plants , especially in warmer climates . This gives C4 plants , such as corn , an advantage over their competitors and also helps them to colonize harsh environments that other plants struggle to thrive in . However , it remains unclear how C4 photosynthesis evolved in some plants living in wet habitats , or why other plants use forms of photosynthesis that are intermediate between C4 and C3 photosynthesis . C4 photosynthesis uses pathways containing enzymes that are found in all plants; therefore , C4 plants evolved by changing how they used enzymes they already had . To understand how these different enzyme pathways may have evolved , Blätke and Bräutigam used an approach known as constraint-based modelling . The researchers built a mathematical model of C3 photosynthesis and used it to predict the optimal enzyme pathways ( for example , pathways involving the fewest enzymes or requiring the least energy ) for photosynthesis under particular conditions . The model predicted that , in addition to shortages in carbon dioxide , shortages in an important plant nutrient known as nitrogen may have driven the evolution of C4 photosynthesis . Furthermore , enzyme pathways that were intermediate between C3 and C4 photosynthesis were predicted to be optimal solutions under particular conditions . Together , the findings of Blätke and Bräutigam may explain why different variations of C4 photosynthesis exist in plants . These findings could be used to breed crops that use the most efficient type of photosynthesis for the conditions they are grown in , leading to better yields .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "computational", "and", "systems", "biology" ]	2019	Evolution of C4 photosynthesis predicted by constraint-based modelling
Stressful experiences potently activate kappa opioid receptors ( κORs ) . κORs in the ventral tegmental area regulate multiple aspects of dopaminergic and non-dopaminergic cell function . Here we show that at GABAergic synapses on rat VTA dopamine neurons , a single exposure to a brief cold-water swim stress induces prolonged activation of κORs . This is mediated by activation of the receptor during the stressor followed by a persistent , ligand-independent constitutive activation of the κOR itself . This lasting change in function is not seen at κORs at neighboring excitatory synapses , suggesting distinct time courses and mechanisms of regulation of different subsets of κORs . We also provide evidence that constitutive activity of κORs governs the prolonged reinstatement to cocaine-seeking observed after cold water swim stress . Together , our studies indicate that stress-induced constitutive activation is a novel mechanism of κOR regulation that plays a critical role in reinstatement of drug seeking . Stress has long been known to be a precipitating factor for the abuse of addictive drugs . Animal models have shown that acute and repeated stressors can escalate intake of addictive substances ( Piazza et al . , 1990; Ramsey and Van Ree , 1993; Goeders and Guerin , 1994; Shaham and Stewart , 1994; Haney et al . , 1995 ) , and that acute stress can reinstate drug seeking in animals that have undergone extinction training ( Shaham et al . , 1994 , 1995; Conrad et al . , 2010; Mantsch et al . , 2016 ) . In recent years , dopaminergic neurons of the VTA have emerged as a significant locus for the overlapping effects of drugs of abuse and stress ( Polter and Kauer , 2014 ) . Synaptic inputs , by shaping the activity of these neurons , are poised to play an important role in drug seeking . Both acute stress and exposure to drugs of abuse induce a concomitant potentiation of excitatory synapses and loss of long term potentiation of inhibitory synapses ( Ungless et al . , 2001; Saal et al . , 2003; Kauer and Malenka , 2007; Nugent et al . , 2007; Chen et al . , 2008; Niehaus et al . , 2010; Polter and Kauer , 2014 ) . Understanding how these synapses are altered by stress will provide key insights into stress-induced drug seeking and provide targets for treating substance use disorders . A major mediator of stress-induced changes in inhibitory VTA synapses is the dynorphin/kappa opioid receptor ( κOR ) system . κORs , and their endogenous ligand , dynorphin , are found throughout the brain and have been highly associated with stressful , aversive , and dysphoric experiences ( Bruchas et al . , 2010; Wee and Koob , 2010; Van't Veer and Carlezon , 2013; Crowley and Kash , 2015 ) . Within the VTA , κORs have a range of physiological effects . κORs decrease the firing rate of dopamine neurons through activation of GIRK channels ( Margolis et al . , 2003 , 2006 ) , inhibit excitatory synaptic transmission onto both dopaminergic and non-dopaminergic VTA neurons ( Margolis et al . , 2005 ) , reduce inhibitory synaptic transmission in a subset of dopamine neurons ( Ford et al . , 2006 ) and inhibit somatodendritic dopaminergic IPSCs ( Ford et al . , 2007 ) . VTA κORs also can control the interactions between stress and reward . Our previous work identified a form of stress-sensitive synaptic plasticity at inhibitory synapses on VTA dopamine neurons ( LTPGABA; Nugent et al . , 2007 , 2009; Niehaus et al . , 2010 ) . LTPGABA is induced via activation of nitric oxide synthase in the dopamine neuron , leading to nitric oxide ( NO ) release , and enhancement of GABA release through cGMP signaling ( Nugent et al . , 2007 , 2009 ) . More recently , we showed that acute stress blocks LTPGABA through activation of κORs , and that preventing this activation via intra-VTA administration of the κOR antagonist , nor-binaltorphimine ( norBNI ) , prevents stress-induced reinstatement of cocaine-seeking ( Graziane et al . , 2013 ) . Remarkably , a single exposure to stress leads to a loss of LTPGABA that lasts for at least five days and is mediated by persistent activation of VTA κORs ( Polter et al . , 2014 ) . We have also shown that treatment with the κOR antagonist after stress can rescue stress-induced reinstatement . These studies highlight the importance of κOR-mediated regulation of LTP at GABAergic synapses in stress-induced drug seeking and underscore the need to better understand the mechanism of this unique and persistent regulation . In the present study , we have now identified the mechanism by which activation of κORs and suppression of LTPGABA in the VTA is maintained for multiple days after an acute , severe stressor . We present evidence that stress blocks LTPGABA by inducing constitutive activation of κORs at VTA inhibitory synapses rather than through persistent increases in dynorphin release . This constitutive activity is likely to be triggered initially by signaling through the endogenous ligand dynorphin , but then is persistently maintained independently of dynorphin release . In parallel , we find that the persistent drug-seeking induced by a single exposure to acute stress is also dependent on constitutive activity of κORs . Our results reveal a novel mechanism of experience-dependent regulation of κOR function , and emphasize the essential role of κORs in mediating stress-induced changes in synaptic plasticity and drug-seeking behavior . As previously shown , bath application of the nitric oxide donor SNAP potentiates GABAergic synapses on dopamine neurons in the VTA , similarly to high-frequency stimulation of VTA afferents; this potentiation is blocked by single exposure to multiple drugs of abuse or acute cold-water swim stress ( LTPGABA; Nugent et al . , 2007; Niehaus et al . , 2010; Graziane et al . , 2013; Polter et al . , 2014; Figure 1A–B ) . Our recent studies indicate that blocking κORs with norNBI prevents and reverses the effects of acute stress on LTPGABA , even when administered several days after stress ( Graziane et al . , 2013; Polter et al . , 2014 ) . We therefore investigated whether stress-induced , persistent activation of κORs could be detected ex vivo in the midbrain slice . We subjected rats to acute cold water forced swim stress and prepared midbrain slices 24 hr later . If after stress , κORs in the VTA are persistently signaling in vitro , we reasoned that bath-applied norBNI could be used to rescue SNAP-induced LTPGABA . Bath application of norBNI ( 100 nM ) indeed allowed us to elicit NO-dependent LTPGABA in slices from stressed animals ( Figure 1E ) , indicating that stress-induced activity of κORs in the VTA persists through brain slice preparation and recovery . It seemed unlikely that sufficient endogenous dynorphin could be released tonically from the denervated brain slices to maintain a block of LTPGABA in vitro . We therefore next sought to establish the mechanism by which norBNI rescued this plasticity . In addition to competing with agonists at the κOR agonist binding site , norBNI acts as an inverse or collateral agonist , and its interactions with the κOR can non-competitively inhibit further activity of κORs via activation of the JNK signaling cascade ( Bruchas et al . , 2007; Melief et al . , 2010 , 2011 ) . We hypothesized that the rescue of LTPGABA by norBNI might also occur non-competitively via JNK signaling ( Figure 1C ) . Slices were treated with the JNK inhibitor SP600125 ( 20 µM ) for 10 min prior to bath application of norBNI ( Figure 1D ) . In contrast to the robust SNAP-induced potentiation observed in slices treated with norBNI alone , we found that LTPGABA remained blocked in slices pretreated with SP600125 ( Figure 1F–H ) . Importantly , bath application of SP600125 did not interfere with expression of LTPGABA in slices from naïve animals or the loss of LTPGABA in slices from stressed animals ( Figure 1—figure supplement 1A–B ) . Therefore , JNK activity has no role in LTPGABA induction or in the block of this plasticity by κORs , but is required for norBNI to rescue LTPGABA following stress . 10 . 7554/eLife . 23785 . 003Figure 1 . norBNI rescues LTPGABA through activation of JNK . ( A ) Summary data showing the blockade of LTPGABA after stress . ( B ) Comparison of the magnitude of LTPGABA10–15 min after SNAP application . ( IPSC amplitudes , control: 140 ± 5% of baseline values , n = 13; stress: 94 ± 11% of baseline values , n = 6; unpaired t-test , p=0 . 0005 . ( C ) Schematic of norBNI’s competitive and non-competitive inhibition of κOR signaling . ( D ) Experimental design . ( E ) Representative single experiment showing that bath application of norBNI ( 100 nM ) rescues LTPGABA in a slice prepared 24 hr after stress . ( F ) Representative single experiment from a slice prepared 24 hr after stress showing that norBNI does not rescue LTPGABA in the presence of the JNK inhibitor SP600125 ( 20 µM ) . ( G ) Summary data from both groups . ( H ) Comparison of the magnitude of LTPGABA10–15 min after SNAP application . ( IPSC amplitudes , norBNI only: 139 ± 7% of baseline values , n = 6; norBNI+SP600125: 106 ± 9% of baseline values , n = 11; unpaired t-test , p=0 . 029 . ) Insets for this and all figures: IPSCs before ( black trace , control ) and 15 min after drug application ( red trace , SNAP , 400 µM ) . Scale bars: 20 ms , 100 pA . Insets are averages of 12 IPSCs . DOI: http://dx . doi . org/10 . 7554/eLife . 23785 . 00310 . 7554/eLife . 23785 . 004Figure 1—figure supplement 1 . Inhibition of JNK does not affect LTPGABA or its block by stress in the absence of norBNI . ( A ) Summary data showing that LTPGABA is expressed in slices from naïve animals in the presence of the JNK inhibitor SP600125 ( 20 µM ) . Average magnitude of LTPGABA10–15 min after SNAP = 144 ± 13% of baseline values , n = 5; one-sample t-test , p=0 . 0257 . ( B ) Summary data showing that LTPGABA remains blocked in slices from stressed animals in the presence of the JNK inhibitor SP600125 ( 20 µM ) . Average magnitude of LTPGABA10–15 min after SNAP = 111 ± 11% of baseline values , n = 6; one-sample t-test , p=0 . 38 . DOI: http://dx . doi . org/10 . 7554/eLife . 23785 . 004 Our data indicate that following stress , κOR activation persists even in the brain slice , and is rescued in a JNK-dependent manner . This suggests that non-competitive actions of norBNI , rather than its block of dynorphin binding , are relevant to the loss of LTPGABA . To test this hypothesis further , we again utilized pharmacological tools in slices from stressed animals . We treated such slices with either norBNI or 6β-naltrexol , a neutral antagonist that only inhibits agonist-stimulated κOR activity ( Figure 2A–B; Wang et al . , 2007 ) . If norBNI rescues LTPGABA only because it can activate JNK signaling , we would predict that a neutral antagonist that only inhibits κOR agonist binding would be ineffective ( Figure 2A , Wang et al . , 2007 ) . While norBNI treatment rescued LTPGABA , bath application of the neutral antagonist did not reverse the stress-induced block of LTPGABA ( Figure 2C–F ) . Bath application of 6β-naltrexol was sufficient to prevent depression of EPSCs onto VTA dopamine neurons induced by the κOR agonist U50488 ( Figure 2—figure supplement 1B , Margolis et al . , 2005 ) , indicating that this concentration of the drug is sufficient to block κORs in the VTA slice . 6β-naltrexol did not have any effects on basal inhibitory transmission in slices from stressed or naïve rats ( Figure 2—figure supplement 1A ) . These results show that a κOR competitive antagonist cannot effectively rescue LTPGABA following stress , and suggest that the persistent block of LTPGABA is maintained by constitutive activation of κORs in the VTA rather than a prolonged increase in dynorphin release . 10 . 7554/eLife . 23785 . 005Figure 2 . The neutral antagonist 6β-naltrexol fails to rescue LTPGABA in slices from stressed animals . ( A ) Schematic of norBNI and 6β-naltrexol inhibition of κOR signaling . ( B ) Experimental design . ( C ) Representative experiment showing that bath application of norBNI ( 100 nM ) rescues LTPGABA in a slice prepared 24 hr after stress . ( D ) Representative experiment from a cell 24 hr after stress showing that 6β-naltrexol ( 10 µM ) fails to rescue LTPGABA . ( E ) Summary data from both groups . ( F ) Comparison of the magnitude of LTPGABA10–15 min after SNAP application . ( IPSC amplitudes , norBNI: 141 ± 20% of baseline values , n = 10; 6β-naltrexol: 100 ± 8% of baseline values , n = 10; unpaired t-test , p=0 . 048 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23785 . 00510 . 7554/eLife . 23785 . 006Figure 2—figure supplement 1 . 6β-naltrexol does not affect basal inhibitory synaptic transmission but does block κORs . ( A ) Summary data showing that 6β-naltrexol ( 10 µM ) does not affect basal inhibitory transmission in cells from control or stressed rats . Normalized IPSC amplitude 5–10 min after 6β-naltrexol: control = 97 ± 6% of baseline values , n = 6 ( 4 animals ) ; stress = 92 ± 17% of baseline values , n = 4 ( 2 animals ) ; unpaired t-test p=0 . 79 . ( B ) Summary data showing that U50488 depresses EPSC amplitudes from VTA dopamine neurons and that 6β-naltrexol ( 10 µM ) prevents this depression . Normalized EPSC amplitude 5–10 min after U50488: U50488 alone = 75 ± 2% of baseline values , n = 3; U50488 + 6β-naltrexol = 95 ± 1% of baseline values , n = 4; unpaired t-test p=0 . 0003 . DOI: http://dx . doi . org/10 . 7554/eLife . 23785 . 006 How might acute stress cause constitutive activation of κORs ? While the results of our slice experiments rule out a requirement for elevated dynorphin in maintaining persistent activity of VTA κORs following stress , dynorphin release during or immediately following stress may be needed to trigger a change in the receptor leading to prolonged constitutive activation . If this model is correct , preventing binding of dynorphin to the κOR during stress would prevent the loss of LTPGABA . However after stress , when the block of LTPGABA is no longer dynorphin-dependent , preventing dynorphin binding would not rescue LTPGABA . To test this idea , we treated animals with the competitive antagonist 6β–naltrexol either 30 min before or one day after FSS ( Figure 3A ) . Consistent with our hypothesis , cells from animals treated with 6β-naltrexol before stress exhibited LTPGABA , while those treated one day after stress did not , similarly to the vehicle-treated animals ( Figure 3B-F ) . In contrast , our previous studies have shown that treating rats with norBNI at the same time point after stress ( one day ) rescues LTPGABA ( Polter et al . , 2014 ) . These data strongly support the idea that the persistent block of LTPGABA following acute swim stress is mediated by dynorphin-dependent activation of the κOR followed by a transition to dynorphin-independent constitutive activity of the receptor . 10 . 7554/eLife . 23785 . 007Figure 3 . 6β-naltrexol rescues LTPGABA when administered pre-stress , but not post-stress . ( A ) Experimental design . ( B ) Representative experiment showing that a cell from a vehicle-treated stressed animal does not exhibit LTPGABA . ( C ) Representative experiment showing that a cell from an animal treated with 6β-naltrexol ( 10 mg/kg ) 30 min pre-stress exhibits LTPGABA . ( D ) Representative experiment showing that a cell from an animal treated with 6β-naltrexol 24 hr post-stress does not exhibit LTPGABA . ( E ) Summary data showing compiled data from all groups . ( F ) Comparison of the magnitude of LTPGABA 10–15 min after SNAP application . ( 1-way ANOVA followed by Dunnett’s multiple comparison test . F2 , 30=4 . 231 , p=0 . 024 . IPSC amplitudes , 6β-naltrexol pre-stress: 136 ± 12% of baseline values , n = 12 , p<0 . 05 from vehicle; 6β-naltrexol post-stress: 100 ± 9% of baseline values , n = 11 , n . s . from vehicle; vehicle+stress: 102 ± 8% of baseline values , n = 10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23785 . 007 To investigate whether a brief activation of κORs is sufficient to produce persistently activated κORs , we treated rats with a single dose of the κOR agonist , U50488 , and measured LTPGABA at various time points after injection ( Figure 4A ) . Upon injection , U50488 rapidly enters the CNS and is metabolized and undetectable in the brain by 24 hr after administration ( Russell et al . , 2014 ) , and we therefore expect that within our experimental time frame , U50488 was no longer occupying the κOR . In neurons from saline-treated animals , bath application of SNAP robustly potentiated IPSCs ( Figure 4B ) . In contrast , SNAP was unable to elicit LTPGABA in neurons from rats either one or five days after U50488 administration ( Figure 4C–F ) . Notably , this time course closely mirrors that of the in vivo block of LTPGABA following acute stress ( Polter et al . , 2014 ) . 10 . 7554/eLife . 23785 . 008Figure 4 . Single treatment with a κOR agonist leads to prolonged blockade of LTPGABA . ( A ) Experimental design . ( B ) Representative experiment showing that a cell from a saline-treated animal exhibits LTPGABA . ( C ) Representative single experiment showing a cell prepared 24 hr after a single treatment with U50488 ( 5 mg/kg ) does not exhibit LTPGABA . ( D ) Representative experiment showing that a cell prepared five days after a single treatment with U50488 does not exhibit LTPGABA . ( E ) Summary data from all groups . ( F ) Comparison of the magnitude of LTPGABA10–15 min after SNAP application . ( 1-way ANOVA followed by Dunnett’s multiple comparison test . F2 , 27=12 . 21 , p=0 . 0002 . IPSC amplitudes , Saline: 137 ± 6% of baseline values , n = 11; U50488 1 day: 108 ± 6% of baseline values , n = 9 , p<0 . 05 vs . saline; U50488 5 days: 100 ± 5% of baseline values , n = 10 , p<0 . 05 vs . saline ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23785 . 008 We next addressed the question of whether κORs at other brain synapses are also persistently activated after acute stress . Bath application of the κOR agonist U69593 has been reported to depress the amplitude of glutamatergic EPSCs in both VTA Ih positive ( presumptive dopamine neurons ) and Ih negative ( presumptive non-dopamine neurons ) , and norBNI reverses this depression ( Margolis et al . , 2005 ) . Therefore if κORs at excitatory synapses become constitutively activated after swim stress , reducing their activity with norBNI should be detectable as potentiation of excitatory VTA synapses . To test this , we prepared VTA slices 24 hr after FSS . We recorded EPSCs from Ih positive and Ih negative neurons from both stressed and unstressed animals and bath-applied norBNI . NorBNI had no effect on EPSC amplitude in Ih-positive neurons in slices from either naïve or stressed animals ( Figure 5A–C ) , and norBNI did not increase EPSC amplitudes in VTA Ih-negative neurons in slices from either naïve or stressed animals ( Figure 5D–F ) . Therefore , the persistent constitutive κOR activation we observe at GABAergic synapses after acute stress does not occur at all κORs , even within the VTA . 10 . 7554/eLife . 23785 . 009Figure 5 . κORs at VTA excitatory synapses are not constitutively activated by stress . ( A ) Representative experiment showing that norBNI ( 100 nM ) does not potentiate excitatory synapses on Ih+ VTA neurons in a slice prepared from a control animal . ( B ) Representative experiment showing that norBNI does not potentiate excitatory synapses on Ih+ VTA neurons in a slice prepared from a stressed animal . ( C ) Summary data from Ih+ neurons . No significant difference in IPSC amplitude 10–15 min after norBNI application ( t-test p=0 . 81 IPSC amplitudes , control: 94 ± 2% of baseline values , n = 5; stressed: 92 ± 6% of baseline values , n = 6 ) . ( D ) Representative experiment showing that norBNI does not potentiate excitatory synapses on Ih− VTA neurons in a slice prepared from a control animal . ( E ) Representative single experiment showing that norBNI does not potentiate excitatory synapses on Ih− VTA neurons in a slice prepared from a stressed animals . ( F ) Summary data from Ih− neurons . No significant difference in IPSC amplitude 10–15 min after norBNI application ( t-test p=0 . 49 IPSC amplitudes , control: 112 ± 2% of baseline values , n = 5; stressed: 110 ± 4% of baseline values , n = 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23785 . 009 Numerous studies from our lab and others have shown a close association between κOR activation and stress-induced drug-seeking behavior ( McLaughlin et al . , 2003; Redila and Chavkin , 2008; Land et al . , 2009; Wee and Koob , 2010; Graziane et al . , 2013; Zhou et al . , 2013; Polter et al . , 2014 ) . We recently reported that blocking κORs with norBNI reverses the modest but prolonged reinstatement of cocaine-seeking induced by swim stress ( Conrad et al . , 2010; Graziane et al . , 2013 ) . This rescue is seen even when norBNI is administered two hours after stress ( Polter et al . , 2014 ) . These findings are consistent with the hypothesis that reinstatement of cocaine-seeking after swim stress requires activation of VTA κORs and suppression of LTPGABA . Having now shown that the blockade of LTPGABA by swim stress is dependent on constitutive activity of κORs , we next tested whether reinstatement of cocaine seeking is similarly dependent on constitutively active κORs . Rats were trained to self-administer cocaine for a minimum of 10 days . Animals then underwent extinction training , and after the final extinction session , they were subjected to forced swim stress , and then returned to their home cages . Twenty-four hours after stress , one group of animals was treated with norBNI and a second group with saline . A third group was treated with 6β–naltrexol 2 days after stress and 60 min prior to reinstatement testing ( Figure 6A ) . Due to the differing pharmacokinetic profiles of norBNI and 6β–naltrexol , time of administration was varied to optimize block of the κOR during the reinstatement test and to ensure that all animals were tested for reinstatement at the same time point ( Endoh et al . , 1992; Raehal et al . , 2005 ) ; thus , all animals were tested for reinstatement 48 hr after stress . 10 . 7554/eLife . 23785 . 010Figure 6 . Post-stress rescue of reinstatement by norBNI but not 6β-naltrexol . ( A ) Experimental design . ( B ) Lever pressing during the final extinction session ( white bar ) and reinstatement session ( colored bar ) . Saline ( black ) : last extinction session: 6 . 4 ± 1 . 7 lever presses; reinstatement session: 13 . 8 ± 2 . 4 lever presses; n = 8 , p=0 . 011 , paired t-test . norBNI ( green ) : last extinction session: 4 . 8 ± 1 . 4 lever presses; reinstatement session: 5 . 2 ± 1 . 1 lever presses; n = 6 , p=0 . 76 , paired t-test . 6β-naltrexol ( blue ) : last extinction session: 5 . 7 ± 1 . 8 lever presses; reinstatement session: 11 . 2 ± 3 . 4 lever presses; n = 11 , p=0 . 033 , paired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 23785 . 010 As previously shown , after acute stress , vehicle-treated animals showed a significant elevation of lever pressing compared to the final extinction session ( Figure 6B ) . Although the reinstatement was modest , this was measured two full days after the stress , demonstrating the prolonged increase in cocaine-seeking ( Conrad et al . , 2010 ) . In contrast , animals given norBNI 24 hr post-stress did not increase their lever pressing two days after stress ( Figure 6B ) . Moreover , the neutral antagonist 6β-naltrexol did not prevent reinstatement , as 6β–naltrexol treated animals significantly increased lever pressing compared to the final extinction session ( Figure 6B ) . These data suggest that while persistent activation of κORs underlies the prolonged reinstatement induced by swim stress , this is mediated by constitutively active receptors rather than by long-term increases in the level of the endogenous ligand dynorphin . It is widely accepted that GPCRs can adopt agonist-independent conformations that are constitutively active ( Seifert and Wenzel-Seifert , 2002; Sadée et al . , 2005;Young et al . , 2013 Meye et al . , 2014 ) . In addition to κORs , the other members of the opioid receptor subfamily , µOR and δOR , have both been shown to exhibit constitutive activity ( Wang et al . , 1994 , 2004; Chiu et al . , 1996; Wang et al . , 1999; Liu and Prather , 2001; Wang et al . , 2007; Corder et al . , 2013 ) . κORs themselves have been shown to exhibit constitutive activity , both in heterologous expression systems ( Becker et al . , 1999; Wang et al . , 2007 ) and in the rat PFC ( Sirohi and Walker , 2015 ) . A decrease in fear and anxiety behaviors in rats after acute footshock has also been reported that is reduced by post-shock norBNI , suggesting the possibility of constitutive κOR activation , although a high dose was required and the effect of norBNI was not compared to a neutral antagonist ( Rogala et al . , 2012 ) . Very little is known , however , about the processes that regulate transitions between constitutively active and inactive states , presumably representing distinct receptor conformations ( Seifert and Wenzel-Seifert , 2002; Sadée et al . , 2005 ) . We present two critical pieces of data indicating that stress induces constitutive activity of κORs . First , brief application ( ~15 min ) of a κOR inverse agonist to VTA slices from stressed rats rescues LTPGABA in a JNK-dependent fashion . Second , the neutral antagonist does not rescue LTPGABA . Signaling through the JNK pathway is thought to be responsible for the long-lasting non-competitive inhibition of the κOR ( Bruchas et al . , 2007; Melief et al . , 2010 , 2011 ) . In our experiments , the requirement of JNK for norBNI to rescue LTPGABA is evidence that norBNI acts through non-competitive means , and suggests that the persistent activation of the κOR following stress does not require continuous receptor binding by ligand . Importantly , inhibition of JNK signaling alone did not prevent LTPGABA induction in slices from naïve animals , nor did it restore LTPGABA in slices from stressed animals , indicating that the role of JNK is limited to inhibition of the receptor by norBNI . The failure of the JNK inhibitor to rescue LTPGABA indicates that inhibition of LTPGABA by κORs is not mediated by the JNK pathway , but instead most likely through one of the other pathways downstream of κORs , such as the p38 or ERK MAPK pathways , or through activation of Gαi ( Bruchas and Chavkin , 2010; Iñiguez et al . , 2010; Ehrich et al . , 2015 ) . The inability of the neutral antagonist , 6β–naltrexol , to rescue LTPGABA is consistent with stress promoting constitutive κOR activity at inhibitory VTA synapses . NorBNI , through activation of JNK , reduces the signaling capacity of the κOR regardless of whether this occurs through constitutive activity or increased dynorphin binding . In contrast , a neutral antagonist like 6β-naltrexol could only reverse the loss of LTPGABA by preventing agonist binding to the receptor . In contrast to the rapid restoration of LTPGABA by bath application of norBNI , bath application of 6β–naltrexol did not rescue LTPGABA . This discrepancy cannot be explained by insufficient concentration or time of application of 6β–naltrexol , as a similar bath perfusion of 6β–naltrexol was sufficient to block the depression of EPSCs onto VTA dopamine neurons induced by bath application of the κOR agonist , U50488 . The simplest explanation of our data is that acute stress induces constitutive activity of the κOR . Alternatively , norBNI may promote JNK signaling via an unknown mechanism independent of κOR receptors . Acute stress appears to trigger a shift towards constitutively active κORs through a transient release of the endogenous κOR ligand , dynorphin ( Figure 7 ) . Our strongest evidence for this model is the ability of the neutral antagonist 6β–naltrexol to prevent the loss of LTPGABA when administered before , but not after stress . Although 6β–naltrexol has equivalent affinity for µ and κ ORs , ( Wang et al . , 2007 ) our previous work has shown that the block of LTPGABA by stress is unaffected by pre-stress administration of the µOR antagonist cyprodime ( Graziane et al . , 2013 ) . Therefore , the ability of 6β–naltrexol to prevent the stress-induced loss of LTPGABA is unlikely to involve µOR signaling and instead occurs by blocking dynorphin binding to the κOR . A single in vivo systemic administration of the κOR agonist U50488 also blocks LTPGABA for at least five days , supporting the idea that brief agonist exposure alone is sufficient to trigger lasting constitutive κOR activity . How might activation of κORs by its endogenous ligand shift the receptor towards constitutive activity ? In a heterologous cell-culture system , previous exposure to a κOR agonist alone significantly increased constitutive activity of the receptor ( Wang et al . , 2007 ) . More is known regarding regulation of constitutive activity of µORs . In either cultured cells heterologously expressing µORs ( Wang et al . , 1994 , 2000; Liu and Prather , 2001 ) or in intact animals ( Wang et al . , 2004; Shoblock and Maidment , 2006; Meye et al . , 2012 ) , exposure to the µOR agonist morphine triggers an increase in constitutive activity of µORs . Morphine-induced constitutive activity of µORs is regulated by calmodulin and protein kinases . Under basal conditions , calmodulin binding to the µORs prevents constitutive association with G-proteins . Following morphine exposure , calmodulin dissociates from the µOR , allowing constitutive activation ( Wang et al . , 1999 , 2000 ) . Although it is unknown whether calmodulin regulates the activity of κORs , an intricate scaffolding complex regulates κOR signaling ( Bruchas and Chavkin , 2010 ) , and future studies investigating the role of these signaling complexes in κOR activity in response to stress will be important and intriguing . Most of the work investigating constitutive activation of GPCRs has focused on enhancement of constitutive activity by administration of exogenous ligands ( Meye et al . , 2014 ) , while considerably less is known about induction of constitutively active states of GPCRs by endogenous signaling . However , it was recently reported that inflammatory pain increases constitutively active µORs in the mouse spinal cord , leading to hyperalgesia and cellular dependence ( Corder et al . , 2013 ) . Our data indicate that treatment with the κOR agonist U50488 alone is sufficient to produce constitutively active κORs on inhibitory VTA synapses . A remaining puzzle is why κOR activation by endogenous ligand can produce constitutively active receptors at some synapses but not at their neighbors , and in response to certain environmental cues ( acute stress ) but not to others during which dynorphin may also be released . One possibility is that receptors in different cell types may couple to different signaling cascades or scaffolding molecules . Another possibility is that coordinated signaling between κORs and another neurotransmitter system may be required . Our prior work indicates that activation of glucocorticoid receptors is required for the block of LTPGABA by stress ( Niehaus et al . , 2010; Polter et al . , 2014 ) . Although persistent activation of these receptors is not seen after stress , it is possible that coincident activation of glucocorticoid and kappa opioid receptors leads to constitutive activation of the latter . Additionally , it has been reported that the orexin-1 receptor attenuates κOR inhibition of cAMP production , but enhances recruitment of β-arrestin and p38 MAPK activation , and both effects are prevented by the JNK inhibitor SP600125 used in our study ( Robinson and McDonald , 2015 ) . Both orexin and dynorphin are co-released from hypothalamic projections to the VTA ( Chou et al . , 2001; Muschamp et al . , 2014; Baimel et al . , 2015 ) . This arrangement raises the possibility that release of both peptides together , or perhaps simultaneous release of dynorphin and an unknown neurotransmitter acting similarly to orexin , may initiate signaling events not triggered by dynorphin alone . The putative dual receptor signaling might be one way to induce synapse- or experience- selective constitutive κOR activity . One unanswered question is how κORs suppress the expression of LTPGABA . Our previous studies have shown that LTPGABA is triggered by nitric oxide-mediated activation of cGMP-protein kinase G ( PKG ) signaling ( Nugent et al . , 2007 , 2009; Niehaus et al . , 2010 ) . Because an exogenous source of nitric oxide ( SNAP ) does not rescue potentiation following stress , the blockade is likely to occur in the presynaptic terminal between activation of guanylate cyclase and enhancement of GABAergic release . While it is possible that κOR activation generally depresses GABA release , our previous work ( Graziane et al . , 2013 ) found no change in mIPSC frequency following cold water swim stress . These data suggest that κORs do not alter basal GABA release . Moreover , LTPGABA is also lost 24 hr after a single morphine exposure , and in this situation a cGMP analog or strong activation of sGC potentiates GABA release ( Nugent et al . , 2007; Niehaus et al . , 2010 ) . We therefore favor a mechanism by which after acute stress , constitutively-active κORs similarly act on a substrate that limits induction of plasticity without affecting basal release mechanisms , perhaps through downregulation of soluble guanylyl cyclase , or scaffolding changes that sequester PKG from its substrates . Although it remains unknown under what conditions LTPGABA is activated in an intact animal , our prior studies shed some light on its potential roles . LTPGABA is a heterosynaptic form of plasticity that can be triggered by a high-frequency tetanus that activates NMDAR-dependent activation of calcium-sensitive nitric oxide synthase ( Nugent et al . , 2007 ) . We therefore expect that LTPGABA would be induced when there is robust activation of excitatory inputs onto dopamine neurons . LTPGABA may play a homeostatic role to enhance inhibition of dopamine neurons after strong NMDAR-activating excitation . Loss of LTPGABA , therefore , would result in an imbalance between inhibitory and excitatory input onto the dopamine neuron . As GABAergic synapses on dopamine neurons strongly control their spontaneous firing ( van Zessen et al . , 2012 ) , the loss of LTPGABA is likely to prolong or enhance firing in response to salient stimuli . The critical role of the VTA in reinstatement of drug seeking has been repeatedly underscored ( McFarland et al . , 2004; Briand et al . , 2010; Graziane et al . , 2013; Mantsch et al . , 2016 ) , and within the VTA , GABAergic synapses on VTA dopamine neurons powerfully regulate DA cell firing ( Tan et al . , 2012; van Zessen et al . , 2012; Polter and Kauer , 2014 ) . Our work shows that stress produces long-lasting κOR constitutive activity that is restricted to inhibitory synapses on dopamine cells , thereby affecting information stored or processed here for far longer than at excitatory synapses . We might therefore predict two phases of stress-induced κOR activation . We hypothesize that dynorphin is released during and/or immediately after stress , depressing EPSCs onto dopaminergic neurons and hyperpolarizing dopaminergic neurons , on balance decreasing dopaminergic neuron excitability ( Margolis et al . , 2003 , 2005; Ford et al . , 2006 ) . However , as dynorphin is degraded , we would expect that the strength of excitatory synapses would return to normal levels while LTPGABA would become blocked by constitutive activity of κORs , a state lasting at least five days after swim stress . This would instead increase the firing rate of dopaminergic neurons , particularly in response to excitatory stimuli . This increased excitability could contribute to the increased drive towards drug-seeking behavior upon exposure to spatial cues associated with past drug experience ( i . e . , return to the operant chamber ) , and could create a state of vulnerability to further stressors . Indeed , in rats subjected to the same cold water stress used in this study , the firing rate of dopaminergic neurons remains elevated for several days afterwards ( Marinelli , 2007 ) . Interestingly , a single dose of the κOR agonist , salvinorin A , has biphasic effects on reward function: immediately after administration , rats exhibit an anhedonic increase in reward thresholds to intracranial self-stimulation . However , 24 hr after salvinorin A administration , rats exhibit decreased reward thresholds , indicating an increase in reward sensitivity ( Potter et al . , 2011 ) . This biphasic effect is consistent with a split between short- and long-term effects of κOR activation , perhaps due to differential mechanisms of regulation and constitutive activation of subsets of receptors . The circuitry of the VTA is highly complex , and dopamine neurons within the VTA exhibit physiological and functional heterogeneity that correlates with projection target . While disagreement remains about the most appropriate pharmacological , physiological , and anatomical markers of different subclasses of dopamine neurons ( Ford et al . , 2006; Margolis et al . , 2006; Lammel et al . , 2008 , 2011; Ungless and Grace , 2012; Baimel et al . , 2017 ) , the electrophysiological markers used here and the lateral location of our recordings within the VTA suggest to us that our population of cells may largely comprise dopamine neurons that project to the nucleus accumbens . This may be significant for drug reward , as activation of these neurons has been shown to be rewarding in mice ( Lammel et al . , 2012 ) . Therefore , our study indicates that an acute stressor induces a long-lasting loss of inhibitory plasticity in circuitry that may drive rewarding behavior . GABAergic afferents on VTA dopamine neurons can release GABA onto either GABAA or GABAB receptors . Previous studies including more recent optogenetic approaches have suggested that GABAB receptor-targeting neurons arise from the nucleus accumbens and regulate drug-induced behaviors ( Sugita et al . , 1992; Cameron and Williams , 1993; McCall et al . , 2017; Edwards et al . , 2017 ) . However , our earlier work found no LTPGABA at GABAB synapses on dopamine neurons ( Nugent et al . , 2009 ) , suggesting that the effects of persistently activated κORs are unlikely to involve the nucleus accumbens-VTA GABAergic afferents . Our data provide the first demonstration that constitutively active κORs in the VTA are required for stress-induced reinstatement of cocaine-seeking . Post-stress ( at least 24 hr ) administration of norBNI prevents reinstatement , while post-stress administration of the neutral antagonist 6β-naltrexol does not . The ability of norBNI to modify drug-seeking behavior even when given significantly after the stressor is remarkable , and indicates the therapeutic potential of targeting κORs to reverse stress-induced neuroadaptations . The failure of 6β–naltrexol to prevent reinstatement at time points when norBNI is effective strongly suggests that the persistent increase in drug seeking induced by swim stress is mediated by constitutive activity of κORs rather than a prolonged increase in dynorphin release . Furthermore , this result is consistent with an important role for GABAergic synapse plasticity in stress-induced drug-seeking behavior . While considerable attention has been given to the role of LTP at excitatory synapses in the VTA , the κOR block by norBNI does not prevent stress from potentiating excitatory synapses on dopamine neurons ( Graziane et al . , 2013 ) . Our current work confirms that the loss of LTP at GABAergic synapses in the VTA is highly correlated with stress-induced drug-seeking . κORs have shown promise as a potential drug target for substance use and mood disorders ( Bruchas et al . , 2010; Van't Veer and Carlezon , 2013; Crowley and Kash , 2015 ) and our work suggests a novel way in which κOR signaling may go awry . In preclinical models , κOR antagonists have shown potential efficacy for depression and for compulsive and stress-induced drug use ( Bruchas et al . , 2010; Wee and Koob , 2010 ) . Numerous clinical trials are in progress using κOR ligands to target substance use disorders and depression ( Ehrich et al . , 2015; Karp et al . , 2014; Chavkin and Koob , 2016; Ling et al . , 2016; Nasser et al . , 2016 ) . However , many of these trials use buprenorphine , a partial agonist at κORs , or novel compounds which may lack inverse agonist activity , neither of which would reduce activity of constitutive κORs ( Karp et al . , 2014; Rorick-Kehn et al . , 2014 ) . Our study suggests that future drug development should consider excess κOR activity through receptor signaling as well as at the level of ligand binding . Similarly , disappointing results or minimal effects in clinical trials may not represent failure of κORs as a pharmaceutical target , but a need to consider drugs that target specific conformations of the κOR that promote constitutive signaling . An alternative strategy would be to target JNK signaling in the VTA , as norBNI appears to rescue κOR function by activating JNK . An intriguing implication of our studies comes from our data that constitutive activity of κORs at inhibitory synapses in the VTA lasts only five to ten days following acute stress ( Polter et al . , 2014 ) , a considerably shorter time period than the 14–21 days typical for turnover of κORs ( McLaughlin et al . , 2004 ) . This suggests that constitutive activity of the κORs may be terminated by an unidentified active mechanism . Future studies investigating such a mechanism could identify targets that could be recruited to promote resilience to stress . Our work demonstrates a novel mechanism of experience-dependent regulation of κORs , and highlights the ability of modulation of κORs to reverse stress-induced neuroadaptations and behavioral deficiencies well after the stressor has occurred . Further study of the mechanisms of constitutive activation of κORs may yield numerous potential targets for the treatment of substance use disorders and other stress-linked illnesses . All procedures were carried out in accordance with the guidelines of the National Institutes of Health for animal care and use , and were approved by the Brown University Institutional Animal Care and Use Committee or by the University of Wyoming Institutional Animal Care and Use Committee . For slice electrophysiology studies , male and female Sprague-Dawley rats ( P16-25 ) were maintained on a 12 hr light ⁄ dark cycle and provided food and water ad libitum . For self-administration studies , male Sprague-Dawley rats ( 350–450g ) were bred in-house and individually housed in a temperature-controlled room with a 12 hr reverse light/dark cycle . All animals were given ad libitum access to water throughout experimentation , except during times in which they were in the operant chambers ( described below ) . Rats were 60–70 days old at the start of behavioral experiments . Stress was administered by a modified Porsolt forced swim task ( Niehaus et al . , 2010 ) . Rats were placed for 5 min in cold water ( 4–6°C ) , then dried and allowed to recover in a warmed cage for two hours before returning to the home cage . U50488 ( 5 mg/kg ) and 6β-naltrexol ( 10 mg/kg ) were dissolved in saline or 10% DMSO in saline , respectively . Vehicle-injected animals were given an injection of the equivalent volume . For some experiments , animals given vehicle injections at varying time points were collapsed into a single group . Brain slices were prepared at several time points after stress exposure , as described below . Horizontal midbrain slices ( 250 µm ) were prepared as previously described from deeply anesthetized Sprague-Dawley rats ( Nugent et al . , 2007; Niehaus et al . , 2010; Polter et al . , 2014 ) . Slices were stored for at least 1 hr at 34°C in oxygenated HEPES holding solution ( in mM ) : 86 NaCl , 2 . 5 KCl , 1 . 2 NaH2PO4 , 35 NaHCO3 , 20 HEPES , 25 glucose , 5 sodium ascorbate , 2 thiourea , 3 sodium pyruvate , 1 MgSO4 . 7H2O , 2 CaCl2 . 2H2O ( Ting et al . , 2014 ) . Slices were then transferred to a recording chamber where they were submerged in ACSF containing ( in mM ) : 126 NaCl , 21 . 4 NaHCO3 , 2 . 5 KCl , 1 . 2 NaH2PO4 , 2 . 4 CaCl2 , 1 . 0 MgSO4 , 11 . 1 glucose . General methods were as previously reported ( Niehaus et al . , 2010; Polter et al . , 2014 ) . Midbrain slices were continuously perfused at 1 . 5–2 mL ⁄ min . Patch pipettes were filled with ( in mM ) : 125 KCl , 2 . 8 NaCl , 2 MgCl2 , 2 ATP-Na+ , 0 . 3 GTP-Na+ , 0 . 6 EGTA , and 10 HEPES . To record IPSCs , the extracellular solution was ACSF ( 28–32°C ) containing: 6 , 7-dinitroquinoxaline- 2 , 3-dione ( DNQX; 10 µM ) and strychnine ( 1 µM ) , to block AMPA and glycine receptors respectively . To record EPSCs , 100 µM picrotoxin was added to the ACSF . Dopaminergic neurons , which comprise about 70% of all VTA neurons , were identified by the presence of a large Ih ( >50 pA ) during a voltage step from −50 mV to −100 mV . GABAA receptor-mediated IPSCs were stimulated using a bipolar stainless steel stimulating electrode placed 100–300 µm rostral to the recording site in the VTA . Cells were voltage-clamped at −70 mV and input resistance and series resistance were monitored throughout the experiment; cells were discarded if these values changed by more than 15% during the experiment . 3-isobutyl-1-methylxanthine ( IBMX; 100 µM ) was used to inhibit phosphodiesterase-mediated degradation of cGMP and applied via perfused ACSF for at least 10 min prior to induction of LTPGABA by application of the NO donor , SNAP ( S-nitroso-N-acetylpenicillamine , 400 µM ) . Control animals ( vehicle-injected stressed or unstressed animals ) were interleaved with experimental animals ( drug-injected stressed animals ) . Where indicated , NorBNI ( 100 nM ) , 6β-naltrexol ( 10 µM ) , and SP600135 ( 20 µM ) were bath applied to slices at least 10 min prior to induction of LTPGABA . Rats were anesthetized with ketamine HCl ( 87 mg/kg , i . m . ) and xylazine ( 13 mg/kg , i . m . ) and implanted with intravenous jugular catheters . In order to protect against infection and maintain catheter patency , catheters were flushed daily with 0 . 2 mL of a mixed cefazolin ( 0 . 1 gm/ml ) and heparin ( 100 IU ) solution . Rats were allowed to recover for one week before behavioral testing . All self-administration procedures were conducted in standard operant chambers ( Med Associates , St . Albans , VT; 30 . 5 cm x 24 . 1 cm x 21 . 0 cm ) . Each box contained a house light ( illuminated throughout behavioral testing ) two retractable levers , a cue light , and tone generator . Prior to beginning cocaine self-administration training , animals were food deprived for 24 hr and subsequently placed into the operant chambers overnight for 14 hr . During this session , a response to the active lever ( the left lever ) resulted in the delivery of a single 45 mg food pellet ( #F0165 , Bio-Serv , Flemington , NJ ) and the presentation of a compound cue ( illumination of light above the active lever +5 s tone , 2900 Hz ) , followed by a 25 s timeout period . Responses to the inactive lever ( the right lever ) had no programmed consequences but were recorded . Total rewards received were also recorded . On the next day , cocaine self-administration training began . During this time , a response to the active lever yielded a 0 . 05 ml infusion of 0 . 20 mg of cocaine ( dissolved in 0 . 9% saline ) as well as the presentation of the compound cue ( light + tone ) . Self-administration continued ( 2 hr/daily ) until animals reliably pressed the active lever ( 3d with minimum of 10 cocaine infusions received ) . Following the acquisition of cocaine self-administration , all animals underwent extinction training , during which responses to the previously active lever yielded the compound cue but no longer produced drug infusion . Animals were food-restricted to 80% of their body weight during self-administration training . During extinction , animals were given ad libitum access to food in the home cage . Extinction procedures continued until animals reached extinction criteria ( 3d with less than 10 active lever presses ) . The day following the last extinction session , rats were subjected to a 5 min forced swim stress in cold water ( 4-6˚C , Saal et al . , 2003; Niehaus et al . , 2010 ) . Rats were then split into three groups , receiving ( i . p . ) injections of either saline ( 1 ml/kg ) , norBNI ( 10 mg/kg ) , or 6β-Naltrexol ( 10 mg/kg ) . 24 hr after swim stress , rats in the saline and norBNI groups were injected and left undisturbed in their home cage for one day . Rats in the 6β-naltrexol group were injected one hour prior to reinstatement testing . At 48 hr after swim stress , all animals were subjected to reinstatement testing , which similarly to extinction yielded the compound cue but no drug infusion . Magnitude of LTP was determined as mean IPSC amplitude for 5 min just before application of SNAP compared with mean IPSC amplitude from 10–15 min after SNAP application , unless otherwise noted . Data are presented as means ± SEM of the percent IPSC amplitude normalized to IPSCs in the 10 min prior to SNAP application . Statistical methods were not used to determine sample size . Sample size was based on our prior experience and previously published studies ( Graziane et al . , 2013; Polter et al . , 2014 ) . All reported n’s are the number of animals ( biological replicates ) , unless otherwise noted . Significance was determined using a two-tailed Student’s t-test or a one-way ANOVA with a significance level of p<0 . 05 . All post-hoc comparisons were done using Dunnett’s test unless otherwise noted . Self-administration data were analyzed using paired t-tests . IBMX was obtained from Enzo Life Sciences . NorBNI , U50488 , and SNAP were obtained from Tocris Biosciences . DNQX , picrotoxin , strychnine , and 6β-naltrexol were obtained from Sigma-Aldrich . SP600125 was obtained from Calbiochem .	People who are recovering from drug addiction are more vulnerable to cravings and relapse when under stress . This ability of stress to boost drug relapse can also be shown in animals previously exposed to addictive drugs . Rats can learn to press a lever to administer themselves a dose of cocaine and , during withdrawal , rats previously exposed to the drug will press the lever more often if they are stressed . Indeed , just a few minutes of stress is enough to increase lever pressing for several days . Stress and addictive drugs both act on a region of the brain called the ventral tegmental area , or VTA , which is part of the brain’s reward system . Stress indirectly increases the activity of the VTA . It does so by activating a protein on the surface of VTA neurons called the kappa opioid receptor ( κOR for short ) . Previous studies revealed that five minutes of stress increases the activity of κORs in the VTA of rats for five days . Conversely , blocking κORs stopped stressed rats from pressing the lever more often for cocaine . Together , these findings suggested that activating κORs in the VTA contributes to stress-induced drug relapse . Polter et al . have now discovered how stress activates κORs . It turns out that stressful or unpleasant experiences cause the brain to produce a protein called dynorphin , which binds to and activates the κORs . After a stressful event , the receptors are said to have become constitutively active , and blocking this constitutive activity prevents stress from inducing drug-seeking behavior . Polter et al . show that binding of dynorphin is needed to change the shape of the receptors so that they remain active even after dynorphin has detached , but it is likely that other molecules are also involved . This is the first study to show a link between stress , constitutive activation of κORs , and drug relapse . The next step is to work out why this process occurs on only some and not all occasions when the brain releases dynorphin , and why only certain κORs in the VTA respond in this way . Whether constitutive kOR activity drives stress-related craving in people with a history of drug abuse and how to halt these cravings also remain to be determined .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Constitutive activation of kappa opioid receptors at ventral tegmental area inhibitory synapses following acute stress
The increase in activity of the two-pore potassium-leak channel Kcnk5b maintains allometric juvenile growth of adult zebrafish appendages . However , it remains unknown how this channel maintains allometric growth and how its bioelectric activity is regulated to scale these anatomical structures . We show the activation of Kcnk5b is sufficient to activate several genes that are part of important development programs . We provide in vivo transplantation evidence that the activation of gene transcription is cell autonomous . We also show that Kcnk5b will induce the expression of different subsets of the tested developmental genes in different cultured mammalian cell lines , which may explain how one electrophysiological stimulus can coordinately regulate the allometric growth of diverse populations of cells in the fin that use different developmental signals . We also provide evidence that the post-translational modification of serine 345 in Kcnk5b by calcineurin regulates channel activity to scale the fin . Thus , we show how an endogenous bioelectric mechanism can be regulated to promote coordinated developmental signaling to generate and scale a vertebrate appendage . Tissue scaling involves the coordinated control of developmental programs , since anatomical structures consist of different tissues that form in a coordinated manner and grow proportionally with each other and with the body . While there are several developmental signals known to regulate cell proliferation and tissue formation , mechanisms that concomitantly activate several developmental signals to synchronize the growth of multi-tissue appendages and organs in a manner that is coordinated with body proportions remain poorly defined . There is growing evidence that several biological phenomena involved in tissue generation and growth are influenced by electrophysiological changes in ‘non-excitable’ cells ( Sundelacruz et al . , 2009 ) . Several cell behaviors are affected by the addition of electric currents ( McCaig et al . , 2005 ) : cell migration , cell proliferation , cell differentiation , gene transcription and consequently tissue formation are all altered by the application of an exogenous current ( Baer and Colello , 2016; Bartel et al . , 1989; Blackiston et al . , 2009; Borgens et al . , 1977; Geremia et al . , 2007; Sundelacruz et al . , 2009; Yasuda , 1974; Zhao et al . , 2002 ) . The culmination of these findings have led to the hypothesis that bioelectrical fields exist that have higher order organizational non-cell-autonomous properties in the development of anatomical structures ( for review , see Levin , 2014; Messerli and Graham , 2011] . As a regulator of membrane potential , K+ conductance is an important component of the electrophysiological properties of cells . Evidence that illustrates the importance of K+ conductance in tissue formation comes from studies in which disruption of inward rectifying K+ channels of the Kir2 family can cause cranial facial defects , abnormal number of digits and reduced digit size ( Andersen et al . , 1971; Canún et al . , 1999; Sansone et al . , 1997; Tawil et al . , 1994; Yoon et al . , 2006a; Yoon et al . , 2006b; Zaritsky et al . , 2000 ) . A striking finding concerning the coordinated control of cell behavior is the formation of eye structures by overexpressing different ion channels that alter membrane potential in early Xenopus embryos ( Pai et al . , 2012 ) : overexpression and activation of a glycine-gated chloride channel in cells that form the eye interferes with eye formation , while overexpression of a dominant-negative K+-ATP channel simulates ectopic eye formation even in unexpected locations on the body ( Pai et al . , 2012 ) . These findings illustrate that changes in the membrane potential of cells can have significant impacts on the development of anatomical structures . However , how electrophysiological information associates with the multiple necessary signals that control formation and/or growth of multi-tissue structures remains unclear . The development of body structures not only involves forming tissues , it also involves coordinating the growth of each contributing tissue cell . To form organs that correctly scale with the body , each tissue grows either isometrically ( grows with the same rate as the body ) or allometrically ( disproportionally grows in relation to the growth of the body ) . The zebrafish mutants another long fin ( alf ) , long fin ( lof ) , and schleier ( schl ) display continued allometric growth of each appendage from the juvenile stage into the adult stage ( Lanni et al . , 2019; Perathoner et al . , 2014; Stewart et al . , 2020 ) . The dominant allometric growth phenotype of alf is due to mutations in the transmembrane pore region of kcnk5b ( Perathoner et al . , 2014 ) , encoding a two-pore K+-leak channel that regulates membrane potential by outward flow of K+ from the cell ( Goldstein et al . , 2001 ) . The dominant phenotype of lof is linked to elevated expression of a voltage-gated potassium channel Kcnh2a ( Stewart et al . , 2020 ) . The phenotype of schl is due to dominant-negative mutations in kcc4 ( Lanni et al . , 2019 ) , encoding a K+-Cl- cotransporter that regulates intracellular K+ levels in a chloride-dependent manner ( Marcoux et al . , 2017 ) . alf , lof , and schl demonstrate the importance of K+ conductance and ultimately of electrophysiological signals for the correct body-to-appendage proportions . Despite the connection between K+ conductance and the proportional growth of the fins , it remains unclear how K+-mediated signal translates into coordinated growth of the fish appendage and how any K+ channel is regulated to scale tissue . We show that activity of the single two-pore K+-leak channel Kcnk5b is sufficient to induce the activation of at least two components , Shh and Lef1 , of two important morphogen pathways , not only in the adult fin but also in the larva . Our data also indicates that this induction is cell autonomous , arguing that increases in membrane potential caused by Kcnk5b regulate growth through intracellular regulation of these developmental pathways . Overexpression of kcnk5b or one of two other two-pore kcnk channels in different mammalian cell lines showed variable activation of genes belonging to different developmental signal transduction mechanisms , supporting the conclusions that the different developmental mechanisms needed to scale fish appendages can be regulated by the same electrophysiological change induced by potassium leak and that the combinatorial activation of the different developmental mechanisms by Kcnk5b is cell-type dependent . Lastly , we provide evidence for how post-translational modification of Kcnk5b at Serine345 by calcineurin regulates its electrophysiological activity and consequently the scaling of zebrafish fins . Thus , we describe an endogenous cell-autonomous mechanism through which electrophysiological signals can induce and coordinate specific morphogen and growth factor signals to mediate the scaling of an anatomical structure . Mutations in the two-pore K+-leak channel Kcnk5b that increase its activity lead to enhanced growth of the zebrafish appendages ( Perathoner et al . , 2014 ) . While this finding implicates the importance of bioelectric signaling in appendage scaling , it remains unknown how the activity of a single K+ channel is integrated with the developmental controls that generate new appendage tissues . Growth of any appendage involves the coordinated activation of specific morphogen and growth factor pathways: Shh , β-catenin-dependent Wnt , Bmp , Fgf , and Retinoic acid . Therefore , to begin to determine how this channel is involved in the coordinated growth of the entire fin , we generated transgenic zebrafish that expresses kcnk5b under the control of a conditionally inducible promoter ( heat-shock promoter ) to temporally activate this channel in adult fins . After a single 10-min heat-shock pulse of the Tg[hsp70:kcnk5b-GFP] transgene , we observed significant activation of shh and lef1 ( β-catenin-dependent Wnt ) ( Figure 1Aa; Figure 1—figure supplement 1A ) , as well as an increase in aldh1a2 ( retinoic acid ) ( Figure 1Ab ) within 6 hr by qRT-PCR , while pea3 ( Fgf ) was slightly increased and msxb ( BMP ) was not significantly changed ( Figure 1A ) . All genes returned to control levels by 12 hr after the single pulse ( Figure 1B ) . The induction of the developmental genes coincided with the temporal expression of the kcnk5b-GFP transgene ( Figure 1C ) , whose expression emerged as a lattice pattern ( Figure 1—figure supplement 1B–D ) , indicating that Kcnk5b-GFP was localized and functioned at cell membranes . When we maintained chronic expression of the transgene by heat shocking the caudal fin for 10 min once per day for 3 days , we observed expression of lef1 , shh , aldh1a2 as well as pea3 and msxb over controls ( Figure 1D ) , which included Shh’s patched receptors and slight up-regulation of bmp2b ( Figure 1—figure supplement 1E ) . Together , these data show that Kcnk5b is sufficient to induce the transcription of certain developmental genes as though it were a part of their signaling mechanisms . Furthermore , the initial activation of shh , lef1 , and aldh2a ( Figure 1A ) followed by later upregulation of pea3 and msxb after continued kcnk5b overexpression ( Figure 1C ) suggest that there is a hierarchical activation of developmental mechanisms that will ultimately lead to the complete allometric growth program . To examine the spatial expression of the two genes most responsive to Kcnk5b , we performed in situ hybridization experiments for shh and lef1 . Compared to control fins ( Figure 3Ea , c ) , localization of shh and lef1 was in the distal tip of the fin ( Figure 1Eb , d ) where growth normally occurs , and cross-sections through the fins showed that Kcnk5b-mediated induction of these genes occured in the epidermal/epithelial tissues ( Figure 1Fb , d ) . We also assessed increases in Shh and Lef1 protein levels after the heat-shock induction of kcnk5b-GFP ( Figure 1G , H; Figure 1—figure supplement 2A , B ) . Because Lef1 conveys β-catenin-dependent Wnt signaling by acting as a transcriptional platform for β-catenin , we examined protein expression of β-catenin and observed no significant differences in its overall levels ( Figure 1G , H; Figure 1—figure supplement 2C ) . However , when we examined β-catenin protein distribution in the fin tissues by immunohistochemistry staining , we observed an increased number of nuclei with β-catenin co-staining in the outer epidermis layers and basal epithelium of the fin ( Figure 1I , J; Figure 1—figure supplement 2E–J ) despite no significant differences in the measured β-catenin-associated fluorescence intensities ( Figure 1—figure supplement 2K ) . To test whether the redistribution of β-catenin leads to the activation of known β-catenin-dependent genes , we performed qRT-PCR of axin2 , cyclin D , and c-myc . We observed that none of these candidates were activated ( Figure 1K ) , which indicates that while Lef1 expression increases , β-catenin-dependent Wnt signaling is not directly activated by Kcnk5b . Together , all the expression results indicate that Kcnk5b activity promotes the transcription of specific components of a limited number of developmental pathways in the adult fin . To test whether increasing the activity of Kcnk5b has the same transcriptional effect on these developmental pathways in another in vivo context , we induced the expression of kcnk5b in the zebrafish larva ( Figure 2—figure supplement 1A ) . We observed that lef1 , shh , adlh2a , pea3 , and msxb expression levels were increased by induction of kcnk5b compared to heat-shocked non-transgenic control fish ( Figure 2A ) . To further explore the effect of Kcnk5b activity on shh transcription in developing fins , we performed in situ hybridization experiments on fish embryos and examined shh expression in the developing pectoral fin buds . Compared to the emerging expression of shh in the fin buds of heat-shocked non-transgenic embryos ( Figure 2Ba , c ) , we observed kcnk5b expression induced increases in the rate of shh mRNA detection ( Figure 2Bb , d ) and in the area of shh expression ( Figure 2Be-h ) . Thus , Kcnk5b activity appears to increase both the intensity ( Figure 2Bi ) and the range of shh expression ( Figure 2Bj ) . We further assessed Lef-1-dependent transcription by crossing the heat-shock-inducible transgenic Tg[hsp70:knck5b-GFP] line with the transgenic Lef1 reporter line Tg[7XTCF-Xla . sam:mCherry] . While the double-transgenic fish Tg[hsp70:kcnk5b-GFP; 7XTCF-Xla . sam:mCherry] displayed limited expression of mCherry before heat-shock induction of the kcnk5b-GFP transgene ( Figure 2Ca , b , g ) , after heat shock , double-transgenic Tg[hsp70:kcnk5b-GFP; 7XTCF-Xla . sam:mCherry] fish showed a broad increase in reporter mCherry expression over single-transgenic Tg[7XTCF-Xla . sam:mCherry] ( Figure 2Cd , e , g ) . From histological cross-sections of double-transgenic larva , we observed that mCherry was upregulated broadly in the body of the animal ( Figure 2D ) . Furthermore , we observed an increase in the length of the caudal finfold along the anterioposterior axis of 5 dpf larva after induction of kcnk5b-GFP by single daily 10 min pulses over 3 days ( Figure 2F; Figure 2—figure supplement 1C ) , while the caudal finfold lengths along the dorsoventral axis of appeared not change ( Figure 2G ) . However , we observed that when we compared both anterioposterior and dorsoventral axes measurements in relation to body length , we observed that in both instances the finfold dimensions proportionally increased ( Figure 2H , I ) , which was in part associated with decreases in the lengths of the bodies ( Figure 2J ) . These results showed that in relation to the body , Kcnk5b activity promoted proportional increases in the finfold dimensions . Our observation that the activity of this channel decreased the length of the body , suggests that the growth-promoting effects of Kcnk5b activity may be limited to the finfolds and appendages . Together , these results indicate that Kcnk5b activity is sufficient to promote the transcription of specific developmental genes in several different tissue types to promote allometric growth of the finfold . Previous work implicates bioelectric intercellular communication as a mechanism for how bioelectricity can influence tissue growth ( McLaughlin and Levin , 2018 ) , and changes in K+ channel activity have been shown to regulate different cell behaviors in a non-cell-autonomous manner ( Morokuma et al . , 2008; Pai et al . , 2015 ) . The broad activation of the Lef1-dependent Wnt reporter in several tissues ( Figure 2 ) and Kcnk5b’s ability to scale all the tissues of the fin appendages suggest that Kcnk5b acts via non-cell autonomous communication among cells . To determine whether the observed Kcnk5b-mediated induction of gene expression is due to intercellular communication ( e . g . through extracellular ligands such as Wnt ) or due to cell autonomous activation of transcription , we transplanted cells from Tg[hsp70:kcnk5b-GFP; 7XTCF-Xla . sam:mCherry] transgenic embryos into embryos harboring only the Tg[7XTCF-Xla . sam:mCherry] transgene and then raised mosaic embryos and larva ( Figure 3A ) . Analyses of the mosaic larva showed the previously reported developmental expression of the Lef1-dependent reporter before heat shock ( Figure 3Ba , e , i , m ) ( Moro et al . , 2012 ) . However , after heat-shock induction of the Tg[hsp70:kcnk5b-GFP] transgene ( Figure 3Bb , f , j , n ) , we observed ectopic activation of 7XTCF-Xla . sam:mCherry reporter only in donor cells of chimeric 48 hpf and 72 hpf fish ( recipient 7XTCF-Xla . sam:mCherry fish harboring transplanted cells from Tg[hsp70:kcnk5b-GFP;7XTCF-Xla . sam:mCherry] embryos ) . GFP-mCherry-positive cells appeared in tissues in the head ( Figure 3Bb , c , d ) , in skeleton surrounding the eye ( Figure 3Bf , g , h ) , in trunk muscles ( Figure 3Bj , k , l ) and in skin ( Figure 3Bn , o , p ) . From closer inspection , we observed co-expression in neurons in the head ( Figure 3C , a-c ) , in the ectodermal bones of the skull ( Figure 3C , d-f ) , mandible bone and cartilage ( Figure 3C , g-i ) , mesenchyme surrounding the otic vesicle ( Figure 3C , j-l ) , epithelial cells in the finfold ( Figure 3C , o-r ) and individual striated muscle cells of the trunk ( Figure 3C , r-t ) . We counted the number of GFP and mCherry positive cells in the different tissues and observed that all Kcnk5b-GFP-positive cells were mCherry positive ( Figure 3C ) . Moreover , in all tissues , the ectopic mCherry expression was always limited to the Kcnk5b-positive cells ( Figure 3B–E; Figure 3—figure supplement 1G–J ) . Together , these data support two conclusions: one , the activation of the Lef1-dependent reporter by Kcnk5b is cell autonomous; and two , Kcnk5b is able to promote the expression of the Lef1 reporter in diverse tissue types . As a K+-leak channel , Kcnk5b’s activity should decrease intracellular K+ levels . We performed Fluorescence Lifetime Microscopy ( FLIM ) analysis with an established genetic sensor for K+ to measure intracellular K+ levels ( Shen et al . , 2019 ) . This sensor uses the FRET potential between two fluorophores that are joined by a K+-binding linker . Changes in FRET due to K+ binding results in changes in the fluorescence lifetime of the fluorophores , which allows for the assessment of intracellular K+ levels . Transfection of the channel in Human Embryonic Kidney HEK293T cells ( Figure 4—figure supplement 1A–L ) resulted in significant increase in CFP fluorescence lifetime due to decreased FRET of the sensor compared to control transfected cells ( Figure 4A , a-c , g ) , which indicated reduced intracellular K+ levels in the cells that express Kcnk5b ( Figure 4A , d-g ) . Additional higher resolution assessments along the lateral borders of cells showed similar increases in CFP fluorescence lifetime along the plasma membrane , indicating expected reduction of K+ levels at the cell membrane by active Kcnk5b ( Figure 4—figure supplement 1M–O ) . To test whether activity kcnk5b promotes the gene expression profile in mammalian cells that we observed in the zebrafish , we expressed kcnk5b either by establishing stable HEK293T ( HEK ) cells lines that either express GFP or zebrafish kcnk5b-GFP or by transient transfections . From qRT-PCR analyses comparing HEK cells expressing either GFP or kcnk5b-GFP , we observed an increase in SHH and PEA3 expression ( Figure 4Ba; Figure 4—figure supplement 2A ) and the down-regulation of LEF1 , ALDH1a2 and MSX1 ( Figure 4Bb; Figure 4—figure supplement 2B ) . To determine whether this transcriptional response is specific to Kcnk5b or is a general response to two-pore K+-leak channels , we transfected cells with one of two K+-leak channels Kcnk9 and Kcnk10 ( Figure 4—figure supplement 2C , D ) . Transfection of HEK cells with these two other channels resulted in a similar HEK-cell transcriptional profile as Kcnk5b for SHH and FGF ( Figure 4C ) , indicating that this transcriptional response to Kcnk5b is a response to the electrophysiological changes associated with intracellular K+ leak . The differences between the transcriptional responses of the zebrafish adult , larva , and HEK cells indicate that different cell types will have different responses to Kcnk5b electrophysiological activity . Therefore , we examined the transcriptional responses to Kcnk5b in other mammalian cell lines . In HeLa cells , Kcnk5b induced PEA3 and LEF ( Figure 4D ) . In the N2A ( neural carcinoma ) cell line , we observed the increase of ALDH1a2 ( Figure 4E ) but decreases in SHH , LEF1 and PEA3 ( Figure 4F ) . In the MCF7 epithelial carcinoma cell line , Kcnk5b induced ALDH1a2 , PEA3 , and MSX1 ( Figure 4G ) . When we tested whether Kcnk9 and Kcnk10 produce similar transcription profiles as Kcnk5b in HELA and N2A cells , we observed similar profiles between Kcnk9 and Kcnk10 ( Figure 4H–L ) , but differences between their outcomes and the transcriptional outcome of Kcnk5b . In HELA cells , the transcriptional response of PEA3 and ALDH2A to Kcnk9 and −10 were different than that of Kcnk5b ( Figure 4D , I–J ) . In N2A cells , while we observed similar increases in Aldh2A transcription and similar trends for Pea3 ( Figure 4E , F , K , L ) , we also observed the lack of down regulation of Shh , Lef1 and Pea3 ( Figure 4E , F , K , L ) . Together , these results reveal that Kcnk channel activity is sufficient to induce the transcription of genes associated with different developmental pathways in different mammalian cells types , and indicate that the downstream consequences of membrane potential changes are not intrinsic to specific signal transduction pathways . We propose that the variability in gene transcription in different cell types may explain why the solitary change in the activity of this channel in all cells of the fin leads to the variable transcriptional responses that promote coordinated growth of a multi-tissue anatomical structure . We previously showed that the phosphatase calcineurin acts as a molecular switch between isometric and allometric proportional growth of the zebrafish fins ( Kujawski et al . , 2014 ) . The similarities in the phenotypes produced by calcineurin inhibition and by the mutations in Knck5b that enhance Kcnk5b channel activity suggest a direct functional relationship between them ( Kujawski et al . , 2014; Perathoner et al . , 2014 ) . Based on whole-cell patch-clamp experiments , the mutations in kcnk5b that maintain allometric growth of the fins also increased K+ conductance at the plasma membrane ( Perathoner et al . , 2014 ) . Therefore , we hypothesized that calcineurin inhibition will increase in vivo K+ conductance and promote allometric growth of the zebrafish fins ( Figure 5A ) . To test whether calcineurin alters the channel activity of Kcnk5b , we examined the activity of Kcnk5b in the presence of calcineurin . Comparison of whole-cell patch-clamp measurements using HEK cells showed that Kcnk5b expression increases current density due to K+ leak from the cells ( Figure 5B ) , as we observed from FRET-FLIM intracellular K+ measurements ( Figure 4A ) . However , cells co-expressing both knck5b and calcineurin decreased the K+ conductance of the cells compared to cells expressing kcnk5b alone ( Figure 5B ) . We then tested whether inhibition of the endogenous calcineurin activity in the HEK cells by the calcineurin inhibitor FK506 affects Kcnk5b channel activity . We found that FK506 treatment of cells expressing kcnk5b resulted in a significant increase in K+ current compared with DMSO-treated kcnk5b-expressing cells ( Figure 5C ) . These results show that changes in calcineurin activity alter Kcnk5b channel activity in a manner that is constant with the enhanced fin growth induced by calcineurin inhibition and the increased channel activity of the kcnk5b zebrafish mutants , which indicates a functional interaction between calcineurin and Kcnk5b . Calcineurin interacts with its substrates at particular amino acid sequence sites ( Grigoriu et al . , 2013 ) . Our analysis of the amino acid sequence in the C-terminal cytoplasmic tail of Kcnk5b suggests a functional calcineurin binding site ( LVIP ) is present ( Figure 5A , red letters ) . To test for functional interaction at this site , we mutated the amino acid sequence ( Figure 5D , red letters ) and assessed how the mutation affected the ability of calcineurin to regulate the channel . Compared to the decrease in activity of the wild-type channel after co-transfection with calcineurin , co-transfection of the Kcnk5b mutant lacking the calcineurin binding site ( Kcnk5bmut + CaN ) showed that the mutation made the channel resistant to calcineurin-mediated inhibition ( Figure 5D ) . The resistance of the Kcnk5bmut to calcineurin indicated that the repression of channel activity on Kcnk5b by calcineurin is due to the interaction of these to proteins at the LVIP site . The regulation of Kcnk5b by calcineurin suggests that changes in calcineurin activity will have an effect on the Kcnk5b-dependent gene expression . To assess whether the activation of SHH by Knck5b can be altered by calcineurin , we compared the expression of SHH between HEK cells stably expressing GFP and HEK cells stably expressing the Kcnk5b channel as well as between HEK cells stably expressing the channel after transfection with calcineurin . We observed that compared to channel expression alone , co-expression of calcineurin decreased the Kcnk5b-mediated induction of SHH ( Figure 5E ) and PEA3 ( Figure 5F ) . To determine whether calcineurin effect on Kcnk5b-medidated SHH expression is specific to Kcnk5b , we transfected HEK cells with Kcnk9 or Kcnk10 . Both Kcnk9 and Kcnk10 lack identifiable calcineurin-binding sites ( Figure 4—figure supplement 2C , D ) , and we observed that unlike the effect on Kcnk5b , calcineurin had no effect on the induction of SHH ( Figure 5G ) or PEA3 ( Figure 5H ) by Kcnk9 or by Kcnk10 , indicating that calcineurin’s regulation of the electrophysiological induction of SHH and PEA3 transcription is specific to Kcnk5b . As a phosphatase , calcineurin should regulate Kcnk5b by dephosphorylating the channel at specific serine or threonine residues . A specific serine in the C-terminal tail represented a typical consensus serine-proline ( Ser345-Pro346 ) phosphorylation site for calcineurin ( Figure 6A ) . Therefore , we hypothesized that calcineurin inhibits the activity the Kcnk5b channel by dephosphorylating this serine . We tested whether rendering the Kcnk5b channel unphosphorylatable at this serine by alanine substitution ( S345A ) would decrease the channel’s activity . Whole-cell patch-clamp experiments of the kcnk5bS345A showed a significant decrease in K+ conductance of the channel compared to the wild-type ( kcnk5bS345 ) control ( Figure 6B ) . To assess the specificity of the reduction effect for this serine , we also systematically substituted adjacent serines with alanines and subsequently measured channel activity ( Figure 6—figure supplement 1A ) . While kcnk5bS345A showed reduction in activity , the substitution of other serines did not ( Figure 6C , Figure 6—figure supplement 1A–D ) . To determine whether the activity of the Kcnk5b channel is associated with the phosphorylation state of this serine , we exchanged the serine for a glutamic acid in order to mimic serine phosphorylation ( kcnk5bS345E ) ( Figure 6C ) . Expression of this mutant displayed elevated K+ conductance compared to knck5bS345 wild-type channel ( Figure 6D ) . In addition , the kcnk5bS345E mutant was resistant to calcineurin-mediated inhibition ( Figure 6D ) . Moreover , the substitution of other serines with glutamic acid had no effect on channel activity , and calcineurin could still regulated the channel ( Figure 6—figure supplement 1A , E , F ) . Together , these results indicate that S345 is the important post-translational regulatory serine involved in calcineurin-mediated regulation of Kcnk5b activity . To determine whether there is a functional relationship between fin scaling and S345-mediated Kcnk5b channel activity , we placed the cDNA of each channel version ( wild-type kcnk5bS345 , kcnk5bS345E , or kcnk5bS345A ) under the control of the heat-shock inducible hsp70 promoter to generate conditionally inducible transgenes for in vivo expression in the fish . We heat shocked the caudal fins of non-transgenic and transgenic fish lines once daily and subsequently measured the length of each fin in relation to the length of each body ( fin-to-body ratio ) . Non-transgenic siblings were included in the same heat-shock regimen to serve as controls . After 12 days of the heat-shock regimen , we noticed differences between the rates of the regenerating caudal fins lobes of the different transgenic lines ( Figure 6E ) after standardizing the length of each fin to the length of the body ( Figure 6—figure supplement 1G ) . By assessing regenerating lobe-to-body measurements over time , we observed that Tg[hsp70:kcnk5bS345E] fish maintained the highest rates of allometric regenerative growth , while Tg[hsp70:kcnk5bS345A] displayed the lowest growth rates of the transgenic lines ( Figure 6E ) . There was no significant change in the rates of growth of the bodies ( Figure 6—figure supplement 1G ) . We also assess the final proportional size of unamputated fin lobes between the different transgenic lines , and we observed a linear relationship between the proportional length of the unamputated lobes and the putative phosphorylation status of the channels: the S345A dephosphorylation mimic displayed the smallest growth proportions ( Figure 6I ) , the S345E phosphorylation mimic displayed the largest growth proportions ( Figure 6I ) , and the average value of the wild-type regulatable version of the channel was between the highly active S345E and marginally active S345A mutants ( Figure 6I ) . To determine whether similar growth effects are generated by other two-pore K+ leak channels , we generated a transgenic fish line that overexpresses kcnk9 , Tg[hsp70:kcnk9-GFP] . Compared to transgenic fish that were not heat-shocked ( Figure 6Ja , K ) and heat-shocked of non-transgenic fish ( Figure 6K ) , we observed that heat-shock-induced overexpression of kcnk9 in the adult fins increased fin outgrowth ( Figure 6Jb , K; Figure 6—figure supplement 1H ) , which further supports the conclusion that Kcnk5b-mediated growth is due to its regulation of its electrochemical properties . Our ability to control the rate of growth by mimicking a specific post-translational modification that can be mediated by calcineurin and that correspondingly determines the level of Kcnk5b activity supports the conclusion that calcineurin regulation of Kcnk5b is an in vivo electrophysiological mechanism through which controlling the potassium conductance of cells scales a vertebrate appendage . Anatomical structures consist of a combination of different tissue types that develop and grow in a coordinated manner . Recent discoveries show that K+ channels regulate the scaling of fish appendages , but it is still unclear how this electrophysiological signal controls several diverse developmental phenomena within this anatomical structure to achieve coordinated developmental growth . Our results reveal that this in vivo electrical signal has a hierarchical effect on specific developmental genes in the fish fins and larva . Two-pore K+-leak channels such as Kcnk5b allow K+ to cross the membrane to establish an electrochemical equilibrium , this activity directly affects the membrane potential of the cell ( Goldstein et al . , 2001 ) . Normally , the concentration of K+ is higher on the cytoplasmic side of the plasma membrane due to continual active transport of K+ into the cell by the ATP-dependent Na+/K+ pumps ( Shattock et al . , 2015 ) . As a leak channel , opening of Kcnk5b causes a flow of K+ out of the cell , which hyperpolarizes the membrane potential ( Goldstein et al . , 2001 ) . The previous findings that mutations that increase Kcnk5b channel activity are associated with maintaining allometric growth ( Perathoner et al . , 2014 ) argue that such changes in membrane potential promote disproportional growth . Our current findings further these previous findings by showing that conditional induction of Kcnk channel activity is sufficient to induce morphogen pathways ( Figures 1 , 2 and 4 ) in different in vivo and in vitro contexts , demonstrating transcriptional control of particular developmental mechanisms by different two-pore K+-leak channels . In addition to K+-leak channels , cells regulate intracellular K+ through different channels and exchangers . Inward rectifying K+ channels allow K+ to enter the cell along the ion’s electrochemical gradient . Exchangers will exchange K+ with different substrates ( e . g . Na+ ) to facilitate the entry or removal of K+ based on the concentration gradient of K+ and the exchanged substrate . Previous findings show the importance of the inward rectifying K+ channel Kir2 for cranial-facial and digit defects in humans ( Andersen et al . , 1971; Canún et al . , 1999; Sansone et al . , 1997; Tawil et al . , 1994; Yoon et al . , 2006a; Yoon et al . , 2006b ) . Knockout of the mouse Kir2 channels results in similar head and digit defects ( Zaritsky et al . , 2000 ) , and dominant-negative inhibition of the Drosophila Kir2 leads to wing appendage defects that are analogous to the human and mouse appendage defects ( Dahal et al . , 2012 ) . While the mammalian phenotypes remain unexplained , the defects in the Drosophila wings have been linked to reduced Dpp ( BMP ) signaling ( Dahal et al . , 2012 ) , suggesting that intracellular K+ homeostasis is important for BMP signaling . Removal of an ATP-sensitive K+ channel in the early Xenopus embryo disrupts eye formation , while ectopic expression of this channel will produce ectopic eyes in the head and in locations that were not considered to be competent for producing eyes ( Pai et al . , 2012 ) . The ability to ectopically generate eyes was linked to electrophysiological hyperpolarization of the cells and the activation of Pax6-eyeless gene ( Pai et al . , 2012 ) , a master regulator for eye development ( Chow et al . , 1999; Halder et al . , 1995 ) . In planaria , shortly after wounding , membrane depolarization acts as an early anterior signal that is sufficient ( even when induced on the posterior side ) to promote the consequent formation of all the anterior structures of the planarian head by inducing notum expression , which inhibits β-catenin-dependent Wnt signal transduction ( Durant et al . , 2019 ) . Furthermore , it was recently revealed that in the Drosophila wing discs , depolarization events promote membrane localization of the transmembrane receptor smoothened to promote Shh signaling and that membrane potential values are patterned within the wing disc ( Emmons-Bell and Hariharan , 2021 ) . These discoveries show that electrophysiological changes are important signals in the formation and growth of anatomical structures . A recent finding showed that calcineurin inhibition or increased activity of the voltage-gated Kcnh2 channel ( lof mutant ) promotes regenerative outgrowth from the proximal side of a mid-fin excavation , a site that normally does not mount a regenerative outgrowth response ( Cao et al . , 2021 ) . Our findings help explain how calcineurin-regulated and electrophysiological changes from potassium channels can lead to broader tissue organizing phenomena by showing the inductive effect that increasing K+ conductance ( and calcineurin’s regulation of K+ conductance ) can have on a broad number of developmental pathways , which is important for coordinating the organized formation of tissues and organs . Furthermore , we posit that the effect of the activity of the Kcnk5b channel is broader than the traditional mechanisms of growth factor/morphogen signaling pathways , because it is not confined to specific signal transduction mechanisms; rather , it has variable broad effects , such as activation of several developmental signals in the adult fin ( Figure 1 ) , the larva ( Figure 2 ) and different mammalian cell lines ( Figure 4 ) . While we observed that Kcnk5b is sufficient to promote lef1-mediated transcription ( Figures 1–3 ) , we did not observe direct activation of axin2 , a down-stream target gene of β-catenin activity ( Figure 1—figure supplement 1C ) . We posit that Kcnk5b activity does not directly activate canonical Wnt , but primes cells for increased β-catenin activity upon the reception of a Wnt signal through up-regulation of Lef1 . Ultimately , continued Kcnk5b activity must lead to increased β-catenin-dependent Wnt signaling , since all evidence indicates that Wnt signaling is required for allometric fin outgrowth ( Kawakami et al . , 2006; Stoick-Cooper et al . , 2007 ) , and allometric fin growth is the phenotype that we get by kcnk5b or kcnk9 overexpression in amputated and unamputated fins ( Figure 6E–L ) . We propose that the competence to activate different developmental pathways by electrophysiological changes is because the responding cells are either primed to activate them or the pathways are already active . It will be important to find out how this electrophysiological signal coordinates the activity of these developmental signals . In this regard , only few factors are known that regulate shh and lef1 transcription . Thus , our finding that an electrophysiological mechanism is involved not only provides a new understanding of how electrophysiology acts as an inductive signal , it also may lead to the discovery of molecular mechanisms that control the expression of these mediators of important morphogen signals . The scaling activity of Kcnk5b includes all the tissues of the entire appendages of the fish ( Perathoner et al . , 2014 ) . Previous findings implicate broader intercellular electrophysiological gradients as a mechanism for tissue growth ( Adams and Levin , 2013 ) . Electrophysiological measurements of animal tissues show that electric fields are generated and are important in vivo ( Borges et al . , 1979; Jenkins et al . , 1996; McGinnis and Vanable , 1986 ) , which suggests the existence of in vivo bioelectric information that regulates physiological phenomena . However , from our transplantation experiments , we observe that Kcnk5b’s effect is cell autonomous ( Figure 3 ) . Consequently , the question arises about how the activity of this K+-leak channel relates to a broad , coordinated phenotype of scaling the several tissues of the fin . An answer is that the autonomous transcriptional programs include morphogens . What is unclear is whether a limited number of cells in the fin control the growth and organizing information so that Kcnk5b only needs to act on a limited number of cell types , or whether Kcnk5b regulates proportional growth at multiple levels and that the cell autonomous transcriptional response that we observe is one outcome of a combination of intracellular and intercellular responses induced by Kcnk5b . Changes in membrane potential from alterations in K+ conductance are also associated with the progression through the cell cycle ( Blackiston et al . , 2009; Urrego et al . , 2014 ) , because K+ channel activity increases at specific cell cycle phases ( Urrego et al . , 2014 ) , and inhibition of K+ channel activity leads to cell cycle arrest in many different tissue cell types ( Blackiston et al . , 2009 ) . It is possible that this phenomenon explains part of Kcnk5b’s ability to promote allometric growth . We do not yet know whether other phenomena linked to the activity of mammalian Kcnk5 [influence cell tonicity ( Niemeyer et al . , 2001 ) , metabolic acidosis and alkalinization ( Warth et al . , 2004 ) , CO2/O2 chemosensing in retrotrapezoid nucleus neurons ( Flores et al . , 2011 ) and apoptosis in lymphocytes and neurons ( Göb et al . , 2015; Nam et al . , 2011 ) ] are involved in appendage scaling . We previously showed that calcineurin inhibition shifts isometric growth to allometric growth ( Kujawski et al . , 2014 ) . Subsequently , Daane et al . showed that this effect is reversible in that removal of calcineurin inhibitors restores isometric growth . The authors also suggest an alternative mechanism for calcineurin regulation of Kcnk5b in its C-terminus ( Daane et al . , 2018 ) ; however , their model posits that active calcineurin promotes the activity of Kcnk5b , which does not yet explain how calcineurin inhibition and increased Kcnk5b activity each promote allometric growth ( Kujawski et al . , 2014; Perathoner et al . , 2014 ) . In either case , these data implicate calcineurin as a molecular switch governing isometric versus allometric growth control . Our findings provide a mechanism for how this switch acts to scale the fish appendages by directly regulating the activity of Kcnk5b through the dephosphorylation and phosphorylation of a specific serine ( Figure 7 ) . The ability to mimic or block calcineurin regulation of this K+-leak channel ( Figure 5 ) , whose activity levels directly translate into the extent of allometric growth ( Figure 6 ) , defines how calcineurin inhibition expands clonal populations during fin regeneration ( Tornini et al . , 2016 ) . However , as we observed from both calcineurin inhibition ( Kujawski et al . , 2014 ) and from conditionally inducing Kcnk5b activity ( Figure 6 ) , the induced allometric growth of the entire fin is more than expanding clonal populations , since the outcome is not tumorigenesis . Instead , the growth is coordinated among all the tissues ( Figure 6G–I , K , L; Kujawski et al . , 2014; Perathoner et al . , 2014 ) , and our finding that Kcnk5b activates several developmental pathways ( Figures 1 , 2 and 4 ) argues that calcineurin activity acting on Kcnk5b regulates more than cell cycle progression . An important next step is to learn how the calcineurin-Kcnk5b circuit is integrated into the broader mechanisms that scale the appendages . Calcineurin is a Ca2+-dependent enzyme which suggests that intracellular Ca2+ is involved in scaling information . Ca2+ is a broad second messenger that can activate several downstream Ca2+-dependent enzymes , so broad changes in its subcellular levels likely have multiple effects . It remains unclear whether there is a specific intracellular distribution pattern that leads to calcineurin-mediated control of scaling . It is also possible that Ca2+-mediated activation of calcineurin—and consequent restoration of isometric growth—is so dominant that other Ca2+-mediated activities have little effect . Two mechanisms that regulate proportional growth of organs are vitamin D and Hippo signaling . Increasing vitamin D signaling enhance the growth of the entire body , including the fins ( Han et al . , 2019 ) . We propose that vitamin D is a systemic body signal that ultimately leads to the increase in Kcnk5b signaling . It is also possible that this hormone acts independently of Kcnk5b . In Drosophila , the Hippo pathway regulates brain size and size of the imaginal discs ( Poon et al . , 2016; Rogulja et al . , 2008 ) . Mice overexpressing a nuclear version of the Hippo-signaling component Yap1 in the adult liver develop significantly enlarged livers ( Camargo et al . , 2007; Dong et al . , 2007 ) . The Hippo signal transduction pathway consists of several core components that can be regulated by different factors at plasma membrane and within the cell ( Yu and Guan , 2013 ) , so there are several possible nodes of interaction between of Kcnk5b and Hippo cascade . It is also possible the Hippo-mediated transcription regulates kcnk5b expression or channel activity . Connexin43 also regulates proportional growth of the fins , since mutations that reduce the intercellular connectivity of connexin43 produce adult fins that are half the size as the fins of wild-type siblings ( Hoptak-Solga et al . , 2007; Iovine et al . , 2005 ) . The connective nature of these intercellular junction proteins indicate that direct communication between intracellular compartments of tissue cells is an important component of the scaling mechanism of the fins . Our observation that Kcnk5b cell-autonomously activates the 7XTCF-Xla . sam:mCherry reporter ( Figure 3 ) indicates that it is not due to intracellular transfer of K+ . It is still unclear whether the disruption of intracellular trafficking of other ions ( such as Na+ or Ca2+ ) or of other factors is responsible for the connexin43’s effect on scaling . Kcnk5b’s ability to activate the Lef1-dependent reporter cell autonomously in different tissue types supports the conclusion that K+ conductance has the potential to regulate developmental transcription in a broad range of tissues ( Figure 3B–E ) . The observation that neuronal cells in the brain and myocytes in the trunk muscle respond similarly to non-excitable cells elsewhere in the body suggests that even cells that harbor action potentials use K+ conductance to regulate gene expression . However , when we either stably expressed or transient transfected Kcnk5b in different cell lines , we did not observe consistent activation of Lef1 or Shh ( Figure 4 ) . We posit that the competencies of the cultured cell lines are different from the developmental competencies of the in vivo cells . Also , the mammalian cell lines are cancer cells , so they may already have an increased transcriptional base line for the selected developmental genes that the two-pore potassium-leak channels can only partially affect . While we observed similarities between Kcnk5b and Kcnk9 or Kcnk10 in HEK cells , in Hela and N2A cells , we observed only partially similar profiles for the selected genes ( while Kcnk9 and Kcnk10 were similar ) ( Figure 4 ) . We postulate that the observed differences may come from one or both of the following explanations . First , different responses from channels of the same ion-type are due to different levels of membrane potential changes , which results in different levels of gene transcription . Second , these channels have different intracellular sequences , which may determine other unknown intermolecular interactions that have different signal transduction properties . In any case , we did observe that transgenic overexpression of Kcnk9 produces a similar allometric growth of the caudal fin ( Figure 6K , L ) . It remains to be explored whether intermolecular interactions between the channels and another protein contribute to the scaling of other organs as well as how other electrophysiological mechanisms that control membrane potential have the same growth effect . In conclusion , we show how a specific electrophysiological mechanism activates important morphogen pathways to scale tissues in different in vivo contexts . We propose the observed diversity in morphogen and growth factor expression to Kcnk5b activity explains why the increased activity of Kcnk5b produces the diverse transcriptional response in the different tissues associated with the observed coordinated outgrowth of the entire fin . Also , we show how changes in phosphorylation of S345 in the cytoplasmic C-terminus can be regulated by calcineurin to directly control electrophysiological activity of the channel to scale the fin . Thus , we offer an in vivo paradigm in which membrane potential acts as potent regulator of coordinated developmental signaling , and we show how the two-pore K+-leak channel Kcnk5b is involved in the initial activation of specific developmental mechanisms that lead to the scaling of the fish fin appendages . Constructs were designed either using standard restriction enzyme or by homologous recombination methods . kcnk5b cDNA was isolated by RT-PCR from regenerating adult fin cDNA library and cloned into MCS region of pcDNA6-myc-6xHIS-tag plasmid ( Invitrogen ) or pBluescript harboring the hsp70 zebrafish promoter and GFP coding sequence surrounded by two miniTol2 sites . Mutagenesis of the Serine345 codon of Kcnk5b was performed using QuikChange Mutagenesis kit ( Agilent ) . AB strain fish were raised in 10L tanks with constantly flowing water , 26°C standard light-dark cycle ( Brand et al . , 2002 ) in either a Schwarz ( DFG-Center for Regeneration , TU Dresden ) or HaiSheng aquarium ( ShanghaiTech University ) systems . Transgenic lines harboring the different kcnk5b transgenes were created by injecting 300 µg of each construct together with mRNA of Tol2 transposase ( Balciunas et al . , 2006 ) . Fish harboring Tg[7XTCF-Xla . sam:mCherry] transgene are deposited in the European Zebrafish Resource Center . Fish embryos and larva were raised in 1xE3 medium ( 5 mM NaCl , 0 . 17 mM KCl , 0 . 33 mM CaCl2 , 0 . 33 mM MgSO4 , 10−5% Methylene Blue ) until 10-12dpf , then transferred to aquarium water tanks to grow . Transgenic lines established by screening for GFP expression after heat shock . Experiments used male and female fish equally . Fish experiments were compliant to the general animal welfare guidelines and protocols approved by legally authorized animal welfare committees ( Technische Universität Dresden , Landesdirektion Dresden , and ShanghaiTech University , ShanghaiTech Animal Welfare Committee ) . Parents of heat-shock-driven transgenic lines were either outcrossed to same-strain wild-type fish or to fish harboring Tg[7XTCF-Xla . sam:mCherry]transgene . Progeny were collected in 1xE3 and raised at 28°C . Carriers were confirmed positive for their respective transgenes by a single heat shock at 37°C for 1 hr . For embryo experiments , heat shock was at 12 hpf or 32 hpf ( pectoral fin bud ) in 37°C E3 medium for 30 min . For larva the heat shock was in 37°C E3 medium for 30 min at 2 dpf . For adult fin , 6-month-old fish underwent a daily heat-chock regimen: first , sedated in 0 . 04% tricane in aquarium water , then placed in conical tubes containing 0 . 04% tricane in aquarium water to allow continued gill movement in oxygenated water and allow the caudal fins to be exposed to 37°C water for 7 min . After heat shocking the caudal fin , the fish were returned to flowing aquarium water and monitored daily . Caudal fin measurements were made from the base of the fin to the distal tip along the third fin ray of each fin lobe counting in from the dorsal-most and ventral-most sides . Body lengths were determined by measuring the most anterior point of the jaw to the base of the caudal fin . Larval finfold measurements were from the ventral-most to dorsal-most points of the caudal finfold or from the anterior-most to the posterior-most points of the ventral finfold along the body length . Body lengths of larva were measured from the anterior-most point of the jaw to the posterior-most point of the somatic musculature . Fins were amputated and snap frozen in liquid nitrogen and homogenized in RIPA buffer ( Proteintech , B100020 ) with protease inhibitor ( Pierce , 88666 ) and then rotated at 4℃ for 2 hr to lysate cells thoroughly . Protein was collected and concentration was quantified by BCA assay ( MDBio , KT054-200rxn ) after centrifuge at 12000 rpm at 4℃ for 20 min . Then gel electrophoresis was conducted at 120V for 90 min using page gels ( Genscript , M81615C ) , protein were then transferred to PVDF membranes ( Bio-Rad , 1620177 ) . Anti-lef1 ( 1:1000 , Cell Signaling Technology , 2230T , ( RRID: AB_823558 ) ) , Anti-shh ( 1:1000 , Novus , NBP2-22139 , RRID: AB_331149 ) , Anti-β-catenin ( 1:1000 , Cell Signaling Technology , 9562L ) , Anti-β-Actin pAb-HRP-DirecT ( 1:2000 , MBL , PM053-7 , RRID: AB_10697035 ) were used as primary antibodies to incubate overnight at 4℃ . Anti-Rabbit IgG ( 1:20000 , Jackson , 711-036-152 , RRID: AB_2340590 ) were used as secondary antibody and blots were detected using ECL detection ( PerkinElmer ( PE ) , NEL105001EA ) , and Image J software ( RRID:SCR_003070 ) was used for density quantification . mRNA probes were made from RT-PCR products isolated from two dpf zebrafish embryos 6 hr after 37 . 5°C heat shock . The primer sequences for generating the probes are F- shha:5’- TGCGGCTTTTGACGAGAGTGC-3’R-shha: 5’-GGTAATACGACTCACTATAGGG TTTCCCGCGCTGTCTGCCG-3’ F-lef1: 5’-GAGTTGGACAGATGACCCCTCCTC-3’; R-lef1: 5’-GGTAATACGACTCA CTATAGGGGCAGACCATCCTGGGTAAAG-3’ . in vitro transcription reagents are from Promega . Isolated fish fins were surgically isolated and incubated in 4% PFA in 1xPBS at 4°C overnight with gentle rocking . Samples were subsequently washed 5 times in 1xPBS and then dehydrated by incubation for 15 min in a graded series of increasing methanol/1xPBS solutions ( 25% , 50% , 75% , 3 × 100% ) on ice . Fins were then incubated in 100% methanol for ≥2 hr at −20°C . Samples were then rehydrated using the reversed dehydration series of ( methanol/1xPBS solutions ) . Samples were then incubated more than 4x in 1xPBS to remove all methanol , and subsequently incubated in 10 µM Proteinase K for 10 min at RT . Samples were then incubated 20 min . in 4% PFA/1xPBS to inactivate the Proteinase K . Samples were incubated in 1xPBS 6 × 10 mins to remove the PFA , then incubated in pre-hybridization buffer for 2 hr at 65°C . Samples were subsequently incubated in the hybridization solution containing 200 ng/ml of each mRNA probe ≥14 hr at 65°C . Samples then were washed with successive wash steps to remove unbound probe and prepare for antibody incubation: twice 2xSSC/50% deionized formalin at 65°C , twice 2xSSC/25% deionized formalin at 65°C , 2xSSC at 65°C , twice 0 . 2xSSC at RT , 6 times 1xPBST ( 1xPBS with Tween-20 ) , once in blocking solution ( 2% Bovine albumin [Sigma-Aldrich , A3294-100G] , 2% Sheep Serum [Meilunbio , M134510] ) at RT for 4 hr . Samples were incubated with Anti-digoxigenin-AP Fab Fragment ( Sigma-Aldrich , 11093274910 , RRID: AB_514497 ) in blocking solution ≥14 hr at 4°C . Samples were then washed 6 times with 1xPBST , subsequently incubated in [0 . 1 M Tris-HCl , pH 9 . 5 , 0 . 1 M NaCl , 0 . 05 M MgCl2] 3 times for 30 min , and then in Nitro Blue Tetrazolium ( Sigma-Aldrich , N6639-1G ) and 5-bromo-4-chloro-3-indolyl phosphate ( Sigma-Aldrich , 136149 ) in [0 . 1 M Tris-HCl , pH 9 . 5 , 0 . 1 M NaCl , 0 . 05 M MgCl2] at RT ≥8 hr . Samples images under Stemi508 stereoscope ( Zeiss ) with Axiocam ERc5s digital color camera ( Zeiss ) and Zen2 . 3 software ( Zeiss , RRID: SCR_013672 ) . For signal areas and signal intensities analyses , images of the shh in situs were transferred to Fiji ImageJ ( RRID:SCR_003070 ) . The stained regions were traced using ‘free ROI’ and then quantitated using the ‘measure’ function under the ‘analysis’ menu to calculate the number of pixels contained in the stained area . For signal intensity analysis , the signal region and adjacent unstained regions were measured by ‘free ROI’ selection and the mean intensity pixel values were determined from the ‘measure’ function under the analysis menu by subtracting the mean value of the unstained region from the mean value of the signal region for each fin bud . Zebrafish fins or larvae euthanized in 1% Mesab were fixed in 4%PFA/1xPBS and embedded in 1% agarose ( CryoStar NX50 , Thermofisher ) or tissue freezing medium ( Leica , 14020108926 ) on dry ice . ( CryoStar NX50 , Thermofisher ) before cryosectioning . 10 μm sections were mounted on glass slides ( Titan , 02036398 ) and dried . The tissue freezing medium ( Leica ) was removed in ddH2O for 10 min . Sections were permeabilized in 0 . 1%Triton-X for 5 min and incubated in 1% BSA/1xPBS/0 . 1%Tween-20 ( PBST ) at room temperature for 30 min . Sections were incubated in a mouse-anti-GFP ( Invitrogen , MA5-15349 , RRID:AB_10987186 ) solution ( 1:400 ) , Anti-β-catenin ( Cell Signaling , 9562L , RRID: AB_331149 ) , Anti-Lef1 ( Cell Signaling , 2230T , RRID: AB_823558 ) solution ( 1:1000 ) , Anti-Shh ( Novus , NBP2-22139 , RRID: AB_2883969 ) solution ( 1:1000 dilution ) , rabbit-anti-mCherry antibody ( Invitrogen , P5-34974 , RRID: AB_2552323 ) solution ( 1:2000 dilution ) in PBST at 4°C overnight ( >12 hr ) . The primary antibody solution was replaced by a goat-anti-mouse-GFP ( Abcam , ab150113 , RRID: AB_2576208 ) , goat-anti-rabbit-mCherry secondary antibody ( Abcam , ab150078 , RRID: AB_2722519 ) solution ( 1:2000 dilution ) in PBST and incubated at RT in the dark for 60 min . The secondary antibody solution was then replaced with DAPI solution ( Roche , 10236276 ) in the dark at room temperature for 5 min . DAPI were washed away in 1xPBST 3 × 5 min incubations at RT . Coverslips ( Titan , 02036401 ) were mounted with a 40% glycerol solution and sealed with nail polish . The sections were visualized using an LSM710 upright scanning confocal ( Zeiss ) or a LSM880 inverted scanning confocal ( Zeiss ) with ZENBlue software ( Zeiss , RRID: SCR_013672 ) . Images were processed with Fiji software ( RID:SCR_003070 ) . Using Fiji ImageJ , multiple nuclei of cells in IHC fin cross-section image of control fins were manually measured , and a mean mRFP fluorescence intensity value was calculated . This mean value was used as the baseline for assessing nuclear β-catenin levels . β-catenin nuclear values for all the nuclei in each cross-section were assessed with ImageJ by splitting the combined images of β-catenin and DAPI and using DAPI to have Fiji ImageJ define and select all nuclei in the image . The nuclear β-catenin levels were determined in selected nuclei by intensity analysis in Fiji ImageJ ( RID:SCR_003070 ) , which provided a numeric value for the β-catenin channel in all the nuclei of each section . Transgenic fish lines [hsp70:kcnk5b-GFP] and [7XTCF-Xla . sam:mCherry] were inbred for 20–30 min before embryos were collected in E3 . The genotypes of the parents and of the imaged progeny were confirmed to be homozygous for the 7XTCF-Xla . sam:mCherry transgene by qPCR ( Table 1 ) . Primers for genomic β-actin locus were used as DNA content standardization . The cycle procedure was at 50 . 0°C for 2 min , 95 . 0°C for 10 min in the hold step; 95 . 0°C for 15 s , 55 . 0°C for 20 s for 40 routine in the elongation step; 95 . 0°C for 15 s , 60 . 0°C for 1 min , 95 . 0°C for 15 s in the melting curve step . The embryos were left to develop in E3 for 3–3 . 5 hr at 28⁰C until the blastula stage . Embryos were then placed in agarose ridges for easy access under a Zeiss stereomicroscope and cells from the [hsp70:kcnk5b-GFP] embryos were isolated by air suction via a glass needle mounted on a Precision Instruments piston . The [hsp70:kcnk5b-GFP] cells were then transplanted into the [7XTCF-Xla . sam:mCherry] embryos . The transplants were carefully moved from the agarose into clean 90 mm dishes with fresh E3 ( ±20–25 per dish ) and incubated at 28°C . One or two days after transplantation , fish were heat shocked for 1 hr at 37°C . Four hours after heat shock , 48- or 72 hr fish were embedded into 1% low melting agarose ( Sigma-Aldrich-Aldrich , A9414-250G ) supplemented with Mesab in 35 mm glass bottom confocal dishes ( Cellvis: D35-20-1-N ) and turned to their side . Visualized with a LSM710 confocal argon laser microscope ( Zeiss ) with ZENBlue software ( Zeiss , RRID: SCR_013672 ) . The cDNA that was used for qRT-PCR was extracted from the HEK293T cells and zebrafish . The mRNA was isolated using Tri-Reagent ( CWBio , 03917 ) . Then 1 μg mRNA was used for the reverse transcription to cDNA using 4x gDNA wiper Mix , 5x HiScript III qRT SuperMix ( Vazyme , L/N 7E350C9 ) . qRT-PCR was performed using 2x ChamQ Universal SYBR qPCR Master Mix ( Vazyme , L/N TE342F9 ) with QuantStudio3 machine ( Thermofisher ) with the following primers ( Tables 2 , 3 , 4 ) ) . The cycle procedure was at 50 . 0°C for 2 min , 95 . 0°C for 10 min in the stage 1; 95 . 0°C for 15 s , 60 . 0°C for 20s for 40 routine in the stage 2; 95 . 0°C for 15s , 60 . 0°C for 1 min , 95 . 0°C for 15s in the Melt Curve . Samples were standardized using the detected β-actin expression for fish mRNA isolation and GAPDH for human cell line mRNA isolations . The data was analyzed in the ΔΔCt method . All cell culture lines were incubated at 37°C , 5% CO2 , 95% humidity in incubators ( Thermofisher , FORMA STERI-CYCLE i160 ) in DMEM medium ( Gibco , 1199506 ) with 10% FBS ( Gibco , 1009914 ) and 1% penicillin streptomycin ( Gibco , 15140122 ) . The identity of the cell lines was not authenticated , and mycoplasma was not determined . Cell were split to 50% density and transfected with Lipofectamine ( Invitrogen , 11668–019 ) 12 hr later . Expression for the transfected constructs was evaluated by expression of fluorescent marker . Two μl of DMEM medium from different cultured cells are freshly taken as template . Primers to detect mycoplasma are used as F: 5’-GGGAGCAAACAGGATTAGATACCCT-3’; R: 5’- TGCACCATCTGTCACTCTGTTAACCTC-3’ , then standard PCR was conducted ( TOYOBO , KOD-401 ) . The procedure is: Step 1: 98°C for 2 min . Step 2: 98°C for 10 s , 55°C for 30 s , and 68°C for 30 s for 30 cycles . Step 3: 68°C for 10 min , 4°C to hold . Then gel electrophoresis was conducted at 120 v for 30 min . Hek293T cells were transfected with 1 µg of pcDNA-kcnk5b-GFP and 1 µg of the pcDNA6-Kirin-FRET sensor ( Shen et al . , 2019 ) . Fluorescence lifetime imaging measurements were made by photon counting the fluorescence emission of CFP using a two-photon-confocal Hyperscope ( Scientifica , UK ) and PMT-hybrid 40 MOD five photon detectors ( Picoquant , Germany ) . Counted photon emissions were calculated and analyzed using SymPhoTime 64 , version 2 . 4 ( Picoquant , Germany , RRID: SCR_016263 ) . Transfected HEK293T cells were seeded on glass coverslips ( Fisher Brand ) , and incubated in cell culture medium at 37°C , 5% CO2 , 95% relative humidity for 4–6 hr . The seeded coverslips were transferred into Tyrode’s solution ( 138 mM NaCl , 4 mM KCl , 2 mM CaCl2 , 1 mM MgCl2 , 0 . 33 mM NaH2PO4 , 10 mM Glucose , 10 mM HEPES ) . Cells were assessed in the ruptured-patch whole-cell configuration of the patch-clamp technique using and EPC9 or EPC10 amplifier ( HEKA ) with borosilicate glass pipettes ( Sutter Instruments ) with 3–6 MΩ resistance when filled with pipette solution ( 130 mM glutamic acid , 10 mM KCl , 4 mM MgCl2 , 10 mM HEPES , 2 mM ATP , pH to 7 . 2 ) . After gigaseal formation , cells were voltage-clamped at −80 mV . Potassium conductance was elicited by test pulses from −100 mV to 70 mV ( in 10 mV increments ) of 600 ms duration at a cycle length of 10 s . The resulting tracings were converted into itx files by the ABF Software ( ABF Software , Inc , RRID: SCR_019222 ) and then analyzed using Clampfit Software ( Molecular Devices , RRID: SCR_011323 ) . Currents were measured at the end of the test pulses .	Organs , limbs , fins and tails are made of multiple tissues whose growth is controlled by specific signals and genetic programmes . All these different cell populations must work together during development or regeneration to form a complete structure that is the right size in relation to the rest of the body . Growing evidence suggests that this synchronicity might be down to electric signals , which are created by movements of charged particles in and out of cells . In particular , previous work has identified two factors that control the development of fins in fish: the Kcnk5b potassium-leak channel , which allows positive ions to cross the cell membrane; and an enzyme called calcineurin , which can modify the activity of proteins . Kcnk5b and calcineurin seem to play similar roles in the proportional growth of the fins in relation to the body , but exactly how was unknown . To investigate this question , Yi et al . used genetically modified zebrafish to show how the Kcnk5b channel could control genes responsible for appendage growth . However , their tests on different cell types revealed that potassium movement through the Kcnk5b channel leads to different sets of developmental genes being turned on , depending on the tissue type of the cell . This could explain how one type of signal ( in this case , movement of ions ) can coordinate the growth of a wide range of tissues that use different combinations of developmental genes to form . Kcnk5b therefore appears to coordinate the regulation of the various combinations of genes needed for different fin tissues to develop , so that every component grows in a proportional , synchronized manner . Yi et al . also showed that calcineurin can modify the Kcnk5b channel to control its activity . In turn , this affects the movement of potassium ions across the membrane , changing electrical activity and , as a consequence , the proportional growth of the fin . Further work should explore how Kcnk5b and calcineurin link to other signals that regulate the size of fins and limbs . Ultimately , a finer understanding of the molecules controlling the growth of body parts will be useful in fields such as regenerative medicine or stem cell biology , which attempt to build organs for clinical therapies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2021	A calcineurin-mediated scaling mechanism that controls a K+-leak channel to regulate morphogen and growth factor transcription
Tumor-associated macrophage ( TAM ) significantly contributes to cancer progression . Human cancer is enhanced by PPARγ loss-of-function mutations , but inhibited by PPARγ agonists such as TZD diabetes drugs including rosiglitazone . However , it remains enigmatic whether and how macrophage contributes to PPARγ tumor-suppressive functions . Here we report that macrophage PPARγ deletion in mice not only exacerbates mammary tumor development but also impairs the anti-tumor effects of rosiglitazone . Mechanistically , we identify Gpr132 as a novel direct PPARγ target in macrophage whose expression is enhanced by PPARγ loss but repressed by PPARγ activation . Functionally , macrophage Gpr132 is pro-inflammatory and pro-tumor . Genetic Gpr132 deletion not only retards inflammation and cancer growth but also abrogates the anti-tumor effects of PPARγ and rosiglitazone . Pharmacological Gpr132 inhibition significantly impedes mammary tumor malignancy . These findings uncover macrophage PPARγ and Gpr132 as critical TAM modulators , new cancer therapeutic targets , and essential mediators of TZD anti-cancer effects . How immune cells in the tumor microenvironment modulate cancer malignancy is a fundamental and fascinating question with tremendous therapeutic significance for cancer intervention . Emerging evidence supports a functional association between inflammation and cancer . Chronic inflammation is implicated in >15% of cancers ( Coussens and Werb , 2002 ) and shown to promote tumorigenesis ( Blaser et al . , 1995; Kuper et al . , 2000; Scholl et al . , 1994; Shacter and Weitzman , 2002 ) . As a key player in inflammation and cancer progression , TAM strongly correlates with poor cancer prognosis ( Bingle et al . , 2002; Noy and Pollard , 2014; Qian and Pollard , 2010; Ruffell et al . , 2012 ) . For example , overexpression of macrophage colony-stimulating factor 1 ( CSF1 or M-CSF ) leads to accelerated tumor progression in mice and human ( Lin et al . , 2001; Scholl et al . , 1994 ) . Moreover , TAM also modulates therapeutic responses ( Ruffell and Coussens , 2015 ) . Although numerous clinical studies and experimental mouse models support that macrophages generally play a pro-cancer role , anti-tumor property has also been reported for certain subtypes of macrophages , suggesting that macrophage regulation of cancer malignancy is pleotropic and context-dependent ( Krzeszinski and Wan , 2015; Noy and Pollard , 2014; Qian and Pollard , 2010; Ruffell et al . , 2012; Ruffell and Coussens , 2015 ) . Peroxisome proliferator-activated receptor gamma ( PPARγ ) is a nuclear receptor and a transcription factor that regulates a myriad of physiological processes ( Ahmadian et al . , 2013; Lefterova et al . , 2014 ) . PPARγ loss-of-function mutations have been associated with human cancer development ( Aldred et al . , 2003; Sarraf et al . , 1999 ) . Synthetic PPARγ agonists , such as the TZD anti-diabetic drugs rosiglitazone and pioglitazone , have been implicated to inhibit tumor malignancy ( Apostoli et al . , 2015; Bosetti et al . , 2013; Drzewoski et al . , 2011; Feng et al . , 2011; Fenner and Elstner , 2005; Fröhlich and Wahl , 2015; Kumar et al . , 2009; Monami et al . , 2014; Skelhorne-Gross et al . , 2012; Uray et al . , 2012 ) . These evidences suggest that PPARγ exerts anti-tumor effects , although a lack of correlation or pro-tumor effects have also been reported ( Saez et al . , 2004 , 1998 ) . Limited clinical trials to date are inconclusive on the effects of TZDs on human cancer outcome ( Burstein et al . , 2003; Mueller et al . , 2000 ) . Provocatively , a meta-analysis of randomized clinical trials reveal that the incidence of cancer malignancies was significantly lower in rosiglitazone-treated patients than in control groups , although rosiglitazone did not significantly modify the risk of cancer ( Monami et al . , 2008 ) . These findings not only support an anti-tumor role of TZDs but also suggest that TZDs may act to impede tumor progression rather than tumor initiation . Importantly , epidemiological studies suggest a bidirectional association between diabetes and cancer: diabetes ( especially type 2 ) correlates with higher risk of cancers including breast cancer ( Park et al . , 2014; Smith and Gale , 2009 , 2010 ) ; conversely , 8–18% of newly diagnosed cancer patients are diabetic , and cancer patients with preexisting diabetes are 50% more likely to die after surgery ( Barone et al . , 2010; Richardson and Pollack , 2005 ) . Therefore , it is of paramount importance to understand how anti-diabetic drugs influence cancer for a better treatment of both cancer and diabetes . Previous studies largely focused on the direct effects of PPARγ on cancer cells . TZDs were shown to promote terminal differentiation , reduce proliferation and trigger lipid accumulation in human breast cancer cells and liposarcoma cells ( Mueller et al . , 1998; Tontonoz et al . , 1997 ) . More recently , the anti-proliferative effect of pioglitazone was reported to involve a metabolic switch in lung and breast cancer cells ( Srivastava et al . , 2014 ) . However , whether and how PPARγ in macrophages modulates cancer progression is unknown . Moreover , previous studies heavily relied on the usage of PPARγ ligands , which may exert PPARγ-independent and/or physiologically irrelevant effects; whereas in vivo genetic dissection of the specific PPARγ functions in each cell type in the cancer milieu is lacking . We previously reported that female mice with PPARγ deletion in the hematopoietic and endothelial cells developed inflammation in their lactating mammary gland . This led to the production of inflammatory milk , which triggered systemic inflammation in the nursing neonates manifested as a transient fur loss ( Wan et al . , 2007b ) . These intriguing observations suggest that PPARγ plays an anti-inflammatory role in macrophage and mammary gland , which may influence breast cancer . We hypothesize that PPARγ in macrophages impedes breast cancer development by inhibiting inflammation . Using a series of genetic and pharmacological , gain- and loss-of-function , in vitro and in vivo approaches , here we uncover macrophage PPARγ as an important suppressor of breast cancer progression and a key mediator of the anti-tumor effects of rosiglitazone that functions by repressing a novel target Gpr132 in macrophages . We generated macrophage PPARγ knockout mice ( mf-g-KO ) by breeding PPARγ flox mice with Tie2Cre or Lysozyme-Cre ( LyzCre ) . Tie2Cre deleted PPARγ in hematopoietic cells and endothelial cells as we previously described ( Wan et al . , 2007a , 2007b ) . LyzCre deleted PPARγ in the myeloid lineage ( Clausen et al . , 1999 ) . These two mf-g-KO models are complementary with different pros and cons: although Lyz-g-KO mice permit a more specific macrophage PPARγ deletion , Tie2-g-KO mice confer a more complete macrophage PPARγ deletion ( 89% ) compared with Lyz-g-KO mice ( 79% ) . Thus , we compared Ppargflox/flox;Cretg/+ KO mice with Ppargflox/flox;Cre+/+littermate controls using both models . To determine the effects of macrophage PPARγ deletion on breast cancer progression , we performed mammary fat pad orthotopic injections of C57BL/6J-compatible mouse breast cancer cells EO771 in female mice , and followed tumor growth by measuring tumor size . Compared to the littermate controls , both Tie2Cre-induced and LyzCre-induced mf-g-KO mice showed enhanced tumor development as indicated by earlier onset and larger tumor volume ( Figure 1A–B ) . These results indicate that the pro-tumor effect observed was largely caused by PPARγ deletion in myeloid cells such as macrophages . Staining for Ki67 and phospho histone H3 ( PH3 ) in the tumor sections showed increased cell proliferation in mf-g-KO mice ( Figure 1C–D ) . To assess a different cancer model , we obtained another C57BL/6-compatible mouse cell line ( Py230 ) that was derived from spontaneous mammary tumors in C57BL/6 MMTV-PyMT female transgenic mice . Py230 cell injection into the mammary fat pad showed that tumor growth was also exacerbated in mf-g-KO mice compared with control mice ( Figure 1—figure supplement 1A ) . These findings suggest that macrophage PPARγ inhibits tumor growth in vivo . 10 . 7554/eLife . 18501 . 003Figure 1 . Macrophage PPARγ deletion enhances mammary tumor growth in vivo . ( A ) Tie2Cre-induced mf-g-KO mice ( n = 26 ) showed enhanced tumor growth compared to control mice ( n = 16 ) as indicated by earlier onset and larger tumor volume . EO771 mouse mammary tumor cells were injected into the mammary fat pad of 6–8 weeks old female mice . ( B ) LyzCre-induced mf-g-KO mice ( n = 6 ) showed augmented tumor growth compared to control mice ( n = 6 ) as indicated by earlier onset and larger tumor volume . ( C–D ) Quantification of cell proliferation markers Ki67 ( C ) and phosphor histone H3 ( PH3 ) ( D ) in tumor sections showed increased cell proliferation in mf-g-KO mice ( n = 4 ) . ( E–G ) RT-QPCR analyses showed an increased expression of pro-inflammatory genes in tumor tissues ( E ) , bone marrow ( BM ) ( F ) and spleen ( G ) from mf-g-KO mice ( n = 3 ) . ( H ) Immunofluorescence staining of tumor sections for macrophage marker CD11b showed an enhanced macrophage recruitment in the tumors from both Tie2Cre- and LyzCre-induced mf-g-KO mice compared with control mice ( n = 4 ) . All tumors were collected 21 days after cancer cell injection . Error bars , SD; p<0 . 05; p<0 . 01; p<0 . 005; *p<0 . 001; n . s . non-significant . DOI: http://dx . doi . org/10 . 7554/eLife . 18501 . 00310 . 7554/eLife . 18501 . 004Figure 1—figure supplement 1 . Additional analyses of tumors . ( A ) Tumors from PPARγ-deficient mice exhibited higher pro-inflammatory M1-like markers ( TNFα and IL1β ) but lower M2-like markers ( Arginase 1 ) . ( B ) Tumor growth from another C57BL/6-compatible mouse breast cancer cell line Py230 was also increased in mf-g-KO mice compared with control mice . Luciferase-labelled Py230 cells were injected into mammary fat pad , and tumor growth was quantified by bioluminescence imaging ( BLI ) ( left ) and tumor volume ( right ) ( n = 3 ) . Error bars , SE . ( C–F ) Immunofluorescence staining of tumor sections . ( C–D ) Macrophage recruitment was increased in both Tie2Cre-PPARγ-KO and LyzCre-PPARγ-KO mice compared to WT control mice . Tumor sections were stained for a macrophage marker F4/80 ( n = 4 ) . ( E–F ) Angiogenesis was increased in Tie2Cre-PPARγ-KO mice but not in LyzCre-PPARγ-KO mice . Tumor sections were stained for an endothelial marker endomucin ( n = 4 ) . All tumors were collected 21 days after cancer cell injection . ( C , E ) Representative images of staining; scale bars , 25 μm . ( D , F ) Quantification of F4/80 intensity ( D ) and number of blood vessels ( F ) . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18501 . 004 We collected tumor tissues , bone marrow cells and spleen cells from tumor-bearing mf-g-KO or control mice , and compared gene expression . The results showed that the expression of pro-inflammatory genes was increased in these PPARγ-deficient cells and tissues , including COX-2 , MMP9 , MCP-1 , TNFα and IL-1β ( Figure 1E–G ) ( Figure 1—figure supplement 1B ) . In contrast , the expression of M2 macrophage markers such as Arginase 1 was decreased ( Figure 1—figure supplement 1B ) . These observations were consistent with previous reports from several laboratories including our own group that PPARγ deficiency promotes inflammatory macrophage activation but attenuates M2 phenotype ( Odegaard et al . , 2007; Ricote et al . , 1998; Straus and Glass , 2007; Wan et al . , 2007b ) . Macrophage infiltration into tumors is a strong indicator for cancer malignancy and poor prognosis ( Komohara et al . , 2014; Ruffell and Coussens , 2015; Zhang et al . , 2012 ) . Immunofluorescence staining using CD11b and F4/80 markers revealed enhanced TAM recruitment in both Tie2-g-KO and Lyz-g-KO mice compared with control mice ( Figure 1H ) ( Figure 1—figure supplement 1C–D ) . This is in line with previous findings that PPARγ-deficient macrophages exhibit increased migration and CCR2 expression ( Babaev et al . , 2005 ) , whereas TZD treatment suppresses macrophage migration and CCR2 expression ( Barlic et al . , 2006; Chen et al . , 2005; Han et al . , 2000; Shah et al . , 2007 ) . Consistent with the reports that PPARγ agonists inhibit angiogenesis ( Goetze et al . , 2002; Keshamouni et al . , 2005; Scoditti et al . , 2010 ) , we found that the number of blood vessels in tumor sections was increased in Tie2-g-KO mice but unaltered in Lyz-g-KO mice ( Figure 1—figure supplement 1E–F ) , further indicating that PPARγ deficiency in macrophage alone is sufficient to augment tumor growth independent of changes in angiogenesis . Together , these findings suggest that macrophage PPARγ deletion changes both the number and property of TAMs to establish a pro-inflammatory tumor environment . To determine if PPARγ-deficient macrophages regulate cancer cell behavior in the absence of other components in the tumor microenvironment such as fibroblasts and extracellular matrix , we performed macrophage and cancer cell co-culture experiments in vitro ( Figure 2A ) . Mouse macrophages were differentiated from the progenitors in bone marrow or spleen and then co-cultured with a luciferase-labelled subline of the MDA-MB-231 human breast cancer cell line ( 1833 cells ) . Specific quantification of tumor cell proliferation was achieved by the luciferase output as only the cancer cells , but not the macrophages , were tagged with a luciferase reporter . The results showed that tumor cell proliferation was significantly augmented by PPARγ-deficient macrophages compared with WT control macrophages ( Figure 2B ) . Consistent with this observation , co-culture with PPARγ-deficient macrophages also led to an increased tumor cell colony formation ( Figure 2C ) . Since mouse macrophages and human cancer cells were from different species , mRNA expression in these two cell types in the co-culture setting could be distinguished by species-specific QPCR primers . We found that co-culture with PPARγ-deficient macrophages resulted in higher expression of proliferation markers and lower expression of apoptosis markers in cancer cells compared with WT control macrophages ( Figure 2D–E ) . 10 . 7554/eLife . 18501 . 005Figure 2 . Macrophage PPARγ deletion exacerbates breast cancer cell proliferation and attenuates the anti-tumor effect of rosiglitazone . ( A ) A diagram of mouse macrophage and human breast cancer cell co-culture . Progenitors in bone marrow or spleen were differentiated into macrophages with M-CSF for nine days before the seeding of luciferase-labelled 1833 human breast cancer cells to the cultures . For rosiglitazone ( Rosi ) pre-treatment , macrophages were treated with 1 μM Rosi or vehicle control for the last 24 hr of macrophage differentiation; after medium was removed and cells were washed , cancer cells were added to the macrophage cultures in fresh medium without Rosi or vehicle . ( B ) Cancer cell proliferation was increased when co-cultured with PPARγ-deficient macrophages derived from bone marrow ( left ) or spleen ( right ) of mf-g-KO mice compared with WT control macrophages ( n = 3 ) . Cancer cell growth was quantified by luciferase signal for 2–6 days . ( C ) PPARγ-deficient macrophages promoted tumor cell colony formation in the co-cultures ( n = 3 ) . Tumor cells were cultured for 11–12 days for the colonies to form . Left , representative images of crystal violet staining . Right , quantification of colony formation . ( D–E ) Co-culture with PPARγ-deficient macrophages resulted in higher expression of proliferation markers ( D ) and lower expression of apoptosis markers ( E ) in breast cancer cells ( n = 3 ) . Human gene expression in cancer cells was quantified by RT-QPCR and human-specific primers . ( F ) PPARγ-deficient macrophages exhibited a higher expression of pro-inflammatory genes ( n = 3 ) . BMMf , bone marrow macrophage; SpMf , spleen macrophage . ( G ) PPARγ-deficient macrophages displayed higher levels of anti-apoptotic genes ( left ) and lower levels of pro-apoptotic genes ( right ) ( n = 3 ) . ( H ) PPARγ-deficient macrophages showed increased proliferation ( n = 3 ) . The number of metabolically active cells was determined by ATP content using the CellTiter-Glo Assay . ( I ) Co-culture with Rosi pre-treated macrophages inhibited breast cancer cell growth compared with vehicle ( Veh ) pre-treated macrophages in a macrophage-PPARγ-dependent manner ( n = 3 ) . ( J ) The ability of Rosi to suppress tumor growth in vivo was significantly attenuated in mf-g-KO mice ( n = 6 ) . Four days after EO771 cell mammary fat pad injection , mf-g-KO mice or control mice were treated with Veh or Rosi ( 10 mg/kg ) every two days before tumor volume measurement . Error bars , SD; p<0 . 05; p<0 . 01; p<0 . 005; **p<0 . 001; n . s . non-significant . DOI: http://dx . doi . org/10 . 7554/eLife . 18501 . 00510 . 7554/eLife . 18501 . 006Figure 2—figure supplement 1 . Additional analyses of co-cultures and rosiglitazone treatment . ( A ) PPARγ-deficient macrophages exhibited increased expression of pro-inflammatory M1-like markers ( TNFα and IL1β ) but decreased expression of M2-like markers ( Arginase 1 and Ym2 ) in macrophage-cancer cell co-cultures ( n = 3 ) . BMMf , bone marrow macrophage; SpMf , spleen macrophage . ( B ) As a positive control , rosiglitazone treatment ( 1 μM , 24 hr ) increased the expression of previously reported PPARγ target gene LXRα in WT macrophages but not g-KO macrophages ( n = 3 ) . Error bars , SD . ( C ) Rosiglitazone treatment reduced the abundance of tumor associated macrophages in WT control mice but not in mf-g-KO ( LyzCre ) mice ( n = 6 ) . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18501 . 006 In accordance with our in vivo observations ( Figure 1 ) , PPARγ-deficient macrophages exhibited elevated expression of pro-inflammatory genes such as COX-2 , MCP-1 and MMP-9 but decreased M2 macrophage markers such as Arginase-1 ( Figure 2F ) ( Figure 2—figure supplement 1A ) . In addition , PPARγ-deficient macrophages displayed higher levels of anti-apoptotic genes and lower levels of pro-apoptotic genes ( Figure 2G ) , indicating an augmented survival . Moreover , PPARγ-deficient macrophages showed increased proliferation , measured by ATP content ( Figure 2H ) or MTT assay ( not shown ) . Our in vitro findings further support our in vivo observations that the increased number and pro-inflammatory property of PPARγ-deficient macrophages are sufficient to promote tumor progression . As a complementary approach to our loss-of-function genetic approach , we next performed gain-of-function pharmacological experiment to assess the effect of rosiglitazone activation of macrophage PPARγ on cancer cells . Mouse macrophages were pre-treated with rosiglitazone or vehicle control; rosiglitazone was removed by the medium change before human cancer cells were seeded for co-culture ( Figure 2A ) . The results showed that cancer cell growth was significantly inhibited when co-cultured with rosiglitazone-treated WT macrophages compared with vehicle-treated WT macrophages ( Figure 2I ) . Importantly , this rosiglitazone effect was macrophage-PPARγ-dependent because tumor cell proliferation was increased equally when co-cultured with PPARγ-deficient macrophages regardless of rosiglitazone or vehicle treatment ( Figure 2I ) . As a positive control , rosiglitazone induction of a previously reported PPARγ target gene LXRα ( Chawla et al . , 2001 ) was observed in WT macrophages but not g-KO macrophages ( Figure 2—figure supplement 1B ) . Together , these findings indicate that activation of macrophage PPARγ by either endogenous or synthetic agonists suppresses tumor growth . To assess the functional significance of macrophage PPARγ in the pharmacological effects of rosiglitazone , we treated mf-g-KO mice and littermate controls with rosiglitazone or vehicle control starting four days after cancer cell injection . The results show that the ability of rosiglitazone to suppress breast cancer growth was significantly attenuated in mf-g-KO mice ( Figure 2J ) . Consistent with this observation , the ability of rosiglitazone to reduce tumor-associated macrophages was also impaired in mf-g-KO mice ( Figure 2—figure supplement 1C ) . This indicates that the macrophage is an essential cell type that is required for the anti-tumor function of rosiglitazone . To understand how PPARγ alters the transcription program in macrophages to control cancer cell proliferation , we next set out to identify the key PPARγ target genes . Our experiments reveal that tumor cell proliferation could be significantly enhanced by co-culture with PPARγ-deficient macrophages but not by the conditioned medium from PPARγ-deficient macrophages ( Figure 3A–B ) , indicating that physical contact between macrophages and cancer cells may be required and thus the key tumor-modulating PPARγ target gene in macrophages likely encodes a membrane protein . By searching published microarray databases comparing PPARγ-deficient vs . control macrophages ( Hevener et al . , 2007 ) and rosiglitazone- vs . vehicle-treated macrophages ( Welch et al . , 2003 ) , we selected several candidate membrane proteins that might be regulated by PPARγ in macrophages . Upon examining their expression in our macrophage cultures , we found that G protein-coupled receptor 132 ( Gpr132 , also known as G2A ) was consistently and significantly upregulated in PPARγ-deficient macrophages compared with WT control macrophages ( see below ) , whereas the expression of 11 other candidates was unaltered ( Figure 3—figure supplement 1 ) . Therefore , we decided to further investigate whether Gpr132 is a functional PPARγ target gene in macrophages . 10 . 7554/eLife . 18501 . 007Figure 3 . PPARγ represses Gpr132 transcription in macrophages . ( A–B ) Physical contact is required for the pro-tumor effects of PPARγ-deficient macrophages . ( A ) A schematic diagram of the co-culture vs . trans-well systems . ( B ) Tumor cell proliferation was enhanced by co-culture with PPARγ-deficient macrophages but not by their conditioned medium delivered via trans-well ( n = 3 ) . ( C ) Gpr132 was predominantly expressed in the hematopoietic cell types and tissues ( n = 3 ) . ( D ) Gpr132 was expressed in macrophages but largely absent in breast cancer cells . mBMMf , mouse bone marrow macrophage; mSpmf , mouse spleen macrophage; mBC , mouse breast cancer cells; hBC , human breast cancer cells . ( E ) Gpr132 mRNA levels were significantly higher in PPARγ-deficient macrophages compared with control macrophages , either in macrophage cultures alone or in macrophages co-cultured with human breast cancer cells ( n = 3 ) . ( F ) Gpr132 protein expression was significantly higher in PPARγ-deficient macrophages ( n = 3 ) . ( G ) PPARγ activation by rosiglitazone reduced Gpr132 mRNA in WT macrophages but not in PPARγ-deficient macrophages ( n = 3 ) . ( H ) Transcriptional output from both 0 . 5 Kb and 1 Kb Gpr132 promoters was reduced by the co-transfection of PPARγ and further diminished by rosiglitazone ( n = 3 ) . HEK293 cells were transfected with PPARγ and its heterodimer partner retinoic X receptor α ( RXRα ) , together with a luciferase reporter driven by 0 . 5 Kb or 1 Kb Gpr132 promoter , and compared with vector-transfected controls . Next day , cells were treated with rosiglitazone or vehicle control for 24 hr before harvest and reporter analyses . ( I ) ChIP assay of PPARγ binding to the endogenous Gpr132 promoter in macrophages . A PPRE region in the Gpr132 promoter was pulled down with anti-PPARγ antibody or an IgG control antibody in RAW264 . 7 mouse macrophages and detected by QPCR ( n = 3 ) . An upstream Gpr132 promoter region served as a negative control . ( J ) ChIP assay of H3K9Ac active transcription histone mark at the Gpr132 transcription start site showed that rosiglitazone represses the transcriptional activity from a Gpr132 promoter in a PPARγ-dependent manner ( n = 3 ) . Control or PPARγ knockdown ( KD ) RAW264 . 7 macrophages were treated with 1 μM rosi or vehicle control for 4 hr before harvest . Error bars , SD; p<0 . 05; p<0 . 01; p<0 . 005; **p<0 . 001; n . s . non-significant . DOI: http://dx . doi . org/10 . 7554/eLife . 18501 . 00710 . 7554/eLife . 18501 . 008Figure 3—figure supplement 1 . Expression of other candidate genes was unaltered in PPARγ-deficient macrophages . Candidate genes that encode membrane proteins and are potentially regulated by macrophage PPARγ were selected from published microarray databases . Their expression in bone marrow macrophages ( BMMf ) or spleen macrophages ( SpMf ) derived from mf-g-KO mice or littermate control mice were quantified by RT-QPCR ( n = 3 ) . LGR5 and CCR4 expression was also examined , but the expression was too low to detect . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18501 . 008 Gpr132 has been previously described as a stress-inducible seven-pass transmembrane receptor that functions at the G2/M checkpoint of the cell cycle ( Weng et al . , 1998 ) , which modulates immune cell function ( Kabarowski , 2009; Radu et al . , 2004; Yang et al . , 2005 ) . We found that Gpr132 was predominantly expressed in the hematopoietic cell types/tissues and highly expressed in macrophages , but largely absent in other tissues or tumor cells ( Figure 3C–D ) , indicating that it may play an important role in macrophage function . Gpr132 expression was significantly higher in PPARγ-deficient macrophages compared with control macrophages , either in macrophage cultures alone or in macrophages co-cultured with cancer cells ( Figure 3E–F ) . In line with this observation , PPARγ activation by rosiglitazone reduced Gpr132 expression in WT macrophages but not PPARγ-deficient macrophages ( Figure 3G ) . These findings suggest that PPARγ represses Gpr132 expression . To determine whether Gpr132 is a direct PPARγ transcriptional target , we investigated if PPARγ can bind to the Gpr132 promoter and regulate its transcription . Gpr132 promoter regions ( 0 . 5 kb and 1 kb ) were cloned into a luciferase reporter vector . Transient transfection and reporter assays reveal that luciferase output from both 0 . 5 Kb and 1 Kb Gpr132 promoter was reduced by the co-transfection of PPARγ and further diminished by rosiglitazone treatment ( Figure 3H ) . These results indicate that PPARγ represses Gpr132 promoter via critical element ( s ) within the 500 base pairs upstream of Gpr132 transcription start site . Indeed , we identified a PPAR response element ( PPRE ) half site in this region ( -188: CATCCGAGCAAGGTCAGAC ) . Chromatin-immunoprecipitation ( ChIP ) assay showed that PPARγ could bind to the endogenous Gpr132 proximal promoter in macrophages but not an upstream negative control region ( Figure 3I ) ; this binding was not significantly altered by rosiglitazone ( not shown ) . Moreover , ChIP assay revealed that H3K9Ac active transcription histone mark at the Gpr132 transcriptional start site was enhanced upon PPARγ knockdown and reduced by rosiglitazone in a PPARγ-dependent manner ( Figure 3J ) . These mechanistic studies indicate that PPARγ directly represses Gpr132 transcription in macrophages upon activation by either endogenous or synthetic ligands . Gpr132 expression in human macrophages derived from human peripheral blood mononuclear cells ( hPBMN ) was also blunted by rosiglitazone ( Figure 4A ) . This indicates that the PPARγ repression of Gpr132 is evolutionally conserved and our findings in mice may translate to human physiology and disease . To explore the significance of Gpr132 in human breast cancer , we analyzed the RNA-Seq and clinical data of breast invasive carcinoma ( BRCA ) from The Cancer Genome Atlas ( TCGA ) database . Because Gpr132 is highly expressed in human macrophages ( Figure 4A ) but absent in human breast cancer cells ( Figure 3D ) , the Gpr132 expression in tumors mainly originates from hematopoietic cells in the microenvironment such as macrophages . Compared with normal breast samples , the majority of breast cancer lesions displayed significantly higher Gpr132 expression ( Figure 4B ) ; in addition , compared with ER-positive breast cancers , the more aggressive ER-negative breast cancers also exhibited higher Gpr132 expression ( Figure 4C ) . Immunohistochemistry staining confirmed that human breast cancer tissues expressed significantly higher Gpr132 compared with normal breast control tissues ( Figure 4D–E ) . Moreover , linear regression analyses showed that higher Gpr132 expression was significantly correlated with higher expression of pro-inflammatory markers including CCL2 ( MCP-1 ) , MMP9 and PTGS2 ( COX-2 ) in breast cancer lesions ( Figure 4F ) . These findings further suggest that macrophage Gpr132 may promote inflammation and tumor progression . 10 . 7554/eLife . 18501 . 009Figure 4 . Gpr132 is repressed by PPARγ in human macrophages and correlates with human breast cancer . ( A ) Human Gpr132 expression in hPBMN-derived macrophages was blunted by rosiglitazone treatment ( n = 3 ) . Macrophages were treated with 1 μM rosiglitazone or vehicle for 4 hr . ( B ) TCGA BRCA data analysis showed that compared with normal breast samples , breast cancer lesions displayed higher Gpr132 expression . Normal Breast ( n = 111 ) ; Infiltrating Ductal Carcinoma ( n = 750 ) ; Infiltrating Lobular Carcinoma ( n = 168 ) ; Medullary Carcinoma ( n = 5 ) ; Mixed Histology ( n = 29 ) ; Mucinous Carcinoma ( n = 14 ) ; Other Histology ( n = 44 ) . Error bars , SE . ( C ) TCGA BRCA data analysis showed that compared with ER+ breast cancers ( n = 746 ) , ER- breast cancers ( n = 221 ) exhibited higher Gpr132 expression . Error bars , SE . ( D–E ) Immunohistochemistry ( IHC ) of human tissue microarrays showed higher Gpr132 expression in breast cancer tissues ( n = 16 ) compared with normal breast tissues ( n = 13 ) . Tissues were stained with anti-Gpr132 ( brown ) and hematoxylin ( blue ) . ( D ) Representative images . Scale bars , 200 μm . ( E ) Quantification of relative IHC scores . Error bars , SE . ( F ) Linear regression analyses of TCGA BRCA data showed that Gpr132 expression was positively correlated with the expression of CCL2 ( MCP-1 ) , MMP9 and PTGS2 ( COX-2 ) in breast cancer lesions ( n = 805 ) . Error bars , SD ( A ) or SE ( B , C , E ) ; p<0 . 05; p<0 . 01; p<0 . 005; **p<0 . 001; n . s . non-significant . DOI: http://dx . doi . org/10 . 7554/eLife . 18501 . 009 We next examined the function of macrophage Gpr132 in regulating cancer cells using our in vitro co-culture system . Gpr132 knockdown in macrophages significantly reduced cancer cell growth ( Figure 5A–C ) . Conversely , Gpr132 over-expression in macrophages increased cancer cell growth ( Figure 5D–F ) . The anti-Gpr132 antibody was validated using Gpr132-KO mice ( Figure 5—figure supplement 1A ) . We then compared macrophages derived from the bone marrow or spleen of Gpr132-KO mice vs . littermate WT control mice . Gene expression analyses reveal that Gpr132-KO macrophages displayed the opposite phenotype from PPARγ-deficient macrophages , with lower pro-inflammatory genes ( Figure 5G ) , higher pro-apoptotic genes and lower anti-apoptotic genes ( Figure 5H ) compared with WT macrophages . In vitro macrophage-tumor cell co-culture experiments showed that Gpr132-KO macrophages exhibited a significantly reduced ability to promote cancer cell colony formation and growth ( Figure 5I–J ) . These results indicate that Gpr132 enhances inflammation and macrophage survival , and the upregulated Gpr132 in PPARγ-deficient macrophages may confer their tumor-promoting effects . 10 . 7554/eLife . 18501 . 010Figure 5 . Macrophage Gpr132 promotes tumor growth and mediates the anti-tumor effect of rosiglitazone . ( A–B ) Gpr132 knockdown decreased Gpr132 mRNA ( A ) and protein ( B ) in macrophages ( n = 3 ) . ( C ) In co-cultures , Gpr132 knockdown in macrophages reduced cancer cell growth ( n = 3 ) . ( D–E ) Gpr132 over-expression increased both mRNA ( D ) and protein ( E ) in macrophages ( n = 3 ) . ( F ) In co-cultures , Gpr132 over-expression in macrophages enhanced cancer cell growth ( n = 3 ) . Cancer cell alone without macrophages ( no mf ) served as a negative control . ( G ) Gpr132-KO macrophages exhibited lower expression of pro-inflammatory genes compared with WT controls ( n = 3 ) . ( H ) Gpr132-KO macrophages displayed higher levels of pro-apoptotic genes and lower levels of anti-apoptotic genes ( n = 3 ) . ( I–J ) In vitro co-cultures showed that Gpr132 deletion in macrophages significantly reduced the ability of macrophages to promote cancer cell colony formation ( I ) and proliferation ( J ) ( n = 3 ) . ( K ) In vivo mammary fat pad tumor grafts showed that tumor growth was significantly diminished in Gpr132-KO mice compared with WT or Gpr132-Het mice ( n = 6 ) . ( L ) In in vitro co-cultures , Rosi pre-treated WT macrophages but not Rosi pre-treated Gpr132-KO macrophages were able to inhibit cancer cell growth ( n = 3 ) . ( M ) The ability of Rosi to suppress tumor growth in vivo was abolished in Gpr132-KO mice ( n = 6 ) . Four days after EO771 cell mammary fat pad injection , Gpr132-KO or WT mice were treated with Veh or Rosi ( 10 mg/kg ) every two days . ( N ) The ability of macrophage PPARγ deletion to exacerbate tumor growth in vivo was abolished in Gpr132-KO mice ( n = 4 ) . DKO , mf-g/Gpr132 double KO . ( O–P ) Pharmacological Gpr132 inhibition impeded mammary tumor growth . Female mice ( six-week-old ) were treated with si-Gpr132 ( n = 8 ) or si-Ctrl ( n = 6 ) for 18 days via intravenous injection at 10 μg/mouse twice/week , three days before and 15 days after EO771 cell mammary fat pad injection . ( O ) Tumor volume was significantly decreased by si-Gpr132 treatment . ( P ) Gpr132 expression in tumors was effectively depleted . Error bars , SD; p<0 . 05; p<0 . 01; p<0 . 005; **p<0 . 001; n . s . non-significant . ( Q ) A simplified working model for how macrophage PPARγ inhibits inflammation and tumor growth by repressing the transcription of macrophage Gpr132 , a novel pro-inflammatory and pro-tumor membrane receptor . Upon sensing and activation by tumor signals , macrophage Gpr132 may modulate macrophage intracellular signaling and downstream targets , which in turn promotes cancer cell proliferation ( indicated by the dashed line ) . Macrophage PPARγ deficiency increases Gpr132 level to afford better tumor sensing by macrophages , thereby promoting tumor growth . Pharmacological Gpr132 inhibition via either PPARγ agonist or Gpr132 blockade attenuates breast cancer progression . Moreover , both macrophage PPARγ and Gpr132 are key mediators of the anti-tumor effects of the clinically used TZD drug rosiglitazone . DOI: http://dx . doi . org/10 . 7554/eLife . 18501 . 01010 . 7554/eLife . 18501 . 011Figure 5—figure supplement 1 . Additional analysis of Gpr132 . ( A ) Western blot of Gpr132 protein in spleen and bone marrow ( BM ) from WT and Gpr132-KO mice . ( B ) Gpr132-null macrophages were refractory to the anti-tumor effects of rosiglitazone pre-treatment . Rosiglitazone pre-treatment of WT macrophages , but not Gpr132-KO macrophages , inhibited the growth of luciferase-labeled 4T1 . 2 mouse mammary tumor cells in the co-cultures ( n = 3 ) . Error bars , SD . ( C ) Western blot of tumors from mice treated with si-Gpr132 or si-Ctrl showed that si-Gpr132 conferred ~62% knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 18501 . 011 To examine the effects of Gpr132 deletion in the tumor environment on cancer growth in vivo , we injected EO771 mouse breast cancer cells into the mammary fat pad of Gpr132-KO mice and WT littermate controls . In this system , Gpr132 was deleted in macrophages as well as other Gpr132-expressing tissues such as bone marrow , spleen and thymus , but not in the injected cancer cells which had essentially no Gpr132 expression ( Figure 3C–D ) . Previous study show that Gpr132-KO mice display a normal pattern of T and B lineage differentiation , appearing healthy and indistinguishable from WT littermates throughout young adulthood , but develop progressive secondary lymphoid organ enlargement associated with abnormal expansion of both T and B lymphocytes that become pathological when older than one year of age ( Le et al . , 2001 ) . Therefore , our experiments were initiated in young mice and terminated before Gpr132-KO mice aged to prevent any potential effects of lymphoid defects on cancer growth . Compared with WT and Gpr132 heterozygous ( Het ) controls , Gpr132-KO mice exhibited significantly diminished tumor growth ( Figure 5K ) . Compared with WT controls , Gpr132-Het mice also showed attenuated tumor growth at a later stage ( Figure 5K ) , indicating that Gpr132 regulation is dosage-sensitive . Together , our in vitro and in vivo results indicate that macrophage Gpr132 promotes tumor growth , suggesting that Gpr132 inhibition may impede cancer progression . To further examine whether Gpr132 is a functional PPARγ target in macrophage that is required for PPARγ cancer regulation , we conducted pharmacological and genetic experiments . As a pharmacological gain-of-function strategy , we treated the macrophage-cancer cell co-cultures or tumor-grafted mice with rosiglitazone or vehicle control . Pre-treating Gpr132-KO macrophages with rosiglitazone before cancer cell seeding no longer showed any inhibition of cancer cell proliferation in the co-cultures ( Figure 5L ) ( Figure 5—figure supplement 1B ) . Consistent with this in vitro observation , the anti-tumor effect of rosiglitazone in vivo was also abolished in Gpr132-KO mice ( Figure 5M ) . These pharmacological findings support that the PPARγ repression of Gpr132 in macrophages is a significant contributor to the anti-tumor effects of rosiglitazone . As a genetic loss-of-function strategy , we bred Gpr132-KO mice with mf-g-KO mice to generate mf-g/Gpr132 double KO ( DKO ) mice . Mammary fat pad tumor graft experiments demonstrated that Gpr132 deletion in the DKO mice impaired the ability of macrophage PPARγ deficiency to exacerbate tumor growth because DKO mice showed similar tumor volume as Gpr132-KO mice ( Figure 5N ) . This genetic rescue further supports that Gpr132 is an essential mediator of macrophage PPARγ regulation of breast cancer progression . To further explore Gpr132 as a potential cancer therapeutic target , we next examined whether acute pharmacological inhibition of Gpr132 could attenuate breast cancer progression . Because macrophage precursors reside in hematopoietic tissues such as blood , bone marrow and spleen that can be efficiently targeted by siRNA ( Larson et al . , 2007 ) , we chose to employ siRNA-mediated Gpr132 knockdown . We treated WT female mice with si-Gpr132 or si-Ctrl for 18 days via intravenous injection at 10 μg/mouse twice/week , three days before and 15 days after cancer cell graft . The results showed that si-Gpr132 significantly reduced tumor volume compared with si-Ctrl ( Figure 5O ) , as the result of depleted Gpr132 expression ( Figure 5P ) ( Figure 5—figure supplement 1C ) . Body weight was unaltered by si-Gpr132 ( not shown ) , indicating a lack of overt toxicity by Gpr132 inhibition . These results support the Gpr132 inhibition as a novel anti-cancer strategy . Given the pleotropic and important roles of PPARγ in physiology and disease , as well as the wide-spread usage of TZD drugs for the treatment of insulin resistance and type II diabetes , it is of paramount importance to elucidate the mechanisms for how PPARγ and TZDs affect cancer . Here we have uncovered a crucial yet previously unrecognized role of macrophage PPARγ in suppressing cancer progression and mediating the anti-tumor effects of rosiglitazone ( Figure 5Q ) . Mechanistically , PPARγ activation in macrophages tunes down inflammatory programs by repressing the transcription of a novel target gene Gpr132 , which is a pro-inflammatory membrane receptor ( Figure 5Q ) . Consequently , tumor growth is inhibited when the macrophage Gpr132 level is low by either Gpr132 deletion/inhibition or PPARγ activation via rosiglitazone; whereas tumor growth is exacerbated when the macrophage Gpr132 level is high as the result of macrophage PPARγ deficiency . Importantly , Gpr132 deletion abolishes the cancer regulation by macrophage PPARγ or rosiglitazone , indicating that Gpr132 is an essential mediator of PPARγ functions in macrophages and tumor progression . These findings reveal PPARγ and Gpr132 as fundamental key players in TAM , providing new mechanisms how macrophages interact with tumor cells to promote cancer malignancy . Although our co-culture experiments suggest that tumor cell-macrophage contact is required for macrophage PPARγ regulations , other possible mechanisms may exist . For example , ( 1 ) tumor cell-macrophage close proximity ( rather than direct contact ) may be required so that an increased macrophage Gpr132 expression can better sense a gradient of the signal sent from tumor cells - this is supported by previous studies indicating Gpr132 as a pH and lipid sensor that may regulate immune cell trafficking ( Justus et al . , 2013; Vangaveti et al . , 2010 ) ; ( 2 ) upon activation by tumor signals , macrophage Gpr132 may modulate macrophage intracellular signaling and downstream targets , which in turn promotes cancer cell proliferation ( Figure 5Q ) . Our findings have opened an exciting new path for future investigations to delineate how macrophage Gpr132 senses tumor signals and then exacerbates tumor malignancy . Both inflammatory macrophages ( M1-like ) and M2-like macrophages have been shown to be potentially pro-tumor . Recent literature strongly support that inflammation exacerbates tumor progression , as highlighted in several reviews ( Coussens and Werb , 2002; Noy and Pollard , 2014; Qian and Pollard , 2010 ) . Thus , in terms of how macrophage PPARγ impacts tumor progression , only the phenotype speaks of the truth . Our in vivo and in vitro findings all support an anti-inflammatory and anti-tumor role of macrophage PPARγ activation , and a pro-inflammatory and pro-tumor effect of macrophage PPARγ deletion . Among the many downstream events that mediate the pro-tumor effects of inflammatory macrophages , it is possible that macrophage PPARγ not only regulates cancer cell behavior but also modulates tumor-infiltrating lymphocytes ( TILs ) to alter immune surveillance ( Engblom et al . , 2016 ) . PPARγ may act on several cell types to exert anti-tumor effects including previously described cancer cells ( Mueller et al . , 1998; Srivastava et al . , 2014; Tontonoz et al . , 1997 ) and now reported immune cells . Our findings reveal macrophage as a key cell type , if not the only cell type , that is essential for the tumor suppressive effects of TZDs . Interestingly , although Tie2-g-KO mice harbored more complete PPARγ deletion than Lyz-g-KO mice and both mouse models showed enhanced tumor growth and tumor macrophage recruitment , the phenotype was more pronounced in Lyz-g-KO . It is possible that the PPARγ deletion in other hematopoietic cells outside of the myeloid lineage exerted opposite albeit minor effect on tumor growth in the Tie2-g-KO mice , which was absent in the Lyz-g-KO mice . Similarly , PPARγ may affect many genes in macrophage , and targets other than Gpr132 may also contribute to the anti-tumor effects of PPARγ . Nonetheless , our genetic rescue and pharmacological experiments indicate that Gpr132 is a very important part of the puzzle and an essential mediator of macrophage PPARγ regulation , because in the absence of Gpr132 , the tumor-modulating function of macrophage PPARγ or rosiglitazone is abolished . This uncovers Gpr132 as not only a novel key PPARγ target but also a new cancer therapeutic target . Our luciferase reporter assays indicate that PPARγ directly suppresses the transcriptional activity from Gpr132 promoter , which is further supported by ChIP assays showing PPARγ binding to Gpr132 promoter , leading to a decreased level of H3K9Ac active transcription histone mark at the Gpr132 transcriptional start site . Nonetheless , additional mechanisms such as changes in mRNA stability may also contribute to PPARγ down-regulation of Gpr132 mRNA level . Cancer cells form an intimate relationship with TAMs to proliferate and survive . As such , targeting the infiltrating macrophages to alter their number and properties can lead to a significant inhibition of cancer malignancy . Our data suggest that this can be achieved by either PPARγ activation or Gpr132 inhibition in the macrophage . By elucidating the mechanisms that macrophages use to promote cancer and inflammation , effective diagnostic tools as well as innovative anti-tumor and anti-inflammatory therapeutics can be designed . For example , macrophage levels of PPARγ and Gpr132 may predict not only tumor aggressiveness but also the pharmacological responses to rosiglitazone or Gpr132 inhibitors . Our findings may explain why rosiglitazone exerts anti-tumor effects in certain cancers but not others – cancers with abundant PPARγ-positive macrophages may be sensitive whereas cancers with limited macrophages or PPARγ-negative macrophages may be resistant . Remarkably , the positive association of Gpr132 with inflammation and breast cancer in human ( Figure 4B–F ) , the repression of Gpr132 expression by rosiglitazone in human macrophage ( Figure 4A ) and the anti-tumor effects of pharmacological Gpr132 inhibition ( Figure 5O–P ) highlight the exciting potential of Gpr132 blockade as a new therapeutic . Moreover , the observation that Gpr132 expression is significantly increased in the majority of human breast cancers ( Figure 4B ) suggests that Gpr132 may serve as a useful marker for breast cancer prognosis , similar to the 21 genes typically examined in the commercially available Oncotype Dx panel . In summary , the significance of our findings resides in the following aspects: ( 1 ) it reveals macrophage as an important cell type that contributes to PPARγ suppression of cancer and the anti-tumor effects of rosiglitazone; ( 2 ) it identifies Gpr132 as a novel PPARγ direct target gene in macrophages that mediates PPARγ functions; ( 3 ) it uncovers Gpr132 as a pro-inflammatory and pro-tumor factor in macrophages , and thus a novel therapeutic target . Ultimately , these new knowledge will enhance our understanding of macrophage regulation , cancer microenvironment as well as PPARγ and Gpr132 biology , which may translate to a better intervention of diseases such as cancer , diabetes and inflammatory disorders . PPARγ flox mice on a C57BL/6J background ( RRID:IMSR_JAX:004584 ) were described ( He et al . , 2003 ) . Gpr132 knockout mice on a C57BL/6J background ( RRID:IMSR_JAX:008576 ) ( Le et al . , 2001 ) were obtained from the Jackson Laboratory . Mice were fed standard chow ad libitum and kept on a 12-h light , 12-h dark cycle . PPARγ flox mice were bred with Tie2-Cre ( RRID:IMSR_JAX:008863 ) ( Kisanuki et al . , 2001 ) or Lysozyme-Cre ( RRID:IMSR_JAX:004781 ) ( Clausen et al . , 1999 ) transgenic mice to generate mf-g-KO mice . Tie2-g-KO was bred with Gpr132-KO to obtain mf-g/Gpr132 double KO mice . Representative results for Tie2-g-KO are shown unless specified as Lyz-g-KO . All experiments were conducted using littermates . Sample size estimate was based on power analyses performed using SAS 9 . 3 TS X64_7PRO platform . All protocols for mouse experiments were approved under Animal Protocol Number 2008–0324 by the Institutional Animal Care and Use Committee of UTSW . For bone marrow- and spleen-derived macrophage , mouse bone marrow or splenocyte were collected with serum-free DMEM . After passing through a 40 μm cell strainer , the cells were cultured in macrophage differentiation medium ( DMEM + 10% FBS + 20 ng/ml M-CSF ) for six days . Gpr132 overexpression was performed with lentiviral transduction . The RAW264 . 7 mouse macrophage cell line ( RRID:CVCL_0493 ) was from ATCC . The EO771 cell line originally derived from a spontaneous mammary tumor in a C57BL/6 mouse ( RRID: CVCL_GR23 ) ( Casey et al . , 1951 ) was from CH3 BioSystems ( Amherst , NY ) . The luciferase-labeled MDA-MB-231 human breast cancer sub-line ( MDA-BoM-1833; RRID:CVCL_DP48 ) ( Kang et al . , 2003 ) was provided by Joan Massagué ( Memorial Sloan-Kettering Cancer Center ) . The luciferase-labeled 4T1 . 2 mouse mammary tumor subline ( RRID:CVCL_GR32 ) ( Lelekakis et al . , 1999 ) was provided by Robin Anderson ( Peter MacCallum Cancer Centre ) and Yibing Kang ( Princeton University ) . Cell lines were authenticated by STR profiling and verified negative for mycoplasma . For macrophage and cancer cell co-cultures , mouse bone marrow and spleen cells were plated in 96-well plate and differentiated into macrophages with 20 ng/ml M-CSF for nine days . Luciferase-labeled 1833 cells or 4T1 . 2 cells were then added to the culture dish . At the end point of experiment , cell lysates were collected for luciferase assay to assess cancer cell growth . For the pre-treatment , macrophages were cultured with 1 μM rosiglitazone ( Cayman Chemical , Ann Arbor , MI ) for the last 24 hr; the medium was removed and the macrophages were washed before cancer cell seeding . EO771 cells ( 2 . 5 × 105 or 5 × 105 ) were injected into the mammary fat pad of 6–8 weeks old female mice . EO771 cells were prepared with a 1:1 ratio in the blank RPMI-1640 medium and matrigel ( BD Biosciences , San Jose , CA ) . Every 2–3 days , tumor length and width were measured with a caliper and tumor volume was calculated using the formula V = ( L × W × W ) / 2 , where V is tumor volume , L is tumor length , and W is tumor width . The Py230 cell line was derived from spontaneous mammary tumors in C57BL/6 MMTV-PyMT female transgenic mice ( RRID:CVCL_AQ08 ) ( Biswas et al . , 2014 ) . Tumor tissues were isolated from tumor-bearing mice three weeks after cancer cell injection . Tumors were frozen in OCT compound ( Tissue-Tek ) , cryo-sectioned , and fixed with acetone before staining with antibodies . The tumor sections were blocked with 2% BSA , and then incubated with FITC anti-CD11b antibody ( BD Pharmingen; RRID:AB_394774; 1:50 dilution ) or FITC anti-F4/80 antibody ( AbD Serotec , Raleigh , NC; RRID:AB_1102553; 1:50 dilution ) . For antibodies without FITC conjugate , the tumor sections were incubated with rat monoclonal anti-endomucin ( Santa Cruz Biotechnologies , Dallas , TX; RRID:AB_2100037; 1:50 dilution ) , rabbit monoclonal anti-Ki67 ( Cell Signaling , Danvers , MA; RRID:AB_2620142; 1:400 dilution ) , or rabbit polyclonal anti-Phospho-Histone H3 ( Ser10 ) ( Cell Signaling; RRID:AB_331534; 1:200 dilution ) . After washing with PBS , the sections were incubated with goat-anti-rat IgG-FITC antibody ( RRID:AB_631753 ) or goat-anti-rabbit IgG-FITC antibody ( RRID:AB_631744 ) ( Santa Cruz Biotechnologies; 1:100 dilutions ) for detection . After washing with PBS , cover slips were mounted with the Vectashield medium containing DAPI ( Vector Laboratories , Burlingame , CA ) . Tissue samples were snap frozen in liquid nitrogen and stored at −80°C . RNA was extracted using Trizol ( Invitrogen , Carlsbad , CA ) according to the manufacturer’s protocol . RNA was first treated with RNase-free DNase I using the DNA-free kit ( Ambion , Austin , TX ) to remove all genomic DNA , and then reverse-transcribed into cDNA using an ABI High Capacity cDNA RT Kit ( Invitrogen ) . The cDNA was analyzed using real-time quantitative PCR ( SYBR Green , Invitrogen ) with an Applied Biosystems 7700 Sequence Detection System . Each reaction was performed in triplicate in a 384-well format . The expression of mouse gene was normalized by mouse L19 . The expression of the human gene was normalized with human GAPDH . Anti-Gpr132 antibody ( Sigma , St . Louis , MO; RRID:AB_10745673 ) was validated using Gpr132-KO cells and used for western blot detection of Gpr132 protein . Tissue microarrays were purchased from US Biomax , Inc . ( Rockville , MD ) , which contain human normal breast tissues and breast cancer tissues . The immunohistochemistry ( IHC ) staining was performed as previously described ( Su et al . , 2014; Zhou et al . , 2014 ) . Briefly , after dewaxing with xylence and dehydration with gradient ethanol , the tissue microarrays were incubated with antigen retrieval buffer ( BD Biosciences ) for 1 hr at 95°C , followed by treatment with 3% hydrogen peroxide ( Sigma ) for 10 min . Specimens were blocked with 5% defatted milk for 1 hr at room temperature , and incubated with anti-human-Gpr132 antibodies ( Sigma; RRID:AB_10745673; 1:100 ) overnight at 4°C and then with HRP-conjugated secondary antibodies ( 1:200 ) for 30 min at room temperature . Immunostaining was performed using a diaminobenzidine ( DAB ) kit ( Thermo scientific , Waltham , MA ) . The expression levels of Gpr132 were scored in a blind fashion semi-quantitatively according to the staining intensity and distribution using the immunoreactive score as described previously ( Su et al . , 2014; Zhou et al . , 2014 ) . Briefly , the IHC score = staining intensity ( negative = 0; weak = 1; moderate = 2; and strong = 3 ) × percentage of positive cells ( 0% = 0; 0–25% = 1; 25–50% = 3; and 75–100% = 4 ) . RNA-Seq and clinical data of breast invasive carcinoma ( BRCA ) were downloaded from The Cancer Genome Atlas ( TCGA ) data portal ( Cancer Genome Atlas Network , 2012 ) and tested for associations . Gene expression for GPR132 , CCL2 ( MCP-1 ) , MMP9 and PTGS2 ( COX-2 ) were analyzed by linear regression . All statistical analyses were performed with Student's t-Test and represented as mean ± standard deviation ( SD ) unless noted otherwise . For in vivo experiments with ≥3 groups , statistical analyses were performed with ANOVA followed by the post hoc Tukey pairwise comparisons . The p values were designated as p<0 . 05; p<0 . 01; p<0 . 005; ***p<0 . 001; n . s . non-significant ( p>0 . 05 ) .	The immune system can both contribute to cancer progression and restrain the growth of tumors . Some immune cells – called macrophages – create an inflammatory environment around a tumor , which can support the spread of the cancer cells . Independent observations and experiments have shown that a protein called PPARγ can suppress the development and growth of tumors . Drugs called thiazolidinediones ( or TZDs for short ) , which are normally used to treat type 2 diabetes , activate PPARγ and therefore have anti-tumor effects . However , it is not fully understood how PPARγ and TZDs suppress tumor development . Cheng et al . hypothesized that the PPARγ protein and TZDs can inhibit the activity of the inflammatory macrophages that help tumors to develop . To test this , mice were genetically engineered so that their macrophages could not produce the PPARγ protein . These engineered mice were more likely to develop breast cancer than normal . Furthermore , the breast tumors in the modified mice did not shrink when they were treated with TZDs , whereas the tumors of normal mice did . Cheng et al . also found that PPARγ inhibits the ability of macrophages to produce a protein called Gpr132 , which itself contributes to inflammation and allows breast cancer cells to grow . Mice that were unable to produce Grp132 displayed less inflammation , and cancer growth was blocked . Drugs that inhibited the activity of Grp132 in normal mice also reduced the ability of breast tumors to spread . Future experiments will need to examine exactly how the Gpr132 proteins produced by macrophages communicate with the cancer cells . Furthermore , developing new drugs that can inhibit Gpr132 could ultimately lead to more effective treatments for cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2016	Macrophage PPARγ inhibits Gpr132 to mediate the anti-tumor effects of rosiglitazone
While axon fasciculation plays a key role in the development of neural networks , very little is known about its dynamics and the underlying biophysical mechanisms . In a model system composed of neurons grown ex vivo from explants of embryonic mouse olfactory epithelia , we observed that axons dynamically interact with each other through their shafts , leading to zippering and unzippering behavior that regulates their fasciculation . Taking advantage of this new preparation suitable for studying such interactions , we carried out a detailed biophysical analysis of zippering , occurring either spontaneously or induced by micromanipulations and pharmacological treatments . We show that zippering arises from the competition of axon-axon adhesion and mechanical tension in the axons , and provide the first quantification of the force of axon-axon adhesion . Furthermore , we introduce a biophysical model of the zippering dynamics , and we quantitatively relate the individual zipper properties to global characteristics of the developing axon network . Our study uncovers a new role of mechanical tension in neural development: the regulation of axon fasciculation . In the developing nervous system , complex neural networks are built through the growth of axons from their neuronal cell body of origin toward their target ( s ) , according to specific pathfinding patterns ( Chédotal and Richards , 2010 ) . These patterns are genetically controlled by molecular cues mediating interactions between axons and their environment , including other axons , cells and the extracellular matrix ( Kolodkin and Tessier-Lavigne , 2011 ) . The establishment of axonal projections from a given neural tissue to its final target is in many cases a multistep process , in which individual axons are sequentially guided from one area to another by a series of cues inducing specific decisions at the level of axonal growth cones ( Mann et al . , 2004 ) . In many organisms and especially in vertebrates , given the generally high number of neurons generated in the various areas , and the need for their massive projections from one area to another , a fundamental principle governing axon pathfinding resides in the control of the fasciculation and defasciculation of their axons . This control is believed to be exerted at the level of axonal growth cones , which may choose to grow along other axons ( fasciculation ) or to detach from other axons ( defasciculation ) ( reviewed in [Honig et al . , 1998] ) . On the one hand , fasciculation ensures robust coordinated growth of a number of axons along the paths initially established by pioneering axons ( Raper and Mason , 2010 ) . On the other hand , axon defasciculation , often associated to branching , is in many cases required for individual axons to reach with precision their specific target ( s ) , which can be distributed in large areas ( Tang et al . , 1994; Schneider and Granato , 2003 ) . For example , motor axons emerging from spinal somatic motoneurons fasciculate in tight bundles , they migrate in fascicles within spinal nerves , and they thereafter defasciculate to allow each individual axon subpopulation to innervate a specific muscle cell group ( Bonanomi and Pfaff , 2010; Huettl et al . , 2012 ) . While the interaction of growth cones with other axons has been the focus of numerous studies ( Honig et al . , 1998; Tang et al . , 1994; Lin and Forscher , 1993; Van Vactor , 1998; Kalil , 1996 ) , other aspects of the process of axon fasciculation have received much less attention . In particular , while it seems obvious that tight fasciculation of axons is aided by adhesion between their shafts , very little is known about the dynamics of shaft-shaft interactions , the underlying biophysical mechanisms , and their potential role in the regulation of axon fasciculation . The aim of the present paper was to address these issues , by analyzing axon-axon interactions and the resulting fasciculation/defasciculation processes in a convenient setting . We chose the mouse olfactory epithelium as a model system . This system has the advantage of comprising a single population of neurons , the olfactory sensory neurons ( OSNs ) . During their normal development from the olfactory epithelium ( OE ) toward their target in the olfactory bulb ( OB ) , OSN axons undergo a massive fasciculation step to form branches of the olfactory nerve , followed by their defasciculation and rearrangement in the OB to project to their specific target cells distributed throughout the OB glomerular layer ( Key et al . , 2002; Nedelec et al . , 2005; Strotmann and Breer , 2006; Mombaerts , 2006; Mori and Sakano , 2011 ) . Since it is currently technically challenging to image mouse OSN axon fasciculation and defasciculation dynamics in vivo , and impossible to manipulate in vivo the individual axons in order to assess their biophysical properties , we chose to perform our study on embryonic OE cultured explants , grown on a permissive planar substrate . Using time lapse imaging , we recorded OSN axons as they grow from the explants , and characterized their dynamic interactions . Surprisingly , we observed that OSN axons interact extensively with each other through their axon shafts , leading to zippering and unzippering behaviors that trigger their fasciculation or defasciculation , respectively . In the present paper , we characterize the dynamics of these axon-axon shaft interactions , assess quantitatively the biophysical parameters of these processes , and develop a biophysical model of this dynamics . Micromanipulations of individual zippers , as well as pharmacologically induced perturbations of the fasciculated network , are used to demonstrate unzippering by forces of functionally relevant magnitude . Our analysis supports a framework in which axonal mechanical tension regulates fasciculation through the control of axon shaft zippering . OSN axons grow in cultures as unbranched axons . In our experimental conditions ( see Materials and methods ) , the growth of these axons from OE explants was characterized by a sequence of three main stages: ( 1 ) advance of the growth cones and initiation of an axon network ( first 24–48 hr ) , ( 2 ) maintenance of the growth cones at distance from the explants , but with little or no further growth ( 48–72 hr ) , and ( 3 ) retraction of the growth cones and collapse of the network ( 3–5 d ) . We analyzed in detail the intermediate stage , during which axon shafts interact and form bundles . During this stage , the initial axon network , composed of individual axons or bundles of few axons , progressively evolves into a less dense network with thicker bundles , indicating that individual axons and/or small bundles fasciculate together to form larger bundles ( Figure 1 and Video 1 ) . To characterize this process quantitatively , we selected a typical area of the network and manually segmented ( see Materials and methods ) the images into vertices ( junction points and crossings of axon segments ) and edges ( lines connecting the vertices ) . Figure 1D–G shows the results of such a segmentation over a 178-min time interval , the red dashed lines representing the segmented edges , and star symbols representing the vertices of the network . Based on this image analysis , we determined the total length of the network and the total number of vertices ( junction points ) , and found that both these quantities decreased approximately linearly with time in the interval examined ( Figure 1H ) . This trend was observed in five out of six quantitatively analyzed experiments from different cultures , with an average reduction of ( 20 ± 16 ) % in length of the network over the duration of the recordings ( 178 to 295 min ) . Topologically , such dynamics is reminiscent of the well-known coarsening of two-dimensional foams ( Glazier and Weaire , 1992; Weaire and Hutzler , 2001 ) ; the underlying structures and forces , however , are substantially different ( see Discussion ) . 10 . 7554/eLife . 19907 . 003Figure 1 . Progressive formation of fascicles in the evolving axon network . ( A–C ) Evolution of the axonal network growing from an explant during 400-min time lapse recording , after 2 days of incubation; the red dashed outline delineates a travelling ensheating cell . Progressive coarsening of the network and decrease of total length and density can be seen . ( D–G ) Red dashed lines outline the edges of the network , while yellow stars indicate junctions between axons or axon bundles , and green stars indicate crossings . ( H ) Quantification of total length and number of vertices of the network area depicted in panels ( D–G ) , as a function of time ( based on seven manually segmented video frames ) . Segmentation coordinates for panels D-G and data from panel H data are available in Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 00310 . 7554/eLife . 19907 . 004Figure 1—source data 1 . Segmentation coordinates ( D–G ) , plot data ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 00410 . 7554/eLife . 19907 . 005Figure 1—figure supplement 1 . An example of spontaneous defasciculation correlated with explant contraction . ( A–C ) Full field view of network evolution , from an experiment with no added drugs . The edge of the explant is marked by the red dashed line . After t=65 min , the explant edge starts to move out of the field and pulls on the outgrown axon network . This causes defasciculation and an increase of network length in the area marked by the red square , labeled ( Di–Diii ) . The panels ( Di–Diii ) show magnified views of the marked area . An increase in network density is apparent in the panel Diii . The time-lapse recording spanning t=1 min through t=135 min is provided as Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 00510 . 7554/eLife . 19907 . 006Figure 1—figure supplement 1—source data 1 . Development of axon network over 135 min , with decoarsening visible in the lower right quadrant after t=65 min , video corresponding to Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 00610 . 7554/eLife . 19907 . 007Video 1 . Development and coarsening of axon network over 12 h , corresponding to Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 007 To understand the processes that lead to this coarsening of the axon network , we examined its dynamics on finer time scales . On time intervals of the order of minutes , we observed elementary zippering processes , as shown in Figure 2A , B . In an advancing zippering process , two axons or axon fascicles progressively adhere to each other in a longer segment of contact and form a larger fascicle . Receding zippers leading to defasciculation of axons were also observed . The zipper vertex at the meeting point of the axons moves with a velocity of the order of 1μmmin until it reaches a position of equilibrium ( Figure 2A ) . In the example of Figure 2B , two adjacent advancing zippers lead to a clear decrease in the total length of the network . 10 . 7554/eLife . 19907 . 008Figure 2 . High-magnification images of individual axon zippers and their evolution in time . Zipper vertices are marked by arrowheads . ( A ) advancing zipper , ( B ) two associated advancing zippers . The total length of the network segments in B decreases during the zippering process . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 008 Numerous and frequent zippers were observed throughout the network , as demonstrated in Figure 3 , showing a selected time interval from the network dynamics of Figure 1 . Blue arrows in Figure 3A–C point to vertices that will zipper in the following frames . Red dashed lines with arrows show the resulting zippered segment . The zippering processes in the upper left corner ( the area marked by rectangle in Figure 3E and enlarged in Figure 3G–J ) lead to a reduction in the number of vertices , from 3 to 1 ( star symbols in Figure 3G and J ) . 10 . 7554/eLife . 19907 . 009Figure 3 . Zippering events drive the progressive formation of fascicles . ( A–F ) Six frames extracted from the time sequence shown in Figure 1D–G . The blue arrows indicate vertices that will start to zippper in the following frame , the red dashed arrows illustrate the direction and the increase in length of the advancing zipper . If two arrowheads are present , there are two vertices extending a single segment . Frames G–J are enlargements of the inset in panel E in the period between the frames E and F , illustrating three vertices ( marked by stars in frames G to I ) merging into a single vertex ( in frame J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 009 To assess the axonal structure of the zippers , we performed scanning electron microscopic analyses of our cultures ( Figure 4A ) . We observed at high magnification that numerous network vertices displayed structures as shown in Figure 4B , C , in which individual or small bundles of axons are adherent to each other along a defined segment , while remaining parallel to each other . A zipper with such structure is free to increase or decrease the length of the zippered segment , depending on the balance of the forces acting at the zipper vertex; we call such zippers ‘simple zippers’ . 10 . 7554/eLife . 19907 . 010Figure 4 . Fine morphological characterization of zippers with scanning electron microscopy . Panel ( A ) illustrates a large area of the culture observed at low magnification . ( B–C ) illustrate a laminar vertex structure formed between small axon bundles ( B ) or between individual axons ( C ) . ( D ) illustrates crossing of axons . ( E ) and ( F ) illustrate more complex , entangled vertices . Such configurations are unlikely to easily unzipper . ( G ) shows thin lateral protrusions , often seen along axon shafts . These protrusions can attach to nearby axons and pull on them . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 01010 . 7554/eLife . 19907 . 011Figure 4—figure supplement 1 . Quantification of abundance of axon crossings , simple zippers and entangled zippers . The SEM image was examined to assign simple zippers ( mobile; marked in blue ) , entangled zippers ( unable to recede; marked in red ) and crossings of the axons ( marked in green ) . The corresponding counts are indicated in the legend . The same type of analysis of a second SEM image provided the following results: 65 simple zippers , 37 entangled zippers and 24 crossings . Coordinates of selected points are available in Figure 4—figure supplement 1—source data 1 . . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 01110 . 7554/eLife . 19907 . 012Figure 4—figure supplement 1—source data 1 . Coordinates of marked points . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 012 More rarely , we observed zippers with entangled structure ( Figure 4E , F ) . Such structure may prevent the zipper from unzippering past the entanglement point; further zippering , however , remains possible . In some instances , axons would cross on top of each other , without forming a zippered segment ( Figure 4D ) . Such crossings ( identified at the light microscopy level by a lack of visible adherent segment and by no change in the axon direction ) are marked by green symbols in Figure 1D–G and were not counted in the totals of Figure 1H . While the entangled zippers cannot be distinguished from the simple zippers with light microscopy , we examined high-magnification ( 1000× ) SEM pictures ( Figure 4—figure supplement 1 ) to find the following abundances of the three types of network vertices: 54% ( 134 out of 247 vertices ) were simple zippers , 28% ( 69 out of 247 ) entangled zippers , and 18% ( 44 out of 247 ) crossings . Besides axon shafts and their bundles , thin lateral protrusions emerging from the shafts are observed in the network ( Figure 4G ) . These protrusions appear highly dynamic and occasionally mediate interactions at a distance when they extend and touch another axon shaft . The observations reported above indicate that the progressive coarsening of the axon network results from zippering events driven by adhesion between the axon shafts . In recorded time lapse sequences of network evolution in 13 explants , we typically observed that the axon network coarsened in a manner similar to Figure 1 , or in some cases appeared stable when the recording was performed over shorter time intervals . This indicates that , in general , zippering events dominated over unzippering events . In a few cases , however , a de-coarsening dynamics was seen in a limited area of the network . Upon examining these cases , we noticed that they were associated with an apparent contraction of the explant , thus generating a pulling force on the axonal network ( Figure 1—figure supplement 1 ) . Stimulated by this observation , we sought a pharmacological manipulation by which a similar effect could be induced in a controlled manner . First , we envisaged treatments aiming at rapidly enhancing growth cone motility , in view of increasing the pulling force exerted by the GC on axon shafts . Since the molecular cues having such effects on OSN explant cultures remain unknown , we tested in a first approach Foetal Bovine Serum ( FBS ) , assuming that some of growth factors it contains may stimulate axon outgrowth . Interestingly , while no obvious effect on the growth cones was observed upon FBS addition to the culture , we found that the application of 5% FBS reliably induced the explant pull . This is likely due to a cell-rounding effect of FBS on cultured neurons , previously demonstrated in Jalink and Moolenaar ( 1992 ) . In Figure 5A–D , an example is shown , with a resulting de-coarsening in the axon network . Often , however , the FBS-induced pull resulted instead in a rapid collapse of the entire axon network onto the explant , due presumably to a disturbance of the attachments of the axons to the substrate . 10 . 7554/eLife . 19907 . 013Figure 5 . Defasciculation resulting from FBS-induced pull on the network . The schemes indicate the protocol of drug addition for the experiments that are shown on the frames below the schemes . ( A–D ) FBS was added to the culture at t=0 min . Decoarsening of part of the network ( marked by the red rectangle ) is visible . At a later stage , the network collapses . The full recording is provided as Figure 5—source data 1 . ( E–H ) The culture was pretreated by blebbistatin added before t=−60 min . Little change is visible between −60 and −1 min . At t=0 min FBS is added , after which progressive movement of the explant border to the left can be observed , exerting a pulling force on the axons . As a result , unzippering occurs , the network defasciculates and several new loops appear in the area marked by the red rectangle . The full recording is provided as Figure 5—source data 2 . ( I–L ) The culture was pretreated with blebbistatin ( t=−79 min ) and the network remained mostly unchanged until FBS was added ( t=0 min ) . Defasciculation is visible in the frames K and L , where the area of interest is marked by the red rectangle . The full recording is provided as Figure 5—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 01310 . 7554/eLife . 19907 . 014Figure 5—source data 1 . Development of axon network over 240 min , treated with FBS at t=90 min , video corresponding to Figure 5A–D . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 01410 . 7554/eLife . 19907 . 015Figure 5—source data 2 . Development of axon network over 142 min , pretreated with blebbistatin , and treated with FBS at t=79 min , video corresponding to Figure 5E-H . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 01510 . 7554/eLife . 19907 . 016Figure 5—source data 3 . Development of axon network over 142 min , pretreated with blebbistatin , and treated with FBS at t=79 min , video corresponding to Figure 5I-L . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 01610 . 7554/eLife . 19907 . 017Figure 5—figure supplement 1 . Stretching of the axons due to FBS-induced pull on the network . ( A–D ) Network configurations in the first 15 min after FBS was added to the culture . Three candidate paths of axons growing from the right edge of the explant are marked in green , blue and red , with the green and blue paths terminating in a growth cone . ( E ) Time course of the total length of the three segmented paths , each normalized to the initial path length . ( F ) Time course of the path straightness , defined as the ratio of the direct-line distance between the initial and final points of the path to the total path length ( i . e . straightness of 1 corresponds to a straight line ) . The colors of the datapoints in ( E ) and ( F ) correspond to the colors of the paths in ( A–D ) . The straightness of the blue and red paths approaches 1 . 0 , while the green path is prevented from such marked straightening by the immovable obstacle in the middle of the path . The coordinates of the path segments in panels ( A–D ) are available in Figure 5—figure supplement 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 01710 . 7554/eLife . 19907 . 018Figure 5—figure supplement 1—source data 1 . Coordinates of paths ( A–D ) , plot data ( E , F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 018 As an alternative means to influence axon tension , we tested blebbistatin , a well known inhibitor of neuronal Myosin II ( NMII ) , previously shown to decrease cell cortex/membrane tension in a variety of non-neuronal cells ( Fischer-Friedrich et al . , 2014; Ayala et al . , 2017 ) . Somewhat surprisingly , in our culture system 10 μM blebbistatin ( dissolved in dimethyl sulfoxide ( DMSO ) ) did not show visible effect on axon tension , but rather had a stabilizing effect on the network: the spontaneous coarsening was inhibited while the individual zippers remained mobile . No such effect was observed in control experiments in which only DMSO was added . We took advantage of the stabilizing effect of blebbistatin to facilitate the analysis of the coarsening or de-coarsening effects generated by subsequent treatments . In Figure 5E–H and I–L , two examples are shown in which FBS was applied to the pre-stabilized network and rapidly induced clear de-coarsening in parts of the network , with unzippering /defasciculation dominating over zippering events . In the example of Figure 5E–H , the edge of the explant ( visible near the left border of each frame ) retreated by approximately 20 μm to the left , thus stretching the axon network in the horizontal direction . Figure 5—figure supplement 1 evaluates how three candidate axon paths were deformed during the first 20 min after FBS was added ( i . e . in between the frames F and G of Figure 5 ) . The lengths of the paths increased by ( 8–23 ) μm , that is by 4–10% ( panel E in Figure 5—figure supplement 1 ) , while at the same time , the paths tended to become more straight ( panel F ) . The axon segments thus became significantly stretched and also aligned in the direction of the pull , as expected for an object under increased tension . The stretching of the axons by ∼15 μm is expected to result in a significant tension increase of over 1 nN ( see Discussion ) . This tension increase is achieved within 20 min of FBS addition and precedes the changes in network configuration seen in Figure 5G–H . As a complement to the observed de-coarsening induced by a pulling force , we sought to perturb the network dynamics by decreasing the tension in the axons . In previous literature , cytochalasin , an inhibitor of actin polymerization , was shown to significantly lower the tension of PC-12 neurites ( Dennerll et al . , 1988 ) . Indeed we found that in our system , 2 μM cytochalasin B ( dissolved in DMSO ) induces a change in network dynamics consistent with a drop of average axon tension . In the example in Figure 6A–C , the application of cytochalasin B induced coarsening ( panel C ) in a network that was previously stable ( panels A and B ) . As the networks generally have a tendency to coarsen , we sought to better isolate the effect of cytochalasin by applying it to networks that were pre-stabilized by blebbistatin . As shown in the example frames in Figure 6D–F and in the graphs in Figure 6G–I , cytochalasin B induces strong network coarsening within 30 min of application . No such effect was observed in control experiments in which only DMSO was applied . 10 . 7554/eLife . 19907 . 019Figure 6 . Cytochalasin-induced fasciculation of axon shafts . The schemes indicate the protocol of drug addition for the experiments that are shown on the frames below the schemes . ( A–C ) Cytochalasin was added to the culture at t=0 min . While there is little visible change between −65 and −1 min , the network exhibits coarsening between 0 and 35 min . The full recording is provided as Figure 6—source data 2 . ( D–F ) The culture was pretreated with blebbistatin before t=−65 min . Little change is visible between −65 and −1 min . After cytochalasin addition at t=0 min , the culture exhibits coarsening . The full recording is provided as Figure 6—source data 3 . The red arrows in frames C and F indicate prominent lamellipodia , which appear after the addition of cytochalasin . ( G–I ) The network statistics for the experiment of panel ( A–C ) ( red squares ) , panel ( D–F ) ( blue half-circles ) , and three other experiments with protocol equivalent to D–F , shown as Figure 6—source data 4 ( orange half-circles ) , Figure 6—source data 5 ( purple half-circles ) and Figure 6—source data 6 ( cyan half-circles ) . ( G ) Total length of the axon network in the field . ( H ) Total number of vertices of the axon network in the field . ( I ) Average area of cordless closed loop in the axonal network . The networks were manually segmented and analyzed as indicated in Materials and methods . In ( G–I ) , the data was aligned by the time of cytochalasin addition marked t=0 min and normalized by the value of the last measured timepoint before the drug was added . A sharp decrease of total length and of the number of vertices , as well as increase of average loop area , is seen within 30 min after t=0 min , indicating coarsening of the network triggered by cytochalasin addition . Segmentation data and frames are available in Figure 6—source data 8 ( please consult Materials and methods ) , source code used to generate the network statistics and input data is in Figure 6—source data 1 , the data points plotted in panels G–I are in Figure 6—source data 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 01910 . 7554/eLife . 19907 . 020Figure 6—source data 1 . ZIP archive; contains source code ( Figure 6_source_code . m ) and five ZIP archives with selection input data . Running the code ( with the five input archives in the same directory ) performs data processing and statistical analysis , and outputs the data shown in plots G , H and I ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 02010 . 7554/eLife . 19907 . 021Figure 6—source data 2 . Development of axon network over 165 min , treated with cytochalasin at t=65 min , video corresponding to Figure 6A–C ( red in graphs G–I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 02110 . 7554/eLife . 19907 . 022Figure 6—source data 3 . Development of axon network over 159 min , pretreated with blebbistatin , and treated with cytochalasin at t=65 min , video corresponding to Figure 6D–F ( blue in graphs G–I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 02210 . 7554/eLife . 19907 . 023Figure 6—source data 4 . Video of development of axon network over 159 min , pretreated with blebbistatin , and treated with cytochalasin at t=65 min , orange data points in Figure 6G–I . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 02310 . 7554/eLife . 19907 . 024Figure 6—source data 5 . Video of development of axon network over 129 min , pretreated with blebbistatin , and treated with cytochalasin at t=67 min , purple data points in Figure 6G–I . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 02410 . 7554/eLife . 19907 . 025Figure 6—source data 6 . Video of development of axon network over 166 min , pretreated with blebbistatin , and treated with cytochalasin at t=75 min , cyan data points in Figure 6G–I . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 02510 . 7554/eLife . 19907 . 026Figure 6—source data 7 . Plot data ( G , H and I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 02610 . 7554/eLife . 19907 . 027Figure 6—source data 8 . ZIP archive; contains video frames and segmentation data underlying the analysis shown in the Figure 6G–I . The archive contains analyzed frames ( TIFF format ) and corresponding segmentation selection data ( ImageJ-generated ZIP archives ) . Data can be displayed using ImageJ , please refer to Materials and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 027 To understand the conditions leading to zippering or unzippering , we analyze the force balance in a zipper of two axons ( Figure 7 ) . The contributing forces originate from the mechanical tension in each axon , the adhesion between each axon and the substrate , and the adhesion between axons in the zippered segment . In the following , we combine the mechanical tension T0 ( i . e . tensile energy per unit length of the axon ) and the axon-substrate adhesion ( i . e . energy of adhesion per unit length of the axon ) into an effective tension parameter T . The zipper will be in static equilibrium when the effective tensile forces are in balance with the force of adhesion between the axons . Consider for simplicity a symmetric zipper , in which the tensions in both axons are equal to each other , T=T1=T2 . At the vertex of the zipper ( Figure 7A ) , the force balance condition in the direction parallel to the zippered segment is given by ( 1 ) S=2T ( 1−cos⁡β2 ) 10 . 7554/eLife . 19907 . 028Figure 7 . Force balance in static axon zippers . ( A ) Illustration of a symmetric zipper . The zipper angle β is marked in blue . The arrows denote the vectors of tension T and axon-axon adhesive force S . ( B ) illustration of an asymmetric zipper , the markings are the same as in A , but the tensions within the axons differ ( T1≠T2 ) . ( C ) distribution of initial and final equilibrium angles of measured zippers ( 17 zippers , 34 measurements ) originating from four distinct cultures ( each obtained from a different mother animal ) , transformed into a probability distribution using convolution with Normal kernel . The red dashed line marks the average angle value ( 51 . 2° ) and the solid red box delimits the interquartile range ( 34°–60° ) . The values correspond to the full zipper angle β , which equals β=α1+α2 in asymmetric case . Individual distributions of the angles α1 , α2 were not recorded , because of prevailing symmetry of measured zippers . The distribution includes only those zippers , which were stable at least 5 min before and after the dynamics . The measured angles and the distribution of panel C are available in Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 02810 . 7554/eLife . 19907 . 029Figure 7—source data 1 . Estimated angle distribution ( C ) and underlying experimental angle data . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 029 where β is the zipper angle ( Figure 7A ) and S is the force arising due to the adhesion between the axon shafts . The force S may also be understood as the axon-axon adhesion energy per unit length of the zippered segment , and Equation 1 derived from the minimization of total energy ( see Materials and methods ) . If the tension T changes so that the equilibrium condition ( Equation 1 ) is no longer satisfied , zippering ( in the case of tension decrease ) or unzippering ( in the case of tension increase ) will result until a new equilibrium value of the zipper angle β is reached . Such changes in axon tension may occur due to a rearrangement of the network configuration in the vicinity of the zipper ( and hence a change in forces pulling on the segments of the zipper ) , or directly from changes in the basal tension generated by the pull of the growth cones and/or shaft cytoskeletal activity . The balance of forces at a junction of three axon segments was previously considered in ( Bray , 1979; Condron and Zinn , 1997; Shefi et al . , 2004; Anava et al . , 2009 ) , and used to analyze the distribution of tensions in a branched axon network . Provided that the junction is not strongly attached to the substrate ( Shefi et al . , 2004 ) , at a static branch point the tension force vectors in the three segments must add to zero ( and there is no axon-axon adhesion force ) . Breaking this balance results in a fast shift in the position of the branch point and adjustment of the branch angles ( Condron and Zinn , 1997 ) ; however , the material composition of the branches does not immediately change . In contrast , when the force balance of Equation 1 is broken , new portions of the two unbranched zippering axons are brought into contact , increasing the length of the zippered segment at the expense of the unzippered segments . To support the explanatory framework presented above , we carried out micro-manipulation experiments designed to measure the magnitude of the inter-axon adhesion force S and to investigate the dynamics of individual zippers . To determine S , we used observations of zippers in static equilibrium combined with measurements of the axon tension T . As seen from Equation 1 , the knowledge of T and of the zipper angle β permits to calculate the magnitude of the adhesion force S . It is known from previous literature that the typical value of mechanical tension in an isolated axon grown in culture is of the order of 1 nN ( [Dennerll et al . , 1988] reports a wide range of rest tension values , with the most common tension around 0 . 5 nN ) . Approaches using calibrated needles or Microelectromechanical systems ( MEMS ) had been successfully used to measure the tension of axons of dorsal root ganglia ( DRG ) neurons and PC-12 neurites ( Dennerll et al . , 1988 , 1989 ) , as well as motor neuron axons in Drosophila embryo ( Rajagopalan et al . , 2010 ) . In our case , the small diameter of OSN axons ( about 200 nm ) makes the use of such approaches difficult , because of the likely physical contact of the manipulator with the substrate resulting in an incorrect force reading , as well as in the detachment of the axon from the substrate . Optical tweezers technique would in principle allow the manipulation using microbeads attached to axons without touching the substrate , but does not permit to achieve manipulation forces comparable to 1 nN . Therefore , we decided to use the Biomembrane Force Probe ( BFP ) technique , in which a red blood cell is used as a force transducer ( Figure 8 ) . In this technique , streptavidin-coated glass microbeads ( 3 μm diameter ) attached to biotinylated axons are manipulated by a biotinylated red blood cell aspirated in a micropipette ( Evans et al . , 1995; Gourier et al . , 2008 ) . By measuring the deformation of the red blood cell , one can calculate the force with which the bead is manipulated . 10 . 7554/eLife . 19907 . 030Figure 8 . BFP measurements of axon tension . ( A–C ) illustrate a BFP experiment ( the full recording is in the video Figure 8–source data 1 ) . ( A ) The bead is slightly pushed against the axon ( with deflection angle δ<0 ) , negative deformation ( compression ) of the RBC is recorded . ( B , C ) Different stages of the probe exerting a pulling force on the axon; the RBC undergoes positive deformation ( extension ) , the axon deflection angle δ>0 . Index i of δi corresponds to the numbering of plateaux in panel D and data points in panel E . The tracked point on pipette and the tracked bead are marked by blue and red circles . ( D ) Time dependence of the force measured on the probe ( evaluated for each frame at 65 fps ) , and the angle ( evaluated each second ) . The deflection angle δ<0 corresponds to deflection by pushing , δ>0 means deflection by pulling . The deflection angle determines the lateral projection of axial tension acting at the apex , that is lateral tensile force 2⁢T⁢sin⁡δ . ( E ) Blue data points represent time-averaged qualities of individual plateaux ( labelled by appropriate numbers ) , abscissa corresponds to average deformation sin⁡δ and ordinate to average perpendicular probe force FBFP⊥ . The error bars represent a standard deviation of the quantities during each plateau . The red line is a linear fit of BFP data points , i . e . FBFP⟂ vs . sin⁡δ , the slope corresponds to axon tension 2⁢T . Goodness of the fit is R2=0 . 97 . ( F ) Distribution of axon tensions , calculated as a normalized sum of linear fit results from all BFP experiments—each fit j was represented by a Normal distribution , with mean at T¯j given by the fit slope , and standard deviation σ ( Tj ) , given by the standard deviation of the fit . The tension mode value is 678 pN , mean 679 pN ( designated by the dashed red line ) , interquartile range ( 529–833 ) pN ( delimited by the red box ) . The distribution of tension in based on N=7 measurements , containing at least three force plateaux each , originating from four distinct cultures ( each obtained from a different mother animal ) . The time course of force and angle ( D ) , plateaux points and the fit ( E ) , mean values of tension of all experiments and the distribution values ( F ) are available in the Figure 8—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 03010 . 7554/eLife . 19907 . 031Figure 8—source data 1 . Illustration of a BFP experiment with overlays that mark the results of pipette and bead tracking ( Video ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 03110 . 7554/eLife . 19907 . 032Figure 8—source data 2 . Time course of force and angle ( D ) , plateaux averages and fit parameters ( E ) , estimated tension distribution ( F ) and underlying experimental tension data . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 03210 . 7554/eLife . 19907 . 033Figure 8—figure supplement 1 . Tension increase during a BFP experiment with axon stretch . ( A ) Axon stretching observed during a BFP manipulation experiment . The two red dots serve as tracer points . The explant in the upper part of the field gradually moves upwards , pulling the proximal section of the measured bundle along , while no immediate retraction of the rest of the bundle is observed . ( B–D ) Distance between the two points increases by 3 μm , demonstrating the stretching of the axon . The stretching is likely responsible for observed increase in tension within the axon between the frames ( A–D ) . Comparing the tension measurements at the beginning of the video and at the later stages , we observe an increase from ( 432 ± 157 ) pN to ( 1665 ± 219 ) pN . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 03310 . 7554/eLife . 19907 . 034Figure 8—figure supplement 2 . Estimation of the axon-axon adhesion parameter S . ( A ) Distribution of static equilibrium angles measured from initial and final equilibria of measured zippers in vitro ( blue line ) , and expected distribution of equilibrium angles based on transformation of the tension probability distribution ( Equation 6 in Materials and methods ) ( red line ) . The adhesion parameter S of the PDF ( T ) →PDF ( β ) transformation was optimized to achieve maximal correlation between the distributions , with the result S=88 pN and correlation coefficient 0 . 813 . ( B ) Contour plot of the joint probability density of axon tension T and static equilibrium vertex angle β . The two distributions were considered independent , the joint probability is the product of marginal PDFs , which are shown along the corresponding coordinate lines . ( C ) Distribution of adhesion parameter S , calculated by screening the values obtained from Equation 1 and the joint distribution in panel B . Interquartile range of S is ( 52–186 ) pN ( delimited by the solid red box ) , while the median of the distribution is 102 pN , designated by the dashed line . The data of experimental and transformed distributions of angles ( A ) , experimental distributions of angles and tension ( B ) and probability distribution of adhesion coefficient ( C ) are available in Figure 8—figure supplement 2—source data 1 , source code used to process the data is in Figure 8—figure supplement 2—source data 2 . . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 03410 . 7554/eLife . 19907 . 035Figure 8—figure supplement 2—source data 1 . Data plotted in Figure 8—figure supplement 2A–C . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 03510 . 7554/eLife . 19907 . 036Figure 8—figure supplement 2—source data 2 . Source code to process input data from Figure 8—figure supplement 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 036 Using the BFP technique , we determined the tension in a collection of thin network segments , presumably individual axons , even though we cannot exclude that some of them might have been fascicles of several axons . The basis of the measurement is force equilibrium between the calibrated force of the probe acting on and deforming the axon , and a restoring force , which arises from the tension in the axon shaft . The measurement is described in Figure 8 . The initially straight axon ( Figure 8A ) is deformed by displacing the micropipette and holding it in a fixed position ( Figure 8B ) . The force equilibrium is reached: the pulling force FBFP is balanced by the projection of the axon tension in the transverse direction 2⁢T⁢sin⁡δ , where δ is the angle of deflection of the axon ( i . e . 180∘−2δ is the angle at the apex of axon deformation ) . This operation is repeated for larger displacements ( Figure 8C ) , until the red blood cell detaches from the bead , which generally occurs at a deformation angle of about δ≈5∘ . Figure 8D shows the time course of the pulling force measured on the probe during this experiment , as well as the measured deformation angle . The force plateaux ( labelled 1 to 5 in Figure 8D and marked by black boxes ) correspond to the time intervals during which the micropipette position was held fixed . To extract the value of the tension in the axon , a linear fit is performed on the transverse projection of pulling force vs . sin⁡δ ( Figure 8E ) . The slope of the fit line gives the tension T=906pN in the case of the experiment shown in Figure 8 . The non zero intercept of the fit arises from calibration effects described in Materials and methods . Out of a several dozen measurements performed , we obtained a collection of eight measurements from seven axons that included at least three plateaux in each . For one of the axons , two distinct values of tension were measured early and late in the experiment: ( 432 ± 157 ) pN and ( 1665 ± 219 ) pN . This increase in tension was likely caused by the strong stretching of the axon that occurred during this particular experiment—see Figure 8—figure supplement 1 . Such stretching is unlikely to occur during spontaneous dynamics of the axon network ( without added drugs ) , and we excluded the post-stretching data from the analysis . This experiment indicates , however , that the FBS-induced pulling ( Figure 5 ) may have lead to very significant increases in axon tension . Using the slope values and their errors calculated from the seven remaining linear fits , we estimated the distribution of the tensions in the axon population , shown in Figure 8F . The distribution is sharply peaked near 678 pN , with the mean value of 679 pN and the interquartile range ( 529–833 ) pN . Technical limitations were encountered in these experiments , including the uncontrollable bead localization along the axons and with respect to zippers , as well as early detachment of beads from the red blood cell upon pulling . The seven measurements included in Figure 8F correspond to the most robust ones and were obtained with beads that were not necessarily in the vicinity of a zipper vertex . It was therefore not feasible to correlate the measured tension values with measured zipper angles on the level of individual axons . Rather , we chose to obtain a separate set of measurements of equilibrium zipper angles . As zippers that are entangled ( as in Figure 4E , F ) may remain static without satisfying the equilibrium Equation 1 , we restricted our measurements to zippers that were observed to be mobile before reaching a static configuration . In the videorecordings of the developing network , we selected 17 such zippers that were approximately symmetric and appeared to consist of single axons ( or possibly thin fascicles ) . We measured the zipper angles of the equilibrated configurations ( requiring stability over at least 5 min ) , and based on these values estimated the distribution of equilibrium zipper angles in the zipper population ( Figure 7C ) . The distribution is sharply peaked around 42° , with mean of 51 . 2° and interquartile range ( 34–60 ) ° . Based on the measured distributions of axon tensions and of equilibrium zipper angles , we then estimated the axon-axon adhesion force S . First , we assumed that the two distributions are related to each other through Equation 1 and determined the value of S resulting in their best mutual match ( see Materials and methods ) , obtaining S=88 pN . In an alternative procedure , we estimated a joint distribution of the axon tensions and equilibrium zipper angles ( treating the two variables as independent ) , and used Equation 1 to compute the corresponding distribution of adhesion parameters S ( see Materials and methods ) . This procedure allows for the expected variability of the values of S among zippers ( e . g . due to different areas of contact ) , and gives a maximum interquartile range of S= ( 52–186 ) pN , with a median of 102 pN . To determine the axon adhesion force more directly , not relying on the measurement of axon tension , we attempted to unzipper selected zippers using a calibrated pulling force . These attempts were not successful , due to insufficient strength of the bond between the bead and the red blood cell . This resulted in the detachment of the red blood cell before any significant effect on the zipper . To overcome this limitation , we bypassed the red blood cell and bead and used the pipette to drag the axon directly . This allowed us to use forces sufficiently large to induce unzippering at the price of losing the knowledge of the force magnitude . Figure 9 and the corresponding video Figure 9—source data 1 show an example . By dragging one of the axons of a zipper , we increased the zipper angle beyond its equilibrium value , leading to unzippering accompanied by a decrease of angle ( Figure 9A–D ) . Then , the axon was released by lifting the pipette . The axons snapped back to a smaller zipper angle which initiated a re-zippering process accompanied by an increase of the angle ( Figure 9E–F ) , leading finally to the recovery of the initial configuration . Similar manipulations performed on other zippers either gave analogous results ( Video 2 ) , or in some instances , no unzippering ( Video 3 ) . However , this latter case is likely to be due to the structural organization of these particular zippers involving entangled axons ( Figure 4E , F ) . 10 . 7554/eLife . 19907 . 037Figure 9 . Example of induced unzippering and rezippering . ( A ) Initial state of the zipper before the manipulation . ( B ) Angle increases as vertex shifts to the side with the initial pipette displacement . ( C–D ) The axons unzipper , slowly . ( E ) Pipette is removed and axon released , the vertex shifts strongly to the left to equilibrate the lateral force imbalance . ( F ) Axons zipper back toward the initial configuration . ( G ) Blue line ( with time stamps ) shows the velocity and angle of the vertex during the manipulation . Data points belonging to the full line were fitted using linear regression ( dashed red line ) . The goodness of the fit is R2=0 . 48 . The pale-blue dashed line corresponds to transients arising during manipulation ( excluded from the regression ) . The values of angle were smoothed by a 20-frame Gaussian filter , and the velocity was calculated using convolution of positional data with derivative of the same Gaussian filter . The blue arrows show the direction of increasing time . Velocity , angle data and fit of panel G are available in Figure 9—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 03710 . 7554/eLife . 19907 . 038Figure 9—source data 1 . Induced unzippering experiment video corresponding to Figure 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 03810 . 7554/eLife . 19907 . 039Figure 9—source data 2 . Velocity and angle data , fit parameters ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 03910 . 7554/eLife . 19907 . 040Video 2 . Induced unzippering of a zipper segment delimited by two vertices on either side . In this case the unzippering becomes complete and the two constitutive bundles separate . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 04010 . 7554/eLife . 19907 . 041Video 3 . Induced unzippering experiment . In this case the vertex does not recede , despite the large increase in zipper angle resulting from the manipulation by micropipette . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 041 Similarly to these cases of induced unzippering/rezippering , we view the numerous individual zippering processes observed in the developing network ( Figures 2 and 3 ) as arising from force perturbations that act on a zipper and move it to a new equilibrium configuration . These perturbations may consist in changes in the network geometry in the vicinity of the zipper , or in changes in mechanical tension within the axons that constitute the zipper . To characterize such spontaneous zippering dynamics , we tracked 17 individual zippering processes within the developing network and measured how the zipper configuration evolved . All 17 zippers selected for this analysis started from approximately stationary initial configurations , and reached a final configuration that remained stationary for at least five min . Selected typical examples are shown in Figure 10 . The distance of the zipper vertex from the final equilibrium position is plotted as a function of time in Figure 10A , B ( the time point when equilibrium is reached is defined as t=0 ) . It can be seen that in both advancing ( Figure 10A ) and receding ( Figure 10B ) zippers , the zipper vertex moves with a velocity in the range ( 0 . 3−2 ) μmmin . Figure 10C shows that while some zippers ( R3 and A6 ) converge with an approximately constant velocity , others ( R4 and A5 ) have a weakly exponential velocity profile , with the velocity gradually decreasing as equilibrium is approached . The former case , in which the zipper stops rather abruptly near the equilibrium position , is observed in roughly 2/3 of the evaluated examples . In Figure 10D , the smoothed zipper angle is plotted as a function of time for three advancing and two receding zippers . In these examples , the angle increases with time for advancing zippers ( A1 , A4 , A6 ) and decreases with time for receding zippers ( R4 , R5 ) . In some other cases ( typically those in which the zipper configuration was complex , e . g . influenced by side processes ) , the time dependence of the zipper angle was more irregular . The full dynamics of the zippers R4 and R5 is shown in the videos Figure 10—source data 1 and Figure 10—source data 2 . 10 . 7554/eLife . 19907 . 042Figure 10 . Dynamics of spontaneous zippering events in the evolving network . ( A ) and ( B ) show the convergence to equilibrium for selected advancing and receding zippers , respectively . The distance between the zipper vertex location at the given time and the final equilibrium position is given . The lines with slopes ( 0 . 3−2 . 0 ) μmmin delimit the typical zippering and unzippering velocities . ( C ) Fits illustrating approximately linear or exponential convergence in time . Linear fit equations: dR3 ( t ) =−1 . 11 ( t+6 . 94 ) and dA6 ( t ) =−0 . 45 ( t+7 . 59 ) ; exponential fit equations: dR4 ( t ) =14 . 95exp⁡{−0 . 20 ( t+19 . 14 ) } and dA5 ( t ) =10 . 42exp⁡{−0 . 22 ( t+19 . 12 ) } . ( D ) Time course of zipper angles , smoothed by five-frame window . Note that the angle increases for advancing zippers and decreases for receding zippers . Data plotted in panels A , B and D are available in Figure 10—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 04210 . 7554/eLife . 19907 . 043Figure 10—source data 1 . Video of receding zipper R4 from Figure 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 04310 . 7554/eLife . 19907 . 044Figure 10—source data 2 . Video of receding zipper R5 from Figure 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 04410 . 7554/eLife . 19907 . 045Figure 10—source data 3 . Plot data ( A , B and D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 045 Our analysis of equilibrium zipper configurations ( cf . Equation 1 ) was based on viewing the zippers as arising from the interplay of mechanical tension and inter-axon adhesion forces . To assess if the observed zipper dynamics is consistent with this framework , we developed a basic biophysical dynamical model , formulated as an effective equation of motion for the zipper vertex ( see Materials and methods for the underlying assumptions and a full derivation ) . Consider the instantaneous configuration shown in Figure 11A; here , the axons are fixed at the points A , B , C ( these may correspond to entangled connections with the rest of the network , to immobile adhesion points with the substrate , or to the soma or the growth cone ) , while the zipper vertex V⁢ ( x , y ) is mobile . The condition for static equilibrium of the vertex ( given by Equation 1 in the case of a symmetric zipper and by Equations 9 , 10 in the general asymmetric case ) takes into account the mechanical tension in the axons and the force arising from axon-axon adhesion . When the vertex is moving , however , additional forces arise from energy dissipation . As shown below , including these frictional forces in the force balance condition permits to obtain an equation of motion , specifying the velocity of the vertex . 10 . 7554/eLife . 19907 . 046Figure 11 . Geometry and notation for the dynamical model of axon zippering . ( A ) Illustration for the zippering dynamics model . L1 and L2 denote the lengths of the two axons . The red dotted line represents the adhered zipper segment and its extension beyond the vertex ( i . e . zipper axis ) , aligned with the y-axis in the figure . The blue vector F→v represents the conservative forces ( i . e . tension and adhesion ) and the red vector u→ the resulting vertex velocity limited by friction . Projection of the velocity to the zipper axis , uZ , determines the vertex-localized dissipative force , as fZ=−ηZuZ . The strain rate , L˙L , determines the elongational viscous dissipative force , that is f⇕=−η⇕L˙L . ( B ) Illustration for the Appendix . δx→ represents a small displacement of the vertex . Vector v→ is the velocity of the element d⁢l , in contrast to the velocity of the vertex , u→ , in panel A . Axial and transverse substrate friction forces f→∥ and f→⊥ are proportional to the element velocity components v→∥ and v→⊥ . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 046 We first describe the frictional force arising from the stretching or shortening of axons ( which necessarily occurs during zippering or unzippering ) . Within the linear viscoelasticity framework , the viscous stress in each axon is proportional to the local strain rate . Assuming a uniform elongation strain in between the axon fixed points , the strain rate is simply expressed as L˙L , where L is the total length of the segments of the axon . During axon elongation or shortening , the total force acting in a cross-section of the axon is therefore ( 2 ) τ=T+η⇕⁢L˙L where T denotes , as before , the axon tension , and η⇕ is the elongation viscosity constant . In addition to axon elongation/shortening , another possible source of energy dissipation consists in changes in the axon configuration in the immediate vicinity of the vertex . When the vertex advances during zippering , new regions of the axons undergo bending/unbending ( internal structural changes ) , with corresponding viscoelastic losses . Possible non-equilibrium binding effects at the newly adhering membrane region may also result in dissipation . These energy losses are expected to result in a localized frictional force that acts at the vertex and is anti-parallel to the vertex velocity component along the axis of the zipper; that is , this frictional force is collinear with the adhesion force −SVC^ . The magnitude of the combined zipper adhesion/friction force is ( 3 ) χ=S−ηZu→⋅ ( −VC^ ) =S−ηZuZ where ηZ is a friction constant and uZ is the ‘zippering velocity’ , given by the projection of vertex velocity in the direction of advancing zipper ( see Figure 11A ) . Thus , the friction force acts in the direction of the adhesion force during unzippering and in the opposite direction during zippering . The balance of forces at a moving vertex may now be readily expressed . Consider for simplicity the case of a symmetric zipper ( the asymmetric case is treated in Materials and methods and the Appendix ) . The dynamics preserves the symmetry , that is , an initially symmetric configuration ( T1=T2=T , α1=α2=β2 , L1=L2=L ) will remain symmetric during the course of zippering . Aligning the zippering direction ( i . e . the direction of the zippered segment ) with the y axis , we have uZ=y˙ in Equation 3 and L˙= ( 1−cos⁡β2 ) y˙ in Equation 2 . Replacing now , in the equilibrium equation Equation 1 , T by τ ( Equation 2 ) and S by χ ( Equation 3 ) , we obtain the condition expressing the total force balance in a moving vertex . Rearranging to express the zippering velocity y˙ , we get the equation of motion for a symmetric zipper ( 4 ) y˙=S−2T ( 1−cos⁡β2 ) 2η⇕ ( cos⁡β2−1 ) 2L+ηZ . The terms cos⁡β/2 and L on the right hand side are nonlinear functions of y and are straightforwardly expressed in terms of the coordinates of the fixed points A , B , C . The resulting differential equation ( Equation 4 ) cannot be solved in closed analytical form , but the predicted vertex trajectory y⁢ ( t ) can be obtained by numerical integration . We tested the equation of motion Equation 4 by comparing it with the experimental recordings of induced zippering/unzippering dynamics in our system . We measured the zippering velocity y˙ and the zipper angle β during the experiment shown in Figure 9A–F; these quantities were evaluated at 1 s intervals and smoothed using a Gaussian kernel of half-width 10 s . Figure 9G demonstrates that the zipper velocity is linearly related to 1−cos⁡β2 . In this plot , the fast transients resulting from the axon manipulation are shown as pale blue dashed curves , while the zippering/unzippering dynamics induced by the manipulation ( once the axons relaxed into an approximately symmetric configuration ) is shown as solid curves . The straight red line indicates the best linear fit ( from which the fast transient manipulation segments were excluded ) . Comparing now to Equation 4 , we see that such linear dependence is predicted when the friction in the vertex ( i . e . term proportional to ηZ ) dominates over the elongation friction ( ηZ⁢uZ≫η⇕⁢L˙L ) . The slope of the linear fit is predicted to equal −2T/ηZ , while the predicted intercept is S/ηZ . From the ratio of the intercept ( 0 . 0825μms ) and the slope ( −4 . 1692μms ) in Figure 9G , we therefore obtain an estimate of the ratio of the axon adhesion force S to the axon tension T . This dynamical estimate gives ST=0 . 04 , as compared with the typical value ST=162pN679pN=0 . 24 that we obtained from the analysis of static configurations . The induced zippering experiment of Figure 9 was performed with axon bundles located close to the explant boundary; these have larger total tension T than the single axons forming the zippers used in our static analysis , while the adhesion parameter S is expected to scale sub-linearly with the number of axons in the bundle; this may explain the lower dynamic S/T ratio . Assuming a tension of order 2 nN , the slope of the fit indicates a value for the vertex friction constant of order ηZ∼10−3Nsm . We now discuss the zippering dynamics in the case of an asymmetric zipper . The general equation of motion for the vertex is presented in the Appendix , and includes ( in addition to the elongation and zippering friction introduced above ) the friction of the axons with the substrate . Figure 12 shows representative trajectories of the vertex obtained by numerical integration of the general equation of motion . The panel 12A displays a contour plot of the energy landscape E⁢ ( x , y ) , defined as the total tensile and adhesive energy of the zipper configuration with vertex located at ( x , y ) ; this energy is given by Equation 7 ( in Materials and methods ) . The energy landscape plotted in Figure 12A corresponds to a zipper constituted by axons with tensions T1=1 nN and T2=1 . 5 nN and mutual adhesion strength S=0 . 2 nN . The marked ‘final point’ denotes the static equilibrium point of the landscape . The initial point of the trajectories in Figure 12A corresponds to the equilibrium zipper configuration for T1=T2=1 nN . Following a rapid increase ( between time t=0 sand t=5 s ) of the tension in the right axon by 0 . 5 nN , the zipper undergoes relaxation to the new equilibrium , driven by the force given by the gradient of the energy landscape displayed in Figure 12A . It is seen that different forms of dominating friction ( black for viscous elongation , red for substrate friction , blue for vertex-localized friction ) lead to distinct paths ( Figure 12A ) as well as time courses ( Figure 12B , D ) of the trajectory . For comparison , the red dashed curve in Figure 12A shows the gradient path , which would correspond to an isotropic and geometry-independent vertex friction tensor H↔ ( see Materials and methods ) . 10 . 7554/eLife . 19907 . 047Figure 12 . Predicted zippering dynamics resulting from applying a perturbation to a zipper initially in equilibrium , converging to a new equilibrium . ( A ) The landscape of tensile and adhesive energy ( Equation 7 ) for the new equilibrium condition ( specified by the parameter values: left tension T1=1nJm , right tension T2=1 . 5nJm ( up from T2=1nJm in the initial equilibrium ) , axon-axon adhesion S=0 . 2nJm ) . Blue contours indicate locations of equal energy . Gray dashed lines show axons in the final equilibrium , dashed red line is the gradient trajectory between equilibria , . The full lines indicate zipper vertex trajectories following a rapid increase of the right tension . Red: trajectory with dominant substrate friction ( f∥+f⟂ ) , blue: trajectory with dominant zippering friction ( fZ ) , black: trajectory with dominant elongation friction ( f⇕ ) . The trajectories with dominant friction types are represented by the same color code across all panels . ( B ) The same trajectories as in A , with time stamps . The tension in the right axon increased rapidly over 5 s and then was kept constant ( see green line in panel D ) . ( C ) Trajectories during gradual perturbation , with dominant zippering friction ( blue ) , substrate friction ( red ) and elongation friction ( black ) over 1000 s . The tension was gradually growing in the right axon over 500 s and then was kept constant ( see green line in panel E ) . ( D , E ) velocity of vertex during transition , color code corresponds to panels B and C , green line represents the prescribed tensile force in the right axon during the transition . For each model run , one friction constant was set to a particular value to probe its effect on the trajectory , others were set to zero . The following values were used: axial substrate friction η∥=200 Pa s , transverse substrate friction η⟂=200 Pa s , elongation friction η⇕=3000 nN s , zippering friction ηZ=nNsμm ( in this case , a small substrate friction value η∥=η⟂=1 Pa s was introduced to avoid a singularity when the motion direction was perpendicular to the zippering axis ) . Source code of the implemented zipper model used to generate the data is available in Figure 12—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 04710 . 7554/eLife . 19907 . 048Figure 12—source data 1 . Source code of zipper dynamical model to generate plot data ( A–E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 048 Our experimental observations ( as in Figure 9 ) show that a typical response of a zipper to a fast asymmetric perturbation consists of a fast lateral equilibration , followed by a slower dynamics during which the vertex moves parallely to the zippered segment . Such trajectory arises from our model in case of dominant zipper friction ( blue line ) , while it cannot be achieved through the other friction mechanisms alone . We conclude that the velocity of zippering is primarily limited by the internal friction localized at the zipper vertex . Our observations of spontaneous zippering processes in the developing network showed that the zippering velocity typically remained approximately constant ( for ∼2/3 of the events ) , with abrupt stop near the equilibrium point ( see Figure 10 ) . The velocity profiles obtained from the model in case of zippering resulting from abrupt perturbation , in contrast , are exponential or double-exponential ( Figure 12D ) . An approximately constant velocity of zippering is obtained in the model , however , when the tension is assumed to increase gradually over an extended interval of minutes , rather then abruptly ( Figure 12E ) . The corresponding trajectories are shown in Figure 12C . In this case , the paths obtained for different dominating forms of friction are similar to each other . This is a consequence of the gradual increase in tension: for all of the friction types considered , the relaxation dynamics is then sufficiently fast to allow the zipper vertex to closely track the equilibrium point of the energy landscape , which evolves on the time scale of minutes . These results suggest that in the developing network , the zippering is driven by gradual , rather than abrupt , changes in the forces that act at the zipper vertex . The resulting reconfiguration of the zipper may then act as a gradual perturbation acting on the zippers in the immediate vicinity . To summarize , the comparison of predictions of the dynamical model with experimental observations supports a framework in which the zippering arises from an imbalance of tension and adhesion forces at the zipper vertex , and in which the zippering velocity is limited predominantly by friction arising from internal energy dissipation in the immediate vicinity of the moving vertex . Following the analysis of the statics and dynamics of individual zippers , we establish a connection to the global dynamics of developing axon network . The gradual decrease of the total network length with time ( Figure 1H ) indicates that in our experimental setting , zippering is overall more frequent than unzippering . The observed decrease of the total number of vertices ( Figure 1H ) is likewise a natural consequence of zippering . An advancing zipper vertex may eventually encounter another vertex and combine with it , resulting in a zipper consisting of thicker fascicles . A process of this type repeatedly observed in the developing network ( Figure 3G–J ) consisted of a gradual collapse of triangular loops , with the three vertices eventually converging into a single-vertex quasi-stable configuration . A possible underlying zipper structure is illustrated in Figure 13A . During this process , the loops typically retained their shape , that is the three zipper angles remained approximately constant during the collapse . Such dynamics is expected to result from a decrease in the tension of the axons that constitute the loop , such that the equilibrium zipper angle becomes larger than the current zipper angles . In such case , no stable redistribution of angles is possible and the vertices advance synchronously , keeping the loop shape invariant . This combined dynamics is therefore distinct from the elementary zippering process we considered in Figure 11 , where it was assumed that the fixed points A , B , C were immobile , and consequently the zipper angle gradually increased as the zipper approached equilibrium . A strong support for this interpretation of the mechanism of loop collapse is provided by the experiments in which we used FBS to generate a pull on the network , hence increasing axon tension . As seen in Figure 5F–H , this manipulation leads to the rapid opening and expansion of triangular loops in the de-coarsening areas of the network . 10 . 7554/eLife . 19907 . 049Figure 13 . Elementary dynamical processes in fasciculating axon networks and in coarsening liquid foams ( froths ) . ( A1–A3 ) Two vertices are lost during the process of closing of a loop formed by three axon segments . The initial configuration starts to zipper at one or more vertices , gradually decreasing the total network length . A single junction formed by three pairs of fully zippered axons remains . ( B1–B4 ) Possible outcomes of initial contact of two axons . ( B1 ) Growth cone ( GC ) interacts with the shaft of another axon , ( B2 ) small initial incidence angle allows incoming GC to adhere and follow the shaft , ( B3 ) the two shafts zipper , increasing the contact angle , ( B4 ) if the initial incidence angle exceeds the equilibrium zipper angle , no stable zippered segment can be formed and the growth cone crosses over . ( C ) Photograph of structure formed by a liquid foam restricted between two glass plates . The gas bubbles are separated by liquid walls that meet at triple junctions . ( D1–D4 ) Schemes illustrating the elementary topological processes in liquid foams . ( D1 ) A three-sided bubble with curved walls , containing gas under excess pressure . ( D1–D2 ) The gas diffuses to neighboring cells and the three-sided bubble gradually collapses . This process is called T2 . ( D3–D4 ) In foams , the T1 process leads to a reconnection of bubble walls , preserving the number of vertices of the network . ( E1–E2 ) In the axon network , separation rather than reconnection results from unzippering . ( E1 ) Two vertices delimiting a zippered segment start to recede , ( E2 ) Once the adhered segment length decreases to zero , the two axons detach and separate ( see experimental example in Video 2 ) . Panel C is adapted from https://commons . wikimedia . org/wiki/Category:Foam\#/media/File:2-dimensional_foam . jpg:Foam\#/media/File:2-dimensional_foam . jpg by Klaus-Dieter Keller , released into the public domain by the author . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 049 Our induced zippering experiments and model analysis showed that the zippering transients resulting from sudden perturbations last for minutes , while the coarsening of the network develops over hours . Such separation of time scales indicates that the network is locally near the quasi-equilibrium state corresponding to the momentaneous values of the axon tensions . The network statistics reported in Figure 1H exhibit robustly monotonous time course and low volatility , which is consistent with this assumption and shows that large abrupt perturbations do not dominate the network dynamics . At a given time , the majority of vertices in the network are seen to be approximately static or fluctuating around an equilibrium position , while the proportion of steadily advancing or receding zippers is minor ( see Video 1 ) . In the following analysis , we will assume that the majority of the zippers have a zipper angle that is close to the equilibrium value given by Equation 1 ( see also Discussion ) . The observed decrease in total network length implies that larger fascicles are gradually formed . The limits of optical microscopy resolution did not allow us to reliably determine the size of fascicles forming individual zippers . However , the structure of the fascicles determines their tension and is therefore expected to be reflected in the equilibrium zipper angles ( cf . Equation 1 ) . To examine this relation , we extracted the distribution of the zipper angles in the network . At each analyzed time point , the network was manually segmented as in Figure 1D–G and the angles between the graph edges were measured . At each zipper vertex , the zippering angle was selected as the sharpest of the three angles between the edges , unless the observed configuration indicated otherwise . Crossings ( marked by green stars in Figure 1D–G ) were excluded from the statistics . The analysis included a total of five experiments in which the network coarsened ( each lasting for 178 to 295 min ) , with 7–10 time points per experiment at which the zipper angle distribution was extracted . The typical shape of the distribution is shown in Figure 14A . Note the marked under-representation of sharp zippering angles ( below 20° ) . In the example of Figure 14A ( which corresponds to Figure 1D-H ) , the median angle of the distribution shifted to lower values during the 3 hr interval ( from 60° to 49° ) . 10 . 7554/eLife . 19907 . 050Figure 14 . Evolution of the network distribution of zipper angles . ( A ) The distribution of zipper angles in the network configurations of Figure 1D ( t=0 min , total 66 vertices ) and Figure 1G ( t=178 min , total 44 vertices ) . ( B ) Correlation between median angle βM and the total network length L in two experiments; r denotes the Pearson correlation coefficient . ( C ) Predicted equilibrium zipper angle distribution PDF ( β ) obtained as a transformation of distribution of fascicle tension PDF ( T ) using Equation 6 ( see main text and Materials and methods ) . The distribution of tensions was approximated by a lognormal distribution PDFlog ( T¯ , σ ( T ) ) . The distribution plotted in orange corresponds to the values of tension from BFP experiments , T¯BFP=0 . 68 nN , σBFP⁢ ( T ) =0 . 25 nN , and adhesion parameter S0=0 . 17 nN adjusted to match the initial median angle of experiment 1 . The distribution plotted in green corresponds to parameters rescaled with mean fascicle size n as T¯∼n , σ ( T ) ∼n , S∼n , with increase in mean fascicle size ( 1 . 50× ) corresponding to Figure 1D–G ( see main text ) . The change in median angle in panel C is 6 . 5° , as compared to 7 . 5° given by the trendline in panel B ( experiment 1 ) . The data of histograms ( A ) , correlations ( B ) and distributions ( C ) are available in Figure 14—source data 1 . The source code used to generate distributions in panel C is available in Figure 14—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 05010 . 7554/eLife . 19907 . 051Figure 14—source data 1 . Data of histograms ( A ) , correlations ( B ) and distributions ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 05110 . 7554/eLife . 19907 . 052Figure 14—source data 2 . Source code to generate angle distributions ( C ) using Equation 6 , see Materials and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 052 Evaluating the relation between the zipper angle distribution and the network coarsening , we found a consistent trend in the five analyzed experiments . The median zipper angle βM overall showed a positive correlation with the total network length L ( with the five correlation coefficients in the range ( 0 . 26–0 . 69 ) ) . Two examples are shown in the scatter plots of Figure 14B , where experiment 1 corresponds to the time interval in Figure 1H . To propose an explanation of this observation , we return to the distribution of single-axon tensions obtained using the BFP technique ( Figure 8F ) . We assume that the distribution of tensions of zipper-forming axons ( either single or fasciculated ) matches the distribution from the BFP experiments . Treating the tensions of individual axons in a fascicle of size n as independent random variables , it follows that the mean of the fascicle tension distribution scales as T¯∼n and its standard deviation as σ ( T ) ∼n . To evaluate how this is reflected in the distribution of zippering angles , we use Equation 1 with an appropriately rescaled adhesion strength S . The adhesion force between two fascicles scales with their contact area and therefore with the fascicle surface . For a fascicle composed of n axons , the surface is expected to scale as ∼n ( assuming that the cross-section of the fascicle remains approximately circular , rather than flattened by strong adhesion to the substrate , which is supported by the SEM micrographs presented in Figure 4 ) . Using these scaling rules and Equation 6 ( which follows from Equation 1 , see Materials and methods ) , we can transform the distribution of tensions p⁢ ( T ) into the distribution of zippering angles q⁢ ( β ) . To qualitatively asses the changes of q⁢ ( β ) with fascicle growth , we made two simplifications: ( i ) we replaced the experimental distribution of tensions with a lognormal distribution , p ( T ) =PDFlog ( T¯ , σ ( T ) ) , of the same mean T¯=T¯BFP=0 . 68nN and std σ ( T ) =σBFP=0 . 25nN and ( ii ) we used a single value of S ( appropriately scaled with n ) , ignoring its possible variance . We verified numerically that the lognormal approximation of tension distribution for fascicles of size n≥2 closely corresponds to the tension distribution obtained by n-fold convolution of the single-axon distribution . The distribution of zipper angles is then given by Equation 6 , using the lognormal distribution of tensions with the two parameters related through scaling with n as T¯=n⁢T¯BFP , σ ( T ) =nσBFP , and using the scaled value of adhesive strength S=n⋅0 . 17nN . As shown in Figure 14C , this analysis predicts that a coarsening-induced increase in mean fascicle size n leads to a lower median zipper angle , in agreement with the trend seen in the experimental data . The orange curve in Figure 14C is the angle distribution with parameters set to the BFP-derived values , while the green curve is the predicted distribution after rescaling of fascicle size n by factor 1 . 50 . This factor was obtained from the data in Figure 1H and from the expected scaling n∼1/D , where D is the total network length per unit area . Strong connection points can be established between our dynamical observations and the in vivo observations of Roberts and Taylor ( Roberts and Taylor , 1982 ) , who studied the formation of the sensory neurite plexus on the basal lamina of trunk skin in Xenopus embryos . In Roberts and Taylor ( 1982 ) , the neurite network on the trunk and the inside skin surface was examined using electron microscopy at magnification 1000–2500 , and the angles between neurites that fasciculated or crossed ( ‘incidence angles’ ) were determined . As shown in the inset of Figure 15 , the distribution of the fasciculation incidence angles in ( Roberts and Taylor , 1982 ) is similar to the distribution of zipper angles measured in our system . Small angles ( between 0° and 30° ) are notably absent from the recorded angle distributions ( see also Figure 14A ) , while these angles would be a priori expected to be equally represented in an isotropically growing network ( and overrepresented in a network with a preferred direction of growth ) . Roberts and Taylor proposed that this was a result of zippering processes analogous to the ones that we directly observed in our study . Thus , if a growing axon encounters another axon at an initially small incidence angle and starts following it ( Figure 13B1 , B2 ) , the segment behind the growth cone subsequently zippers and the incidence angle increases until the equilibrium zipper angle is reached ( Figure 13B3 ) . Our observations of zippering dynamics are consistent with this proposal . The under-representation of small angles in the zipper angle distributions ( Figures 14A and 15 ) thus further supports our inference that most zippers are close to local equilibrium during the development of the network . 10 . 7554/eLife . 19907 . 053Figure 15 . Prediction for the probability of crossing ( rather than zippering ) of two neurites , and its comparison to in vivo experimental data . The red dots show data from Fig . 14 of Ref . ( Roberts and Taylor , 1982 ) ( referred to as Roberts in the figure ) : the observed probability of crossing of two neurites as function of observed incidence angle βinc . The blue line is the cumulative distribution function of angles of incidence obtained from Roberts' data . It was calculated as Π ( βinc ) =∫0βincp ( β ) dβ , where the PDF of angles of incidence , p ( β ) , was constructed by kernel-smoothing the Roberts’ histogram . The crossing probability of interval ( 0–10 ) ° is an outlier ( marked by the red ring ) based on a single observed case . Inset: Histogram of angles of incidence as observed by Roberts ( upper panel of Fig . 13 in Roberts and Taylor ( 1982 ) ) . The red line is the PDF of vertex angles measured in our in vitro system ( Figure 1 at 60 min ) . The crossing probabilities and angles , experimental data and the distributions , are available in Figure 15—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 05310 . 7554/eLife . 19907 . 054Figure 15—source data 1 . Crossing probabilities and angles—data ( Roberts and Taylor , 1982 ) and distributions estimates . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 054 In addition to extracting the distribution of incidence angles for fasciculated neurites , Roberts et al . determined the probability for two neurites to cross ( rather than fasciculate ) . This crossing probability Π⁢ ( βinc ) was found to depend strongly on the incidence angle βinc ( Figure 15 ) . Using our analysis framework , we can quantitatively explain this observed dependence . A given pair of axons will not fasciculate ( zipper ) if their equilibrium zippering angle βeq is smaller than their initial incidence angle βinc . Any zippered segment formed in this situation would be unstable towards unzippering; the axons are therefore expected to cross while maintaining the initial incidence angle ( Figure 13B4 ) . Given that there is a distribution of equilibrium zippering angles in the network ( see the previous section ) , the probability that two randomly chosen axons with initial incidence angle βinc will cross ( rather than zipper ) equals the probability of their equilibrium angle βeq being smaller than the incidence angle βinc . This probability—the cumulative distribution function evaluated at βinc—is computed in Figure 15 , starting from the angle distribution taken from Roberts and Taylor ( 1982 ) . A good agreement with the crossing probabilities reported in Roberts and Taylor ( 1982 ) is seen . We thus successfully applied our framework to explain the network properties observed in the in vivo study of Roberts and Taylor ( 1982 ) , indicating that this framework is not limited to dynamics in culture . Using a combination of experimental observations and biophysical modeling , we showed that axon zippering arises from the competition of two principal forces: axon-axon adhesion and mechanical tension . The adhesion force favors an increase in the length of the zippered segment . The mechanical tension tends to minimize the total length of the axons , thus favoring unzippering . The relative strength of these two forces determines the vertex angle between axons in a zipper that reached static equilibrium . We used the BFP technique to measure the mechanical tensions of the OSN axons grown ex vivo , and obtained values ( interquartile range ( 529–833 ) pN ) comparable to tensions reported in the previous literature for PC-12 neurites grown in culture , on average around 650 pN ( Dennerll et al . , 1988 ) . Combining this information with measurements of the geometry of zippers in static equilibrium , we extracted the magnitude of the axon-axon adhesion force , obtaining approximately S≈100 pN ( with an upper bound on its spread , the interquartile range ( 52–186 ) pN ) . To our knowledge , this is the first experimental estimate of the force of adhesion between axon shafts , in any system . From the EM images in Figure 4 , we estimate that the fraction of circumference participating in contact between two axon shafts is in the range ( 15–35 ) % . Assuming that 25% of the circumference adhered and converting the adhesion force S to the adhesion energy per unit membrane area , one obtains 6 × 10−16Jμm2 . This is comparable to the energy density for E-cadherin-mediated cell-cell adhesion , which we estimate from the separation force measurements of Chu et al . ( 2004 ) to be ( 2 × 10−16 to 4 × 10−15 ) Jμm2 ( obtained as F3πR , where F is the separation force and R is the cell radius ) . We were able to unzip selected zippers by manipulating them with micropipettes , with consequent re-zippering after the manipulation was stopped . By fitting such induced zippering/unzippering dynamics to a basic biophysical model , we obtained an independent estimate of the axon-axon adhesion force , comparable with the estimate based on static observations . Comparing the shape of the observed zipper trajectories with the trajectories predicted by the dynamical model , we inferred that the zippering dynamics is limited by energy dissipation arising at the zipper vertex , and estimated the corresponding friction coefficient ηZ . Taking into account the typical zipper vertex velocity of ( 0 . 3–2 . 0 ) μmmin , the rate of energy dissipation during zippering or unzippering is of order ∼10−17Jmin . Our results provide a first systematic characterization of the statics and dynamics of individual axon zippers . The dynamical biophysical model that we developed ( see Materials and methods ) makes it possible to include axon zippering in mathematical models of axon guidance and bundling . Such studies previously focused on the dynamics of the growth cone , and modeled growth cone guidance by diffusible guidance cues ( Goodhill and Urbach , 1999 ) , the influence of tension forces and anchor points/focal adhesions on growth cone trajectory ( Li et al . , 1995; Nguyen et al . , 2016 ) , as well as contact interactions of growth cones with other axons ( Hentschel et al . , 1999; Chaudhuri et al . , 2011 ) . The previous modeling studies did not , however , consider the dynamics of axon shafts . In the previous literature , the tendency to fasciculate was often interpreted as arising from differential adhesion , that is , the growth cone having stronger adhesion to another axon than to the substrate ( Acheson et al . , 1991; Roberts and Taylor , 1982 ) . The structure of the small fascicles observed in our system , with axons travelling on top of each other ( Figure 4 ) , does indicate that adhesion between axon shafts is stronger than axon-substrate adhesion . We note , however , that for the zippering of axon shafts to occur , such differential advantage is in principle not necessary . Two axon shafts ( or axon fascicles ) adhered to the substrate can gain mutual adhesion energy by initiating zippering , while remaining adhered to the substrate ( as in the configurations shown in Figure 4B–C ) . In our biophysical analysis , we assumed that adhesion to the substrate was preserved during zippering , and we did not model the possible subsequent slower rearrangements ( involving loss of substrate contact for some axons ) in the internal structure of the zippered fascicle . A striking observation , in our culture system , is that the axon network , initially established as a complex network of individual axons or small bundles of axons , progressively coarsens in time , leading to the formation of large fascicles of axons . The coarsening persists over time scales of about 10 hr , which is much longer than the typical time scale for an individual zippering process ( 10 min ) . It is therefore unlikely that the coarsening is a result of protracted equilibration of zipper configurations under stationary force conditions . A possible explanation of the slow coarsening dynamics may lie in a gradual decrease of the average axon tension as the culture matures , which would lead to increasing domination of axon-axon adhesion forces over tension , hence favoring zippering . This proposal is supported by the observed sequence of stages of the culture maturation: axon elongation in the early stage , insignificant elongation in the intermediate stage and axon retraction in the final stage . According to Dennerll et al . ( 1989 ) , axonal elongation is possible only when the axon tension exceeds a certain threshold ( estimated as 1 nN for PC-12 axons ) ; in our system , the arrest of growth in the intermediate stage may thus have resulted from a decrease of tension below such threshold . Further indications of decreased tension are observed near the end of the intermediate stage , when some growth cones visibly loose their grip to the substrate ( see Video 4 ) , and hence can no longer generate axon tension by pulling ( Lamoureux et al . , 1989 ) . 10 . 7554/eLife . 19907 . 055Video 4 . Growth cone losing grip on the substrate . DOI: http://dx . doi . org/10 . 7554/eLife . 19907 . 055 To directly test if a decrease in average axon tension leads to coarsening , we used cytochalasin , a drug that was previously shown to significantly decrease the tension of PC-12 neurites ( Dennerll et al . , 1988 ) . Indeed , we found that when the drug was applied to a slowly coarsening or stabilized network , a marked increase in coarsening rate resulted within 30 min of the application ( Figure 6 ) . Apart from the more pronounced coarsening , cytochalasin did not change the structure of the network , and the axons did not become visibly slack; this is consistent with the expected reduction of tension in all directions in the evolving network , with enough tension remaining to keep the axons taut . While we did not measure the axon tension in cytochalasin-treated cultures , we did observed morphological changes ( the growth cone acquiring a stub-like shape with suppressed filopodia , and a reduction in number of axonal side processes ) consistent with previous studies ( Dennerll et al . , 1988; Letourneau et al . , 1987 ) in which cytochalasin-induced reduction of axon tension was assessed . Similarly to ( Letourneau et al . , 1987 ) , we also observed that in cytochalasin-treated cultures , the axons took a longer time ( 16 min in two experiments , compared to 6 and 10 min for untreated network ) to detach from the substrate and retract when exposed to trypsin , presumably because longer proteolysis of adhesion molecules is needed before the reduced tension can detach the axons ( Letourneau et al . , 1987 ) . As a second strategy aiming to perturb axon tension , we tested blebbistatin , an inhibitor of NMII , previously shown to decrease cell cortex/membrane tension in non-neuronal cells ( Fischer-Friedrich et al . , 2014; Ayala et al . , 2017 ) and to decrease tension generated in smooth muscle ( Ratz and Speich , 2010 ) . In growth cones of isolated DRG neurons , Sayyad et al . ( Sayyad et al . , 2015 ) showed that blebbistatin reduced the force exerted by lamellipodia , but surprisingly increased the force exerted by filopodia . Other studies found that the blebbistatin-dependent NMII inhibition may have opposite effects on axon extension , depending on the substrate . For example , in Ketschek et al . ( 2007 ) , while inhibition of NMII promoted peripheral dorsal root ganglia ( DRG ) axon extension on polylysine , it decreased axon extension on laminin . Hur et al . ( Hur et al . , 2011 ) , who observed a positive effect of blebbistatin on DRG axon extension on a laminin substrate ( thus the opposite result ) , discussed how these puzzling differences may be due to differences in laminin concentrations , or to the different adhesive properties of polylysine ( used in [Hur et al . , 2011] and in our study ) and polyornithine ( used by [Ketschek et al . , 2007] ) . These studies suggest that while blebbistatin decreases cell cortex contractility , its effect on axon shaft tension may depend on additional factors . In our model system , we did not observe an obvious effect of blebbistatin on axon tension or on axon extension . Further analyses would be required to explain why blebbistatin has a stabilizing effect on OSN axons grown in our experimental conditions . Since the biological agents having the ability to specifically stimulate axon growth or motility in our cultured explants are currently unknown , we tested FBS for this purpose , because of its established content of a variety of bioactive molecules and growth factors . It turned out that FBS did not have any significant boosting effect on OSN growth cones but , very interestingly , it induced the apparent contraction of the whole explant itself , leading to the generation of pulling forces on axons from the explant core . This apparent contraction of the explant is likely to be the result of changes in the shape of the individual cells constituting the explant , if such changes lead to an overall rounding of previously flattened or elongated cells . In line with this hypothesis , in ( Jalink and Moolenaar , 1992 ) , the serum induced a rapid rounding of cultured differentiating neural cells , an effect which may likely be due to lysophosphaditic acid , which by itself induces both cell rounding and neurite retraction ( Jalink et al . , 1993 ) . A quantitative estimate of the increase in axon tension due to the FBS-induced pull may be obtained from the measured stretch of the axons and the expected axon elongation stiffness . The measurements of ( Dennerll et al . , 1988 , 1989 ) showed the spring constant of PC-12 and DRG neurites ( which have baseline tension and length comparable to the axons in our system ) to be of order 100 pNμm . Assuming a similar axon stiffness for our system , the axon stretch of ∼15 μm in the FBS experiments ( Figure 5—figure supplement 1 ) is expected to generate a tension increase of over 1 nN . This is further supported by our tension measurements in the experiment of Figure 8—figure supplement 1 , where a pull of 6 μm was correlated with increase of tension by about 1 . 2 nN . The estimated FBS-induced tension increase of at least 1 nN ( possibly several nN ) is very significant compared to the typical axon tension ( under 1 nN ) that we recorded in the untreated networks . This tension increase , which was achieved within 20 min of the application of FBS , preceded a marked and rapid de-coarsening in parts of the network ( Figure 5F-H ) . We have thus shown on the network level that the extent of axon fasciculation can be regulated by changes in axon tension: an overall decrease in tension leads to zippering-driven coarsening , while an overall increase in tension leads to unzippering and de-coarsening . The estimated tension increase ( of order 1 nN ) generated endogenously by the FBS-induced explant pull is comparable in magnitude to active changes in axon tension demonstrated in previous literature , such as the tension recovery within 30 min after axon unloading in vitro ( Dennerll et al . , 1989 ) and in vivo ( Rajagopalan et al . , 2010 ) . Our results therefore demonstrate the control of axon zippering by tension changes of functionally relevant magnitude . As the axon network coarsens , the zippers become predominantly formed by axon fascicles , rather than by individual axons . We derived the expected distribution of tensions among the fascicles and combined it with the expected scaling of the fascicle-fascicle adhesion force ( proportional to the surface area of the fascicle ) to predict how the distribution of equilibrium zipper angles in the network depends on the mean number of axons per fascicle . This theoretical prediction was consistent with the observed relation between the median zipper angle and network coarsening , thus supporting the framework in which zippering is controlled by the competition between tension and adhesion not only for individual axons , but also on the fascicle level . A direct test of our theoretical models , be it on the single-zipper level or on the network level , would require a measurement of tension changes in the evolving network and following pharmacological manipulations . The tension generated by the pull of the growth cone may be efficiently assessed using traction force microscopy ( Style et al . , 2014 ) . In this technique , the net traction force is determined from the deformation of a suitable hydrogel substrate with embedded tracer beads , and corresponds to the axon tension directly behind the growth cone ( Koch et al . , 2012 ) . One may expect , however , that far behind the growth cone , where the zipper vertices are located , the axon tension can differ , due to force dissipation at substrate attachments along the axon and due to the tension generated directly within the axon shaft . A recently developed contact-less technique that may allow the monitoring of axon tension near the zipper vertex is thermal fluctuation spectroscopy ( TFS ) , adapted to transverse fluctuations of long protrusions . In Gárate et al . ( 2015 ) , this technique was used to obtain time-resolved estimates of axial tension in PC-12 neurites . Each tension measurement by TFS requires , however , to expose the axon to hundreds of short laser pulses , and relies on a fitting of the measurements to a phenomenological biomechanical model of the axon shaft , in order to extract combinations of viscoelastic parameters . A TFS procedure validated in our system would potentially allow the monitoring of increases or decreases in axon tension , with sufficient temporal resolution ( ∼10 s ) to correlate these tension changes with the dynamics of individual zippering events in the developing network . Topologically , the structure of the axon network observed in our study is similar to the structure of foams ( froths ) with a low liquid fraction . A froth consists of gas bubbles that are separated by liquid film walls , with surface tension in the walls ( Weaire and Hutzler , 2001 ) . The typical structure of a ‘two-dimensional froth’ ( obtained when the foam is restricted between two closely spaced glass plates ) is shown in Figure 13C . Similarly to the axon network ( Figure 1 ) , the structure is defined by vertices ( junctions ) at which three segments under mechanical tension meet . In froths ( Weaire and Hutzler , 2001; Glazier and Weaire , 1992 ) , the tensions in the three walls are equal , resulting in approximately 120° angles at the junction . In the axon network , the triple junctions are formed predominantly by axon zippers . At each junction , the mechanical tension in the zippered segment is necessarily the largest ( being the sum of tensions in the unzippered segments ) but is effectively lowered by the axon-axon adhesion force ( see Equation 1 ) ; the three angles between the segments are generally unequal . Closed loops that are formed by axon or fascicle segments are the topological equivalent of bubbles in the froth . In froths , the walls of a bubble with fewer than six sides are on average curved , making the bubble slightly rounded rather than polygonal; this is associated with a surplus of air pressure ( Weaire and Hutzler , 2001 ) . Gas diffusion out ( into the bubbles with lower pressure ) leads to the shrinking and disappearance of the rounded bubbles . This T2 process is illustrated for a three-sided bubble in Figure 13D1 , D2 . A second type of topological rearrangement ( denoted T1 in the literature ) observed in coarsening froths is shown in Figure 13D3 , D4 . In this process , two bubbles ( at top and bottom ) come into contact , pushing out the bubbles at left and right; the bubble walls are reconnected . Through a combination of T2 and T1 rearrangements , the froth coarsens ( Weaire and Hutzler , 2001; Glazier and Weaire , 1992 ) ; typically , the coarsening will proceed indefinitely ( until the size of the container is reached ) . The froth coarsening is not driven by changes in wall tension . No comparable coarsening mechanism , associated with pressure differences , exists in the axon network . However , we have repeatedly observed the shrinking and disappearance of loops formed by three axon/fascicle segments ( see Results and Figure 13A ) ; this is a topological analogue of the T2 process known from foam dynamics . The reconnection of bubble walls ( the T1 process ) has no topological analogue in the axon network . Rather , an attempt to implement such a process leads to a complete unzippering ( Figure 13E1 , E2 ) , which is topologically equivalent to the third elementary process of foam dynamics—a wall rupture ( Glazier and Weaire , 1992 ) . Such a process was observed in the axon network only infrequently , consistent with the overall predominance of zippering over unzippering . As explained in Results , the shrinking or expansion of a loop of axon segments can result from the zipper angles at the loop vertices being smaller than or larger than the equilibrium zipper angle , respectively . This is directly supported by our observations of expanding triangular loops in the experiments in which a pull was generated on the network ( Figure 5F-H ) . Basic geometry implies that on average , the zipper angle is 60° at vertices that form a triangular loop ( as in Figure 13A ) , and 90° at vertices that form an analogous rectangular loop . The equilibrium angles recorded from static zippers formed by single axons or small fascicles ( Figure 7C ) are predominantly below 90° , with the mean value of 51° . This suggests that at the early stages of network coarsening , rectangular loops ( with sides formed by single axons ) are unstable toward expansion , while triangular loops are closer to equilibrium and may either slowly retract or slowly expand . Such tendency may change in the later stages of network development , however , as axon fascicles can form loops with complex structure and modified equilibrium angles . A more detailed study would be required to characterize the stability of loops in the evolving axon network and its role in the coarsening process . Analogies between froths and biological tissues consisting of closely packed cells have been investigated in previous literature ( e . g . [Kafer et al . , 2007; Corson et al . , 2009] ) . Such analogies are complicated by two aspects: ( i ) the active cortical mechanics of the cells ( which make the interfacial tension dependent on the cell configuration [Kafer et al . , 2007; Manning et al . , 2010] ) , ( ii ) and restrictions on the volume of animal cells ( which prevent pronounced coarsening ) . However , in ( Corson et al . , 2009 ) , a growing plant tissue ( meristem of Arabidopsis thaliana ) was converted into a ‘living froth’ when oryzalin was used to depolymerize microtubules attached to the cell walls . In the resulting tissue , the topology and geometry of the cell interfaces was consistent with a typical froth , and pronounced coarsening of the structure was observed during the plant growth . The system we investigated—the zippering axon network—presents a remarkable example of an ex vivo system exhibiting both topological and dynamical analogies to froths . In our system , the coarsening is not limited by any cell volume restrictions and can proceed rapidly , on the time scale of hours . However , structural features such as complex loop configurations and entangled zippers can limit the final extent of coarsening . We report in the present paper that OSN axons grown on a planar glass substrate covered homogeneously with polylysine and laminin display extensive zippering behavior , raising the possibility that these axons , and more generally other types of axons with adhesive interactions , may zipper in vivo . While a direct imaging of zippering dynamics in vivo is difficult to achieve in mice with current methods , strong indirect evidence for axon zippering has been obtained in some other model organisms . In Xenopus embryos , the geometry of the sensory neurite plexus on the basal lamina of trunk skins shows features that are likely to be the result of axon zippering , as proposed by Roberts and Taylor ( Roberts and Taylor , 1982 ) and discussed in the last subsection of Results . This in vivo configuration has strong similarities to the ex vivo axonal network we studied—a result of the shared planar character and absence of obstacles to zippering in these two systems . In C . elegans , axon fascicles from the left and right ventral nerve cords fuse into a single fascicle if a specific medial interneuron is ablated at a late stage , when the axons have already reached their targets ( Aurelio et al . , 2002 ) . This ‘axon flip-over’ phenotype appears very likely to be due to axon shaft zippering , as evidenced by the abnormal fasciculation profiles observed between shafts which never fasciculate in non-manipulated animals . It was further shown that this phenotype was absent in immobilized animals , indicating that axon zippering was facilitated by mechanical forces exerted during the wriggling locomotion of the worm ( Aurelio et al . , 2002 ) . From a mechanistic point of view , the inhibition of zippering in wild-type animals is due to secretion by the medial interneuron of a 2-immunoglobulin-domain protein , which was proposed to bind and inhibit the activity of homophilic molecules expressed by the left/right contralaterally analogous axons ( Aurelio et al . , 2002 ) . This study nicely illustrates the ability of axon shafts to zipper in vivo , in this case with detrimental developmental consequences . It further provides a framework in which inhibition of axon-axon adhesion negatively regulates zippering . We next discuss the zippering potential of axons in the mammalian nervous system . In the developing neural systems , the high volume of extracellular space ( Lehmenkühler et al . , 1993 ) is highly compatible with zippering/unzippering of axons . Despite its reduction as the development proceeds , this extracellular space ( representing 15% of the volume of the adult cortical tissue [Korogod et al . , 2015] ) still provides a suitable environment for axon shaft dynamic interactions , as far as individual axons are not separated from each other by glial or parenchyme boundaries . Central and peripheral myelinated axons will obviously lose their zippering/unzippering abilities as soon as the myelination process begins . Similarly , unmyelinated axons within peripheral nerves , which become separated from each other by Schwann cell cytoplasmic processes as the nerves mature ( Elfvin , 1958 ) , will not be able to zipper thereafter . However , no such isolation by glial cells of unmyelinated axons occurs in the central nervous system , where shaft-shaft contacts persist in numerous areas . Our work with OE explants was motivated by the projection pattern of the in vivo olfactory system , where the growth of OSN axons from the OE toward the OB leads to the formation of fascicles . Interestingly , the type of glial cell associated to these axon bundles , the olfactory ensheathing cells ( OECs ) , never myelinate , cover nor even separate individual axons from each other: they instead wrap large bundles containing up to thousands of naked axons ( Li et al . , 2005 ) . Within these bundles , olfactory sensory axons are free to interact directly with each other , from early development onwards , because of the continuous production of OSNs throughout life ( Farbman , 1994 ) . One can thus speculate that the zippering we observed in vitro may occur and serve some functions in vivo . An appealing hypothesis is that the zippering of olfactory axons may participate in the sorting of olfactory axons . This sorting is indeed critical given the facts that each OSN expresses one odorant receptor ( OR ) gene , picked out of a large repertoire of roughly 1 , 000 OR genes , and that the axons of all OSNs expressing the same OR ( these OSNs are distributed across a large zone of the OE ) converge into a few glomeruli of the OB ( reviewed in [Mombaerts , 2006; Mori and Sakano , 2011] ) . This projection pattern results from a multistep process involving the regulated expression of adhesion and guidance cues , some of which under the control of ORs ( Key et al . , 2002; Nedelec et al . , 2005; Strotmann and Breer , 2006; Mombaerts , 2006; Nishizumi and Sakano , 2015; Zapiec et al . , 2016; Assens et al . , 2016 ) . In addition , an OR-independent pre-sorting of OSN axons , leading to the segregation of Class I vs . Class II OSN axon types within the nerve branches , has been reported ( Bozza et al . , 2009 ) . An OR-dependent sorting ultimately ensures that the axons of OSNs expressing the same OR are segregated from each other in purely innervated glomeruli . We did not carry out labeling to distinguish axon subtypes expressing for example specific adhesion molecules in our cultures , and therefore , we do not know if the dynamic axon-axon interactions we observed in vitro are related to any sorting process . Since it takes about 4 days for newborn OSNs to express their OR ( Rodriguez-Gil et al . , 2015 ) , the OSN growing in our cultures probably do not express their OR at the time point of our analyses , precluding that any OR-dependent sorting would occur in these cultures . It remains to be investigated if zippering of olfactory axons , in conjunction with OR-specific adhesion between axon shafts or OR-specific tension differences , may play a role in the axon sorting process in vivo . More generally , what are the expected functional consequences of regulated axon shaft zippering in vivo ? First , early on as a growth cone navigates toward its target , the ability to zipper can regulate the probability with which it would cross a fascicle or not , as we illustrated in Results with the Roberts and Taylor data ( Roberts and Taylor , 1982 ) . Second , once growth cones are already at distance on the way toward their target area , the extent to which their shafts are fasciculated may be regulated through zippering or unzippering . During both the development and maturation of neural networks , ephaptic interactions between axons may be favored in tightly fasciculated segments , thus influencing the synchrony of transmitted action potentials , or generating ectopic spikes ( Bokil et al . , 2001 ) . One could speculate that controlling the degree of fasciculation of axons through a regulation of zippering may be used to modulate such ephaptic interactions . Third , the resulting structure of the fascicles may have important consequences for subsequent steps of development and maturation of the networks . Indeed , while tightly fasciculated small bundles of pioneer axons constitute a robust path for follower axons , loosening the axons within fascicles might be beneficial for their myelination . Finally , in pathological contexts of axon regeneration following injury , or of axon demyelination , the unmyelinated axons or axon segments may zipper up in tracts . In tightly bundled tracts of partially demyelinated axons , ephaptic interactions are predicted to permit recovery of robust conduction ( Reutskiy et al . , 2003 ) . In light of our analysis , there are two principal ways by which zippering and hence the extent of fasciculation may be regulated in vivo . In a developing network , the growth cone activity or shaft cytoskeleton activity ( O'Toole et al . , 2015 ) can change the axon tension ( Rajagopalan et al . , 2010 ) and hence influence fascicle structure on fast time scales ( dozen minutes ) . On a slower time scale , fasciculation can be regulated by changes in CAM expression or their post-translational modifications . For example , it has long been established that axonal N-CAM is involved in axon-axon adhesion and regulated post-translationally by addition or removal of polysialic acid ( PSA ) ; high levels of PSA on N-CAM decrease cell-cell adhesion ( Hoffman and Edelman , 1983; Sadoul et al . , 1983; Rutishauser et al . , 1983 ) . Axon fasciculation is also regulated by external guidance cues , through a variety of signalling pathways . For example , matrix metalloproteases promote motor axon fasciculation in Drosophila ( Miller et al . , 2008 ) , secreted Slit2 promotes motor axon fasciculation via an autocrine and/or juxtacrine mechanism in the mouse embryo ( Jaworski and Tessier-Lavigne , 2012 ) , EphA4 expressed by otic mesenchyme cells regulates in a non-cell autonomous manner the spiral ganglion axon fasciculation in the mouse auditory system ( Coate et al . , 2012 ) , and Neuropilin1 mediates inter-axonal communications before and within the plexus region of the limb , thus regulating the fasciculation of sensory and motor projections ( Huettl et al . , 2011 ) . It remains to be established if and how these signals affect axon-axon adhesion or possibly axon tension . The tension of an axon shaft is influenced by the traction force generated by its growth cone , which in turn depends on the mechanical properties of its environment ( reviewed in [Athamneh and Suter , 2015] ) , likely through micro-scale elastic deformation of adhesion complexes between the axon actin network and the substrate ( Athamneh et al . , 2015; Mejean et al . , 2013 ) . In ( Koch et al . , 2012 ) , the growth cone traction force was found to increase with the substrate stiffness ( except for very rigid substrates ) ; a similar relation was found in non-neuronal cells ( Ghibaudo et al . , 2008; Yip et al . , 2013 ) ( reviewed in [Kerstein et al . , 2015] ) . Spatial changes in substrate stiffness may therefore regulate the distal axon shaft tension and hence the extent of zippering , potentially triggering fasciculation/defasciculation of a population of axons during development , when their growth cones arrive to a specific target area . In ( Koch et al . , 2012 ) , a ∼1 nN gradual increase in growth cone net traction force was recorded for DRG neurons plated onto a stiff substrate . In our experiments with FBS-induced explant pull , we observed marked defasciculation following an estimated tension increase of comparable magnitude ( see previous subsection ) . This suggests that significant changes in fasciculation may result from growth cone transitions between tissues with distinct elastic properties . Similarly , the general increase in stiffness of brain tissue during development ( Franze , 2013 ) may gradually increase the GC traction force and as a consequence facilitate unzippering and defasciculation as the growing tracts differentiate . In comparison , the substrate stiffness in our ex vivo experiments is more homogeneous and static , simplifying the zippering-driven dynamics . In conclusion , our work shows that adhesion-driven zippering of axon shafts can induce the formation of axon fascicles without a direct involvement of the growth cones . However , active changes in the pulling force at the growth cone , and hence in axon tension , may be used as a mechanism to control the extent of zippering and to regulate fasciculation/defasciculation . Mechanical tension has been shown to play important roles in neural development ( reviewed in [Franze , 2013; Franze et al . , 2013] ) , and recent studies have demonstrated that changes in axon tension can affect the formation of neural circuits by regulating neurite differentiation ( Lamoureux et al . , 2002 ) , axon branch survival ( Anava et al . , 2009 ) , as well as synaptic structure ( Siechen et al . , 2009 ) . Our work introduces a novel role of axon tension in neural circuit assembly: the regulation of fasciculation/defasciculation through the control of axon shaft zippering . All animal procedures were approved by the Île de France Ethics Committee . Pregnant female Swiss mice were sacrificed by cervical elongation at embryonic day 13 . 5 ( E13 . 5 ) , embryo were extracted from the uterus , and olfactory epithelium explants were prepared from the posterio-dorsal quarter of the septum and turbinates as follows . First , these posterior and dorsal parts of septa and turbinates were cut into pieces in L15 medium ( Gibco 21083 , Gibco Thermo Fisher Scientific , Waltham , Massachusets ) maintained on ice at 4°C , before being subsequently incubated for 30 min at 25°C in a solution of 1:1 of Trypsin 0 . 25% ( Gibco 25050 ) and Pancreatin 4X USP ( Gibco 02-0036DG ) to allow the OE to separate from the lamina propria . Enzymatic reactions were stopped by adding 10% Fetal Bovine Serum ( FBS , Gibco 10270 ) , and the biological material was rinsed in ice-cold L15 containing 5% FBS . Pieces of tissue were transferred into a glass Petri dish in which the OE sheets were cut , using a micro-scalpel , into small pieces of about ( 100–200 ) μm diameter each . Explants were then carefully transferred into 50 mm diameter IBIDI video dishes ( Biovalley , Illkirch , France ) that included a 35 mm glass coverslip ( for BFP experiments ) , or into IBIDI μ-slide eight well #1 . 5 polymer coverslip ( Biovalley 80826 ) ( for time lapse acquisition ) , previously coated with poly-L-lysine ( 0 . 2 mgml , Sigma P1524 , Sigma , St-Louis , Missouri ) and Laminin ( 0 . 02 mgml , Sigma L2020 ) , and maintained in culture ( 37°C , 5% CO2 ) until the day of experiment in a culture medium of DMEM/F12 ( Gibco 31331 ) containing 1% N2 ( Gibco 17502 ) , 0 . 1 mgml Gentamycin ( Sigma G1272 ) , 1 . 5% D-Glucose ( Sigma G8769 ) , 1% BSA ( Sigma A4161 ) and 7 μgml Ascorbic acid ( Sigma A4403 ) . We typically prepared and put in culture 40 to 60 explants per set of experiments coming from 10 to 12 embryos . 10 mM Hepes was added to the explant cultures 1 hr before starting time lapse acquisitions . In some experiments , the cultures were treated with FBS ( Gibco , 5% final concentration ) , blebbistatin ( Sigma B0560 , 10 μM in culture medium containing a final concentration of 0 . 1% DMSO ) , cytochalasin B ( Sigma C6762 , 2 μM in culture medium containing a final concentration of 0 . 1% DMSO ) , or trypsin ( Gibco 25050 , 0 . 25% in culture medium ) . Videomicroscopy was performed on a Leica DMI 6000B ( Leica , Wetzlar , Germany ) inverted microscope in a thermostated chamber ( 37°C , 7% CO2 at the rate 10 l/h , ( 87–95 ) % relative humidity ) using a DIC 63× NA 1 . 40 IMM , or a dry phase contrast objective 40× NA 0 . 75 Leica HCX PL APO , and a CCDcoolSNAP HQ2 camera ( Photometrics , Tucson , Arizona ) driven by Metamorph 7 . 1 , in a multiple acquisition mode . Typically , 9 Z steps with an interval of 1 μm were acquired each minute for each of the 8 to 10 positions chosen around explants . Recording of each experiment lasted 2 to 19 hr . The pool of recordings of network evolution contained 13 explants where no drug was added , 15 explants where cytochalasin was added ( pre-treated with blebbistatin in 11 cases ) , and 10 explant where FBS was added ( pre-treated with blebbistatin in five cases ) . Major criteria for selection for quantitative analysis were good contrast , culture survival , and sufficient area and density of the network . The reported data are based on analyses performed for ( i ) N=6 recorded experiments with no added drug ( originating from three individual animals ) , of which network coarsening did not occur in one experiment , which was therefore excluded from statistical analyses , and ( ii ) N=12 recorded experiments in which cytochalasin was applied ( originating from six individual animals ) , eight of which were pretreated with blebbistatin . The quantitative analysis was performed on video recordings in which the axonal network showed clear evolution lasting over 1 hr . Initial preprocessing and manual segmentation were performed using the distribution Fiji ( Schindelin et al . , 2012 ) of the project ImageJ ( Schneider et al . , 2012 ) . The field was cropped to restrict the region of interest and 6 to 10 images ( frames ) uniformly spaced in time were chosen from the course of the recording . The network of axons was then manually segmented by drawing individual selection lines over the image . In some cases , successive frames were consulted to decide whether a line is an axon to include or a transient side-process . The list of segmentation selections was exported to Matlab ( The Mathworks , Inc , 2015 ) , where a set of custom-made functions was used to: ( i ) convert the list of selection lines into a graph data structure , ( ii ) detect cordless loops in the graph , ( iii ) semi-automatically measure zipper angles or determine crossing points , ( iv ) calculate the network statistics ( notably the zipper angle distribution , total network length , number of vertices , average area of cordless loops ) , and ( v ) determine correlations between these statistics . The segmentation selections underlying the analysis shown in Figure 6G-E can be found in the Figure 6—source data 8 source file . They can be displayed in ImageJ by opening the TIFF file and the corresponding ZIP file with the segmentation; the segmentation can be laid over the image by checking the box ’Show All’ on the ’ROI Manager’ window . The Matlab script and input data which can be used to perform the fully automatic post-segmentation steps are provided in Figure 6—source data 1 . Segmentation coordinates for Figure 1 are provided in Figure 1—source data 1 . Regarding the data of dynamics of individual zippers , 17 events were measured by manually tracking the coordinates of the zipper vertex in consecutive frames . The measurement was accepted only if the vertex remained in static equilibrium 5 min before and after the transition . Zipper measurements were obtained from 14 explants originating from four mother animals . They were chosen from the networks of low density to minimize disturbance from the areas adjacent to the zipper . Scanning electron microscopy explants were cultured on a 14 mm diameter coverslip ( as described above ) , fixed for 1h at 4°C in 2% glutaraldehyde prepared in 0 . 1 M sodium cacodylate buffer , rinsed in cacodylate buffer , dehydrated in a series of graded ethanol baths , and dried using a critical point dryer ( Quorum Technologies CPD7501 , Laughton , UK ) . They were finally mounted on a carbon stub and sputter-coated . Observations were made using a Cambridge Instruments Stereoscan 260 scanning electron microscope equipped with a digital camera . The implementation of the BFP method ( Evans et al . , 1995 ) was adapted from ( Gourier et al . , 2008 ) . Basically , this method uses a force transducer composed of a biotinylated red blood cell ( RBC ) held by a glass micropipette ( treated with BSA ) , and a streptavidin-coated glass microbead ( 3 μm diameter ) , linked to the RBC by a streptavidin-biotin bond . In our experimental design , the bead was attached to axons of the culture previously treated for surface biotinylation . Within the range of forces measured in our experiments , the RBC force-deformation relation is linear , and the RBC behaves as a spring of stiffness k determined by the geometry of the probe and by the aspiration pressure Δ⁢P within the pipette: ( 5 ) k=RpΔPπ ( 1−R^p ) 1log⁡ ( 4R^cR^p ) − ( 1−14R^p−38R^p2+R^c2 ) where Rp ( 0 . 6–1 . 0 μm ) and RC ( 0 . 75–1 . 2 μm ) are the internal pipette radius and the radius of contact between the RBC and the bead respectively . The hat designates the corresponding radius divided by the radius of the aspirated unstrained RBC ( 2–3 μm ) . An adjustment of the pressure allows to set up the desired stiffness , k=100−400pNμm . By measuring the extension of the RBC , we could calculate the force exerted by the probe on an attached axon . For these experiments , 2 days in vitro ( DIV ) OE explants , cultured in 50 mm IBIDI dishes , were biotinylated using EZ-Link Sulfo-NHS-SS-Biotin ( Thermo Fisher Scientific 21328 , Waltham , Massachusets ) according to the manufacturer instructions . Dishes were then transferred into the thermostated chamber ( 37°C ) of the Leica DMIRB inverted microscope equipped with micropipette manipulators and a CCD digital camera ( purchased form JAI , Yokohama , Japan ) . Streptavidin beads were added to the culture , a micropipette ( 1 . 5–2 μm inner diameter ) was filled with the culture medium and fixed onto the mechanical micropipette manipulator ( Gourier et al . , 2008 ) . The diameter of the pipette was measured using the 40× objective and the CCD camera . Biotinylated RBCs were added to the culture medium ( Gourier et al . , 2008 ) . Then , a RBC was aspirated into the micropipette ( ΔP=200–250 Pa ) and put in contact for at least two 2 min with a bead attached to an axon or a small axon bundle . After an adhesive contact had been formed between the bead and the RBC , the pipette was slowly moved ( see Figure 8—source data 1 ) in order to pull or to push the axon ( s ) , being recorded by the CCD camera . In favorable cases , when the bead-axon contact adhesion was sufficiently strong , the pulling or pushing of the pipette lead to a deformation ( elongation or compression respectively ) of the RBC , often resulting in lateral deflection of the axon ( see Figure 8—source data 1 ) . After the pipette movement , we paused to let system relax; an equilibrium would be reached between the force induced by the probe and the transverse projection of the reaction force of the axon axial tension . Several steps of pulling or pushing were performed for each bead , gradually increasing applied force and axon deflection , until the bead detached from the RBC . The whole process was recorded on the CCD camera and the recording analyzed . The analysis was based on the captured recordings , recorded at rate of 65 fps . The BFPTool software package ( Šmít et al . , 2017 ) was used to subdivide each recorded video into intervals suitable for automated analysis , and then to track the pipette and bead position with sub-pixel precision . For the algorithms used , please refer to ( Šmít et al . , 2017 ) . The distance of the centre of the bead and a fixed point on the pipette tip were used to represent the length of RBC . The length of unloaded RBC was determined from a frame where the bead first touches the approaching RBC . Having established the RBC stiffness ( from geometry and pressure , using Equation 5 ) , the applied force was calculated for every frame . The force was corrected by a projection to the direction normal to manipulated axon . Then , the stable plateaux would be identified in the force time course , and for each , the average force Fi applied by the probe ( over the duration of the plateau i ) and the angle of axon deflection δi would be determined . Such pairs of values , ( 2⁢sin⁡δi , Fi⁢sin⁡ϕi ) , constituted our data points for each experiment ( ϕi represents the angle between pipette axis and axon axis ) . Finally , linear interpolation of the acquired data points was performed to obtain the tensile force within the axon as the slope of the interpolating line . Non-zero intercept of the interpolation line was often present; this happens when the selected reference distance does not truly correspond to the unstrained size of the RBC . With the slope determined from several plateaux ( we chose experiments having at least three ) , this offset does not influence the resulting calculated axonal tension . The uncertainty of the tension measurement has three sources . The uncertainty of stiffness of the BFP , δ ( k ) ≈14% , given by the limited precision of measurement of probe radii and aspiration pressure ( see Equation 5 ) . The uncertainty of RBC deformation measurement; while pipette pattern matching is generally very robust and precise , the tracking of the bead centre is more sensitive to perturbations and can introduce an error of ( 10–50 ) nm—see ( Šmít et al . , 2017 ) . Lastly , the most important source of measurement uncertainty is the deflection angle δi; the value of the deflection angle is small ( <5° ) while the precision of measurement is limited by diffuse edges of axons at Δ⁢δi≈0 . 5° . The change of angle is small between consecutive frames ( at 65 fps ) , so the precision can be improved by averaging several measurements , giving an upper limit on the relative error , δ⁢ ( δi ) ≤25% for the smallest angles measured . The axon tension is obtained by a linear regression of time-averaged quantities; the variability of applied force and deformation angle over the duration of each plateau is shown by error bars in Figure 8E . The final reported error of axon tension measurement is the standard deviation of the slope of regression . Each BFP experiment resulted in a value of tension and its experimental uncertainty for the given axon . This pair of parameters was used to construct the corresponding normal distribution , representing the tension of each axon . These Gaussian distributions were added and their sum normalized , to approximate the distribution of tensions within the whole axonal population . Similarly , the set of the measured equilibrium zipper angles ( described in Results ) was transformed into a distribution , by convolving the dataset with a Gaussian kernel ( using Matlab's kernel distribution functions ) . Two complementary approaches were used to estimate the value of axon-axon adhesion strength S: in the first , the measured distribution of tensions and the distribution of zipper angles are fully determined by each other , while in the second approach , the tensions and angles are treated as statistically independent variables . In the first approach , a fixed value of S was assumed , and Equation 1 was used to transform the measured distribution of tensions p⁢ ( T ) into a distribution of angles q⁢ ( β ) : ( 6 ) q ( β ) =p ( T ( β ) ) \|dβdT\|=p ( S2Φ ) S42−ΦΦ32 , whereΦ=1−cos⁡β2 where the relation between T and β is specified by Equation 1 . The correspondence ( evaluated as correlation ) between the distribution q⁢ ( β ) and the experimentally obtained angle distribution was maximal for S=88 pN ( correlation coefficient 0 . 813 ) . As shown in Figure 8—figure supplement 2A , the experimentally determined distribution was wider than the transformed distribution , suggesting that in reality the zippers do not all have the same value of S . In the second approach , we estimated an upper bound on the spread of S values . We constructed the joint distribution , Figure 8—figure supplement 2B , of tensions and angles as the product of the measured tension and angle distributions ( thus treating the tension and the angle as mutually independent ) , and computed the distribution of adhesion strengths S defined by this joint distribution and Equation 1 . To do so , the values of S were discretized in 1 pN bins and the probabilities of ( tension , angle ) pairs that gave S in the given bin were integrated . The resulting distribution of S is shown in Figure 8—figure supplement 2C . As the tensions and the angles are in reality expected to be partially dependent , the obtained interquartile range S= ( 52–186 ) pN should be viewed as the upper bound on the spread of S values . The obtained median ( 102 pN ) is consistent with the value of S obtained from the first approach . The Matlab scripts for performing the calculations described in this subsection are provided as source files associated with Figure 8—figure supplement 2 . For the general asymmetric axon zipper ( as in Figure 11A , with mobile vertex V and fixed points A , B , C ) , the static equilibrium condition and the equation of motion were derived as follows . We assumed that the vertex motion is sufficiently slow to allow the tension forces to keep the axon segments straight during zippering or unzippering; this is consistent with the experimental observations ( Figure 3 ) . Mechanical stresses were assumed to be uniform along each axon . We neglected elastic forces arising from axon bending in the immediate vicinity of the zipper; in zippers formed by single axons or small fascicles , the axons form a sharp bend ( Figure 3 ) , indicating that the bending rigidity is low . This assumption is further supported by the following quantitative arguments . Considering the flexural rigidity of a microtubule ( EI ) MT≲1×10−1nNμm2 ( Pampaloni et al . , 2006 ) , at most 10 microtubules in each axon ( Fadić et al . , 1985 ) , and a radius of curvature of the axons R≈1μm at the vertex , the density of energy of flexure can be estimated as 10⋅ ( EI ) MT2R2≲1×10−1nN ( [Roark et al . , 2002 , p . 127 ) , that is an order of magnitude lower than the axial tensile energy density ( the axon tension , see Results ) . For a bundle of axons , the bending energy is expected to scale quadratically with the number of axons , while the tension scales linearly . The two energy densities are therefore expected to become comparable in the vicinity of the zipper vertex only for bundles of ≳10 axons . We note that while the energy stored in the elastic flexure is neglected in our model , the energy dissipation resulting from the disruption of microtubule-associated cross-linking proteins and other bending-related structural changes is included in the empirical vertex-localized friction force introduced in Results . When formulating the dynamical model of zippering , we further assume that the tension in the constituent axons is constant in time . As the time scale for a simple zippering or unzippering process is of order 10–20 min ( see Results ) , one cannot in principle exclude active adjustments of axon tension accompanying the zippering or unzippering , or a coupling to active intracellular transport processes . In previous literature , a recovery of tension within 15–60 min was shown for axons that were made slack following a large rapid distension ( Rajagopalan et al . , 2010 ) . Compared to such distension experiments , however , the unzippering dynamics is gradual , and we assume no active tension regulation . Under these assumptions , the instantaneous zipper configuration is fully specified by the Cartesian coordinates ( x , y ) of the vertex . The total tension/adhesion energy of the configuration is given by ( 7 ) E ( x , y ) =T1 ( \|VA\|+\|VC\| ) +T2 ( \|VB\|+\|VC\| ) −S\|VC\| where \|V⁢X\| denotes the length of the given axon segment , T1 and T2 the values of ( effective ) tension in the two axons ( equivalent to the tensile energy per unit length ) , and S the energy of inter-axon adhesion per unit length of the adhered segment V⁢C . ( Treating the tensions T1 and T2 as constants independent of the axon length , we neglect possible Hookean elasticity contributions . ) The spatial gradient of the potential energy E defines the mechanical conservative force F→v that effectively acts at the vertex and drives the dynamics . The vector F→v thus points in the direction along which the energy decreases fastest upon a displacement of the zipper vertex . One straightforwardly obtains ( 8 ) F→v=−∇E ( x , y ) =T1VA^+T2VB^+ ( T1+T2−S ) VC^ where V⁢A^ indicates the unit vector in the V⁢A direction ( and similarly for V⁢B , V⁢C ) . The right-hand side of Equation 8 can be interpreted as the vector sum of the forces with which the axon segments V⁢A , V⁢B and V⁢C pull at the vertex . The last term , -S⁢V⁢C^ , in Equation 8 is the force of inter-axon adhesion , which has magnitude S and is always oriented anti-parallely to the zippered axon segment V⁢C . A zipper is in a static equilibrium when F→v=0→ . Spatial components of the force F→v can be conveniently expressed in terms of the zipper angles α1 and α2 ( see Figure 11A ) . In the direction along the zippered segment , the force equilibrium condition then becomes ( 9 ) − ( T1+T2−S ) +T1cos⁡α1+T2cos⁡α2=0 while in the perpendicular direction ( 10 ) T1sin⁡α1−T2sin⁡α2=0 . Given the parameters T1 , T2 and S , the Equations 9 , 10 specify the angles α1 and α2 , and hence the equilibrium vertex position ( x , y ) . It is readily shown that the equilibrium defined by Equations 9 , 10 is stable ( i . e . E⁢ ( x , y ) has a local minimum at the equilibrium point ) . In the special case of a symmetric zipper ( i . e . T1=T2 ) , Equation 10 implies α1=α2 and Equation 10 becomes equivalent to Equation 1 used in our static data analysis . A nonzero driving force F→v will result in motion of the vertex , with a velocity u→= ( x˙ , y˙ ) such that F→v is balanced by an effective frictional force acting at the zipper vertex . ( Expressed in terms of energy , the rate of change of E⁢ ( x , y ) when the vertex moves must equal the rate of energy dissipation in the entire zipper configuration—see Appendix ) . Assuming a frictional force proportional to the vertex velocity , the resulting equation of motion is ( 11 ) F→v=H↔ ( x , y ) u→ where the friction tensor H↔ is independent of u→ but may in general depend on the zipper configuration geometry , specified by the vertex position ( x , y ) . In the simplest case of isotropic and geometry-independent friction , H↔=c⋅1 is a constant multiple of unit tensor and the integration of Equation 11 results in a trajectory that follows the gradient of E⁢ ( x , y ) . In case of anisotropic and/or geometry-dependent friction , however , the vertex trajectory deviates from this path . The form of the vertex friction tensor depends on the dominant mechanism of energy dissipation . In the main text , we introduced two forms of internal energy dissipation in the axons—the viscosity of elongation/shortening , and the vertex-localized dissipation . As the corresponding frictional forces are collinear with the axon tension and with the axon-axon adhesion force , respectively , one may substitute the ‘dynamically corrected’ tension and adhesion magnitudes ( Equations 2 , 3 ) into the static equilibrium condition ( Equations 9 , 10 ) , and obtain two coupled equations for the zipper velocity components x˙ , y˙ . It is straightforward to see that in general , the vertex friction tensor resulting from either of these frictional forces is anisotropic . H↔ depends on the geometry ( i . e . the vertex position relative to the fixed points A , B , C ) in the case of elongation viscosity but is geometry-independent in the case of vertex-localized friction . A third form of energy dissipation—friction between the axons and the substrate—is evaluated in the Appendix . In this case , no simple prescription for generalizing the static equilibrium equation at the zipper vertex is available . The corresponding equation of motion is derived by integrating the energy dissipated along the axons , and equating the total rate of dissipative energy loss with the rate of gain of tension/adhesion energy . The Rayleigh dissipation function formalism is used for the unified treatment of all three forms of friction we consider .	As an animal develops , neurons within the nervous system connect with one another to form complex networks . Each neuron has a long cable-like protrusion known as an axon that establishes connections with other neurons . The axon has a structure called the growth cone at its tip , which navigates toward its target in response to signals produced by the surrounding tissues . Newly growing axons may bundle together or with previously grown axons , which helps them to move along a common path . Individual axons can later detach from the bundle to reach their specific target . It is generally thought that the growth cone controls axon bundling by latching on to the shaft of a neighboring axon and then moving along it . However , this viewpoint does not take into account possible dynamic adjustments in the adhesion of the shafts behind the growth cone . Šmít , Fouquet et al . have now grown neural explants taken from the nasal tissue of mouse embryos in the laboratory and used video microscopy to record how the axons grew . The growing axons formed progressively larger bundles without direct involvement from the growth cones . Instead , the shafts of the axons stuck together in a way that resembles fastening a zipper . Šmít , Fouquet et al . manipulated the ‘axon zippers’ and observed that zippering arises from a competition between two forces: the contact force that causes two axons to adhere to each other ( which favors zippering ) and the mechanical tension that arises from internal or external pulls on the axon ( which favors unzippering ) . More research is now needed to directly observe zippering in developing animals in order to understand how it helps the nervous system to assemble .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2017	Axon tension regulates fasciculation/defasciculation through the control of axon shaft zippering
HIV-1 assembles at the plasma membrane ( PM ) of infected cells . PM association of the main structural protein Gag depends on its myristoylated MA domain and PM PI ( 4 , 5 ) P2 . Using a novel chemical biology tool that allows rapidly tunable manipulation of PI ( 4 , 5 ) P2 levels in living cells , we show that depletion of PI ( 4 , 5 ) P2 completely prevents Gag PM targeting and assembly site formation . Unexpectedly , PI ( 4 , 5 ) P2 depletion also caused loss of pre-assembled Gag lattices from the PM . Subsequent restoration of PM PI ( 4 , 5 ) P2 reinduced assembly site formation even in the absence of new protein synthesis , indicating that the dissociated Gag molecules remained assembly competent . These results reveal an important role of PI ( 4 , 5 ) P2 for HIV-1 morphogenesis beyond Gag recruitment to the PM and suggest a dynamic equilibrium of Gag-lipid interactions . Furthermore , they establish an experimental system that permits synchronized induction of HIV-1 assembly leading to induced production of infectious virions by targeted modulation of Gag PM targeting . Human immunodeficiency virus type 1 ( HIV-1 ) particles assemble and bud at the plasma membrane ( PM ) of infected cells . This requires trafficking of the main structural component Gag from its site of synthesis at cytosolic polysomes to the inner leaflet of the PM . Here , Gag assembles into a multimeric lattice comprising ~2500 Gag molecules . Gag also recruits the other constituents of the infectious virion as well as components of the cellular endosomal sorting complex required for transport ( ESCRT ) to nascent budding sites ( reviewed in Sundquist and Kräusslich , 2012 ) . Gag consists of the individually folded MA ( matrix ) , CA ( capsid ) , NC ( nucleocapsid ) and p6 domains , which separate upon proteolytic maturation to the infectious virion . CA and NC mediate protein and RNA interactions during virion morphogenesis , while p6 recruits ESCRT components . PM targeting of Gag depends on the N-terminal MA domain . Mutational analyses revealed that the N-terminal myristoyl moiety and a surface exposed patch of basic residues in MA ( the highly basic region , HBR ) are important for Gag targeting , membrane association and virus formation ( Lorizate and Kräusslich , 2011; Freed , 2015 ) . However , these two features would not be sufficient to explain specific PM targeting . The phosphoinositide phosphatidylinositol 4 , 5-bisphosphate ( PI ( 4 , 5 ) P2 ) is known to act as a cue for specific recruitment of many peripheral PM proteins and was also shown to be important for Gag PM targeting . Depleting PI ( 4 , 5 ) P2 abolished PM targeting of Gag , while increasing PI ( 4 , 5 ) P2 at intracellular membranes redirected Gag to those sites ( Ono et al . , 2004 ) . Consistently , several studies reported that presence of PI ( 4 , 5 ) P2 enhances binding of Gag-derived proteins to synthetic liposomes in vitro ( Chukkapalli et al . , 2008 , 2010; Dick et al . , 2012; Olety and Ono , 2014; Mercredi et al . , 2016 ) , and PI ( 4 , 5 ) P2 appeared to be enriched in the HIV-1 lipidome ( Chan et al . , 2008 ) . Gag seems to dominate in a monomeric form or in small oligomers in the cytoplasm ( Kutluay and Bieniasz , 2010; Hendrix et al . , 2015 ) , and its N-terminal myristate group is proposed to be buried within the globular domain of MA at this stage ( Spearman et al . , 1997; Saad et al . , 2006; Tang et al . , 2004 ) . Gag interaction with PI ( 4 , 5 ) P2 has been suggested to trigger exposure of the myristate moiety followed by its insertion into the inner leaflet of the PM , thereby anchoring Gag . Recent models propose a regulatory role of RNA for this myristoyl switch ( Chukkapalli et al . , 2010; Alfadhli et al . , 2011; Chukkapalli et al . , 2013; Kutluay et al . , 2014 ) . In the cytosol , specific cellular tRNAs are bound to the basic region of MA in the absence of PI ( 4 , 5 ) P2 , thus preventing premature Gag association with intracellular membranes . This MA-tRNA interaction may be outcompeted by PI ( 4 , 5 ) P2 at the PM , thereby triggering the myristoyl switch . The NMR structure of HIV-1 MA bound to a PI ( 4 , 5 ) P2 molecule carrying truncated acyl chains revealed the 2´ acyl chain to be buried in the myristoyl-binding pocket of MA ( Saad et al . , 2006 ) . This observation suggested a model for Gag membrane anchoring , which was subsequently challenged by a recent report from the same group , however ( Mercredi et al . , 2016 ) . Upon arrival of Gag at the PM , the 2´ acyl chain of PI ( 4 , 5 ) P2 may be flipped outward and displace myristate from the acyl binding pocket of MA . Myristate in turn would insert into the PM . The consequent exchange of the unsaturated 2´ acyl chain of PI ( 4 , 5 ) P2 by the saturated myristic acid would increase local lipid saturation and membrane order . If occurring in all or a majority of the ~2500 tightly packed Gag molecules , this mechanism could also explain the observed liquid-ordered state of the HIV envelope ( Lorizate et al . , 2009; Brügger et al . , 2006; Chan et al . , 2008 ) . This model would further suggest that Gag clusters are stably anchored to the PM via their myristoyl moiety and PI ( 4 , 5 ) P2 would then be expected to be dispensable once anchoring has occurred . PI ( 4 , 5 ) P2 is clearly important for Gag PM targeting , but the dynamics of PI ( 4 , 5 ) P2 Gag interaction and the role of PI ( 4 , 5 ) P2 in later stages of HIV assembly have not been examined so far . To do this , we need to monitor Gag localization in real time while manipulating PI ( 4 , 5 ) P2 levels in living cells . Here , we made use of a recently developed reversible chemical dimerizer system ( abbreviated rCDS ) allowing rapid and controlled depletion and reconstitution of PM PI ( 4 , 5 ) P2 levels in living , virus-producing cells by a small molecule ( Feng et al . , 2014; Schifferer et al . , 2015 ) . Using this system , we showed that the nascent HIV-1 Gag assembly site is highly dependent on PI ( 4 , 5 ) P2 during the entire assembly process , and membrane association of apparently complete assembly sites remained PI ( 4 , 5 ) P2 dependent . PI ( 4 , 5 ) P2 removal from the PM not only abolished Gag PM targeting , but also caused dissociation of pre-assembled Gag clusters from the membrane . Reconstitution of bona fide Gag assembly sites at the PM was observed upon re-establishment of PM PI ( 4 , 5 ) P2 levels . Our findings are inconsistent with stable anchoring of large Gag clusters at the PM through myristoyl moieties alone and suggest a highly dynamic mode of Gag PM binding . Furthermore , this approach allows synchronization of assembly and release of infectious HIV-1 . These processes are usually highly asynchronous , both within a cell population and on the level of individual cells ( Ivanchenko et al . , 2009 ) . We first analyzed Gag assembly at the ventral PM in co-transfected HeLa cells that were subjected to PM PI ( 4 , 5 ) P2 depletion through rCD1 prior to Gag accumulation . For visualization of assembly sites in live cells , Gag tagged with EGFP was expressed in the viral context from a non-infectious subviral construct expressing all HIV-1 proteins except for Nef ( equimolar mixture of pCHIV and pCHIVEGFP [Lampe et al . , 2007] ) . Untagged Gag from proviral plasmids was detected in fixed cells by immunostaining . Confocal z-stacks covering the whole cell volume were acquired and analyzed in order to discriminate between assemblies at the ventral PM and punctuate signals arising from intracellular Gag clusters . Gag clusters at the PM corresponding to HIV-1 assembly sites were rarely observed when rCD1 was added at 4 hr after transfection ( Figure 1B , right ) whereas DMSO treated control cells showed a high number of Gag clusters ( Figure 1B , left ) . Quantification of Gag clusters at the ventral PM of DMSO ( n = 27 ) or rCD1 treated ( n = 26 ) cells revealed a pronounced ( 80% ) and highly significant ( p<0 . 0001 ) reduction of Gag clusters in PM PI ( 4 , 5 ) P2 depleted cells compared to control cells ( Figure 1F ) . Interaction of Gag with PI ( 4 , 5 ) P2 in vitro is mediated by the N-terminal MA domain and mutations in MA affect Gag PM localization . In order to gain further insight into the role of MA for the observed PI ( 4 , 5 ) P2-dependent Gag recruitment , we employed a Gag variant that retains the myristoylation signal and C-terminal cleavage site , but carries a deletion of the entire globular MA domain ( Δ8–126 SR ) . This protein has been shown to assemble both at the PM and at intracellular membranes of MT-4 cells and is competent for particle release ( Reil et al . , 1998 ) . As can be seen in Figure 1C , PM localization and assembly site formation of Gag ( Δ8–126 SR ) was not affected by PI ( 4 , 5 ) P2 depletion . Quantification of Gag clusters in DMSO ( n = 15 ) or rCD1 treated ( n = 18 ) cells showed no statistically significant difference in the number of Gag clusters at the ventral PM upon PI ( 4 , 5 ) P2 depletion ( p=0 . 8392 ) ( Figure 1F ) . These results provide direct evidence for the requirement of the globular MA domain for PI ( 4 , 5 ) P2 dependent Gag recruitment to the PM . Next , we analyzed the relevance of PM PI ( 4 , 5 ) P2 depletion for PM assembly site formation of two HIV-1 variants with mutations in the highly basic region of MA . These variants were previously characterized with respect to membrane interaction by in vitro liposome binding and membrane flotation assays . Changing MA residues K25 , K26 to T25 , T26 was reported to exhibit enhanced PI ( 4 , 5 ) P2 independent membrane binding ( Chukkapalli et al . , 2010 ) . Consistent with this in vitro observation , PM recruitment and assembly site formation of the 25/26KT variant was not significantly affected by PI ( 4 , 5 ) P2 depletion in live cells ( p=0 . 1242 ) ( Figure 1D and F ) . Changing MA residues K29 , K31 to E29 , E31 had been reported to cause complete loss of PI ( 4 , 5 ) P2 sensitivity with Gag mislocalization and impaired virus release; virus production was partially restored upon an additional change of MA residue E16 to K16 ( Tedbury et al . , 2015 ) . This correlated with enhanced membrane binding of the MA variant in vitro , but the additional mutation did not appear to restore PI ( 4 , 5 ) P2 dependence ( Tedbury et al . , 2015 ) . In agreement with these results , we observed that PI ( 4 , 5 ) P2 depletion did not have a statistically significant effect on assembly of 16EK 29/31KE Gag ( p=0 . 5948 ) ( Figure 1E and F ) . We then asked whether HIV-1 assembly site formation from cytosolic Gag molecules can be induced at the PM by restoring PI ( 4 , 5 ) P2 levels . Similar to the previous experiments , PI ( 4 , 5 ) P2 was depleted from the PM in rCDS expressing HeLa cells by rCD1 addition starting 4 hr after transfection . At 22 hr after transfection , Gag-EGFP was found to be mainly cytosolic with very few Gag assemblies at the PM ( Figure 2A , left panel ) . Addition of FK506 and consequent restoration of PM PI ( 4 , 5 ) P2 levels rapidly induced Gag cluster formation at the PM ( Figure 2A and C and Video 2 ) , whereas no major changes were observed upon addition of DMSO ( Figure 2B and C , Video 3 ) . After 90 min of FK506 treatment , the ventral PMs were covered with nascent HIV assembly sites and/or particles trapped between the cell and the coverslip ( Figure 2A ) . 10 . 7554/eLife . 25287 . 006Figure 2 . Gag assembly can be induced by PI ( 4 , 5 ) P2 reconstitution . ( A , B ) Representative time-lapse SDC fluorescence images of the ventral PM of HeLa Kyoto cells transfected with plasmids expressing the rCDS and HIV-1 derived constructs pCHIV and pCHIVEGFP . Cells were treated with 1 µM rCD1 at 4 hpt and PI ( 4 , 5 ) P2 rescue was induced at 22 hpt by addition of 1 µM FK506 ( A ) or 1% DMSO solvent control was added ( B ) . Arrows indicate Gag-EGFP clusters . Scale bar represents 20 µm . ( See also Videos 2 and 3 ) ( C ) Quantitative analysis of the increase in number of Gag-EGFP clusters at the PM following DMSO ( grey ) or FK506 ( black ) treatment . Error bars represent the standard error of the mean of n = 37 FK506 treated cells from four independent experiments and n = 16 DMSO treated cells from two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 00610 . 7554/eLife . 25287 . 007Figure 2—figure supplement 1 . The MA deletion mutant Δ8–126 SR does not respond to PI ( 4 , 5 ) P2 depletion . ( A , B ) Representative time-lapse fluorescence images of the ventral membrane of HeLa Kyoto cells transfected with the plasmids expressing the rCDS and HIV-1 MA deletion constructs pCHIVΔ8–126 SR and pCHIVEGFP ( Δ8–126 SR ) . Cells were treated with 1 μM rCD1 at 4 hpt and 1 μM FK506 ( A ) or 1% DMSO solvent control ( B ) was added at 22 hpt . Arrows indicate Gag-EGFP clusters . Scale bar represents 20 μm . ( C ) Quantitative analysis of the increase in number of Gag ( Δ8–126 SR ) clusters at the PM following DMSO ( grey ) or FK506 ( black ) treatment . Error bars represent SEM of data from n = 8 and n = 9 FK506 and DMSO treated cells , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 00710 . 7554/eLife . 25287 . 008Video 2 . Induction of HIV-1 assembly in PI ( 4 , 5 ) P2 depleted cells by PI ( 4 , 5 ) P2 reconstitution . Representative time-lapse SDC fluorescence microscopy sequence of the ventral PM of a HeLa Kyoto cell transfected with the rCDS and HIV-1 derived constructs pCHIV and pCHIVEGFP . Cells were treated with 1 μM rCD1 at 4 hpt , followed by addition of 1 μM FK506 at 22 hpt . The sequence was acquired at a time resolution of 1 min/frame and is displayed with 12 fps . See Figure 2A for corresponding still images . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 00810 . 7554/eLife . 25287 . 009Video 3 . Addition of DMSO solvent control to PI ( 4 , 5 ) P2 depleted cells does not induce Gag assembly . Representative time-lapse SDC fluorescence sequence of the ventral membrane of a HeLa Kyoto cell transfected with the rCDS and HIV-1 derived constructs pCHIV and pCHIVEGFP . Cells were treated with 1 μM rCD1 4 hpt , followed by addition of 1% DMSO solvent control at 22 hpt . The sequence was acquired at a time resolution of 2 min/frame and is displayed with 4 . 6 fps . See also Figure 2B for corresponding video still images . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 009 Analogous experiments performed with the MA deletion variant Δ8–126 SR confirmed dependence of the PI ( 4 , 5 ) P2 responsive phenotype on the MA domain of Gag . Assembly site formation at the PM was not impaired for this variant in PM PI ( 4 , 5 ) P2 depleted cells , and no significant difference was observed between cells subsequently treated with FK506 or DMSO ( Figure 2—figure supplement 1 ) . Deletion of the globular MA domain thus makes Gag independent of and unresponsive to PI ( 4 , 5 ) P2 . In order to determine whether the rate of Gag assembly differed between induced and native Gag assembly sites , we applied Total Internal Reflection Fluorescence ( TIRF ) microscopy , which had been used to track single nascent HIV assembly sites over time with low background and high time resolution ( Jouvenet et al . , 2008; Ivanchenko et al . , 2009 ) . Cells were co-transfected with the rCDS components and pCHIV derivatives and either left untreated ( Video 4 ) , or subjected to PI ( 4 , 5 ) P2 depletion at 4 hr after transfection followed by rescue of PI ( 4 , 5 ) P2 PM levels by FK506 addition at 22 hr after transfection ( Video 5 ) . Fluorescence intensity was recorded over time for individual Gag assembly sites . Figure 3A and B show normalized and averaged HIV-1 assembly traces for 175 native and 186 induced assembly sites from three different cells each . No obvious difference in assembly behavior was apparent from these data sets . For a quantitative comparison , data from each individual trace were fitted to a single exponential equation to extract the assembly rate constant k ( Figure 3C ) . Mean assembly rate constants derived from averaging all individual k values did not differ significantly between induced and native assembly , with k = 4 . 5 ± 0 . 4710−3 s−1 for native and k = 3 . 49 ± 0 . 2610−3 s−1 for induced assembly sites ( p=0 . 8563 ) and a similar distribution of rate constants ( Figure 3—figure supplement 1A ) . This translates into half-times of approximately 150 s and 200 s for native and induced assembly , respectively , or 8 . 5 min and 11 min for 90% completion of native and induced assembly , respectively . These values are in agreement with previously published data ( Ivanchenko et al . , 2009 ) . 10 . 7554/eLife . 25287 . 010Figure 3 . Kinetics of native and induced Gag assembly do not differ significantly . ( A , B ) Normalized and averaged HIV-1 assembly traces of 175 individual native ( A ) and 186 individual induced ( B ) assembly sites at the ventral PM of n = 3 cells each . The standard deviation is shown in grey , while a single exponential fit is shown in red . ( C ) Assembly rate constants derived from the individual assembly traces by fitting to single exponential equations . Mean assembly rate constants for native and induced assembly were 4 . 5 ± 0 . 4710−3 s−1 and 3 . 49 ± 0 . 2610−3 s−1 , respectively . Whiskers plots represent 5–95 percentile ( statistical significance was assessed with the Mann-Whitney U test; differences were considered significant when p≤0 . 05 ) . ( D ) Quantitative analysis of nascent native and induced Gag clusters ( imaged by high time-resolution TIRF microscopy ) forming at the ventral membrane over time . Numbers represent the mean amount of Gag-GFP clusters per 1000 µm² membrane . Error bars represent SEM . ( See also Figure 3—figure supplement 1 for relative frequency distributions of assembly rates and Videos 4 and 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 01010 . 7554/eLife . 25287 . 011Figure 3—figure supplement 1 . Relative frequency distributions of assembly rates do not differ between native and induced assembly . ( A , B ) Relative frequency distribution of native and induced assembly rates k extracted from HIV-1 assembly traces of 175 individual native ( left ) and 186 individual induced ( right ) assembly sites at the ventral membrane from n = 3 cells each , fitted to single exponential equations . The frequency distribution was plotted using a bin width of 0 . 0005 s−1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 01110 . 7554/eLife . 25287 . 012Video 4 . High time resolution TIRF imaging of native Gag assembly . Representative time-lapse TIRF microscopy sequence of the ventral PM of a HeLa Kyoto cell transfected with the HIV-1 derived constructs pcHIV and pCHIVEGFP . Cells were imaged at 22 hpt . The cell shows only few assembly sites in the beginning of the movie and progresses in assembly throughout the sequence . The sequence was acquired at a time resolution of 5 s/frame and is displayed with 140 fps . See Figure 3 , showing the assembly kinetics derived from this and additional movies . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 01210 . 7554/eLife . 25287 . 013Video 5 . High time resolution TIRF imaging of induced Gag assembly . Representative time-lapse TIRF microscopy sequence of the ventral PM of a HeLa Kyoto cell transfected with the rCDS and HIV-1 derived constructs pcHIV and pCHIVEGFP . Cells were treated at 4 hpt with 1 μM rCD1 for 90 min , followed by addition of 1 μM FK506 at 22 hpt . The sequence was acquired at a time resolution of 5 s/frame and is displayed with 140 fps . See Figure 3 , showing the assembly kinetics derived from this and additional movies . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 013 Visual inspection of the movies suggested that PM accumulation of assembly sites occurred more rapidly for induced compared to native assembly sites , while assembly rates at individual sites did not significantly differ between the two conditions . We therefore determined the number of native and induced PM assembly sites over time . Figure 3D shows that assembly sites accumulated faster and more synchronously when induced by PI ( 4 , 5 ) P2 restoration compared to the native situation . Formation of 100 new assembly sites per 1000 µm2 PM required only ~15 min for induced versus 45 min for native assembly . Live cell TIRF imaging yielded assembly rates , but did not provide information on the morphology of induced assemblies , since the size of HIV-1 particles is below the diffraction limit of conventional light microscopy . Previous studies using PALM/dSTORM super resolution microscopy had revealed a distinct HIV-1 assembly site architecture , with compact Gag clusters of 100–120 nm diameter surrounded by larger patches of Env glycoprotein molecules recruited to the budding site in a Gag dependent manner ( Lehmann et al . , 2011; Muranyi et al . , 2013 ) . In order to obtain insight into the morphology of induced assembly sites , we applied Stimulated Emission Depletion ( STED ) super resolution microscopy to analyze the distribution of Gag and Env at the ventral membrane of virus expressing cells with or without expression and induction of the rCDS . To allow for detection by STED nanoscopy , we employed Gag tagged with the stainable protein tag CLIP ( Gautier et al . , 2008; Hanne et al . , 2016 ) instead of the EGFP tag , while Env was detected by indirect immunofluorescence . Imaging at the ventral PM does not distinguish late virus budding structures from trapped extracellular particles , and we therefore refer to late HIV-1 assemblies here . As can be seen in Figure 4 , STED nanoscopy strongly increased resolution compared to confocal images and HIV-1 assembly subdomains became discernible ( compare confocal and STED panels ) . STED images revealed that neither the overall appearance of the cell nor the spatial distribution of Gag and Env at late assemblies differed notably between native and induced assembly . Both structures were characterized by a densely-packed Gag cluster with a diameter of about 120 nm , surrounded by a clustered accumulation of Env ( Figure 4 ) , in line with previous observations ( Muranyi et al . , 2013; Roy et al . , 2013 ) . The morphology of induced late assemblies was indistinguishable from native sites in micrographs ( compare Figure 4A and B ) as well as in averaged line profiles of individual assemblies ( compare Figure 4C and D ) . 10 . 7554/eLife . 25287 . 014Figure 4 . Native and induced HIV-1 assembly sites are indistinguishable by STED nanoscopy . ( A , B ) Confocal ( left ) and STED images ( right ) of the ventral PM of HeLa Kyoto cells transfected only with the HIV-1 derived constructs pCHIV and pCHIVCLIP ( native assembly , ( A ) ) or additionally with the plasmids expressing the rCDS ( induced assembly , ( B ) ) . In B , HIV-1 assembly was first inhibited by addition of rCD1 at 4 hpt and then induced by addition of FK506 for 90 min at 22 hpt . Gag ( cyan ) was detected via Atto 590 CLIP and Env ( red ) was detected via indirect immunolabeling . Some individual assembly sites are highlighted with arrows . Gag clusters ( filled arrowheads ) were surrounded by Env accumulations ( open arrowheads ) . Scale bar represents 1 µm ( overview images ) or 200 nm ( enlargements ) . ( C , D ) Averaged line profiles of selected native ( C ) or induced ( D ) Gag assembly sites ( n = 9 and n = 7 , respectively ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 014 While the previous experiments established that induced Gag assemblies appeared indistinguishable from their native counterparts , the recovery of infectious virus would be ultimate proof for synchronized induction of functional HIV-1 assembly . To this end , we established a stable HeLarCDS cell line constitutively expressing Anchor and Enzyme tagged with autofluorescent proteins . The presence of both components of the rCDS in all cells prior to transfection with the HIV-1 expressing plasmid prevents constitutive production of infectious virus from cells only transfected with the HIV-1 plasmid , but lacking one or both components of the rCDS , while an uneven distribution of components between cells is expected to occur when the three plasmids are cotransfected . As shown in Figure 5A , HeLarCDS cells constitutively express both fusion proteins , with the Enzyme detected in the cytoplasm and the Anchor at the PM . Next , we assessed functionality of the rCDS in these cells by expressing EGFP-PLCδ-PH , whose PM localization serves as indicator for PI ( 4 , 5 ) P2 . In the absence of rCD1 , the indicator was detected mainly at the PM ( Figure 5B , left panel ) . Upon rCD1 addition ( Figure 5B , middle panel ) , the Enzyme translocated to the PM ( see insert ) , resulting in relocalization of the indicator protein to the cytoplasm as a consequence of PI ( 4 , 5 ) P2 depletion from the PM . This effect was reversed upon addition of FK506 ( Figure 5B , right panel ) . 10 . 7554/eLife . 25287 . 015Figure 5 . Infectivity of virions produced under native and induced conditions in HeLarCDS cells . ( A ) Representative SDC fluorescence images of the central section of HeLarCDS cells expressing the Anchor and Enzyme . Single planes are shown . ( B ) HeLarCDS cells were transfected with pEGFP-PLCδ-PH and imaged 22 hpt ( left panel ) . Cells were treated with 1 µM rCD1 ( middle panel ) , followed by addition of 1 µM FK506 ( right panel ) . Maximum intensity projections of three focal planes acquired with an axial spacing of 0 . 5 µm are shown . Scale bars in A and B represent 20 µm . ( C , D ) HeLarCDS cells were transfected with pNL4-3 and treated with DMSO ( native ) or 1 µM rCD1 ( inhibited ) 5 hpt . ( C ) Relative RT activity as a measure of virus release was determined from supernatants harvested at 22 hpt . ( D ) Cells were subsequently treated with DMSO ( native ) or 1 µM FK506 ( induced ) . Infectivity on TZM-bl reporter cells was determined from supernatants harvested 2 hr after addition of DMSO ( native ) /1 µM FK506 ( induced ) and used to calculate normalized relative infectivity values . Data from two independent experiments [experiments ( 1 ) and ( 2 ) ] are shown . Please refer to Figure 5—figure supplement 1 for non-normalized data . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 01510 . 7554/eLife . 25287 . 016Figure 5—figure supplement 1 . Release , infectivity and relative infectivity of virions produced under native and induced conditions . ( A–C ) HeLarCDS cells were transfected with pNL4-3 ( native/induced ) . Untransfected cells were used as an additional control ( neg . ctr . ) . Cells were treated with DMSO ( native ) or 1 µM rCD1 ( induced ) 5 hpt , followed by treatment with DMSO ( native ) or 1 µM FK506 ( induced ) 22 hpt . RT activity as a measure of virus release ( A ) and infectivity on TZM-bl reporter cells ( B ) were determined from supernatants harvested before addition of DMSO or FK506 at 22 hpt ( 0 hr ) or 2 hr after addition of DMSO ( native ) or 1 µM FK506 ( induced ) ( 2 hr ) and used to calculate relative infectivity values shown in ( C ) . Data from two independent experiments [experiments ( 1 ) and ( 2 ) ] are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 016 After validation of the stable cell line , constitutive or induced virus release and formation of infectious HIV-1 was analyzed following transfection of HeLarCDS cells with the complete proviral plasmid pNL4-3 . rCD1 was added 4 hr after transfection . Comparing virus production from rCD1 and control cells at 22 hr after transfection showed a strong inhibition by PI ( 4 , 5 ) P2 depletion ( 80–85% reduction compared to control cells ) ( Figure 5C; see also Figure 5—figure supplement 1A for non-normalized data ) . Virus assembly was then induced in rCD1-treated cells by addition of 1 µM FK506 at 22 hr after transfection . Extracellular virus was collected 2 hr after addition of FK506 since a short harvest time limits the analysis mainly to virions formed by induced assembly . Virus production was restored after FK506 addition ( Figure 5—figure supplement 1A ) . Most importantly , relative infectivity of virions produced under inducing conditions was only mildly reduced ( ~65% compared to virions produced under native conditions from DMSO treated HeLarCDS cells ) ( Figure 5D; see also Figure 5—figure supplement 1B and C for non-normalized data ) . These results indicate that induced PM recruitment of Gag allows faithful assembly of infectious HIV-1 particles . Our results so far confirmed the essential role of PI ( 4 , 5 ) P2 for Gag recruitment to the PM and established an inducible system for assembly of infectious HIV-1 . To determine whether PI ( 4 , 5 ) P2 is also required to retain multimeric Gag assemblies at the PM , we performed reversible depletion of PM PI ( 4 , 5 ) P2 in cells displaying pre-formed HIV-1 assembly sites . Assuming insertion of the N-terminally attached myristic acid into the inner membrane leaflet for most or all Gag molecules in nascent assembly sites , PI ( 4 , 5 ) P2 would not be expected to play a role at this stage . Rapid and reversible PM PI ( 4 , 5 ) P2 depletion in living cells offers a unique opportunity to address this issue . HIV-1 expressing HeLa cells carrying the rCDS were incubated without compound addition until a large number of Gag assemblies had accumulated at the PM . Gag assemblies at the ventral PM ( towards the coverslip ) comprise nascent assembly sites as well as extracellular particles trapped between the cell and the substrate . These extracellular particles can obviously not be affected by rCD1 mediated PI ( 4 , 5 ) P2 PM depletion , and microscopy of the ventral PM is therefore not suitable for such an experiment . Instead , a central cell section was imaged in the subsequent experiments to focus on nascent assembly sites still connected to the cytosol . PI ( 4 , 5 ) P2 depletion was induced in cells with numerous Gag assembly sites by addition of rCD1 . Unexpectedly , the vast majority of cells ( 52 of 59 ) lost most pre-formed Gag assemblies from the lateral PM upon PI ( 4 , 5 ) P2 depletion ( Figure 6A and Video 6 ) . For quantitative analysis , we determined the relative number of Gag clusters over time following PI ( 4 , 5 ) P2 depletion in the cell shown in Figure 6A ( Figure 6C ) . This number decreased to 50% at ~30 min after rCD1 addition and reached a minimum level of ~30% . Averaging the relative amount of Gag clusters detected over time in n = 10 cells yielded a very similar result ( Figure 6E ) . In contrast , DMSO treated cells retained or slowly increased the number of Gag assembly sites at the lateral PM over time ( Figure 6C and E , Figure 6—figure supplement 1A , Video 7 ) , excluding any unspecific effect of the carrier or the imaging procedure . PM PI ( 4 , 5 ) P2 depletion also had no effect on preformed assembly sites for the Δ8–126 SR Gag variant ( Figure 6—figure supplement 1B and C ) , and this was confirmed by quantitative analysis of Gag ( Δ8–126 SR ) cluster number over time ( Figure 6—figure supplement 1D ) . 10 . 7554/eLife . 25287 . 017Figure 6 . The partially assembled Gag lattice dissociates reversibly upon PI ( 4 , 5 ) P2 depletion . ( A , B ) Representative time-lapse SDC fluorescence images of the central volume of HeLa Kyoto cells transfected with plasmids expressing the rCDS and HIV-1 derived constructs pCHIV and pCHIVEGFP . Maximum intensity projections of four focal planes acquired with an axial spacing of 0 . 5 µm are shown . Cells were treated at 22 hpt with 1 µM rCD1 for 90 min ( A ) and FK506 was added subsequently ( B ) . Arrows indicate Gag-EGFP clusters . Scale bar represents 20 µm . ( C , D ) Relative number of Gag clusters in the cell shown in ( A , B ) and the control cell shown in Figure 6—figure supplement 1A following rCD1 ( black ) or DMSO ( grey ) ( C ) and FK506 ( black ) or DMSO ( grey ) ( D ) addition , respectively . ( E , F ) Quantitative analysis of the mean relative number of Gag-EGFP clusters following rCD1 ( black ) or DMSO ( grey ) ( E ) and FK506 ( black ) or DMSO ( grey ) ( F ) addition , respectively . Error bars represent SEM for n = 10 rCD1+FK506 treated cells from six independent experiments and n = 13 DMSO treated cells from four independent experiments . ( See also Videos 6 and 7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 01710 . 7554/eLife . 25287 . 018Figure 6—figure supplement 1 . Addition of DMSO solvent control does not affect already assembled Gag clusters and the MA deletion mutant Δ8–126 SR does not respond to PI ( 4 , 5 ) P2 depletion and rescue . ( A–C ) Representative time-lapse SDC fluorescence images of the central volume of HeLa Kyoto cells transfected with the rCDS and HIV- 1 derived constructs pCHIV and pCHIVEGFP ( A ) or pCHIVΔ8–126SR and pCHIVEGFP ( Δ8–126 SR ) ( B , C ) . Maximum intensity projections of four focal planes acquired with an axial spacing of 0 . 5 μm are shown . Arrows indicate Gag-EGFP clusters . ( A ) Cells were treated at 22 hpt with 1% DMSO for 90 min , followed by addition of additional 1% DMSO . See Video 7 and Figure 6C and D for quantitative analysis of the number of Gag clusters over time in the cell shown in A . ( B ) Cells were treated at 22 hpt with 1 μM rCD1 for 90 min , followed by addition of 1 μM FK506 . ( C ) Cells were treated at 22 hpt with 1% DMSO for 90 min , followed by subsequent addition of 1% DMSO . Scale bar represents 20 μm . ( D , E ) Quantitative analysis of the mean relative number of Gag ( Δ8–126SR ) clusters following rCD1 ( black ) or DMSO ( grey ) ( D ) and FK506 ( black ) or DMSO ( grey ) ( E ) addition , respectively . Error bars represent SEM for data from n = 3 rCD1+FK506 treated cells and n = 4 DMSO treated cells . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 01810 . 7554/eLife . 25287 . 019Figure 6—figure supplement 2 . Influence of rate of particle release and new particle formation on number of Gag clusters . ( A–D ) Representative SDC fluorescence images of the ventral PM of HeLa Kyoto cells transfected with HIV-1 derived constructs pCHIV and pCHIVEGFP ( A ) , pCHIVΔvpu and pCHIVEGFPΔvpu ( B ) or the rCDS and HIV- 1 derived constructs pCHIVΔvpu and pCHIVEGFPΔvpu ( C , D ) . Single focal planes are shown . Dashed lines show Gag-positive cells . Cells were either left untreated ( A , B ) , treated with 1 µM rCD1 4 hpt ( C ) or treated at 22 hpt with 1 µM rCD1 for 90 min followed by 1 µM FK506 for additional 90 min ( D ) . Cells were fixed 24 . 5 hpt and Tetherin was stained via immunolabeling . Scale bar represents 20 µm . ( E ) Representative time-lapse SDC fluorescence images of the central volume of HeLa Kyoto cells transfected with the rCDS and HIV- 1 derived constructs pCHIVΔvpu and pCHIVEGFPΔvpu . Maximum intensity projections of four focal planes acquired with an axial spacing of 0 . 5 μm are shown . Arrows indicate Gag-EGFP clusters . Cells were treated at 22 hpt with 1 µM rCD1 for 90 min . ( F ) Quantitative analysis of the mean relative number of Gag clusters following rCD1 addition . Error bars represent standard error of the mean for n = 6 cells . ( G ) Representative time-lapse SDC fluorescence images of the central volume of HeLa Kyoto cells transfected with the HIV- 1 derived constructs pCHIV and pCHIVEGFP . Maximum intensity projections of four focal planes acquired with an axial spacing of 0 . 5 μm are shown . Arrows indicate Gag-EGFP clusters . Cells were treated at 22 hpt with 10 µg/ml cycloheximide ( CHX ) for 90 min . Scale bar indicates 20 µm . ( H ) Quantitative analysis of the mean relative number of Gag clusters following CHX addition . Error bars represent SEM for data from n = 47 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 01910 . 7554/eLife . 25287 . 020Figure 6—figure supplement 3 . G2AGag does not respond to PI ( 4 , 5 ) P2 rescue following PI ( 4 , 5 ) P2 depletion . ( A , B ) Representative time- lapse SDC fluorescence images of the central volume of a HeLa Kyoto cell transfected with the plasmids expressing the rCDS and HIV-1 derived constructs pCHIV and pGag-EGFP ( G2A ) . Maximum intensity projections of four focal planes acquired with an axial spacing of 0 . 5 μm are shown . Cells were treated at 22 hpt with 1 μM rCD1 for 90 min ( A ) , followed by addition of FK506 ( B ) . Arrows indicate Gag-EGFP ( G2A ) clusters . Scale bar represents 20 μm . Quantitative analysis of the relative number of Gag-EGFP ( G2A ) clusters over time following rCD1 ( C ) and subsequent FK506 ( D ) treatment . Quantitative analysis of the relative cluster size of Gag-EGFP ( G2A ) clusters over time following rCD1 ( E ) and subsequent FK506 ( F ) treatment . Quantitative analysis of the mean fluorescence intensity of Gag-EGFP ( G2A ) clusters over time following rCD1 ( G ) and subsequent FK506 ( H ) treatment . Error bars represent the SEM of data from n = 6 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 02010 . 7554/eLife . 25287 . 021Figure 6—figure supplement 4 . Kinetics of reinduced and native assembly at the lateral PM . ( A , B ) Normalized and averaged HIV- 1 assembly traces of 96 individual reinduced assembly sites ( A ) and 169 individual native assembly sites ( B ) at the lateral PM of n = 3 and n = 6 cells , respectively . The SD is shown in grey and a single exponential fit is shown in red . ( C ) Assembly rate constants k of native and reinduced assembly at the lateral PM extracted from each individual assembly trace by fitting to single exponential equations . Data for native assembly were taken from main Figure 3 and shown here in grey to allow for easier comparison . Mean assembly rate constants for native and reinduced assembly at the lateral PM were 5 . 9 ± 0 . 410−3 s−1 and 5 . 8 ± 0 . 610−3 s−1 . Whiskers plots represent 5–95 percentile ( statistical significance was assessed with the Mann-Whitney U test , **p≤0 . 001 ) . See also Video 8 for a representative high time-resolution SDC microscopy movie of reinduced assembly . ( D ) Relative frequency distribution of native ( lateral ) and reinduced assembly rates k extracted from HIV-1 assembly traces of 169 individual native ( left ) and 96 individual induced ( right ) assembly sites at the lateral membrane from n = 6 cells and n = 3 cells , respectively , fitted to single exponential equations . The frequency distribution was plotted after binning with a bin width of 0 . 0005 s−1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 02110 . 7554/eLife . 25287 . 022Video 6 . The partially assembled Gag lattice dissociates reversibly upon PI ( 4 , 5 ) P2 depletion . Representative time-lapse SDC fluorescence microscopy sequence of the central volume of a HeLa Kyoto cell transfected with the rCDS and HIV-1 derived constructs pcHIV and pCHIVEGFP . Maximum intensity projections of four focal planes acquired with an axial spacing of 0 . 5 μm are shown . Cells were treated at 22 hpt with 1 μM rCD1 for 90 min , followed by addition of 1 μM FK506 . The sequence was acquired at a time resolution of 4 min/frame and is displayed with 4 fps . See Figure 6A and B for corresponding still images . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 02210 . 7554/eLife . 25287 . 023Video 7 . Addition of DMSO solvent control does not affect already assembled Gag clusters . Representative time-lapse SDC fluorescence microscopy sequence of the central volume of HeLa Kyoto cells transfected with the plasmids expressing the rCDS and HIV-1 derived constructs pcHIV and pCHIVEGFP . Maximum intensity projections of four focal planes acquired with an axial spacing of 0 . 5 μm are shown . Cells were treated at 22 hpt with 1% DMSO for 90 min , followed by addition of additional 1% DMSO . The sequence was acquired at a time resolution of 2 . 5 min/frame and is displayed with 8 fps . See Figure 6—figure supplement 1A for corresponding still images . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 023 The number of Gag assembly sites at the plasma membrane at a given time reflects the equilibrium between assembly site formation and loss of assembly sites , either by extracellular release of budded virions or by dissociation of pre-formed assembly structures from the PM into the cytosol . To assess the contribution of virus release to the observed reduction of PM assemblies following PI ( 4 , 5 ) P2 depletion , we performed the depletion experiment with a Vpu-defective HIV-1 derivative . Vpu counteracts the host restriction factor tetherin ( Figure 6—figure supplement 2A and B ) . In the absence of Vpu , tetherin retains fully budded HIV-1 particles at the PM and thereby inhibits virus release ( Neil et al . , 2008; Van Damme et al . , 2008 ) . High amounts of PM tetherin in cells producing Vpu ( - ) particles should therefore largely prevent virus release , while removal of assembly sites from the PM to the cytosol would be unaffected . Control experiments confirmed that neither the presence of the rCDS nor its activation affected the localization of tetherin ( Figure 6—figure supplement 2C and D ) . Importantly , the loss of Gag clusters from the PM upon PI ( 4 , 5 ) P2 depletion was unaffected by lack of Vpu ( Figure 6—figure supplement 2E and F ) , supporting the conclusion that Gag from nascent assembly sites re-localizes to the cytosol upon PI ( 4 , 5 ) P2 depletion . The observed reduction of Gag clusters at the PM upon PI ( 4 , 5 ) P2 depletion could also be caused by constitutive loss of Gag from the PM , which is normally overcome by PM binding of newly synthesized Gag molecules . In this case , one would expect a reduction of preformed assembly sites when blocking synthesis of new proteins by addition of cycloheximide ( CHX ) . Adding CHX to cells with established Gag clusters did not affect the appearance or number of Gag assemblies over up to 90 min ( Figure 6—figure supplement 2G and H ) , suggesting that loss of Gag assemblies from the cell membrane is caused by PI ( 4 , 5 ) P2 dependence of their PM association . We next asked whether Gag molecules that had been dissociated from PM assembly sites by PI ( 4 , 5 ) P2 depletion could form new assembly sites upon restoration of PM PI ( 4 , 5 ) P2 . For this experiment , we treated cells displaying pre-formed assembly sites with rCD1 for 90 min as described above , followed by addition of FK506 to allow PM PI ( 4 , 5 ) P2 reconstitution . FK506 treatment yielded a rapid increase in the number of Gag assembly sites at the PM ( Figure 6B , Video 6 ) . The number of Gag clusters reached the initial level ( prior to rCD1 mediated PI ( 4 , 5 ) P2 depletion ) at 30–40 min after addition of FK506 ( Figure 6D and F ) . No effect on the number of Gag clusters was observed in control cells treated with DMSO ( Figure 6—figure supplement 1A , Figure 6D and F , Video 7 ) or when the PI ( 4 , 5 ) P2 independent MA variant Δ8–126 SR was used ( Figure 6—figure supplement 1B , C and E ) . Visual inspection of movie sequences indicated that newly induced assembly sites generally did not appear at those positions where previously dissociated sites had been observed . Quantitative analyses indicated that accumulation of reinduced HIV-1 Gag assemblies following a cycle of PI ( 4 , 5 ) P2 depletion and restoration ( Figure 6D and F ) occurred twice as fast as initial accumulation of budding sites under induced conditions ( compare Figure 2C ) . To analyze whether reinduced Gag assembly sites at the PM upon PI ( 4 , 5 ) P2 reconstitution comprised only newly synthesized Gag molecules or whether ‘recycled’ molecules previously depleted from the membrane contribute , we blocked protein translation by cycloheximide ( CHX ) treatment 60 min prior to rCD1 addition . As shown in Figure 7A and C , PI ( 4 , 5 ) P2 depletion resulted in loss of clusters from the PM that was fully reverted by restoring PI ( 4 , 5 ) P2 levels in the presence of CHX ( Figure 7B and D ) . These data indicate that Gag molecules synthesized before assembly site dissociation are sufficient for formation of new assembly sites following PI ( 4 , 5 ) P2 reconstitution and new protein synthesis is not required . 10 . 7554/eLife . 25287 . 024Figure 7 . Assembly of Gag can be reinduced in the presence of cycloheximide . ( A , B ) Representative time-lapse SDC fluorescence images of the central volume of a HeLa Kyoto cell transfected with plasmids expressing the rCDS and HIV-1 constructs pCHIV and pCHIVEGFP . Maximum intensity projections of four focal planes acquired with an axial spacing of 0 . 5 µm are shown . Cells were pre-treated at 21 hpt with 10 µg/ml CHX for 60 min . Subsequently cells were treated with 1 µM rCD1 for 90 min ( A ) , followed by addition of FK506 ( B ) . Arrows indicate individual Gag-EGFP clusters . Scale bar represents 20 µm . ( C , D ) Relative number of Gag-EGFP clusters over time following rCD1 ( C ) or FK506 ( D ) treatment . Error bars represent SEM of data from n = 23 cells from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 024 Based on these observations , we asked whether the more rapid accumulation of assembly sites upon dissociation and reinduction may be explained by multimeric Gag lattice remnants following assembly site dissociation . To address this issue , we employed a Gag variant lacking the N-terminal myristoylation site ( Gag G2A ) . This variant cannot assemble at the PM ( Bryant and Ratner , 1990; Göttlinger et al . , 1989 ) , but can be rescued by a low amount of co-expressed wild-type Gag ( Park and Morrow , 1992; Morikawa et al . , 1996 ) . Accordingly , co-expression of Gag-EGFP ( G2A ) and wild-type Gag ( in 2:1 molar ratio ) led to formation of fluorescent assembly sites at the PM ( Figure 6—figure supplement 3A ) of cells expressing the Anchor and Enzyme , while Gag-EGFP ( G2A ) alone failed to assemble . As expected , addition of rCD1 resulted in a loss of Gag-EGFP ( G2A ) clusters from the PM ( Figure 6—figure supplement 3A and C ) . However , only few Gag-EGFP ( G2A ) clusters reformed at the PM and the quantitative analysis revealed , that Gag-EGFP ( G2A ) was not efficiently re-recruited to the PM upon PI ( 4 , 5 ) P2 rescue ( Figure 6—figure supplement 3B and D ) , while wild-type Gag readily formed new assembly sites under these conditions ( Figure 6 ) . This result is consistent with complete dissociation of Gag clusters following PI ( 4 , 5 ) P2 depletion . Quantitative analysis revealed that neither the size nor the fluorescence intensity of residual clusters increased following PI ( 4 , 5 ) P2 restoration ( Figure 6—figure supplement 3E–H ) , arguing against recruitment of Gag ( G2A ) molecules to residual assembly sites . To compare the dynamics of assembly site formation for induced and reinduced assembly , we analyzed assembly rates following PI ( 4 , 5 ) P2 depletion and reconstitution . Spinning disk confocal microscopy was performed at high time resolution in a 3D volume to identify and track single assembly sites at the lateral PM ( Video 8 ) . Figure 6—figure supplement 4A shows normalized and averaged traces of 96 individual reinduced assembly sites from three cells . 10 . 7554/eLife . 25287 . 025Video 8 . High time resolution SDC imaging of reinduced Gag assembly . Representative time-lapse SDC fluorescence microscopy sequence of the central volume of a HeLa Kyoto cell transfected with the plasmids expressing the rCDS and HIV-1 derived constructs pcHIV and pCHIVEGFP . Maximum intensity projections of four focal planes acquired with an axial spacing of 0 . 5 μm are shown . Cells were treated at 22 hpt with 1 μM rCD1 for 90 min , followed by addition of 1 μM FK506 . The sequence was acquired at a time resolution of 6 s/frame and is displayed with 100 fps . See Figure 6—figure supplement 4 , showing the assembly kinetics derived from this and additional movies . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 025 For a quantitative comparison , data from each individual trace were fitted to a single exponential equation to extract the assembly rate constant k ( Figure 6—figure supplement 4C ) . The mean assembly rate constant of k = 5 . 8 ± 0 . 610−3 s−1 for reinduced assembly , which was derived from averaging all individual k values , was slightly , but significantly , increased compared to native assembly ( native: k = 4 . 5 ± 0 . 510−3 s−1 , p<0 . 0001 ) ( Figure 6—figure supplement 4C ) . The assembly constant for reinduced assembly translates into a half-time of ~120 s ( native: t1/2 ~150 s ) . Native and induced assembly kinetics were both determined at the ventral membrane , while reinduced assembly had to be measured at the lateral membrane , and this difference could influence the observed assembly rates . We therefore determined assembly kinetics of native Gag assembly site formation at the lateral membrane as well . Figure 6—figure supplement 4B shows normalized and averaged traces of 169 individual native assembly sites at the lateral PM of six cells . The average assembly constant derived from single exponential fits for all native assembly traces under these conditions of k = 5 . 9 ± 0 . 410−3 s−1 ( Figure 6—figure supplement 4C ) was very similar to that observed for reinduced assembly ( no significant change , p=0 . 1569 ) and was again significantly faster than observed for assembly at the ventral membrane ( p<0 . 0001 ) . The half-time of native assembly at the lateral PM was changed accordingly ( t1/2 ~115 s ) . The observed difference may reflect effects of the adjacent glass substrate on the kinetics of HIV-1 Gag assembly at the ventral membrane , but clearly shows that reinduced and native assembly site formation occur with very similar kinetics . The distribution of rate constants also did not reveal any major difference between reinduced and native assembly ( Figure 6—figure supplement 4D ) . Finally , we asked whether the reinduced HIV-1 Gag assemblies were morphologically normal using STED nanoscopy . Since depletion from the ventral PM was inefficient , we again focused on the lateral PM for these analyses . As shown in Figure 8A , late native HIV-1 assemblies at the lateral PM appeared very similar to those detected at the ventral PM ( compare Figure 4 ) . In both cases , condensed Gag clusters with a diameter of roughly 120 nm surrounded by larger Env clusters were detected in untreated cells . Reinduced assembly sites also displayed this distinct phenotype ( Figure 8B ) and were undistinguishable from native assembly sites at the lateral PM with respect to diameter and morphology in micrographs and line plot diagrams of Gag and Env ( Figure 8C and D ) . 10 . 7554/eLife . 25287 . 026Figure 8 . Native and reinduced HIV-1 assembly sites are indistinguishable by STED nanoscopy . ( A , B ) Confocal ( left ) and STED images ( right ) of the lateral PM of fixed HeLa Kyoto cells expressing only the HIV-1 constructs pCHIV and pCHIVCLIP ( native assembly , ( A ) ) or in addition the plasmids expressing the rCDS ( reinduced assembly , ( B ) ) . In B , dissociation of assembled Gag clusters was induced by addition of 1 µM rCD1 for 90 min at 22 hpt; subsequently Gag assembly was reinduced by addition of 1 µM FK506 for 90 min . Gag ( cyan ) was detected via Atto 590 CLIP and Env ( red ) was detected via indirect immunolabeling . Some individual HIV-1 assembly sites are highlighted by arrows . Gag clusters ( filled arrowheads ) were surrounded by Env accumulations ( open arrowheads ) . Scale bar represents 1 µm ( overview images ) or 200 nm ( enlargements ) . ( C , D ) Averaged line profiles of selected native ( C ) or induced ( D ) Gag assembly sites ( n = 7 and n = 6 for native and reinduced assembly sites , respectively ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 25287 . 026 PM association of HIV-1 Gag - and consequently assembly site formation and virus production - have previously been shown to depend on PM PI ( 4 , 5 ) P2 . Here , we made use of a recently established chemical biology tool allowing rapid and reversible manipulation of PI ( 4 , 5 ) P2 levels at the PM in living cells ( Feng et al . , 2014; Schifferer et al . , 2015 ) to study the role of PI ( 4 , 5 ) P2 for HIV-1 assembly site formation and maintenance . This rCDS has several advantages compared to alternative strategies . First , PI ( 4 , 5 ) P2 depletion is strictly dependent on rCD1 addition and can be rapidly achieved within minutes . This allows maintaining normal cell metabolism prior to compound addition despite the presence of all rCDS components in the cell , and should thus prevent induction of escape pathways . Second , the system is completely and rapidly reversible , allowing restoration of PM PI ( 4 , 5 ) P2 levels within minutes of FK506 addition . Here , we show that long-term depletion of PM PI ( 4 , 5 ) P2 is compatible with cell viability and viral protein expression , while completely preventing Gag membrane targeting and assembly site formation . Furthermore , PI ( 4 , 5 ) P2 depletion also caused loss of pre-assembled Gag lattices from the PM and large clusters removed from the membrane appeared to dissociate into monomers or small oligomers in the cytosol . Restoration of PM PI ( 4 , 5 ) P2 in these cells reinduced assembly site formation even in the absence of new protein synthesis indicating that the dissociated Gag molecules remained assembly competent . These results reveal an important role of PI ( 4 , 5 ) P2 beyond Gag recruitment to the PM and suggest that Gag interactions with PM lipids may be more dynamic than previously thought . PI ( 4 , 5 ) P2 and MA-dependent PM targeting of Gag and assembly site formation is consistent with previous studies showing that continuous overexpression of 5Ptase causes loss of Gag targeting in Gag-producing cells , that Gag-derived proteins preferentially interact with PI ( 4 , 5 ) P2 in vitro in an MA-dependent manner and that PI ( 4 , 5 ) P2 was found to be enriched in virions ( Ono et al . , 2004; Saad et al . , 2006; Chukkapalli et al . , 2008; Chan et al . , 2008 ) . Applying an inducible system in the current study allowed us to investigate the kinetics of assembly site formation . Our results show that rate constants for native and induced assembly , and even for reinduced assembly following depletion and subsequent restoration of PI ( 4 , 5 ) P2 , are virtually identical . Given that a large pool of myristoylated and assembly competent Gag is present in the cells prior to induction ( or reinduction ) , one may have expected assembly to occur faster than under native conditions , but this was not the case . This observation is consistent with the assumption that PI ( 4 , 5 ) P2 is important to retain Gag once it reaches the PM , but does not affect its PM trafficking , which may be rate-limiting under these conditions . Interestingly , Gag assembly site formation was not faster , but more synchronous under induced or reinduced assembly conditions compared to native assembly . This finding most likely reflects the presence of a large pool of assembly competent Gag in PI ( 4 , 5 ) P2 depleted cells , which can only be retained at the PM upon PI ( 4 , 5 ) P2 restoration . Gag molecules may transiently associate with all cellular membranes including the PM in the absence of PI ( 4 , 5 ) P2 , but are stably retained only in the presence of this phosphoinositide . Assembly site formation therefore is synchronized by the rCDS , which provides a promising tool to study the role of host components during HIV-1 assembly and the induction of virus maturation . Analysis of these processes is so far limited by the inherently asynchronous nature of virus production – between cells and within an individual cell – and synchronizing this process via the described approach can overcome this limitation . The observation that virions released under those conditions are infectious and the opportunity to use stable HeLarCDS cells for bulk assays broadens the range of applications for this system . The finding that PI ( 4 , 5 ) P2 is also required to retain partially or completely assembled Gag lattices at the PM was unexpected , since a triggered myristoyl switch in conjunction with ionic interactions between the HBR and negatively charged lipids of the inner leaflet would be expected to retain Gag at the PM , especially considering the high avidity of the multimeric Gag lattice ( ca . 2 , 500 Gag molecules per assembly site ) . Large PM Gag clusters resembling fully assembled HIV-1 budding structures were rapidly and completely lost from the PM upon PI ( 4 , 5 ) P2 depletion , however . Thus , PI ( 4 , 5 ) P2 is clearly required throughout assembly and budding of HIV-1 particles and not only for Gag membrane targeting . In vitro , MA binding to liposomes is significantly weaker for PI ( 4 ) P compared to PI ( 4 , 5 ) P2 ( Mercredi et al . , 2016 ) , suggesting that Gag dissociation from the PM may be directly caused by dephosphorylation of PI ( 4 , 5 ) P2 and consequent loss of Gag interaction . It is not entirely clear , however , whether Gag-associated phosphoinositides are good substrates for the phosphatase since the enzyme would have to enter the tight protein lattice of the assembling virion . Accordingly , effects of general depletion of PM PI ( 4 , 5 ) P2 with lipid exchange of Gag molecules must also be considered . Our results are not easily reconciled with the suggested lipid flip , where the 2´acyl chain of PI ( 4 , 5 ) P2 would be pulled from the membrane and exchanged for the myristic acid once the myristate switch is triggered . As discussed by Summers and colleagues ( Mercredi et al . , 2016 ) , extrusion of the 20-carbon 2´-acyl chain of PI ( 4 , 5 ) P2 would require ~24 kcal/mol ( Tanford , 1979; Israelachvili et al . , 1977 ) , while inserting 10 carbons of myristate into the lipid bilayer would gain only 8 kcal/mol . More importantly , our results appear inconsistent with the continuous insertion of the myristate moiety of all or the majority of Gag molecules into the inner leaflet . While the binding energy of the myristate anchor is not sufficient to retain individual Gag molecules at the PM , the avidity effects of up to 2500 Gag molecules with all or the majority of their N-terminal myristate groups inserted into the inner leaflet should make membrane association of the Gag lattice almost irreversible , independent of PI ( 4 , 5 ) P2 . Given that the binding energy of myristate interaction with MA is the same as for extrusion of myristate from the PM ( 8 kcal/mol ) ( Charlier et al . , 2014 ) , we suggest a dynamic equilibrium between fully membrane-inserted and MA-bound PI ( 4 , 5 ) P2 molecules at the assembly site . Accordingly , myristate may readily flip between the inner leaflet of the PM and the lipid binding pocket of MA . Once Gag reaches the PM , PI ( 4 , 5 ) P2 association triggers the myristate switch , and this effect is enhanced by the presence of acidic phospholipids . The combination of specific PI ( 4 , 5 ) P2 binding and myristate insertion into the PM then anchors the Gag molecule and makes it available for virus assembly . Subsequently , a dynamic exchange of lipid interactions may occur , where depletion of PM PI ( 4 , 5 ) P2 causes loss of Gag molecules once their myristate moieties flip out from the membrane . Finally , our results also shed some light on the question whether Gag assembly is directed to pre-existing assembly-prone membrane microdomains or whether Gag creates its own microdomain . Comparison of the membrane localization of Gag assembly sites prior to PI ( 4 , 5 ) P2 depletion with assembly sites in the same region of the cell after PI ( 4 , 5 ) P2 restoration indicated that they occur in different positions . The conclusion that there are no stable assembly-prone membrane microdomains is consistent with earlier observations showing that Gag initially binds to non-raft regions of the PM and subsequently either laterally associates with raft like domains or induces their formation ( Ono and Freed , 2001; Mercredi et al . , 2016 ) . Collectively , our study shows that PI ( 4 , 5 ) P2 at the PM is crucial throughout the entire HIV-1 assembly process and suggests a dynamic equilibrium of Gag-lipid interactions . Furthermore , it establishes an experimental system that permits synchronized induction of HIV-1 assembly by targeted modulation of Gag PM targeting . All chemicals and reagents were purchased from commercial sources unless otherwise noted . rCD1 was synthesized according to previously described procedures ( Feng et al . , 2014 ) . FK506 was purchased from LC laboratories ( Woburn , MA , USA ) . Plasmid pCHIV , expressing all HIV-1 NL4-3 proteins except for Nef under the control of a CMV promotor and its derivatives pCHIVEGFP , pCHIVSNAP and pCHIVCLIP were described previously ( Lampe et al . , 2007; Eckhardt et al . , 2011; Hanne et al . , 2016 ) . Derivatives pCHIV ( Δvpu ) and pCHIVEGFP ( Δvpu ) carry a deletion within the vpu coding region ( bp 3–120 ) , resulting in truncation of 40 amino acids followed by a frameshift . The design of pCHIV ( Δ8-126 SR ) , kindly provided by Martin Obr , was based on the previously published MA deletion Δ8-126 ( Reil et al . , 1998 ) , in which the globular domain of MA was replaced by two foreign amino acids ( Ser-Arg ) . A SNAP-tagged variant , pCHIVSNAP ( Δ8-126 SR ) , was generated by gene synthesis ( Thermo Fisher Scientific Geneart , Regensburg Germany ) based on pCHIVSNAP . The EGFP-tagged variant was generated by exchanging a ClaI fragment comprising the SNAP-tag coding region from pCHIVSNAP ( Δ8-126 SR ) by a ClaI fragment of pCHIVEGFP ( Lampe et al . , 2007 ) that comprises the GFP coding region . In the resulting plasmid pCHIVEGFP ( Δ8-126 SR ) the globular domain of MA is replaced by two foreign amino acids ( Ser-Arg ) , which are connected to EGFP by a flexible linker . EGFP and CA are , similar to pCHIVEGFP , connected via the HIV-1 protease cleavage site . Plasmid sequence information of the pCHIV ( Δ8-126 SR ) constructs are provided as supplementary files . For the generation of pGag-EGFP ( G2A ) , Gag was PCR amplified from pGag-EGFP ( Hermida-Matsumoto and Resh , 2000 ) using the primer 5’-CCC AAG CTT ATG GCT GCG AGA GCG TCG-3’ , which contains a HindIII restriction site and introduces the G2A mutation and the primer 5’-CGG GAT CCC CTT GTG ACG AGG GGT CGC-3‘ , containing a BamHI cleavage site . The PCR product was digested using BamHI and HindIII restriction endonucleases and ligated with a BamHI/HindIII cleavage product of pGag-EGFP . pNL4-3 was described previously ( Adachi et al . , 1986 ) . pNL4-3 16EK/29/31KE ( Joshi et al . , 2009 ) and pNL4-3 25/26KT ( Freed et al . , 1994 ) mutants were kindly provided by Eric Freed ( NCI , Frederick ) . pEGFP-PLCδ-PH was described previously ( Várnai and Balla , 1998 ) . The reversible chemical dimerizer system consisting of plasmids pLCK-ECFP-SNAPf and pmRFP-FKBP-5Ptase were described previously ( Feng et al . , 2014; Varnai et al . , 2006 ) . The EGFP-tagged version pEGFP-FKBP-5Ptase was kindly provided by Martina Schifferer . Plasmids pWPI_Puro and pWPI_BLR were obtained from Oliver Fackler ( Trotard et al . , 2015 ) . For generation of the lentiviral expression vectors pWPI-Puro-LCK-ECFP-SNAPf and pWPI-BLR-mRFP-FKBP-5Ptase , the Anchor ( LCK-ECFP-SNAPf ) and the Enzyme ( mRFP-FKBP-5Ptase ) coding regions were PCR-amplified from pLCK-ECFP-SNAPf and pmRFP-FKBP-5Ptase using the primers 5’-atcgaGGCGCGCCATGGGCTGCGTGTGCAG-3’ with 5’-atcgaACTAGTTTAATTAACCTCGAGTTTAAACGC-3’ for amplification of LCK-ECFP-SNAPf and 5’-gataaGGCGCGCCATGGCCTCCTCCGAGGA-3’ with 5’-gctacACTAGTTCAAGAAACGGAGGCGATG-3’ for amplification of mRFP-FKBP-5Ptase , introducing AscI and SpeI restriction sites at the 5’ and 3’ end , respectively . The fragments were digested using AscI and SpeI and subcloned into the vectors pWPI-Puro ( Anchor ) and pWPI-BLR ( Enzyme ) via AscI and SpeI restriction sites . Vesicular stomatitis virus G protein expression plasmid pMD2 . G ( Addgene#12260 ) and the lentiviral packaging plasmid psPAX2 ( Addgene#12259 ) were a gift from Didier Trono ( EPFL , Lausanne , Switzerland ) . pAdVantage was obtained from Promega ( Mannheim , Germany ) . The cell line HeLarCDS was generated using a lentiviral vector system . 293T cells seeded in 6-well plates were co-transfected with psPax2 , pMD2 . G , pAdvantage and either pWPI-Puro-LCK-ECFP-SNAPf or pWPI-BLR-mRFP-FKBP-5Ptase at a molar ratio of 0 . 27: 0 . 25: 0 . 13: 0 . 35 using Turbofect ( Thermo Scientific , Waltham , USA ) according to the manufacturer’s instructions . Tissue culture supernatants were harvested at 48 hpt , cleared by centrifugation and sterile filtered ( 0 , 45 µm pore size ) . 100 µl of both supernatants were mixed and used to transduce HeLa Kyoto cells in 6-well plates . Two days after transduction the growth medium was replaced by growth medium containing 2 µg/ml Puromycin ( Merck Millipore , Billerica , USA ) and 5 µg/ml Blasticidin ( Thermo Scientific , Waltham , USA ) for selection . Cells were passaged in selection medium and expression of the Anchor and Enzyme was monitored by spinning disc confocal microscopy and flow cytometry . Three weeks post transduction 100% of the HeLarCDS cell pool expressed the Enzyme , while 93% expressed the Anchor . HeLa Kyoto ( RRID:CVCL_1922 ) , HEK293T ( RRID: CVCL_0063 ) and TZM-bl reporter cells ( RRID:CVCL_B478 ) ( Wei et al . , 2002 ) were cultured at 37°C and 5% CO2 in Dulbecco’s modified Eagle’s medium ( DMEM; Invitrogen ) supplemented with 10% fetal calf serum ( FCS; Biochrom ) , 100 U/ml penicillin and 100 µg/ml streptomycin . Medium for cultivation of HeLarCDS cells additionally contained 2 µg/ml Puromycin and 5 µg/ml Blasticidin . Cell line identity has been authenticated using STR profiling ( Promega PowerPlex 21 Kit; carried out by Eurofins Genomics , Ebersberg , Germany ) . Cell lines were grown from mycoplasma free liquid nitrogen stocks . Passaged cells in culture in the lab are monitored regularly ( every 4 months ) for mycoplasma contamination using the MycoAlert mycoplasma detection kit ( Lonze Rockland , USA ) and cell lines used here were contamination free . For microscopy experiments , cells were seeded on slides within eight-well Lab-Tek chambered coverglass systems ( Thermo Scientific , Waltham , USA ) . At about 50% confluence , cells were transfected using Turbofect ( Thermo Scientific , Waltham , USA ) according to the manufacturer’s instructions . 0 . 5 µg DNA was transfected per well with pLCK-ECFP-SNAP , pmRFP-FKBP-5Ptase or pEGFP-FKBP-5Ptase , as indicated , and pCHIV derivatives ( molar ratio of 1 . 8:1:0 . 5 ) . Tagged pCHIV derivatives were transfected in equimolar ratio with their non-labeled counterpart resulting in a ratio of 1 . 8:1:0 . 25:0 . 25 . In the case of pNL-43 variants a molar ratio of 1 . 8:1:0 . 43 ( pLCK-ECFP-SNAP: pmRFP-FKBP-5Ptase: pNL4-3 ) was applied . pGag-EGFP ( G2A ) was used at a molar ratio of 2:1 with pCHIV and DNA amount of HIV-derivatives was increased 3 . 6-fold , while pLCK-ECFP-SNAP and pmRFP-FKBP-5Ptase quantities were kept constant , resulting in a total DNA amount of 0 . 76 µg/well and a molar ratio of pLCK-ECFP-SNAP:pmRFP-FKBP-5Ptase:pCHIV:pGag-EGFP ( G2A ) 1 . 8:1:0 . 9:1 . 8 . If pEGFP-PLCδ-PH was used , total DNA amount was reduced to 0 . 41 µg/well , while pLCK-ECFP-SNAP and pmRFP-FKBP-5Ptase quantities were kept constant , resulting in a molar ratio for pLCK-ECFP-SNAP: pmRFP-FKBP-5Ptase: pEGFP-PLCδ-PH of 1 . 8: 1: 0 . 05 . At 4 hr post transfection ( hpt ) the transfection mixture was replaced by imaging medium ( DMEM high glucose w/o phenol red w/o glutamine supplemented with 10% FCS , 4 mM GlutaMAX , 2 mM sodium pyruvate , 20 mM HEPES pH 7 . 4 , 100 U/ml penicillin and 100 µg/ml streptomycin ) . In order to achieve PM PI ( 4 , 5 ) P2 depletion prior to Gag accumulation , 1 µM rCD1 or 1% dimethyl sulfoxide ( DMSO ) vector control was added to the imaging medium at 4 hpt . During image acquisition ( performed at 22 hpt ) , samples were additionally treated with 1 µM FK506 to achieve PM PI ( 4 , 5 ) P2 reconstitution or with 1% DMSO , as indicated . In order to achieve PM PI ( 4 , 5 ) P2 depletion after Gag assembly at the PM , transfected samples were treated with 1 µM rCD1/1% DMSO at 22 hpt for a period of 90 min during live cell imaging . Subsequently 1 µM FK506 or 1% DMSO , as indicated , was added for an additional period of 90 min to achieve PM PI ( 4 , 5 ) P2 reconstitution . The medium was not exchanged between treatments . For Cycloheximide ( CHX ) treatment , cells were incubated with 10 µg/ml CHX ( Sigma-Aldrich , St . Louis , USA ) . For STED imaging HIV-1 Gag was detected via a CLIP-tag ( expressed from pCHIVCLIP[Hanne et al . , 2016] ) . Living cells were stained with Atto 590 BC-CLIP ( kindly provided by Janina Hanne ) in imaging medium for 30 min . Cells were washed twice with imaging medium , incubated for further 30 min at 37°C 5% CO2 in imaging medium , washed with phosphate buffered saline ( PBS ) and fixed for immunostaining . For immunostaining , cells were fixed at 24 hpt for 15 min with 4% PFA in PBS . For samples expressing the pNL4-3 mutants 16EK/29/31KE or 25/26KT fixation was prolonged to 90 min for biosafety reasons . Samples were washed with PBS , permeabilized with 0 . 1% Triton-X100 in PBS for 5 min and blocked for 30 min with 2% BSA in PBS . Permeabilization was omitted in case of Env labeling . Cells were incubated with the indicated primary antibody in 2% BSA in PBS for 2 hr , washed with PBS and incubated with the respective secondary antibody or Fab in 2% BSA in PBS for 1 hr . Antibody/Fab combinations were as follows: MA: polyclonal rabbit antiserum ( in house ) 1:500 / anti rabbit Alexa Fluor 647 1:500 ( Thermo Fisher Scientific Cat# A-21244 RRID:AB_2535812 ) , Env: monoclonal anti-gp120 antibody 2G12 1:50 ( Polymun Scientific Cat#AB002 RRID:AB_2661842 ) / Fab Fragment Goat Anti-Human IgG ( H+L ) ( Jackson ImmunoResearch Labs Cat# 109-007-003 RRID:AB_2337555 ) coupled to Abberior STAR RED NHS ( Abberior Instruments GmbH , Göttingen , Germany ) 1:50 , Tetherin: monoclonal anti-CD317 antibody 26F8 1:200 ( Thermo Fisher Scientific Cat# 16-3179-82 , RRID:AB_1518775 ) / anti mouse Alexa 647 1:500 ( Thermo Fisher Scientific Cat# A-21236 RRID:AB_2535805 ) . Finally , cells were washed and kept in PBS . Spinning disk confocal ( SDC ) imaging was performed at a PerkinElmer UltraVIEW VoX SDC microscope ( Perkin Elmer , Waltham , USA ) using a 60x Apo TIRF ( NA 1 . 49 ) oil immersion objective and Hamamatsu C9100-23B EM-CCD camera . Stacks were acquired with a z-spacing of 500 nm . Live-cell imaging was performed at 37°C , 5% CO2 , 40% humidity using multiposition imaging with an automated stage and the Perfect Focus System ( Nikon , Tokio , Japan ) for automated focusing at each time point with a time-resolution of 6 s - 5 min/frame . Total internal reflection fluorescence ( TIRF ) live cell imaging was performed at objective type TIRF setup ( Visitron Systems , Puchheim , Germany ) based on a Zeiss Axiovert 200M fluorescence microscope equipped with a 100 x Zeiss Alpha Plan Apochromat ( NA 1 . 46 ) oil immersion objective and Hamamatsu EM-CCD 9100–50 camera . The TIRF angle was manually controlled . Live-cell imaging was performed at 37°C 5% CO2 , 40% humidity with a time-resolution of 5 s/frame . Stimulated emission depletion ( STED ) imaging was performed at a λ = 775 nm STED system ( Abberior Instruments GmbH , Göttingen , Germany ) , using a 100 x Olympus UPlanSApo ( NA 1 . 4 ) oil immersion objective with 590 and 640 nm excitation laser lines at room temperature . Nominal STED laser power was set to ~60% of the maximal power of 1200 mW with 10 µs pixel dwell time and 15 nm pixel size . Representative still images or single frames of image sequences were chosen . Super resolution STED images were deconvolved with a Lorentzian function ( full width half maximum = 60 nm ) using the software Imspector ( Abberior Instruments GmbH , Göttingen , Germany ) . For all images shown , the camera offset value was subtracted and the contrast and brightness were adapted for optimal display of the image . To eliminate background noise , a 0 . 5 px median filter was applied to all SDC and TIRF images . Images are shown in greyscale or pseudo colors . In the latter case , the Fire and Green Fire Blue lookup tables ( LUTs ) were used for SDC and TIRF images , respectively , while different channels of super resolution STED images are shown with the LUTs Red Hot ( referred to as ‘red’ ) and Green Fire Blue ( referred to as ‘cyan’ ) . Particle analysis of TIRF microscopy data was done in FIJI ( RRID:SCR_002285 ) ( Schindelin et al . , 2012 ) in an overall similar manner as described for particle analysis of SDC microscopy data . The following parameters were changed in order to achieve optimal particle identification in TIRF image sequences: radius of rolling ball was 20 px , radius for Niblack automated thresholding was 10 , parameter 2 was −10 , a 1 px median filter was applied , the size in FIJI’s Analyze Particles function was set to 3-infinity px2 . Values were exported and further analyzed in Excel ( Microsoft , Redmond , USA ) ore GraphPad Prism ( GraphPad Software , Inc . , La Jolla , USA; RRID:SCR_002798 ) . All values obtained were divided by membrane area . For temporal alignment of movie sequences , the frame in which ~100 particles/1000 µm² were detected was defined as t = 0 . Finally , values were plotted as number of Gag clusters/1000 µm² over time . Cluster intensity analysis was done in FIJI ( RRID:SCR_002285 ) ( Schindelin et al . , 2012 ) . In contrast to the particle analysis , sum intensity projections of four slices acquired with 500 nm spacing in the middle of the cell were used . Clusters were thresholded as described above ( see Particle analysis of SDC microscopy data ) . The mean fluorescence intensity in clusters was quantified using the thresholded sum projected images and plotted over time . Images were linearly deconvolved with a Lorentzian function ( full width half maximum = 60 nm ) using the software Imspector ( Abberior Instruments GmbH , Göttingen , Germany ) . Line profiles of selected assembly sites were generated manually in FIJI ( RRID:SCR_002285 ) ( Schindelin et al . , 2012 ) . The intensity values in the Gag and Env channel were exported to Excel . To align the line profiles of different assembly sites , the Gag intensity peak of an assembly site was set to x = 0 nm and the corresponding Env intensity profile was adjusted accordingly . Intensity values were exported to GraphPad Prism ( GraphPad Software , Inc . , La Jolla , USA; RRID:SCR_002798 ) , normalized ( smallest value = 0 , highest value = 100 ) and the average normalized fluorescence intensities ± SEM were plotted . Time courses of native or induced Gag assembly at individual sites at the ventral PM were extracted from single plane TIRF microscopy time lapse sequences at a time resolution of 5 s/frame . Time courses of reinduced and native assembly at individual sites at the lateral PM were extracted from 3D volume time lapse ( five slices , 0 . 5 µm spacing ) acquired by SDC microscopy at a time resolution of 6 s/frame . The camera offset value was subtracted from all image sequences using FIJI ( RRID:SCR_002285 ) ( Schindelin et al . , 2012 ) and image sequences were imported to Imaris 8 ( Bitplane AG , Zurich , Switzerland ) . Spot detection and tracking was performed using Imaris’ Spot detection module . Within this process the background was subtracted and the estimated diameter for spot detection was set to 500 nm ( TIRF ) or 700 nm ( SDC ) . The quality parameter for spot detection was in the range of 35–250 ( TIRF ) or 75–200 ( SDC ) , depending on the dataset . Tracking was performed using the Autoregressive Motion algorithm , assuming a maximum distance between frames of 500 nm ( TIRF ) or 700–1000 nm ( SDC ) , allowing for a maximum gap size of 1 ( TIRF ) or 4 ( SDC ) and a track duration above 300 s . Filling gaps was disabled . Out of the detected spots , those which increased in mean intensity and reached a plateau phase were selected for further analysis . The mean intensity values over time were exported to Excel , temporally aligned ( the beginning of each track was set to t = 0 ) and normalized ( smallest value = 0 , highest value = 100 ) . The average normalized fluorescent values ( a . u . ) including standard deviation over time were plotted . In order to calculate assembly rate constants and half-times , single exponential fits to the data were performed using GraphPad Prism software ( GraphPad Software , Inc . , La Jolla , USA; RRID:SCR_002798 ) . The frequency distribution of k-values was plotted after binning with a bin width of 0 . 0005 s−1 and 120 s , respectively . HeLarCDS cells were seeded in 6-well plates and transfected with 2 µg pNL4-3 at about 50% confluence using Turbofect ( Thermo Scientific , Waltham , USA ) according to the manufacturer’s instructions . At 5 hpt cells were treated with 1 ml growth medium containing 1 µM rCD1 ( induced/inhibited assembly ) or equivalent amounts of DMSO ( native assembly ) at 5 hpt . At 22 hpt the tissue culture supernatant was harvested ( ‘0 hr’ sample ) and cleared by centrifugation at 1000 g for 10 min . The cells were washed with PBS twice and virus attached to the cell surface was inactivated by acid wash ( 40 mM citric acid , 135 mM NaCl , 10 mM KCl , pH 3 . 0 ) for 1 min followed by a washing step with pre-warmed medium . Subsequently , 1 ml of fresh , pre-warmed growth medium containing 1 µM rCD1 ( inhibited assembly ) , 1 µM FK506 ( induced assembly ) or DMSO solvent control ( native assembly ) was added to the cells and incubation was continued for 2 hr . Supernatants were harvested again ( ‘2 hr’ samples ) and cleared by centrifugation at 1000 g for 10 min . Cells were processed for immunofluorescence and flow cytometry to control for transfection efficiency and Enzyme localization . Relative virus infectivity was assessed using TZM-bl reporter cells . 5*103 TZM-bl cells/well were plated in 96-well plates and infections with serial 2-fold dilutions of the cleared supernatants obtained from virus producing cells were carried out on the following day . At 48 hpi ( hours post infection ) , infected cells were lysed and luciferase activity was measured using the Steady-Glo-Assay ( Promega , Madison , USA ) according to the manufacturer’s instructions . Relative infectivity of supernatants was calculated from the linear range of the titration curves and normalized to the amount of virus present in the supernatants as assessed by SG-PERT ( SYBR Green based Product Enhanced Reverse Transcriptase ) assay ( Pizzato et al . , 2009 ) . Data analysis was performed using GraphPad Prism ( GraphPad Software , Inc . , La Jolla , USA; RRID:SCR_002798 ) . Values are expressed as mean ±SEM or mean ±SD , as indicated . Statistical significance of the data presented was assessed with the two-tailed unpaired Student’s t-test or the nonparametric Mann-Whitney U test , as indicated . Mann-Whitney U test was applied for statistical analysis of the data , which did not follow a Gaussian distribution ( determined by inspection of Box and Whiskers graphs and Histograms of frequency distributions ) . Values of p<0 . 05 were considered significant . Supplementary files 1 , 2 , and 3 . Files pCHIV ( Δ8-126 SR ) . txt , pCHIVGFP ( Δ8-126 SR ) . txt and pCHIVSNAP ( Δ8-126 SR ) . txt contain the plasmid sequences of the respective constructs . Figure 1—figure supplement 1 shows reversible PI ( 4 , 5 ) P2 depletion from the PM by the rCDS . Figure 2—figure supplement 1 shows that the MA deletion mutant Δ8-126SR is not responsive to PI ( 4 , 5 ) P2 depletion . Figure 3—figure supplement 1 shows the relative frequency distributions of assembly rates of native and induced Gag assembly at the ventral PM . Figure 5—figure supplement 1 shows the individual data sets determined for virus release and infectivity used for calculation of the normalized numbers shown in Figure 5 . Figure 6—figure supplement 1 shows that addition of DMSO does not affect assembled Gag clusters and that the MA deletion mutant Δ8-126SR does not respond to PI ( 4 , 5 ) P2 depletion and rescue . Figure 6—figure supplement 2 shows that virus release or impairment of new particle formation do not significantly contribute to the loss of Gag clusters from the PM upon PM PI ( 4 , 5 ) P2 depletion . Figure 6—figure supplement 3 shows that the Gag mutant G2A is not responsive to PI ( 4 , 5 ) P2 rescue following PI ( 4 , 5 ) P2 depletion . Figure 6—figure supplement 4 shows the assembly kinetics of reinduced and native assembly at the lateral PM . Video 1 shows reversible PI ( 4 , 5 ) P2 depletion from the PM by the rCDS , supporting Figure 1—figure supplement 1 . Video 2 shows rCDS-induced Gag assembly and Video 3 shows the respective DMSO control , together supporting Figure 2 . Videos 4 and 5 show TIRF microscopy time-lapse of native and induced assembly , respectively , supporting Figure 3 . Video 6 show loss of Gag clusters upon PI ( 4 , 5 ) P2 depletion and subsequent reinduction of Gag assembly by the rCDS and Video 7 shows the respective DMSO control , together supporting Figure 6 . Video 8 shows reinduced assembly acquired with high timeresolution spinning disc confocal microscopy , supporting Figure 6—figure supplement 4 .	Viruses are parasites that must infect the living cell of a host in order to grow and replicate . To do so , the virus attaches to the host’s cell and transfers its genetic material to the inside . The virus then hijacks the cell and forces it to build proteins and more genetic material that will assemble into new copies of the virus , and the completed virus particles are released to infect new cells . The HIV-1 virus encodes a protein called Gag that coordinates the assembly of new copies of this virus . Thousands of Gag proteins accumulate into a lattice at the inside of the host cell membrane by an unknown mechanism , where they gather all essential components to form a new infectious virus . Once completed , the virus particle detaches itself from the host cell and Gag gives the virus its structure . Previous research by several groups has shown that a lipid molecule found in cell membranes called PIP2 can bind to Gag and helps the virus to assemble . However , until now , it was unclear if PIP2 anchors Gag to the cell membrane of living cells and if it plays any other roles in the later stages of assembly . Now , Mücksch et al . have developed a new approach to study this question by rapidly manipulating the levels of PIP2 in living cells and monitoring Gag assembly through a fluorescent marker . The experiments showed that PIP2 was needed to initiate the assembly process , but also to maintain the partially assembled Gag lattice at the membrane . When PIP2 was removed , Gag proteins could not gather at the cell membrane and the already assembled Gag lattices broke down . When PIP2 levels were increased again , the Gag proteins that had disappeared from the membrane formed new lattices . This suggests that PIP2 has a much broader role in HIV-1 particle formation than previously assumed . A next step will be to use this new experimental approach to study how particles of HIV are assembled and released . This knowledge may help scientists to develop antiviral drugs that interfere with the assembly step of the virus that could be used to prevent or treat HIV infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "microbiology", "and", "infectious", "disease" ]	2017	Synchronized HIV assembly by tunable PIP2 changes reveals PIP2 requirement for stable Gag anchoring
Cell-based studies showed that several Mdm2-binding ribosomal proteins , upon overexpression , stabilize and activate p53 . In contrast , here we show in a mouse knockout study that Mdm2-binding ribosomal protein S27-like ( Rps27l ) , upon disruption , activates p53 . Germline inactivation of Rps27l triggers ribosomal stress to stabilize Mdm2 , which degrades Mdm4 to reduce Mdm2-Mdm4 E3 ligase towards p53 , leading to p53-dependent apoptotic depletion of hematopoietic stem cells and postnatal death , which is rescued by Trp53 deletion . Paradoxically , while increased p53 is expected to inhibit tumorigenesis , Rps27l−/−;Trp53+/− mice develop lymphomas at higher incidence with p53 loss-of-heterozygosity and severe genome aneuploidy , suggesting that Rps27l disruption impose a selection pressure against p53 . Thus , Rps27l has dual functions in p53 regulation: under Trp53+/+ background , Rps27l disruption triggers ribosomal stress to induce p53 and apoptosis , whereas under Trp53+/− background , Rps27l disruption triggers genomic instability and Trp53 deletion to promote tumorigenesis . Our study provides a new paradigm of p53 regulation . Tumor suppressor p53 is a key regulator of cell growth and cell death ( Ko and Prives , 1996; Levine , 1997 ) and is activated by many environmental stimuli , including DNA damaging agents ( Giaccia and Kastan , 1998 ) . Activated p53 acts as a guardian of the genome by inducing growth arrest , allowing cells to repair the damage , or apoptosis when the damage is too severe and irreparable ( Vogelstein et al . , 2000; Levine and Oren , 2009 ) . p53 is frequently inactivated during human carcinogenesis either by point mutation which occurs in 50% of human cancers ( Greenblatt et al . , 1994 ) , or by Mdm2/Mdm4-mediated ubiquitylation and degradation ( Kruse and Gu , 2009; Wade et al . , 2013 ) . Mdm2 is a direct p53 target . Upon induction by p53 , Mdm2 inactivates p53 through at least two main mechanisms: ( a ) binding to p53 at its transactivation domain and blocking its transactivation activity , and ( b ) serving as an E3 ubiquitin ligase to promote a rapid degradation of p53 ( Haupt et al . , 1997; Honda et al . , 1997; Kubbutat et al . , 1997; Fang et al . , 2000; Honda and Yasuda , 2000 ) . Thus , p53-Mdm2 forms an auto-regulatory feedback loop to keep p53 levels under control ( Wu et al . , 1993 ) . Both in vitro and in vivo studies indicated that oncogenic activity of Mdm2 is mainly attributable to its binding to and degrading p53 ( Jones et al . , 1995; Montes de Oca Luna et al . , 1995; de Rozieres et al . , 2000 ) . The key role of Mdm2 E3 ligase activity in controlling p53 levels in a physiological setting was further demonstrated by a knock-in study in which introduction of a ligase dead Mdm2C462A mutant results in embryonic lethality , that like Mdm2 deletion , can be fully rescued by simultaneous Trp53 deletion ( Itahana et al . , 2007 ) . In addition to Mdm2 , p53 is also subject to negative regulation by Mdm4 ( also known as MdmX ) , an Mdm2 family member ( Shvarts et al . , 1996 ) . Although Mdm4 itself does not have an intrinsic E3 ligase activity toward p53 ( Linares et al . , 2003 ) , it does bind to p53 transactivation domain to block its transcription activity ( Shvarts et al . , 1996 ) . Moreover , Mdm4 forms a tight 1:1 complex with Mdm2 via their respective C-terminal RING finger domains ( Sharp et al . , 1999; Tanimura et al . , 1999 ) , and the Mdm2-Mdm4 heterodimers are the preferred dimer form , compared to the Mdm2-Mdm2 or Mdm4-Mdm4 homodimers ( Kostic et al . , 2006 ) . Furthermore , Mdm4 is a direct substrate of Mdm2 for targeted ubiquitylation and degradation ( de Graaf et al . , 2003; Kawai et al . , 2003; Pan and Chen , 2003 ) . More importantly , both in vitro cell culture studies using Mdm2 mutants ( Kawai et al . , 2007; Poyurovsky et al . , 2007; Uldrijan et al . , 2007 ) and in vivo studies using knock-in mice of Mdm2 and Mdm4 mutants ( Itahana et al . , 2007; Huang et al . , 2011; Pant et al . , 2011; Wang et al . , 2011 ) demonstrated that the Mdm2-Mdm4 heterodimer has an optimal E3 ligase activity and is required for p53 degradation . Thus , the Mdm2-Mdm4 complex is interconnected and cross-regulated to keep p53 levels precisely in check under physiological conditions ( such as during embryogenesis ) and in response to various stresses ( Wade et al . , 2013 ) . Accumulated biochemical and cellular studies have shown that the p53-MDM2-MDM4 axis is further regulated by various ribosomal proteins ( Zhang and Lu , 2009 ) . Specifically , the ribosomal proteins , such as RPL11 ( Lohrum et al . , 2003; Zhang et al . , 2003; Bhat et al . , 2004; Sasaki et al . , 2011 ) , RPL5 ( Dai and Lu , 2004 ) , RPL23 ( Dai et al . , 2004; Jin et al . , 2004 ) , RPL26 ( Zhang et al . , 2010 ) , RPS7 ( Chen et al . , 2007; Zhu et al . , 2009 ) , RPS3 ( Yadavilli et al . , 2009 ) , RPS27/S27L ( Xiong et al . , 2011 ) , S27a ( Sun et al . , 2011 ) , RPS25 ( Zhang et al . , 2013 ) , RPS26 ( Cui et al . , 2014 ) and RPS14 ( Zhou et al . , 2013 ) , as well as RPL37 , RPS15 and RPS20 ( Daftuar et al . , 2013 ) , were found to bind to MDM2 upon ribosomal stress and inhibit MDM2-mediated p53 ubiquitylation and degradation , leading to p53 activation to induce growth arrest and apoptosis , thus acting as p53 activators ( Zhang and Lu , 2009 ) . However , whether and how these Mdm2-binding ribosomal proteins indeed regulate p53 by modulating Mdm2 activity has not been explored previously using an in vivo mouse model . RPS27L ( NM_015920 ) is an 84-amino acid ribosomal like protein , which differs from its family member RPS27 ( NM_001030 ) by only three amino acids ( R5K , L12P , K17R ) at the N-terminus . We and the others found that RPS27L is a direct p53 inducible target ( He and Sun , 2007; Li et al . , 2007 ) . Our recent cell-based study showed that RPS27L directly binds to MDM2 and is subjected to MDM2-mediated ubiquitylation and degradation ( Xiong et al . , 2011 ) . Furthermore , RPS27L competes with p53 for MDM2 binding , consequently inhibiting MDM2-mediated p53 ubiquitylation and degradation ( Xiong et al . , 2011 ) . Thus , RPS27L interplays with the MDM2-p53 axis to regulate p53 activity . Although several ribosomal proteins have been previously shown to bind and inhibit MDM2 , causing p53 activation ( Zhang and Lu , 2009 ) , with RPS7 and RPL26 being MDM2 substrates as well ( Ofir-Rosenfeld et al . , 2008; Zhu et al . , 2009 ) , RPS27L is the first and only known ribosomal-like protein that is a direct p53 inducible target , a MDM2 substrate , and a regulator of the MDM2-p53 axis . However , the physiological function of Rps27l and whether Rps27l plays a physiological role in regulation of the p53-Mdm2-Mdm4 axis in mouse remain entirely unknown . Here we present in vivo evidence that , unlike in vitro cell culture studies which showed that several Mdm2-binding ribosomal proteins act as p53 activators , Rps27l , under the Trp53+/+ background , appears to be a physiological p53 inhibitor that stabilizes the Mdm2-Mdm4 heterodimer for effective p53 ubiquitylation and degradation . Unexpectedly , we also found that Rps27l , under the Trp53+/− background , acts as a tumor suppressor by maintaining the genomic integrity to prevent the loss of Trp53 heterozygosity and subsequent development of lymphoma . Thus , Rps27l regulates p53 either negatively or positively in a manner dependent of Trp53 dosage . Our previous studies showed that RPS27L is a direct p53 target ( He and Sun , 2007 ) and regulates p53 activity by interacting with MDM2 ( Xiong et al . , 2011 ) . To test the physiological function of Rps27l , we generated the gene-trap based Rps27l+/− mice through Texas Institute for Genomic Medicine ( TIGM ) in a pure C57BL/6 background from an ES clone , IST11658B7 , with a targeted BGEO vector inserted at the intron 1 to disrupt the open reading frame of Rps27l ( Figure 1—figure supplement 1A ) . Intercrossing of Rps27l+/− mice gave rise to the F1 offspring with three Rps27l genotypes , as confirmed by Southern blotting , PCR genotyping , and immunoblotting ( Figure 1—figure supplement 1B–D ) . Thus , Rps27l deletion is not embryonic lethal . However , genotyping of 351 offspring at weaning of the Rps27l+/− intercrossing failed to identify any Rps27l−/− mice ( Figure 1A ) . The ratio of Rps27l+/+ vs Rps27l+/− mice was about 1:2 , indicating that Rps27l mutation is homozygous lethal ( Figure 1A ) . In fact , all Rps27l−/− pups in a pure C57BL/6 background died within 12 days of birth . Under a mixed sv129/B6 background , about 50% mice die at 18 day after birth with the longest life-span of 35 days ( Figure 1—figure supplement 1E ) . Thus , Rps27l disruption causes postnatal death . 10 . 7554/eLife . 02236 . 003Figure 1 . Rps27l disruption causes postnatal death as a result of increased apoptosis . ( A ) Disruption of Rps27l causes postnatal death . ( B–D ) Reduced body size , body weight , and organ weight in Rps27l−/− mice . An Rps27l−/− pup ( bottom ) and two Rps27l+/− littermates ( top ) at P8 were photographed ( B ) . The body ( C ) and the organs ( D ) of P8–10 pups with genotypes of Rps27l+/+ ( n = 3 ) , Rps27l+/− ( n = 6 ) , and Rps27l−/− ( n = 10 ) were weighed . Shown are mean ± SD . p < 0 . 05 , p < 0 . 01 , and p < 0 . 001 , as compared to Rps27l+/+ counterparts . ( E ) Representative H&E staining of bone marrows in femurs of P6 pups . Scale bars represent 200 µm ( top ) and 40 µm ( bottom ) . ( F ) Representative H&E staining of thymuses from P8–10 pups . Arrows point to apoptotic cells . Scale bar represents 10 µm . ( G and H ) Representative immunofluoresent staining of cleaved caspase-3 in thymuses ( G ) of P8–10 pups and bone marrows ( H ) of P6 pups . Scale bars represent 20 µm ( G ) and 100 µm ( H ) . ( I ) Accumulation of cleaved caspase-3 and Puma in Rps27l−/− thymuses . The thymuses of P6 pups were lysed for immunoblotting ( IB ) . ( J ) Increased Annexin V-positive myeloid progenitors ( MP ) in Rps27l−/− fetal livers . Cells from E14 . 5 Rps27l+/+ ( n = 6 ) , Rps27l+/− ( n = 12 ) , and Rps27l−/− ( n = 11 ) fetal livers were stained with antibodies ( Abs ) against surface markers and Annexin V-FITC , followed by FACS analysis . Data shown are mean ± SD . p < 0 . 001 , as compared to Rps27l+/+ counterparts . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 00310 . 7554/eLife . 02236 . 004Figure 1—figure supplement 1 . Generation of Rps27l gene trap mice and phenotypes of Rps27l−/− mice . ( A ) Genomic structure of mouse Rps27l gene . Four exons ( E1-E4 ) and three introns ( between exons ) are shown , along with the insertion site in intron 1 of the gene trap vector . ( B ) Southern blotting confirmed the genotypes . Mouse genomic DNA was digested with EcoRI ( 5′ probe ) or PstI ( β-Gal probe ) , followed by Southern blot analysis . ( C ) PCR-based genotyping . Mouse tail DNA was isolated and amplified by PCR . Shown are two fragments with sizes of 525 bp ( wt ) and 325 bp ( mutant ) . ( D ) Lack of Rps27l protein expression in Rps27l−/− brains . The brains from P8 mice were homogenized and subjected to IB with indicated antibodies ( Abs ) . ( E ) Kaplan–Meier survival curves of mice with three Rps27l genotypes under a mixed Sv129/B6 background . ( F ) Increased apoptosis in myeloid progenitors ( MP ) in fetal livers upon Rps27l disruption . Fetal liver cells of three genotypes were stained with Abs against surface markers and Annexin V-FITC , followed by FACS analysis . Shown is a representative FACS profile of Annexin V staining . ( G ) No change of proliferation of MP in fetal livers upon Rps27l disruption . Pregnant mice with indicated genotypes were i . p . injected with BrdU labeling reagent 2 hr before being sacrificed . Cells from E14 . 5 fetal livers of Rps27l+/+ ( n = 6 ) , Rps27l+/− ( n = 5 ) , and Rps27l−/− ( n = 3 ) were stained with surface markers and BrdU-FITC Abs , followed by FACS analysis . ( H ) Reduced size of thymus and spleen in Rps27l−/− mice . Representative thymuses and spleens from three genotypes of P18 mice under a mixed Sv129/B6 background were photographed . ( I ) Hypo-cellularity of organs in Rps27l−/− mice . The total cell numbers of bone marrow ( femur and tibia from two hind limbs ) , spleen , and thymus were counted from P18 Sv129/B6 mice with the genotypes of Rps27l+/+ ( n = 3 ) , Rps27l+/− ( n = 4 ) , or Rps27l−/− ( n = 7 ) . Shown are mean ± SD . p < 0 . 05 , *p < 0 . 01 , as compared with Rps27l+/+ counterparts . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 004 Compared to the wild type and heterozygous littermates , Rps27l−/− mice were about 50% in size and body weight ( Figure 1B , C ) . At the organ levels after normalization with the body weight , the weight in the thymus and spleen was significantly reduced in Rps27l−/− mice at P8–10 ( Figure 1D ) . The H&E staining revealed that the bone marrow derived from Rps27l−/− mice have much reduced cellularity ( Figure 1E ) . Likewise , the Rps27l−/− thymus showed a remarkable reduction in the thymic cortex , and thymocytes in cortical and cortical-medullary junction areas undergo apoptosis with accumulation of small irregular nuclear fragments ( Figure 1F , arrows ) . Induction of apoptosis was further confirmed by the immunofluorenscent staining with antibody against cleaved caspase-3 in both thymocytes ( Figure 1G ) and bone marrow cells ( Figure 1H ) . Moreover , apoptosis in thymus was readily detectable by immunoblotting , which showed increased levels of cleaved caspase-3 and pro-apoptotic protein Puma ( Figure 1I ) . An increased apoptosis , as reflected by Annexin V staining ( Figure 1J , Figure 1—figure supplement 1F ) , but not reduced proliferation , as reflected by BrdU incorporation ( Figure 1—figure supplement 1G ) was also seen in myeloid progenitors ( MP ) , derived from E14 . 5 Rps27l−/− fetal livers . Finally , hypo-cellularity of bone marrow , thymus , and spleen was readily observed in Rps27l−/− mice with a mixed 129/B6 background ( Figure 1—figure supplement 1H , I ) . Thus , Rps27l is required for postnatal development of some organs , particularly thymus , spleen and bone marrow , and postnatal death of Rps27l knockout mice is likely associated with enhanced apoptosis , leading to bone marrow depletion . To determine potential sources of bone marrow depletion in Rps27l−/− pups , we compared the fetal livers from E14 . 5 embryos among three Rps27l genotypes and found that Rps27l−/− fetal livers are visibly smaller with significant reduction in cell number ( Figure 2A , B ) . The FACS analysis of Rps27l−/− fetal livers revealed a substantial reduction in number of hematopoietic stem and progenitor cells ( HSPCs ) , including hematopoietic stem cells ( HSCs ) -containing LSK ( Lin−/Sca-1+/c-Kit+ ) population , and MP population , consisting of common myeloid progenitor ( CMP ) , granulocyte-monocyte progenitor ( GMP ) , and megakaryocytic-erythroid progenitor ( MEP ) cells ( Figure 2C–F , Figure 2—figure supplement 1A ) . To determine the viability and functionality of Rps27l−/− HSPCs , we performed a non-competitive reconstitution assay to see whether these stem cells , while reduced in number , are still sufficient to rescue the bone marrow failure of recipient mice which were sterilized by a lethal dose of radiation . A total of 2 × 106 fetal liver cells from either Rps27l+/+ or Rps27l−/− embryos were used . Remarkably , while HSPCs from wild type fetal liver cells completely reconstituted sterilized bone marrow and fully rescued the recipient mice , cells from Rps27l−/− fetal livers were unable to do so , leading to a 100% death of recipient mice within 10 days of reconstitution ( Figure 2G , H , Figure 2—figure supplement 1B ) . To eliminate possibility that the failure in rescue is due to an insufficient number of stem cells in Rps27l−/− fetal livers , which is about 50% of the Rps27l+/+ control ( Figure 2C , 0 . 005% vs 0 . 01% of total liver ) , we repeated this non-competitive reconstitution assay , using three times more fetal liver cells ( 6 × 106 ) from the Rps27l−/− embryos than that ( 2 × 106 ) from the Rps27l+/+ or Rps27l+/− embryos . Again , the stem cells from Rps27l−/− fetal livers failed to rescue sterilized recipient mice with 100% death within 12 days post reconstitution , whereas those from wild type or heterozygous fetal liver caused a 100% rescue ( Figure 2—figure supplement 1C ) , indicating an intrinsic defect . Peripheral blood profiling of rescued mice showed a normal distribution of various blood cells ( Figure 2—figure supplement 1D , E ) . 10 . 7554/eLife . 02236 . 005Figure 2 . Reduced hematopoietic stem and progenitor cells in Rps27l−/− fetal livers . ( A and B ) Reduced size and total cell numbers in Rps27l−/− fetal livers . Representative fetal livers of three genotypes were photographed ( A ) , and the numbers of fetal liver cells from E14 . 5 embryos were counted . Shown are mean ± SD with embryo numbers as follows: Rps27l+/+ ( n = 7 ) , Rps27l+/− ( n = 10 ) , and Rps27l−/− ( n = 8 ) . p < 0 . 05 , **p < 0 . 001 , as compared with Rps27l+/+ counterparts . ( C–F ) Decreased percentages and absolute numbers of HSPCs in Rps27l−/− fetal livers . Cells from E14 . 5 fetal livers of Rps27l+/+ ( n = 7 ) , Rps27l+/− ( n = 10 ) , and Rps27l−/− ( n = 8 ) were stained with antibodies against various surface markers . The populations of LSK ( C and D ) , MP , CMP , GMP , and MEP ( E and F ) were analyzed by flow cytometry . Shown are mean ± SD . p < 0 . 05 , p < 0 . 01 , and p < 0 . 001 , as compared with Rps27l+/+ counterparts . ( G ) Diagram of non-competitive reconstitution assay . Fetal liver cells ( 2 × 106 cells ) from E14 . 5 embryos ( CD45 . 2 ) were injected into lethally irradiated recipient mice ( CD45 . 1 ) . Peripheral blood from recipients was analyzed by flow cytometry at 4 weeks after transplantation . ( H ) Kaplan–Meier survival curves of recipient mice after transplantation . Rps27l+/+ or Rps27l−/− fetal liver cells were injected , respectively , into recipient mice ( n = 5 , for each genotype ) . p = 0 . 0026 . ( I ) Diagram of competitive reconstitution assay . Fetal liver cells ( 2 × 106 cells ) from E14 . 5 embryos ( CD45 . 2 ) were injected into lethally irradiated recipient mice ( CD45 . 1 ) together with bone marrow cells ( 0 . 5 × 106 cells ) from normal recipient mice . Peripheral blood from recipients was analyzed by flow cytometry at 4 , 12 , and 20 weeks after transplantation . ( J ) Representative FACS profiles of donor-type ( CD45 . 2 ) and recipient-type ( CD45 . 1 ) blood cells at 4 weeks post transplantation . ( K ) Dramatic reduction of donor-type ( CD45 . 2 ) blood cells in recipients transplanted with Rps27l−/− fetal livers . The percentages of CD45 . 2+ cells in peripheral blood at 4 , 12 , and 20 weeks post transplantation were summarized . Rps27l+/+ or Rps27l−/− fetal liver cells were injected into recipient mice ( n = 3 , for each genotype ) . Shown are mean ± SD . p < 0 . 01 , *p < 0 . 001 , as compared to Rps27l+/+ counterparts . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 00510 . 7554/eLife . 02236 . 006Figure 2—figure supplement 1 . Depletion of hematopoietic stem and progenitor cells ( HSPCs ) upon Rps27l disruption . ( A ) Representative FACS profiles of HSPCs in fetal livers . Fetal liver cells were isolated from E14 . 5 embryos with indicated genotypes and subjected to FACS analysis using indicated surface marker Abs . ( B ) Representative FACS profiles of blood cells from donor-type ( CD45 . 2 ) and recipient-type ( CD45 . 1 ) mice in recipient mice transplanted with Rps27l+/+ fetal liver cells at 4 weeks post transplantation . ( C ) Kaplan–Meier survival curves of recipient mice after transplantation . The fetal liver cells from Rps27l+/+ ( 2 × 106 cells ) , Rps27l+/− ( 2 × 106 cells ) , or Rps27l−/− ( 6 × 106 cells ) embryos were injected , respectively , into recipient mice ( n = 8 for Rps27l+/+ livers , n = 9 for Rps27l+/− livers , n = 7 for Rps27l−/− livers ) . p < 0 . 0001 . ( D and E ) The fetal livers from Rps27l+/+ and Rps27l+/− embryos have similar reconstitution activity . CBC classification of peripheral blood from recipient mice was performed at 4 weeks post transplantation ( D ) . WBC , white blood cells; NE , neutrophils; LY , lymphocytes; MO , monocytes; HCT , hematocrit; RBC , red blood cells; Hb , hemoglobin; PLT , platelets . The percentages of donor-derived leukocytes at 4 weeks post transplantation were summarized ( E ) . ( n = 5 for each genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 006 To further confirm the inability of stem/progenitor cells from Rps27l−/− fetal livers in re-building the recipient bone marrow , we performed a competitive reconstitution assay in which a 1:4 mixture of recipient bone marrow cells with donor fetal liver cells derived from Rps27l+/+ vs Rps27l−/− embryos was given to sterilized recipient mice ( Figure 2I ) . In this case , all recipient mice survived , as expected . Bone marrow profiling of surviving chimeric mice at 4 , 12 and 20 weeks post reconstitution revealed that while 65–85% cells were derived from the Rps27l+/+ donor fetal livers , none of cells were derived from Rps27l−/− donor fetal livers ( Figure 2J , K ) . Together , our study clearly demonstrated in vivo that HSPCs from Rps27l−/− fetal livers must gradually die or fail to repopulate sterilized bone marrow . Thus , defective hematopoiesis is most likely contributable to postnatal lethality of Rps27l−/− mice . We next investigated potential mechanisms by which Rps27l disruption induces apoptosis and causes the loss of HSPCs and depletion of bone marrow . We focused on p53 , since ( 1 ) our recent study showed that RPS27L could modulate MDM2 to regulate the level and activity of p53 ( Xiong et al . , 2011 ) , and ( 2 ) growth retardation in Rps27l−/− mice and apoptosis in Rps27l−/− organs ( e . g . , thymus and bone marrow ) are the typical phenotypes of p53 activation . Indeed , by immunohistochemistry analysis , we detected more p53 positively stained cells in the fetal livers and bone marrows derived from Rps27l−/− mice than those from Rps27l+/+ control littermates ( Figure 3A ) . By immunoblotting , we found that both basal and radiation-induced levels of p53 and p53 target protein Puma are higher in fetal livers derived from Rps27l−/− embryos than in those from Rps27l+/+ embryos ( Figure 3B ) . Likewise , the levels of p53 and Puma are also higher in Rps27l−/− spleen and brain tissues ( Figure 3C , D ) . Furthermore , we observed a moderate increase in the levels of p53 and its two well-known targets , Mdm2 and p21 in MEFs derived from Rps27l−/− embryos than in those from Rps27l+/+ embryos ( Figure 3E ) . Finally , the p53 levels , induced by ( a ) ribosomal stress inducer , actinomycin D ( Act D ) , ( b ) DNA damaging agent , etoposide ( Figure 3—figure supplement 1A ) , or ( c ) ionizing radiation ( Figure 3—figure supplement 1B ) , were also higher in Rps27l−/− MEFs than in Rps27l+/+ MEFs . It is worth noting that increased p53 in Rps27l−/− MEFs is unlikely due to enhanced DNA damage response , given the similar phosphorylation levels of γH2AX and Chk1 between MEFs of the two genotypes ( Figure 3—figure supplement 1B ) . Taken together , our results clearly showed that Rps27l disruption causes an increase in p53 levels in multiple organs and cell types , and that increased p53 is transcriptionally active to induce the expression of its downstream targets , namely p21 , Mdm2 and Puma . 10 . 7554/eLife . 02236 . 007Figure 3 . Rps27l disruption causes a moderate increase of p53 and p53 targets . ( A ) Representative p53 staining in fetal livers and bone marrows . Fetal livers at E14 . 5 and bone marrows at P6 were immuno-stained with p53 Ab . Arrows point to p53 positive staining . Scale bar represents 40 µm . ( B ) Accumulation of p53 and Puma in Rps27l−/− fetal livers . Fetal livers isolated from embryos of non-irradiated or irradiated pregnant Rps27l+/− females ( E14 . 5 ) were lysed for IB at 5 hr after ionizing radiation at 5 Gy . denotes a nonspecific band . ( C ) Accumulation of p53 and p53 targets in Rps27l−/− spleens . Several spleens with the same genotype from P6 pups were harvested , pooled , homogenized , and subjected to IB . ( D ) Accumulation of p53 and p53 targets in Rps27l−/− brains . Brains from P8–10 pups were harvested and lysed for IB . ( E ) Accumulation of p53 and p53 target proteins in Rps27l−/− MEFs . Two independent pairs of MEFs from embryos at E13 . 5 were lysed for IB with indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 00710 . 7554/eLife . 02236 . 008Figure 3—figure supplement 1 . Increased p53 levels in responsive to various stresses in Rps27l−/− MEFs . ( A ) Higher p53 levels induced by ribosomal stress and DNA damaging agent , not by endoplasmic reticulum ( ER ) stress in Rps27l−/− MEFs . MEFs were treated with 5 nM Actinomycin D ( ActD ) for 24 hr , 5 μg/ml tunicamycin ( Tm ) for 8 hr , or 50 μM etoposide ( Etop ) for 24 hr , followed by IB . LE: long exposure . ( B ) Higher p53 levels induced by ionizing radiation in Rps27l−/− MEFs . MEFs were harvested for IB at various time points post 6 Gy of ionizing radiation . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 008 We next determined the mechanism by which Rps27l disruption causes p53 increase . Quantitative RT-PCR analysis using four independent sets of MEFs revealed that Rps27l disruption did not increase p53 mRNA level , but as expected , increased mRNA levels of two p53 targets , Mdm2 and p21 , due to p53 transactivation ( Figure 4—figure supplement 1A ) . The [35S]-methionine labeling experiment in two independent pairs of MEFs showed that the rate of p53 synthesis is similar regardless of Rps27l status ( Figure 4—figure supplement 1B ) , indicating that Rps27l disruption does not alter the p53 protein synthesis . We then focused our attention on the rate of p53 degradation by measuring the p53 protein half-life . We found that p53 half-life was doubled from ∼10 min to ∼25 min in MEFs upon Rps27l disruption ( Figure 4A ) due to reduced ubiquitylation and degradation ( Figure 4B , Figure 4—figure supplement 1C ) . Because the major E3 ligase for p53 degradation is Mdm2 , we performed an in vitro ubiquitylation assay , using , as the E3 source , the Mdm2 complex immuno-precipitated from the paired Rps27l MEFs , and found that the p53 ubiquitylation was reduced in Rps27l−/− MEFs substantially ( Figure 4C ) , indicating a reduced Mdm2 ligase activity toward p53 . 10 . 7554/eLife . 02236 . 009Figure 4 . Rps27l disruption alters the levels of the p53-Mdm2-Mdm4 axis . ( A ) Extension of protein half-lives of p53 and Mdm2 upon Rps27l disruption . Rps27l+/+ or Rps27l−/− MEFs were harvested at various time points post CHX treatment for IB ( top ) . Densitometry quantification was performed with ImageJ , and the decay curves are shown ( bottom ) . ( B ) Rps27l disruption impairs the ubiquitylation of endogenous p53 in vivo . Rps27l+/+ or Rps27l−/− MEFs were harvested after 4 hr of MG132 treatment for IP with p53 Ab or normal IgG control , followed by IB with p53 Ab ( top ) , or for direct IB with p53 or Rps27l Ab ( bottom ) . * denotes a nonspecific band . WCE: whole cell extract . ( C ) Rps27l disruption impairs p53 ubiquitylation in vitro . The Mdm2-Mdm4 E3 and p53 ( substrate ) complex was immunoprecipitated with Mdm2 Ab from MG132 treated Rps27l+/+ or Rps27l−/− MEFs , and added into an in vitro ubiquitylation reaction mixture containing ATP , ubiquitin , E1 , and E2 ( UbcH5b ) . After 60 min incubation with continuous vortexing , the reaction mixture was subjected to IB using p53 Ab . SE: short exposure . ( D ) Reduced Mdm2 self-ubiquitylation upon Rps27l depletion . Rps27l+/+ or Rps27l−/− MEFs were treated with MG132 for 4 hr before being harvested for IB using indicated Abs . ( E ) Negative regulation of MDM2 protein levels by RPS27L . p53-null H1299 lung cancer cells were transfected with MDM2 alone , or in combination with two concentrations of FLAG-tagged RPS27L for 48 hr ( top ) , or infected with lenti-virus targeting RPS27L or scrambled control siRNA for 3 days ( bottom ) , followed by IB with indicated Abs . ( F ) Reduced Mdm4 protein level in Rps27l−/− MEFs . Two independent pairs of MEFs were harvested and subjected to IB with indicated Abs . ( G ) Reduced Mdm2-Mdm4 complex in Rps27l−/− MEFs . Whole cell extracts of Rps27l+/+ or Rps27l−/− MEFs were subjected to IP with Mdm2 Ab or normal IgG control , followed by IB with Mdm4 Ab ( left ) , or subjected to IP with Mdm4 Ab or normal IgG control , followed by IB with Mdm2 Ab ( right ) . WCE were also subjected to direct IB with indicated Abs . Densitometry quantification was performed with ImageJ . ( H ) Sequential change in the protein levels of Mdm2 , Mdm4 , and p53 upon acute depletion of Rps27l . Rps27lfl/fl MEFs were harvested at various time points following adenoviral infection and subjected to IB . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 00910 . 7554/eLife . 02236 . 010Figure 4—figure supplement 1 . Rps27l regulates p53 ubiquitylation and Mdm2 protein half-life . ( A ) No change in p53 transcription , but an increase in transcription of p53 targets . Four independent pairs of MEFs were harvested for quantitative real-time PCR using indicated primers . Shown are mean ± SEM ( n = 4 ) . p < 0 . 01 , as compared with Rps27l+/+ MEFs . ( B ) No change in p53 protein synthesis . Two independent pairs of MEFs were treated with 50 μM MG132 for 1 hr , and then labeled with [35S]-methionine , followed by IP with p53 Ab . Immunoprecipitates ( top ) , along with whole cell extract ( bottom ) , were then subjected to SDS-PAGE and autoradiography . ( C ) Reduced p53 poly-ubiquitylation upon Rps27l depletion . Rps27l+/+ or Rps27l−/− MEFs were treated with MG132 for 4 hr before harvesting for IB using indicated Abs . SE: short exposure . ( D ) Negative regulation of MDM2 protein half-life by RPS27L . H1299 cells were transfected with MDM2 alone , or in combination with RPS27L for 48 hr ( top , left ) , or infected with lenti-virus targeting RPS27L or scrambled control siRNA for 3 days ( bottom , left ) . Cells were then treated with CHX for various time periods before being harvested for IB . Densitometry quantification was performed with ImageJ , and the decay curves are shown ( right panels ) . ( E ) Sequential change in the protein levels of Mdm2 , Mdm4 , and p53 upon acute depletion of Rps27l . Rps27lfl/fl;Trp53−/− MEFs were harvested at various time points following infection with indicated adenovirus and subjected to IB . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 010 We have shown that Rps27l disruption increases the levels of both Mdm2 mRNA ( Figure 4—figure supplement 1A ) and protein ( Figure 3C–E ) . We then measured the Mdm2 protein half-life and Mdm2 self-ubiquitylation in Rps27l+/+ vs Rps27l−/− MEFs and found that Rps27l disruption extends Mdm2 protein half-life from ∼18 min to 40 min ( Figure 4A ) , which is likely attributable to a decreased Mdm2 self-ubiquitylation ( Figure 4D ) . We further used p53-null H1299 cells to determine this RPS27L-MDM2 inverse relationship in a p53-independent manner . Consistently , we found that ectopic expression of RPS27L decreases MDM2 protein in a dose dependent manner , whereas RPS27L knockdown increases MDM2 protein ( Figure 4E ) . Furthermore , ectopic RPS27L expression shortens the MDM2 protein half-life , while RPS27L knockdown extends it ( Figure 4—figure supplement 1D ) . Thus , Rps27l disruption stabilizes Mdm2 by extending its protein half-life likely via reducing its self-ubiquitylation . It appears paradoxical that an increased Mdm2 leads to a decreased p53 ubiquitylation and degradation . Given that Mdm2 is known to bind to its family member , Mdm4 , to form the most active heterodimer E3 ligase toward p53 ( Tanimura et al . , 1999; Kostic et al . , 2006; Kawai et al . , 2007 ) , we determined whether reduced Mdm2 E3 toward p53 upon Rps27l disruption is due to a reduced amount of Mdm4 . Indeed , total cellular levels of Mdm4 were significantly reduced in several independent isolates of Rps27l−/− MEFs , as compared to the Rps27l+/+ littermates ( Figure 4F ) . We then determined the binding affinity of Mdm2 and Mdm4 by the two-reciprocal immunoprecipitation ( IP ) assay in Rps27l+/+ vs Rps27l−/− MEFs under unstressed native condition in the absence of proteasome inhibitor MG132 . Although the total cellular level of Mdm4 was much lower in Rps27l−/− than in Rps27l+/+ MEFs with a ratio of 0 . 12 vs 1 , the level of Mdm2-bound Mdm4 ( pulled-down by Mdm2 IP ) was much higher with a ratio of 0 . 75 vs 1 ( Figure 4G , left ) . Reciprocally ( Mdm4 IP ) , the level of Mdm4-bound Mdm2 was also higher with a ratio of 0 . 32 vs 1 ( Figure 4G , right ) . Thus , although a much lower total levels of Mdm4 in Rps27l−/− MEFs , most of Mdm4 molecules were found in the complex with Mdm2 . These results clearly demonstrated that in the absence of Rps27l , Mdm2 actually has a higher binding affinity towards Mdm4 , which might facilitate Mdm2-mediated Mdm4 degradation , leading to a decreased level of Mdm4 , compromised Mdm2-Mdm4 ligase activity towards p53 , and consequent p53 accumulation . To elucidate the complicated interplays among p53-Mdm2-Mdm4-Rps27l , we generated Rps27lfl/fl mice ( unpublished data ) and determined the initial and/or sequential event ( s ) that cause ( s ) the change in the Mdm2-Mdm4-p53 proteins upon acute depletion of Rps27l . Rps27lfl/fl MEFs were infected with adenovirus expressing Cre recombinase ( Ad-Cre ) to deplete Rps27l . Compared to the Ad-GFP controls , acute depletion of Rps27l caused an elevated Mdm2 , starting at 24 hr post Ad-Cre infection and lasting up to 96 hr , followed by depletion of Mdm4 , starting at 48 hr . A moderate accumulation of p53 was not seen until the later time points , starting at 72–84 hr following Ad-Cre infection ( Figure 4H ) . Furthermore , Mdm2 accumulation , followed by Mdm4 reduction , is independent of p53 , since a similar changing pattern was seen when Rps27lfl/fl;Trp53−/− MEFs were infected with Ad-Cre ( Figure 4—figure supplement 1E ) . Taken together , these results clearly demonstrated that upon Rps27l depletion ( not necessary for complete elimination ) , Mdm2 increases first , followed by an Mdm4 decrease , and finally a p53 increase . These sequential changes supported the notion that Rps27l depletion somehow stabilizes Mdm2 , which promotes Mdm4 degradation , as also shown previously ( de Graaf et al . , 2003; Kawai et al . , 2003; Pan and Chen , 2003 ) , leading to a suboptimal Mdm2-Mdm4 complex with reduced ligase activity toward p53 , and eventually p53 accumulation . By amino sequence comparison , Rps27l is a family member of ribosomal protein S27 ( Rps27 ) ( Xiong et al . , 2011 ) . However , it has never been determined previously whether Rps27l is indeed a ribosomal protein . By ribosomal profiling we found that like its family member Rsp27 , Rps27l is localized in the ribosomes , along with other known Mdm2-binding ribosomal proteins , including Rpl5 , Rpl11 , and Rps7 , and that Rps27l disruption does not cause obvious alterations in ribosomal profile , given the profiling patterns are largely overlapping with each other , except for the 40S peak , which is relatively lower in Rps27l−/− MEFs ( Figure 5A ) . 10 . 7554/eLife . 02236 . 011Figure 5 . Rps27l is a ribosomal protein whose depletion triggers ribosomal stress . ( A ) Rps27l is a ribosomal protein . Rps27l+/+ or Rps27l−/− MEFs were treated with CHX for 30 min before harvesting for ribosomal profiling . The cytoplasmic extracts were loaded on sucrose gradients ( 10%–50% ) and subjected to ultra-centrifugation . Gradients were then fractionated and measured by optical density at 254 nm ( top ) . The fractions from Rps27l+/+ MEFs were subjected to IB using indicated Abs ( bottom ) . ( B ) B23 is released from nucleoli upon Rps27l depletion . MEFs with indicated genotypes were left untreated or treated with 5 nM Act D for 24 hr , followed by immunofluoresent staining of B23 ( left ) . Scale bar represents 20 µm . Cells with nucleolus and/or nucleoplasmic B23 staining were counted and expressed as percentage of total cells ( at least 200 ) counted ( right ) . Shown are mean ± SD . p < 0 . 01; p < 0 . 05 . ( C ) The change in Mdm2 binding of various ribosomal proteins in Rps27l−/− MEFs . Rps27l+/+ or Rps27l−/− MEFs were treated with 5 nM Act D for 4 hr before being harvested for IP with Mdm2 Ab or normal IgG control , followed by IB with indicated Abs ( left ) , or for direct IB with indicated Abs ( right ) . Densitometry quantification was performed with ImageJ . ( D and E ) Activation of p53 by Rps27l-deficiency is dependent on Rpl11 , but not Rpl5 . Two independent pairs of MEFs were infected with lentivirus expressing short hairpin RNA ( shRNA ) against GFP as a negative control or against Rpl11 ( D ) or Rpl5 ( E ) before being harvested for IB with indicated Abs . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 01110 . 7554/eLife . 02236 . 012Figure 5—figure supplement 1 . Rps27l disruption has minimal effects on rDNA transcription , rRNA processing and/or protein synthesis . ( A ) Rps27l loss does not impair the transcription of rDNA . Four independent pairs of MEFs were harvested for quantitative real-time PCR using primers for 45S rRNA . Shown are mean ± SEM ( n = 4 ) . ( B ) No change in 28S and 18S rRNA levels upon Rps27l disruption . Four pairs of MEFs from two independent litters were harvested for RNA extraction . Equal amounts of RNA were analyzed by formaldehyde gel electrophoresis with ethidium bromide staining . ( C and D ) Rps27l loss moderately affects the kinetics of rRNA processing . Two independent pairs of MEFs were pulse labeled with L-[methyl-3H]-methionine , and chased for the indicated times . Total RNA was isolated and subjected to Northern blotting . Membrane was autographed . Ethidium bromide staining of 28S and 18S rRNA is shown in the bottom panel as a loading control ( C ) . The band density of 18S rRNA and 47S pre-RNA were quantified with ImageJ . The relative ratios of 18S rRNA to 47S pre-RNA , with the value of the 18S/47S ratio in Rps27l+/+ cells at 0 min arbitrarily set at 1 , are shown ( D ) . ( E and F ) Rps27l loss had no effect on general protein synthesis . Three independent pairs of MEFs were labeled with [35S]-methionine for 1 hr . Total [35S]-Met incorporation was measured using liquid scintillation counting ( E ) . Equal amounts of protein were subjected to SDS-PAGE ( F ) . Cycloheximide ( CHX ) ( 100 µg/ml ) was used to inhibit new protein synthesis , serving as a labeling control . ( G ) No change in the levels of various ribosomal proteins known to bind to Mdm2 . Two independent pairs of MEFs were harvested and subjected to IB with indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 012 We next determined whether Rps27l disruption causes aberrant ribosomal assembly . We first used qRT-PCR analysis and found that Rps27l disruption had no effect on the levels of 45S rRNA ( Figure 5—figure supplement 1A ) , indicating Rps27l loss does not impair the transcription of rDNA . Ethidium bromide staining also showed a similar level of mature forms of 28S and 18S rRNAs ( Figure 5—figure supplement 1B ) . We then performed pulse-chase labeling of rRNA and found that Rps27l loss caused a moderate decrease in the kinetics of rRNA processing from 47S to 18S ( Figure 5—figure supplement 1C , D ) . Finally , we found that Rps27l loss had no effect on global protein synthesis , as determined by a [35S]-methionine pulse-chase experiment ( Figure 5—figure supplement 1E , F ) . Taken together , we conclude that p53 induction upon Rps27l disruption is unlikely due to aberrant protein synthesis , but reduced rRNA processing may contribute to some extent . It is established that upon ribosomal stress a number of ribosomal proteins are released from nucleoli to bind to Mdm2 , leading to its stabilization ( Zhang and Lu , 2009 ) . We then determined if Rps27l disruption triggers ribosomal stress by staining MEFs with B23/NPM , a nucleolus protein known to be released to nucleoplasm upon ribosomal stress , as a marker ( Jin et al . , 2004 ) . While only ∼7% of Rps27l+/+ MEFs had nucleoplasm staining , ∼25% of Rps27l−/− MEFs had positive nucleoplasm staining ( p <0 . 01 ) ( Figure 5B ) . Positive control Act D caused nucleolar B23 release in nearly all of MEFs , regardless of Rps27l status ( Figure 5B ) . Since p53 has been shown to constrain the ribosome biogenesis ( Golomb et al . , 2014 ) , we determined potential involvement of p53 in the process and found that nucleolar B23 release was also significantly increased in Rps27l−/−;Trp53−/− MEFs , although to a lesser extent ( 7 vs 17% ) ( Figure 5B ) . Thus , depletion of Rps27l triggers ribosomal stress to cause B23 release from nucleoli , which is largely independent of p53 . We further determined the levels of Mdm2-bound ribosomal proteins as an independent marker for ribosomal stress ( Zhang and Lu , 2009; Zhou et al . , 2012 ) . Although there is no measurable difference between Rps27l+/+ and Rps27l−/− MEFs ( two independent pairs ) in total cellular levels of several ribosomal proteins known to bind to Mdm2 , including Rpl5 , Rpl11 , Rpl23 , Rps7 , Rps14 , Rps19 ( Figure 5—figure supplement 1G ) , we did detect in Rps27l−/− MEFs an increased Mdm2 binding of Rpl11 , Rpl23 , Rps7 , Rps14 , and Rps27 , but not Rpl5 , Rps27a , and Rps19 ( Figure 5C ) . Moreover , since recent studies have documented the central role of RPL5 and RPL11 in ribosomal stress-induced p53 activation ( Bursac et al . , 2012; Golomb et al . , 2014 ) , we determined whether activation of p53 by Rps27l-deficiency is dependent on Rpl5 and Rpl11 by siRNA-based silencing approach . In two pairs of MEFs derived from two independent littermates , silencing of Rpl11 ( with enhanced Mdm2 binding upon Rps27l disruption ) , but not Rpl5 ( without changing Mdm2 binding ) abrogated the p53 activation in Rps27l−/− MEFs , suggesting activation of p53 by Rps27l-deficiency is dependent on Rpl11 , but not Rpl5 ( Figure 5D , E ) . Interestingly , we reproducibly observed that Rpl5 silencing increased p53 levels in MEFs using three independent Rpl5-targeting shRNAs ( Figure 5E and data not shown ) . Taken together , our results suggest that Rps27l disruption triggers ribosomal stress to increase the binding of Mdm2 to selective sets of ribosomal proteins , particularly Rpl11 , leading to Mdm2 stabilization . We then addressed mechanistically how stabilized Mdm2 failed to promote p53 degradation in Rps27l−/− MEFs with focus on Mdm4 . We found that substantial reduction of Mdm4 in Rps27l−/− MEFs was not due to decreased Mdm4 mRNA transcription ( Figure 6—figure supplement 1A ) , nor decreased Mdm4 mRNA translation ( data not shown ) , but enhanced degradation , since treatment with proteasome inhibitor , MG132 for 4 hr caused its accumulation ( Figure 6—figure supplement 1B ) . It is noteworthy that MG132-induced accumulation of Mdm2 and p53 is much more substantial , given that both Mdm2 and p53 have a much shorter protein half-life . Consistent with the fact that Mdm2 promotes Mdm4 ubiquitylation and degradation ( de Graaf et al . , 2003; Kawai et al . , 2003; Pan and Chen , 2003 ) under some stressed conditions including ribosomal stress ( Gilkes et al . , 2006 ) , we found Mdm4 polyubiquitylation was indeed increased with accompanied reduction of Mdm4 levels in Rps27l−/− MEFs ( Figure 6A ) . Furthermore , in p53-null H1299 cells , MDM2-mediated MDM4 polyubiquitylation was substantially inhibited by ectopically expressed RPS27L ( Figure 6B ) . A similar result was seen in 293 cells ( Figure 6—figure supplement 1C ) . Consistently , the Mdm4 protein half-life was significantly shortened upon Rps27l disruption in MEFs ( Figure 6C ) or upon RPS27L knockdown in H1299 cells ( Figure 6—figure supplement 1D ) , but significantly extended upon ectopic expression of RPS27L in H1299 cells ( Figure 6—figure supplement 1E ) . Taken together , our study demonstrated that stabilized Mdm2 upon Rps27l depletion indeed promotes ubiquitylation and degradation of Mdm4 to reduce its cellular levels . Finally , we found a consistent increase of p53 and Mdm2 and a decrease of Mdm4 in several independent Rps27l-null tissues of brain and liver ( Figure 6D , E ) . 10 . 7554/eLife . 02236 . 013Figure 6 . Rps27l depletion leads to Mdm2-mediated Mdm4 degradation . ( A ) Increased Mdm4 ubiquitylation upon Rps27l disruption . Rps27l+/+ or Rps27l−/− MEFs were treated with MG132 before being harvested for IP with Mdm4 Ab or normal IgG control , followed by IB with Mdm4 Ab ( top ) , or for direct IB with indicated Abs ( bottom ) . SE: short exposure . ( B ) MDM2-mediated MDM4 ubiquitylation is inhibited by ectopic RPS27L . H1299 cells were transfected with indicated plasmids . At 24 hr post-transfection , cells were treated with MG132 for 4 hr , and then harvested for purification of His-tagged ubiquitylated proteins by Ni-NTA beads , followed by IB with Myc-tag Ab to detect MDM4 ubiquitylation ( top ) , or for direct IB with indicated Abs ( bottom ) . ( C ) Shortened Mdm4 protein half-life upon Rps27l disruption . Rps27l+/+ or Rps27l−/− MEFs were harvested at various time points post CHX treatment for IB ( top ) . Densitometry quantification was performed with ImageJ , and the decay curves are shown ( bottom ) . ( D and E ) Increased protein levels of p53 and Mdm2 and decreased protein level of Mdm4 in Rps27l−/− brains ( D ) and livers ( E ) . Brain and liver tissues from P6 pups with indicated genotypes were harvested and lysed for IB . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 01310 . 7554/eLife . 02236 . 014Figure 6—figure supplement 1 . Rps27l regulates Mdm2-mediated Mdm4 degradation . ( A ) No change in Mdm4 mRNA transcription . Four independent pairs of MEFs were harvested for qRT-PCR using indicated primers . Shown are mean ± SEM ( n = 2 ) . ( B ) Reduced Mdm4 protein level in Rps27l−/− MEFs is due to enhanced degradation . The paired MEFs were left untreated or treated with MG132 for 4 hr , followed by IB . ( C ) MDM2-mediated MDM4 ubiquitylation is inhibited by ectopically expressed RPS27L . The 293 cells were transfected with indicated plasmids . At 48 hr post-transfection , cells were treated with MG132 for 4 hr , lysed , followed by purification of His-tagged ubiquitylated proteins by Ni-NTA beads , followed by IB with Myc-tag Ab ( top ) , or for direct IB with indicated Abs ( bottom ) . ( D and E ) Regulation of MDM4 protein half-life by RPS27L . H1299 cells were infected with lenti-virus targeting RPS27L or scrambled control siRNA for 3 days ( D ) , or transfected with MYC-MDM4 and MDM2 alone , or in combination with RPS27L for 48 hr ( E ) . Cells were then treated with CHX for various time periods before being harvested for IB . Densitometry quantification was performed with ImageJ , and the decay curves are shown ( bottom panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 014 We next determined whether increased p53 plays a causal role in developmental defects seen in Rps27l−/− mice by simultaneous deletion of Trp53 . Intercrossing of Rps27l+/−;Trp53+/− mice gave rise to mice with the Rps27l−/− background in combination with three p53 genotypes . All mice with Rps27l−/−;Trp53+/+ background die within 3 weeks of age with one mouse living up to 35 days ( Figure 1—figure supplement 1E , Figure 7A ) . Consistently , HSPCs in bone marrow were significantly reduced ( Figure 7—figure supplement 1A , B ) , as well as the peripheral blood cells ( Figure 7—figure supplement 1C , D ) . In contrast , both Rps27l−/−;Trp53+/− and Rps27l−/−;Trp53−/− mice developed normally and are fertile with expected Mendelian distribution ( Figure 7A ) , indicating a complete rescue of the postnatal death by either heterozygous or homozygous deletion of Trp53 . Note that the life-span greater than 5 weeks was used in Figure 7A , given that the longest life-span of a Rps27l−/−;Trp53+/+ mouse is 5 weeks ( Figure 1—figure supplement 1E ) . At the organ and cellular levels , Trp53 deletion of either one or both alleles rescued overall defective phenotypes seen in Rps27l−/− mice as follows: ( 1 ) growth retardation , as reflected by the recovery of reduced body size and weight ( Figure 7B , C ) ; ( 2 ) organ specific hypo-cellularity , as reflected by the recovery of total cell numbers in spleen , thymus and bone marrow ( Figure 7D , E ) . On the other hand , however , the recovery of bone marrow HSPCs depletion was completely rescued by homozygous but not by heterozygous Trp53 deletion , since some of HSPCs were only partially recovered in Rps27l−/−;Trp53+/− mice ( Figure 7F , G ) , although it does not affect the survival of mice . Partial to complete recovery was seen in peripheral blood cells ( Figure 7H ) as well as in differentiated lineages from peripheral blood ( Figure 7—figure supplement 1E ) . 10 . 7554/eLife . 02236 . 015Figure 7 . Simultaneous deletion of Trp53 rescues growth retardation and HSPCs depletion . ( A ) Deletion of Trp53 rescues postnatal death by Rps27l disruption . Lower than expected number of mice with Trp53−/− genotype ( regardless of Rps27l genotype ) is due to high frequency of developmental abnormalities during embryonic and neonatal stages which cause the premature death ( Armstrong et al . , 1995; Sah et al . , 1995 ) . ( B–D ) Deletion of Trp53 rescues growth retardation and organ hypocellularity . Representative mice at P18 of three genotypes were photographed ( B ) . The bodies ( C ) were weighed; and the total cell numbers ( D ) of bone marrow ( femur and tibia from two hind limbs ) , spleen , and thymus were counted from P18 mice with genotypes of Rps27l+/+;Trp53+/+ ( n = 3 ) , Rps27l−/−;Trp53+/+ ( n = 7 ) , Rps27l−/−-;Trp53+/− ( n = 10 ) , Rps27l−/−;Trp53−/− ( n = 5 ) . Shown are mean ± SD . p < 0 . 05 , p < 0 . 01 , and p < 0 . 001 . ( E ) Representative H&E staining of bone marrows in femurs from P18 mice . Scale bars represent 200 µm ( top ) or 40 µm ( bottom ) . ( F and G ) Deletion of Trp53 rescues HSPCs depletion in Rps27l−/− bone marrow . The percentage of LSK , MPP , ST-HSC , and LT-HSC ( F ) ; and the percentage of MP , CMP , GMP , and MEP ( G ) in bone marrow from P18 mice with genotypes of Rps27l+/+;Trp53+/+ ( n = 4 ) , Rps27l−/−;Trp53+/+ ( n = 5 ) , Rps27l−/−;Trp53+/− ( n = 7 ) , and Rps27l−/−;Trp53−/− ( n = 5 ) . LSK: Lin−/Sca-1−/c-Kit+; MPP: Lin−/Sca-1−/c-Kit+/CD48+/CD150−; ST-HSC: Lin−/Sca-1−/c-Kit+/CD48+/CD150+; LT-HSC: Lin−/Sca-1−/c-Kit+/CD48−/CD150+ . Shown are mean ± SD . p < 0 . 05 , p < 0 . 01 , and p < 0 . 001 . ( H ) Deletion of Trp53 rescues defects in Rps27l−/− peripheral blood . CBC classification of peripheral blood from Rps27l+/+;Trp53+/+ ( n = 3 ) , Rps27l−/−;Trp53+/+ ( n = 7 ) , Rps27l−/−;Trp53+/− ( n = 10 ) , Rps27l−/−;Trp53−/− ( n = 5 ) mice at P18 was performed . WBC , white blood cells; NE , neutrophils; LY , lymphocytes; MO , monocytes; HCT , hematocrit; RBC , red blood cells; Hb , hemoglobin; PLT , platelets . Shown are mean ± SD . p < 0 . 05 , p < 0 . 01 , and p < 0 . 001 . ( I and J ) Deletion of Trp53 rescues HSPCs depletion in Rps27l−/− fetal livers . Flow cytometry analysis was performed to measure the percentage of HSPCs including LSK ( I ) , MP , CMP , GMP , and MEP ( J ) in E14 . 5 fetal livers with genotypes of Rps27l−/−;Trp53+/+ ( n = 5 ) , Rps27l−/−;Trp53+/− ( n = 7 ) , and Rps27l−/−;Trp53−/− ( n = 6 ) . Shown are mean ± SD . p < 0 . 05 , p < 0 . 01 , and p < 0 . 001 , as compared to Rps27l−/−;Trp53+/+ . ( K ) Kaplan–Meier survival curves of recipient mice after transplantation . Fetal liver cells ( 2 × 106 cells ) from E14 . 5 embryos with indicated genotypes were respectively injected into lethally irradiated recipient mice ( n = 7 for each genotype ) . p < 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 01510 . 7554/eLife . 02236 . 016Figure 7—figure supplement 1 . Simultaneous deletion of Trp53 rescues defective phenotypes caused by Rps27l disruption . ( A and B ) Significant HSPCs decrease in bone marrow of Rps27l−/− mice . Bone marrow cells from Rps27l+/+ ( n = 4 ) , Rps27l+/− ( n = 4 ) , and Rps27l−/− ( n = 5 ) mice at P18 under a mixed Sv129/B6 background were stained with Abs against surface markers , followed by FACS analysis . The total cell numbers of LSK , MPP , ST-HSC , and LT-HSC ( A ) and the total numbers of MP , CMP , GMP , and MEP ( B ) were summarized . Shown are mean ± SD . p < 0 . 05 , *p < 0 . 01 , as compared to Rps27l+/+ counterparts . ( C and D ) Depletion of peripheral blood cells in Rps27l−/− mice . Shown are CBC classification of peripheral blood cells ( C ) and the cell density of leukocytes with different surface markers in peripheral blood ( D ) from Rps27l+/+ ( n = 3 ) , Rps27l+/− ( n = 4 ) , and Rps27l−/− ( n = 7 ) mice at P18 under a mixed Sv129/B6 background . Shown are mean ± SD . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 , as compared with Rps27l+/+ counterparts . ( E ) Deletion of Trp53 rescues defects in peripheral blood cells . The cell density of leukocytes with different surface markers in peripheral blood from Rps27l+/+;Trp53+/+ ( n = 3 ) , Rps27l−/−;Trp53+/+ ( n = 7 ) , Rps27l−/−;Trp53+/− ( n = 10 ) , Rps27l−/−;Trp53−/− ( n = 5 ) mice at P18 were summarized . Shown are mean ± SD . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 . ( F ) Deletion of Trp53 rescues the reconstitution defects . The percentage of CD45 . 2+ , B220+ , CD3+ , and Mac-1+ cells derived from the indicated donor mice at 4 weeks post transplantation were measured and plotted . Shown are mean ± SD . p < 0 . 01 , *p < 0 . 001 , as compared with Rps27l+/+;Trp53+/+ ( n = 7 for each genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 016 We further determined whether Trp53 deletion rescues hematopoietic failure of Rps27l−/− fetal liver . Indeed , the percentage of HSPCs was significantly increased upon deletion of one or both alleles of Trp53 ( Figure 7I , J ) . More importantly , in a non-competitive reconstitution assay , inability of Rps27l−/− fetal liver cells to reconstitute sterilized bone marrow was rescued by heterozygous or homozygous deletion of Trp53 ( Figure 7K ) . The percentage of donor derived cells in rescued chimeric mice approached the wild type control ( Figure 7—figure supplement 1F ) . Thus , p53 increase upon Rps27l disruption is fully responsible for phenotypic defects observed in Rps27l−/− mice . The long-term animal survival studies revealed that compared to Rps27l+/+;Trp53−/− mice , Rps27l−/−;Trp53−/− mice had a similar life-span with ∼50% of mice dying mainly of lymphoma at age of ∼150 days ( Figure 8A , p = 0 . 713 , log-rank test ) , indicating Rps27l genotype has no effect on the life-span and lymphomas development in the Trp53−/− mice . However , compared to Rps27l+/+;Trp53+/− and Rps27l+/−;Trp53+/− mice , Rps27l−/−;Trp53+/− mice had a significantly shortened life-span ( Figure 8B , p = 0 . 0036 , log-rank test ) . Whole body necropsy revealed that out of 14 Rps27l−/−;Trp53+/− mice died within a period of 400 days ( 50% death rate ) , seven developed T-lymphoblastic lymphoma ( Figure 8C , D ) , one developed T-cell lymphoma detected in tissues of thymus , lymph node and spleen ( Figure 8—figure supplement 1A ) , and one developed B-cell lymphoma seen in a much enlarged lymph node ( Figure 8—figure supplement 1B ) . The remaining mice had enlarged liver and/or spleen ( not shown ) . The results indicate that Rps27l disruption accelerated the formation of spontaneous lymphomas under the Trp53+/− background . We then collected more lymphoma tissues from Rps27l−/−;Trp53+/− mice at age of 4–6 months for Trp53 genotyping and found that the wild type Trp53 allele was deleted in 29 out of 30 ( 97% ) lymphomas genotyped ( Figure 8—figure supplement 1C , D ) , indicating that Rps27l disruption imposed the selection pressure against wild-type Trp53 . We further measured the genome integrity of these lymphoma cells by metaphase chromosome spread and found a high degree of aneuploidy in up to 63% of total population ( Figure 8E , Figure 8—figure supplement 1E ) . Centromere–centromere fusions were also seen in some diploid lymphoma cells ( Figure 8E , arrow ) . Thus , Rps27l disruption causes genomic instability and Trp53 deletion , eventually leading to lymphoma development . To determine which event ( genomic instability vs Trp53 deletion ) occurs first , we performed the same metaphase chromosome spread in four independent pairs of early passage primary MEFs and found a significant higher rate of chromosomal aneuploidy in Rps27l−/−;Trp53+/− cells than in Rps27l+/+;Trp53+/− cells ( 30% vs 15% ) ( Figure 8F ) , although wild-type Trp53 was retained , as measured by qPCR ( Figure 8G ) . Thus , Rps27l disruption triggers genomic instability prior to Trp53 deletion and the cells with subsequent Trp53 deletion are selected , which outgrow to form spontaneous lymphoma . We further performed the same metaphase chromosome spread in MEFs with the genotype of Rps27l−/−;Trp53+/+ ( n = 7 ) vs Rps27l+/+;Trp53+/+ ( n = 3 ) and found a same low rate of aneuploidy at ∼10% in both group of MEFs ( Figure 8H ) . Thus , Rps27l disruption triggers genomic instability only when one allele of Trp53 is deleted or mutated . 10 . 7554/eLife . 02236 . 017Figure 8 . Rps27l disruption induces genomic instability and spontaneous lymphoma . ( A and B ) Kaplan–Meier survival curves of Trp53−/− ( A ) and Trp53+/− ( B ) mice with three Rps27l genotypes and indicated numbers of mice . p = 0 . 713 ( A ) ; p = 0 . 0036 ( B ) . ( C ) Representative pictures of lymphomas developed from Rps27l−/−; Trp53+/− mice . ( D ) Representative FACS profiles of T cells from lymphomas . Lymphoma cells were isolated and subjected to FACS analysis using Abs against indicated surface markers . ( E ) Representative pictures of metaphases from four tumors developed from Rps27l−/−;Trp53+/− mice ( top ) . Chromosome numbers ( n ) were counted . Scale bar represent 10 μm . Frequency of diploid and aneuploidy from four tumors developed in Rps27l−/−;Trp53+/− mice with 100 metaphases counted ( bottom ) . ( F ) Frequency of diploid and aneuploidy from four pairs of primary MEFs at P2 and P3 derived from Rps27l−/−;Trp53+/− vs Rps27l+/+;Trp53+/− embryos with at least 100 metaphases counted in each sample . Shown are mean ± SD . p < 0 . 05 , p < 0 . 01 , and *p < 0 . 001 . ( G ) Trp53 dosage in four pairs of primary MEFs at P3 derived from Rps27l−/−;Trp53+/− ( R−/−;T+/− ) vs Rps27l+/+;Trp53+/− ( R+/+;T+/− ) embryos . Amounts of p53 DNA in individual MEFs were quantified by qPCR using three sets of primers for exons 5 , 6 , and 7 . The combined results ( n = 3 for each primer set , mean ± SD ) were presented with the values from Rps27l+/+;Trp53+/− MEFs averaged and set as 1 . ( H ) Frequency of diploid and aneuploidy from primary MEFs at P2 and P3 derived from Rps27l−/−;Trp53+/+ ( n = 7 ) vs Rps27l+/+;Trp53+/+ ( n = 3 ) individual embryos with at least 100 metaphases counted in each sample . Shown are mean ± SD . p = 0 . 15 . ( I and J ) Increased levels of p53 and p53 targets and decreased levels of Arf upon Rps27l disruption in Trp53+/− MEFs . Primary MEFs at P2 and P3 derived from Rps27l+/+;Trp53+/− and Rps27l−/−;Trp53+/− embryos were left untreated ( I ) or treated with 5 nM ActD for indicated time periods ( J ) before being harvested for IB with indicated Abs . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 01710 . 7554/eLife . 02236 . 018Figure 8—figure supplement 1 . Rps27l disruption induces genomic instability and spontaneous lymphoma . ( A ) Representative staining with H&E and CD3 , a T cell marker of indicated tissues from a Rps27l−/−;Trp53+/− mouse . Scale bar = 50 µm . ( B ) Representative staining with H&E and B220 , a B cell marker of the spleen tissue from a Rps27l−/−;Trp53+/− mouse . Scale bar = 50 µm . ( C and D ) Deletion of wild type Trp53 allele in T-cell ( C ) and B-cell lymphoma ( D ) . Genomic DNA from tails ( C ) and lymphomas ( T ) were extracted and amplified with primers to detect indicated Trp53 exons . ( E ) Representative pictures of metaphases from four tumors developed in Rps27l−/−;Trp53+/− mice . Chromosome numbers ( n ) were counted . Scale bar = 10 μm . ( F ) p53 level and activity are higher in Rps27l−/−; Trp53+/+ than in Rps27l−/−; Trp53+/− MEFs . Primary MEFs derived from embryonic littermates with indicated genotypes were harvested for IB with indicated Abs . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 018 Our results implied that Rps27l depletion is responsible for genomic instability by a mechanism that appears to be independent of p53 loss . To further test this , we measured the levels of p53 and its upstream and downstream regulators in three independent pairs of MEFs of Rps27l−/−; Trp53+/− vs the littermates of Rps27l+/+; Trp53+/− , and found that in every case , both basal and induced ( by ActD ) levels of p53 and p53 downstream ( p21 and Mdm2 ) are higher in Rps27l-null MEFs than in Rps27l-wt MEFs , whereas the p19/Arf level was lower in the former ( Figure 8I , J ) . These results demonstrated that Rps27l disruption induces genomic instability under Trp53+/− background , which triggers p53 to balance the genome , and is indeed independent of p53 loss . On the other hand , Rps27l disruption also confers a selective pressure against p19/Arf , a p53 upstream regulator , although the underlying mechanism is unclear . Nevertheless , p19/Arf reduction would likely contribute to observed Mdm2 increase to degrade Mdm4 , leading to p53 accumulation . Finally , it is worth noting that p53 level and activity were higher in Rps27l−/−; Trp53+/+ than in Rps27l−/−;Trp53+/− MEFs ( Figure 8—figure supplement 1F ) , and this difference , likely extendable to other organs , such as bone marrow , determines the life and death of mice at the postnatal stages . In this study , we demonstrated , using a gene-trap mouse model that Rps27l , a p53 downstream ribosomal protein and an Mdm2 E3 ligase substrate , is essential for mouse development . Rps27l disruption causes a moderate increase of p53 which is sufficient to deplete HSPCs in fetal livers and bone marrows , eventually to cause the postnatal death . By both noncompetitive and competitive reconstruction assays , we clearly demonstrated that the HSPCs from Rps27l−/− fetal livers were unable to reconstitute sterilized bone marrow . Rescuing of these defects by simultaneous deletion of a single or both alleles of Trp53 indicates that p53 plays a causal role and that a moderate p53 increase is responsible for and sufficient to cause the depletion of these HSPCs via inducing apoptosis . It is noting worthy that although , unlike homozygous Trp53 deletion , heterozygous Trp53 deletion fails to completely rescue the defects in number or percentage of HSPCs in the bone marrows in p18 weaning mice , it is still sufficient to rescue the postnatal growth retardation and death . An increasing list of ribosomal proteins ( RPs ) has been shown in numerous in vitro cell culture studies to bind to Mdm2 and stabilize p53 , thus acting as a p53 activator ( Zhang and Lu , 2009; Zhou et al . , 2012 ) . It was also reported that siRNA silencing of some ribosomal proteins , such as L23 , triggered ribosomal stress , leading to p53 stabilization and activation ( Dai et al . , 2004; Jin et al . , 2004 ) . We previously showed that RPS27L is an MDM2 binding protein whose overexpression inhibits MDM2-mediated p53 ubiquitylation , whereas its silencing reduces p53 in some cancer lines ( e . g . , A549 and SJSA ) , but not in others ( e . g . , SY5Y , MCF7 , U2OS , HCT116 , and RKO ) ( Xiong et al . , 2011 ) ( data not shown ) , suggesting RPS27L might act as a p53 activator in a cancer cell dependent manner . Here we reported that Rps27l is indeed a ribosomal protein , whose depletion triggers ribosomal stress ( as evidenced by B23 nucleoplasm staining and enhanced RPs-Mdm2 binding ) with moderate reduction in ribosomal RNA processing , but no effect on general protein synthesis . We further found , using Rps27l KO model , that unlike in vitro cell culture studies which showed that almost all known MDM2-binding ribosomal proteins acts as p53 activators upon overexpression ( Zhang and Lu , 2009; Zhou et al . , 2012 ) , Rps27l actually activates p53 upon disruption , given the fact that ( a ) Rps27l deletion caused p53 accumulation in multiple organs as well as in primary MEFs , and ( b ) all Rps27l-null phenotypes , such as depletion of HPSCs and bone marrow , growth retardation and postnatal death , can be fully rescued by Trp53 deletion . It is conceivable that the discrepancy between in vitro and in vivo studies is likely attributable to the fact that most cell culture studies were conducted ( a ) in human cancer cell lines with multiple genetic alterations , ( b ) under artificial overexpression and/or ( c ) partial RNA silencing conditions . Nevertheless , through this knockout study we conclusively showed that Rps27l disruption induces p53 under the physiological conditions in a manner independent of DNA damage . It is also interesting in our finding that consistent with enhanced binding of Mdm2-Rpl11 , but not Mdm2-Rpl5 , p53 induction triggered by Rps27l disruption is dependent of Rpl11 , but independent of Rpl5 , although both are key ribosomal proteins , released from nucleolus upon ribosomal stress , to inhibit Mdm2-mediated p53 degradation ( Golomb et al . , 2014 ) . Furthermore , we made a novel observation that Rps27l disruption caused an increased binding of Mdm2 selectively to Rpl11 , Rpl23 , Rps7 , Rps14 , and Rps27 , but not to Rpl5 , Rps27a , and Rps19 . Whether these RPs with enhanced Mdm2 binding indeed contribute to p53 activation in Rps27l-deficient MEFs is an interesting topic warranting future investigation . Our Rps27l acute depletion experiment conducted in both Trp53-wt and Trp53-null MEFs clearly showed a sequential change in the levels of Mdm2 ( increase ) , Mdm4 ( decrease ) and p53 ( increase ) , indicating that ( 1 ) Rps27l depletion induces p53-independent Mdm2 increase and ( 2 ) increased Mdm2 fails to degrade p53 . Our follow-up experiments mechanistically addressed these two fundamental questions: how Mdm2 is increased upon Rps27l disruption and why increased Mdm2 fails to degrade p53 in the absence of Rps27l . For first question , we found that while Rps27l has no direct effect on Mdm2 mRNA transcription , it indeed negatively regulates the Mdm2 protein stability with Mdm2 protein half-life being extended or shortened upon Rps27l depletion or overexpression , respectively . It is conceivable that in wt cells under unstressed conditions , Rps27l binds to Mdm2 on its acidic domain ( Xiong et al . , 2011 ) , which has potential to block the binding of other ribosomal proteins ( e . g . , L11/S7/S27a/S14 ) to Mdm2 on the same domain ( internal competition ) . Upon Rps27l disruption , which triggers ribosomal stress , other ribosomal proteins are released from nucleoli and bind to Mdm2 on an unoccupied acidic domain to inhibit Mdm2 self-ubiquitylation , leading to its stabilization . To address the second question , we showed that the total Mdm4 level was significantly lower in Rps27l−/− than in Rps27l+/+ MEFs . We also showed by both loss-of-function and rescue-of-function studies that Mdm2-mediated Mdm4 ubiquitylation and degradation is negatively regulated by Rps27l , being enhanced upon Rps27l depletion and reduced upon RPS27L overexpression . Thus , enhanced Mdm2-RPs binding ( such as Mdm2-Rpl11 binding ) as a result of ribosomal stress in response to Rps27l disruption would facilitate Mdm2-dependent Mdm4 degradation , which is consistent with a previous observation ( Gilkes et al . , 2006 ) . Furthermore , we found that although Mdm2-Mdm4 has a better affinity upon Rps27l deletion , a much reduced amount of Mdm4 in Rps27l−/− MEFs results in a reduced level of Mdm2-bound Mdm4 , thus lesser formation of the Mdm2-Mdm4 heterodimer . Given that Mdm2-Mdm4 heterodimer is the most stable and active form of E3 ligase for targeted p53 degradation , as shown in both in vitro ( Kawai et al . , 2007; Wang et al . , 2011 ) and in vivo ( Francoz et al . , 2006; Garcia et al . , 2011; Huang et al . , 2011; Pant et al . , 2011 ) studies , compromised Mdm2-Mdm4 ligase activity in Rps27l−/− MEFs leads to a reduced p53 ubiquitylation and degradation , consequently an extended p53 protein half-life and increased p53 levels . While numerous gene knockout studies have shown an involvement of p53 during embryogenesis , the causal role played by p53 was best demonstrated by the total knockout of Mdm2 ( Jones et al . , 1995; Montes de Oca Luna et al . , 1995 ) or Mdm4 ( Parant et al . , 2001; Finch et al . , 2002; Migliorini et al . , 2002 ) and by the knock-in ( KI ) of mutant Mdm2C462A ( Itahana et al . , 2007 ) , Mdm4C462A ( Huang et al . , 2011 ) , or Mdm4ΔRING ( Pant et al . , 2011 ) . In all cases , p53 is accumulated to robust levels to kill the embryos at the early stage of embryogenesis , which can only be rescued by homozygous , but not heterozygous , deletion of Trp53 ( Jones et al . , 1995; Montes de Oca Luna et al . , 1995; Parant et al . , 2001; Itahana et al . , 2007; Huang et al . , 2011; Pant et al . , 2011 ) . In addition , the embryonic lethality of Mdm2−/− mice can also be rescued by Trp53515C , a hypomorphic allele of p53 , which encodes p53R172P ( Abbas et al . , 2010 ) . While Mdm2−/−;Trp53515C/515C mice also die by postnatal day 13 , the death is mainly attributable to depletion of HPSCs in postnatal bone marrows , but not in fetal liver due to p53R172P-induced ROS generation , and subsequent senescence and cell death ( Abbas et al . , 2010 ) . In contrast , Rps27l disruption causes a moderate increase of p53 which is insufficient to induce embryonic lethality , but sufficient to induce postnatal lethality , resulting from apoptotic depletion of HPSCs in both fetal liver and neonatal bone marrow . The Trp53 dosage effect has also been seen in Mdm2+/− and Mdm4+/− double heterozygous embryos and mice ( Terzian et al . , 2007 ) . Besides its engagement in the Mdm2/Mdm4 KO/KI studies , p53 was found to be activated upon impairment of ribosome biogenesis in several RP-deficient mouse models ( Bursac et al . , 2014 ) . For example , liver specific deletion of Rps6 prevented hepatocytes from re-entering the cell cycle after partial hepatectomy due to increased p53 , which can be rescued by Trp53 deletion ( Volarevic et al . , 2000; Fumagalli et al . , 2009 ) . Developmental defects derived from tissue specific deletion of Rps6 in T cells or oocytes was largely due to p53 activation , and rescued by Trp53 deletion ( Sulic et al . , 2005; Panic et al . , 2006 ) . Likewise , p53 activation was involved in defective αβ T cell development in Rpl22 knockout mice and this deficiency was completely rescued by Trp53 deletion ( Anderson et al . , 2007 ) . p53 activation was also involved in congenital malformations in Rpl24-deficient mice which were largely rescued by p53 ablation ( Barkic et al . , 2009 ) . Furthermore , phenotypes such as cerebellar ataxia , pancytopenia and epidermal hyperpigmentation seen in Rpl27a mutant mice was rescued in a haploinsufficient Trp53 background ( Terzian et al . , 2011 ) , whereas the dark skin as a result of p53-dependent epidermal melanocytosis seen in Rps6 , Rps19 , or Rps20 mutant mice was also rescued by Trp53 deletion ( McGowan et al . , 2008 ) . Finally , p53 appears to be causally related to the defects seen in the Treacher Collins syndrome ( TCS ) , a congenital disorder of craniofacial development arising from mutations of TCOF1 , which encodes the nucleolar phosphoprotein Treacle and whose haploinsufficiency perturbs mature ribosome biogenesis to trigger p53 activation ( Jones et al . , 2008 ) . Only one mouse KO model involving the deletion of an Mdm2-binding ribosomal protein Rps14 , along with other seven known genes ( deletion of Cd74-Nid67 interval ) in the 5q-syndrome , was previously reported ( Barlow et al . , 2010 ) . Hematological abnormalities associated with increased apoptosis in bone marrow progenitor cells can be rescued only by homozygous Trp53 deletion ( Barlow et al . , 2010 ) . To our best knowledge , our study is the first targeted inactivation of a single Mdm2-binding ribosomal protein in mouse and demonstrated that ( 1 ) Rps27l is required in vivo to keep p53 in check and ( 2 ) developmental defects and postnatal death upon Rps27l disruption are rescued by the deletion of single Trp53 allele , indicating that moderate increase of p53 is sufficient to tip the life-death balance to the death . It is worth noting that postnatal death upon Rps27l disruption occurs under the wt background of its family member , Rps27 as well as other ribosomal genes encoding all known Mdm2-binding ribosomal proteins . Thus , for hematopoiesis , Rps27l plays a non-redundant role in reducing p53 levels via stabilizing the Mdm2-Mdm4 complex . The viability of Rps27l−/−;Trp53+/− mice provided us an opportunity to study the role of Rps27l in spontaneous tumorigenesis . Surprisingly , while moderate p53 increase is expected to decrease the tumor incidence as shown in Mdm2 hypomorphic mice ( Mendrysa et al . , 2006 ) and in Mdm2+/− or Mdm4+/− mice ( Alt et al . , 2003; Terzian et al . , 2007 ) , our Rps27l−/−;Trp53+/− mice have a much higher incidence than Rps27l+/+;Trp53+/− mice to develop spontaneous T-lymphoblastic and in rare case B-cell lymphomas , resulting in a shortened life-span . In 97% of tumor tissues genotyped , wt Trp53 allele was deleted , indicating that Rps27l depletion imposes a selection pressure against p53 . Furthermore , the genome of these lymphoma cells is highly unstable with a strike aneuploidy rate of ∼63% , which is remarkably higher than those derived from Trp53-null mice with an aneuploidy rate of 30–35% ( Kibe et al . , 2012 ) . Even in diploid population , abnormal centromere–centromere fusions were found . Thus , our study provides an in vivo demonstration that Rps27l is required for the maintenance of genomic stability , which is consistent with a previous report , showing that RPS27L knockdown in HCT116 colon cancer cells may trigger genomic instability ( Li et al . , 2007 ) . One more striking observation we made in this study is that the genomic instability occurs in early passage primary MEFs with the genotype of Rps27l−/−;Trp53+/− , but not of Rps27l−/−;Trp53+/+ . In other word , Rps27l plays a limited role in the maintenance of genomic stability , if p53 is normal . However , when one allele of Trp53 is deleted or mutated ( such as in the case of Li-Fraumeni syndrome or at various stage of tumorigenesis ) , Rps27l becomes critically essential in preventing the loss of p53 heterozygosity . Thus , its loss or decreased expression by any means would exacerbate genomic instability and provide selection pressure against p53 , eventually leading to tumor development . Potential tumor suppressive function of RPS27L is further supported by an observation that high level of RPS27L expression predicted a better prognosis in colon cancer patients ( Huang et al . , 2013 ) . In summary , our results provide strong pieces of in vivo evidence that Rps27l precisely regulates p53 threshold . In normal cells with wt p53 ( Trp53+/+ status ) , Rps27l appears to be a physiological inhibitor of p53 through optimizing Mdm2-Mdm4 E3 ligase activity towards p53 to keep p53 in check . Rps27l disruption triggers ribosomal stress to increase Mdm2 for targeted Mdm4 degradation , leading to p53 activation ( in an Rpl11 dependent manner ) and subsequent apoptosis induction and postnatal death . Under Trp53+/− status , Rps27l acts as a tumor suppressor by maintaining the genomic stability and preventing Trp53 deletion . Rps27l disruption triggers genomic instability and confers selection pressure for p53 inactivation , leading to lymphamagenesis ( Figure 9 ) . Whether the RPS27L level determines the early-onset of human cancers in Li-Fraumeni syndrome patients with a germ-line TP53+/− status is certainly an intriguing question that deserves further investigation . 10 . 7554/eLife . 02236 . 019Figure 9 . Rps27l is a p53 regulator and also a p53 ‘goalkeeper’—a working model . In normal cell with wild-type p53 ( Trp53+/+ ) , Rps27l , upon induction by p53 in response to various stresses , stabilizes the Mdm2-Mdm4 heterodimer to form an optimal E3 ligase complex for effective p53 ubiquitylation and degradation , thus keeping p53 level in check , leading to normal growth and development ( dotted area ) . Rps27l disruption causes an imbalance in ribosomal protein levels and triggers ribosomal stress to stabilize Mdm2 . Increased Mdm2 adapts a conformation that favors the Mdm4 ubiquitylation , leading to a reduced Mdm2-Mdm4 complex and compromised p53 ubiquitylation with an ultimate increase in p53 . Moderately increased p53 is sufficient to induce growth retardation and apoptosis by transactivating p21 and Puma , respectively , leading to the depletion of HSPCs and bone marrow , and eventually postnatal death . In cancer-prone cells with a Trp53+/− status , Rps27l plays an essential role in keeping genome integrity . Rps27l disruption triggers genomic instability , followed by selection for Trp53 deletion . The cells with Trp53 deleted are selected for and outgrow to form spontaneous lymphoma . DOI: http://dx . doi . org/10 . 7554/eLife . 02236 . 019 Germline-transmitted heterozygous Rps27l mice generated from an ES cell clone ( IST11658B7 , C57BL/6 ) were obtained from the Texas A&M Institute for Genomic Medicine ( TIGM , College Station , TX ) . Germline transmission was confirmed by PCR and Southern blotting . Mice bearing the Trp53-null allele ( deletion of exons 2–7 ) ( Jacks et al . , 1994 ) were provided by Dr Yuan Zhu ( University of Michigan , Ann Arbor , MI ) . The genetic background of Rps27l mutant mice is pure C57/BL6 , while Rps27l/Trp53 mice are the hybrids of 129 Svj and C57BL/6 . C57BL/6 Ly5 . 2 ( CD45 . 1 ) mice were purchased from the National Cancer Institute . All procedures were approved by the University of Michigan Committee on Use and Care of Animals . Animal care was provided in accordance with the principles and procedures outlined in the National Research Council Guide for the Care and Use of Laboratory Animals . Genomic DNA was isolated from mouse tail tips according to University of Michigan Transgenic Animal Model Core protocol ( http://www . med . umich . edu/tamc/tDNA . html ) . Mice were genotyped using the primer set of LTR-Rev 2: 5′-ACC TGA AAT GAC CCT GTG CCT TA-3′ and IST11658B7-f: 5′-TTG ATG GCT ACC CAG CCA AAC G-3′ for Rps27l mutant ( 325 bp ) and IST11658B7-f and IST11658B7-r2: 5′-ACG TAT CCT TTA CCT GGC TCC C-3′ for Rps27l wt ( 525 bp ) . The primer sets for Trp53 genotyping are p53x6 . 5: 5′-ACA GCG TGG TGG TAC CTT AT-3′; p53x7: 5′-TAT ACT CAG AGC CGG CCT-3′; and p53 Neo19: 5′-CAT TCA GGA CAT AGC GTT-3′ . Genomic DNA was isolated from mouse tails and digested with EcoRI or PstI . The 5′ end probe ( 644 bp ) was generated by PCR using the primer set , S27LKO-5pb-F: 5′-TAA GCC AGG GGG TCA ATA-3′ and S27LKO-5pb-R: 5′-CTC CCC TGT TCA TTG TGC-3′ . The β-Gal probe ( 686 bp ) was generated by PCR using the primer set , S27LKO-BG-F: 5′-GGC GTA ATA GCG AAG AGG-3′ , and S27LKO-BG-R: 5′-TTC ACC CTG CCA TAA AGA-3′ . All the probes were confirmed by DNA sequencing . Southern blot analysis was carried out as described ( Tan et al . , 2009 ) . MEF cells were generated from day E13 . 5 embryos with indicated genotypes as described ( Tan et al . , 2009 ) , and cultured in DMEM with 15% FBS , 2 mM L-Glutamine , 0 . 1 mM MEM non-essential amino acids at 37°C in a 5% CO2 humidified chamber . Paraffin sections were deparaffinized , rehydrated , and analyzed by immunofluorescence ( Wang et al . , 2009a ) . The tissue sections were incubated with antibody against cleaved caspase-3 ( Cell signaling Technology , Danvers , MA ) in blocking solution overnight . MEF cells were left untreated or treated for 24 hr with ribosomal stress inducer , actinomycin D ( 5 nM ) , followed by immune-staining with antibody against B23 ( Sigma , St . Louis , MO ) . The secondary antibodies were conjugates of Alexa Fluor 488 or Alexa Fluor 594 ( Life Technologies , Grand Island , NY ) . DAPI ( Life Technologies ) was used as nuclear counterstaining . Stained tissues or cells were examined under a fluorescence microscope ( Olympus , Center Valley , PA ) . Cells or tissues were harvested , lysed and subjected to Western blotting or immunoprecipitation ( Macias et al . , 2010 ) , using various antibodies as follows: RPS27L or RPS27 polyclonal rabbit antibody was raised and purified as described ( He and Sun , 2007 ) , p53 ( 1C12 from Cell signaling technology and CM5p from Leica Microsystems , Buffalo Grove , IL ) , Mdm2 4B2 and Mdm4 7A8 ( gifts from Dr Jiandong Chen ) , rabbit polyclonal Mdm4 ( gift from Dr Aart Jochemsen , used for IP ) , Rpl5 , Rpl11 , and Rpl23 ( gifts from Drs Yanping Zhang and Hua Lu ) , Rps7 ( gift from Dr Ruiwen Zhang ) , Rps27a ( gift from Dr Mushui Dai ) , Rps14 and Rps19 ( Santa Cruz Biotechnology , Santa Cruz , CA ) , rabbit polyclonal Ab for mouse Mdm2 ( raised against mouse Mdm2 peptide EQTPLSQESDDYSQPSTSSS , made by Yenzym , San Francisco , CA , used for IP ) , human MDM2 ( Ab-1 , Calbiochem , San Diego , CA ) , p21 ( BD Biosciences , San Jose , CA ) , and β-actin ( Sigma ) . For surface staining , cells were incubated with the indicated antibodies in staining buffer ( HBSS with 2% FBS ) for 20 min at 4°C . The following antibodies were purchased from eBioscience , San Diego , CA: Fluorescein isothiocyanate ( FITC ) -conjugated anti-CD3e ( clone 145-2C11 ) , anti-CD4 ( GK1 . 5 ) , anti-CD8a ( 53–6 . 7 ) , anti-CD16/CD32 ( 93 ) , anti-CD43 ( eBioR2-60 ) , anti-CD45 . 1 ( A20 ) , and anti-CD45 . 2 ( 104 ) ; phycoerythrin ( PE ) -conjugated anti-CD3e ( M1/69 ) , anti-B220 ( RA3-6B2 ) , anti-Mac-1 ( M1/70 ) , anti-Gr-1 ( RB6-8C5 ) , anti-Ter119 ( TER-119 ) , and anti-CD45 . 2 ( 104 ) ; PE-Cy7–conjugated anti-CD4 ( RM4-5 ) , and anti-B220 ( RA3-6B2 ) ; allophycocyanin ( APC ) -conjugated anti-CD8a ( 53–6 . 7 ) , anti-CD45 . 1 ( A20 ) , anti-CD48 ( HM48-1 ) , anti-CD71 ( R17217 ) , and anti-Gr-1 ( RB6-8C5 ) ; APC-eFluor780–conjugated anti-CD117 ( c-Kit , clone 2B8 ) ; and eFluor660-conjugated anti-CD34 ( RAM34 ) . The following antibodies were purchased from BD Biosciences: PerCP-Cy5 . 5–conjugated anti-B220 ( RA3-6B2 ) , and anti-Mac-1 ( M1/70 ) ; PE-Cy7–conjugated anti-CD45 . 2 ( 104 ) , and anti-Ly-6A/E ( Sca-1 , clone D7 ) . Lineage markers included B220 ( B cells ) , CD3 ( T cells ) , Gr-1 ( granulocyte ) , Mac-1 ( myeloid cells ) , and Ter119 ( erythrocytes ) . PerCP-Cy5 . 5-conjugated anti-CD150 Ab ( clone TC15-12F12 . 2 ) was purchased from BioLegend , San Diego , CA , and FITC-conjugated Annexin V and FITC-conjugated anti-BrdU antibody were from BD Biosciences . All FACS analyses were performed on an LSR II flow cytometer ( BD Biosciences ) , and data were analyzed with FlowJo software ( Tang et al . , 2012 ) . 6–8-week-old C57BL/6 Ly5 . 2 ( CD45 . 1 ) recipient mice were lethally irradiated with a 137Cs source delivering 170 rad per min for a total dose of 1100 rads . Given numbers of fetal liver cells from E14 . 5 embryos ( CD45 . 2 ) were either injected alone or mixed with recipient-type ( CD45 . 1 ) competitive BM cells . The cells were injected into recipients through the tail vein within 24 hr after irradiation . Reconstitutions were measured by flow cytometry of peripheral blood at the time points indicated ( Chen et al . , 2009; Tang et al . , 2012 ) . Immunohistochemical staining was performed on the DAKO Autostainer ( DAKO , Carpinteria , CA ) using diaminobenzadine ( DAB ) as the chromogen . After dewaxing and rehydration , serial sections were labeled with p53 ( CM5p , Leica Microsystems ) , after 10 mM citrate buffer , pH6 microwave epitope retrieval . LSAB+ ( DAKO ) was employed as the detection system . Appropriate negative ( no primary Ab ) and positive controls were stained in parallel . MEF cells treated with MG132 for 4 hr were lysed and IP with Mdm2 Ab . The Mdm2/Mdm4-p53 complex were then incubated with 10 μg ubiquitin , 375 ng UBE1 , 150 ng Ubc5Hb ( Boston Biochem , Cambridge , MA ) , and 30 μl reaction buffer ( 50 mM Tris pH7 . 5 , 2 . 5 mM MgCl2 , 15 mM KCl , 1 mM DTT , 0 . 01% Triton-X-100 , 1% glycerol ) in the presence of 4 mM ATP . The mixture was incubated at 37°C for 60 min with continuous vortexing , and subjected to IB after boiling in SDS sample buffer ( Cheng et al . , 2009 ) . The profiling was conducted as described ( Zhu et al . , 2012 ) with modifications . Briefly , MEFs were treated with cycloheximide ( 100 μg/ml ) in growth medium for 30 min at 37°C . Cells were washed with PBS containing cycloheximide and then lysed in extraction buffer containing 20 mM HEPES ( pH 7 . 5 ) , 5 mM MgCl2 , 150 mM KCl , 100 μg/ml cycloheximide , 1 mM DTT , 0 . 5% Triton-X 100 , 0 . 5% Sodium deoxycholate , and RNase inhibitor . Extracts were spun for 10 s to pellet nuclei and then cleared by centrifugation at 10 , 000×g for 10 min at 4°C . The cytoplasmic extracts were loaded onto 4 . 5 ml sucrose gradients ( 10%–50% ) buffered in 20 mM HEPES ( pH 7 . 5 ) , 100 mM KCl , 5 mM MgCl2 , 1 mM DTT . Gradients were subjected to ultra-centrifugation using a Beckman SW50 . 1 Rotor at 40 , 000 rpm for 100 min at 4°C . Gradients were then fractionated measured by optical density at 254 nm . MEF cells on 60 mm dish were starved of methionine in methionine-free medium for 30 min and then pulse-labeled for 30 min in 1 ml medium containing 50 μCi of L-[methyl-3H]-methionine ( MP Biochemicals , Santa Ana , CA ) . After rinsing with 10 × methonine ( 0 . 3 mg/ml ) medium , cells were incubated in 10 × methonine medium for indicated time periods . Total RNA was isolated using RNeasy kit ( Qiagen , Valencia , CA ) and subjected to 1% Agarose formaldehyde gel electrophoresis and then transferred on Nylon membrane . The membrane was dried and sprayed with EN3HANCE ( PerkinElmer , Waltham , MA ) and then exposed at −80°C for a week ( Itahana et al . , 2003 ) . Human lung H1299 cells were transiently transfected with various plasmids . Cells were harvested 24 hr post transfection after last 4 hr MG132 treatment and split into two aliquots with one for direct Western blotting analysis and the other for in vivo ubiquitylation assay as described ( Gu et al . , 2007 ) . Briefly , cell pellets were lysed and incubated with Ni-NTA beads ( Qiagen ) at room temperature for 4 hr . Beads were then washed and incubated with elution buffer at room temperature for 20 min . The eluted proteins were analyzed by Western blotting . The sequences of scrambled control siRNA and RPS27L siRNA have been described ( Xiong et al . , 2011 ) . Short hairpins targeting mouse Rpl11 ( targeting sequence: 5′-CGG GAG TAT GAG TTG CGG AAA-3′ ) or Rpl5 ( targeting sequence: 5′-CCC TCA TAG TAC CAA ACG ATT-3′ ) was cloned into pLKO . 1-puro vector . Two additional Rpl5-silencing clones ( TRCN0000104427 and TRCN0000104429 ) were purchased from Thermo Fisher Scientific , Waltham , MA . Lentiviral particles were produced by University of Michigan Vector Core . Primary thymic lymphoma cells from Rps27l−/−;Trp53+/− mice or primary MEFs at early passage were treated with 0 . 2 μg/ml KaryoMAX colcemid solution ( Life Technologies ) for 2 hr before harvesting for metaphase preparation . Metaphase spreads were stained by incubation in 4% KaryoMAX Giemsa solution ( Life Technologies ) for 15 min , followed by observation under a light microscope . Chromosome numbers were counted ( Wu et al . , 2011 ) . Total RNA was isolated from MEF cells with Trizol reagent ( Life Technologies ) . Complementary DNA was made from RNA with Superscript III ( Life Technologies ) . Real-time PCR was performed on a 7500 Real Time PCR system ( Life Technologies ) . The cycling program was set as follows: 50°C 2 min , 95°C 10 min for the PCR initial activation and 45 cycles of denaturation at 95°C for 15 s , annealing and extension at 60°C for 1 min . The sequences of p53 , Mdm2 , p21 , Mdm4 , 45S rRNA , and GAPDH are as follows: p53-RT-F: 5′-GAG AGT ATT TCA CCC TCA AGA TCC G-3′ , p53-RT-R: 5′-CCC CAC TTT CTT GAC CAT TGT TT-3′; Mdm2-RT-F: 5′-GAT GAG GAT GAT GAG GTC TAT CGG-3′ , Mdm2-RT-R: 5′-TCT GGA AGC CAG TTC TCA CGA A-3′; p21-RT-F: 5′-ACT TCC TCT GCC CTG CTG CA-3′ , p21-RT-R: 5′-CGC TTG GAG TGA TAG AAA TCT GTC A-3′; Mdm4-RT-F: 5′-TTT ACA GAC AAA TCA GGA TAT AGG TA-3′ , Mdm4-RT-R: 5′-GTA CAC TGC CAC TCA TCC TCA-3′; 45S-F: 5′-ACA CGC TGT CCT TTC CCT ATT AAC ACT AAA-3′ , 45S-R: 5′-AGT AAA AAG AAT AGG CTG GAC AAG CAA AAC-3′; GAPDH-F: 5′- GCC GCC TGG AGA AAC CTG CC-3′ , GAPDH-R: 5′-GGT GGA AGA GTG GGA GTT GC-3′ ( Poortinga et al . , 2004 ) . For quantitative measurement of Trp53 dosage , genomic DNA were isolated from MEFs and amplified with the following primer sets: p53-del-F2: 5′-CCT GAT CGT TAC TCG GCT TGT C-3′ , p53-del-R2: 5′-CAA CTG CAC AGG GCA CGT CT-3′; p53-del-F4: 5′-GGC TTC TGA CTT ATT CTT GCT CTT A-3′ , p53-del-R4: 5′-AGA CCT CGG GTG GCT CAT AA-3′; p53-del-F5: 5′-GAG GTA GGG AGC GAC TTC ACC-3′ , p53-del-R5: 5′-GGT AAG GAT AGG TCG GCG GTT-3′; GAPDH-F: 5′-GTA TGA CTC CAC TCA CGG CAA A-3′ , GAPDH-R: 5′-GGT CTC GCT CCT GGA AGA TG-3′ . For determination of p53 protein synthesis , MEF cells were treated for 1 hr in methionine-free and cysteine-free DMEM , containing 10% dialyzed FCS and 50 μM MG132 . Cells were then labeled with 250 μCi/ml of [35S]-methionine ( MP Biochemicals ) , followed by immunoprecipitation with anti-p53 antibody ( 1C12 ) . Immunoprecipitates , along with whole cell extract , were then subjected to SDS-PAGE and autoradiography ( Wang et al . , 2009b ) . For determination of global nascent protein synthesis , MEF cells on 60 mm dish were starved in methionine-free and cysteine-free DMEM with dialyzed FCS for 30 min and then pulse-labeled with 1 ml [35S]-methionine ( 200 μCi/ml ) for 1 hr . After PBS washing , cells were lysed in 0 . 5% NP-40 buffer . Equal amounts of protein extracts were precipitated with 20% ice-cold TCA for 30 min on ice , and dissolved in 0 . 1 N NaOH . Total [35S]-methionine incorporation was measured with a liquid scintillation counter and plotted as counts per min/μg protein . Equal amounts of protein were also subjected to SDS-PAGE and autoradiography ( Lindstrom and Zhang , 2008 ) . To determine the deletion of Trp53 wt allele , genomice DNA were isolated and amplified with the following primer sets: Exon-1-F: 5′-ATC GGT TTC CAC CCA TTT TG-3′ , Exon-1-R: 5′-ATA CAC TCC CGT TCA TCC CG-3′; Exon-2-F: 5′-TAC CTC TGC TCA GCC CCC G-3′ , Exon-2-R: 5′-TTA CAG ACA CCC AAC ACC ATA CCA-3′; Exon-3-F: 5′-GCA GGG TCT CAG AAG TTT GAG G-3′ , Exon-3-R: 5′-GTG GAT GGG ACA AAG AAG AAC C-3′; Exon-4-F: 5′-TTG GGC TTT GGT GTT GGG-3′ , Exon-4-R: 5′-AGG CTG AAG AGG AAC CCC C-3′; Exon-5-F: 5′-CGG GGA GTT GTC TTT CGT GT-3′ , Exon-5-R: 5′-TAA GAG CAA GAA TAA GTC AGA AGC C-3′; Exon-6-F: GTA AGC CCT CAA CAC CGC C-3′ , Exon-6-R: 5′-GAC TCA GCG TCT CTA TTT CCC G-3′; Exon-7-F: 5′-TCC AGC AGG TGT GCC GAA-3′ , Exon-7-R: 5′-AAC CCC GAG AAG CCA CAG A-3′; Exon-8-F: 5′-TCT GTG GCT TCT CGG GGT TC-3′ , Exon-8-R: 5′-GGA AGG AGA GAG CAA GAG GTG AC-3′; Exon-9-F: 5′-CGG AGG AGC CTG TTG AGC TT-3′ , Exon-9-R: 5′-CAG CCT CAG AGC ATG AGC TC-3′; Exon-10-11-F: GAG CCA GCT TAA GTT GGG AAC-3′ , Exon-10-11-R: 5′-GAC AGC AAG GAG AGG GGG-3′ . The two-tailed Student's t test was used for the comparison of parameters between groups . Survival analysis was performed by Kaplan–Meier analysis . Statistical significance was determined as p < 0 . 05 .	There are over a hundred different types of cancer that can affect humans; but , in general , all cancers are caused by mutations that cause cells to grow and divide abnormally . ‘Tumor suppressor genes’ are genes that normally protect a cell from genetic changes that can lead a cell towards becoming cancerous . About half of all cancers in humans have a mutation in one of the two copies of a tumor suppressor gene that encodes a protein called p53 , which helps to control how and when cells grow and divide . In normal cells , the p53 protein can be activated in various ways . Damage to a cell's DNA triggers p53 to stop the cell growing , which gives the cell time to repair the DNA damage . However , if the damage is too severe and cannot be repaired , p53 essentially causes the cell to kill itself , via a process called apoptosis . Furthermore , if a cell has problems building new copies of its protein-making machinery , some of the parts ( called ribosomal proteins ) that make up these molecular machines can also lead to p53 being activated . By deleting the gene for a protein called Rps27l that is a newly characterized ribosomal protein , Xiong et al . have discovered that , in mice , Rps27I regulates the p53 protein in two different ways . In normal cells , Rps27l appears to inhibit p53 , which is likely to encourage cancer to develop . But , if a cell has already lost a copy of the p53 gene—a situation that would normally encourage the cells to accrue further mutations and become cancerous—Rps27l acts as a tumor suppressor . In these mutated cells , the Rps27l protein helps to maintain the stability of the genome and prevent the loss of the second copy of gene for p53 , and so protects the cell from becoming cancerous . Thus Rps27l can either activate or inactivate p53 activity depending on how many copies of the gene for p53 remain intact . The next challenge is to investigate if Rps27l levels determine the early-onset of tumor development in cancer-prone cells seen in patients with Li-Fraumeni syndrome , who are born with a mutated copy of the p53 gene .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2014	Ribosomal protein S27-like is a physiological regulator of p53 that suppresses genomic instability and tumorigenesis
The neural control of social behaviors in rodents requires the encoding of pheromonal cues by the vomeronasal system . Here we show that the typical preference of male mice for females is eliminated in mutants lacking oxytocin , a neuropeptide modulating social behaviors in many species . Ablation of the oxytocin receptor in aromatase-expressing neurons of the medial amygdala ( MeA ) fully recapitulates the elimination of female preference in males . Further , single-unit recording in the MeA uncovered significant changes in the sensory representation of conspecific cues in the absence of oxytocin signaling . Finally , acute manipulation of oxytocin signaling in adults is sufficient to alter social interaction preferences in males as well as responses of MeA neurons to chemosensory cues . These results uncover the critical role of oxytocin signaling in a molecularly defined neuronal population in order to modulate the behavioral and physiological responses of male mice to females on a moment-to-moment basis . A fundamental goal of neuroscience is to understand how brain circuits control behavior . New advances in genetic , imaging and functional approaches applied to small and large brains are uncovering wiring diagrams and neural activity ensembles underlying specific behaviors at increasingly high resolution . Proper regulation of the functional properties of these neural circuits requires cohorts of still poorly understood neuromodulators , such as biogenic amines and neuropeptides ( Bargmann and Marder , 2013; Marder , 2012 ) . Released from the axons , dendrites , and cell bodies of specific neuronal subpopulations , neuropeptides signal on time scales ranging from seconds to minutes and hours , and act on local microcircuits as well as on diffuse targets distributed throughout the brain . Thus , neuromodulation adds temporal and spatial dynamics to the function of neural circuits according to the animal’s external environment and internal physiological state ( Brezina , 2010; Hökfelt et al . , 2000 ) . The vertebrate hypothalamus has emerged as a particularly rich source of neuropeptides . These neuromodulators shape intricate allostatic and behavioral functions such as temperature and energy balance , thirst , hunger , sleep , as well as aggression , reproduction , and parenting . In particular , the nonapetides oxytocin ( OXT ) and vasopressin ( AVP ) represent an important set of neuromodulators that are produced by discrete populations of hypothalamic neurons in response to social signals and according to an animal's physiological state ( Donaldson and Young , 2008; Knobloch and Grinevich , 2014; Landgraf and Neumann , 2004; Veenema and Neumann , 2008 ) . These nonapeptides are evolutionarily conserved across vertebrate and invertebrate species , and it has been proposed that sex- and species-specific differences OXT and AVP systems may underlie genetic variations in social behavior control ( Bendesky et al . , 2017; Caldwell , 2017; Johnson and Young , 2017; Knobloch and Grinevich , 2014; Lockard et al . , 2017; Vaidyanathan and Hammock , 2017 ) . Genetic models of mice deficient in OXT or in OXT receptor ( OXTR ) have helped uncover the requirement of OXT signaling in myriad social behaviors including recognition of familiar and novel conspecifics ( Ferguson et al . , 2001 , Ferguson et al . , 2000Ferguson et al . , 2000; Takayanagi et al . , 2005; Wersinger et al . , 2008 ) , parental care ( Takayanagi et al . , 2005 ) , social interactions of female with male mice during the estrous cycle ( Nakajima et al . , 2014 ) , aggression ( Harmon et al . , 2002 ) , and anxiety-related behaviors in male mice ( Li et al . , 2016 ) . OXT can act directly via OXTR signaling on postsynaptic neurons to alter the activity of key components of neural circuits regulating social behaviors as seen in circuits controlling female social approach to males at specific phases of the estrus cycle ( Nakajima et al . , 2014 ) , anxiety-related behaviors in male mice ( Li et al . , 2016 ) , and maternal responses to pup calls ( Marlin et al . , 2015 ) . OXT has also been shown to modulate social behaviors indirectly by promoting the processing of social odor cues in the main olfactory bulb via top-down projections from the anterior olfactory nucleus ( Oettl et al . , 2016 ) , by acting on other neuromodulatory systems , for example regulating the activities of dopamine neurons ( Xiao et al . , 2017 ) , facilitating norepinephrine release in the olfactory bulb to promote social recognition ( Dluzen et al . , 2000; Dluzen et al . , 1998 ) or reinforcing social interaction with conspecifics by regulating presynaptic serotonin release in the nucleus accumbens ( Dölen et al . , 2013 ) . In mice , genetic and surgical ablation of the vomeronasal organ ( VNO ) , a specialized organ required for the detection of pheromones , results in atypical behaviors towards conspecifics , with female and male mutants displaying sexual behaviors indiscriminately towards conspecifics of both sexes , suggesting that the vomeronasal system plays a critical role in decoding sex-specific chemosensory cues ( Kimchi et al . , 2007; Liman et al . , 1999; Stowers et al . , 2002 ) . The medial amygdala ( MeA ) , located two-synapses downstream of VNO chemosensory neurons , and projecting to hypothalamic areas controlling innate behavioral responses , is an important brain region implicated in the processing of social cues ( Kevetter and Winans , 1981a; Kevetter and Winans , 1981b; Petrovich et al . , 2001 ) . The MeA displays specific , topographically organized , and sexually dimorphic responses to chemosensory cues ( Bergan et al . , 2014; Choi et al . , 2005; Lehman et al . , 1980 ) . Functional manipulation of the MeA has been shown to affect various social behaviors , such as aggression , mating , social-grooming , and social recognition ( Choleris et al . , 2007; Ferguson et al . , 2001 ) . OXT signaling in the MeA is critical for mice to distinguish familiar from novel conspecifics; however , the involvement of OXT in modulating other vomeronasal functions , as well as the mechanisms by which OXT sculpts the discrimination of socially relevant cues are not yet fully understood ( Choleris et al . , 2007; Ferguson et al . , 2001; Gur et al . , 2014 ) . Here , we investigated OXT action on neural circuits regulating sex-specific behavioral responses to conspecific signals . We found that , during social interactions , OXT mutant males do not display the preference for female conspecifics that is typical of wild-type males , and are impaired in chemosensory discrimination of sex-specific cues . To dissect the underlying cellular components of OXT function in behavioral sex-discrimination , we conducted genetic ablation as well as virus-mediated manipulation of OXTR , and demonstrate the critical role of OXT signaling in a sub-population of MeA neurons expressing the steroid converting enzyme aromatase . We further show that impaired chemosensory discrimination in OXT mutants correlates with altered chemosensory response profiles in the MeA . Moreover , acute modulation of OXT signaling in adults is sufficient to alter sensory representation in the MeA and sex discrimination in social interactions . Our study thus reveals that acute neuromodulatory function of OXT in a single , molecularly defined , population of neurons shapes the sensory representation of the MeA and is critical for tuning an animal's preference for male versus female social cues . Social behavior responses are fine-tuned according to the species , gender , and endocrine status of individual animals . To investigate a potential role of OXT in the modulation of sex-specific social preference behaviors , we compared the preference of C57BL/6J OXT knockout male mice ( Oxt-/- ) and wild-type littermate controls ( Oxt+/+ ) in investigating female versus male conspecifics ( Figure 1A ) . Subject male mice were group-housed with littermates of the same sex after weaning , individually housed for one week before behavioral tests , and sexually naïve at the onset of testing . After an initial habituation period of 10 min in a 3-chamber paradigm with an empty wire cup in each side chamber ( Yang et al . , 2011 ) , Oxt+/+ and Oxt-/- mice were allowed to explore the arena with a novel male confined in the wire cup in one side chamber and a novel female confined in the wire cup in the other side chamber . The female and male sides were randomized between experiments . Both stimulus mice were confined in wire cups that permit the exchange of visual , chemosensory , and acoustic signals between subject and stimulus mice . The amount of time subjects spent adjacent to each enclosure cage was quantified , and a 'social preference score' ( see Materials and methods ) was calculated that ranges from −1 . 0 ( all time spent next to the male cage ) to 1 . 0 ( all time spent next to the female cage ) . Data show that control Oxt+/+ male mice spent significantly more time investigating females than males ( female interaction zone = 192 . 8 ± 12 . 2 s; male interaction zone = 128 . 0 ± 10 . 8 s; p<0 . 001 , t test; n = 10 ) resulting in a positive social preference score that is significantly higher than one obtained with empty wire cups ( 0 . 21 ± 0 . 03 vs . −0 . 001 ± 0 . 05; p<0 . 001 , t test; n = 10; Figure 1B ) . By contrast , Oxt-/- male mice displayed little preference for investigating female over male conspecifics ( male interaction zone = 105 . 1 ± 13 . 2 s , female interaction zone = 121 . 8 ± 13 . 8 s; p=0 . 39 , t test; n = 10 ) and had a social preference score that was not significantly different from one obtained with empty wire cups ( 0 . 08 ± 0 . 05 vs . 0 . 03 ± 0 . 05; p=0 . 52 , t test; n = 10; Figure 1B ) . When tested with empty enclosure cages , both Oxt+/+ and Oxt-/- males spent equal time in the two interaction zones ( Oxt+/+: empty male interaction zone = 141 . 7 ± 13 . 9 s , empty female interaction zone = 137 . 8 ± 7 . 7 s; p=0 . 81 , t test; Oxt-/-: empty male interaction zone = 108 . 8 ± 7 . 7 s , empty female interaction zone = 115 . 9 ± 7 . 5 s; p=0 . 52 , t test ) . Thus , our results suggest that loss of OXT alters the normal preference of male mice to interact with female rather than male conspecifics . Next , we measured the preference of OXT mutant mice for investigating female versus male bedding , a rich source of conspecific chemostimuli . In this preference paradigm ( Figure 1C ) , individually caged naïve Oxt+/+ and Oxt-/- male littermates were presented in their home cages with male and female bedding separated by a divider . Subject mice were allowed to investigate the two types of bedding for a total of 5 min . As expected , Oxt+/+ males spent a significantly longer time on average investigating female than male bedding ( male bedding = 34 . 3 ± 5 . 1 s , female bedding = 76 . 3 ± 11 . 2 s; p<0 . 001 , t test; n = 9 ) , whereas Oxt-/- male mice presented lost preference in investigating female chemostimuli ( male bedding = 43 . 5 ± 5 . 7 s , female bedding = 50 . 8 ± 7 . 4 s; p=0 . 46 , t test; n = 13; Figure 1D ) . Accordingly , the mean odor preference score ( see Materials and methods ) of Oxt+/+ male mice ( 0 . 38 ± 0 . 05 ) was significantly higher than that of Oxt-/- male mice ( 0 . 03 ± 0 . 08; p<0 . 01 , t test; Figure 1D ) . However , the overall time spent investigating either bedding was similar between Oxt+/+ and Oxt-/- male mice ( 110 . 6 ± 15 . 8 s vs . 94 . 24 ± 10 . 3 s; p=0 . 37 , t test; Figure 1D ) , suggesting that the loss of male preference for female cues is not simply due to general loss of interest in socially relevant chemostimuli . Our results thus demonstrate that OXT signaling is required for male mice to display normal preference for female mouse chemosensory cues . Next , we investigated gonadally intact females for the effect of OXT signaling on preference for conspecific cues . Socially naïve C57BL/6J Oxt+/+ and Oxt-/- female mice were group-housed since weaning with littermates of the same sex and genotype , and individually caged for one week before being tested for their preference in investigating female versus male bedding . In contrast to the observations made with male mice , loss of OXT did not cause significant changes in odor preference in females . Both Oxt+/+ and Oxt-/- female littermates spent slightly , but statistically significant , longer durations investigating female bedding than male bedding ( Oxt+/+: male bedding = 32 . 6 ± 1 . 8 s , female bedding = 47 . 4 ± 5 . 7 s , n = 9; Oxt-/-: male bedding = 26 . 9 ± 2 . 8 s , female bedding = 42 . 7 ± 6 . 4 s , n = 11; p<0 . 05 , t test; Figure 1D ) , and there was no significant difference between the two groups in odor preference score ( Oxt+/+:0 . 16 ± 0 . 06 , n = 9; Oxt-/-:0 . 19 ± 0 . 05 , n = 11; p=0 . 75 , t test; Figure 1D ) . Importantly we investigated sex preference at different stages of estrous cycle in wild-type females ( Figure 1—figure supplement 1 ) in the 3-chamber social interaction assay that allowed visual , auditory and chemosensory interactions ( Figure 1E ) . No significant preference was observed in either estrous or di-estrous C57BL/6J female mice ( t test for comparisons of time spent in interaction zones and beddings , and for comparison of preference scores and total sniffing time ) . Based on these data , we focused our subsequent investigation on the robust behavioral preference observed by males for female cues . To explore a more general role of OXT in the recognition of animal cues , including heterospecific signals , we utilized a Y-maze paradigm to test the responses of Oxt+/+ and Oxt -/- mice to predator cues ( Figure 1F ) . After initial habituation in the Y-maze , subject mice were allowed to explore the side chambers ( goal zones ) of the Y-maze containing only a food pellet on one side and both fox urine and a food pellet on the other side . We measured the time mice spent in each goal zone and found that Oxt+/+ ( fox +food = 35 . 8±2 . 4 s , food = 73 . 4 ± 5 . 5 s; p<0 . 001 , t test; n = 10 ) as well as Oxt -/- ( fox +food = 28 . 9±7 . 3 s , food = 63 . 9 ± 9 . 0 s; p<0 . 001 , t test; n = 10 ) mice spent significantly less time in the goal zone containing fox urine in additional to a food pellet ( Figure 1G ) . Therefore , mice with impaired OXT signaling readily recognize and adapt behaviors in response to predator cues , supporting a specific role for OXT in modulating social interaction of male mice with conspecifics . The loss of behavioral preference for female cues observed in Oxt-/- males could be due to circuit changes at any stage along the sensorimotor transformation . Consistent with previous findings ( Ferguson et al . , 2000 ) , our results show no significant difference in the combined time that Oxt-/- and Oxt+/+ mice spent investigating male and female beddings , suggesting that mice with impaired OXT signaling have similar ability and motivation to investigate conspecific chemical cues . To exclude the possibility that diminished cue preference results from changes in signal detection , we carried out an olfactory habituation and dishabituation assay ( Crawley et al . , 2007; Yang and Crawley , 2009 ) ( Figure 1H ) . Individually caged male mice received three successive 2 min presentations of mouse urine samples from one sex at an interval of 1 min , followed with samples from the other sex in three subsequent presentations . Data show that both wild-type and OXT mutant males habituate to repeated presentation of the urine from the same sex , as indicated by reduced duration spent investigating the odor source , and that they subsequently dishabituate to urine from the other sex ( Fisher’s LSD test , *p<0 . 001 , p<0 . 01 , p<0 . 05; Figure 1I ) . Thus , mice lacking OXT still have the ability to detect and remember conspecific cues , suggesting that the lack of behavioral bias towards females in OXT-/- male mice is likely due to abnormal processing of conspecific cues rather than a mere defect in signal detection . OXT and AVP peptides are highly similar and bind to the OXT receptor ( OXTR ) and the AVP receptors V1aR , V1bR and V2R ( Albers , 2015; Gimpl and Fahrenholz , 2001 ) . To investigate the specificity of OXT function in the discrimination of sex-specific cues , we characterized the sex-specific cue preference of male mice with impaired AVP signaling . Mutant mice lacking AVP die within 1 week after birth , thus preventing further analysis of adult social behaviors ( Hayashi et al . , 2009; Russell et al . , 2003 ) . Instead , we next attempted to investigate the potential role of AVP in sex-specific cue discrimination by ablating AVP neurons in mice generated by the cross of Avp-Cre mice with the ROSA26-eGFP-DTA transgenic mouse line ( Figure 2—figure supplement 1A–B ) , which expresses the cellular toxin diphtheria toxin subunit A ( DTA ) after Cre-mediated recombination . As expected , the DTA-mediated ablation led to significant reduction in the number of AVP neurons with little effect on OXT-expressing neurons ( S Figure 2C ) . Unfortunately , DTA-mediated ablation of AVP neurons induced excessive water intake and urination and grossly reduced body weight ( Figure 2—figure supplement 1C ) , preventing the characterization of behavioral impairments resulting from lack of AVP . In turn , we tested the behavior of mice lacking one or both of the two central AVP receptors , V1aR and V1bR . Three mutant mouse lines , V1aR knockout , V1bR knockout ( Wersinger et al . , 2002 ) and a double knockout line lacking both V1aR and V1bR , were tested for male preference toward female or male beddings . Data show that males from all three knockout lines displayed normal preference for female chemo-stimuli compared to controls ( V1aR: ncontrol = 10 , nmutant = 6; V1bR: ncontrol = 10 , nmutant = 11; V1aR V1bR: ncontrol = 11 , nmutant = 11 ) , such that both control and mutants spent more time investigating female beddings than male beddings ( V1aR: Pcontrol <0 . 05 , Pmutant <0 . 05; V1bR: Pcontrol <0 . 001 , Pmutant <0 . 001; V1aR V1bR: Pcontrol <0 . 001 , Pmutant <0 . 001; t test ) and had positive mean odor preference scores ( Figure 2A , and Figure 21—figure supplement 1D–I ) . Thus , our results suggest that , unlike OXT , AVP is dispensable for sex-specific chemosensory preference and that the effect of OXT in sex-cue preference is mediated exclusively by OXTR . The MeA , previously shown to express moderate level of OXTR ( Yoshida et al . , 2009 ) , is a key center for decoding socially relevant chemo-stimuli in the VNO pathway . In particular , the posterior MeA is an attractive candidate area to mediate the behavioral effects we observed in mice lacking OXT because it receives direct inputs from the accessory olfactory bulb ( AOB ) , and displays topographically organized responses to social interactions ( Bergan et al . , 2014; Choi et al . , 2005; Lehman et al . , 1980 ) . To investigate the potential role of OXT signaling in modulating the chemosensory discrimination of social cues in the posterior MeA , we genetically ablated OXTR expression from two distinct subpopulations of neurons in the posterior MeA defined by the expression of thyrotropin-releasing hormone ( TRH ) and aromatase , respectively , using a conditional OXTR knockout mouse line ( Oxtrflox/flox ) ( Lee et al . , 2008 ) . TRH is predominantly expressed in the posteroventral subdivision ( MeApv ) ( Allen Mouse Brain Atlas , http://mouse . brain-map . org/gene/show/21801 ) ; while aromatase , which converts circulating testosterone to estrogen , is mainly found in neurons of the posterodorsal subdivision of the MeA ( MeApd ) ( Wu et al . , 2009 ) . To perform this experiment , we used a previously described Trh-IRES-Cre knock-in line ( Krashes et al . , 2014 ) , in which Cre expression in the MeA is limited to the posteroventral subnucleus ( Figure 2—figure supplement 2A . We also generated a Cyp19a1 ( aromatase ) -Cre BAC transgenic line that faithfully recapitulates the endogenous aromatase expression in the MeApd , as well as in the other described sites of endogenous aromatase expression across the brain , such as the posteromedial BNST , preoptic hypothalamus , and lateral septum ( Balthazart et al . , 1991; Lauber and Lichtensteiger , 1994; Wu et al . , 2009 ) ( Figure 2—figure supplement 2B , C ) . We then used the 3-chamber paradigm to examine the social interaction preference of mice lacking OXTR in subsets of neurons that express either aromatase ( Cyp19a1-Cre; Oxtrflox/flox ) or TRH ( Trh-Cre; Oxtrflox/flox ) . In all cases , male mice with intact OXT signaling were used as controls for littermates lacking the expression of OXTR in specific neuron types . Data show that male mice lacking OXTR expression in aromatase-positive neurons failed to show a significant preference for social interactions with female conspecifics , in a manner similar to what we observed previously with Oxt-/- mice ( male interaction zone = 101 . 5 ± 10 . 0 s , female interaction zone = 134 . 7 ± 9 . 1 s; not significant , t test; social preference score = 0 . 15 ± 0 . 05 , preference score with empty wire cups = 0 . 10 ± 0 . 06 , p=0 . 60 , t test; n = 16; Figure 2C ) . These results suggest that OXT signaling in aromatase-expressing neurons is necessary for normal discrimination of male and female conspecific cues . In contrast , male mice with intact OXT signaling ( male interaction zone = 98 . 2 ± 12 . 0 s , female interaction zone = 211 . 4 ± 16 . 9 s; p<0 . 001 , t test; social preference score = 0 . 36 ± 0 . 07 , preference score with empty wire cups = 0 . 12 ± 0 . 03; p<0 . 01 , t test; n = 10 ) and male mice with impaired OXT signaling restricted to TRH neurons ( male interaction zone = 76 . 5 ± 6 . 9 s , female interaction zone = 167 . 4 ± 20 . 2 s; p<0 . 001 , t test; social preference score = 0 . 34 ± 0 . 07 , preference score with empty wire cups = 0 . 02 ± 0 . 06; p<0 . 01 , t test; n = 8 ) all spent significantly longer durations in the female interaction zone and had a strong preference for female conspecifics ( Figure 2C and Figure 2D ) . Next , we tested the various mutant lines and littermate controls in bedding investigation assays . Normal preference in investigating female odor was observed in Oxtrflox/flox ( male bedding = 39 . 3 ± 5 . 6 s , female bedding = 96 . 1 ± 9 . 2 s; p<0 . 001 , multiple t-tests; preference score = 0 . 41 ± 0 . 08; n = 10 ) and Cyp19a1-Cre ( male bedding = 40 . 7 ± 4 . 9 s , female bedding = 88 . 1 ± 12 . 5 s; p<0 . 001 , t test; preference score = 0 . 36 ± 0 . 03 , n = 9 ) male mice , both of which have intact OXT signaling . In contrast , the homozygous offspring of these two lines , which have impaired OXT signaling in aromatase-positive neurons , showed loss of preference to investigate female bedding ( n = 13; male bedding = 59 . 0 ± 8 . 7 s , female bedding = 58 . 4 ± 5 . 9 s; p=0 . 95 , t test; preference score = 0 . 02 ± 0 . 08 , significantly lower than that of Oxtrflox/flox and Cyp19a1-Cre , p<0 . 05 , t test; Figure 2G ) . We also utilized the Y-maze to measure the willingness of mice lacking OXTR expression in specific neuron types to explore attractive cues versus predator cues . Consistent with the apparently normal ability of Oxt-/- male mice to detect predator cues , male mice with impaired OXT signaling in aromatase-positive ( Food zone = 82 . 7 ± 14 . 0 s , Food +fox urine zone = 51 . 8 ± 8 . 8 s; p<0 . 05 , t test; n = 10 ) or TRH positive neurons ( Female chamber = 65 . 8 ± 9 . 7 s , Fox urine zone = 41 . 1 ± 6 . 3 s; p<0 . 05 , t test; n = 11 ) still spent significantly longer durations in the goal chamber with the attractive cue of either food or female urine than the goal chamber containing fox urine ( Figure 2E and Figure 2F ) . Taken together , our results demonstrate that aromatase-positive neurons represent a critical site for OXT action in the control of chemosensory discrimination of male versus female conspecifics . Yet , the ablation of OXTR from aromatase-positive neurons using the Cyp19a1-Cre line also affects other sites expressing this gene . Of particular interest are the aromatase-expressing neurons of the bed nucleus of the stria terminalis ( BNST ) , another first order target of neurons in the AOB , and a key region critical for the processing the chemosensory information ( Figure 2—figure supplement 2C ) . To determine whether the function of OXT signaling in chemosensory discrimination of conspecifics is restricted to aromatase-positive neurons in the MeA , we used a virus-assisted strategy to selectively abolish OXT signaling in specific brain areas of adult mice . Adult Oxtrflox/flox male mice were stereotaxically injected with a recombinant AAV virus with CMV-driven expression of a GFP-Cre fusion protein ( AAV-GFP-Cre ) in the bilateral BNST or MeA ( Figure 3A ) . Mice receiving similar injection of an AAV virus with CMV driven expression of GFP ( AAV-GFP ) served as controls . The specificity of targeting was verified by examining GFP expression along the rostrocaudal axis of virally injected brains . No diffusion of virus between the two targets was observed . Using the chemostimuli-preference paradigm , we tested the ability of male mice with targeted manipulation of OXTR expression to discriminate between male and female conspecific cues . We found that mice with specific impairment of OXT signaling in the MeA displayed loss of preference for female chemostimuli ( male bedding = 40 . 8 ± 3 . 8 s , female bedding = 47 . 6 ± 5 . 7 s; p=0 . 29 , multiple t-tests; preference score = 0 . 04 ± 0 . 05 , n = 24 ) , whereas control ( GFP in MeA: male bedding = 25 . 7 ± 2 . 6 s , female bedding = 52 . 9 ± 6 . 8 s; p<0 . 01 , t test; preference score = 0 . 32 ± 0 . 06 , n = 10; GFP in BNST: male bedding = 31 . 6 ± 4 . 7 s , female bedding = 71 . 7 ± 9 . 9 s; p<0 . 01 , t test; preference score = 0 . 36 ± 0 . 08 , n = 10 ) and male mice with impaired OXT signaling in the BNST ( GFP in MeA: male bedding = 28 . 4 ± 2 . 8 s , female bedding = 79 . 0 ± 9 . 6 s; p<0 . 001 , t test; preference score = 0 . 42 ± 0 . 05 , n = 18 ) presented normal preference in investigating female bedding ( Figure 3B–G ) . The odor preference score of male mice with impaired OXT signaling in the MeA was significantly lower than that of control mice with GFP expression in the MeA ( p<0 . 01 , t test ) , whereas impaired OXT signaling in the BNST had no effect on the odor preference ( p=0 . 52 , t test ) . Thus , our results suggest that aromatase-positive neurons of the MeA , but not of the BNST , mediate the effect of OXT on the discrimination of sex-specific cues by male mice . To provide further evidence for the specific role of MeA aromatase-positive neurons in sex-cue discrimination , we performed a series of experiments in which OXTR expression was restored in either the MeA or BNST of Cyp19a1-Cre; Oxtrflox/flox male mice . A recombinant AAV virus in which the CMV promoter drives constitutive expression of OXTR ( AAV-OXTR ) or a control virus with CMV-driven expression of GFP was stereotaxically delivered ( Figure 3H ) and behaviors were measured using the chemostimuli preference paradigm ( Figure 3I–N ) . We found that rescue of OXTR expression in the MeA , but not the BNST , was sufficient to restore the preference for female chemostimuli in male mice . Bilateral injection of AAV-OXTR to the MeA led to increased investigation of female bedding in mutant mice ( male bedding = 35 . 7 ± 4 . 6 s , female bedding = 71 . 0 ± 6 . 9 s; p<0 . 01; t test; preference score = 0 . 33 ± 0 . 06 , n = 13; Figure 3I–K ) , whereas virus-mediated OXTR expression in the BNST had no effect ( male bedding = 35 . 3 ± 2 . 6 s , female bedding = 46 . 9 ± 4 . 0 s; p>0 . 05; t test; preference score = 0 . 13 ± 0 . 04 , n = 21; Figure 3L–N ) . In summary , our data indicate that OXT signaling is required in MeApd aromatase-positive neurons for the discrimination of sex-specific cues . This role may rely on one or two of the following mechanisms . OXT may have a developmental role on the aromatase-positive neurons of the MeA to ensure the proper configuration of this neuronal population to control sex discrimination . Alternatively , or in addition , OXT signaling may be acutely required in adults to modulate sex discrimination . OXT release is under dynamic regulation in adults , and can be triggered by sensory stimulations and psychosocial stressors ( Landgraf and Neumann , 2004; Veenema and Neumann , 2008 ) , consistent with the possibility that OXT is capable of acutely modulating the function of neural circuitries in adults in response to sensory cues . Although our results cannot rule out a possible role of OXT in configuring neural circuits regulating sex discrimination during development , the virus-mediated ablation and rescue of OXTR expression in adult MeA lead to corresponding changes in the ability of male mice to discriminate female and male conspecific cues , thus strongly suggesting that defects in chemosensory discrimination are a direct consequence of acute impairment of OXT signaling in the adult MeA . The MeA relays chemosensory inputs from the VNO pathway to hypothalamic nuclei controlling behavioral and endocrine responses , thus playing a key role in the regulation of social behaviors . To investigate how OXT signaling in the MeA alters sensory responses to social stimuli on a moment-to-moment basis , we used multisite extracellular recording and monitored neuronal activation patterns in the MeA upon odor stimulation in the presence or absence of OXT signaling ( Bergan et al . , 2014 ) . Multisite extracellular recordings of the MeA were conducted in anesthetized wild-type and mutant mice , with the VNO pump activated by electrical stimulation of the sympathetic nerve trunk to allow nonvolatile stimuli , including female urine , male urine and predator urine , to access the VNO neuroepithelium ( Figure 4; Ben-Shaul et al . , 2010 ) . Each sensory stimulus was presented with 6–12 randomly interleaved repetitions for all experiments . Electrophysiological probes consisted of 32 recording sites evenly distributed dorso-ventrally over 1 . 55 mm ( 50 μm between sites; Neuronexus ) . Identification of single unit activity was based on spike shape , clustering of principal component projections , and a clear refractory period between successive spikes ( Harris et al . , 2000; Hazan et al . , 2006 ) . Consistent with MeA anatomy , units responsive to VNO sensory stimuli were typically found from 0 to ~1 . 4 mm from ventral surface of the brain . Electrode probes were dipped in a fluorescent dye , and accurate targeting of probes was confirmed with postmortem histology ( Figure 4C ) . Sensory-evoked responses were identified by comparing the spike rates before stimulus presentation ( 20 s prior to stimulus presentation ) to spike rates following stimulus presentation ( 40 s following stimulus presentation ) using a non-parametric ANOVA . Units were considered 'responsive' if any stimulus elicited a p-value less than 0 . 01 . MeA neurons from control Oxtrflox/flox male mice were compared to that of mutant male mice lacking OXTR signaling in aromatase-positive neurons . A total of 106 single units were recorded from the MeA of control male mice ( 9 animals; 10–13 weeks old ) , with 40 units responding to VNO cues with a p value <= 0 . 01 ( non-parametric ANOVA ) and 191 single units were recorded from the MeA of mutant male mice ( 13 animals; 10–13 weeks old ) , with 70 units responding to VNO cues with a p value <= 0 . 01 ( non-parametric ANOVA ) . Single units responsive to VNO cues were categorized by the sensory stimulus that elicited the strongest response: female , male or predator ( Figure 4D ) . In accord with previous c-fos and electrophysiological findings ( Bergan et al . , 2014; Choi et al . , 2005; Kang et al . , 2011; Samuelsen and Meredith , 2009 ) , MeA units in male mice with wild-type OXTR function were most responsive to female ( 23 . 6% ) and predator ( 13 . 2% ) cues , while less than 1% of MeA units from control males responded to male stimuli ( Figure 4D–G ) . By contrast , in male mice lacking OXT signaling in the aromatase-expressing neurons , the proportion of MeA units responsive to female stimuli was reduced ( 6 . 2% ) , while the proportions of units responsive to male ( 6 . 7% ) and predator ( 23 . 3% ) stimuli were increased compared to control male mice ( Figure 4D ) . The observed difference in response rates between control mice and male mice lacking OXT signaling in the aromatase-expressing neurons was significant for each stimulus ( female: p<=0 . 0001; male: p<=0 . 02; predator: p<=0 . 02; permutation test ) . We found that the response strength ( spikes/second ) to conspecific cues was not dramatically changed by permanent loss of OXT signaling although the average response strength of predator neurons was reduced by ~10% in mutant male mice as compared to control male mice ( Figure 4E; p<0 . 05 ) . These results suggest that the inability to discriminate sex-specific cues observed in mice with impaired OXT signaling ( Figure 2 ) coincides with a clear reorganization of sensory responses to social stimuli in the MeA . To further elucidate the temporal requirement of OXT signaling in the chemosensory discrimination of female and male cues , we transiently suppressed OXT signaling by intraperitoneal injection of OXTR antagonist ( L-368 , 899 hydrochloride ) at a dosage of 5 mg/kg−1 ( Dölen et al . , 2013; Nakajima et al . , 2014 ) , followed by behavioral analysis ( Figure 5A–C ) and electrophysiological recording ( Figure 5D–J ) . We found that OXTR antagonist administration specifically reduced the time that male mice spent in the female interaction zone ( Figure 5A–B ) , without affecting the time spent in the male interaction zone or the total distance travelled ( saline: female interaction zone = 159 . 8 ± 18 . 1 s , male interaction zone = 93 . 8 ± 8 . 9 s; p<0 . 001; preference score = 0 . 24 ± 0 . 04 , total distance travelled = 3709 ± 160 cm; OXTR antagonist: female interaction zone = 106 . 4 ± 11 . 0 s , male interaction zone = 101 . 6 ± 12 . 7 s; p=0 . 79; preference score = 0 . 04 ± 0 . 08 , total distance travelled = 3360 ± 247 cm; n = 12; Figure 5C ) . The changes at the behavioral level could be observed 30 min after IP injection , demonstrating that normal social interactions with conspecifics rely on OXT signaling in adults and that acute modulation in OXT signaling can promptly modify behavioral responses . To assess the acute function of OXT signaling at the sensory response level , we compared the responses of single MeA neurons before and after systemic OXTR antagonist injection ( Figure 5D–J ) . Before pharmacological manipulation , responses of MeA units to repeated sensory stimulation were determined as described previously . Neurons were monitored without sensory stimulation for 10 min before and 10 min after OXTR antagonist IP injection to identify a direct effect of the compound on neural activity . Data show that , in the absence of sensory stimulation , injection of the OXTR antagonist decreased the firing rate of MeA units ( Figure 5E ) . Sensory responses were then reassessed in the presence of OXTR antagonist ( Figure 5D ) . We restricted our analyses to single MeA units that were maintained before , during , and after antagonist injection . A total of 129 single units from the MeA of Oxtrflox/flox male mice were maintained for the entire experiment , which typically lasted 2 hr ( 11 animals; 10–13 weeks old ) . Data show that OXTR antagonist reduced the number of neuronal responses categorized as 'female' ( Figure 5G; 16 . 4% responding before; 3 . 3% responding after; p<=0 . 001 , permutation test ) , increased the number of neuronal responses categorized as 'male' ( 0 . 8% responding before; 5 . 7% responding after; p<=0 . 026 , permutation test ) , and increased the number of neuronal responses categorized as 'predator' ( 0 . 8% responding before; 5 . 7% responding after; p<=0 . 026 , permutation test ) . Because units were classified based only on the stimulus that elicited the strongest response , these categorization changes could reflect a reduction in female responses and increase in male and predator responses , or both . A single MeA unit , showing a particularly dramatic antagonist mediated change in sensory responses , is shown in Figure 5F . Prior to drug injection this unit responded approximately evenly to both predator and female stimuli . Following drug injection , the unit responded nearly exclusively to predator stimuli . Next , we compared the response strengths for all units that responded in either the pre-drug or post-drug epoch ( see Materials and methods ) . OXTR antagonist application dramatically reduced the strength of female-responsive units ( p=0 . 003; Wilcoxon signed rank test; Figure 5H , red ) , modestly increased the strength of male-responsive units ( p=0 . 07; Wilcoxon signed rank test; Figure 5H , blue ) , and showed no clear effect on predator responsive units ( p=1; Wilcoxon signed rank test; Figure 5H , green ) . This suggests that the OXT-dependent changes in MeA sensory responses persists into adulthood , and that the neural circuits regulating social discrimination can be manipulated acutely by repressing OXT signaling ( Figure 5I , J ) . Moreover , in view of the decrease in response strength to female stimuli following OXTR antagonist ( Figure 5H ) , but not males or predators , it is clear that the reduced response to female stimuli is the major effect of OXTR antagonist in the MeA . To further confirm the critical role of OXT-mediated acute neuromodulation in the discrimination of female and male conspecifics , we investigated the effects of direct manipulation of OXT-expressing neurons in conspecific social interaction . OXT-expressing neurons are mainly distributed in two hypothalamic areas , the paraventricular nucleus ( PVN ) and the supraoptic nucleus ( SON ) . To systematically map the distribution and origins of OXT fibers from the two populations of OXT-expressing neurons , a Cre-dependent AAV-ChR2-YFP virus that readily enables visualization of neuronal fibers was stereotaxically delivered to either the PVN or the SON of an OXT-iCre line ( 2 female and 2 male virgin mice for each target site; Figure 5—figure supplement 1 ) ( Gradinaru et al . , 2009; Wu et al . , 2012 ) , and fine OXT fibers could be observed after GFP immunostaining . The distribution of OXT fibers did not exhibit any obvious sexual dimorphism between virgin female and male mice , and female and male results were combined when analyzing the fiber distribution patterns from OXT-expressing neurons in the PVN or the SON . Consistent with previous reports ( Buijs et al . , 1983; Knobloch et al . , 2012 ) , our results show that OXT-expressing neurons in the PVN , but not those in the SON , are the major contributor of OXT fibers in all 27 brain areas examined ( Figure 5—figure supplement 1 ) . Medium to high levels of OXT fibers originated from the PVN can be found in the BNST and MeA , two chemosensory nuclei with dense populations of aromatase-expressing neurons , whereas only sparse OXT fibers from the SON were found in the two areas . Although OXT can be released from dendrites and somas , local axonal release of OXT is suggested to specifically control region-associated behaviors ( Knobloch et al . , 2012 ) . OXT-expressing neurons in the PVN , which are the major contributor of OXT fibers in the MeA , may also be the major provider of OXT peptides modulating neuronal responses of MeApd neurons to chemostimuli . To examine the effect of reversible modulation of OXT peptide levels on a longer time scale , we aimed to directly control the activities of OXT-expressing neurons by using the Designer Receptor Exclusively Activated by Designer Drugs ( DREADD ) . Cre-dependent AAV virus conditionally expressing the inhibitory DREADD ( AAV8-hSyn-DIO-hM4Di-mCherry ) was stereotaxically delivered either to the PVN ( Figure 6A , B , D , E , H , I , L and M ) or the SON ( Figure 6B , C , F , G , J , K , N and O ) of the Oxt-Cre line ( Figure 6—figure supplement 1 ) ( Armbruster et al . , 2007; Krashes et al . , 2011; Krashes et al . , 2014; Wu et al . , 2012 ) . After binding of the designer drug clozapine-N-oxide ( CNO ) , the inhibitory DREADD reduces neuronal activities via the Gi pathway , therefore preventing the release of OXT peptides . We assessed the effects of chemogenetic inhibition of OXT-expressing neurons in social interactions by comparing the preference of virally infected male mice after IP injection of saline and at various intervals after IP injection of CNO . As expected , male mice with saline injection showed a strong preference for investigating a female mouse over a male mouse in the 3-chamber paradigm ( Figure 6D–G ) . After a single injection of CNO , OXT-iCre mice with inhibitory DREADD virus infection in the PVN presented reduced preference in investigating female conspecifics 1 hr after drug injection ( habituation: empty male interaction zone = 121 . 2 ± 11 . 8 s , empty female interaction zone = 119 . 1 ± 8 . 0 s; p=0 . 92 , t test; preference score with empty wire cups = 0 . 01 ± 0 . 04; saline: male interaction zone = 90 . 2 ± 12 . 2 s , female interaction zone = 171 . 5 ± 21 . 2 s; p<0 . 001 , t test; social preference score = 0 . 31 ± 0 . 06 , significantly higher than that with empty wire cups , p<0 . 001 , paired t test; CNO: male interaction zone = 108 . 2 ± 13 . 3 s , female interaction zone = 114 . 7 ± 18 . 9 s; p=0 . 76 , t test; social preference score = −0 . 002 ± 0 . 08 , not significantly differing from that with empty wire cups , p=0 . 92 , paired t test; n = 10; Figure 6D–E ) . This phenomenon of impaired preference could still be observed 2 days after CNO injection ( habituation: empty male interaction zone = 121 . 5 ± 10 . 0 s , empty female interaction zone = 116 . 3 ± 7 . 2 s; p=0 . 76 , t test; preference score with empty wire cups = −0 . 02 ± 0 . 03; Choice: male interaction zone = 145 . 6 ± 13 . 7 s , female interaction zone = 142 . 6 ± 15 . 6 s; p=0 . 86; social preference score = −0 . 02 ± 0 . 05 , not significantly differing from that that with empty wire cups , paired t test; n = 10; Figure 6H–I ) , while mice appeared to regain normal preference within 1 week of drug injection ( habituation: empty male interaction zone = 130 . 1 ± 10 . 1 s , empty female interaction zone = 121 . 70 . 1±9 . 4 s; p=0 . 68 , t test; preference score with empty wire cups = −0 . 03 ± 0 . 03; choice: male interaction zone = 111 . 8 ± 13 . 7 s , female interaction zone = 187 . 3 ± 20 . 8 s; p<0 . 001 , t test; social preference score = 0 . 25 ± 0 . 05 , significantly higher than that with empty wire cups , p<0 . 001 , paired t test; n = 10; Figure 6L–M ) . Thus , direct modulation of the neuronal source of OXT in the PVN can lead to fast and reversible changes in social interaction preference . Consistent with the PVN being the main origin of OXT fibers in the MeA , OXT-iCre mice with DREADD virus infection in the SON did not show any significant change in social interaction after CNO injection ( differences in time spent in interaction zones tested via the t test , and differences in social preference scores tested via paired t test; p<0 . 001 , p<0 . 01 , *p<0 . 05; n = 7; Figure 6F–G , J–K and N–O ) . Two sets of controls confirmed that the observed DREADD-mediated effects of CNO were specific to the activation or inhibition of OXT-expressing neurons . First , the behavior of a given mouse was compared before and after CNO injection . Prior to CNO injection , Oxt-iCre male mice with DREADD virus infection in either PVN or SON presented normal preference for female mice , excluding the possibility of ligand-independent effect of DREADDs . Second , in contrast to the effects observed post CNO injection with Oxt-iCre male mice expressing the inhibitory DREADD in the PVN , Oxt-iCre male mice expressing the inhibitory DREADD in the SON presented normal preference towards female mice before and after CNO injection , excluding the possibility that the impaired social preference is due to an off-target effect of CNO or its metabolite clozapine ( Gomez et al . , 2017; Saloman et al . , 2016 ) . As a complement to pharmacological and genetic manipulations inhibiting OXT signaling , we explored whether enhancing the activity of OXT-expressing neurons in the PVN would alter sexual preference behavior and electrophysiological response profiles of neurons in the MeA of males . To achieve this , we targeted the PVN of Oxt-iCre mice with excitatory DREADD ( AAV8-hSyn-DIO-hM3Dq-mCherry ) to allow the activation of PVN OXT-expressing neurons by IP injection of CNO ( Figure 6—figure supplement 1 ) . Elevated central OXT levels are known to induce self-grooming of the facial , truncal , and genital areas in rodents ( Amico et al . , 2004; Caldwell et al . , 1986; Drago et al . , 1986 ) , which can be blocked by OXTR antagonists ( Amico et al . , 2004 ) . Indeed , IP injection of CNO caused intense self-grooming in Oxt-iCre mice expressing excitatory DREADD in the PVN OXT-expressing neurons , indicating acute elevation of central OXT signaling . This intense grooming behavior starting minutes after CNO injection was seen to gradually diminish only after about 15 min , impairing our ability to detect the immediate changes in social preference behaviors triggered by acute elevation of OXT levels . When tested 15 min after CNO injection , that is after the phase of intense grooming had subsided , Oxt-iCre mice expressing excitatory DREADD in the PVN showed a loss preference for female stimuli ( Figure 7—figure supplement 1A ) , which did not recover 2 days , 1 week and even 2 weeks post CNO injection ( Figure 7—figure supplement 1B–C ) . To investigate the impact of acute increase in OXT levels on neuronal responses to sensory stimuli , we profiled the responses of single units in the MeA of male Oxt-iCre mice injected with excitatory DREADDs in the PVN before and after CNO injection ( Figure 7D ) . CNO modestly increased the firing rate of MeA units in the absence of sensory stimulation ( Figure 7E ) . A total of 98 single units were recorded from the MeA to compare the effect of chemogenetic activation of OXT-expressing neurons in the PVN . A single MeA unit , illustrating an increase in responses to female stimuli is shown in Figure 7F . Though several spikes are evident prior to CNO injection , this unit was relatively unresponsive . Following CNO injection , a strong response to female stimuli is revealed , demonstrating that increased response strength to female stimuli can be observed in units that were almost entirely silent prior to DREADD-mediated activation ( Figure 7F ) . We found that the OXT activation by CNO increased the number of neuronal responses categorized as 'female' ( Figure 7G; 16 . 9% responding before; 29 . 3% responding after; p<=0 . 03 , permutation test ) , modestly decreased the number of neuronal responses categorized as 'male' ( 3 . 8% responding before; 0 . 0% responding after; p<=0 . 069 , permutation test ) , and had little effect on the number of neurons categorized as predator ( 12 . 3% responding before; 10 . 3% ) . We next compared the response strengths for all units that responded to sensory stimuli before or after CNO injection . CNO injection increased the strength of female-responsive units ( p=0 . 0005; Wilcoxon signed rank test; Figure 7H , red ) , while having little effect on the response strength of male ( p=0 . 63; Wilcoxon signed rank test ) or predator ( p=0 . 82; Wilcoxon signed rank test ) evoked neural responses . This suggests that the impact of PVN-derived OXT on MeA sensory processing is largely restricted to opposite sex stimuli . Interestingly , electrophysiological recordings enabled us to assess the effect of DREADD activation of OXT neurons in the PVN more immediately than was possible in our behavior tests . The enhancement of female stimulus evoked responses by chemogenetic activation was most pronounced immediately after CNO injection and declined during subsequent stimulus presentations . The time range during which DREADD activation most strongly influenced sensory responses was marked by intense grooming that prohibited tests of social preference in the corresponding behavior experiments . These findings suggest that DREADD activation initially impacted MeA responses strongly , but led to a subsequent inactivation of OXT signaling . In support of this hypothesis , a second dose of CNO administered at the end of each electrophysiology experiment did not noticeably influence the firing rate of MeA units ( data not shown ) . These results are consistent with a relatively short duration ( ~15 min ) of CNO driven enhanced OXT activity , followed by a long-lasting inhibition of OXT signaling , which may result from depletion of presynaptic OXT and/or desensitization of MeA OXTR . AVP and OXT regulate overlapping social behaviors with the specific function of the two peptides differing according to the sex and age of the animal , and the target neuronal population ( Johnson and Young , 2017; Stoop , 2012 ) . For example , early studies using the monogamous prairie voles suggest that OXT and AVP act to facilitate the formation of pair-bonding in both male and female voles , but male voles are more sensitive to AVP manipulations while female voles are more sensitive to OXT manipulations ( Cho et al . , 1999; Young et al . , 2011 ) . OXT and AVP have also been shown to act via different neural circuits to regulate a given behavior . For instance , mice with impaired OXT or OXTR function and mice with genetic ablation of V1aR all present impairment in social recognition of familiar and novel conspecifics ( Bielsky et al . , 2004; Ferguson et al . , 2001; Ferguson et al . , 2000; Wersinger et al . , 2008 ) . However , pharmacological manipulation using receptor antagonists and virus-mediated rescue demonstrate that signaling critical for social recognition acts via OXTR in the MeA for , and via V1aR in the lateral septum ( Bielsky et al . , 2005; Ferguson et al . , 2001 ) . Our results reveal a robust specificity in the role played by OXT in regulating the sensory processing of sex-specific social cues in males . We extend previous findings by showing that in the 3-chamber-social-investigation paradigm OXT , but not AVP , plays a unique role in the discrimination of sex-specific cues by males and that this signaling is exclusively mediated by OXTR in the MeA , and does not involve V1aR nor V1bR . Moreover , despite the well-established cross reactivity of OXT and AVP on all of OXTR , V1aR and V1bR , and the presence of rich AVP fiber tracks in the MeA , it appears that endogenous AVP does not compensate the defects in social interactions resulting from loss of OXT . Our behavioral experiments revealed little bias in the time female mice spent interacting with male versus female mice . We note that this result differs from previous reports in which a preference of intact female mice for male-derived chemo-stimuli or intact males was demonstrated ( Bakker et al . , 2002; Brock and Bakker , 2011; Chalfin et al . , 2014; D'Udine and Partridge , 1981; Ramm et al . , 2008 ) . This discrepancy may result from differences in the genetic backgrounds of the mouse lines or slight differences ( e . g . , stimulus source ) in the behavioral paradigms used by each study . Regardless , we found that presence or absence of OXT did not noticeably impact the preference of female mice for male versus female stimuli . This suggests that OXT plays a sex-specific role in sculpting the social preferences of male versus female mice . OXT has been shown to lead to sexually dimorphic behavioral outcomes due to the sex-specific sensitivities of local circuits to hormones ( Li et al . , 2016; Nakajima et al . , 2014 ) . In our study , the sex specificity in OXT function coincides with a sexual dimorphism in MeA organization and function . The site of OXT action has been mapped to the MeApd aromatase-expressing neurons , with more MeApd aromatase-expressing neurons in males than females ( Wu et al . , 2009 ) . The observed sex-specific role of OXT in social preference might be due to gender-specific differences in MeA physiology or circuitry . In further support of this idea , OXT was also found to selectively affect the plasticity of neuronal population activity in the MeA of awake behaving males , but not females , after sexual experience ( Li et al . , 2016 ) . Since we observed no clear effect of OXT on social preference displayed by females , we focused our subsequent studies on male mice where the impact of OXT on social preference was clear and reliable . OXT has been implicated in the modulation of a large number of social behaviors , including social recognition and social learning ( Ferguson et al . , 2001; Ferguson et al . , 2000; Takayanagi et al . , 2005; Wersinger et al . , 2008 ) , the formation of pair bonds ( Young et al . , 2011 ) , parental care ( Takayanagi et al . , 2005 ) , social defeat ( Guzmán et al . , 2013 ) and reinforcement of social reward ( Dölen et al . , 2013 ) . Recently , impaired OXT function has been implicated in the abnormal social communications seen in certain human psychiatric diseases such as autism and schizophrenia , although a causal role of OXT in specific human behaviors and associated disorders is still to be clearly established ( De Berardis et al . , 2013; Gordon et al . , 2013; Heinrichs and Domes , 2008; Owen et al . , 2013; Pobbe et al . , 2012 ) . In rodents , chemosensory cues emitted by individuals are essential to encode the animal’s social and physiological status , in turn leading to specific social behavioral responses between conspecifics such as mating , fighting or parenting . OXT mutant mice have been shown to display impaired social memory and altered responses to socially relevant chemosensory cues in the MeA and in MeA projections ( Ferguson et al . , 2001 ) , supporting the notion that OXT modulates social behaviors by directly affecting the MeA processing of chemosensory cues encoding social information . Importantly , previous studies and the experiments we report here suggest that sensory modulation by OXT is specific for social cues and does not affect the detection of odor cues not relevant for social behavior . To further investigate mechanisms underlying the role of OXT in male mouse social preference , we used genetic and viral tools to conditionally impair OXT signaling in distinct brain regions and neuronal populations , and mapped the site of OXT action to the sexually dimorphic aromatase-expressing neurons of the MeA . Furthermore , virus-mediated expression of OXTR in the MeA rescues sex-specific odor discrimination impairment in Cyp19a1-Cre;Oxtrflox/flox male mice , whereas expression of OXTR in the BNST alone has no effect , thus helping to further delineate the site of OXT action to aromatase-expressing neurons of the MeA . This in turn led us to perform in vivo multisite extracellular recordings and directly monitor the responses of individual MeA neurons to conspecific and predator stimuli with and without OXT signaling . Our results showed that , in the absence of OXTR , MeA neurons display overall reduced responses to chemosensory cues , while the percentage of neurons responding to male cues is increased , suggesting that the lack of preference of OXT mutant mice for sex-specific cues results from abnormal patterns of MeA neural activity . Moreover , consistent with a specific role of OXT in the processing of conspecific cues , neuronal responses to mouse cues are reduced more than those to predator cues in OXT mutants . Importantly , the observed behavioral and neural phenotype observed in OXT mutants may result from defects in early OXT-dependent developmental processes , or from the acute lack of OXT-mediated signaling . Our data showing that similar sex-preference behavioral defects are observed in constitutive as well as in conditionally virus-mediated OXTR ablation in the adult MeA strongly suggest a role for OXT signaling in adult . This conclusion is further supported by the ability of OXTR-expressing viral injection in the adult MeA , but not BNST , to rescue the phenotype of OXTR mutants , and by the observation of behavioral and neural phenotypes resulting from acute reduction of OXT activity in adults using chemogenetic or pharmacological approaches . Therefore , the effect of OXT mediated through OXTR in aromatase-expressing neurons is not developmental , but rather acts on a moment-to-moment basis even in adults . Our previous study showed that the establishment of the sexually dimorphic sensory representation in the MeA requires steroid signaling near the time of puberty to organize the functional representation of sensory stimuli ( Bergan et al . , 2014 ) . In contrast to the organizational effect of steroids , we found here that the OXT-mediated acute modulation of MeA neuronal response can alter social behavior in the adult on a moment-to-moment basis , while maintaining the integrity of circuit connectivity . Social recognition is similarly modulated by acute variation in OXT signaling ( Ferguson et al . , 2001 ) . Our results further highlight the important role of neuropeptides in allowing the functional flexibility of neuronal circuits driving genetically pre-programmed instinctive behaviors . Recent studies have shown that specific activation of aromatase-expressing neurons in the MeA promotes aggressive behaviors ( Unger et al . , 2015 ) , and that reproductive sensory cues are processed by both the MeApv ( Ishii et al . , 2017 ) and the MeApd ( Bergan et al . , 2014; Choi et al . , 2005 ) . The overlapping control of aggressive and reproductive behaviors in the MeA mirrors the neuroethological argument that aggression is best understood as a component of reproductive behavior ( Tinbergen , 1951 ) . In further support of this view , our data show that OXT signaling through MeA aromatase-expressing neurons modulates the preference of male mice for interacting with female versus male mice , thus mediating a change in social preference that is likely to alter the balance of reproductive versus aggressive behaviors . The change in sensory responses observed in the MeA parallels the changes observed in male social preference , suggesting that the observed behavioral effect of OXT signaling is achieved by filtering which sensory responses have access to behavioral centers in the hypothalamus on a moment-to-moment basis . We showed that OXT enhances the responses of neurons in the MeA that encode female stimuli while suppressing sensory responses to both male and predator stimuli . This increased salience of female stimuli observed in this study is reminiscent of the strengthened responses for attended auditory and visual stimuli ( Desimone and Duncan , 1995; Winkowski and Knudsen , 2006 ) . OXT is dynamically regulated by the external environment and an animal's internal physiological state ( Devarajan and Rusak , 2004; Higashida et al . , 2017; Kalin et al . , 1985; Kendrick et al . , 1986; Wathes et al . , 1992 ) , and the OXT-mediated filtering of sensory cues shown in our study likely supports a fine-tuning of behavioral responses according to the social context . An intriguing question raised by our data is how is the specificity of OXT action to conspecific responses being achieved in the MeA . Our study identifies aromatase-expressing neurons in the MeA as a necessary and sufficient site for OXT to shape the social preference of male mice to explore female stimuli . OXT has been shown to regulate specific brain functions by facilitating the release of other neurotransmitters , in particular dopamine and serotonin ( Dluzen et al . , 2000; Dluzen et al . , 1998; Dölen et al . , 2013; Liu and Wang , 2003; Olazábal and Young , 2006; Ross et al . , 2009 ) . In addition OXT has been documented to increase neuronal firing rates , mainly , though not exclusively in fast-spiking interneurons , resulting in lower background activity and enhanced information transfer in cortical and non-cortical circuits ( Marlin et al . , 2015; Oettl et al . , 2016; Owen et al . , 2013; Xiao et al . , 2017 ) . These mechanisms may similarly operate in the MeA , and understanding how OXT signaling in aromatase-expressing neurons alters social preferences and the representation of social cues is an important line of research for future studies . Altogether , our results uncover aromatase-expressing MeA neurons as a hub for OXT-dependent modulation of female-evoked neural responses and behaviors in the male mouse . OXT selectively increases MeA responses to female stimuli , indicating that OXT signaling has the ability to differentially modulate the specific sensing of female cues through its action on MeA neurons co-expressing aromatase and OXTR . The MeA receives direct sensory input from the AOB and is reciprocally connected with hypothalamic behavior centers . Although sex-specific sensory cues are relayed separately at the receptor level , AOB mitral cells , which send input to MeA neurons , often respond to multiple sensory stimuli . It is possible that MeA neurons co-expressing aromatase and OXTR receive input from AOB neurons that are preferentially responsive to female stimuli . Alternatively , the specificity of OXT in modulating female-evoked neural responses may be achieved by regulating the local MeA circuitry , for example by acting on a specific set of local aromatase- and OXTR-expressing interneurons that boost responses to female cues while dampening responses to male and predator signals . These possibilities are not mutually exclusive . Understanding the molecular identity and circuit connectivity of neurons both directly and indirectly modulated by OXT will be essential for understanding how OXT regulates female-evoked neural responses and behaviors in male mice . The data presented here provide significant new insights into the role of OXT signaling in the control of social behavior . OXT selectively enhanced neural and behavioral responses to female stimuli , suggesting that subsets of female-responsive and male-responsive neurons in the MeA likely express different levels of OXTR . Thus , by dynamically and selectively tuning the neural representation of social cues in a genetically defined neuronal population of the MeA , OXT may alter the behavioral significance of social stimuli on a moment-to-moment basis . All mice were maintained in 12 hr:12 hr light:dark cycles with food and water available ad libitum . Animal care and experiments were carried out in accordance with the NIH guidelines and approved by the Harvard University Institutional Animal Care and Use Committee . The conditional oxytocin receptor knockout mice ( B6 . 129 ( SJL ) -Oxtrtm1 . 1Wsy/J , stock number: 008471 ) were from Jackson laboratories . The line had been backcrossed to the C57BL/6J inbred mice for at least 11 generations . V1aR knockout mice ( B6 . 129P2-Avpr1atm1Dgen/J , stock number: 005776 ) and V1bR knockout mice ( B6;129 × 1-Avpr1btm1Wsy/J , stock number: 006160 ) were obtained from Jackson laboratories and were backcrossed to the C57BL/6J inbred mice for at least six generations in our facility . The Oxt-iCre and Trh-Cre transgenic mouse lines were previously reported ( Krashes et al . , 2014 ) . The lines were previously maintained on a mixed C57BL/6 × 129S background and were backcrossed to C56BL/6J mice for our study . The Cyp19a1-Cre line was generated by BAC recombination , such that the Cre coding sequence was inserted in front of the start codon of the Cyp19a1 gene , which encodes aromatase . The ATG codon of the Cyp19a1 gene in the BAC was additionally mutated into a TTG stop codon . The absence of germline recombination events due to the presence of Cre and floxed alleles together in the germline was confirmed by two genotyping strategies . One pair of genotype primers ( P1: 5’-TGAAGAAGGATGGGCTTTTG-3’; P2: 5’- GGTCCCAGGAAAGAGTCAGC-3’ ) was designed to amplify both WT and floxed loci of the conditional OXTR knockout allele ( 244 bp and 341 bp , respectively ) . Another pair of primers ( P1; P3: 5’- TGGGAGTCCAGAGATAGTGGAA-3’ ) was designed to amplify the recombined locus ( 485 bp ) . The germline Cre-mediated deletion of floxed loci can be detected due to the lack of PCR products corresponding to the floxed loci ( 341 bp ) , and the presence of PCR products corresponding to the recombined locus ( 485 bp ) . We did not observe any germline recombination in our crosses . The absence of germline recombination and partial recombination due to activity in the early embryo was additionally confirmed for the Cyp19a1-Cre line by generating a double transgenic line , consisting of the Cyp19a1-Cre line and the Rosa26-lsl-tdTomato reporter . Mice containing both Cyp19a1-Cre and Rosa26-lsl-tdTomato were bred to maintain the double-transgenic line . The expression of tdTomato was observed for multiple generations: no germline-mediated recombination was observed in this double-transgenic line , and the tdTomato expression pattern faithfully reflected the endogenous expression pattern of the aromatase gene , thus excluding the possibility of germline recombination or partial recombination . The AAV-GFP and AAV-GFP-Cre viruses of serotype one were ordered from Penn Vector Core; the AAV-hSyn-DIO-hM4Di-mCherry virus and AAV-hSyn-DIO-hM3Dq-mCherry of serotype eight were obtained from the University of North Carolina Vector Core . To generate AAV-OXTR-IRES-GFP virus , the CMV-OXTR-IRES-GFP-BGH PolyA cassette was inserted into the XbaI site of pSub801 , and the resulting plasmid was used to generate high titer virus of serotype one in the Harvard virus Core . Mice 6–8 weeks old were anesthetized by ketamine/xylazine . A total of 200 nl virus was stereotaxically injected into the target areas . The coordinates used for the PVN injections were: bregma: anterior-posterior , –0 . 90 mm; dorsal-ventral , –4 . 75 mm; lateral , ±0 . 25 mm , those for the SON were: bregma: anterior-posterior , –0 . 80 mm; dorsal-ventral , –5 . 50 mm; lateral , ±1 . 3 mm , those for the posterior medial amygdala were: bregma: anterior-posterior , –1 . 90 mm; dorsal-ventral , –5 . 0 mm; lateral , ±1 . 75 mm , and those for the BNST were: bregma: anterior-posterior , –0 . 34 mm; dorsal-ventral , –4 . 50 mm and −4 . 00 mm; lateral , ±0 . 75 mm . Mice recovered for 4 weeks after surgery before being individually caged for social behavior tests . Mice brains were dissected and fixed overnight in 4% paraformaldehyde at 4°C , and 50 μm sections were prepared using a vibratome . Brain sections were blocked in PBX ( PBS with 0 . 05% Triton X-100 ) containing 10% FBS for 1 hr at room temperature , following by overnight incubation in the appropriate primary antibodies at 4°C . The sections were then washed with PBX , and incubated with either Fluorophore-conjugated secondary antibodies or Biotin-SP-labeled secondary antibodies for 1 hr at room temperature . The sections were then incubated with fluorophore-labeled streptavidin when necessary , and mounted after washing in PBX . The following primary antibodies were used: rabbit anti-oxytocin ( Immunostar , AB911 , 1:1000 ) , rabbit anti-vasopressin ( Immunostar , 20069 , 1:1000 ) , and rabbit anti-GFP ( Invitrogen , Waltham , MA , A-11122; 1:1000 ) . Biotin-SP-AffiniPure Goat Anti-Rabbit IgG ( 111-065-144 ) , and CY3 streptavidin ( 016-160-084 ) were from Jackson ImmunoResearch Laboratories , West Grove , PA . Alexa Fluo 488 Anti-mouse IgG , Alexa Fluo 488 streptavidin was from Invitrogen , Waltham , MA , ( Catalog Number: S11223 ) . RNA in situ hybridization was performed as described previously ( Isogai et al . , 2011 ) . Briefly , mice brains were dissected and immediately embedded in OCT , and 20 μm sections were prepared using a cryostat . Brain sections on slides were washed with PBS once , fixed in 4% paraformaldehyde for 10 min at room temperature , washed three times with PBS , treated with acetylation solution ( 0 . 1 M triethanolamine with 2 . 5 μl ml−1 acetic anhydride ) for 10 min , and then incubated with the pre-hybridization solution ( 50% formamide , 5 × SSC , 5 × Denhardt’s , 2 . 5 mg ml−1 yeast RNA , 0 . 5 mg ml−1 herring sperm DNA ) for 2 hr . The slides were then added with the hybridization buffer containing the appropriate DIG or FITC labeled probes and subsequent hybridization was carried out in a sealed chamber at 68°C overnight . The brain sections were then washed sequentially in 5XSSC , 0 . 2XSSC and TNT buffer ( 100 mM Tris , pH 7 . 5 , 150 mM NaCl , 0 . 05% Tween 20 ) . After the post-hybridization washes , slides were incubated with anti-FITC-POD ( Roche , 1:250 ) and/or anti-DIG-POD ( Roche , 1:500 ) , followed by TSA amplification ( Perkin Elmer; Waltham , MA ) . A stock solution was made by dissolving 5 mg of CNO ( Sigma Aldrich , St . Louis , MO ) in 1 . 333 ml dH2O to make 3 . 75 mg/ml solution ( 10 mM ) . A working solution was prepared by diluting the working solution with normal saline at a ratio of 1:100 . Mice received IP injection of CNO at a dose of 0 . 3 mg/kg , followed by behavioral analysis . To observe acute behavioral effects , mice were injected with CNO once , followed by behavioral test one hour later . The OXTR antagonist ( OXTR-A , L-368 , 899 hydrochloride ) was purchased from Tocris Bioscience , Bristol , United Kingdom . A 10 mg/ml stock solution of OXTR-A was prepared in dH2O , and was delivered by intraperitoneal injection at a dosage of 5 mg/kg−1 , followed by behavioral analysis and electrophysiological recording . All behavioral tests were performed in the dark phase of the light/dark cycle . Mice aged between 2 to 4 months were used for behavioral tests . All behavioral comparisons were analyzed using the Prism6 software to determine the statistical significance of behavioral results . Statistical tests for odor and social preference were performed by comparing the bias in time an animal spent near stimulus A ( ex . female conspecific ) versus stimulus B ( ex . male conspecific ) to the bias in time spent in the same physical area without the respective stimuli ( ex . empty cups ) . In all cases , a family-wise error rate correction for multiple comparisons was made using the Holm-Sidak's method . Thus , our statistical tests account for any observed bias in an animal's preference for one stimulus while accounting for any innate spatial biases the animal may display . Mice were anesthetized , tracheotomized , and prepared for electrophysiological recording as previously described ( Ben-Shaul et al . , 2010; Bergan et al . , 2014 ) . A stimulating cuff electrode was placed around the rostral sympathetic nerve trunk to control the VNO pump and access of stimuli to the VNO sensory epithelium . A ~ one mm2 craniotomy was opened dorsal to the MeA based on stereotaxic coordinates , to allow the insertion of Neuronexus probes ( NeuroNexus Technologies , Ann Arbor , Michigan: a1 × 32–10 mm 50-500-413 ) coated with a fluorescent dye ( DiI or DiD; Invitrogen , Carlsbad , CA ) into the MeA ( Bergan et al . , 2014 ) . Accurate stereotaxic targeting to the MeA was confirmed by post mortem histological analysis of the electrode tract . Urine was collected from adult singly housed estrus female and singly housed male mice of the balb/C , CBA , or C57Bl6 strains and immediately placed in liquid nitrogen , for subsequent storage in −80C . Urine samples from fox , mountain lion , and bobcat were obtained from PredatorPee ( Lincoln , Maine ) . All sensory stimuli were presented at least six times in a randomized order for each dataset . VNO stimuli were applied by placing 1 ul of stimulus ( 1:100 dilution in Ringer's ) directly into the nostril , followed by activation of the sympathetic nerve cuff electrode to facilitate VNO pumping and stimulus entry to the VNO lumen . The nasal cavity and VNO was cleaned with Ringer's solution after each stimulus presentation ( Ben-Shaul et al . , 2010 ) . Animals were maintained under anesthesia while CNO or the OXTR antagonist were injected ( see above for additional injection details ) , and the electrophysiological recording was maintained throughout to allow identification of electrode movement due to the injection . Only units that were maintained throughout the recording were analyzed . All recording channels were band-pass filtered ( 300–5000 Hz ) and digitized at 25 kHz using an RZ2 processor , PZ2 preamplifier , and two RA16CH head-stage amplifiers ( TDT , Alachua , FL ) . Custom MATLAB ( Mathworks , Natick , MA ) scripts ( https://github . com/joebergan/Spike-sorting ) were used to extract 3 . 5 ms spike waveforms from the continuous data ( Bergan and Ben-Shaul , 2017 ) . A copy is archived at https://github . com/elifesciences-publications/Spike-sorting . Waveforms were extracted for the eight nearest electrode channels and units were typically visible on 2–4 contiguous electrode channels , which enhanced the isolation of single units . Single unit clusters were determined based on the principal components using KlustaKwik ( Harris et al . , 2000; Bergan et al . , 2014 ) and manually verified , and adjusted using Klusters ( Hazan et al . , 2006 ) . Single unit spike clusters were determined based on spike shapes , projections on principal component space ( calculated independently for each recording session ) , and autocorrelation functions ( interspike interval ) . Single units displayed a distinct spike shape that was fully separated from both the origin ( noise ) and other clusters ( multi-unit ) with respect to at least one principal component projection , and also displayed a clear refractory period in the interspike interval histogram . Valid sensory responses were identified by comparing the spike rate after sensory stimulation to the spike rate immediately preceding sensory stimulation for each unit . Responsive units passed a non-parametric ANOVA statistical test performed at the significance level of p≤0 . 01 . Response magnitude was quantified as the change in average firing rate during the 40 s following stimulus presentation relative to the firing rate during the 20 s prior to stimulus presentation . Unless otherwise noted , statistics comparing two populations of units were performed using a nonparametric permutation test ( Efron and Tibshirani , 1993 ) as observed distributions were not typically normally distributed .	Oxytocin is a hormone that promotes milk production , contractions during childbirth , and many social interactions in humans and other creatures . It has also been implicated in conditions like autism or schizophrenia , which show altered social interactions . Oxytocin is made and released by cells in the brain called neurons . The oxytocin-producing neurons are clustered in a brain region called the hypothalamus , and oxytocin can act over a long distance in the brain or in the body . Many mammals detect chemical signals called pheromones that are involved in social interactions . These chemicals are detected by neurons in a structure within the cartilage of the nose called the vomeronasal organ . Pheromone-sensing neurons in the vomeronasal organ connect with another part of the brain called the medial amygdala . The medial amygdala , in turn , connects with regions of the brain that control behavior . Mice in particular rely on pheromones for social communication . Male and female mice respond differently to pheromones . Male mice prefer to investigate female mice to other males . The neurons in medial amygdala of male mice also become more active in response to scents from females than from males . Oxytocin is known to act on the medial amygdala , but its exact role in the male’s preference for females and their scents is not known . Now , Yao et al . show that oxytocin controls male preference for interacting with females and their scents by turning on neurons in the medial amygdala . In the experiments , male mice genetically engineered to lack oxytocin do not prefer female mice to other males , and they also appear unable to distinguish male and female scents . These mice also have less activity in the neurons of the medial amygdala when exposed to females and their scents . Directly manipulating these neurons and the oxytocin receptors on them also altered sex-preferences in male mice . The experiments show that oxytocin alters the behaviors of male mice in response to females or their scents by manipulating a specific set of brain cells . More studies of these cells or their interactions with oxytocin might help scientists understand oxytocin-liked diseases that impair social interactions or develop new treatments for conditions like autism or schizophrenia .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Oxytocin signaling in the medial amygdala is required for sex discrimination of social cues
Cholesterol is a major structural component of the plasma membrane ( PM ) . The majority of PM cholesterol forms complexes with other PM lipids , making it inaccessible for intracellular transport . Transition of PM cholesterol between accessible and inaccessible pools maintains cellular homeostasis , but how cells monitor the accessibility of PM cholesterol remains unclear . We show that endoplasmic reticulum ( ER ) -anchored lipid transfer proteins , the GRAMD1s , sense and transport accessible PM cholesterol to the ER . GRAMD1s bind to one another and populate ER-PM contacts by sensing a transient expansion of the accessible pool of PM cholesterol via their GRAM domains . They then facilitate the transport of this cholesterol via their StART-like domains . Cells that lack all three GRAMD1s exhibit striking expansion of the accessible pool of PM cholesterol as a result of less efficient PM to ER transport of accessible cholesterol . Thus , GRAMD1s facilitate the movement of accessible PM cholesterol to the ER in order to counteract an acute increase of PM cholesterol , thereby activating non-vesicular cholesterol transport . Sterol is one of the major membrane lipids in eukaryotes . In metazoans , cholesterol represents ~20% of total cellular lipids and is therefore essential for the structural integrity of cellular membranes and for cell physiology ( van Meer et al . , 2008; Vance , 2015 ) . Sterol is distributed among cellular membranes primarily via non-vesicular transport , a process that is independent of membrane traffic ( Baumann et al . , 2005; Hao et al . , 2002; Heino et al . , 2000; Ikonen , 2008; Urbani and Simoni , 1990 ) . Levels of sterol vary considerably between different cellular membranes . Between 60% and 80% of total cellular cholesterol is concentrated in the plasma membrane ( PM ) , where it represents up to ~45% of total lipids in this bilayer ( de Duve , 1971; Lange et al . , 1989; Ray et al . , 1969 ) . Cellular cholesterol levels are maintained by regulated delivery and production , primarily through receptor-mediated endocytosis of low-density lipoproteins ( LDLs ) ( Goldstein and Brown , 2015 ) and de novo synthesis in the endoplasmic reticulum ( ER ) that is controlled by the activation of SREBP transcription factors ( Brown et al . , 2018; Goldstein and Brown , 1990 ) . Cholesterol is also supplied to cells via high-density lipoproteins ( HDL ) through the reverse cholesterol flux pathway ( Acton et al . , 1996; Phillips , 2014 ) . Cholesterol within the bilayer membranes exists in two distinct chemical states: one being free and ‘accessible’ ( also known as ‘unsequestered’ or ‘chemically active’ ) , and the other being ‘inaccessible’ ( also known as ‘sequestered’ or ‘chemically inactive’ ) owing in part to the formation of complexes with other membrane lipids , including sphingomyelin and phospholipids ( Chakrabarti et al . , 2017; Das et al . , 2014; Gay et al . , 2015; Lange et al . , 2013; Lange et al . , 2004; McConnell and Radhakrishnan , 2003; Ohvo-Rekilä et al . , 2002; Radhakrishnan and McConnell , 2000; Sokolov and Radhakrishnan , 2010 ) . Most cholesterol in the PM is sequestered , but a small fraction of PM cholesterol ( ~15% of PM lipids ) remains accessible for extraction and transport ( Das et al . , 2014 ) . Although the majority of cellular cholesterol resides in the PM , the biosynthesis of cholesterol occurs exclusively in the ER . Thus , the ER must communicate with the PM to monitor levels of PM cholesterol and to adjust cholesterol biosynthesis in order to maintain lipid homeostasis . To achieve this , cells sense transient increases in the accessible pool of PM cholesterol and rapidly transport the newly expanded pool of accessible PM cholesterol to the ER . This suppresses cholesterol biosynthesis by inhibiting SREBP-2 , a master regulator of de novo cholesterol synthesis , thereby avoiding cholesterol overaccumulation while maintaining PM cholesterol levels ( Das et al . , 2014; Infante and Radhakrishnan , 2017; Lange and Steck , 1997; Lange et al . , 2014; Scheek et al . , 1997; Slotte and Bierman , 1988 ) . Artificially trapping the accessible pool of cholesterol in the PM results in dysregulated activation of SREBP-2 ( Infante and Radhakrishnan , 2017; Johnson et al . , 2019 ) . Despite its critical importance , the intracellular transport machinery that senses the accessibility of PM cholesterol is unknown . This machinery is likely to respond to a sharp change in the accessibility of cholesterol on the cytoplasmic leaflet of the PM and to facilitate transport of accessible cholesterol from the PM to the ER , thereby helping the ER to communicate with the PM . Such a homeostatic system would also allow cells to monitor PM cholesterol accessibility in order to help to maintain cellular cholesterol homeostasis . The ER extends throughout the cytoplasm , forming physical contacts with virtually all other cellular organelles and the PM ( Phillips and Voeltz , 2016; Wu et al . , 2018 ) . Growing evidence indicates that these membrane contact sites play critical roles in cellular physiology , including lipid exchange and delivery via non-vesicular lipid transport that is facilitated by lipid transfer proteins ( LTPs ) ( Antonny et al . , 2018; Drin , 2014; Elbaz and Schuldiner , 2011; Holthuis and Menon , 2014; Jeyasimman and Saheki , 2019; Kumar et al . , 2018; Lahiri et al . , 2015; Lev , 2012; Luo et al . , 2019; Nishimura and Stefan , 2019; Petrungaro and Kornmann , 2019; Saheki et al . , 2016; Saheki and De Camilli , 2017a; Saheki and De Camilli , 2017b; Wong et al . , 2018 ) . Thus , LTPs may participate in intracellular cholesterol transport and may help to maintain PM cholesterol homeostasis by regulating non-vesicular cholesterol transport between the PM and the ER at ER–PM contact sites . Decades of biochemical and genetic research into cholesterol metabolism has identified several key LTPs that bind to cholesterol and mediate its non-vesicular transport ( Luo et al . , 2019; Wong et al . , 2018 ) . These proteins include a family of 15 proteins that contain a StAR-related lipid transfer ( StART ) domain , which binds and transports a wide variety of lipids , including cholesterol , glycerolipids , and sphingolipids ( Alpy and Tomasetto , 2014 ) . Five members of this family , namely STARD1 , STARD3 , STARD4 , STARD5 , and STARD6 , bind and transport cholesterol ( Alpy et al . , 2013; Iaea et al . , 2017; Lin et al . , 1995; Mesmin et al . , 2011; Soccio et al . , 2002; Stocco , 2001; Wilhelm et al . , 2017 ) , but they are not conserved in yeast . This lack of conservation suggests that there may be a more ancient family of sterol transfer proteins that control cholesterol homeostasis in all eukaryotes . A bioinformatics search for proteins that possess StART-like domains identified a novel family of evolutionarily conserved proteins that includes six Lam/Ltc proteins in budding yeast ( Gatta et al . , 2015; Murley et al . , 2015 ) , and five GRAM domain-containing proteins ( GRAMDs ) in metazoans . These GRAMDs include the StART-like domain-containing GRAMD1s , also known as Asters ( GRAMD1a/Aster-A , GRAMD1b/Aster-B , and GRAMD1c/Aster-C ) , and two highly related proteins that lack a StART-like domain ( GRAMD2 and GRAMD3 ) . Lam/Ltc proteins and GRAMDs all possess an N-terminal GRAM domain , which has structural similarity to the PH domain and thus may sense or bind lipids ( Begley et al . , 2003; Tong et al . , 2018 ) , and a C-terminal transmembrane domain , which anchors the proteins to the ER . Structural and biochemical studies of yeast and mammalian StART-like domains have identified a hydrophobic cavity that can bind sterol ( Gatta et al . , 2018; Horenkamp et al . , 2018; Jentsch et al . , 2018; Sandhu et al . , 2018; Tong et al . , 2018 ) . The StART-like domains of GRAMD1s bind and transport sterols in vitro ( Horenkamp et al . , 2018; Sandhu et al . , 2018 ) . Recent studies have demonstrated that some GRAMDs , including GRAMD1a , GRAMD1b , and GRAMD2 , localize to ER–PM contact sites ( Besprozvannaya et al . , 2018; Sandhu et al . , 2018 ) . GRAMD2 facilitates STIM1 recruitment to ER–PM contacts and potentially regulates Ca2+ homeostasis ( Besprozvannaya et al . , 2018 ) , whereas GRAMD1b facilitates the transport of HDL-derived cholesterol to the ER in the adrenal glands of mice ( Sandhu et al . , 2018 ) . By contrast , yeast Lam/Ltc proteins sense cellular stress and potentially regulate cholesterol exchange between the ER and other membranes ( Gatta et al . , 2015; Murley et al . , 2015; Murley et al . , 2017 ) . However , the role of these proteins in PM cholesterol or sterol homeostasis has been elusive . In this study , we provide evidence that GRAMD1s sense a transient expansion of the accessible pool of PM cholesterol and facilitate its transport to the ER at ER–PM contact sites , thereby contributing to PM cholesterol homeostasis . We found that GRAMDs form homo- and heteromeric complexes via their transmembrane domains and predicted the existence of luminal amphipathic helices that interact with each other within these complexes . We also found that GRAMD1s rapidly move to ER–PM contacts upon acute hydrolysis of sphingomyelin in the PM . We characterized the mechanisms of this acute recruitment and found that the GRAM domain acts as a coincidence detector of unsequestered/accessible cholesterol and anionic lipids in the PM , including phosphatidylserine , allowing the GRAMD1s to sense a transient expansion of the accessible pool of PM cholesterol once it increases above a certain threshold . We generated HeLa cells that lacked GRAMD1a/1b/1c ( i . e . , all of the GRAMDs that contain a StART-like domain ) and determined the effect of knocking out these proteins on cholesterol metabolism , using a combination of cholesterol-sensing probes for live cell imaging and lipidomics of membrane extracts . Upon treatment with sphingomyelinase , which liberates the sphingomyelin-sequestered pool of PM cholesterol into the ‘accessible’ pool and thus stimulates its PM to ER transport , GRAMD1 triple knockout ( TKO ) cells exhibited exaggerated accumulation of the accessible pool of PM cholesterol and reduced suppression of SREBP-2 cleavage compared to wild-type control cells . This accumulation resulted from less efficient transport of accessible cholesterol from the PM to the ER . Using structure–function analysis , we demonstrated that GRAMD1s couple their PM-sensing property and cholesterol-transport function via their GRAM and StART-like domains , and that GRAMD1 complex formation ensures the progressive recruitment of GRAMD1 proteins to ER–PM contacts . Finally , we observed striking expansion of the accessible pool of PM cholesterol in GRAMD1 TKO cells at steady state . Drug-induced acute recruitment of GRAMD1b to ER–PM contacts was sufficient to facilitate removal of the expanded pool of accessible cholesterol from the PM in GRAMD1 TKO cells . Collectively , our findings provide evidence for novel cellular mechanisms by which GRAMD1s monitor and help to maintain PM cholesterol homeostasis in mammalian cells . As one of the key homeostatic regulators , GRAMD1s sense a transient expansion of the accessible pool of PM cholesterol and facilitate its transport to the ER , thereby contributing to PM cholesterol homeostasis at ER–PM contact sites . Previous studies identified GRAMD1s as ER-resident proteins that are distributed throughout ER structures in a punctate pattern ( Sandhu et al . , 2018 ) . GRAMDs ( namely GRAMD1a , GRAMD1b , GRAMD1c , GRAMD2 , and GRAMD3 ) all possess an N-terminal GRAM domain and a C-terminal transmembrane domain . In addition , the three GRAMD1 proteins ( GRAMD1s ) possess a StART-like domain ( Figure 1A ) . Some LTPs are known to form homo- and heteromeric complexes . Thus , we reasoned that GRAMD1s may also interact with one another to form complexes . To further analyze the dynamics of these proteins on the ER at high spatial resolution , we tagged the GRAMD1s , as well as GRAMD3 , with fluorescent proteins and analyzed their localization using spinning disc confocal microscopy coupled with structured illumination ( SDC-SIM ) . Analysis of COS-7 cells expressing individual EGFP-tagged GRAMD1s or GRAMD3 ( EGFP-GRAMD1a , EGFP-GRAMD1b , EGFP-GRAMD1c , or EGFP-GRAMD3 ) and a general ER marker ( RFP-tagged Sec61β ) revealed enrichment of GRAMD1s and GRAMD3 in similar discrete patches along ER tubules . By contrast , RFP-Sec61β localized to all domains of the ER , including the nuclear envelope and the peripheral tubular ER network ( Hoyer et al . , 2018 ) ( Figure 1B and Figure 1—figure supplement 1A ) . When individual EGFP–GRAMD1s and either mRuby-tagged GRAMD1b ( mRuby-GRAMD1b ) ( Figure 1C ) or mCherry-tagged GRAMD3 ( mCherry-GRAMD3 ) ( Figure 1—figure supplement 1B ) were co-expressed in COS-7 cells , the patches of EGFP and mRuby/mCherry significantly overlapped , indicating potential complex formation between these proteins on tubular ER . To test whether these proteins form complexes , we examined biochemical interactions between GRAMD1s and GRAMD3 using co-immunoprecipitation assays . HeLa cells co-transfected with individual EGFP–GRAMD1s together with either myc-tagged GRAMD1b ( Myc–GRAMD1b ) ( Figure 1D and Figure 1—figure supplement 1C ) or myc-tagged GRAMD3 ( Myc–GRAMD3 ) ( Figure 1E and Figure 1—figure supplement 1D ) were lysed , and either anti-GFP ( Figure 1D , E ) or anti-Myc nanobodies ( Figure 1—figure supplement 1C , D ) were used to perform immunoprecipitation . Analysis of the resulting immunoprecipitates by western blotting ( i . e . immunoblotting ) revealed robust interaction between GRAMD1s and GRAMD1b ( Figure 1D and Figure 1—figure supplement 1C ) , as well as between GRAMD1s and GRAMD3 ( Figure 1E and Figure 1—figure supplement 1D ) . These results demonstrate that these proteins form both homo- and heteromeric complexes . The formation of homo- and heteromeric complexes between GRAMD1s and GRAMD3 suggested the presence of amino-acid sequence within these proteins that facilitate their interaction . Secondary structure predictions indicated the presence of a conserved alpha helix within the luminal region of GRAMD1s ( Figure 2A ) . Furthermore , helical wheel analysis of the luminal helix from GRAMD1b predicted that this protein contained an amphipathic helix , with charged and hydrophobic amino acids occupying opposite sides of the helix ( Figure 2B and Figure 2—figure supplement 1A , B ) . It is known that some amphipathic helices mediate protein–protein interactions through their hydrophobic surfaces ( Segrest et al . , 1990 ) . Therefore , we first asked whether the luminal helix was necessary for these proteins to form discrete patches on tubular ER . We focused on GRAMD1b as a model protein for analysis of the properties of the GRAMD1 luminal helices , generating a version of GRAMD1b that lacked the luminal helix ( Δhelix ) , and a second version in which the five hydrophobic residues within the luminal helix were mutated to glutamic acid ( 5E ) , thereby disrupting the hydrophobic surface ( Figure 2B and Figure 2—figure supplement 1C ) . Whereas GRAMD1b ( wild-type control ) formed patches on tubular ER , both GRAMD1b ( Δhelix ) and GRAMD1b ( 5E ) exhibited diffuse localization patterns , with fewer discrete patches on tubular ER ( Figure 2C ) . By contrast , a version of GRAMD1b in which the four hydrophobic residues preceding the luminal helix were mutated to glutamic acid ( 4E ) formed patches that were similar to those formed by the control ( Figure 2C ) , demonstrating that the 5E mutation specifically disrupted patch formation . The potential ability of the luminal helices to interact directly with one another was examined using cell-free assays . Wild-type luminal helices ( GRAMD1b674–718 ) and luminal helices with the 5E mutation ( GRAMD1b674–718 5E ) were purified individually as EGFP fusion proteins and analyzed by size exclusion chromatography ( SEC ) . Whereas the predicted molecular weights of the fusion proteins were the same ( ~35 kDa ) , wild-type luminal helices ( EGFP–helix: EGFP–GRAMD1b674–718 ) eluted at a much lower elution volume compared to 5E mutant luminal helices [EGFP–helix ( 5E ) : EGFP–GRAMD1b674–718 5E] ( Figure 2D ) . Blue native PAGE analysis ( BN-PAGE ) of the purified proteins revealed that wild-type helices migrated slower than the 5E mutants , indicating that interaction between luminal helices depended on the hydrophobic surface of GRAMD1b ( Figure 2E ) . By contrast , in the presence of SDS , the denatured forms of these proteins migrated similarly ( SDS-PAGE ) . Slightly slower migration of 5E mutants on the gel was possibly due to the increased hydrophilicity of this fragment compared to wild-type ( Guan et al . , 2015 ) ( Figure 2E ) . These results suggest that the luminal helix is probably amphipathic and is important for the formation of GRAMD1b complexes through its hydrophobic surface . Finally , the formation of GRAMD1 complexes was examined biochemically in cells using co-immunoprecipitation assays . Homomeric interactions between GRAMD1bs and heteromeric interactions between GRAMD1b and GRAMD1a were greatly reduced when the luminal helix of GRAMD1b was either removed ( Δhelix ) or mutated to the 5E version , supporting the important role of the luminal helix in homo- and heteromeric interactions of the GRAMD1s ( Figure 2F , G ) . Residual interactions were mediated by the transmembrane domain of GRAMD1b , as replacing this domain and its luminal region with those from Sec61β ( TM swap ) ( Figure 2J ) completely abolished the ability of GRAMD1b to form homo- and heteromeric complexes ( Figure 2F , G ) . Accordingly , GRAMD1b with the TM swap exhibited a diffuse localization pattern compared to that of wild-type GRAMD1b ( Figure 2H ) , and failed to interact with wild-type GRAMD1b on tubular ER ( Figure 2I ) . Thus , both transmembrane domains and luminal helices contributed to the formation of GRAMD1 complexes ( Figure 2J ) . Taken together , these results revealed the biochemical mechanisms by which GRAMDs form homo- and heteromeric complexes . As key residues contributing to the hydrophobic surface of the luminal helix are conserved among GRAMD1s ( Figure 2A and Figure 2—figure supplement 1A ) , they probably play a role in the heteromeric interactions of all of these proteins . Recent studies demonstrated that ‘cholesterol loading’ leads to the accumulation of GRAMD1s at ER–PM contact sites ( Sandhu et al . , 2018 ) . Within 20 min of treating cells with a complex of cholesterol and methyl-β-cyclodextrin ( cholesterol/MCD ) , GRAMD1b was indeed recruited to the PM ( Figure 3A , B; Video 1 ) . In addition , we found that GRAMD1a , GRAMD1c , and GRAMD3 were all recruited to ER–PM contacts upon cholesterol loading , with kinetics similar to GRAMD1b recruitment ( Figure 3B ) . However , a version of GRAMD1b that lacked the GRAM domain ( GRAMD1b ΔGRAM ) failed to localize to the PM , even after 30 min , indicating the essential role of this domain in sensing PM cholesterol ( Figure 3—figure supplement 1A; Video 2 ) . Although these results suggest that PM cholesterol plays a critical role in recruiting GRAMDs to ER–PM contacts , all of the GRAMDs localize to tubular ER at rest , even though a significant amount of cholesterol is already present in the PM ( Lange et al . , 1989; Ray et al . , 1969 ) . Thus , their GRAM domains may possess unique abilities to sense the accessibility of PM cholesterol , rather than detecting the total levels of PM cholesterol . However , it is not known whether the GRAM domains are able to sense accessible cholesterol in the PM . To elucidate the biochemical properties of the GRAMD1 GRAM domain , we first purified the GRAM domain of GRAMD1b and performed liposome sedimentation assays to test its ability to bind lipids . In this assay , purified GRAM domains were mixed with sucrose-loaded heavy liposomes in sucrose-free buffer . After incubation , free liposomes and the liposomes that bound to GRAM domains ( P ) were pelleted by centrifugation; the supernatant contained only unbound GRAM domains ( S ) ( Figure 3C , E , Figure 3—figure supplement 1B , D , and Figure 3—figure supplement 2A , B ) . The GRAM domain did not bind liposomes when the liposome contained only phosphatidylcholine ( Figure 3C and Figure 3—figure supplement 1B ) . By contrast , the GRAM domain bound liposomes that contained free cholesterol , although such binding was rather weak , and only ~25% of purified GRAM domains bound liposomes even when the liposome contained high levels of cholesterol ( Chol ) ( 60% ) ( Figure 3—figure supplement 1B ) . The GRAM domain also bound liposomes when the liposomes contained phosphatidylserine ( PS ) , the predominant anionic phospholipid in the PM . However , such binding only occurred when the liposomes contained non-physiological high levels of phosphatidylserine ( 50% or 80% ) ( Figure 3—figure supplement 1B ) . Thus , we explored the possibility that the GRAM domain may bind to membranes more efficiently in the presence of both lipids , thereby acting as a coincidence detector of unsequestered/accessible cholesterol and phosphatidylserine . Little binding was observed when the liposomes contained 50% cholesterol or 20% phosphatidylserine ( Figure 3C and Figure 3—figure supplement 1B ) . However , strong binding was observed when 50% cholesterol and 20% phosphatidylserine were both present in the liposomes ( ~80% of the GRAM domains bound to liposomes ) ( Figure 3C ) . Thus , the addition of free cholesterol dramatically enhanced binding of the GRAM domain to phosphatidylserine-containing membranes . Replacing cholesterol with a non-bilayer forming lipid , phosphatidylethanolamine ( PE ) , abolished the binding of the GRAM domains to liposomes , confirming the specific effect of cholesterol ( Figure 3—figure supplement 2A ) . Similar synergistic effects were observed with the GRAM domain of GRAMD1a ( Figure 3D ) , suggesting the conserved function of GRAMD1 GRAM domains . Despite the presence of phosphatidylserine ( ~10% of PM lipids ) and high levels of cholesterol ( ~45% of PM lipids ) in the PM of mammalian cells , GRAMD1s are not enriched at ER–PM contacts at rest ( Figure 1B , Figure 3A and Figure 4—figure supplement 3B ) . The majority of cholesterol in the PM ( ~27% of PM lipids ) is sequestered and ‘inaccessible’ to cytosolic proteins , and only ~15% of PM lipids remain unsequestered and accessible ( Das et al . , 2014 ) . Thus , interactions between GRAM domains and the PM could be suppressed by ‘the factors that sequester cholesterol’ in this bilayer in cells . One of the major factors that mediate the direct sequestration of PM cholesterol is sphingomyelin , which forms a complex with cholesterol and makes it inaccessible ( Endapally et al . , 2019; Finean , 1953; McConnell and Radhakrishnan , 2003; Radhakrishnan and McConnell , 2000; Slotte , 1992 ) . The sphingomyelin-sequestered pool of PM cholesterol consists of ~15% of PM lipids , while the rest of the inaccessible pool is sequestered by other membrane factors ( Das et al . , 2014 ) . To test whether the sequestration of cholesterol by sphingomyelin affects the binding of GRAM domains to artificial membranes , we incorporated increasing amounts of sphingomyelin ( SM ) ( 10% or 25% ) into liposomes that contained 50% cholesterol and 20% phosphatidylserine ( Figure 3E ) . When these liposomes contained 25% sphingomyelin , the percentage of GRAMD1b GRAM domains that bound to the liposomes decreased from ~80% to ~45% ( Figure 3E and Figure 3—figure supplement 1C ) . Similar results were obtained with the GRAM domain of GRAMD1a ( Figure 3F ) . Thus , the binding of GRAM domains to artificial membranes that contain cholesterol and phosphatidylserine can be modulated by the presence of sphingomyelin ( Figure 3—figure supplement 1C ) . These results suggest that sphingomyelin helps to suppress the binding of GRAM domains to the PM at rest by reducing the accessibility of the cholesterol in this bilayer . In addition to sphingomyelin , phospholipid acyl chain saturation has profound effects on the accessibility of cholesterol in membranes ( Chakrabarti et al . , 2017; Gay et al . , 2015; Lange et al . , 2013; Radhakrishnan and McConnell , 2000; Sokolov and Radhakrishnan , 2010 ) . If the GRAM domain binds to the PM by sensing the accessibility of cholesterol , its binding to artificial membranes should also be influenced by the acyl chain diversity of the phospholipids . To test this possibility , we generated liposomes containing fixed amounts of phosphatidylserine ( 20% ) with varying ratios of cholesterol and phosphatidylcholine ( Figure 3—figure supplement 2B ) . We individually tested the three types of phosphatidylcholine that possesses different acyl chain structures , namely POPC , DOPC , and DPhyPC ( Figure 3—figure supplement 2C ) . Branched ( DPhyPC ) and more unsaturated ( DOPC ) acyl chains lower the tendency to form ordered conformation in the membranes , and thus , POPC has the strongest cholesterol sequestration effect of these three lipids , followed by DOPC and DPhyPC ( Sokolov and Radhakrishnan , 2010 ) . The binding of the GRAM domain of GRAMD1b to liposomes shifted to lower cholesterol concentration as the ordering tendency of phosphatidylcholine is lowered ( i . e . as the cholesterol sequestration effect is reduced ) ( Figure 3—figure supplement 2B ) . These results are consistent with the ability of the GRAM domain to sense the accessibility of cholesterol in membranes . Finally , to determine whether GRAM domains bind more broadly to other anionic lipids , we replaced phosphatidylserine with other anionic lipids , namely phosphatidic acid ( PA ) , PI ( 4 ) P , and PI ( 4 , 5 ) P2 , and asked whether they affected GRAM domain binding similarly . In this assay , we used 5% anionic lipids , including phosphatidylserine , because even 5% phosphatidylserine was sufficient to mediate the binding of the GRAM domain to liposomes that also contained 50% free cholesterol , albeit less efficiently than 20% phosphatidylserine ( Figure 3—figure supplement 1D ) . No or little binding was observed when GRAM domains were mixed with liposomes that contained 5% of these anionic lipids ( each was tested individually ) ( Figure 3—figure supplement 1D ) . However , as seen when phosphatidylserine and cholesterol were combined , the addition of free cholesterol to these anionic-lipid-containing liposomes enhanced the binding of GRAM domains to the liposomes ( Figure 3—figure supplement 1D ) . As anionic lipids , including phosphatidylserine , are enriched in the inner leaflet of the PM ( Yeung et al . , 2008 ) , these results indicate that the recruitment of GRAMD1s to the PM is regulated by interactions between GRAM domains and anionic lipids , and that these interactions are enhanced by the additional presence of accessible/unsequestered cholesterol in the PM . To examine the physiological role of sphingomyelin in GRAM domain-dependent recruitment of GRAMD1s to the PM , HeLa cells expressing either EGFP–GRAMD1b or EGFP–GRAMD1b ΔGRAM were treated with sphingomyelinase , which hydrolyzes PM sphingomyelin , and imaged under total internal reflection fluorescence ( TIRF ) microscopy . Although sphingomyelin is enriched in the outer leaflet of the PM bilayer , it also contributes to suppressing the accessibility of cholesterol in the inner leaflet of the PM , because unsequestered cholesterol can spontaneously flip flop between the outer and inner leaflets of this bilayer ( Leventis and Silvius , 2001; Steck and Lange , 2018 ) . Within 30 min of sphingomyelinase treatment , GRAMD1b was indeed recruited to the PM ( Figure 3G ) , albeit this recruitment was less efficient than cholesterol loading to the PM ( Figure 3B ) . GRAMD1b ΔGRAM , however , failed to localize to the PM , even after 60 min ( Figure 3G ) . EGFP-tagged GRAMD1 GRAM domains ( namely EGFP–GRAM1a , EGFP–GRAM1b , and EGFP–GRAM1c ) were all recruited to the PM upon sphingomyelinase treatment ( Figure 3H ) , revealing a direct role of the GRAM domain in detecting the unsequestered/accessible pool of PM cholesterol in cells . These results are also consistent with the lack of enrichment of GRAMD1s at ER–PM contact sites at rest ( Figure 1B , Figure 3A and Figure 4—figure supplement 3B ) . Taken together , these data demonstrate that GRAMD1s are recruited to the PM by sensing an increase in the accessibility of PM cholesterol ( i . e . acute expansion of the accessible pool of PM cholesterol that exceeds a certain threshold at which the GRAM domain interacts with the PM ) . Furthermore , the data show that this recruitment depends on the GRAM domain , which acts as a coincidence detector for both unsequestered/accessible cholesterol and anionic lipids in the PM ( Figure 3I ) . As GRAMD1s move to ER–PM contact sites upon acute expansion of the accessible pool of PM cholesterol ( Figure 3G , H ) , they may also contribute to the extraction of accessible PM cholesterol in order to maintain homeostasis . To investigate the potential functions of GRAMD1s in this process , we used the CRISPR/Cas9 system to disrupt GRAMD1 function by targeting all three GRAMD1 genes ( GRAMD1A , GRAMD1B and GRAMD1C ) in HeLa cells . Guide RNAs specific to exon 13 of GRAMD1A and GRAMD1B and to exon 11 of GRAMD1C were chosen , as they encode the lipid-harboring StART-like domains ( Figure 4A ) . After transfection of plasmids expressing GRAMD1-specific guide RNAs and Cas9 protein , two independent isolates of GRAMD1a/1b double knockout cell clones ( DKO #38 and DKO #40 ) and two independent isolates of GRAMD1a/1b/1c triple knockout cell clones ( TKO #1 and TKO #15 ) were selected . The absence of GRAMD1a and GRAMD1b was confirmed by western blotting and genomic sequencing ( Figure 4B and Figure 4—figure supplement 1A–D ) . Disruption of the GRAMD1C gene was validated by sequencing the targeted genomic region within the GRAMD1C locus ( Figure 4C and Figure 4—figure supplement 1E ) . No obvious defects in cell viability or overall morphology were observed for these KO cells , with the exception that KO cells grew slightly slower than parental HeLa cells . Subsequent experiments were performed using GRAMD1a/1b/1c TKO #15 cells ( hereafter referred to as GRAMD1 TKO cells ) . The incubation of cells with sphingomyelinase reduces the sequestration of PM cholesterol , resulting in a transient expansion of the accessible pool of cholesterol in the PM ( Das et al . , 2014; Endapally et al . , 2019 ) . The newly expanded pool of accessible cholesterol is then extracted and transported to the ER ( Das et al . , 2014; Lange and Steck , 1997; Scheek et al . , 1997; Slotte and Bierman , 1988 ) . Based on the ability of the GRAM domain to sense expansion of the accessible pool of PM cholesterol ( Figure 3H , I ) , EGFP–GRAM1b was used as a probe to detect acute increases in the accessible pool of PM cholesterol . Without stimulation , cytosolically expressed EGFP–GRAM1b was distributed throughout the cytoplasm without particular enrichment in the PM in both wild-type and GRAMD1 TKO HeLa cells ( Figure 4—figure supplement 2A ) . Treatment with sphingomyelinase for 1 hr led to only modest recruitment of EGFP–GRAM1b to the PM in wild-type HeLa cells ( Figure 4D ) . By contrast , the same treatment lead to much more prominent recruitment of EGFP–GRAM1b to the PM in GRAMD1 TKO cells ( Figure 4D ) . TIRF microscopy of cells expressing EGFP–GRAM1b revealed that PM recruitment of EGFP–GRAM1b upon sphingomyelinase treatment was significantly enhanced in GRAMD1 TKO cells , compared to wild-type control cells , over the entire 1 hr treatment ( Figure 4E ) . Importantly , additional treatment of GRAMD1 TKO cells with methyl-β-cyclodextrin ( MCD ) , which extracts cholesterol from cellular membranes , resulted in acute loss of the PM recruitment of EGFP–GRAM1b within 2 min ( Figure 4—figure supplement 2B ) . However , the same treatment resulted in only modest changes in the binding of the phosphatidylserine biosensor ( the mCherry-tagged C2 domain of lactadherin , mCherry-LactC2 ) or the PI ( 4 , 5 ) P2 biosensor ( the iRFP-tagged PH domain of PLCδ , iRFP-PHPLCδ ) , confirming the specificity of EGFP–GRAM1b in sensing the newly expanded pool of accessible cholesterol in the PM upon sphingomyelinase treatment ( Figure 4—figure supplement 2B ) . Taken together , these results demonstrate an exaggerated accumulation of the accessible pool of PM cholesterol in GRAMD1 TKO cells upon sphingomyelinase treatment , and suggest that the extraction and transport of this acutely expanded accessible pool may be impaired in the absence of GRAMD1s . We also assessed the role of GRAMD1s in regulating steady-state PM cholesterol levels by separating and purifying PMs from cultured cells using poly-D-lysine-coated dextran beads ( Saheki et al . , 2016 ) . Cultured cells were attached to the beads and osmotically lysed by vigorous vortexing . Brief sonication was used to remove most organelles , whereas PM sheets remained attached to the bead surface ( visualized by BODIPY-labeled ceramide ) ( Figure 4—figure supplement 3A ) . As shown by western blotting , the PM sheets that remained bound to the beads were highly enriched for PM marker proteins ( such as CD44 ) relative to the starting material . The endosomal marker , EEA1 , was greatly depleted , whereas small amounts of ER proteins ( such as VAPA and VAPB ) were recovered in the PM ( Figure 4—figure supplement 3B ) . This probably reflected the tight attachment of cortical ER ( Saheki et al . , 2016 ) . Importantly , the levels of endogenous GRAMD1a and GRAMD1b on bead-attached PM sheets were similar to those seen for the integral ER protein , VAP ( Figure 4—figure supplement 3B ) . This confirmed that the majority of these two proteins are distributed throughout the ER , with only a very small fraction localizing to ER–PM contact sites at rest ( Figure 1B and Figure 3A ) . Mass spectrometry analysis of whole-cell and purified PM lipid extracts from wild-type control and GRAMD1 TKO HeLa cells did not reveal significant changes in cholesterol and other major lipids , except for very minor increases in cholesterol esters ( Figure 4—figure supplement 3C , D ) . Thus , GRAMD1s are not essential for maintaining total levels of PM cholesterol . This result is also consistent with very little enrichment of GRAMD1s at ER–PM contacts at the steady state . Collectively , these results indicate that GRAMD1s may contribute to PM cholesterol homeostasis by counteracting acute increases in the accessible pool of PM cholesterol through its extraction and transport to the ER . Although GRAMD1 StART-like domains transport cholesterol in vitro , it remains unclear whether this property is relevant to cellular physiology . Our live-cell imaging analysis of EGFP–GRAM1b ( which is a novel biosensor for detecting acute expansion of the accessible pool of PM cholesterol that we identified in this study ) allowed us to conduct a structure–function analysis of GRAMD1s in the context of cellular functions for the first time . We first asked whether the sterol-binding pocket of the StART-like domain is required for the cellular functions of GRAMD1s . As a first step , we characterized the cholesterol-transporting properties of individual StART-like domains in vitro and generated a series of structure-guided mutations in order to identify key amino-acid residues that are essential for cholesterol transport . We purified StART-like domains from all three GRAMD1s and performed cell-free liposome-based lipid transfer assays . In this assay , the amount of dehydroergosterol ( DHE ) ( a fluorescent analog of cholesterol ) in liposomes was quantitatively measured using fluorescence resonance energy transfer ( FRET ) between DHE and Dansyl-PE ( DNS-PE ) ( Figure 5A ) . DHE was initially loaded only into donor liposomes , and its transfer from donor to DNS-PE-containing acceptor liposomes was monitored over time by measuring FRET between transferred DHE and DNS-PE in acceptor liposomes ( Figure 5A and Figure 5—figure supplement 1A ) . In the absence of StART-like domains , very few increases in the FRET signal were observed ( Figure 5—figure supplement 1B , buffer ) . However , when GRAMD1 StART-like domains were mixed with donor and acceptor liposomes , a rapid increase in FRET signal was observed , indicating the efficient extraction of DHE from donor liposomes and its loading onto acceptor liposomes by the StART-like domains ( Figure 5E and Figure 5—figure supplement 1B ) . Increasing amounts of purified proteins ( 0 . 5 µM , 1 µM , and 2 µM ) reduced the time required for the FRET signal to plateau ( Figure 5—figure supplement 1C–E ) . GRAMD1a StART-like domains transferred DHE most efficiently , at a rate corresponding to ~8 DHE molecules per minute . In comparison , GRAMD1b and GRAMD1c transported ~1 DHE molecule per minute , as calculated using a standard curve ( Figure 5—figure supplement 1A , F ) . Our results show the ability of GRAMD1 StART-like domains to transport cholesterol between membranes . Guided by the crystal structures of GRAMD1 StART-like domains in complex with 25-hydroxycholesterol ( Laraia et al . , 2019; Sandhu et al . , 2018 ) , we designed mutations that would potentially block the insertion of cholesterol into the GRAMD1b StART-like domain . Our mutagenesis strategy was to rigidify the loop that was predicted to open or close to capture or release sterol ( 5P ) ( Figure 5B ) . Purified GRAMD1a and GRAMD1b StART-like domains with 5P mutations were unable to transfer DHE in vitro ( Figure 5C , D and Figure 5—figure supplement 1G , H ) . A similar result was also obtained with a version of the GRAMD1b StART-like domain with a point mutation ( T469D ) that was previously shown to be defective in DHE extraction in vitro ( Horenkamp et al . , 2018 ) ( Figure 5—figure supplement 1H ) . Building upon our newly designed 5P mutation , which eliminated the ability of StART-like domains to transport cholesterol , we asked whether the exaggerated accumulation of the accessible pool of PM cholesterol that was observed in GRAMD1 TKO cells upon sphingomyelinase treatment ( using the EGFP–GRAM1b biosensor ) could be rescued by re-expressing wild-type or mutant versions of GRAMD1b . Strikingly , the enhanced PM recruitment of EGFP–GRAM1b was dramatically suppressed by expressing wild-type mRuby–GRAMD1b but not by expressing a mutant version of mRuby–GRAMD1b that is defective in cholesterol transport [mRuby–GRAMD1b ( 5P ) ] ( Figure 5E ) . By contrast , PM recruitment of mRuby–GRAMD1b upon sphingomyelinase treatment of TKO cells was higher for the 5P mutant GRAMD1b than for wild-type GRAMD1b ( Figure 5F ) . These results suggest that the StART-like domain-dependent extraction and transport of accessible PM cholesterol to the ER facilitates the dissociation of GRAMD1b from the PM , as the interaction of the GRAM domain of GRAMD1b with the PM is weakened , owing to a reduction in accessible cholesterol in the PM . Our results to date suggest that GRAMD1b may play a unique role in sensing and controlling the movement of accessible PM cholesterol . To further support this notion , we used GRAMD1 TKO cells to examine whether overexpression of other known cholesterol-transfer proteins , such as STARD4 ( Iaea et al . , 2017; Mesmin et al . , 2011 ) and some ORPs [including OSBP ( Antonny et al . , 2018 ) , ORP4 ( Charman et al . , 2014 ) and ORP9 ( Ngo and Ridgway , 2009 ) ] could substitute the function of GRAMD1s . Specifically , we examined whether their overexpression rescue exaggerated accumulation of the accessible pool of PM cholesterol observed in GRAMD1 TKO cells , as monitored by the EGFP–GRAM1b biosensor , upon sphingomyelinase treatment ( Figure 4D , E ) . Transiently transfected mCherry-tagged STARD4 ( mCherry–STARD4 ) and mRuby-tagged ORPs ( mRuby–OSBP , mRuby–ORP4 , and mRuby–ORP9 ) were all well expressed in TKO cells ( Figure 5—figure supplement 2A ) . However , their expression did not suppress the enhanced recruitment of EGFP–GRAM1b to the PM in TKO cells upon sphingomyelinase treatment , being unable to substitute the function of GRAMD1s ( Figure 5—figure supplement 2B , D; compare with Figure 5E ) . None of these proteins were recruited to the PM by sphingomyelinase treatment , demonstrating a unique property of GRAMD1s in sensing a transient expansion of the accessible pool of PM cholesterol ( Figure 5—figure supplement 2C , E ) . Taken together , our results suggest a critical role of the GRAMD1s in controlling the movement of the accessible pool of PM cholesterol between the PM and the ER via their StART-like domains . Acute expansion of the accessible pool of PM cholesterol results in the suppression of SREBP-2 cleavage and the inhibition of cholesterol biosynthesis as a result of transport of accessible cholesterol from the PM to the ER . However , the intracellular transport machinery by which accessible cholesterol is transported from the PM to the ER remains unknown . GRAMD1s may play a role in this process , as they are able to sense and counteract the acute expansion of the accessible pool of PM cholesterol . TIRF microscopy of cells expressing EGFP–GRAMD1b revealed that sphingomyelinase treatment led to sustained recruitment of GRAMD1b to the PM ( during 3 hr of imaging ) ( Figure 6A; Video 3 ) . As GRAMD1 TKO cells show exaggerated accumulation of the accessible pool of PM cholesterol upon sphingomyelinase treatment compared with wild-type cells ( Figure 4D , E ) , GRAMD1s may be involved in PM to ER transport of the accessible pool of cholesterol via their GRAM and StART-like domains . To examine the role of GRAMD1s in this process , we determined a time-course for the suppression of SREBP-2 cleavage upon sphingomyelinase treatment in wild-type control and GRAMD1 TKO cells as an estimate of the efficiency of the transport of accessible cholesterol from the PM to the ER . In this assay , we first depleted most of the accessible cholesterol from control and TKO cells by treating them with a combination of lipoprotein-deficient serum ( LPDS ) and mevastatin , an HMG-CoA reductase inhibitor , for 16 hr ( a treatment designed to induce maximum SREBP-2 cleavage by cholesterol starvation ) . We then stimulated the cells with sphingomyelinase and , using total cell lysates , we monitored over time the suppression of SREBP-2 cleavage , which results from the PM to ER transport of accessible cholesterol in response to the liberation of the sphingomyelin-sequestered pool of PM cholesterol by sphingomyelinase . Cell lysates were collected at different time points ( 0 , 30 , 60 , 90 , 120 , 150 , and 180 min ) and analyzed by SDS-PAGE followed by immuno-blotting against SREBP-2 ( Figure 6B ) . At time 0 , there were no detectable changes in the cleavage of SREBP-2 in GRAMD1 TKO cells compared to wild-type control cells . Suppression of SREBP-2 cleavage was observed in control cell lysates within 90 min; however , such suppression was delayed and reduced ( but not eliminated ) in TKO cells . Even after 180 min , TKO cells were not able to suppress SREBP-2 cleavage to levels similar to those observed in wild-type control cells ( Figure 6B , C ) . Importantly , re-expression of GRAMD1b in TKO cells was sufficient to suppress SREBP-2 cleavage to an extent similar to that observed in wild-type control cells at the 180 min time point , thereby rescuing the phenotype ( Figure 6D , E and Figure 6—figure supplement 1A , B ) . We hypothesized that both the recruitment of GRAMD1s to ER–PM contact sites and their ability to transport cholesterol are critical for the suppression of the cleavage of SREBP-2 by facilitating transport of the newly expanded pool of accessible PM cholesterol to the ER . To test this hypothesis , we used a GRAMD1b mutant that lacks the GRAM domain ( GRAMD1b ΔGRAM ) , which cannot be recruited to the PM ( Figure 3G ) , and a GRAMD1b with the mutated StART-like domain , which is defective in cholesterol transport ( 5P ) ( Figure 5B–D ) . The expression of GRAMD1b ΔGRAM or GRAMD1b 5P in TKO cells failed to rescue the phenotype ( Figure 6D and Figure 6—figure supplement 1A ) . These data demonstrate that GRAMD1s play a role in the transport of accessible cholesterol from the PM to the ER upon acute expansion of the accessible pool of PM cholesterol and help to suppress SREBP-2 activity . Furthermore , the data show that such functions require the recruitment of GRAMD1s to ER–PM contact sites , which is regulated by the ability of their GRAM domain to sense a transient expansion of the accessible pool of PM cholesterol , and by their StART-like-domain-dependent cholesterol transport . In order to measure changes in the accessible pool of PM cholesterol , we took advantage of the cholesterol-binding domain 4 ( D4 ) of bacterial Perfringolysin O ( PFO ) , which has been widely used as a probe to measure the accessible pool of PM cholesterol ( Das et al . , 2013; Gay et al . , 2015; Shimada et al . , 2002; Sokolov and Radhakrishnan , 2010 ) . Wild-type control and GRAMD1 TKO cells that had been pre-treated with a combination of LPDS and mevastatin for 16 hr were stimulated with sphingomyelinase for a fixed period of time ( 0 , 30 , 60 , 90 , 120 , 150 , and 180 min ) and washed . Cells were then incubated with recombinant EGFP-tagged D4 ( EGFP–D4 ) proteins for 15 min at room temperature . After washing , cell lysates were collected and analyzed by SDS-PAGE followed by immuno-blotting against GFP to detect EGFP–D4 proteins that were bound to accessible cholesterol in the PM ( Figure 6—figure supplement 2A ) . At time 0 , there were no detectable changes in EGFP–D4 signals in TKO cells compared to wild-type control cells . A 30 min treatment with sphingomyelinase induced a similar increase in the binding of EGFP–D4 to both control and TKO cells . A gradual decrease of EGFP–D4 signals was observed in control cell lysates over the time course of 180 min , similar to that reported in a previous report that utilized a mutant form of PFO to assess changes in accessible cholesterol in the PM upon sphingomyelinase treatment ( Das et al . , 2014 ) . TKO cells , however , showed continuous increase in binding of EGFP–D4 to the PM even after 180 min ( Figure 6—figure supplement 2A , B ) , suggesting a sustained accumulation of accessible cholesterol in the PM of TKO cells , due to less efficient transport of accessible cholesterol from the PM to the ER , that does not occur in wild-type control cells . Together with the results obtained with the cytosolically expressed EGFP–GRAM1b biosensor ( Figure 4D , E ) , these data strongly indicate that the extraction and transport of accessible PM cholesterol to the ER by GRAMD1s is able to counteract with acute expansion of the accessible pool of PM cholesterol ( e . g . acute expansion induced by sphingomyelinase treatment ) to prevent the accumulation of accessible cholesterol in the PM in wild-type control cells , and that this homeostatic response is impaired in GRAMD1 TKO cells . It is also important to note that there might be other intracellular cholesterol transport mechanisms that may act in parallel with GRAMD1s to facilitate accessible cholesterol extraction from the PM for its transport to the ER , as suppression of SREBP-2 cleavage is not eliminated even in the total absence of GRAMD1s ( see Discussion ) . A version of GRAMD1b in which the transmembrane domain and luminal region are both replaced by those of Sec61β ( TM swap ) cannot form protein complexes ( Figure 2F–J ) . Remarkably , GRAMD1b TM swap failed to rescue the reduced suppression of SREBP-2 cleavage observed in GRAMD1 TKO cells ( Figure 6E and Figure 6—figure supplement 1B ) and failed to suppress the enhanced recruitment of EGFP–GRAM1b to the PM in TKO cells upon sphingomyelinase treatment , although the mutant protein was still recruited to the PM ( Figure 6—figure supplement 1C , D ) . TIRF microscopy analysis of HeLa cells expressing the GRAMD1b TM swap mutant , however , revealed major differences in how this protein was recruited to the PM compared to wild-type GRAMD1b ( Figure 6F ) . GRAMD1b TM swap remained diffusely distributed on the tubular ER ( which is closely attached to the PM ) even at the end of the 180 min imaging period . By contrast , wild-type GRAMD1b progressively accumulated at ER–PM contacts as discrete patches with much stronger PM recruitment ( Figure 6F , G; Video 3 ) . These results support an important role for GRAMD1 complex formation in facilitating the progressive accumulation of GRAMD1s at ER–PM contacts , thereby supporting efficient accessible cholesterol transport at these contacts . Taken together , we conclude that GRAMD1s play a role in PM to ER transport of the accessible pool of PM cholesterol upon acute expansion of this pool . Loss of GRAMD1 function leads to sustained accumulation of accessible cholesterol in the PM , resulting in less effective suppression of SREBP-2 cleavage and possibly dysregulation of cellular cholesterol homeostasis . Distinct pools of cholesterol co-exist in the PM at steady state: a major pool is ‘inaccessible’ ( i . e . , sequestered or chemically inactive ) and a smaller pool is ‘accessible’ ( i . e . , unsequestered or chemically active ) . Given the role of GRAMD1s in facilitating the transport of accessible cholesterol from the PM to the ER , the impact of GRAMD1 deficiency on steady-state levels of accessible PM cholesterol was examined . We purified EGFP-tagged D4 mutant ( D434S ) proteins ( EGFP–D4H ) , which have a lower threshold for binding to accessible cholesterol compared to D4 in vitro ( Johnson et al . , 2012; Maekawa and Fairn , 2015 ) . Wild-type control and GRAMD1 TKO HeLa cells that express a PM marker ( iRFP-PHPLCδ ) were incubated with buffer containing purified recombinant EGFP–D4H proteins for 15 min at room temperature and washed , and then imaged under spinning disc confocal microscopy . D4H binding was assessed by line scan analysis . Strikingly , EGFP–D4H proteins bound more strongly to the PM of GRAMD1 TKO cells compared to that of control cells ( Figure 7A , B ) . Pre-treatment of GRAMD1 TKO cells with MCD for 30 min resulted in loss of the binding of EGFP–D4H to the PM ( Figure 7—figure supplement 1A , B ) , validating the specificity of this probe in sensing the accessible pool of PM cholesterol . As the total level of PM cholesterol was not elevated in GRAMD1 TKO cells in our lipidomics analysis ( Figure 4—figure supplement 3C , D ) , these results indicate that the chronic expansion of the accessible pool of PM cholesterol occurs in the absence of GRAMD1s . Re-expression of any of the three GRAMD1s in TKO cells was sufficient to reduce the binding of EGFP–D4H to the PM , thereby rescuing the chronic expansion of the D4H-accessible pool of PM cholesterol observed in TKO cells ( Figure 7—figure supplement 2A–C ) . Versions of GRAMD1b in which the StART-like domain was mutated were systematically expressed in TKO cells to determine whether the ability of GRAMD1b to transport accessible cholesterol is required to rescue the phenotype ( Figure 7—figure supplement 2D ) . All mutant versions of GRAMD1b , including a newly designed mutant in which the hydrophobicity of the surface of the sterol-binding pocket is changed ( Y430A , V445A ) , as well as 5P and T469D mutants , failed to reduce the binding of EGFP–D4H to the PM of TKO cells because they are unable to rescue the chronic expansion of the D4H-accessible pool of PM cholesterol in TKO cells ( Figure 7—figure supplement 2E–I ) . Taken together , these results suggest the importance of GRAMD1s in maintaining steady-state levels of accessible PM cholesterol by facilitating its transport from the PM to the ER . Chronic expansion of the accessible pool of PM cholesterol in GRAMD1 TKO cells at steady state , revealed by increased PM binding of the EGFP–D4H probe , indicates that GRAMD1s are important for maintaining PM cholesterol homeostasis through their functions in sensing a transient expansion of the accessible pool of PM cholesterol and by facilitating the transport of accessible PM cholesterol to the ER at ER–PM contact sites . If this is the case , artificial forced recruitment of re-expressed GRAMD1s to ER–PM contacts in GRAMD1 TKO cells should mediate the extraction and transport of accessible cholesterol from the PM to the ER and reduce the binding of the EGFP–D4H probe to the PM . To test whether GRAMD1s can directly act at ER–PM contact sites , rapamycin-induced dimerization of the FK506-binding protein ( FKBP ) and the FKBP-rapamycin-binding domain ( FRB ) ( Muthuswamy et al . , 1999 ) was used to recruit GRAMD1b to these sites acutely . In this assay , GRAMD1 TKO cells were co-transfected with a version of GRAMD1b in which the N-terminus , which contains the GRAM domain , was replaced by a miRFP-tagged FKBP module ( miRFP-FKBP–GRAMD1b ) and a PM-targeted FRB module ( PM-FRB–mCherry ) ( Figure 7C ) . TIRF microscopy revealed rapid recruitment of miRFP-FKBP–GRAMD1b to the PM within 10 min of rapamycin treatment ( Figure 7D and Figure 7—figure supplement 3A; Video 4 ) . To assess accessible pool of PM cholesterol after the acute recruitment of the chimeric GRAMD1b protein to the PM , cells that had been pre-treated with rapamycin for a fixed period of time ( 0 min , 30 min , and 60 min ) were incubated with recombinant EGFP–D4H proteins for 15 min at room temperature , washed and then imaged under spinning disc confocal microscopy . Strikingly , rapamycin-induced PM recruitment of the chimeric GRAMD1b protein led to acute reduction in D4H-accessible PM cholesterol in GRAMD1 TKO cells within 60 min , reducing the binding of EGFP–D4H proteins to the PM ( Figure 7E , F ) . Mutant versions of miRFP-FKBP–GRAMD1b carrying a StART-like domain that cannot transport cholesterol ( 5P and T469D mutants; see also Figure 5B–D , Figure 7—figure supplement 2D–I and Figure 5—figure supplement 1H ) were recruited to the PM with kinetics similar those of the wild-type version ( WT ) ( Figure 7D and Figure 7—figure supplement 3B , C; Video 5 ) . However , recruitment of the mutant versions did not reduce the PM EGFP–D4H signal ( Figure 7F ) , demonstrating a critical role for StART-like domain-dependent PM to ER cholesterol transport in the removal of the expanded pool of accessible PM cholesterol by GRAMD1b at ER–PM contact sites . On the basis of these results , we conclude that GRAMD1s play a direct role in facilitating the transport of accessible PM cholesterol to the ER at ER–PM contact sites . We have demonstrated that the evolutionarily conserved family of ER-anchored GRAMD1s contribute to PM cholesterol homeostasis by sensing a transient expansion of the accessible pool of PM cholesterol and facilitating its transport to the ER at ER–PM contact sites . We have also identified the molecular mechanisms by which GRAMD1s interact with one another to form a complex , and how they are recruited to the PM . Key findings of the current study are the following: ( 1 ) We found that GRAMD1s and GRAMD3 form homo- and heteromeric complexes and localize to discrete patches on tubular ER in mammalian cells at rest . We identified that their transmembrane domains and luminal helices , which are predicted to form amphipathic surfaces , mediated the formation of protein–protein complexes and regulated their progressive recruitment to ER–PM contacts and their functions . ( 2 ) Using in vitro liposome sedimentation assays and live-cell imaging , we found that the GRAM domain of GRAMD1s acts as a coincidence detector , tuned to the presence of both ‘unsequestered/accessible’ cholesterol and anionic lipids , including phosphatidylserine , in the PM ( Figure 3I ) . Importantly , the binding of the GRAM domain to membranes requires that unsequestered/accessible cholesterol exceeds a certain threshold . As the majority of cholesterol in the PM is sequestered ( i . e . , inaccessible ) , this switch-like property allows GRAMD1s to move to ER–PM contact sites only when the accessible pool of PM cholesterol transiently expands , preventing GRAMD1s from accumulating at ER–PM contacts at rest . ( 3 ) We have deciphered the novel cellular mechanisms by which the accessibility of PM cholesterol is monitored by an LTP at ER–PM contacts . We found that GRAMD1s sense a transient expansion of the accessible pool of PM cholesterol through their GRAM domain and facilitate its transport through their StART-like domain . Disruption of their functions leads to less efficient transport of accessible PM cholesterol to the ER and reduced suppression of SREBP-2 cleavage . Importantly , we showed that the formation of GRAMD1 protein complexes , as well as the StART-like and GRAM domains of GRAMD1 proteins , is critical for the cellular functions of GRAMD1s in vivo . ( 4 ) Our results demonstrate that removal of accessible PM cholesterol occurs within ~1 hr , when re-expressed GRAMD1 proteins are artificially recruited to ER–PM contact sites to facilitate transport of accessible cholesterol from the PM to the ER in GRAMD1 TKO cells . These experiments utilized a drug-induced dimerization approach . In addition , this GRAMD1 function requires the sterol-transporting StART-like domain . Previous studies of yeast mutants lacking Lam/Ltc proteins relied on genetic approaches ( over a time scale of ~20 hr ) , making it difficult to interpret the significance of their lipid-transfer functions in vivo . Using the CRISPR/Cas9 gene-editing system , we demonstrated that GRAMD1s are not essential for cell viability . Although cells that lack all three GRAMD1s grow more slowly than wild-type cells , they do not exhibit major abnormalities . Overlapping functions between mammalian GRAMD1s and other sterol-binding STARD proteins may explain the lack of major defects in these cells . Accordingly , PM lipidomics did not reveal major differences in PM cholesterol levels between GRAMD1 TKO cells and wild-type cells . Our analyses suggest , however , that GRAMD1s facilitate the transport of accessible cholesterol from the PM to the ER in response to a transient expansion of the accessible pool of PM cholesterol , thereby contributing to PM cholesterol homeostasis . First , GRAMD1 TKO cells exhibited an exaggerated accumulation of the accessible pool of PM cholesterol in response to acute hydrolysis of sphingomyelin , which liberates the sphingomyelin-sequestered pool of cholesterol into the ‘accessible pool’ . This was revealed by sustained binding of the PFO D4 probe to the PM and by enhanced PM recruitment of the GRAM domain of GRAMD1b , a novel biosensor for detecting acute expansion of the accessible pool of PM cholesterol that we identified in this study , in GRAMD1 TKO cells . Second , GRAMD1 TKO cells showed reduced suppression of SREBP-2 cleavage upon hydrolysis of sphingomyelin , reflecting less efficient PM to ER transport of the newly expanded pool of accessible cholesterol in these cells . Third , GRAMD1 TKO cells exhibited a chronic expansion of the accessible pool of PM cholesterol , as detected by the D4H probe . All of these phenotypes are consistent with defects in efficient PM to ER transport of the accessible pool of cholesterol . Importantly , re-expression of GRAMD1b rescued these phenotypes , with rescue depending on the sterol-binding property of GRAMD1b's StART-like domain , the GRAM domain , and protein complex formation . Although the transition of cholesterol between ‘inaccessible’ and ‘accessible’ pools in the PM plays crucial roles in controlling cellular cholesterol homeostasis ( Das et al . , 2014; Endapally et al . , 2019 ) , the molecular mechanisms by which these transitions are monitored , and the intracellular transport machinery responsible for the PM to ER transport of the accessible pool of cholesterol , have both remained elusive . Our results suggest that these two mechanisms can be coupled by non-vesicular cholesterol transport mediated by an LTP at ER–PM contact sites . We found that GRAMD1s sense a transient expansion of the accessible pool of PM cholesterol and facilitate its transport to the ER at ER–PM contact sites ( Figure 8A–C ) . We found that interactions between the purified GRAM domains of GRAMD1a/b and artificial membranes that contain phosphatidylserine , which is a major acidic phospholipid in the PM , are dramatically enhanced by the presence of unsequestered/accessible cholesterol that exceeds a certain threshold . Furthermore , such interactions are modulated by sphingomyelin and phospholipids , which sequester cholesterol by forming dynamic complexes . Thus , the GRAM domain allows GRAMD1s to sense an increase in accessible PM cholesterol and facilitates the accumulation of GRAMD1s at ER–PM contacts only when the accessible pool of cholesterol transiently expands in this bilayer ( Figure 8B ) . Importantly , such regulation prevents GRAMD1s from depleting PM cholesterol at steady state . Loss of GRAMD1 functions , however , leads to chronic and acute expansion of the accessible pool of PM cholesterol ( Figure 8B , C ) . It has been known that accessible cholesterol , upon reaching a certain threshold , is extracted for transport to the ER and regulates cholesterol biosynthesis to maintain homeostasis . How the intracellular transport machinery senses such a sharp threshold has been unknown . The ability of the GRAMD1s to accumulate at ER–PM contacts in a switch-like fashion , by sensing a transient expansion of the accessible pool of PM cholesterol through their GRAM domain , provides a conceptual framework for this process . Importantly , GRAMD1s themselves directly facilitate the transport of the expanded pool of accessible PM cholesterol from the PM to the ER , thereby contributing to PM cholesterol homeostasis as a critical homeostatic regulator . Levels of PM cholesterol are maintained at an equilibrium by balancing the efflux and influx of cholesterol out of and into the PM , respectively . Thus , inhibition of the efflux pathway only ( e . g . , by loss of GRAMD1s ) disrupts this equilibrium , potentially increasing total levels of PM cholesterol . However , levels of PM cholesterol are unchanged in GRAMD1 TKO cells ( Figure 4—figure supplement 3C , D ) . Thus , alternative backup systems may support the efflux of PM cholesterol ( or reduce the influx of cholesterol to the PM ) to maintain total cholesterol levels in the PM in the absence of GRAMD1s . For example , recent studies found that macrophages are able to dispose of accessible cholesterol by releasing PM-derived particles that are rich in cholesterol ( He et al . , 2018; Hu et al . , 2019 ) . It is also important to note that suppression of SREBP-2 cleavage upon sphingomyelinase treatment ( an estimate of the efficiency of PM to ER transport of accessible cholesterol ) is reduced but not eliminated in the absence of GRAMD1s ( Figure 6B , C ) . Thus , other intracellular cholesterol transport machineries might also participate in the transport of accessible cholesterol from the PM to the ER . Such robust parallel mechanisms may be partially mediated by other non-vesicular sterol transport systems , such as those mediated by STARDs or ORPs; vesicular transport may also be involved in maintaining total levels of PM cholesterol . Further studies are needed to better understand the interplay between GRAMD1s and other sterol efflux/transfer systems in the regulation of cellular cholesterol homeostasis . GRAMD1s interact with each other through their transmembrane domains and predicted luminal amphipathic helices ( e . g . , GRAMD1b homomeric and GRAMD1b/GRAMD1a heteromeric interactions ) . Mutant GRAMD1b proteins that lack these regions fail to interact with one another and are diffusely distributed throughout the tubular ER . Other lipid transfer proteins that are known to localize to membrane contact sites also form complexes or oligomers . These proteins include E-Syts ( Giordano et al . , 2013; Saheki et al . , 2016 ) , ORP2 ( Wang et al . , 2019 ) , and ORP5/8 ( Chung et al . , 2015 ) , as well as the yeast ERMES complex ( AhYoung et al . , 2015 ) . ORP2 oligomerization and ERMES assembly enhance the ability of these proteins to transfer lipids ( Kawano et al . , 2018; Wang et al . , 2019 ) . Accordingly , mutant GRAMD1b proteins that lack the ability to form protein complexes accumulate less robustly at ER–PM contacts , and less effectively rescue phenotypes associated with GRAMD1 TKO cells . Thus , the formation of GRAMD1 complexes plays important roles in regulating GRAMD1 localization and function . A recent study by Sandhu et al . ( 2018 ) reported that GRAMD1s facilitate the transport of PM cholesterol that is additionally loaded from external sources . In particular , they showed that GRAMD1b mediates PM to ER transport of HDL-derived cholesterol that is taken up by adrenal glands via the scavenger receptor SR-B1 in mice . On the basis of these results , they proposed that GRAMD1s might also play more general roles in intracellular cholesterol transport by sensing accessible cholesterol in the PM . Our findings support this notion and further demonstrate that GRAMD1s contribute to PM cholesterol homeostasis and cholesterol metabolism by sensing a transient expansion of the accessible pool of PM cholesterol by their GRAM domain , and facilitate the transport of this cholesterol to the ER by their StART-like domain . Interestingly , GRAMD1a is broadly expressed in many tissues with particular enrichment in the brain . Future studies will be needed to determine the physiological functions of GRAMD1s in other tissues in mammals . In summary , our study has demonstrated that ER-localized GRAMD1s help to maintain PM cholesterol homeostasis in trans via their ability to transport accessible cholesterol from the PM to the ER at ER–PM contact sites . Our results show that GRAMD1s sense a transient expansion of the pool of accessible PM cholesterol and facilitate its transport to the ER through non-vesicular transport . These proteins probably perform additional functions , as yeast mutants that lack Lam/Ltc proteins show pleiotropic defects in cell signaling , including altered mTOR kinase signaling ( Murley et al . , 2017 ) . Furthermore , the StART-like domain of GRAMD1b has been shown to transport PI ( 4 , 5 ) P2 in addition to cholesterol ( Horenkamp et al . , 2018 ) . Thus , potential changes in the properties of other organelles and the PM in GRAMD1 TKO cells deserve future investigations . Further elucidating the physiological functions of these proteins , as well as other sterol-transfer proteins , at membrane contact sites will be important to gain insight into how cellular cholesterol homeostasis is regulated . Primary and secondary antibodies , chemicals , lipids and other reagents used in this study are listed in Supplementary file 1 . DNA plasmids used in this study are listed in Supplementary file 1; the sequences of oligos and primers used are listed in Supplementary file 2 . HeLa and COS-7 cells were cultured in Dulbecco’s modified Eagle’s medium ( DMEM ) containing 10% or 20% fetal bovine serum ( FBS ) and 1% penicillin/streptomycin at 37°C and 5% CO2 . Transfection of plasmids was carried out with Lipofectamine 2000 ( Thermo Fisher Scientific ) . Both wild-type and genome-edited HeLa cell lines were routinely verified as free of mycoplasma contamination at least every two months , using MycoGuard Mycoplasma PCR Detection Kit ( Genecopoeia ) . No cell lines used in this study were found in the database of commonly misidentified cell lines that is maintained by ICLAC and NCBI Biosample . For imaging experiments , cells were plated onto 35 mm glass bottom dishes at low density ( MatTek Corporation ) . All live-cell imaging was carried out one day after transfection . Spinning disc confocal ( SDC ) microscopy ( Figures 3A , 4D , 7A , E , Figure 4—figure supplement 2A , Figure 4—figure supplement 3A , Figure 5—figure supplement 2A , Figure 7—figure supplement 1A , Figure 7—figure supplement 2A , E , G ) and super-resolution SDC-structured illumination microscopy ( SDC-SIM ) ( Figures 1B , C , 2C , H–I and Figure 1—figure supplement 1A , B ) were performed on a setup built around a Nikon Ti2 inverted microscope equipped with a Yokogawa CSU-W1 confocal spinning head , a Plan-Apo objective ( 100 × 1 . 45 NA ) , a back-illuminated sCMOS camera ( Prime 95B; Photometrics ) , and a super-resolution module ( Live-SR; Gataca Systems ) that was based on structured illumination with optical reassignment and image processing ( Roth and Heintzmann , 2016 ) . The method , known as multifocal structured illumination microscopy ( York et al . , 2012 ) , makes it possible to double the resolution and the optical sectioning capability of confocal microscopy simultaneously . The maximum resolution is 128 nm with a pixel size in super-resolution mode of 64 nm . Excitation light was provided by 488 nm/150 mW ( Coherent ) ( for GFP ) , 561 nm/100 mW ( Coherent ) ( for mCherry/mRFP/mRuby ) and 642 nm/110 mW ( Vortran ) ( for iRFP/miRFP ) ( power measured at optical fiber end ) DPSS laser combiner ( iLAS system; Gataca systems ) . All image acquisition and processing was controlled by MetaMorph ( Molecular Device ) software . Images were acquired with exposure times in the 400–500 msec range . Total internal reflection fluorescence ( TIRF ) microscopy ( Figures 3B , G , H , 4E , 5E , F , 6A , F , G , 7D and Figure 3—figure supplement 1A , Figure 4—figure supplement 2B , Figure 5—figure supplement 2B-E , Figure 6—figure supplement 1C , D , Figure 7—figure supplement 3A-C ) was performed on a setup built around a Nikon Ti2 inverted microscope equipped with a HP Apo-TIRF objective ( 100 × 1 . 49 NA ) , and a back-illuminated sCMOS camera ( Prime 95B; Photometrics ) . Excitation light was provided by 445 nm/25 mW ( for CFP ) , 488 nm/70 mW ( for GFP ) , 561 nm/70 mW ( for mCherry/mRFP/mRuby ) and 647 nm/125 mW ( for iRFP/miRFP ) ( power measured at optical fiber end ) DPSS laser combiner ( Nikon LU-NV laser unit ) , coupled to the motorized TIRF illuminator through an optical fiber cable . Critical angle was maintained at different wavelengths throughout the experiment from the motorized TIRF illuminator . Acquisition was controlled by Nikon NIS-Element software . For time-lapse imaging , images were sampled at 0 . 05 Hz with exposure times in the 200–500 msec range . Cells were washed twice and incubated with Ca2+ containing buffer ( 140 mM NaCl , 5 mM KCl , 1 mM MgCl2 , 10 mM HEPES , 10 mM glucose , and 2 mM CaCl2 [pH 7 . 4] ) before imaging with either an SDC microscope or a TIRF microscope . All types of microscopy were carried out at 37°C except for the experiments with recombinant EGFP–D4H proteins , which were performed at room temperature via SDC microscopy . The GRAMD1B , GRAMD1A and GRAMD1C genes were sequentially targeted to generate GRAMD1 triple knockout cells . The sequences of oligos and primers used are listed in Supplementary file 2 . For the generation of HeLa cells lacking GRAMD1b , control wild-type HeLa cells were transfected with a plasmid encoding spCas9 and the GRAMD1b-targeting guide RNA ( Figure 4A ) , followed by isolation of individual clones by dilution cloning . Two clones ( #10 and #17 ) were further characterized by sequencing and immunoblotting ( i . e . western blotting ) . These analyses revealed deletions and insertions within the guide RNA-binding sites , frame-shift and early termination in the open-reading frame of GRAMD1B gene , and the loss of GRAMD1b protein expression ( Figure 4—figure supplement 1A , C ) . To generate GRAMD1a/1b double knockout ( DKO ) cell lines , a subclone of the GRAMD1b KO cell line #10 was transfected with a plasmid encoding spCas9 and the GRAMD1a-targeting guide RNA with ssDNA oligos containing stop codons and homology-arms ( Figure 4A ) . These cells were subjected to single cell sorting , and individually isolated clones [lines #38 and #40 ( hereafter GRAMD1a/1b DKO #38 and GRAMD1a/1b DKO #40 ) ] showed insertion of ssDNA within the guide RNA-targeted locus , resulting in the lack of GRAMD1a protein expression ( Figure 4B and Figure 4—figure supplement 1B , D ) . To generate GRAMD1 triple knockout ( TKO ) cell lines , the GRAMD1a/1b DKO #40 cell line was transfected with two plasmids encoding spCas9 and each one of the two GRAMD1b-targeting guide RNAs ( Figure 4A ) . Two clones ( GRAMD1a/1b/1c TKO #1 and GRAMD1a/1b/1c TKO #15 ) that showed large deletions in the exon 11 of GRAMD1C , as assessed by genomic PCR ( Figure 4—figure supplement 1E ) , were isolated and knock-outs were confirmed by direct sequencing ( Figure 4C ) . In all figures and texts , TKO denotes GRAMD1a/1b/1c TKO #15 unless stated . The modeled structure of GRAMD1bStART was obtained by submitting the primary sequence ( residues 375–545 ) to the I-TASSER server , using the GRAMD1aStART structure ( PDB: 6GQF ) as the template . Primary sequences of luminal helices of GRAMD1s ( GRAMD1a , 657–706; GRAMD1b , 672–720; GRAMD1c , 599–647 ) were submitted to the I-TASSER server without assigning any templates . Only the luminal amphipathic helix region , indicated in Figure 2A , is shown in Figure 2—figure supplement 1B–C . No statistical method was used to predetermine sample size , and the experiments were not randomized for live-cell imaging . Sample size and information about replicates are described in the figure legends . The number of biological replicates for all cell-based experiments and the number of technical replicates for all other biochemical assays are shown as the number of independent experiments within the figure legends for each figure . Comparisons of data were carried out by the two-tailed unpaired Student’s t-test , Holm-Sidak’s t-test or one-way ANOVA , followed by Tukey or Dunnett corrections for multiple comparisons as appropriate with Prism 7 or 8 ( GraphPad software ) . Unless p<0 . 0001 , exact P values are shown within the figure legends for each figure . p>0 . 05 was considered not significant .	The human body contains trillions of cells . At the outer edge of each cell is the plasma membrane , which protects the cell from the external environment . This membrane is mostly made of fatty molecules known as lipids and about half of these lipids are specifically cholesterol . Human cells can either take up cholesterol that were obtained via the diet or produce it within a compartment of the cell called the endoplasmic reticulum . Cells need to monitor the cholesterol levels in both the endoplasmic reticulum and the plasma membrane in order to regulate the uptake or production of this lipid . For example , if there is too much of cholesterol in the plasma membrane , then the cell transports some to the endoplasmic reticulum to tell it to shut down cholesterol production . However , how these different areas of the cell communicate with each other , and transport cholesterol , has remained unclear . Naito et al . set out to look for key regulators of cholesterol transport and identified a group of endoplasmic reticulum proteins called GRAMD1 proteins . Cholesterol in the plasma membrane is either accessible or inaccessible , meaning it either can or cannot be moved back into the cell . The GRAMD1 proteins sense accessible cholesterol , and experiments with human cells grown in the laboratory showed that , specifically , the GRAMD1 proteins work together in a complex to sense accessible cholesterol at or near the plasma membrane . One particular part of the protein senses when the amount of accessible cholesterol reaches a certain level at the plasma membrane; when this threshold is reached , the complex flips a switch to start the transport of cholesterol to the endoplasmic reticulum and tell it to shut down cholesterol production . This coupling of sensing and transporting lipids by one protein complex also helps maintain the right ratio of accessible and inaccessible cholesterol in the plasma membrane to prevent cells from activating unwanted cell-signaling events . Getting rid of the GRAMD1 proteins in cells , or removing sensing part of these proteins , leads to inefficient transport of cholesterol . A better understanding of how GRAMD1 proteins sense the accessibility of cholesterol could potentially help identify new approaches to control cholesterol transport inside cells . This may in turn eventually lead to new treatments that counteract the defects in cholesterol metabolism seen in some forms of neurodegenerative diseases such as Alzheimer’s disease and Parkinson’s disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2019	Movement of accessible plasma membrane cholesterol by the GRAMD1 lipid transfer protein complex
Long-term adult stem cells sustain tissue regeneration throughout the lifetime of an organism . They were hypothesized to originate from embryonic progenitor cells that acquire long-term self-renewal ability and multipotency at the end of organogenesis . The process through which this is achieved often remains unclear . Here , we discovered that long-term hair follicle stem cells arise from embryonic progenitor cells occupying a niche location that is defined by attenuated Wnt/β-catenin signaling . Hair follicle initiation is marked by placode formation , which depends on the activation of Wnt/β-catenin signaling . Soon afterwards , a region with attenuated Wnt/β-catenin signaling emerges in the upper follicle . Embryonic progenitor cells residing in this region gain expression of adult stem cell markers and become definitive long-term hair follicle stem cells at the end of organogenesis . Attenuation of Wnt/β-catenin signaling is a prerequisite for hair follicle stem cell specification because it suppresses Sox9 , which is required for stem cell formation . Long-term adult stem cells ( SCs ) are defined by their ability to continuously generate all downstream differentiated cell lineages as well as regenerate themselves throughout the lifetime of an organism ( Fuchs and Chen , 2013 ) . Although rudimentary tissue progenitor cells in embryos give rise to all cells in adult tissue , these cells are not SCs , because they are destined to change identity as development proceeds . It has been postulated that definitive tissue SCs originate from tissue embryonic progenitor cells that acquire the capacity for long-term self-renewal and multipotency at the end of organogenesis ( Slack , 2008 ) . However , the underlying cellular and molecular mechanisms of these processes remain unknown . Deciphering the process leading to localized SC emergence during normal embryonic development will likely reveal principles that can be used to acquire and maintain the long-term self-renewal and regenerative potentials that are prerequisites for SC-based therapies . Adult SCs in hair follicles ( HFs ) are located at the so-called ‘outer bulge region’ . The bulge region constitutes the bottom of the permanent portion of a HF . It contains two layers of epithelial cells: the CD34- inner layer niche cells and the CD34+ outer layer stem cells . Although HFs begin to develop in the embryo , the bulge structure is only formed when HFs enter the first postnatal telogen ( resting phase ) ; this delineates the end of organogenesis and the emergence of the adult SC niche ( Paus et al . , 1999; Morris et al . , 2004; Tumbar et al . , 2004; Blanpain et al . , 2004; Zhang et al . , 2009; Hsu et al . , 2011; Chen et al . , 2012 ) . Initiation of HF organogenesis is marked by placode emergence . Activation of Wnt/β-catenin signaling in basal epithelial cells by locally expressed Wnt ligands is both necessary and sufficient to induce placode formation ( van Genderen et al . , 1994; Gat et al . , 1998; Huelsken et al . , 2001; Andl et al . , 2002; Jamora et al . , 2003; Millar et al . , 1999; Reddy et al . , 2001; Fu and Hsu , 2013 ) . Concomitant with Wnt ligand production , placode epithelial cells also express the Wnt inhibitor Dkk ( Bazzi et al . , 2007 ) . The combination of the short-range activator Wnt and the long-range inhibitor DKK function in a reaction-diffusion mechanism to suppress Wnt/β-catenin signaling in cells surrounding existing placodes , thereby regulating HF spacing ( Sick et al . , 2006 ) . Contrary to the required activation of Wnt/β-catenin signaling , inhibition of BMP signaling through the dermal BMP inhibitor Noggin is necessary for placode morphogenesis ( Jamora et al . , 2003; Botchkarev et al . , 1999; Kobielak et al . , 2003 ) . After placode formation , signaling events downstream of Wnts/BMPs drive HF to further develop into the hair germ and then the hair peg . Shh is expressed by placode epithelial cells ( Oro et al . , 1997; St-Jacques et al . , 1998; Morgan et al . , 1998 ) . In the absence of Shh , dermal condensate fails to aggregate and HF development arrests at the placode stage instead of developing into the hair germ . This implicates Shh in establishing proper epithelial-mesenchymal crosstalk ( St-Jacques et al . , 1998; Oro and Higgins , 2003; Levy et al . , 2007 ) . Guard hairs , as the first HFs to appear in the embryonic backskin , are uniquely dependent on Eda/EdaR signaling ( Laurikkala et al . , 2002; Mustonen et al . , 2004 ) . Eda , which is itself a Wnt signaling target ( Laurikkala et al . , 2001 ) , can induce the expression of BMP inhibitors and Shh in guard hair ( Pummila et al . , 2007 ) . Placode progenitor cells generate all cells in adult HFs ( Levy et al . , 2007 ) . Some of their progeny cease further development at a particular point and become definitive adult HFSCs . Previous studies using label-retaining methods demonstrated that putative HFSCs are present in postnatal developing HFs as slow cycling cells , and it was shown that their specification requires the transcription factor Sox9 ( Nowak et al . , 2008 ) . Intriguingly , placode cells already express adult HFSC markers such as Lhx2 and Sox9 , although in a largely non-overlapping pattern . Another HFSC marker , Nfatc1 , appears in the subsequent hair peg ( Rhee et al . , 2006; Horsley et al . , 2008; Vidal et al . , 2005 ) . These dynamic expression patterns suggest that cells in the placode and hair peg are heterogeneous . However , whether or not HFSC fate is already pre-determined at these early developmental stages is not clear . Other critical unanswered questions include the following: Are adult HFSCs remnant of embryonic progenitor cells that maintain their embryonic developmental potential , or do they , alternatively , come from progenitor cells that gain long-term potential ? What are the underlying mechanisms ? What determines the niche location and where do HFSCs become localized ? The current study addresses these key questions . To uncover the cellular origin of HFSCs and to identify the time point of their specification , it will first be necessary to perform lineage-tracing experiments . These can be done by labelling distinct cell populations at the rudimentary stages and later determining whether SCs come from these initially labelled progenitor cells ( Figure 1A ) . We chose tail skin HFs for this study . Unlike un-patterned back skin HFs ( Figure 1—figure supplement 1A ) , tail skin HFs are arranged in triplets , and the growth of two secondary outer follicles is typically initiated next to a primary central follicle after it has already developed ( Figure 1—figure supplement 1B ) . By inducing Cre activation at specific time points and focusing on HFs in a chosen area , we can label progenitor cells in defined developmental stages and continue to follow their fates in individual HFs until the end of organogenesis , when the bulge forms ( Figure 1B , C; Figure 1—figure supplement 2A-C ) . 10 . 7554/eLife . 10567 . 003Figure 1 . Embryonic cellular origin of adult hair follicle stem cells . ( A ) Diagram of hair follicle morphogenesis and the lineage-tracing experiment . All lineage-tracing experiments ended at the first telogen , but started at different stages including the placode , hair germ , and hair peg stages . ( B , C ) Representative images of tail skin hair follicle organogenesis . Red boxes indicate the regions used for quantification in the lineage-tracing experiments . The hair cycle in tail skin progresses along the anterior to posterior and in the dorsal to ventral directions . At postnatal day 1 ( P1 ) , in the chosen region , the primary central hair follicles are in the hair peg stage while the secondary outer follicles are in the placode stage . At P15 , in the chosen region , primary central hair follicles are in the telogen phase . Scale bar: 1500 μm for the whole mount image; 100 μm for the enlarged images . ( D ) Summary of the lineage-tracing experiments with: Shh-CreER::Rosa-stop-mTmG , Nfatc1-CreER::Rosa-stop-mTmG , and Lgr5-GFP-CreER::Rosa-stop-tdTomato::K14H2BGFP mice . Percentages of labelled HFSCs at the telogen phase were quantified using whole mount sample images . Representative images are single confocal Z slices from the data used for quantification . N=5 mice , >200 HFs . Raw data are plotted to the left of each box-and-whisker plot: the median and the 25th and the 75th percentiles are denoted by notches and the bottom and top boxes , respectively; the 5th and 95th percentiles are denoted as whiskers . Scale bar: 50 μm . ( E ) In both the hair germ and the hair peg , the progenies of Shh+ cells in the center of HFs express the differentiation marker Keratine 6 ( Krt6 ) . Scale bar: 50 μm . ( F ) Lineage-tracing experiment starting at the hair peg stage and ending at morphogenesis anagen ( growth phase ) using Nfatc1-CreER::Rosa-stop-mTmG mice . Note that the labeled cells stay dormant at their original position without contributing to HF down growth . Scale bar: 50 μm . ( G ) Model depicting the spatial and temporal pattern of hair follicle stem cell emergence . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 00310 . 7554/eLife . 10567 . 004Figure 1—figure supplement 1 . Difference between back and tail skin hair follicle organogenesis . ( A ) In mouse back skin , hair follicle initiation occurs via three separate waves that start , respectively , around E13 . 5 , 15 . 5 , and 18 . 5 . At any given embryonic time point between E13 . 5 to E18 . 5 , there is a mixture of hair follicles at different developmental stages in back skin , lacking any discernable pattern . After birth , the three separate waves synchronize into similar anagen and then telogen . Thus , it is impossible to pinpoint the exact early developmental stage of any telogen back skin hair follicle at the beginning of the lineage-tracing experiments when Tamoxifen was injected at an embryonic time point . ( B ) Tail skin hair follicles are arranged in triplets , and the growth of two secondary outer follicles is typically initiated next to a primary central follicle after it has already begun development . Within the dorsal middle one-third section of tail skin , at P1 , the primary central hair follicles are at the hair peg stage and the secondary outer follicles on the side are at the placode stage . At P6 , all three hair follicles in a triplet progress to anagen . At P15 , the primary central hair follicles have already entered telogen or even the telogen to anagen transition stage , while the secondary outer hair follicles on the side are still in the catagen to telogen transition . At P21 , the primary central hair follicles are in full anagen and the two secondary outer follicles are at the telogen to anagen transition stage . Scale bars: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 00410 . 7554/eLife . 10567 . 005Figure 1—figure supplement 2 . Diagram , quantification , and specificity of CreER lines for the lineage-tracing experiments . ( A ) Diagram of the lineage-tracing experiment using Nfatc1-CreER::Rosa-stop-mTmG mice . ( B ) Diagram of the lineage-tracing experiment using Lgr5-GFP-CreER::Rosa-stop-tdTomato::K14H2BGFP mice . Representative images used here are the same representative images shown in Figure 1D to indicate these diagram are drawn to explain the experimental designs for Figure 1D . Scale bar: 50 µm . ( C ) Diagram depicting the individual time points used for the lineage-tracing experiments starting from different early developmental stages . Green color highlights the target hair follicles used in the designated lineage-tracing experiments . ( D–F ) Percentage of HFs with labeled HFSCs at telogen . Tracings starting from the placode stage ( D ) , the hair germ stage ( E ) , and the hair peg stage ( F ) . ( G ) Nfatc1-CreER::Rosa-stop-mTmG , Shh-CreER::Rosa-stop-mTmG , Lgr5-GFP-CreER::Rosa-stop-tdTomato::K14-H2BGFP mice tail skin hair follicles with or without Tamoxifen induction . Scale bar: 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 005 To test whether distinct HFSC precursors are already present at the initial developmental stages , we used several CreER knock-in lines that can label different cell populations of the placode , hair germ , and hair peg . Shh starts to be expressed in the placode ( St-Jacques et al . , 1998 ) . When we used inducible Shh-CreER::Rosa-stop-mTmG ( Harfe et al . , 2004 ) mice to label Shh-expressing cells in either the placode or the hair germ , we found that in both cases Shh+ progenitors cells can become HFSCs ( Figure 1D ) . Unexpectedly , Shh-expressing hair peg cells lost the ability to form HFSCs and mostly contributed to differentiated lineages ( Figure 1D , E ) . Since Shh-expressing placode cells generate the majority of cells in hair peg , including those still expressing Shh , this intriguing change in cell fate indicates that progenies of Shh-expressing placode cells separate into two different populations: the population still expressing Shh loose the potential to form HFSCs , while the cells with down regulated Shh expression in hair peg can presumably still become HFSCs . Consistent with this hypothesis , two adult HFSC markers , Lgr5 and Nfatc1 , first begin to be expressed in hair peg , with complementary patterns to that of Shh ( Figure 1D ) . Using Nfatc1-CreER::Rosa-stop-mTmG mice ( Tian et al . , 2014 ) , we found that Nfatc1-expressing cells first appear at the upper hair peg and remain relatively dormant there , without actively contributing to HF down growth ( Figure 1F ) . Then , at the end of organogenesis , Nfatc1-expressing hair peg cells become the sebaceous gland above bulge and most of the HFSCs ( Figure 1D ) . A second HFSC marker , Lgr5 , follows a similar pattern , albeit with different cell fate preferences . Using Lgr5-GFP-CreER::Rosa-stop-tdTomato mice ( Jaks et al . , 2008 ) , we found that Lgr5-expressing cells appear first in the middle portion of the hair peg . They eventually become HFSCs in the lower bulge and secondary hair germ cells below the bulge ( Figure 1D ) . The labeling efficiencies of the CreER lines used here were very high; almost all the HFs in the chosen region targeted for quantifications were labeled , and we did not detect any leaky induction in the absence of Tamoxifen injection ( Figure 1—figure supplement 2D-G ) . Collectively , these results suggest that HFSCs originate from progenitor cells in hair peg that lose Shh expression but gain the expression of adult stem cell markers ( Figure 1G ) . The hair peg SC precursors are distinguished from the broader placode and hair germ cells in that they stop further development once they are specified . During subsequent HF down growth , these cells remain dormant in their original position until the end of organogenesis . At that point , they incorporate into the outer layer of the bulge and become HFSCs . This leads to an interesting question: is the reason these cells gain expression of stem cell markers and become dormant due to signals that they receive at the specific upper follicle position ? We used a two-photon laser to precisely ablate SC precursors in hair peg labeled by Nfatc1-CreER::Rosa-stop-tdTomato::K14H2BGFP ( Figure 2A , B ) . Successful and precise cell ablation was achieved without affecting normal development of neighboring HFs ( Figure 2—figure supplement 1A ) ( Rompolas et al . , 2012 ) . Following complete ablation of Nfatc1-CreER-labeled cells , we observed de novo bulge formation at the end of organogenesis ( Figure 2C ) . The newly formed bulge even had the same number of SCs as the control HFs ( Figure 2D ) . To achieve maximum labeling and ablation efficiency of Nfatc1+ cells , we conducted the ablation experiment only in HFs along the anterior-posterior midline of the dorsal tail skin , as this region has the highest Tamoxifen induction efficiency ( Figure 2—figure supplement 1B ) . Under this condition , almost all SCs in neighboring control HFs were tdTomato positive , while the de novo formed SCs in the ablated HFs were all tdTomato negative ( Figure 2E ) . Importantly , the de novo formed HFSCs were functional as indicated by their ability to support HF down growth ( Figure 2F-I and Figure 2—figure supplement 1C ) . These results indicate that niche location in hair peg determines HFSC fate . 10 . 7554/eLife . 10567 . 006Figure 2 . Niche position in the hair peg determines hair follicle stem cell fate . ( A ) Diagram of two-photon-mediated cell ablation experiment using Nfatc1-CreER::Rosa-stop-tdTomator::K14H2BGFP mice . ( B ) Representative images of hair follicles before and after cell ablation in a live animal . ( C ) Representative whole-mount images of hair follicles 15 days after cell ablation . Notice that the control hair follicle has tdTomato+ cells in the bulge while the ablated hair follicle has a normal bulge composed of tdTomato- cells . ( D ) Quantification of hair follicle stem cell number at telogen using whole mount samples . N=6 mice , >12 HFs . ( E ) Quantification of percentage of tdTomato+ outer bulge cells at telogen using whole mount samples . N=6 mice , >12 HFs . ( F–H ) Representative images of hair follicles before ( F ) , immediately after ( G ) , and 21 days after ( H ) cell ablation from the same mouse . Notice that both the control hair follicle and ablated hair follicle enter anagen . ( I ) Quantification of hair follicles that have started regeneration 21 days post cell ablation . N=5 . Scale bars: 50 μmDOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 00610 . 7554/eLife . 10567 . 007Figure 2—figure supplement 1 . Ablation specificity , Tamoxifen induction efficiency , and whole-mount views of the two-photon-mediated cell ablation experiments . ( A ) Representative images of hair follicles before , immediately after , and 2 days post ablation in the same live mouse . Notice the ablated hair placodes disappear , while the follicles adjacent to them develop normally . Scale bar: 50 µm . ( B ) Whole-mount tail skin image of Nfatc1-CreER::Rosa-stop-mTmG mice with lineage tracing starting at the hair peg stage . The percentage of labeled HFSCs decreased along the dorsal-ventral axis starting from the dorsal anterior-posterior midline . Scale bar: 500 µm . ( C ) Whole-mount image of tail skin used for quantification 21 days after ablation . Ablated hair follicles are numbered and highlighted in red . Scale bar: 500 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 007 To understand the underlying mechanism leading to niche-induced HFSC specification , we profiled the expression of mRNA isolated from purified hair peg cells expressing Shh or Nfatc1 ( Figure 3A-C ) . For the purpose of identifying niche-defining factors , we focused on genes associated with extracellular signal paths . Rather than identifying signals that were uniquely present in hair peg niche occupying cells ( Nfatc1+ cells ) , we found signals that were uniquely absent from these cells . Based on DAVID functional gene annotation analysis ( Dennis et al . , 2003 ) , among genes that were expressed ≥2-fold more in Shh+ cells than in Nfatc1+ cells , we found that genes of the Wnt signaling pathway were enriched prominently ( Figure 3D ) . Several canonical Wnts and some well-known Wnt/β-catenin target genes , including Axin2 , were highly expressed in Shh+ cells ( Figure 3—figure supplement 1 , Figure 3—source data 1 ) . Using either real-time PCR or in situ hybridization , we confirmed that Axin2 , Wnt10b , Wnt3 , and Lef1 were all expressed at higher levels in Shh+ cells than in Nfatc1+ hair peg cells ( Figure 3E-G ) . We then used TOPGAL Wnt/β-catenin signal reporter mice to validate the patterns we had observed . Starting in the placode stage and persisting through to subsequent development stages , Shh+ cells showed consistent co-localization with β-gal+ cells ( Figure 3H ) . These results suggest that the hair peg niche position is associated with an absence of Wnt/β-catenin signaling . 10 . 7554/eLife . 10567 . 008Figure 3 . Unbiased RNA-seq analysis reveals factors that define the hair peg niche . ( A ) Diagram of the FACS experiments using Nfatc1- and Shh-CreER::Rosa-stop tdTomato::K14H2BGFP mice . ( B ) FACS isolation of distinct GFP+RFP+ populations to obtain Shh+ and Nfatc1+ epithelial cells . ( C ) Unsupervised hierarchical clustering and heat map display of genes that were differentially expressed between Shh+ cells and Nfatc1+ cells . N=2 ( D ) Gene Ontology analysis of ≥2-fold up-regulated genes in Shh+ cells compared to Nfatc1+ cells . The Wnt signaling pathway is highlighted . ( E ) Validation of differentially expressed genes using qPCR . N=3 . ( F–G ) In situ staining of Axin2 ( F ) and Wnt10b ( G ) in developing hair follicles . ( H ) Shh+ cells are Wnt/β-catenin signal responsive cells . Shh expression was represented by tdTomato in Shh-CreER::Rosa-stop-tdTomato mice . Wnt/β-catenin-responsive cells were detected by β-gal staining in TOPGAL mice . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 00810 . 7554/eLife . 10567 . 009Figure 3—source data 1 . RNA-seq results of differentially expressed genes between Nfatc1+ and Shh+ cells . Genes with significantly different expression levels ( p<0 . 01 and log2FC>1 ) between the Nfatc1+ and Shh+ cells were chosen for further analysis . N=2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 00910 . 7554/eLife . 10567 . 010Figure 3—figure supplement 1 . RNA seq results of representative genes from different populations . ( A ) RNA seq results of Nfact1 and Shh from different populations . N=2 ( B ) Fold enrichment of canonical Wnt pathway genes from the RNA-seq results . N=2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 010 To directly test whether or not attenuated Wnt/β-catenin signaling in epithelial cells was correlated with HFSC specification at the hair peg stage , we conducted experiments with Axin2-CreER::Rosa-stop-mTmG mice ( Lim et al . , 2013 ) . As expected , Wnt/β-catenin signal responsive cells in placode give rise to most of the HFSCs ( Figure 4A , B ) ( Huelsken et al . , 2001 ) . This clear correlation at the placode stage was dramatically decreased at the hair germ stage , and even further decreased at the hair peg stage . Only a very small proportion of HFSCs came from Axin2+ hair peg cells; the majority of HFSCs came from Axin2- hair peg cells ( Figure 4A , B ) . This suggests that from the placode stage to the hair peg stage , cells that maintain active Wnt/β-catenin signaling have dramatically decreased potential to form HFSC . 10 . 7554/eLife . 10567 . 011Figure 4 . Attenuated Wnt/β-catenin signaling is uniquely associated with hair follicle stem cell specification . ( A–B ) Lineage tracing with Axin2CreER::Rosa-stop-mTmG mice . N=5 mice , >120 HFs . ( C ) Descendants of Gli1+ cells were represented by the expression of tdTomato in Gli1-CreER::Rosa-stop-tdTomato::K14H2BGFP mice . Note that Gli1+ descendants are located in almost all the hair follicle cells at the placode , hair germ , and hair peg stages . ( D ) Descendants of Id2+ cells were represented by the expression of tdTomato in Id2-CreER::Rosa-stop-tdTomato::K14H2BGFP mice . Note that Id2+ descendants were only located at the hair bulb stage matrix and pre-cortex area . ( E ) Diagram summarizing the descendants of the Wnt- , Shh- , and BMP-signal responsive cells at different early developmental stages . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 011 During HF down growth , several signaling pathways are known to be involved in promoting the differentiation or changes in cell fates ( Millar , 2002 ) . Although the Wnt/β-catenin signaling pathway is the only such pathway identified from our RNA-seq results , we checked to see if similar associations existed for other pathways . We used Gli1-CreER::Rosa-stop-mTmG mice to detect descendents of Shh signal-responsive cells and Id2-CreER::Rosa-stop-mTmG mice to detect descendents of BMP signal-responsive cells ( Ahn and Joyner , 2004; Rawlins et al . , 2009 ) . Almost all the HF cells were Gli1+ in all the early developmental stages examined ( Figure 4C ) . Id2 only became active in terminally differentiated hair bulb cells after future HFSC fate had been specified ( Figure 4D ) . The results from these experiments indicate that attenuated Wnt/β-catenin signaling is uniquely associated with the HFSC-inducing niche location in hair peg ( Figure 4E ) . To investigate whether the observed correlation of attenuated Wnt/β-catenin signaling with HFSC specification is a functional requirement , we increased Wnt/β-catenin signaling in cells occupying hair peg niche position to see if that would affect HFSC specification . This was achieved by using Nfatc1-CreER::Exon3-Ctnnb1fl/wt mice ( Harada et al . , 1999 ) . Activation of canonical Wnt/β-catenin signaling in Nfatc1+ cells was confirmed by nuclear β-catenin staining ( Figure 5A , B ) . Initial HF development was normal in exon3-Ctnnb1 heterozygous ( Het ) mice . However , after morphogenesis , rather than entering a new hair cycle like the WT , the hair shafts of Het mice were shed ( Figure 5—figure supplement 1A , B ) . Section staining revealed that bulge formation was completely abolished in exon3-Ctnnb1 Het mice . There were disorganized Keratin6+ cells where the bulge should have been , but there were no cells expressing the HFSC markers CD34 or Sox9 ( Figure 5C-E ) . Thus , when Wnt/β-catenin signaling was increased in hair peg niche-occupying cells , neither HFSCs nor the bulge niche formed at the end of organogenesis . 10 . 7554/eLife . 10567 . 012Figure 5 . Elevated Wnt/β-catenin signaling abolishes hair follicle stem cell specification and suppresses Sox9 expression in hair follicles . ( A ) Diagram of the experiments using Nfatc1-CreER::Rosa-stop-tdTomato::Exon3-Ctnnb1fl/wt mice . ( B ) Nuclear β-catenin staining indicates successful activation of Wnt/β-catenin signaling in upper hair follicle . ( C–E ) Abolished bulge niche formation and hair follicle stem cell specification in exon3-Ctnnb1 Het HFs compared to WT HFs . Krt6 ( C ) is a marker for inner layer bulge cells serving as a niche that can maintain quiescence for outer layer HFSCs . Sox9 ( D ) and CD34 ( E ) are adult hair follicle stem cell markers . ( F ) Wnt/β-catenin signal responsive cells , represented by β-gal positive cells in TOPGAL mice , do not express Sox9 during normal development . ( G ) Activation of Wnt/β-catenin signaling suppresses Sox9 expression in vivo . The arrows point to protrusions resulting from elevated Wnt/β-catenin signaling in exon3-Ctnnb1 Het HFs . ( H ) Diagram of the experiments using K14-rtTA::teto-Cre::Ctnnb1fl/fl mice . β-catenin is conditionally deleted in epithelial cells by feeding mice with Doxycyclin from E17 . 5 to P11 . Samples were taken at P1 and P11 . ( I–J ) Loss of Wnt/β-catenin signaling leads to expanded Sox9 expression in HFs at both the hair germ stage ( I ) and in postnatal skin ( J ) . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 01210 . 7554/eLife . 10567 . 013Figure 5—figure supplement 1 . Effects of both gain and loss of function studies targeting Wnt/β-catenin signaling in skin . ( A ) Hematoxylin and eosin ( H&E ) staining of exon3-Ctnnb1 Het and control tail skin sections from P15 to P55 . Control hair follicles enter into the first anagen from P15 to P25 . Exon3-Ctnnb1 Het follicles enter into anagen around P45 . Afterwards they return to telogen at P55 , and new bulge structures reform . Scale bar: 50 µm . ( B ) Tail skin hair coats of control and exon3-Ctnnb1 Het mice at different time points . Notice in Het mice the disappearance of the hair coat at P30 and its reappearance at P100 . Scale bar: 1 mm . ( C ) De novo formation of a normal niche in exon3-Ctnnb1 Het mice at P100 . Note the recovered expression of the stem cell markers CD34 and Sox9; scale bar: 50 µm . ( D ) Spontaneous disappearance of exon3-Ctnnb1 Het-expressing cells shown by genotyping of isolated cells at different time points . ( E ) Conditional ablation of epithelium β-catenin leads to hair follicle development defects . β-catenin can be detected in the control mice but not in the epithelia of knockout mice . Scale bar: 50 µm . ( F ) Hair coat differences between WT and β-catenin cKO mice at P11 . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 013 Interestingly , with increased time , the exon3-Ctnnb1 Het mice started to generate new hair coats with normal bulges and HFSCs that expressed CD34 and Sox9 ( Figure 5—figure supplement 1A-C ) . Cells expressing exon3-Ctnnb1 have spontaneously disappeared from HFs , and the remaining mosaic WT cells organized into a de novo niche after entering the normal hair cycle process ( Figure 5—figure supplement 1D ) . These observation further supports our conclusion that attenuated Wnt/β-catenin signaling is required for HFSC fate specification . It is intriguing that Sox9 expression is not present in cells with elevated Wnt/β-catenin signaling . Sox9 is known to be an essential intrinsic factor that is required for HFSC formation . Ablation of epithelial Sox9 completely blocks HFSC specification ( Nowak et al . , 2008 ) . To determine if the canonical Wnt signal suppresses Sox9 expression in HFs , first we observed that in normal developing HFs , Sox9 expressing cells and Wnt/β-catenin signal responsive cells are mutually exclusive ( Figure 5F ) . Second , soon after activation of Wnt/β-catenin signaling in Nfatc1+ cells , we observed small protrusions from HFs that were positive for nuclear β-catenin . Sox9 expression in these small protrusions was completely absent , while cells immediately next to these protrusions express Sox9 ( Figure 5G ) . Lastly , in the Wnt/β-catenin loss of function mutant ( K14-rtTA::teto-Cre::Ctnnb1fl/fl ) ( Nguyen et al . , 2006; Perl et al . , 2002 ) , HF development is defective but Sox9 expression is expanded to the extent that all remaining cells in HFs express Sox9 ( Figure 5H-J ) . These results suggest that the Wnt/β-catenin signaling suppresses Sox9 expression in HFs . Given that the loss of Sox9 expression completely blocks HFSC specification , the requirement for attenuation of Wnt/β-catenin signaling as a prerequisite for HFSC specification is related to enabling Sox9 expression . It is noteworthy that hair shaft formation is aborted in K14-rtTA::teto-Cre::Ctnnb1fl/fl mice ( Figure 5—figure supplement 1E , F ) . This indicates that epithelial Wnt/β-catenin signaling is required for progenitor cell differentiation , and may further explain why activation of Wnt/β-catenin signaling prevents HFSC formation . Given that attenuated Wnt/β-catenin signaling is required for HFSC specification , it was puzzling to observe that a small portion of HFSCs originated from Axin2+ hair peg cells ( Figure 4A , B ) . This suggests that there are two origins of HFSCs: a small portion of HFSCs come from hair peg cells maintaining active Wnt/β-catenin signaling , while the majority HFSCs come from progenitor cells with attenuated Wnt/β-catenin signaling in hair peg . To investigate the long-term consequences of this heterogeneity , we followed the fates of HFSCs from the two different origins through multiple hair cycles; the Wnt/β-catenin signal positive origin was represented by the Axin+ lineage while the Wnt/β-catenin signal negative origin was represented by the Sox9+ lineage ( Figure 6A ) . 10 . 7554/eLife . 10567 . 014Figure 6 . Embryonic Wnt/β-catenin signaling diminishes the long-term self-renewal ability of hair follicle stem cells in adult . ( A ) Diagram depicting the time points of the lineage tracing experiment . ( B–C ) Long-term lineage tracing experiments using Axin2-CreER::Rosa-stop-mTmG mice . Note the disappearance of labelled HFSCs in the third telogen . Representative images shown are single confocal Z slice from the data used for quantification . N=3 mice , >110 HFs . ***p<0 . 0001 . ( D ) Long-term lineage tracing experiments with Sox9-CreER::Rosa-stop-mTmG mice . Arrowheads indicate ORS and arrows indicate terminally differentiated layers . Note the change in downstream progeny fates of labeled cells at different hair cycles . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 01410 . 7554/eLife . 10567 . 015Figure 6—figure supplement 1 . Long term cell fate of Lgr5-CreER labelled hair peg cells . ( A ) Diagram depicting the time points of the lineage tracing experiment . ( B–C ) Long-term lineage tracing experiments using Lgr5-CreER::Rosa-stop-tdTomato mice . Note the persistence of a small percentage of labelled HFSCs up to the ~6th telogen . Representative images shown are single confocal Z slice from the data used for quantification . N=3 mice , >100 HFs . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 015 Consistent with results presented in Figure 4A , ~15% of HFSCs in the first telogen originated from Axin2+ hair peg cells . In the first growth phase ( anagen ) following the first telogen , these HFSCs contributed to both outer root sheath ( ORS ) cells and to matrix differentiating cells ( Figure 6B ) . Of note , after two full hair cycles , in the third telogen , Axin2+ hair peg cell formed HFSCs have been mostly exhausted ( Figure 6C ) . On the contrary , Sox9+ hair peg precursor cells were the origin of most of the HFSCs at the first telogen . During the first anagen , these HFSCs contributed preferentially to ORS rather than to the differentiated lineages in the matrix . Instead of decreasing , HFSCs originated from Sox9+ hair peg cells persisted and expanded after one round of hair cycle to become essentially all of the HFSCs in the bulge and most of the secondary hair germ cells below the bulge . For the second anagen , the labeled HFSCs generate all of the downstream lineages including , ORS , matrix , and hair shaft cells ( Figure 6D ) . These results suggest that embryonic Wnt/β-catenin signaling diminishes the long-term potential of HFSCs . We have shown that Wnt/β-catenin signaling is required for terminal differentiation of the hair shaft . Also it suppresses Sox9 expression , which is important for HFSC maintenance through inhibition of differentiation ( Kadaja et al . , 2014 ) . So it is possible that embryonic exposure to Wnt/β-catenin signaling primes the embryonic cells for differentiation via both positive induction and lack of Sox9 expression . This ultimately results in diminished long-term potential of the HFSCs . Our study elucidates the cascade of events that lead to long-term SC emergence in HFs during organogenesis ( Figure 7 ) . Wnt/β-catenin signaling is necessary for HF initiation ( Andl et al . , 2002; Ito et al . , 2007 ) . Early placode cells are Wnt/β-catenin signal responsive cells , and they generate all of the cells in adult HFs . At the hair peg stage , a localized Wnt/β-catenin signaling free zone emerges in upper HF . Progenitor cells residing in this embryonic niche position stop further development and become HFSCs at the end of organogenesis . These established HFSC precursor cells are not just remnants of embryonic progenitor cells , because they gain the expression of adult HFSC markers including Nfatc1 . Embryonic cells residing in this Wnt/β-catenin signaling free zone , together with a few other cells located below them that have active Wnt/β-catenin signaling , give rise to the adult HFSC pool . However , only the HFSCs derived from Wnt/β-catenin signaling negative precursor cells have long-term self-renewal ability and can continuously support regeneration . The small portion of HFSCs that are derived from embryonic cells with active Wnt/β-catenin signaling are highly prone to differentiation as the hair cycle progresses and are soon exhausted after regeneration . So , in embryos , the attenuation of Wnt/β-catenin signaling not only defines the adult niche position and HFSC fate , but also has a long lasting effect on the subsequent fate choice and long-term potential of cells in adults . 10 . 7554/eLife . 10567 . 016Figure 7 . Model illustrating that long-term hair follicle stem cell emergence results from progenitors occupying an embryonic niche location , which is defined by the absence of Wnt/β-catenin signaling that would otherwise block the expression of a key factor required for stem cell specification . DOI: http://dx . doi . org/10 . 7554/eLife . 10567 . 016 It’s noteworthy that we observed largely none overlapping labeling patterns of Lgr5+ and Axin2+ cells in hair peg ( Figure 6B , Figure 6—figure supplement 1B ) . Also , Lgr5+ hair peg cells contributed HFSCs persist after multiple hair cycles instead of completely diminishing like Axin2+ cells contributed HFSCs ( Figure 6—figure supplement 1 ) . These results indicate it is possible that in hair follicle Lgr5 expression either reflects very low level of Wnt signaling , or it is induced by a different set of Wnt ligands than those induce Axin2 . Although we were able to pinpoint the main mechanism through which Wnt/β-catenin signaling blocks long-term HFSC formation , i . e . the suppression of Sox9 expression , we suspect that this might not be the only mechanism involved in this process . It has been reported previously that Sox9 expression is induced by Shh signaling ( Vidal et al . , 2005 ) . Given that we observed extended activity of Shh signaling throughout follicle epithelial cells and observed only localized activity of Wnt/β-catenin signaling in lower follicle cells , we can explain the mutually exclusive pattern of Sox9+ and Wnt/β-catenin signaling responsive cells . As development proceeds , the Sox9 expression pattern continues to expand and become wider than the actual future bulge region; for example , Sox9 is expressed in most of the ORS cells . So , following the emergence of the Wnt/β-catenin signaling free zone in the hair peg , there are probably other mechanisms that help maintain the restricted niche position . One observation further supporting this hypothesis is that , once initiated in hair peg , the expression of Nfatc1 is restricted to the upper HF where the future sebaceous gland and bulge will be; Nfatc1 expression does not expand to the broader ORS like Sox9 expression does . Nfatc1 was reported to be a target of BMP signaling in adult HFs ( Horsley et al . , 2008 ) . However , we did not observe active BMP signaling at the hair peg niche location using a BMP reporter , a result similar to a published staining pattern of pSMAD1/5 in developing HFs ( Kandyba et al . , 2014 ) . So , it remains to be determined which , if any , other mechanisms lead to the maintenance of a more restricted niche region that is reflected by the localized expression of stem cell markers . Since attenuated Wnt/β-catenin signaling defines the niche location and long-term HFSC fate , the process of exactly how the Wnt/β-catenin signaling free zone emerges is of great interest . We observed increased expression of Wnt inhibitors such as SFRPs in Nfatc1+ hair peg cells as compared to Shh+ hair peg cells . However , given that HFSC precursor cells can be replaced by embryonic cells occupying the same hair peg niche position , it is unlikely that Wnt inhibitors expressed by the precursor cells themselves are the main causes of attenuated Wnt/β-catenin signaling in that position . Our RNA seq analysis revealed enrichment of Wnt and Dkk expression in Wnt/β-catenin signal positive Shh+ hair peg cells as compared to Nfatc1+ hair peg cells . We hypothesize that the combination of the lack of Wnt ligands , which are short-range signal molecules , and the presence of Wnt inhibitors such as DKK4 , which are long-range signal molecules , at the embryonic niche location is the main reason contributes to the emergence of a Wnt/β-catenin signaling free zone in upper hair peg . A similar model has been used to explain HF density determination and epidermal SC regulation ( Sick et al . , 2006; Lim et al . , 2013 ) . To directly test the suppositions of this hypothesis , the protein localization patterns of Wnt and DKK will need to be dissected . Nfatc1-CreER mice were generated and provided by Dr . Bin Zhou ( Tian et al . , 2014 ) . K14-H2BGFP mice were kindly provided by Dr . Elaine Fuchs . The Shh-CreER ( Harfe et al . , 2004 ) , Lgr5-GFP-CreER ( Barker et al . , 2007 ) , Axin2-CreER ( Lim et al . , 2013 ) , Id2-CreER ( Rawlins et al . , 2009 ) , Gli1-CreER ( Ahn and Joyner , 2004 ) , Rosa-stop-mTmG ( Muzumdar et al . , 2007 ) , Rosa-stop-tdTomato ( Madisen et al . , 2010 ) , TOPGAL ( DasGupta and Fuchs , 1999 ) , Ctnnb1fl/fl ( Huelsken et al . , 2001 ) , Exon3-Ctnnb1fl/fl ( Harada et al . , 1999 ) , K14-rtTA ( Nguyen et al . , 2006 ) , and teto-Cre ( Perl et al . , 2002 ) mice have all been described previously . We generated the Sox9-CreER mice by integrating IRES-CreERT2-SV40pA cassettes into the 3' UTR of the endogenous mouse Sox9 gene , before the aacatggaggacgattggagaatc sequence via Cas9/RNA mediated gene targeting in zygotes . To perform the lineage tracing experiments , female Rosa-reporter mice were mated with male mice of the following genotypes: Shh- , Lgr5-GFP- , Nfatc1- , Axin2- , Gli1- , Sox9- , and Id2-CreER . For the calculation of embryonic time points in the timed pregnancy experiments , the morning of the vaginal plug date was designated as embryonic stage 0 . 5 ( E0 . 5 ) . Cre activity was induced at E16 . 5 or E17 . 5 by a single intraperitoneal injection of Tamoxifen dissolved in sunflower oil/10% ethanol . Solutions ranging from 10–20 mg/ml were used with the following dosage on the basis of body weight: pregnant mothers of the Shh- , Nfatc1- , Axin2- , Gli1- , Sox9- and Id2-CreER::Rosa-reporter mice received a single dose of Tamoxifen totaling 40 µg/g body weight; the dose for the Lgr5-GFP-CreER::Rosa-reporter mice was 100 µg/g body weight . At the indicated time points specified in figures , tail skin was removed and treated with 20 mM EDTA ( pH 8 . 0 ) for 2–4 hr at 37°C . Epidermis with attached hair follicles was removed from the dermis , fixed in 4% paraformaldehyde for 20 min , washed in PBS for >2 hr and then processed for staining or imaging . For the Lgr5-GFP-CreER::Rosa-stop-tdTomato::K14H2BGFP lineage tracing experiment , tail skin was removed and treated with 20 mM EDTA ( pH 8 . 0 ) for 30 min at 37°C . Epidermis without hair follicles was removed from the dermis . The remaining dermal tissue with embedded hair follicles was fixed in 4% paraformaldehyde for 20 min , then washed for 1 hr in PBS and imaged without staining . For the mice more than 100 days old , the remaining dermal tissue was digested additionally in the collagenase ( Sigma ) for 1 hr and then washed with PBS before imaging to allow detection of fluorescence signal . Whole-mount tail skin images were acquired using a 20×0 . 75 objective lens . Z-stacks were acquired at a resolution of 1024×1024 , or at 512×512 for large samples . Tissue and section samples were imaged with a Nikon A1-R confocal microscope . Microscopy data was analyzed using Imaris ( 3D software ) with the 3D visualization module . To label distinct cell populations in hair germs at the chosen area of tail skin , pregnant mice were administered a single dose of Tamoxifen at E16 . 5 , and the labeling pattern was established 48 hr later at E18 . 5 using embryos removed by Caesarian section . At this time point , the primary central follicle in tail skin follicle triplets is in the hair germ stage , while the secondary outer follicles have not yet been initiated . To continue lineage tracing , a foster mother nursed the remaining pups . At postnatal day 15 ( P15 ) , the initially labeled primary central follicles enter into the first telogen and the fates of the labeled cells were analyzed . At this stage the secondary outer follicles are in anagen and block the view of the telogen primary central follicle . Prior to imaging , the secondary outer hair follicles were plucked by tweezers , leaving only the central hair follicle in the P15 whole mount skin . To label specific cell populations at the placode and the hair peg stages , pregnant mice were administered a single dose of Tamoxifen at E17 . 5 and labeling patterns were established 48 hr latter at P1 when the primary central follicle is at the hair peg stage and the secondary outer follicles are at the placode stage . The fate of labeled cell populations was traced to P15 for the primary central follicle and P21 for the secondary outer follicles , when they enter into the resting phase separately . The development of tail skin hair follicles varies slightly , in both the anterior-posterior axis and in the dorsal-ventral axis . To label progenitor cells at defined developmental stages , only the middle one-third section along both the length and the width of dorsal tail skin was used to perform lineage-tracing experiments . Within this chosen area , hair follicle development consistently follows the above-mentioned pattern at the indicated time points . To count the number of HFSCs , multiple Z-stack images were taken for each whole mount of tail skin . For each individual telogen HF , only the cross section through the center of the hair shaft was used for counting . This was done by picking the Z slice with the largest hair shaft for each counted HF . Telogen HFSC number was counted as the outer layer of cells from the U-shaped bulge structure below the SG and above the secondary hair germ . This standard was chosen based on section staining of tail skin hair follicles with HFSC markers ( CD34 and Sox9 ) and niche cell marker ( Krt6 ) ( for reference , please see Figure 5C–E control HFs ) . P1 mice were anesthetized with cotton containing isoflurane and kept anesthetized on a 37°C heat stage during the imaging process with vaporized isoflurane through a gas tube connected to the head . The tail of the mouse was immobilized by tape and imaged directly under a water lens . Imaging was performed with a BX61WI ( Olympus ) microscope equipped with a Chameleon Ultra ( COHERENT ) two photon laser and a 25× water lens ( Olympus , UIS2 , N . A . 1 . 05 ) . A laser beam of 910 nm was used to simultaneously excite both H2BGFP and tdTomato . The step increment of the serial optical sections was 2 μm . Prior to laser ablation , Nfatc1-CreER::Rosa-stop-tdTomato::K14-H2BGFP pregnant mice were injected with a single dose of Tamoxifen at E17 . 5 to label distinct cell populations in hair pegs on tail skin . Laser ablation was carried out with a 910 nm beam scanning a region of 10 μm2 for less than 5 s . The power of the laser intensity was 1 . 86 mW , and less than 80% of the laser power was used . Neil Blue dye was used to tattoo a marker at the vicinity of the ablated area to help us retrieve the same laser ablated follicles more than 15 days later . To express the stabilized form of β-catenin in Nfatc1 expressing cells , Exon3-Ctnnb1fl/fl::Rosa-stop-tdTomatofl/fl mice were mated with Nfatc1-CreER mice . To induce the conditional expression of exon3-Ctnnb1 , nursing mothers of Nfatc1-CreER::Rosa-stop-tdTomatofl/wt::Exon3-Ctnnb1fl/wt pups and WT littermates were intraperitoneally injected with Tamoxifen doses totaling 40 μg/g body weight for 8 days starting from E17 . 5 . To generate K14-rtTA::teto-Cre::Ctnnb1fl/fl mice , K14-rtTA::teto-Cre mice were mated with Ctnnb1fl/fl mice . F1 male K14-rtTA::teto-Cre::Ctnnb1fl/wt progeny were subsequently mated with female homozygous Ctnnb1fl/fl . To induce deletion of Ctnnb1 , the nursing mothers of K14-rtTA::teto-Cre::Ctnnb1fl/fl pups and WT littermates were given water containing 1 mg/ml doxycycline ( Sigma ) and chow containing 2 g/1000 g doxycycline ( Research Diets Company ) for periods specified in the figures . All mice were maintained in an SPF facility , and procedures were conducted in a manner consistent with the National Institute of Biological Sciences guide for the care and use of laboratory animals . For section staining , tissues were embedded in OCT compound and frozen . After cryosection ( 20–30 µm ) and fixed for 10 min in 4% paraformaldehyde in PBS , sections were permeabilized for 10 min in 0 . 5% Triton ( PBST ) and blocked for 1 hr in a solution of 2% normal donkey serum , 1% BSA , and 0 . 3% Triton in PBS . The following antibodies were used: anti-GFP ( Abcam , ab13970 , 1:1000 ) , anti-P-cad ( R&D , BAF761 , 1:500 ) , anti-β-gal ( Abcam , ab9361 , 1:10 , 000 ) , anti-β-catenin ( Sigma C2206 , 1:500 ) , anti-Sox9 ( Chen Lab , 1:50 ) , anti-CD34 ( ebioscience , 50-0341 , 1:300 ) , and anti-K6 ( Chen Lab , 1:2000 ) . For hematoxylin and eosin ( H&E ) staining , after cryosectioning ( 6–8 µm ) and fixation for 10 min in 4% paraformaldehyde in PBS , sections were stained in haematoxylin ( Fisher ) for 30 s and then rinsed in running tap water and 0 . 3% acid alcohol . Eosin ( Sigma ) was used to treat the sections for 10 s . The haematoxylin and eosin dyes were prepared according to the manufacturers’ instructions . For the whole-mounts of tail skin , images were acquired using a 20× 0 . 75 objective lens . Z-stacks were acquired at a resolution of 1024×1024 , or 512×512 for large samples . Tissue and section samples were imaged on a Nikon A1-R confocal microscope . Microscopy data were analyzed using Imaris ( 3D software ) with the 3D visualization module . RBG images were assembled in Adobe Photoshop CS3 and panels were labeled with Adobe Illustrator CS6 . Skin was fixed with 4% paraformaldehyde , embedded in paraffin , and sectioned at 5 µm . After de-waxing , sections were blocked with buffer from a MOM kit ( Vector ) with 0 . 1% Triton . Immunohistochemistry assays were then conducted by incubating sections at 4°C overnight with primary antibody β-catenin ( Sigma C7207 , 1:500 ) , then secondary HRP Donkey-anti-mouse antibody ( Jackson ImmunoResearch ) , followed by use of a DAB kit ( Vector ) , according to the manufacturer’s instructions . RNA in situ hybridization was performed with an RNAscope 2-plex Detection Kit ( Chromogenic ) according to the manufacturer’s instructions . The RNAscope probes used were Axin2 ( Catalog 400331 ) and Wnt10b ( Catalog 401071 ) . Isolation of Shh+ and Nfatc1+ cells for RNA-seq was performed using Shh- and Nfatc1-CreER::Rosa-stop-tdTomato::K14-H2BGFP mice . For purification of Nfatc1 expressing cells in hair pegs , pregnant mice were administered a single dose of Tamoxifen at E15 . 5 . Then , at E17 . 5 , embryonic back skin was removed and the whole skin was cut into 6–9 small pieces and placed , dermis side down , in a 0 . 25% trypsin solution ( Gibco ) at 37°C for 16 min . Single-cell suspensions were obtained by triturating the skin gently . Cells were then filtered with strainers ( 70 µm followed by 40 µm ) . For purification of Shh-expressing hair peg cells at E17 . 5 , pregnant mice were administered a single dose of Tamoxifen at E15 . 5 . Embryonic back skin was cut into 9 pieces and placed , dermis side down , in dispase ( Life Technologies , 0 . 4 mg/ml ) dissolved in PBS for 1 hr at 37°C . Epidermis containing placodes and hair germs were removed . Dermis samples containing only hair pegs were placed in a 0 . 25% trypsin solution at 37°C for 10 min . Single-cell suspensions were obtained by triturating the skin gently . Cells were then filtered with strainers ( 70 µm followed by 40 µm ) . For the cells used in the experiments presented in Figure 5—figure supplement 1D , back skin of Nfatc1-CreER::Rosa-stop-tdTomatofl/wt::K14-H2BGFP::Ctnnb1fl/wt Het and WT mice at either P23 or P100 was used . Skin was cut into 6 to 9 pieces and placed , dermis side down , in a 0 . 25% trypsin solution at 37°C for 30 min . Epithelium cells were harvested by gentle scraping and triturating . After filtering with strainers , cells were stained for 20 min on ice with Alexa647-CD34 antibody ( ebioscience , 50-0341 ) where indicated and then washed . DAPI was used to exclude dead cells . Cell isolations were performed with a BD FACSAria III cell sorter equipped with FACSDiva software ( BD Bioscience ) . Total RNA was isolated from FACS-purified cells lysed with Trizol ( Life Technologies ) followed by extraction using a direct-Zol RNA mini prep kit ( Zymo research ) . Equal amounts of RNA were added to a reverse-transcriptase reaction mix ( Takara ) with Oligo ( dT ) . Expression levels were normalized to the expression of GAPDH or PPIB . Real-time PCR was conducted using a CFX96 Real-Time system ( Bio-Rad ) with Power SYBR Green PCR Master Mix ( Life Technologies ) . All primer pairs were designed for the same cycling conditions , which were: 10 min at 95°C for initial denaturing , 40 cycles of 10 s at 95°C for denaturing , 30 s at 58°C for annealing , and 10 s at 65°C for extension . The primers were designed to produce a product spanning exon-intron boundaries in the target genes; the sequences were as follows: Shh F , AAAGCTGACCCCTTTAGCCTA; Shh R , TTCGGAGTTTCTTGTGATCTTCC; Nfatc1 F , CCATACGAGCTTCGGATCGA; Nfatc1 R , AGTAACCGTGTAGCTGCACAATG; Wnt3 F , AGCTGCCAAGAGTGTATTCG; Wnt3 R , CTAGATCCTGCTTCTCATGGG; Wnt10b F , AAGTCACAGAGTGGGTCAATG; Wnt10b R , GCCACGATAAACCCTAGACAG; Lef1 F , AGCCTGTTTATCCCATCACG; Lef1 R , TGTTACAATAGCTGGATGAGGG; PPIB F , GTGAGCGCTTCCCAGATGAGA; PPIB R , TGCCGGAGTCGACAATGATG; GAPDH F , GGTGTGAACGGATTTGGCCGTATTG; GAPDH R , CCGTTGAATTTGCCGTGAGTGGAGT . RNA from FACS-purified cells was sent to the Biodynamic Optical Imaging Center at Peking University for quantification , RNA-seq library preparation , and sequencing . The libraries were sequenced on the illumina HiSeq 2500 platform using the Pair-End 2x100-bp sequencing strategy . For analysis , data was mapped to the mouse genome ( GRCm38/mm10 ) using TopHat ( v2 . 0 . 13 ) with default settings . Cufflinks ( v2 . 2 . 1 ) was used to quantify changes in gene expression between the Nfatc1+ and Shh+ cells . Genes with significantly different expression levels ( p<0 . 01 and log2FC>1 ) were chosen for further analysis . Gene ontology ( GO ) analysis of genes was done using DAVID ( Database for Annotation , Visualization and Integrated Discovery ) . DNA was isolated from FACS-purified cells using a TIANamp Genomic DNA Kit ( TIANGEN ) according to the manufacturer’s instructions . The primers used to detect the β-catenin sequences flanking the exon3 region were: wild-type/un-recombined allele 900 bp; exon3 F , GACACCGCTGCGTGGACAATG; exon3 R , GTGGCTGACAGCAGCTTTTCT; recombined allele ~700 bp; exon3 F , GACACCGCTGCGTGGACAATG; exon3 R , ACGTGTGGCAAGTTCC GCGTCATCC . Box-and-Whisker plots prepared with Origin 9 ( Origin Lab ) software were used to illustrate the lineage tracing quantification results .	Many tissues and organs in an adult’s body – including bone marrow , skin and intestines – contain a small number of cells called adult stem cells . These cells usually stay dormant within these tissues ( at a site called a ‘niche’ ) until they are required to repair damaged or lost cells . At this point , adult stem cells can specialize , or ‘differentiate’ , into the many different cell types that make up the tissue or organ where they reside . The cells that produce hairs are an example of adult stem cells . In mammals , hairs grow from structures called hair follicles that are found in the skin , and over the life of an animal , old hairs are shed and replaced . Previous research had suggested that certain embryonic cells are set to become hair follicle stem cells before the hair follicles emerge in the adult tissue . However it remained unclear how this decision is made , and which genes and molecules are involved in this process . Xu et al . have now found that , in mice , the fate of hair follicle stem cells is decided at an early stage in development , when the hair follicle is a simpler structure called a ‘hair peg’ . Cells near the upper part of the hair peg tend to become dormant and adopt an adult stem cell fate , while the ones in the lower part are more likely to differentiate straight away . This shows that the position , hence the niche environment , plays a key role in determining these different cells’ fates . Xu et al . went on to discover that the decision for a cell to become a hair follicle stem cell relies on reduced signaling through the so-called Wnt signal pathway . Understanding how adult stem cells become established during development may help future efforts to grow tissues and organs in the laboratory for research purposes or organ transplantation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2015	Embryonic attenuated Wnt/β-catenin signaling defines niche location and long-term stem cell fate in hair follicle
Protein kinases are potentially attractive therapeutic targets for neglected parasitic diseases , including African trypanosomiasis caused by the protozoan , Trypanosoma brucei . How to prioritize T . brucei kinases and quantify their intracellular engagement by small-molecule inhibitors remain unsolved problems . Here , we combine chemoproteomics and RNA interference to interrogate trypanosome kinases bearing a Cys-Asp-Xaa-Gly motif ( CDXG kinases ) . We discovered that hypothemycin , a fungal polyketide previously shown to covalently inactivate a subset of human CDXG kinases , kills T . brucei in culture and in infected mice . Quantitative chemoproteomic analysis with a hypothemycin-based probe revealed the relative sensitivity of endogenous CDXG kinases , including TbGSK3short and a previously uncharacterized kinase , TbCLK1 . RNAi-mediated knockdown demonstrated that both kinases are essential , but only TbCLK1 is fully engaged by cytotoxic concentrations of hypothemycin in intact cells . Our study identifies TbCLK1 as a therapeutic target for African trypanosomiasis and establishes a new chemoproteomic tool for interrogating CDXG kinases in their native context . Human African trypanosomiasis , or sleeping sickness , is a debilitating and fatal parasitic disease endemic to sub-Saharan Africa ( Fevre et al . , 2008; Simarro et al . , 2008 ) . Caused by two subspecies of Trypanosoma brucei , the disease begins in the hemolymphatic system and later crosses the blood–brain barrier , resulting in sleep disturbances that deteriorate to coma and death . During the course of infection , the parasites evade the host immune system through periodic switching of the variable surface glycoprotein coat , making vaccination-based approaches unlikely to succeed ( Taylor and Rudenko , 2006; Horn and McCulloch , 2010 ) . Currently , only four chemotherapeutics are approved for sleeping sickness . These therapies suffer from poor oral bioavailability , severe toxicity , and/or emerging resistance . Eflornithine , the only drug with a known mechanism of action , is also the only therapeutic for sleeping sickness approved in the last 50 years ( Fairlamb , 2003 ) . Although a combination of eflornithine and nifurtimox , a drug for American trypanosomiasis ( Chagas disease , caused by T . cruzi ) , was introduced in 2009 for T . brucei infection , no new drugs are near the clinic ( Priotto et al . , 2009 ) . Protein kinases have been intensely pursued as therapeutic targets for cancer and autoimmune disease , and more than 15 kinase inhibitors have received FDA approval during the past decade . As in humans , protein phosphorylation in T . brucei is an essential regulatory mechanism and plays critical but poorly understood roles in its unique life cycle ( Parsons et al . , 2005; Nett et al . , 2009 ) . The vast majority of the 182 protein kinases in T . brucei remain poorly characterized , and few have been interrogated in their native cellular context with small-molecule inhibitors . High-throughput screening and medicinal chemistry recently led to the discovery of SCYX-5070 , the first reported protein kinase inhibitor with efficacy in a murine model of T . brucei infection ( Mercer et al . , 2011 ) . Affinity chromatography with an immobilized derivative suggested that SCYX-5070 binds at least six trypanosome kinases related to human mitogen activated protein kinases ( MAPKs ) and cyclin-dependent kinases , although the extent to which SCYX-5070 engages these kinases in vivo is not known . Genetic studies in T . brucei have established essential roles for orthologs of several human kinases , including cyclin-dependent kinases ( Tu and Wang , 2004; Gourguechon and Wang , 2009 ) , Aurora kinase ( Tu et al . , 2006 ) , polo-like kinase ( Li et al . , 2010 ) , glycogen synthase kinase-3 ( Ojo et al . , 2008 ) , casein kinase-1 ( Urbaniak , 2009 ) , and DBF-2-related kinases ( Ma et al . , 2010 ) . High-throughput RNAi studies have implicated additional kinases in T . brucei proliferation ( Alsford et al . , 2011; Mackey et al . , 2011 ) . Despite the knowledge gained from these studies , it is not clear which T . brucei kinases make the best therapeutic targets . Achieving sufficient potency to inhibit essential T . brucei kinases , while avoiding related human kinases , poses a significant challenge . Moreover , quantifying intracellular kinase engagement by small-molecule inhibitors in T . brucei is an unsolved problem . Pharmacological and chemoproteomic approaches are therefore needed to complement genetic studies in the search for new therapeutic targets ( Moellering and Cravatt , 2012 ) . Hypothemycin ( 1 , Figure 1A ) is a polyketide natural product that covalently inhibits a diverse subset of human kinases , including MEK , ERK , PDGFR , VEGFR2 , and FLT3 ( Schirmer et al . , 2006; Winssinger and Barluenga , 2007 ) . Hypothemycin ( Tanaka et al . , 1999 ) and related compounds ( Barluenga et al . , 2010 ) have antitumor activity in mouse xenograft models , and one variant has entered clinical trials ( Kumar et al . , 2011 ) . All hypothemycin-sensitive kinases have a common cysteine immediately preceding the catalytic DXG motif ( where X is usually Phe or Leu ) ( Schirmer et al . , 2006 ) . A homologous CDXG motif is present in 48 of 518 human protein kinases ( Leproult et al . , 2011 ) . CDXG kinases are functionally diverse , encompassing Tyr , Ser/Thr , and dual-specificity kinases distributed throughout the kinome phylogenetic tree . The nucleophilic thiol of hypothemycin-sensitive CDXG kinases undergoes conjugate addition to the cis-enone , as revealed by a crystal structure of the ERK2/hypothemycin complex ( Rastelli et al . , 2008 ) . While the CDXG motif is necessary for potent inhibition by hypothemycin , it does not appear to be sufficient . One-third of human CDXG kinases were unaffected by hypothemycin variants screened at a concentration of 1 μM ( Barluenga et al . , 2010 ) . The selectivity of hypothemycin toward endogenous kinases in living cells remains unknown . 10 . 7554/eLife . 00712 . 003Figure 1 . Potent trypanocidal activity of hypothemycin . ( A ) Hypothemycin ( 1 ) inhibits CDXG kinases via conjugate addition to the cis-enone . ( B ) Bloodstream form T . brucei were treated with hypothemycin and cell density was measured after 24 hr ( mean ± SD , n = 3 ) . ( C ) Parasitemia in T . brucei infected mice . Mice received once daily intraperitoneal injections of hypothemycin and parasitemia was measured 5 days post infection ( mean ± SD , n = 4 , ** denotes p<0 . 05 ) . ( D ) Kaplan-Meier analysis of T . brucei infected mice . Hypothemycin or vehicle was administered by intraperitoneal injection once daily for 7 days post infection ( arrowheads , n = 12 ) . Data were accumulated from three studies , p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 00712 . 003 In this study , we exploit a semi-synthetic derivative of hypothemycin to identify potential therapeutic targets in T . brucei . We found that hypothemycin is potently trypanocidal , both in cell culture and in mice . To identify its molecular targets , we synthesized the equipotent propargyl-hypothemycin derivative 2 ( Figure 2B ) . This affinity probe provided a means to purify and identify covalent kinase targets of hypothemycin . More importantly , probe 2 enabled us to quantify the extent of kinase engagement upon exposure to a defined concentration of hypothemycin . In principle , the workflow described in this study— ( 1 ) phenotypic evaluation of mammalian kinase inhibitors , ( 2 ) target identification with chemoproteomic tools , ( 3 ) genetic knockdown of candidate targets , and ( 4 ) correlation of intracellular target engagement with phenotype—can be used to identify and prioritize kinase targets in other therapeutic contexts . 10 . 7554/eLife . 00712 . 004Figure 2 . Design and validation of a hypothemycin-based affinity probe . ( A ) Crystal structure of hypothemycin ( 1 ) bound to ERK2 ( PDB: 3C9W ) , indicating the solvent-exposed C4 methyl ether . ( B ) Structure of probe 2 . ( C ) Effect of 2 on proliferation of cultured BSF T . brucei ( mean ± SD , n = 3 ) . ( D ) T . brucei whole-cell lysates were treated with 2 for 30 min . Labeled proteins were visualized after click conjugation to rhodamine-azide and separation by SDS-PAGE . ( E ) Fluorescence quantification of bands at 90 , 50 , and 43 kDa demonstrating saturation of the 43 kDa band , but not the 90 or 50 kDa bands . ( F ) Lysates were treated with the indicated concentrations 1 for 30 min , followed by 500 nM 2 for 30 min . Labeled proteins were visualized as above . The saturable 43 kDa protein is indicated by the arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 00712 . 004 Similar to humans , approximately ten percent of T . brucei kinases ( 21 of 182 ) have a CDXG motif . Hypothemycin’s selectivity toward CDXG kinases prompted us to test its effects on T . brucei grown in culture . We reasoned that hypothemycin would enable pharmacological interrogation of a focused subset of kinases that are distributed throughout the kinome . Treatment of bloodstream form ( BSF ) parasites with increasing concentrations of hypothemycin for 24 hr caused a sharp reduction in cell density ( EC50 ∼ 170 nM , Figure 1B ) , principally resulting from cell death . Visual inspection revealed that cells treated for 5 hr with 500 nM hypothemycin were round and swollen , with the flagellum detached from the cell body . They died within 2 hr and did not recover even when transferred to media lacking hypothemycin ( data not shown ) . Previous studies with hypothemycin demonstrated activity in murine tumor xenograft models ( Tanaka et al . , 1999 ) . We therefore tested hypothemycin in mice infected with T . brucei . Infected mice showed a dose-dependent reduction in parasitemia ( Figure 1C ) , and a 10 mg/kg dose administered once daily for 7 days prolonged survival of infected mice over 30 days , with a cure rate of 33% ( Figure 1D ) . At a dose of 10 mg/kg or higher , signs of toxicity were evident in infected animals ( weight loss , lethargy ) , precluding more aggressive dosing regimens . Although hypothemycin’s therapeutic window is narrow , likely due to inhibition of essential mammalian CDXG kinases ( e . g . , VEGFR2 , MEK1/2 ) , its potent trypanocidal activity in vitro and in vivo motivated us to search for trypanosomal hypothemycin-binding proteins . We hypothesized that hypothemycin’s trypanocidal effects were mediated by covalent inhibition of one or more T . brucei CDXG kinases and sought to identify its targets in an unbiased manner . In the crystal structure of hypothemycin bound to ERK2 , the C4 methyl ether is solvent exposed ( Rastelli et al . , 2008 ) , suggesting that this and other kinases would accommodate a larger group at this position ( Figure 2A ) . We therefore replaced the methyl ether with a propargyl ether to enable copper-promoted ( ‘click’ ) conjugation to a biotin- or rhodamine-linked azide . Alkylation of 4-O-desmethyl hypothemycin ( Wee et al . , 2006 ) with propargyl bromide was promoted by cesium carbonate to afford the C4 propargyl ether 2 , ( Figure 2B ) . Consistent with our hypothesis , probe 2 was equipotent to hypothemycin against cultured BSF T . brucei ( Figure 2C ) . We next tested the ability of 2 to covalently modify T . brucei proteins . Treatment of whole-cell lysates with 2 , followed by click conjugation to rhodamine-azide , revealed multiple fluorescent bands by SDS-PAGE . Labeling of a 43 kDa protein was detected with 25 nM 2 , reaching maximum intensity at 500 nM ( Figure 2D , E ) . Labeled bands at 50 and 90 kDa were apparent at higher concentrations , but they failed to saturate ( Figure 2E ) and are likely nonspecific adducts . Pretreating lysates with increasing concentrations of hypothemycin abolished subsequent labeling of the 43 kDa band , demonstrating that it is a specific and saturable hypothemycin target ( Figure 2F ) . To identify the 43 kDa protein , we used biotin-azide in the click reaction and affinity purified covalently modified proteins . After SDS-PAGE , analysis of the 43 kDa band by mass spectrometry identified the dominant protein ( based on total peptide counts ) as TbGSK3short ( Tb927 . 10 . 13780 ) , a CDXG kinase previously found to be essential in BSF T . brucei ( Ojo et al . , 2008 ) . Peptides from three other CDXG kinases were also detected in the same gel band ( Supplementary file 1A ) . Despite revealing four distinct CDXG kinases , this experiment provided no information about their relative sensitivity to hypothemycin . Moreover , we suspected that abundant and/or hyper-reactive proteins may have bound 2 nonspecifically , obscuring specific , low-abundance targets in the gel-based analysis ( Figure 2D ) . To identify hypothemycin-binding proteins in a more comprehensive and quantitative manner , we employed a gel-free method with hypothemycin competition and isobaric mass tags , similar to previously reported methods for identifying small-molecule targets ( Bantscheff et al . , 2007; Huang et al . , 2009; Ong et al . , 2009 ) . Four lysate samples were treated in parallel with increasing concentrations of hypothemycin for 30 min ( 0 , 20 , 200 , or 1000 nM ) , followed by probe 2 ( 500 nM , 30 min ) . As before , covalently modified proteins were conjugated to biotin-azide and affinity purified . After trypsinization of the eluted proteins , each sample was derivatized with a unique iTRAQ reagent ( Isobaric Tag for Relative and Absolute Quantitation ) ( Ross et al . , 2004 ) and then pooled for fractionation and mass spectrometry analysis . Using this method , peptides corresponding to 10 protein kinases were identified , including three of the four kinases identified by in-gel digest . Four peptides correspond to two nearly identical TbCLK ( cdc2-like kinase ) genes , TbCLK1 ( Tb927 . 11 . 12410 ) and TbCLK2 ( Tb927 . 11 . 12420 ) , which could not be differentiated on the basis of the identified peptides . In total , we identified 11 potential kinase targets of 2 ( Supplementary file 1A ) . Of note , every protein kinase identified by probe 2 contains the CDXG motif , accounting for over half of the 21 CDXG kinases encoded in the T . brucei genome . To our knowledge , this represents the first unbiased identification of hypothemycin-binding proteins . Quantification of iTRAQ reporter ions revealed the extent of CDXG kinase labeling by 2 after pretreatment with increasing concentrations of hypothemycin ( Figure 3 ) . This experiment thus provides an estimate of each kinase’s sensitivity to hypothemycin . Fragmentation spectra of peptides derived from eight kinases provided iTRAQ reporter ions of sufficient intensity to quantify kinase recovery ( i . e . , labeling by 2 ) as a function of hypothemycin pretreatment ( Figure 3B , C ) . In addition to CDXG kinases , several other proteins ( e . g . , tubulin , heat shock proteins , ribosomal proteins ) were identified . However , hypothemycin pretreatment did not affect their recovery ( Figure 3B , Supplementary file 1A ) , which likely resulted from nonspecific , low-level modification by 2 and/or nonspecific adsorption to the avidin-agarose beads . 10 . 7554/eLife . 00712 . 005Figure 3 . Quantification of hypothemycin binding to CDXG kinases . ( A ) MS/MS fragmentation spectra of peptides from TbCLK1/2 , TbGSK3short , and β-tubulin after labeling with 2 , affinity purification , and trypsinization . Red text: residues bearing an iTRAQ tag; ‘m’: oxidized Met . iTRAQ reporter ions for each peptide are shown to the right ( m/Z 114 , 115 , 116 , and 117 correspond to lysate samples pretreated with 0 , 20 , 200 , or 1000 nM hypothemycin , respectively ) . ( B ) Normalized recovery values , based on iTRAQ quantification , for all identified protein kinases and 25 non-kinases ( blue lines ) with ≥3 unique peptides . ( C ) Recovery of protein kinases as a function of hypothemycin pretreatment . Values for each hypothemycin pretreatment condition are derived from mean iTRAQ reporter ion counts normalized to vehicle across all peptides identified ( ± SD ) . Asterisks indicate values derived from a single spectrum . DOI: http://dx . doi . org/10 . 7554/eLife . 00712 . 005 Among the CDXG kinases quantified , recovery of TbCLK1/2 was uniquely affected by pretreatment with 20 nM hypothemycin ( Figure 3 ) . Recovery of two additional kinases , TbGSK3short ( Tb927 . 10 . 13780 ) and TbMAPK2 ( Tb927 . 10 . 16030 ) , was reduced more than 50% by 200 nM hypothemycin , and recovery of all but one CDXG kinase was reduced by 1 μM . These data indicate a wide range of hypothemycin sensitivity , with TbCLK1/2 exhibiting somewhat greater sensitivity than TbGSK3short and TbMAPK2 . TbMAPK2 is not essential in BSF T . brucei but is required for proliferation of the procyclic ( insect host ) form ( Muller et al . , 2002 ) . TbCLK1 and TbCLK2 , which have identical kinase domains and only diverge in their N-terminal regions , are presumed to be equally sensitive to hypothemycin . We assessed the requirement for each of the 21 CDXG kinases for cell viability using RNA interference . This required the creation of 21 cell lines , each containing a stably integrated cassette under tetracycline control and designed to silence a unique CDXG kinase ( Wang et al . , 2000 ) . After induction of RNAi , cell proliferation was followed for 6 days . Consistent with previous results ( Ojo et al . , 2008 ) , we observed reduced viability after knockdown of TbGSK3short ( Figure 4A ) . Also consistent with previous studies , knockdown of MAP kinases TbMAPK2 ( Muller et al . , 2002 ) , Tb927 . 6 . 1780 ( Guttinger et al . , 2007 ) , and Tb927 . 6 . 4220 ( Domenicali Pfister et al . , 2006 ) had no effect on cell viability ( Figure 4B , Supplementary file 1B ) . 10 . 7554/eLife . 00712 . 006Figure 4 . RNAi analysis of CDXG kinases . ( A–D ) BSF T . brucei were stably transfected with the indicated RNAi constructs and induced with tetracycline ( triangles ) or left uninduced ( squares ) starting on day 0 . Cell density was measured every 24 hr and cumulative cell growth was plotted on a log scale ( mean ± SD , n = 3 ) . mRNA levels were measured by quantitative RT-PCR ( bar graphs ) in the absence ( – ) and presence ( + ) of tetracycline . DOI: http://dx . doi . org/10 . 7554/eLife . 00712 . 006 Our initial RNAi construct did not distinguish between TbCLK1 and TbCLK2 . Induction with tetracycline resulted in reduced proliferation and increased cell death ( data not shown ) . To test for nonredundant functions of TbCLK1 and TbCLK2 , we designed a set of RNAi constructs targeting the unique 5′-UTR of each gene . A lethal phenotype was only observed after knockdown of TbCLK1; knockdown of TbCLK2 had no obvious effect ( Figure 4C , D ) . Individual knockdown of the remaining CDXG kinases similarly had no effect on cell proliferation . In each case , knockdown of the corresponding mRNA was confirmed by quantitative PCR ( 55–80% reduction across 21 mRNAs , Supplementary file 1B ) . With the caveat that insufficient knockdown may underlie the lack of a proliferation defect in certain cases , these results suggest that TbGSK3short and TbCLK1 are the only essential CDXG kinases in BSF T . brucei . Remarkably , these kinases also appeared to be among the most sensitive to hypothemycin ( Figure 3 ) . Because both TbGSK3short and TbCLK1 reacted with nanomolar concentrations of hypothemycin in cell lysates and exhibited strong RNAi phenotypes , we focused on these CDXG kinases and assessed their sensitivity to hypothemycin in enzymatic assays and in living parasites . Full-length TbGSK3short and TbCLK1 were expressed and purified from E . coli . Hypothemycin inhibited TbCLK1 with an IC50 of 150 nM when pre-incubated for 30 min in the presence of 100 μM ATP , whereas inhibition of TbGSK3short required 30-fold higher concentrations under the same conditions ( Figure 5A ) . This is consistent with our chemoproteomic experiment in which 20 nM hypothemycin reduced subsequent labeling of TbCLK1/2 in lysates , while having little effect on TbGSK3short ( Figure 3 ) . The greater difference in IC50 values in the enzymatic assays is likely due to the presence of competing ATP , which had been depleted from cell lysates prior to the labeling experiments . 10 . 7554/eLife . 00712 . 007Figure 5 . Preferential inhibition of TbCLK1 by hypothemycin . ( A ) In vitro assays with recombinant TbGSK3short and TbCLK1 . Kinases were incubated with hypothemycin and 100 μM ATP for 30 min before initiating reactions with substrate and γ32P-ATP . Substrate phosphorylation was quantified and normalized to DMSO control ( mean ± SD , n = 3 ) . ( B ) T . brucei expressing HA-tagged TbGSK3short and TbCLK1 from their endogenous loci were incubated with hypothemycin for 5 hr , and then were either harvested or diluted 1:30 into drug-free media and counted after 24 hr ( bar graph , mean ± SD , n = 3 ) . Harvested cells were lysed , labeled with 2 , and submitted to click conjugation with biotin-azide . HA-tagged proteins were immunoprecipitated , resolved by SDS-PAGE , and analyzed by Western blotting for biotin and HA . DOI: http://dx . doi . org/10 . 7554/eLife . 00712 . 007 The significant difference in hypothemycin sensitivity of recombinant TbGSK3short and TbCLK1 , along with their differential sensitivity in lysates , prompted us to ask whether this difference is also observed in intact cells . To quantify hypothemycin binding without mass spectrometry , we generated a strain of BSF T . brucei that contains a C-terminal hemagglutinin ( HA ) tag inserted into the endogenous loci of both TbGSK3short and TbCLK1 ( hemizygous for each HA-tagged variant ) . HA-tagged TbGSK3short and TbCLK1 are easily distinguished on Western blots by their molecular weight . Cells in log-phase growth were treated with increasing concentrations of hypothemycin . After 5 hr , an aliquot of cells from each treatment condition was diluted 30-fold into fresh media , and cell density was measured after 24 hr . The remaining cells were harvested after hypothemycin treatment , and lysates from these cells were treated with 500 nM 2 , followed by click conjugation with biotin-azide . After immunoprecipitation , the HA-tagged kinases were analyzed for covalently attached 2 by Western blot with streptavidin detection . In control cells , TbCLK1 migrated as two major species ( likely corresponding to differentially phosphorylated forms , our unpublished results ) , which collapsed to a single band upon treatment with hypothemycin ( Figure 5B ) . Probe 2 labeled both TbCLK1 bands in the DMSO-treated cells , but labeling was sharply reduced in cells pretreated with hypothemycin . Moreover , the concentration dependence of this effect ( reduction of labeling ) correlated with the loss of cell viability; the vast majority of cells treated for 5 hr with 500 nM hypothemycin died . In contrast to its potent effect on TbCLK1 , hypothemycin had a negligible effect on HA-tagged TbGSK3short , even at the highest concentration tested . Taken together , our results suggest that inhibition of TbCLK1 , rather than TbGSK3short , plays a more dominant role in mediating the cytotoxic effects of hypothemycin . New drugs are urgently needed to treat human African trypanosomiasis . While protein kinases are attractive targets , elucidating which of the 182 T . brucei kinases to prioritize for drug discovery efforts is nontrivial . Many essential T . brucei kinases show high sequence identity with human orthologs , frustrating attempts to obtain selective inhibitors . Moreover , once a kinase target has been validated genetically and subjected to high-throughput biochemical screens , demonstrating that a given inhibitor engages the endogenous kinase in intact cells is difficult or impossible; to our knowledge , this has not been accomplished for any T . brucei kinase inhibitor . To begin to address these challenges , we have applied a combination of pharmacology , quantitative chemoproteomics , and reverse genetics to the CDXG kinases , a focused yet phylogenetically diverse subset of all eukaryotic kinomes . We initially discovered that hypothemycin and the propargyl ether derivative 2 are potently trypanocidal . Although hypothemycin reduced parasitemia and cured one-third of infected mice , it was toxic at higher doses , and we do not consider it to be a drug lead . Hypothemycin-mediated toxicity may be caused by inactivation of MEK and VEGFR2 , as previous studies have found toxic effects associated with selective inhibitors of these kinases ( Rugo et al . , 2005; Adjei et al . , 2008 ) . Trypanosomes lack orthologs of most hypothemycin-sensitive mammalian kinases , including MEK and VEGFR2 . Conversely , the majority of CDXG kinases in T . brucei lack clear human orthologs . It was therefore imperative to develop an unbiased and quantitative method for identifying direct targets of hypothemycin in T . brucei . Of 21 predicted CDXG kinases , 11 were identified by probe 2 after affinity purification from cell lysates . It is likely that the remaining CDXG kinases are not efficiently labeled by 2 or are not expressed in BSF T . brucei . Mass spectrometry-based quantification revealed the extent of CDXG kinase occupancy as a function of hypothemycin concentration . We observed a wide spectrum of sensitivity , with most CDXG kinases requiring micromolar concentrations of hypothemycin to block labeling by probe 2 . Two of the kinases showing the greatest reduction in probe labeling after exposure to hypothemycin , TbGSK3short and TbCLK1 , are also the only CDXG kinases whose knockdown by RNAi significantly reduced cell growth . TbGSK3short was previously shown to be essential for proliferation and survival of BSF T . brucei ( Ojo et al . , 2008 ) . High-throughput screening with recombinant TbGSK3short has provided inhibitors with high biochemical potency , yet modest selectivity when tested against human GSK3β ( Oduor et al . , 2011; Urbaniak et al . , 2012 ) , consistent with the high sequence similarity shared by these kinases . Although many of these compounds block T . brucei proliferation , it is unclear whether they actually inhibit TbGSK3short in cells . Indeed , hypothemycin was able to saturate TbGSK3short at nanomolar concentrations in cell lysates , yet it had little effect on this kinase in intact cells , even after prolonged treatment at cytotoxic concentrations . Thus , despite TbGSK3short being the most prominent protein labeled by 2 in lysates , it is unlikely that hypothemycin kills T . brucei by inhibiting TbGSK3short . Nevertheless , we cannot exclude the possibility that partial inhibition of TbGSK3short or other CDXG kinases contributes to hypothemycin’s cytotoxic effects . Treatment of intact trypanosomes with cytotoxic concentrations of hypothemycin resulted in full occupancy of TbCLK1 . Together with the demonstration that TbCLK1 silencing by RNAi is sufficient to kill T . brucei , these results validate TbCLK1 as a druggable target . T . brucei and other kinetoplastids , including the human pathogens T . cruzi and Leishmania species , are the only organisms in which a CLK ortholog bears a CDXG motif . The kinase domains of TbCLK1 and human CLK1-4 share only 30% sequence identity . By contrast , T . brucei and human GSK3 share 52% sequence identity , including the CDXG motif . Obtaining inhibitors that discriminate between T . brucei and human CLK may therefore be more straightforward as compared to GSK3 . CLKs have been shown to regulate pre-mRNA splicing in fission yeast ( Tang et al . , 2011 ) , flies ( Du et al . , 1998 ) , and human cells ( Muraki et al . , 2004; Karlas et al . , 2010; Fedorov et al . , 2011 ) , and recent studies have revealed important splicing-independent roles ( Rodgers et al . , 2010 , 2011; Lee et al . , 2012 ) . Further work will be required to elucidate the functions and substrates of TbCLK1 in BSF T . brucei . By employing 2 as an affinity probe , it should now be possible to quantify engagement of both TbGSK3short and TbCLK1 by compounds derived from high-throughput screening campaigns , even without knowledge of their downstream substrates or signaling pathways . We note that TbCLK1 was not among the 57 kinases identified in T . brucei cell lysates using broad-spectrum kinase inhibitors immobilized on ‘Kinobeads’ ( Urbaniak et al . , 2012 ) . Although Kinobeads and other powerful chemoproteomic platforms ( Patricelli et al . , 2011 ) have the advantage of profiling a larger swath of the kinome , a potential disadvantage is that the probes only work in cell lysates and cannot be used to quantify kinase occupancy in intact parasites . More broadly , probe 2 and the chemoproteomic methodology described herein can be used to explore the related CDXG kinomes of T . cruzi and Leishmania , as well as uncharted kinomes in other disease-causing eukaryotes , all of which express diverse CDXG kinases . Finally , many human CDXG kinases are validated or potential drug targets ( e . g . , MEK , ERK , VEGFR2 , c-KIT , FLT3 , PDGFR , TAK1 , MNK ) , and probe 2 may prove valuable in quantifying their intracellular engagement by candidate inhibitors . BSF T . brucei ( strain 221 ) was cultured at 37°C and 5% CO2 in HMI-9 medium with 10% Fetal Bovine Serum , 10% Serum Plus ( Sigma-Aldrich , St . Louis , MO ) , 100 units/ml penicillin , and 100 μg/ml streptomycin . The 90-13 cell line was similarly cultured with addition of 2 . 5 μg/ml G418 , 5 . 0 μg/ml hygromycin . Transgenic cell lines were maintained in medium supplemented with 5 . 0 μg/ml hygromycin , 2 . 5 μg/ml phleomycin , and/or 0 . 1 μg/ml puromycin . The endpoint luciferase-based assay for quantifying cell density was performed as previously described ( Mackey et al . , 2011 ) with the following modifications . BSF T . brucei cells were treated with hypothemycin at 5 × 105 cells/ml in media lacking antibiotics . Cell density was measured after 24 hr using CellTiter-Glo Luminescent Cell Viability Assay ( Promega , Madison , WI ) . Luminescence was measured using a SpectraFluor Plus luminometer ( Tecan , San Jose , CA ) . Results were plotted using GraphPad Prism ( GraphPad Software , La Jolla , CA ) . Adult female Balb/c mice ( Charles River Laboratories , Wilmington , VA ) weighing 18–22 g were infected via intraperitoneal injection with 103 BSF parasites in 100 μl of PBS containing 1% glucose . One day post-infection , hypothemycin ( ChemieTek , Indianapolis , IN ) in 60% DMSO/water was administered once daily via intraperitoneal injection for 7 days . Mice were monitored every 48 hr for parasitemia in tail blood and visually inspected for general health . Surviving aparasitemic mice at day 30 were considered cured . Experiments were carried out in accordance with protocols approved by the Institutional Animal Care and Use Committee at the University of California , San Francisco . 4-O-desmethylhypothemycin ( Wee et al . , 2006 ) ( 10 mg , 0 . 0274 mmol ) was added to a dried glass reaction vessel equipped with a magnetic stir bar and dissolved in dry acetonitrile ( 6 ml ) . Propargyl bromide ( 69 μL 80% in toluene , 0 . 55 mmol ) and Cs2CO3 ( 10 . 7 mg , 0 . 033 mmol ) were added . After 6 hr at rt , the mixture was concentrated , dissolved in minimal methylene chloride , and purified by preparative silica TLC using 2% then 3 . 5% methanol in methylene chloride . Pure 2 ( 4 . 5 mg , 41% yield ) was eluted from the silica with 10% methanol in methylene chloride . NMR ( 1H , DMSO , 400 MHz ) : 0 . 98 ( dd , J=14 . 9 Hz , J=9 . 4 Hz , 1H ) , 1 . 36 ( d , J=6 . 2 Hz , 3H ) , 1 . 86 ( dd , J=14 . 0 , 9 . 8 Hz , 1H ) , 2 . 55 ( m , 1H ) , 2 . 76 ( m , 1H ) , 2 . 94 ( dt , J=17 . 3 , 10 . 9 Hz , 1H ) , 3 . 61 ( t , J=2 . 4 Hz , 1H ) , 3 . 87 ( m , 1H ) , 4 . 32 ( d , J=1 . 7 Hz , 1H ) , 4 . 45 ( dd , J=5 . 1 , 1 . 5 Hz , 1H ) , 4 . 83 ( d , J=2 . 4 Hz , 2H ) , 4 . 93 ( d , J=5 . 1 Hz , 1H ) , 5 . 15 ( d , J=6 . 6 Hz , 1H ) , 5 . 39 ( m , 1H ) , 6 . 11 ( dt , J=11 . 2 , 2 . 6 Hz , 1H ) , 6 . 32 ( d , J=2 . 8 Hz , 1H ) , 6 . 42 ( dd , J=11 . 7 , 2 . 7 Hz , 1H ) , 6 . 51 ( d , J=2 . 8 Hz , 1H ) , 11 . 90 ( s , 1H ) . NMR ( 13C , DMSO , 400 MHz ) : 21 . 02 , 34 . 24 , 36 . 77 , 56 . 39 , 57 . 29 , 63 . 72 , 69 . 54 , 74 . 83 , 79 . 12 , 79 . 51 , 81 . 84 , 102 . 37 , 104 . 00 , 105 . 66 , 128 . 59 , 143 . 10 , 143 . 21 , 162 . 82 , 165 . 09 , 171 . 00 , 201 . 85 . HRMS: predicted [M+H+] m/z 403 . 1387; measured 403 . 1383 . BSF T . brucei at a density of 106—5 × 106 cells/ml were collected and washed twice with PBS . The pellet was suspended in PBS containing Complete Protease Inhibitor Cocktail ( Roche , Basel , Switzerland ) and PhosStop Phosphatase Inhibitor Cocktail ( Roche ) and sonicated on ice . Cellular debris was pelleted , and the supernatant was passed through a NAP-5 column ( GE Healthcare , Buckinghamshire , United Kingdom ) equilibrated with lysis buffer . Whole-cell lysates ( 18 . 75 µl , 1–3 mg/ml protein ) were treated with increasing concentrations of hypothemycin for 30 min , followed by the indicated concentration of 2 for 30 min . Samples were denatured with SDS ( 1 . 25 µl , 20% ) , followed by addition of TAMRA-azide ( 0 . 5 µl , 5 mM ) , TCEP ( 0 . 5 µl , 50 mM , pH ∼7 . 0 ) , TBTA ligand in 1:4 DMSO:tert-butyl alcohol ( 1 . 5 µl , 1 . 67 mM ) , and CuSO4 ( 0 . 5 µl , 50 mM ) . Reactions were incubated at rt for 1 hr , resolved by SDS-PAGE , scanned for fluorescence ( Typhoon Imaging System , Molecular Dynamics , Sunnyvale , CA ) , and stained with Coomassie . T . brucei whole-cell lysates ( 14 mg , 3 mg/ml ) were treated with 0 ( DMSO control ) , 20 , 200 , or 1000 nM hypothemycin for 30 min , followed by 2 ( 500 nM , 30 min ) . Samples were then denatured by adding 20% SDS to a final concentration of 1% . Base-cleavable biotin-azide ( Choy et al . , 2013 ) was added ( 100 μM ) , followed by the remaining click reagents at the concentrations described above . After 1 hr at rt , proteins were precipitated with cold acetone ( 80% vol/vol final ) . Protein pellets were washed three times with acetone and dried . Pellets were resuspended in a minimal volume of 1% SDS in 50 mM Tris pH 8 . 0 . The solution was then diluted with 9 volumes of 1% NP-40 in PBS and passed through a PD-10 column ( GE Healthcare ) , eluting with 1% NP-40 and 0 . 1% SDS in PBS . Avidin-agarose ( 30 µl , Sigma-Aldrich ) was added and the samples were rotated overnight at 4°C . The beads were washed twice for 1 hr at rt with 1% NP-40 , 0 . 1% SDS in PBS; twice for 1 hr at 4°C with 6 M urea in PBS; 1 hr with PBS at 4°C; and 1 hr with PBS at rt . To facilitate the quantitative release of captured proteins , the ester linkage ( Choy et al . , 2013 ) was hydrolyzed with NaOH ( 0 . 4 N , 20 µl ) . After 20 min at room temperature , the solution was neutralized with HCl ( 0 . 8 N , 10 µl ) , brought to 1% SDS , and heated to 90°C for 3 min before collecting the supernatant . Eluted proteins were acetone-precipitated and pellets were washed twice with additional cold acetone . Dried pellets were resuspended in 8 M guanidinium HCl , reduced ( 5 mM TCEP , 50 mM ammonium bicarbonate , 6 M guanidinium HCl ) , and alkylated with 10 mM iodoacetamide . The solution was adjusted to 1 M guanidinium HCl , 100 mM ammonium bicarbonate , 10% acetonitrile and trypsinized overnight . Volatiles were removed , and the samples were resuspended in 0 . 1% formic acid , extracted with C18 OMIXtips ( Varian , Palo Alto , CA ) , and eluted with 50% acetonitrile , 0 . 1% formic acid . Volatiles were removed and peptides were resuspended in 0 . 5 M triethylammonium bicarbonate pH 8 . 5 and labeled with iTRAQ Reagents ( Applied Biosystems , Foster City , CA ) . Samples were mixed , dried , and resuspended in 30% acetonitrile , 5 mM potassium phosphate ( pH 2 . 7 ) . The samples were then fractionated on a 50 . 0 × 1 . 0 mm 5 µM 200 Å Polysulfoethyl A column ( PolyLC , Colombia , MD ) with a 1–40% gradient of 350 mM potassium chloride in 30% acetonitrile , 5 mM potassium phosphate ( pH 2 . 7 ) . Peptide-containing fractions were dried , resuspended in 0 . 1% formic acid , extracted using C18 ZipTips ( Millipore , Billerica , MA ) , eluted with 50% acetonitrile , dried , resuspended in 0 . 1% formic acid , and analyzed by LC/MS/MS . Fractionated tryptic peptides were separated by nano-flow liquid chromatography using a 75 µm × 150 mm reverse phase C18 PepMap column ( Dionex-LC-Packings , Sunnyvale , CA ) at a flow rate of 350 nl/min in a NanoLC-1D Proteomics high-performance liquid chromatography system ( Eksigent Technologies , Dublin , CA ) equipped with a FAMOS autosampler ( Dionex-LC-Packings ) . Peptides were eluted using a 2–30% gradient of acetonitrile/0 . 1% formic acid over 40 min , followed by 50% acetonitrile for 3 min . The eluate was coupled to a microionspray source attached to a QSTAR Elite mass spectrometer ( Applied Biosystems/MDS Sciex , Framingham , MA ) . Peptides were analyzed in positive ion mode . MS spectra were acquired for 1 sec in the m/z range between 350 and 1500 . MS acquisitions were followed by 4 × 2 . 5 sec collision-induced dissociation ( CID ) experiments in information-dependent acquisition mode . For each MS spectrum , the two most intense multiple charged peaks over a threshold of 25 counts were selected for generation of CID mass spectra . Two MS/MS spectra of each were acquired , first on the m/z range 119–1500 , with Q1 resolution set at ‘low’ , and then on the m/z range 112–119 , with resolution set at ‘unit’ . The CID collision energy was automatically set according to mass to charge ( m/z ) ratio and charge state of the precursor ion . A dynamic exclusion window was applied which prevented the same m/z from being selected for 1 min after its acquisition . Peak lists were generated using the mascot . dll script . The peak list was searched against the Trypanosoma subset of the NCBInr database using ProteinProspector . A minimal ProteinProspector protein score of 15 , a peptide score of 15 , a maximum expectation value of 0 . 1 , and a minimal discriminant score threshold of 0 . 0 were used for initial identification criteria . iTRAQ modification of the amino terminus or the epsilon-amino group of lysines , carbamidomethylation of cysteine; acetylation of the N-terminus of the protein and oxidation of methionine were allowed as variable modifications . Peptide tolerance in searches was 100 ppm for precursor and 0 . 2 Da for product ions , respectively . Peptides containing two miscleavages were allowed . The number of modifications was limited to two per peptide . Quantification was based on the relative areas of the reporter ions ( m/z =114 , 115 , 116 and 117 ) generated by the isobaric iTRAQ reagents during CID experiments . Abundance ratios of individual peptides between the different samples were calculated taking as reference the DMSO-treated sample , by dividing the areas of their respective iTRAQ reporter ions . Peptides with peak areas lower than 30 for the most intense reporter ion were discarded . For changes in relative abundance at the protein level , all MS/MS spectra for the different peptides belonging to a particular protein were used to calculate the average and SD of the abundance ratios . Plasmid DNA ( 10 μg ) was linearized ( MfeI , TbGSK3short-HA; Bsu36I , TbCLK1-HA; NotI , pZJM constructs ) and resuspended to 1 μg/μl in water . Cells ( 2 × 107 ) were electroporated using the Human T Cell Nucleofector kit and the Amaxa program X-001 ( Lonza , Basel , Switzerland ) , then diluted in media and allowed to recover for 24 hr before selection for 5–7 days with the appropriate antibiotic . A 250–500 bp fragment of each CDXG kinase gene was amplified from genomic DNA by PCR ( primers listed in Supplementary file 1C ) and cloned into the XhoI/HindIII site of pZJM ( Wang et al . , 2000 ) . Plasmid DNA was stably transfected into BSF T . brucei 90–13 cells as described above . RNAi was induced with 1 μg/ml tetracycline . Parasites were counted every 24 hr using a Coulter Counter ( Beckman Coulter , Brea , CA ) , and cultures were diluted to maintain cell density between 105 and 2 × 106 cells/ml . RNA was extracted 48 hr after tetracycline induction using TRIzol ( Invitrogen ) and RNeasy kit ( Qiagen , Hilden , Germany ) . For qRT-PCR quantification ( relative to TbGAPDH , Tb927 . 6 . 4300 ) , cDNA was generated with the High Capacity RNA-to-cDNA Kit ( AppliedBiosystems ) and quantified using the PowerSYBR kit ( AppliedBiosystems ) . Full-length TbGSK3short and TbCLK1 were amplified from genomic DNA using primers listed in Supplementary file 1C and cloned into the pET100 expression vector ( Invitrogen ) . Proteins were expressed in E . coli ArcticExpress ( DE3 ) ( Stratagene , La Jolla , CA ) in ZY5052 medium at 20°C for 60 hr with agitation . Cells were lysed using a microfluidizer into lysis buffer ( 50 mM Tris pH 8 , 300 mM NaCl , 10 mM imidazole , 5% glycerol , 1 mM CaCl2 , 1 mM MgCl2 , 500 nM PMSF , 1× Protease inhibitor cocktail ( Roche ) , DNase ( Sigma-Aldrich ) , lysozyme ( Sigma-Aldrich ) . The soluble fraction was isolated by centrifugation and incubated overnight at 4°C with Ni-NTA beads . The recombinant protein was eluted with lysis buffer containing 250 mM imidazole and dialyzed into storage buffer ( 30 mM Tris pH 7 . 5 , 50 mM NaCl , 50% glycerol , 5 mM DTT , 1 mM EDTA , 0 . 03% Brij35 ) . Aliquots of the proteins were flash frozen with liquid nitrogen and stored at –80°C . Kinases ( 5 nM TbGSK3short , 10 nM TbCLK1 ) were incubated with hypothemycin in reaction buffer ( 50 mM Tris pH7 . 4 , 10 mM MgCl2 , 0 . 2 mM EGTA , 0 . 2 mg/ml BSA , 1 mM DTT ) and 100 µM ATP for 30 min at rt . A solution of reaction buffer with γ32P-ATP ( 70–150 μCi/ml , Perkin Elmer , Waltham , MA ) and GSM peptide substrate ( TbGSK3short , 0 . 05 mg/ml , Millipore ) or myelin basic protein ( TbCLK1 , 0 . 5 mg/ml , Sigma-Aldrich ) was added to initiate the kinase reaction . After 15 ( TbGSK3short ) or 30 min ( TbCLK1 ) , reactions were spotted onto phosphocellulose paper , washed once with 10% acetic acid , twice with 1% phosphoric acid , and once with methanol , and dried . Kinase activity was quantified using a Typhoon Imaging System ( Molecular Dynamics ) and ImageQuant 5 . 2 ( Molecular Dynamics ) . IC50 values and dosing curves were generated using GraphPad Prism 5 ( GraphPad Software ) . TbGSK3short or TbCLK1 lacking start and stop codons were amplified from genomic DNA using primers listed in Supplementary file 1C and inserted between the KpnI and XhoI sites of pC-PTP-PURO , which was modified to contain a C-terminal HA-tag ( Schimanski et al . , 2005; Gourguechon and Wang , 2009 ) . For the TbCLK1-HA construct , the puromycin resistance cassette was replaced with a hygromycin resistance gene . T . brucei 221 cells were first transfected with the TbGSK3short-HA construct to establish a puromycin resistant cell line in which the vector stably integrated into the endogenous TbGSK3short locus via homologous recombination . This cell line was then transfected with the TbCLK1-HA construct and selected for hygromycin and puromycin resistance . Cells stably expressing HA-tagged TbGSK3short and TbCLK1 ( 5 × 106 cells/ml ) were treated with the indicated concentrations of hypothemycin . After 5 hr , aliquots of the cultures were diluted 30-fold with fresh media and cell density was quantified after an additional 24 hr . The remaining cells were pelleted , washed with trypanosome dilution buffer ( 5 mM KCl , 80 mM NaCl , 1 mM MgSO4 , 20 mM glucose , 22 mM sodium phosphate pH 7 . 7 ) and lysed by sonication in 50 μl PBS containing 0 . 25% NP-40 , 1× Complete EDTA-free protease inhibitor cocktail , and 1× PhosStop phosphatase inhibitors ( Roche ) . Lysates were clarified by centrifugation , normalized for protein content , and treated with 500 nM 2 for 30 min . Samples were subjected to click conjugation with biotin-azide as described above and diluted 10-fold with 1 . 1% NP-40 in PBS . HA-tagged proteins were immunoprecipitated using 12CA5 anti-HA antibody ( Roche ) and Protein A Dynabeads ( Invitrogen ) , eluted with sample buffer , resolved by SDS-PAGE , and transferred to nitrocellulose membranes . HA and biotin were detected with anti-HA ( 1:1000 , Sigma Aldrich , H6908 ) , IRDye ( 680 ) -conjugated anti-rabbit ( 1:10 , 000 ) , IRDye ( 800 ) -conjugated streptavidin ( 1:20 , 000 ) using the Odyssey Imaging System ( Li-Cor Biosciences , Lincoln , NE ) .	Human African trypanosomiasis—commonly known as sleeping sickness—is a debilitating and potentially fatal tropical disease that is widespread in sub-Saharan Africa . It is caused by the single-celled parasite Trypanosoma brucei , which is transmitted to humans by the bite of the tsetse fly . The infection takes its name from the disruption of the circadian clock that occurs early on in the disorder and leads to sleep disturbances . If left untreated , T . brucei infection leads to coma , organ failure and death . Most of the existing pharmaceutical treatments for sleeping sickness were developed more than 50 years ago . However , they are only weakly absorbed into the bloodstream—meaning that high doses must be used—and they lead to unpleasant side effects . Moreover , the T . brucei parasite is developing resistance to existing drugs , so further research is needed to identify new therapeutic targets . One promising option could be the parasite’s protein kinases . These enzymes , which add phosphate-based chemical groups to proteins , have a key role in regulating protein function and many of them are already being investigated as therapeutic targets for cancers and autoimmune diseases . T . brucei has 182 different kinases , suggesting a wealth of potential new targets . However , many of these are similar to human enzymes , and inhibiting the latter could lead to harmful side effects . Now , Nishino et al . have produced a synthetic version of a microbially derived kinase inhibitor , called hypothemycin , and have shown that it kills T . brucei cells grown in culture . Hypothemycin also killed T . brucei in infected mice , completely curing the infection in one third of animals , although high doses of the drug led to side effects . Using a chemical biology approach and quantitative mass spectrometry , Nishino et al . found that the main target of hypothemycin was a previously unknown kinase that is essential for T . brucei survival . Although hypothemycin itself is probably unsuitable as a treatment due to its lack of specificity , the work of Nishino et al . suggests that its kinase targets deserve further investigation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2013	Hypothemycin, a fungal natural product, identifies therapeutic targets in Trypanosoma brucei
Using serial block-face scanning electron microscopy , we report on the internal 3D structures of the brown planthopper , Nilaparvata lugens ( Hemiptera: Delphacidae ) at nanometer resolution for the first time . Within the reconstructed organs and tissues , we found many novel and fascinating internal structures in the planthopper such as naturally occurring three four-way rings connecting adjacent spiracles to facilitate efficient gas exchange , and fungal endosymbionts in a single huge insect cell occupying 22% of the abdomen volume to enable the insect to live on plant sap . To understand the muscle and stylet movement during phloem sap-sucking , the cephalic skeleton and muscles were reconstructed in feeding nymphs . The results revealed an unexpected contraction of the protractors of the stylets and suggested a novel feeding model for the phloem sap-sucking . The morphology and arrangement of organs in the insect body have been studied as early as in the nineteenth century ( e . g Lowne , 1892 ) . These data are still used in current textbooks and atlases of insect morphology and physiology . However , the internal structures and their spatial relationship lack a direct three-dimensional ( 3D ) representation . In recent years , morphological studies using micro-computed tomography ( micro-CT ) served to reveal the internal structures of some insect tissues , such as the copulating Drosophila ( Mattei et al . , 2015 ) or even of the entire organism such as the Drosophila pupa ( Schoborg et al . , 2019 ) . However , for small insects and delicate structures , the resolution of micro-CT is insufficient to distinguish details . Serial block-face scanning electron microscopy ( SBF-SEM ) allows visualization of the insect’s internal structure at the nanometer resolution in 3D . This method relies on reiterated imaging and sectioning of resin embedded sample using a robotic ultramicrotome and digital assembly of the single images to 3D images ( Denk and Horstmann , 2004 ) . Several 3D structures of insect tissues have been resolved by SBF-SEM , such as pericardial nephrocytes in Drosophila ( Kawasaki et al . , 2019 ) , and cyst-like bodies formed by Trypanosoma brucei in the anterior midgut of tsetse flies ( Rose et al . , 2020 ) . Insects of small size are suitable for this technology to obtain an integrated 3D atlas of the whole body with a resolution higher than possible by any other previous method . Resolving the internal structures and their spatial relationship is of great value in understanding the relationship of structure and function , and in conducting physiological , biochemical , and molecular studies in this direction . Although reconstruction of a 3D atlas of the whole body of any insect by EM method is valuable and theoretically possible , it has not been conducted yet . Hemipteran insects are in general small and possess delicate mouthparts with straw-like stylets that are highly specialized for penetrating plant tissues . For virus vectors from the order of Hemiptera , including small/tiny insects like planthoppers , leafhoppers , psyllids , aphids , and whiteflies , studying their phloem sap-sucking mechanism will enable us to understand the way they injure crops and inject their saliva with transmitted viruses into crop tissues . Knowledge about their internal structures such as the mouthparts and alimentary canal of the phloem sap-sucking Diaphorina citri ( Hemiptera: Liviidae ) ( Ammar et al . , 2017 ) , Psammotettix striatus ( Hemiptera: Cicadellidae ) ( Zhang et al . , 2012 ) , and the aphid Acyrthosiphon pisum ( Hemiptera: Aphididae ) ( Guschinskaya et al . , 2020 ) is restricted to studies that relied on some discrete images by either light , confocal , or electron microscopy . Study of phloem–sucking insect/pathogen interactions represents an exciting frontier of plant science ( Jiang et al . , 2019 ) . When sucking in plant sap , hemipterans are proposed to protrude their straw-like stylets into the plant tissue by contracting the respective muscles ( Leopold et al . , 2003; Beardsley and Gonzalez , 1975 ) , but an evidence of the muscle movement has not been provided . Indeed , the commonly assumed feeding mechanism for phloem sap-sucking insects was largely speculative instead of being based on the actual movement of muscles and the feeding apparatuses . The rice sap-sucking brown planthopper ( BPH ) , Nilaparvata lugens ( Stål ) ( Hemiptera: Fulgoroidea: Delphacidae ) , is the most important pest of the rice plant . They damage rice not only by directly feeding and ovipositing on it but also by transmitting two rice viruses , rice rugged stunt virus and rice grassy stunt virus . Dense BPH infestations can cause complete drying and wilting of rice plants , a condition known as ‘hopperburn’ . Since the late 1960s , BPH outbreaks have been re-occurring approximately every 3 years in Asian countries , with their annual outbreak area amounting to approximately 10–20 million hectares of rice , resulting in millions of tons of losses . BPHs seasonally migrate from their overwintering rice fields in tropical Asia northward to temperate areas in China , Korea , Japan , and northern India when rice becomes available , inflicting damage in an even larger area across Asia ( Perfect and Cook , 1994 ) . BPH has been a model insect for ecological research and a model insect for hemipteran insects in biological research in recent years . In this study , we applied SBF-SEM to investigate the structure of various systems involved in food ingestion , digestion , locomotion , gas exchange and perception , and muscle movement during the feeding process of the nymph . A detailed 3D structure for a whole insect is presented for the first time , and a plant phloem sap-sucking model based on the actual movement of muscles and all feeding apparatuses during feeding process is proposed . The length of the insect measures approximately 600 µm ( Figure 1 ) . We reconstructed the 3D structure of the whole insect body ( Figure 1 , Figure 1—video 1 , Figure 1—video 3 , interactive 3D PDF file ) , including the organ systems of food ingestion and digestion , locomotion , gas exchange , and perception . We discover several novel and fascinating internal structures in BPH , such as 24 neuropils in the central nerve system ( Figure 1—figure supplement 1 ) , three four-way rings connecting the spiracles in adjacent segments to facilitate efficient gas exchange ( Figure 2D , H–J ) , and fungal endosymbionts in a single huge mycetocyte occupying 22% of the abdominal volume to enable the host to thrive on a low-nutrient diet provided solely by rice ( Figure 2G , Figure 2—figure supplement 2D , Table 2 ) . The alimentary canal starts with specialized mouthparts and includes the esophagus , the anterior diverticulum ( AD ) ( Figure 2—figure supplement 1D ) , the midgut ( Figure 2—figure supplement 1E ) , and the hindgut ( Figure 2—figure supplement 1F ) and ends with the anus ( Figure 2—figure supplement 1G ) . The anterior diverticulum ( AD ) extends anteriorly into the head ( Figure 2A , B , Interactive 3D-PDF ) . The midgut is divided into an anterior sac and a midgut loop ( Figure 2—figure supplement 1A , E ) . The midgut loop is a narrow tube coiling into a cluster of loops . The anterior part of the loop coils into the inner loop , and the outer loop has its posterior end sticking into the center of the cluster . After that , the midgut descends posteriorly to fuse with the Malpighian tubules ( Figure 2B ) . The inner surface of the whole midgut is covered by a dense layer of microvilli , especially in the loop region where the gut lumen is almost constricted ( Figure 2—figure supplement 1E ) . There is no membrane or sheath enveloping the midgut loop , and the filter chamber , which is a common structure in sap-sucking Hemipteran insects , is not found ( Figure 2—figure supplement 4 ) . The anterior midgut contacts the end of the loop region at one point where the epithelia of the anterior midgut become thinner than in the neighboring area ( Figure 2—figure supplement 1E ) . The midgut loop at position ‘3’ is connected to the hindgut . The 3D reconstruction reveals the distribution of three structurally different endosymbionts in the nymph . The most generally distributed endosymbiont is the Ascomycete fungus Entomomyces delphacidicola ( Fan et al . , 2015 ) , also known as the yeast-like symbionts ( YLSs ) , which have an ellipsoidal shape from a few microns to over 10 microns ( Figure 2G , Figure 2—figure supplement 2D ) . Most of the YLS reside in a huge fat body mycetocyte that surrounds the midgut , taking up 22% of the abdominal volume and 4 . 2% of the total body volume ( Table 2 ) . A few YLSs are dispersed in the body cavity including the head and legs . Another two fat body mycetocytes adjacent to the midgut ( Figure 2C , F ) are full of thread-like bacterial endosymbionts . The anterior diverticulum lumen connects the midgut lumen though a narrow region ( Figure 2—figure supplement 1B ) . Anterior to this narrow region , in the wall we found 10 hollow cells that each accommodates numerous rod-like bacterial endosymbionts ( Figure 2C , E , Figure 2—figure supplement 2A , B ) , whereas in the anterior diverticulum lumen there are no endosymbionts . These cells seem to be the specialized structures to host the symbiont bacteria . 16S rDNA sequence of the AD showed that 95% of the symbionts are Arsenophonus species ( Figure 2—figure supplement 2E ) . The volume of total tracheae in the first instar nymph was 24 , 000 µm3 , accounting only for about 0 . 1% of the body volume ( Table 2 ) . Two pairs of thoracic spiracles ( ts1 and ts2 ) , eight pairs of abdominal spiracles ( as1–as8 ) , and tracheae extending from the spiracles make up the network ( Figure 2D ) . Every abdominal spiracle is embedded in a pitting called the atrium with two valves , one of which separated the trachea and the atrium , while the other one divides the atrium into two parts ( Figure 2—figure supplement 3 ) . All the spiracles are connected by dorsal longitudinal trunks ( dlt ) ( Figure 2D ) . At the ventral side , there are no longitudinal trunks . Three four-way tracheal rings at the ventral side connect the adjacent two spiracles and their counterparts at the other side of the body , like a traffic circle ( Figure 2D , H–J ) . The ts2–as1 four-way tracheal ring and the as7–as8 four-way tracheal ring connect the spiracles in adjacent segments , as well as the left and right spiracles in the same segment . The as1 four-way tracheal ring connects two tracheae branches from the ts1 spiracle and their symmetrical counterparts at the opposite side . In addition , we applied the SBF-SEM technique to explore the mechanism of phloem sap-sucking of N . lugens . Here , we focus on the sap-sucking process and the organs involved in food uptake . The external morphology of the mouthparts in different Hemiptera is roughly identical . The beak is formed of the labium under a conical labrum ( Figure 3E , Figure 3—figure supplement 1C , Figure 3—figure supplement 2A , B ) . The labium is 130 μm in length with three segments , but only the second ( L2 ) and third segments ( L3 ) are seen from the ventral side , as the first segment ( L1 ) is covered by the labrum ( Figure 4A , Figure 3—figure supplement 2A , B ) . A longitudinal groove on the ventral surface of the labium accommodates the stylet bundle , resulting in a longitudinal lumen . The stylet bundle is in this lumen ( Figure 3F ) . During the feeding process , the stylet bundle protrudes from the apex of the labium ( Figure 4B , Figure 3—figure supplement 2A–C ) . The muscles in the labium include three pairs of retractors of the third segment ( RT1–RT3 ) , four pairs of lockers of the stylets ( LS1–LS4 ) , two pairs of rotators of the second segment ( RS1 and RS2 ) , and a pair of rotators of the first segment ( RF ) ( Figure 3E , Interactive 3D PDF ) . RT1 originates from the lateral region of the L2 and inserts into the floor of the labium groove in the L3 . RT2 and RT3 both insert into the apodeme of the L3 . RT2 originates on the dorsal wall of the L2 , while RT3 originates from the lateral wall of the L2 . The four pairs of lockers of the stylets have their origins in the dorsal region of the second labium and insert into the floor of the labium groove on the same segment . RS1 and RS2 originate from the dorsal and lateral regions of the L1 , respectively , and both insert into the apodeme of L2 ( Figure 3E , Figure 3—figure supplement 1C ) . The stylet bundle consists of two mandibular stylets and two maxillary stylets . They emerge from the stylet bases ( Figure 3D ) . As the stylets enter the groove , they become adherent to one another forming a compact bundle . The two maxillary stylets are interlocked by longitudinal ridges and grooves ( Figure 3F ) . The food canal and the saliva canal are formed where the two maxillary stylets meet . The thicker mandibular stylets lie laterally to the maxillary stylets ( Figure 3F , Figure 3—figure supplement 2C , D ) . They encircle the maxillary stylets for tight packaging and at the same time allowing themselves to slide along the longitudinal axis . Every stylet base is enclosed by a cuticular pouch ( MdPC and MxPC ) that partly wraps the sheath ( Figure 3A , C , D , Figure 3—figure supplement 1A ) . The mandibular pouch produces two long arms called the mandibular levers ( MdL ) ( Figure 3A , C , D ) . One of the levers proceeds dorsally , and at the end of the lever are the insertion of two retractor muscles . One ( RMdA ) originates from the dorsal wall of the head capsule , while the other one ( RMdB ) extends anteriorly . The other lever extends anteriorly to the mandibular plate . This lever supports most of its length a wide and thin epidermal inflection , on which the protractor of the mandibular stylet ( PMd ) is inserted ( Figure 3A , C , Figure 3—figure supplement 1A , B ) . The protractor of the mandibular stylet originates in the inner surface of the mandibular plate . The maxillary stylet arises from the maxillary pouch , which is dorsal to its mandibular counterpart . The retractor of the maxillary stylet ( RMx ) is inserted in the dorsal end of the pouch and has its origin on the dorsal wall of the head capsule , positioned lower than the origin of the RMd ( Figure 3C ) . On the lateral side of the pouch , we discerned the insertion point of the protractor of the maxillary stylet ( PMx ) , which originates on the inner surface of the maxillary plate ( Figure 3A ) . Sap-sucking requires the function of the salivary glands that are associated with the mouthparts . These paired organ lies in the anterior part of the thorax , lateral to the alimentary tract ( Figure 1A , B ) . On each side , a principle gland is connected with an accessary gland by the accessary duct ( Figure 5A ) . The efferent salivary ducts arise from the two principle glands and converge anteriorly to form the common salivary duct that transports saliva into the salivary syringe ( Figure 5D–I ) . The principle gland is acinar , with nine kinds of follicles ( A–I ) . It is digitate with the nuclei at the periphery of the cell . Follicles C–I are composed of several compact cells with no clear boundary or interspace between them ( Figure 5—figure supplement 1 ) . Therefore , cells are difficult to distinguish and count . In follicles A and B , by contrast , the boundaries between the cells are clear . Both include six binucleate cells ( A1–A6 and B1–B6 ) ( Figure 5B ) . The follicles are classified according to the density , size , and electron densities of the vesicles in the cells . Among the nine kinds of follicles , follicle D contains the smallest and the highest number of vesicles , while follicle E contains the lowest number of vesicles ( Figure 5—figure supplement 1 ) . Each one of the accessary glands is composed of two symmetrical follicles , and each follicle contains translucent vesicles ( Figure 5—figure supplement 1J ) . Different cell types suggest they may secrete different components of saliva . Ductules from each cell join up into the efferent salivary gland ( Figure 5—figure supplement 1L ) . Follicle H is on a separate branch from other follicles . The two branches merge with the accessary duct into the efferent salivary duct , which is a single layer of cells lined with a thin cuticle ( Figure 5—figure supplement 1K ) . A nerve bundle protrudes from the ventral pharyngeal sensory center , bypasses the accessary duct , penetrates follicles G and A successively , and then bifurcates to two branches at follicle A , finally they terminates in follicle H ( Figure 5B ) . The salivary syringe lies at the base of the hypopharynx ( Figure 5A ) . Its core component is a pocket-like structure , the salivarium ( Baptist , 1941 ) , which connects the common salivary duct , the reservoir , and the anterior salivary duct . A pair of dilators of the salivary syringe are attached on the lateral side of the reservoir . The extreme anterior end of the esophagus , the cibarial pump , is a sclerotized-walled canal with quadrangle cross sections ( Figure 3A ) . The cibarial pump lies vertically in the posterior part of the head . Its anterior wall invaginates into the lumen , which renders the lumen of the cibarial pump flexible and easy to expand . The dorsal dilators of the cibarial pump originate from the dorsal wall of the head and insert into the anterior wall of the cibarial pump . The lateral dilators of the cibarial pump insert into the lateral wall of the cibarium pump , with their origins on the anterior arm of the tentorium ( Figure 3A , Figure 3—figure supplement 1B ) . On the posterior wall of the cibarium pump is the third pair , that is , the posterior dilators of the cibarial pump that have their origins on the posterior arm of the tentorium ( Figure 3B ) . Only the tip of stylets pierce into the plant tissue during feeding . To figure out how the nymph pulls its stylet out of the labium , we investigated the musculature that might be involved in feeding process on two samples . In feeding modality , unexpectedly , the protractors of the stylets are shortened and retractors are all elongated ( Figure 4D , D’ , E , E’ ) . When the muscles contract to retract the stylet bases , the stylets stretch out of the labium ( Figure 4B ) . We observed that adult N . lugens lower their heads during the feeding process while the labium tip touches the surface of the plant ( Figure 6—video 1 ) . We suppose that N . lugens might compress its beak to stretch out the stylet . To test this , we measured the beak of several nymphs after freezing them in liquid nitrogen during feeding . The lengths of the beak ( lb ) and the protruding stylet ( lps ) were divided by the width of the head ( wh ) to eliminate a possible influence of different body sizes ( Figure 4A , B ) . The lb/wh ratio of the 19 feeding nymphs is 0 . 530 ± 0 . 065 ( mean ± SEM ) , significantly smaller than that of the 12 relaxed nymphs with the ratio of 0 . 801 ± 0 . 069 ( mean ± SEM , Figure 4C ) . The result agrees that the beak is shortened when feeding . The beak is supposed to shorten due to the contraction of the intersegmental muscles in the labium , including the retractors of L1 , the rotators of L2 , and the rotators of L3 . The total length of lb and lps in feeding nymphs divided by wh is 0 . 673 ± 0 . 082 ( mean ± SEM ) , significantly smaller than that in relaxing nymphs ( Figure 4C ) . Given that in relaxed individuals the stylets and the labium are of the same length , the result suggests that the stylets of feeding nymphs are shorter than those of relaxed nymphs , which agrees with the observation that stylets of feeding nymphs are pulled back by the retractors of the stylets . In this study , we applied SBF-SEM to assemble a complete reconstruction of the N . lugens nymph with a total length of 600 μm . Compared to micro-CT that was applied to resolve the internal structure of an insect as small as the coffee berry borer ( Hypothenemus hampei ) ( Alba-Alejandre et al . , 2019 ) , SBF-SEM were able to resolve smaller individuals at single cell-scale resolution . Although the majority of SBF-SEM data are collected in neuroscience studies and the sample preparation methods are optimized for brains and neuronal tissues ( Denk and Horstmann , 2004; Tapia et al . , 2012 ) , SBF-SEM is well suitable for entomological studies . SBF-SEM can handle large samples like insects of several hundred microns and small samples like collagen fibrils ( Starborg et al . , 2013 ) . If a higher resolution is desired in a certain sub-volume , SBF-SEM can be combined with some other volume EM imaging technique with higher resolution , such as FIB-SEM and TEMCA ( Guo et al . , 2020; Xu et al . , 2017; Zheng et al . , 2018 ) . However , the destructive cutting-and-imaging mode is a drawback when analyzing the sample with various techniques . For example , sensilla on the antenna of an insect can be analyzed by SBF-SEM imaging on the whole antenna , and the following higher resolution imaging on the selected sub-volume can be only conducted on another sample . The convenient serial sectioning SBF-SEM is also a method of choice for tracking neurons or muscles in studies on 3D structure of connective tissues . Compared with brains and tissue blocks , the waterproof exoskeleton of insects prevents normal fixation and staining . We tried inducing some wounds on the legs to allow rapid fixation and longer staining times to produce better contrasts in images and less damage to the sample . The full-pipeline of sectioning and imaging can generate large volumes of data automatically , while analyzing the images need tedious manual work . Need of time and experience on analyzing images can constrain broad usage of SBF-SEM . Further improvements such as deep learning algorithms for image segmentation and better alignment on defective images will reduce manual intervention in the next future . We have elucidated the musculature in the labium and the head of N . lugens both in relaxing and in feeding modalities . These results helped us to understand muscle movements during the feeding process . Based on reconstruction of the musculature , a novel feeding process model is proposed ( Figure 6 , Figure 6—video 2 ) : at the beginning , three pairs of retractors of L3 and two pairs of rotators of L2 contract to bend the labium in a way that the labium tip touches the plant surface vertically . At the same time , it is also shortened by contraction of the intersegment muscles . The retractors of the maxillary stylets and mandibular stylets also contract to withdraw the stylets into the labium and generates space for the first labial segment that is packed into the head capsule ( Figure 6B ) . Then the two protractors of the mandibular stylets contract alternately to pull out the stylets to pierce further and deeper into the plant tissue ( Figure 6C ) . The labium is compressed both by contracting the intersegment muscles and by packing the first labial segment into the head capsule . At the same time , the insect lowers its head to compress the labium , thereby the stylet bundle sticks further out of the labium . As the two mandibular stylets penetrate into the plant tissue alternately with the maxillary stylets following , all the four stylets reach the phloem tissue ( Figure 6D ) . Finally , the head is at a low position , and the first labial segment of the beak is partly retracted into the head capsule . Retractors of the stylets keep contraction to pull up the pouches to generate space for the first labial segment that is packed into the head capsule . The lockers of the stylets may continue contracting during this process . When they contract , the floor of the labium groove will move dorsally and the lateral side of the labium groove will approach . They are likely to lock the stylet bundle in the groove ( Spangenberg et al . , 2013 ) , so the stylet bundle may be restricted to move along the labium groove . As the stylets continue penetrating , saliva is secreted into the plant tissue in the following way ( Figures 4D–I and 6B ) . There is no muscle on the salivary duct or the salivary gland , so the salivary syringe is the only possible apparatus for saliva ejection . Its unique structures are modified for actively expelling the saliva . Dilators of the salivary syringe contract to expand the reservoir . Saliva produced in the salivary glands is pumped into the salivarium and the reservoir . When the dilator of the salivary syringe relaxes , the reservoir shrinks and squeezes out the saliva into the anterior salivary duct . The saliva then flows out of the stylets through the saliva canal . As the stylets penetrate into the plant tissue , dilators of the salivary syringe alternately contract and relax to pump out the saliva that includes gel saliva and watery saliva . When the gel saliva is released to the plant surface , it solidifies to generate a saliva interface between the plant surface and the labium tip to immobilize the labium . When it is released into the plant tissue , it diffuses surrounding the stylets and solidifies to a sheath that protects the insect against chemical defenses from the plant ( Huang et al . , 2015; Sōgawa , 1982 ) . The maxillary stylets can dig more into the plant tissue ( Figure 3—figure supplement 2 ) . There is some evidence that the tip of the maxillary stylet can explore a wide range in the tissue to find the phloem or a blood vessel for the blood-feeding species like Triatominae ( Hemiptera: Reduviidae ) ( Tull et al . , 2020 ) . Once the maxillary stylets reach the phloem , the sucking process starts . Three pairs of dilator muscles , the dorsal , lateral , and posterior dilator muscles of the cibarial pump , are involved ( Figure 6D , E ) . Contraction of muscles lifts the invaginated wall of the cibarial pump , expands the lumen , and consequently creates the upward suction through the food canal . In this manner the liquid food is drawn into the pump lumen l . When the muscles relax , the elastic energy of lifted pump wall is released and it springs back . The lumen is contracted and the liquid food in the pump flows into the esophagus . The stylets protruding mechanism in hemipteran insects has been discussed in Homalodisca coagulata ( Membracoidea: Cicadellidae ) and diaspidid insects ( Coccoidea: Diaspididae ) ( Leopold et al . , 2003; Beardsley and Gonzalez , 1975 ) . Previously , there is some consensus that the protractor muscles contract to protrude the corresponding stylets ( Leopold et al . , 2003; Beardsley and Gonzalez , 1975 ) . In contrast , we found that the protractor muscles did not contract even when the stylet tips were pushed out of the beak , which was beyond previous conception . Protrusion of the stylet resulted mainly by lowering the head and compression of the labium . Although there are few reports on inner musculation , the exterior structures of the mouthparts of the hemipteran species have been studied a lot . The mouthpart structures of the Hemiptera insects are similar , probably due to evolutionary relatedness ( e . g . Dai et al . , 2014; Mora et al . , 2001 ) . They share similar morphological characteristics , such as a segmented labium that wraps the stylets , interlocking stylets , and two canals in the stylet bundle . Given that function is strictly constrained by structure , we assume that the N . lugens feeding mechanism also applies to other hemipteran species . In some species with bigger body sizes , the labium is bended rather than compressed . The seed bug Pyrrhocoris sibiricus ( Pyrrhocoroidea: Pyrrhocoridae ) and the stink bug Erthesina fullo ( Pentatomoidea: Pentatomidae ) are reported to bend their labium between the first segment and the second segment to protrude the stylet ( Wang and Dai , 2017; Wang and Dai , 2020 ) . These works did not mention if the head was lowered down . We reckon that the head should be lowered in this situation . The bent labium shortens the straight-line distance between the base and the tip of the labium . Theoretically , when the labium bends , the tip of the labium does not reach the surface of the plant unless the head lowers down . It may be common in this family that the head-lowering behavior and bending of the labium are combined when protruding the stylets . The head-lowering behavior in aphids is combined with lifting of the abdomen . They even position their body nearly vertical to the plant surface when sucking in sap ( Guerrieri and Digilio , 2008 ) . Aphids do not bend their labium , so they probably compress the labium as in the case of N . lugens . Scale insects , like Paraputo guatemalensis ( Coccoidea: Pseudococcidae ) and P . odontomachi , have stylet bundles that are much longer than the labium . They usually coil the stylet bundle and hide it in the head capsule ( Beardsley and Gonzalez , 1975 ) , so they are unable to protrude the stylet bundle by lowering the head . An aphid species also have extremely long mouthparts , and their labium can dig into the plant tissue ( Brożek et al . , 2015 ) . Therefore , they probably adopt a different way of piercing their mouthparts into the plant . Overall , the planthopper stylet penetration mechanism might apply to most sap-sucking hemipterans with short stylets . The filter chamber locates in the alimentary canal of most hemipteran insects that feed on fluid food . It connects the anterior and the posterior parts of the midgut and transfers extra water directly from the anterior to the posterior end of the midgut . Without excess water , the sap in the midgut is concentrated 10-fold ( Terra and Ferreira , 2012 ) . This water-disposal mechanism may quickly expel water , ions , and soluble sugars to the hindgut , while amino acids , proteins , and lipids will be retained and digested in the midgut ( Marshall and Cheung , 1974; Salvucci et al . , 1998 ) . However , a filter chamber is not present in N . lugens , and a different water-disposal mechanism is proposed . There are some aphid families such as Aphididae , Adelgidae , Mindaridae , and Chaitophoridae without a filter chamber ( Douglas , 2003 ) . They develop coiled segments of the hindgut adhering to the stomach . Shakesby et al . , 2009 revealed that in the pea aphid Acyrthosiphon pisum ( Aphidoidea: Aphididae ) the stomach and distal intestine are rich in water-transporters called aquaporins . The hypothesis is that water is transferred from the stomach to the distal intestine through the activity of aquaporins at the point where the hindgut contacts the stomach . In N . lugens , we also found a similar structure . The distal end of the anterior midgut contacts the anterior midgut ( Figure 2—figure supplement 1E ) . The epithelia of the anterior midgut become thin at this point , which may facilitate water in the midgut to move to the end of the midgut loop through the epithelium . In the anterior part of the midgut of N . lugens , protuberance on the inner surface may enable the gut lumen to increase by expanding the folds . In contrast , the loop region narrows and develops dense microvilli to increase the absorption surface and the time that food travels through the midgut . When phloem sap is ingested , it moves quickly into the anterior part but slowly into the loop region . As a result , most of the sap may stop in the anterior part , and the gut lumen expands to temporarily store food . At the distal end of the dilated anterior part , the epithelium becomes thinner than at neighboring positions at a point where the end of the midgut loop contacts it . At this place , water in the sap may be transported to the end of the loop region through the epithelia , thereby concentrating proteins and lipid that may then move into the anterior part of the midgut loop for further digestion and absorption . This may be a N . lugens way for concentrating nutriments in the food . The order Hemiptera consists of four higher taxa as Heteroptera ( red bugs , water bugs , seed bugs , lace bugs , bedbugs , and stinkbugs ) , Sternorrhyncha ( aphids , scale insects , whiteflies , and psyllids ) , Auchenorrhyncha ( cicadas , froghoppers , leafhoppers , and planthoppers ) , and Coleorrhyncha ( moss bugs ) ( Johnson et al . , 2018; Figure 2—figure supplement 5 ) . Almost all members of this order rely on a fluid diet , but the morphology of the alimentary canals varies a lot . Their diverse water-disposal mechanisms give some clues in alimentary canal evolution . Species of Pentatomomorpha in the suborder Heteroptera have a discontinuous alimentary canal so that the fluid food does not pass through the midgut . In these species , the intestinal symbiotic organ blocks food flow from the anterior part of the midgut by an extreme narrow region . Food fluids in the anterior regions of the midgut are completely absorbed and transferred into the hemolymph and excreted through the Malpighian tubules back into the hindgut ( Ohbayashi et al . , 2015 ) . Other families of Heteroptera including herbivorous Cimicomoepha and Leptopodmorpha , predatory Nepomorpha and Gerrmorpha have continuous intestines showing no unique water-disposal mechanism ( Habibi et al . , 2008; Nardi et al . , 2019; Rost-Roszkowska et al . , 2017 ) . Their convoluted alimentary canals indicate osmotic water exchanges by juxtapositioning an anterior region of the midgut to the most posterior region as in N . lugens . Members of the suborder Auchenorrhyncha have filter chambers with exception of those species belonging to Fulgoroidea ( Hickernell , 1923; Kershaw , 1914; Zhong et al . , 2015 ) . This indicates a simplification of the alimentary canal in an ancestor of Fulgoroidea . Members of the suborder Sternorrhyncha have a filter chamber , except for most families of Aphidoidea that develop coiled segments of the hindgut adhering to the stomach , as mentioned above ( Kruse et al . , 2017; Mathew et al . , 2011; Peeters et al . , 2017 ) . Members of Coloradoa and Hyalopterus in Aphidoidea still have filter chambers , so loss of the filter chamber occurred only in some families ( Ponsen , 1991 ) . The ancestors of Fulgoroidea and Aphidoidea experienced similar morphological transformations . This may reflect the impact of similar diet and life style on evolution . The association between insects and endosymbionts is widely found in nature and in most cases the endosymbionts are bacteria ( Wernegreen , 2002 ) . The phenomena that symbionts synthesize and supply amino acid for their hosts are well known in hemipteran insects . The aphid bacterial symbiont ( Buchnera aphidicola ) synthesizes essential amino acids ( Shigenobu et al . , 2000 ) . The leafhopper ( Dalbulus maidis ) symbiont Nasuia has the smallest bacterial genome and provides two essential amino acids to the host ( Bennett and Moran , 2013 ) . Burkholderia symbionts found in the species of the superfamilies Lygaeoidea and Coreoidea have been shown to enhance host innate immunity , but do not have any specific nutritional function ( Kikuchi et al . , 2011; Kim et al . , 2015 ) . N . lugens is an exception insofar as a fungus is the dominant endosymbiont . The yeast-like fungus is an obligate endosymbiont in the planthopper . They cannot survive in vitro and usually reside in the fat body mycetocytes , though in this case , we found some of them in legs and heads . Xue et al . , 2014 reported that N . lugens lacks the ability to synthesize essential amino acids that are missing in its diet , the rice phloem sap ( Hayashi and Chino , 1990 ) . Fan et al . , 2015 revealed that genes involved in the synthesis of essential amino acids and cholesterol are found in the genome of the yeast-like symbionts but not in the BPH genome , indicating that the yeast-like symbionts provide essential amino acids and cholesterol to their host . Yeast-like symbionts are maternally transmitted to the eggs and are present in all developmental stages of the host insect ( Cheng and Hou , 2001 ) . We found yeast-like symbionts in a single huge mycetocyte that occupied 22% of the abdominal volume . The mycetocyte is likely to grow bigger as the symbionts thrive , but it does not undergo a cell division during the embryonic and nymphal stages . The anterior diverticulum in hemipterans is often filled with air bubbles; however , its function remains controversial . The Cercopid Tomaspis saccharina ( Cercopoidea: Cercopoidea ) takes in air with the fluid food and excretes air bubbles coated with raucinoid through the anus to produce froths , so AD is likely to separate the air from the food when the fluid passes down into the midgut ( Kershaw , 1914 ) . Kershaw also proposed that the AD worked as a reservoir to store innutritious fluid and wax materials separated from the food . On the other hand , Goodchild found that the AD contains air at the time of moulting and suggested that the AD inflates the thorax during molting ( Goodchild , 1966 ) . Despite these speculations , the AD might serve as a specialized organ to host symbionts in N . lugens . Numerous bacterial symbionts that are much smaller than the yeast-like symbionts are in the wall of the AD . They reside in the hollow AD cells . Our 16 s sequencing results showed that the dominant microbe in the AD is Arsenophonus species , which is reported to kill the male eggs in the wasp , Nasonia vitripennis but in N . lugens shows effect on host insecticide resistance ( Pang et al . , 2018 ) . In previous studies , Arsenophonus was identified as the second most abundant symbiont after the yeast-like symbionts in N . lugens . It mainly resides in the fat body ( Fan et al . , 2016; Ly et al . , 2013 ) . In this case , we believe that Arsenophonus species reside also in the anterior diverticulum cells . The thread-like symbionts in fat body mycetocytes and the symbionts in the AD wall have not been reported in precedent publications . They reside in certain cells in the body instead of being uniformly distributed in the abdomen like the yeast-like symbionts . Therefore , it is difficult to find them in sections in conventional TEM-based studies . We were able to find them probably because we collected serial sections of the entire nymph . The mycetocytes that host the thread-like symbionts lie very close to the midgut . They may be involved in nutrient absorption . Further studies are expected to unravel the relationship between the insect and these intracellular microorganisms . Early studies on the anatomy of the Drosophila and Coleopteran tracheal systems are based on dissections and hand-made illustrations ( Whitten , 1957; Tonapi , 1978 ) . Given that thin tracheae are easily neglected and hand drawing are error-prone , the results are not reliable enough , and quantitative descriptions are difficult . The μCT system provides more reliable and complete results allowing elaborate comparisons ( Iwan et al . , 2015; Socha et al . , 2010 ) . The volume of total tracheae in N . lugens accounted for about 0 . 12% of the body volume . Compared to the tracheal density of 0 . 5% in Tribolium castaneum ( Coleoptera: Tenebrionidea ) , 4 . 8% in Eleodes obscura ( Coleoptera: Tenebrionidea ) calculated using the stereological point count method ( Kaiser et al . , 2007 ) , and 0 . 6% in Tenebrio molitor ( Coleoptera: Tenebrionidea ) larva ( Raś et al . , 2018 ) , the tracheal density in N . lugens is relatively low . This is in agreement with the hypothesis that larger insects invest more on the tracheal system ( Callier and Nijhout , 2011; Kaiser et al . , 2007 ) . The planthopper has a simplified tracheal system without ventral longitudinal trunks and air sacs that are often found in insects with well-developed flight ability ( Gunn , 1931 ) . There is no study on tracheal systems in small insects , so it is unclear whether we can generalize our simplification . N . lugens excels at long-distance movement . Every spring the migration starts from the Indochina peninsula to eastern China , Japan , and Korea with southwesterly monsoons ( Bao et al . , 2000; Sogawa and Cheng CH , 1979 ) . Obviously , a simple tracheal system is able to satisfy the need of oxygen during flight . This may be partly due to the small body size of N . lugens that allows efficient oxygen diffusion . In addition , the novel structure described here may also facilitate gas exchange . The four-way tracheal ring that connects nearby four spiracles is a structure found in insects for the first time . Airflow in the ring is possible both longitudinally and transversely . It is likely to compensate for the lack of ventral longitudinal trunks . The dorsal and ventral longitudinal trunks are present in many other insects . Generally , the airflow in the insect body is longitudinal , which means that the air is taken in through the anterior spiracles and expelled though the posterior spiracles as in the cockroaches ( Heinrich et al . , 2013 ) and the locust ( Miller , 1960 ) , or in the reverse direction as in the wingless dung beetles ( Duncan and Byrne , 2002 ) . If air flows in and out though the same spiracle , the incoming air will mix with the used air and the gas exchange efficiency will decrease . The longitudinal and transverse trunks that connect different spiracles allow airflow between spiracles , resulting in more efficient gas exchange . In N . lugens , oxygen diffuses longitudinally at the dorsal side through the dorsal longitudinal trunks , while at the ventral side , there are no longitudinal trunks . Instead , oxygen diffuses longitudinally through the four-way tracheal rings to efficiently supply oxygen to many tissues located in the thorax and abdomen . The BPH , a member of the superfamily Fulgoroidea , belongs to the dominant group of phytophagous Hemipterans that exceeds 12 , 500 described species . Many species in Fulgoroidea are economically significant pests of major agricultural crops for their direct wounding of plants , high reproductive ability , long-distance migration , and status in virus transmitting . Therefore , detailed studies of the internal structure and feeding mechanism are essential to support research on pest control . In this study , we report on a 3D reconstruction of a whole N . lugens nymph and its sap-sucking mechanism for the first time using SBF-SEM . The new technology is well suitable for a tiny insect and yielded many details with high resolution . We present a set of new findings on the internal structures including 24 neuropils in the central nerve system , nine kinds of follicles in the principle salivary gland with different cell numbers and vesicles , an admirable structure of four-way tracheal rings connecting the spiracles in adjacent segments , fungal endosymbionts in a single huge mycetocyte occupying 22% of the abdominal volume , and symbionts in the crypts of the anterior diverticulum . We also provide videos and interactive three-dimensional PDF versions of the reconstructed structures , which may serve as comprehensible tools for scientists to explore the internal structures of an insect or for teaching . Based on the 3D reconstruction of the cephalic musculature and movement of all feeding apparatuses , we propose a novel feeding model for the phloem sap-sucking mechanism that may also be applied for many other species with similar mouthparts . These results allow an integral understanding of the actual position and relationships of the structures and organ systems in a tiny insect , especially during the food ingestion and digestion , endosymbiosis , and respiration . The information will serve to study how a tiny pest injures its plant host , and the structures as adaptations shaped in the long history of natural selection . The piercing mechanism may even inspire us to new ideas of tunnel excavation in bionics . Further studies on the indispensable relationship of different symbionts with the host may be valuable in developing microbial pesticides . The BPHs used in this study were originally collected in Hangzhou ( 30°16′N , 12°11′E ) , China . They were reared at 26 ± 0 . 5°C on rice seedlings under a 16:8 hr ( light: dark ) photoperiod . We prepared a first instar nymph ( Sample 1 ) and collected 12 , 000 digital images of the whole body ( Figure 1—video 1 ) at a resolution high enough to reconstruct the tracheae and muscles . A head with the prothorax ( Sample 2 , Figure 1—video 2 ) , a head with intact mouthpart ( Sample 3 ) , an abdominal part with several segments ( Sample 4 ) , and a few segments on the posterior end ( Sample 5 ) of other first instar nymphs were prepared to collect digital images at similar resolution as Sample 1 . Images from the whole body were used for reconstruction and the other images provided additional information for locating structural features . To study muscle contraction during the feeding process , the first instar nymphs were immobilized with liquid nitrogen while they were sucking on rice plants . They were frozen fast ensuring that the musculature remained in the feeding state . The insects were then removed from the plant and fixed in 2 . 5% glutaraldehyde ( TED PELLA , Lot No . : 2171002 ) at room temperature immediately ( Sample six and Sample 7 , Figure 4—video 1 , Figure 4—video 2 ) Detailed samples description and image resolution are listed in Table 1 . The samples were fixed quickly for 24 hr in 2 . 5% glutaraldehyde ( TED PELLA , Lot No . : 2171002 ) , and 0 . 003% CaCl2 ( Sigma , C-2661–500G ) in sodium cacodylate ( Sigma , CAS: 6131-99-3 ) buffer ( 0 . 1 M ) . The tissue blocks were then washed in sodium cacodylate buffer ( 0 . 1 M ) and treated with a solution containing equal volumes of 2% OsO4 ( TED PELLA , Lot No: 4008–160501 ) and 3% potassium ferrocyanide ( Sigma , CAS: 14459-95-1 ) in 0 . 3 M sodium cacodylate with 4 mM CaCl2 for 1 hr on ice . After rinsing with double-distilled water ( ddH2O ) , the samples were incubated in 1% thiocarbohydrazide ( Sigma , CAS: 2231-57-4 ) ( in water ) for 20 min at room temperature . Then , the samples were rinsed with ddH2O and treated with 2% aqueous OsO4 for 30 min at room temperature . The tissue blocks were washed in ddH2O and immersed overnight at 4°C in 1% aqueous uranyl acetate . After washing in ddH2O , the samples were incubated in 0 . 66% lead nitrate ( Sigma , CAS: 10099-74-8 ) diluted in 0 . 03 M L-aspartic acid ( Sigma , CAS: 56-84-8 ) ( pH 5 . 5 ) for 30 min at 60°C , then dehydrated in an ascending ethanol series and flat-embedded in EPON 812 resin ( EMS , cat . no: 14900 ) for 48 hr at 60°C . Resin blocks were carefully trimmed using a Leica EM trimmer , until the surface of the black tissue in the block could be observed . Next , the resin blocks were glued on a stub with electrically conductive colloidal silver ( TED PELLA , Lot NO: 169773610 ) , followed by coating with about 30 nm platinum using a sputter coater ( Leica , ACE200 ) . 3D data was obtained by a scanning electron microscope ( Thermo Fisher , Teneo VS ) with one ultramicrotome in the specimen chamber , which allowed synchronous sectioning of resin blocks and imaging of the sample surface . Section thickness and pixel size for each sample are listed in Table 1 . Each serial face was imaged with 2 . 5-kV acceleration voltage and 0 . 2-nA current in backscatter mode with a VS DBS detector . The image store resolution was set to 6144 × 6144 pixels with a dwell time of 2 μs per pixel . The images were aligned , filtered , manually segmented , and used to generate surfaces in Amira 6 . 8 ( Thermo Fisher Scientific ) . The surfaces generated in Amira were reduced to 10% and then exported as Wavefront . obj files . Maxon Cinema 4D R19 ( Maxon Computer GmbH ) was used to reassemble the structures , and the segmentation artifacts were carefully removed using the sculpting tool . The resulting C4D projects were used in the video ( Figure 1—video 3 ) and static images were exported . For the interactive 3D model , the polygon counts were further reduced to 10% or 1% according to the details of the structures . The simplified C4D project was exported as . 3ds file , colored , renamed , saved as . u3d file in Deep Exploration 6 ( Right Hemisphere ) , and embedded into the interactive PDF file using Acrobat Pro DC ( Adobe ) ( Supplementary file ) . Volumes of all reconstructed structures were calculated using the label analysis tool in Amira ( Table 2 ) . The length of muscles was measured using the measurement tool in Amira . To measure the length of the beak , the first instar nymphs were fast frozen by liquid nitrogen when they were sucking in phloem sap on a rice plant . The nymphs were warmed up to room temperature and carefully removed from the plant . A stereoscope was used to measure the distance between two eyes as the width of the head ( wh ) , the length of the beak , and the length of the protruding stylets . The length of the beak ( lb ) refers to the distance between the tip of the labrum and the tip of the labium , and the length of the protruding stylets ( lps ) refers to the distance between the tip of the stylet and the tip of the labium . The measurements are presented as mean ± SEM . Statistical analyses were performed using a two-tailed Student’s t-test . p-values <0 . 001 were considered as statistically significant . For conventional SEM ( CSEM ) , nymphs feeding on rice plants were fixed in liquid nitrogen and post fixed in glutaraldehyde ( 2 . 5% with 0 . 1 M phosphate buffer , at pH 7 . 4 , overnight at 4°C ) . Then the samples were washed in the phosphate buffer ( pH 7 . 4 ) and dehydrated in an ascending ethanol series ( 30% , 50% , 70% , and 90% , followed by 3*100% ) . After critical drying with an automated critical point dryer ( Leica , CPD300 ) , the samples were mounted on double-sided carbon tape on stubs . They were then plasma coated with 10 nm gold using a sputter coater ( Leica , ACE200 ) and viewed under a field emission scanning electron microscope ( Thermo Fisher , Helios UC G3 ) . The image was used in Figure 3—figure supplement 2C . For cryo-SEM , nymphs feeding on rice plants were fixed in liquid nitrogen and transferred to a Deben Cryo-SEM system which preserved samples at −20°C . Samples were examined under a field emission scanning electron microscope ( ZEISS , EVO15 ) at low vacuum mode ( 70 Pa ) . The images were used in Figure 3—figure supplement 2A , B . For structures whose homologous counterparts were reported in other insects , the authors would prefer using the existed terminological systems instead of starting de novo . Nomenclature of neuropils followed Insect Brain Name Working Group et al . , 2014 , and the external sclerites followed that of Wipfler et al . , 2016 . Muscles in the labium are continuously numbered in order of appearance and follow the terminology established in Spangenberg et al . , 2013 . Abbreviations of the structures are presented in Table 3 .	Since the 19th century , scientists have been investigating how the organs of insects are shaped and arranged . However , classic microscopy methods have struggled to image these small , delicate structures . Understanding how the organs of insects are configured could help to identify new methods for controlling pests , such as chemicals that target the mouthparts that some insects use to feed on plants . Most insects that feed on the sap of plants suck out the nutrient via their stylet bundle – a thin , straw-like structure surrounded by a sheath called the labium . As well as drying out the plant and damaging its tissues , the stylet bundle also allows the insect to transmit viruses that cause further harm . To investigate these mouthparts in more detail , Wang , Guo et al . used a method called SBF-SEM to determine the three-dimensional structure of one of the most destructive pests of rice crops , the brown planthopper . In this technique , a picture of the planthopper was taken every time a thin slice of its body was removed . This continuous slicing and re-imaging generated thousands of images that were compiled into a three-dimensional model of the brown planthopper’s whole body and internal organs . Previously unknown features emerged from the reconstruction , including a huge cell in the planthopper’s abdomen which is full of fungi that provide the nutrients absent in plants . Next , Wang , Guo et al . used this technique to see how the muscles in the labium and surrounding the stylet move by imaging planthoppers that were frozen at different stages of the feeding process . This revealed that when brown planthoppers bow their heads to eat , the labium compresses and pushes out the stylet , allowing it to pierce deeper into the plant . This is the first time that the body of such a small insect has been reconstructed three-dimensionally using SBF-SEM . Furthermore , these findings help explain how brown planthoppers and other sap-feeding insects insert their stylet and damage plants , potentially providing a stepping stone towards identifying new strategies to stop these pests from destroying millions of crops .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2021	Three-dimensional reconstruction of a whole insect reveals its phloem sap-sucking mechanism at nano-resolution
How neurons become sensitive to the direction of visual motion represents a classic example of neural computation . Two alternative mechanisms have been discussed in the literature so far: preferred direction enhancement , by which responses are amplified when stimuli move along the preferred direction of the cell , and null direction suppression , where one signal inhibits the response to the subsequent one when stimuli move along the opposite , i . e . null direction . Along the processing chain in the Drosophila optic lobe , directional responses first appear in T4 and T5 cells . Visually stimulating sequences of individual columns in the optic lobe with a telescope while recording from single T4 neurons , we find both mechanisms at work implemented in different sub-regions of the receptive field . This finding explains the high degree of directional selectivity found already in the fly’s primary motion-sensing neurons and marks an important step in our understanding of elementary motion detection . Flies see the world through a hexagonal array of facets each equipped with its own small lens focusing the light onto 8 photoreceptors . Photoreceptors send their axons into the optic lobe , which consists of four consecutive layers of neuropil called lamina , medulla , lobula and lobula plate ( Figure 1a ) . Each neuropil layer is made up of retinotopically organized columns together containing roughly 100 different neurons per column ( Fischbach and Dittrich , 1989 ) . Within the optic lobe , visual motion information is extracted in parallel pathways encoding light increments ( ON ) and decrements ( OFF ) ( Joesch et al . , 2010; Eichner et al . , 2011; Joesch et al . , 2013 ) . Both pathways bifurcate in the lamina and lead , via a set of specific medulla interneurons , onto the dendrites of T4 and T5 cells , respectively ( Takemura et al . , 2013; Shinomiya et al . , 2014 ) . Both T4 and T5 cells exist in 4 subgroups tuned to one of the four cardinal directions of motion and project into four layers of the lobula plate ( Maisak et al . , 2013 ) ( Figure 1b ) . There , they form monosynaptic cholinergic excitatory connections with the dendrites of the large-field lobula plate tangential cells ( Mauss et al . , 2014; Schnell et al . , 2012 ) as well as with lobula plate intrinsic neurons . These intrinsic neurons , in turn , inhibit tangential cells in the adjacent layer ( Mauss et al . , 2015 ) . Since none of the neurons upstream from T4 and T5 respond to visual motion in a direction-selective way ( Behnia et al . , 2014; Strother et al . , 2014; Meier et al . , 2014; Ammer et al . , 2015; Serbe et al . , 2016 ) , T4 and T5 cells are the first neurons in the processing chain where directional information is represented explicitly ( Maisak et al . , 2013; Fisher et al . , 2015 ) . 10 . 7554/eLife . 17421 . 003Figure 1 . Fly optic lobe and visual stimulation . ( a ) Circuit diagram of the ON and the OFF pathway of fly motion vision . Directionally selective signals are carried via T4 and T5 cells to four layers of the lobula plate , where T4 and T5 cells with the same preferred direction converge on the dendrites of the tangential cells ( yellow ) . Inhibition is conveyed via local interneurons LPi ( red ) . From Borst and Helmstaedter , 2015 . ( b ) Confocal image of T4 and T5 cells and their directional tuning . The light green bands indicate the dendrites of T4 and T5 cells . The presynaptic terminals of both T4 and T5 cells form four distinct layers within the lobula plate . The inset shows the result of two-photon calcium imaging , revealing four subgroups of T4 and T5 cells tuned to the four cardinal directions . Scale bar , 20 μm . From Borst and Helmstaedter , 2015 . ( c ) Continuous ( top ) versus apparent motion ( bottom ) , shown as x-t-plots where the luminance distribution is shown along one spatial ( x ) and the time axis ( t ) . During continuous motion at constant velocity , a luminance profile is smoothly drifting along one direction , giving rise to a slanted bar in the x-t-plot . During apparent motion , the luminance profile is stable for a while and then jumps to a new position . ( d ) Two mechanisms proposed to account for direction selectivity . In each model , the signal from one photoreceptor is delayed by a temporal filter ( τ ) and fed , together with the direct signal from the neighboring photoreceptor , into a nonlinearity . In case of preferred direction enhancement ( left ) , the delayed signal ( E ) enhances the direct signal ( D ) , e . g . by a multiplication ( E x D ) , in case of the null direction suppression ( right ) , the delayed signal ( S ) suppresses the direct signal ( D ) , e . g . by a division ( D/S ) . ( e ) Immunostaining of a single T4 dendrite in layer 10 of the medulla ( green ) covering multiple columns ( counterstained against bruchpilot , purple ) . ( f ) Setup for telescopic stimulation of single lamina columns . Antidromic illumination of the eye ( left ) results in parallel beams from the 6+2 photoreceptors in neighboring facets with identical optical axes . These are focused in the back focal plane of the objective projected onto a CMOS camera . In addition , an AMOLED display is coupled into the beam path to precisely stimulate single lamina columns . Lower left inset: The fly eye and the principle of neural superposition . Light rays parallel to each other shown in the same color activate different photoreceptors in neighboring ommatidia that converge onto a single column in the lamina ( ‘neuro-ommatidium’ ) . Lower right inset: Picture from the CMOS camera , showing the far field radiation pattern of the Drosophila eye . Dot stimuli can be precisely positioned such as to stimulate single lamina columns . DOI: http://dx . doi . org/10 . 7554/eLife . 17421 . 003 To investigate the mechanism leading to direction selectivity in T4 cells , we applied apparent motion stimuli where , instead of continuously moving ( Figure 1c , top ) , a bar is abruptly stepped from one location to a neighboring one ( Figure 1c , bottom ) . These stimuli lend themselves well to discriminate between preferred direction ( ‘PD’ ) enhancement ( Hassenstein and Reichardt , 1956 ) ( Figure 1d , left ) and null direction ( ‘ND’ ) suppression ( Barlow and Levick , 1965 ) ( Figure 1d , right ) as the response to the sequence can be compared with the sum of the responses to the luminance pulses given in isolation ( ‘linear expectation’ ) : in case of preferred direction enhancement , the response to the sequence along the preferred direction is larger than the linear sum of responses to the isolated pulses and identical if the sequence is along the null direction; In case of null direction suppression , the response to the sequence along the preferred direction is identical to the linear sum of responses to the isolated pulses and smaller if the sequence is along the null direction . The columnar organization of the optic lobes stretches from the laminar cartridges through the medulla and into the lobula and lobula plate . Individual T4 and T5 neurons extend their dendrites across multiple columns ( Figure 1e ) and receive synaptic inputs from different medulla cell types located in different columns relative to the home columns of the T4 or T5 neuron and to each other ( Takemura et al . , 2013; Shinomiya et al . ; 2014 ) . To understand the particular contribution of those inputs it seemed necessary to precisely place the stimuli onto the columnar raster of the fly’s optic lobe . The structure of the optical system and of the neuronal wiring in flies obeys the neural superposition principle ( Kirschfeld , 1967; Braitenberg , 1967 ) . Those photoreceptors R1-6 from 6 neighboring ommatidia that share the same optical axis converge on the same lamina cartridge , thus forming an optical column that represents the unit of spatial resolution ( ‘neuro-ommatidium’ ) . To visually stimulate these neuro-ommatidia precisely and in isolation , stimulation must consist of parallel rays at angles along the optical axes of those photoreceptors and aligned to the columnar raster . For this , we adopted a telescopic stimulation device ( Franceschini , 1975; Schuling et al . , 1989 ) ( Figure 1f ) . As the fly rhabdomeric photoreceptors work as light guides , the raster of optical columns can be visualized by shining light from within the head capsule ( antidromic illumination ) to align the raster of neuro-ommatidia to the stimulation locations on a micro-display with the help of a CMOS camera . As a proof of principle , we first expressed the genetically encoded calcium indicator GCaMP6m ( Chen et al . , 2013 ) in lamina cells L2 , recorded from their terminals in the medulla by 2-photon microscopy ( Denk et al . , 1990 ) and stimulated them with light spots of 1176 ms duration covering a single optical column at various positions . L2 cells responded maximally to stimuli at a certain column , with less than 30% response amplitude to stimuli positioned on surrounding columns , and negligible responses ( <12% ) to stimuli positioned on columns in the next outer ring ( Figure 2a ) . The slight activation of neighboring columns results directly from the optics of the fly’s eye , where the visual fields of single ommatidia have Gaussian sensitivity profiles with an acceptance angle roughly matching the inter-ommatidial separation ( Götz , 1965 ) . These experiments illustrate the specificity of the telescopic stimulation to single lamina cartridges and thus optic lobe column with minimal cross-stimulation of neighboring columns at the physical limit of the fly optical system . 10 . 7554/eLife . 17421 . 004Figure 2 . Receptive field of L2 ( a ) and T4 ( b ) cells . Three example traces from a single terminal ( top , stimulated ommatidium indicated in black ) and mean responses ( bottom ) of L2 cells ( a , n = 23 cells from six flies ) and T4 cells ( b , n = 10 cells from 10 flies ) to flicker stimuli presented at 19 different columnar positions . The responses of individual cells were averaged after alignment to the maximum and normalization and are shown in false color code ( left ) as well as in 3D bar plots ( right ) . In addition , responses are presented as bar plots along the three axes ( dashed lines ) of the hexagonal array ( mean ± SEM ) . D = dorsal , V = ventral , L = lateral , F = frontal , DL = dorso-lateral , VF = ventro-frontal , VL = ventro-lateral , DF = dorso-frontal . DOI: http://dx . doi . org/10 . 7554/eLife . 17421 . 004 We next used a driver line specific for those T4 and T5 cells sending their axons into layer 3 of the lobula plate that are hence sensitive to upward motion . We recorded from single T4 cells by selecting individual processes in layer three of the lobula plate and confirmed their preference for luminance increments . Repeating the above experiment with T4 cells , we again found maximal responses to the stimulus placed in a single column . Compared to L2 , however , the receptive field was found to be larger , with about 50% amplitude to the stimuli onto the six surrounding columns , and roughly 25% to the next outer ring ( Figure 2b ) . This indicates that T4 neurons pool excitatory synaptic input from more than one column , consistent with their morphology ( Figure 1e ) and expected from a motion detector that is required to integrate information from spatially offset input . In order to discriminate between preferred direction enhancement and null direction suppression , we tested T4 cells with three light pulses of 472 ms duration positioned along the dorso-ventral axis of the eye ( Figure 3a , left ) . T4 cells responded to the individual pulses with different amplitudes , depending on the position of the stimulus ( Figure 3a , ‘Flicker’; see also Figure 2b ) . When stimulated sequentially from ventral to dorsal ( Figure 3a , ‘Sequence’ , top middle ) , the cell responded more strongly ( thick blue line ) than expected from the sum of the responses to the individual stimuli ( thin blue line ) . The opposite was observed when we stimulated the cell sequentially from dorsal to ventral ( Figure 3a , ‘Sequence’ , bottom middle ) : now the cell responded more weakly ( thick red line ) than expected from the sum of the responses to the individual stimuli ( thin red line ) . We then calculated the nonlinear response component by subtracting the linear expectation from the actual response and found that both preferred direction enhancement ( Figure 3a , ‘Sequence’ , top right ) and null direction suppression ( Figure 3a , ‘Sequence’ , bottom right ) contributes to the directionally selective responses of T4 cells . 10 . 7554/eLife . 17421 . 005Figure 3 . Responses of T4 neurons to apparent motion stimulation . ( a ) Response of a single T4 cell recorded in a single sweep to three-step apparent motion stimuli . The image shows the position of the three stimuli . Left: Responses to individual light pulses ( ‘Flicker’ ) delivered at the three different positions . The responses are shifted according to the stimulus protocol used for the subsequent apparent motion stimuli . Middle: Responses of T4 to apparent motion stimuli in preferred and null direction ( thick line = measured response , thin line = linear expectation , i . e . sum of responses to the single light pulses ) . Right: Nonlinear response component defined as the difference between measured response and linear expectation . The responses are the mean obtained from n = 3 stimulus repetitions . Similar data were obtained in 12 experiments . ( b ) Dependence of the nonlinear two-step response component on the position within the receptive field of T4 cells . Left: Responses to individual light pulses at the four positions . Middle: Stimulus arrangement . Right: Nonlinear response component . Data represent the mean ± SEM ( n = 15 cells in 11 flies ) . ( c ) Responses to 2-step apparent motion stimuli as a function of the onset time delay ( top: preferred direction , location −1 -> 0; bottom: null direction , location 2 -> 0 ) . Data represent the mean ± SEM ( upper: n = 13 cells from 8 flies; lower: n = 7 cells from 7 flies ) . ( d ) Average responses to a flicker stimulus of the central neuro-ommatidium ( black ) , to a two-step ( red ) and a three-step ( green ) null-direction apparent motion sequence ( mean ± SEM , n = 11 T4 cells from eight flies ) . ( e ) Mean responses of T4 cells to pulses presented to the central column and simultaneously to one of the 18 surrounding columns . Responses were averaged after alignment to the receptive field center , normalized to the flicker response to the central column and are shown in false color code and as bar plots along the three axes of the hexagonal array ( mean ± SEM , n = 10 T4 cells from 8 flies ) . Abbreviations as in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17421 . 005 We then asked whether these two mechanisms occupied separate or overlapping receptive fields . For that , we presented pulses in four neighboring columns along the dorso-ventral axis , individually ( Figure 3b , left ) as well as sequences of pairs in adjacent columns ( Figure 3b , middle ) . Stepping the stimulus up and down between the uppermost columns , we observed only null direction suppression but no preferred direction enhancement ( Figure 3b , right top ) . The opposite was true for stimuli at the two lower most locations: here , only preferred direction enhancement was observed , but no null direction suppression ( Figure 3b , right bottom ) . For sequences between the two inner locations , both phenomena were visible ( Figure 3b , right middle ) . We conclude that the different mechanisms are offset in their receptive fields: preferred direction enhancement is shifted towards the ‘null-side’ , and null direction suppression towards the ‘preferred side’ with respect to each other . To assess the time-course of both mechanisms , we presented two pulses of 472 ms duration and delayed one with respect to the other by a variable time-lag ( Figure 3c ) . When we placed the pulses onto the ventral part of the receptive field , delaying the upper with respect to the lower one ( PD ) , responses peaked at a delay of about 500 ms ( Figure 3c , top ) . When the pulses were placed onto the dorsal part of the receptive field , delaying the lower with respect to the upper one ( ND ) , responses were suppressed and returned to base line at a delay larger than 500 ms ( Figure 3c , bottom ) . We also noticed that this null direction suppression builds up over longer sequences of steps ( Figure 3d ) . When the stimulation of the central column was preceded by the stimulation of one directly adjacent column on the ‘preferred side’ , the response to the central pulse was strongly suppressed as before ( red trace ) . This suppression led to a response to the sequence that on average not only falls below the sum of both individual stimulations , but even below the flicker response to the central column ( black trace ) , indicating inhibition . When this null-direction sequence was extended from two to three columns , the resulting response was even smaller ( green trace ) . The experiments shown in Figure 3c also revealed that for zero onset time differences , i . e . simultaneous stimulation of two columns , the response was suppressed as compared to stimulation of just one column . We tested the receptive field of this suppression of the central column systematically by simultaneously stimulating the central column together with one of the columns in the 2 rings surrounding it ( Figure 3e ) . We normalized the response to the two stimulated columns with respect to the response to the isolated stimulation of the central column . The resulting response in the T4 neuron was found to be suppressed in comparison to the isolated central column response for simultaneous stimulation of the central column together with another one in the dorsal part of the receptive field . This suggests that employing apparent motion with flicker stimuli of larger bars should only lead to small flicker responses , possibly occluding null-direction suppression . We tested this prediction by presenting bright horizontal bars of different spatial extent at different locations within the receptive field of individual T4 cells on an LED arena ( Figure 4a , c ) . Using a bar size of 180° × 4 . 5° , indeed , flicker responses of T4 cells were almost undetectable ( Figure 4a ) . Apparent motion stimuli along the preferred direction led to pronounced responses , larger than the linear prediction , while null direction sequences did not suppress the negligible sum of flicker responses by a significant amount ( Figure 4b ) . This was much different when repeating the same stimuli using bars of only 4 . 5° × 3° instead: now , as in the experiments employing telescopic stimulation of individual columns , strong flicker responses appeared ( Figure 4c ) . Furthermore , in addition to the preferred direction enhancement , pronounced null direction suppression was observed with peak sensitivity within the dorsal part of the receptive field ( Figure 4d ) . 10 . 7554/eLife . 17421 . 006Figure 4 . Apparent motion stimulation on a LED arena . ( a ) Flicker responses of T4 cells to the presentation of a large horizontal bar ( 180˚ × 4 . 5˚ ) at different elevations on a LED arena , aligned to the elevation evoking the maximum response ( n = 17 cells from 5 flies ) . ( b ) Corresponding non-linear response components to two-step apparent motion sequences at different elevations on an LED arena in the preferred ( blue ) and null direction ( red ) . ( c , d ) as in ( a , b ) for small horizontal bars ( 4 . 5˚ × 3˚ ) ( n = 18 cells from 5 flies ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17421 . 006 What is the virtue of having null direction suppression in addition to the preferred direction enhancement ? To address this question , we constructed a simple algorithmic model of a motion detector ( Figure 5a ) . In this model , the motion detector receives 3 inputs . A central direct input ( D ) is flanked by two low-pass filtered inputs: one enhancing input ( E ) implementing a multiplicative and one suppressing input ( S ) implementing a divisive non-linearity . This way , the detector is designed to combine a preferred direction enhancement on one and a null direction suppression on the other side within a single motion-sensing unit . Testing the model with single pulses , responses were only detected for positions −1 and 0 ( Figure 5b , left ) . Stimulating the model with 2-pulse-sequences at positions −1 and 0 led to pronounced preferred direction enhancement but no null direction suppression ( Figure 5b , top 2 rows ) . When the 2-pulse-sequences were delivered at positions 0 and 1 , no pronounced preferred direction enhancement occurred , but null direction suppression was substantial ( Figure 5b , bottom 2 rows ) . Thus , these simulation results qualitatively match the respective experimental data ( compare with Figure 3a and b ) . We next simulated an array of such units and calculated its responses to grating motion at different speeds and directions . We found that the model exhibits a high degree of directional selectivity over a broad range of velocities , with little responses to null direction motion ( Figure 5c , middle left ) . In contrast , the model without null direction suppression revealed substantial null direction responses almost as large as its preferred direction responses ( Figure 5c , middle center ) . The same was true for a model with no preferred direction enhancement ( Figure 5c , middle right ) . Interestingly , no such qualitative differences were observed at the level of a model tangential cell , i . e . after subtraction of opponent units ( Figure 5c , bottom ) . When we simulated the responses of the model to grating motion into different directions , we found the full model T4 cell to be more sharply tuned to its preferred direction than the one without preferred direction enhancement or the one without null direction suppression ( Figure 5d ) . We conclude that combining preferred direction enhancement with null direction suppression leads to a strong direction selectivity of the primary motion-sensing unit . 10 . 7554/eLife . 17421 . 007Figure 5 . Computer simulations . ( a ) Model T4 cell combining preferred direction enhancement and null direction suppression ( compare with Figure 1d ) . ( b ) Left: Responses to individual light pulses ( ‘Flicker’ ) delivered at the three different positions . The responses are shifted according to the stimulus protocol used for the subsequent apparent motion stimuli . Middle: Responses of the model unit to apparent motion stimuli in preferred and null direction ( thick line = measured response , thin line = linear expectation , i . e . sum of responses to the single light pulses ) . Right: Nonlinear response component defined as the difference between measured response and linear expectation . Preferred direction enhancement occurs only between positions −1 and 0 , null direction suppression only between positions 0 and 1 . ( c ) Comparison of three models ( top row ) : a full model , as in ( a ) , implementing preferred direction enhancement and null direction suppression ( left ) , one with preferred direction enhancement only ( center ) and one with null direction suppression only ( right ) . Temporal frequency tuning of model T4 cells ( middle row ) and tangential cells ( bottom row ) using motion of a sine-grating ( spatial wavelength = 50° , contrast = 1 . 0 ) along the preferred ( PD ) and null direction ( ND ) , based on those 3 models . ( d ) Directional tuning of an array of model T4 cells using the motion of a sine-grating ( spatial wavelength = 50° , contrast = 1 . 0 , temporal frequency = 1 . 0 ) for the same 3 models . DOI: http://dx . doi . org/10 . 7554/eLife . 17421 . 007 We next investigated the direction selectivity of T4 and T5 cells by recording their responses to moving gratings presented on an LED arena moving at various speeds and directions . Over a wide range of velocities spanning more than two orders of magnitude , T4/T5 cells responded almost exclusively to upward motion , i . e . along their preferred direction , with little or no responses at all to motion along their null direction ( Figure 6a ) . When stimulating the cells with 1 Hz grating motion at various directions , we found a rather sharp directional tuning with about 90° half-width around its preferred direction ( Figure 6b; see also Figure 3g in Maisak et al . , 2013 ) . To rule out that the highest degree of directional selectivity in T4/T5 cells is achieved by a hitherto unknown reciprocal inhibition between T4/T5 cells with opposite preferred direction , e . g . via inhibitory lobula plate interneurons ( Mauss et al . , 2015 ) we blocked the synaptic output from all T4 and T5 cells by expression of tetanus toxin light-chain ( Sweeney et al . , 1995 ) . Having confirmed the effectiveness of the block ( Supplementary file 1 ) , we repeated the above experiments . T4/T5 cells revealed the same high degree of directional selectivity as before ( Figure 6c , d ) . From this , we conclude that the mechanism underlying the high degree of directional selectivity in T4 cells does not include the output of oppositely tuned T4 cells but rather originates in the dendrite of the cells . 10 . 7554/eLife . 17421 . 008Figure 6 . Response properties of T4/T5 neurons . ( a ) Temporal frequency tuning of T4/T5 cells in control flies showing the normalized ΔF/F responses to square-wave gratings in the preferred ( up , blue ) and null direction ( down , red ) ( n = 15 flies ) . ( b ) Directional tuning of T4/T5 cells to square-wave gratings moving at a temporal frequency of 1 Hz ( n = 7 ) . ( c , d ) Temporal ( c , n = 9 ) and directional tuning ( d , n = 9 ) of T4/T5 cells with their synaptic output blocked by expression of the tetanus toxin light chain . All data represent the mean ± SEM ( shaded ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17421 . 008 In the Hassenstein-Reichardt model ( Hassenstein and Reichardt , 1956 ) as well as in the Barlow-Levick model ( Barlow and Levick , 1965 ) direction selectivity emerges by a nonlinear interaction of asymmetrically filtered signals from adjacent image points: an enhancement along the preferred direction in the first , and a suppression along the null direction in the second model . Both models lead to weakly direction-selective signals in the first place , which , in the Hassenstein-Reichardt model , are improved downstream by subtraction of oppositely tuned components ( for review , see Borst and Helmstaedter , 2015 . Surprisingly , however , a high degree of directional selectivity is found already at the first stage where directional responses are observed , i . e . in T4 and T5 cells ( Maisak et al . , 2013 ) ( Figure 6 ) . This can now be explained by the fact that both preferred direction enhancement and null direction suppression are implemented in T4 cells . While the output of the full Hassenstein-Reichardt-correlator after the subtraction stage shows a high degree of directional selectivity in the absence of null-direction suppression , the relatively small differences between large , but poorly tuned signals ( see Figure 5c , middle panel ) would be highly prone to noise . Improving the direction-tuning already at the level of the half-detectors by the additional null-direction suppression increases robustness to noise and might in addition be energetically less costly . A recent study addressing the mechanisms underlying the elementary motion detectors in Drosophila concluded preferred direction enhancement as the sole mechanism ( Fisher et al . 2015 ) . However , we find that their stimulation paradigm consisting of flashes at 2 positions simultaneously does not allow to discriminate between different mechanisms including lateral inhibition , adaptation and null direction suppression . Since the operations of preferred direction enhancement and null direction suppression have offset receptive fields , it requires the cells to collect their input from more than two adjacent image points , in line with anatomical observations ( Figure 1e; Takemura et al . , 2013; Shinomiya et al . , 2014 ) . In contrast to the cellular implementation of the motion detector proposed by Behnia et al , 2014 our findings also require that T4 cells receive input from more than two cell types . This again is in agreement with recent connectomic data ( for T4 cells: Louis Scheffer , Janelia Research Campus , personal communication; for T5 cells: Shinomiya et al . , 2014 ) . Finally , functional studies where the synaptic input elements onto T4 and T5 neurons were blocked also suggested the involvement of more than two input elements in the ON- ( Ammer et al . , 2015 ) as well as in the OFF-pathway ( Serbe et al . , 2016 ) . Considering the columnar nature of the input elements and the structure of the T4 dendrites covering multiple columns , we predict , that the inputs representing the enhancing , direct and suppressing input segregate on different sub-regions of the T4 dendrite , with enhancing inputs on the ‘null-side’ and the suppressing inputs on the ‘preferred side’ of the dendrite . Future experiments will have to map the different cell types of the medulla onto the proposed model ( Figure 5a ) as well as to identify the transmitter receptors , responsible for enhancing , exciting and inhibiting the dendrites of T4 and T5 cells . ( Drosophila melanogaster ) were raised at 25°C and 60% humidity on a 12 hr light/12 hr dark cycle on standard cornmeal agar medium . For calcium imaging of T4/T5 cells , flies were used to express the genetically-encoded calcium indicator GCaMP6m ( Chen et al . , 2013 ) in T4/T5 neurons with axon terminals predominantly in layer 3 of the lobula plate ( w-; Sp/cyo; VT50384-lexA , lexAop-GCaMP6m/TM6b ) . For the imaging experiments where synaptic output of T4/T5 activity was blocked , the above flies were crossed to flies driving expression of the tetanus toxin light chain in all T4/T5 cells ( w-; R59E08-AD / UAS-TNT-E; R42F06-DBD / VT50384-lexA , lexAop-GCaMP6m ) . For imaging L2 cells , we used the 21D-Gal4 driver line ( Joesch et al . , 2010 ) to express GCaMP6m . Fly surgeries were performed , and the neuronal activity was measured from the left optical lobe on a custom-built 2-photon microscope as previously described ( Maisak et al . , 2013 ) . Images were acquired at resolutions between 64 × 64 and 256 × 256 pixels and frame rates between 1 . 88 and 15 Hz with the ScanImage software ( Pologruto et al . , 2003 ) in Matlab . Antidromic illumination of the fly’s head through the objective used for two-photon-microscopy visualizes the hexagonal mosaic of the optical axes of the ommatidia of a living Drosophila ( Franceschini , 1975; Schuling et al . , 1989 ) . The far field radiation pattern ( FFRP ) visible in the back focal plane of the objective ( LD Epiplan 50x/0 , 50 , Zeiss ) is projected onto a CMOS camera ( DCC1545M , Thorlabs ) via several lenses , a beam splitter ( CM1-BP145B5 , Thorlabs ) and a diaphragm , to reduce stray light . Visual stimuli are generated on the AMOLED display ( 800 × 600 pixels , pixel size 15 × 15 μm , maximal luminance > 1500 Cd/m2; lambda = 530 nm; refresh rate 85 Hz ) ( SVGA050SG , Olightek ) . Both stimulus pattern and FFRP can be visualized simultaneously by means of the beam splitter and a mirror with the CMOS camera . This allows to precisely position the stimuli onto the FFRP . In order to prevent stimulus light from entering the photomultiplier of the two-photon microscope , light generated by the AMOLED display was filtered with a long pass filter ( 514 LP , T: 529 . 4–900 nm , AHF ) . The AMOLED display was controlled with MATLAB and the psychophysics toolbox ( V 3 . 0 . 11; Brainard , 1997 ) . We determined the receptive field of T4 cells by stimulating single cartridges with light pulses of 472 ms duration at randomized positions . At each position , three stimulus presentations were delivered . The resulting responses were averaged and the peak of the averaged response was taken . We performed all experiments on T4 cells only . The cells were selected based on their response to light-on stimuli . While T4 cells respond to the onset of a light pulse ( Figure 2b ) , the T5 cells respond to the light off . Apparent motion stimuli consisted either of consecutive light stimuli to two or three neighboring cartridges . The second stimulus was presented right after the first turned off , resulting in a delay from onset to onset of 472 ms . The LED arena subtended approximately 180° in azimuth and 90° in elevation with a resolution of 1 . 5° , based on a design modified from Reiser and Dickinson ( 2008 ) as previously described ( Maisak et al . , 2013 ) . Stimuli were presented with 3–5 repetitions per experiment in a randomized fashion . All stimuli were presented in full contrast . To measure the directional and temporal frequency tuning , square-wave gratings with a spatial wavelength of 24° spanning the full extent of the stimulus arena were used . For the direction tuning those gratings were moved in 12 directions separated by 30° at a temporal frequency of 1 Hz . To determine the temporal frequency tuning , gratings were moved at temporal frequencies ranging 0 . 05 to 20 Hz moving in the preferred and null direction . For the apparent motion experiments , either large ( 180° wide × 4 . 5° high ) or small ( 4 . 5 × 3° ) bright horizontal bars were presented for 400 ms either in isolation ( flicker ) or in sequences of 2 pulses ( apparent motion ) offset by 4 . 5° in the preferred or null direction with a ∆t , onset to onset , of 400 ms . Data analysis was performed offline using custom-written routines in Matlab . Regions of interests ( ROIs ) were selected by hand in layer 3 of the lobula plate . The time courses of relative fluorescence changes ( ΔF/F ) were calculated from the raw imaging sequence . Responses to the stimulus were baseline-subtracted , averaged across repetitions , and quantified as the peak responses over the stimulus epochs . For T4 cells , the baseline was determined during one second before the stimulus , for L2 cells ( Figure 2a ) , it was determined during the last second before luminance off-set . Those responses were averaged across experiments . Where indicated , responses were normalized to the maximum average response before averaging . For the apparent motion experiments , non-linear response components were calculated as the differences of the time-courses of the responses to the apparent motion stimuli and the sum of the appropriately time-shifted responses to flicker stimuli at the corresponding positions . Visual stimuli were represented as a two-dimensional array ( 200 x 300 ) at 1° spatial and 10 ms temporal resolution , mapped onto a linear array of 40 visual columns . The signal amplitude ranged from 0 to 1 . The input to the ON pathway , as represented by lamina neuron L1 , was modeled as previously described ( Eichner et al . , 2011 ) . Briefly , the local luminance signal was high-pass-filtered ( 1st order , τ = 250 ms ) , and 10% of the DC value was added with subsequent rectification . The T4 cell was then modeled as receiving input from three adjacent columns ( Figure 5a ) : an enhancing signal E at position −1 , representing the low-pass filtered signal of L1 ( 1st order , τ = 250 ms ) , a direct signal D at position 0 which is identical to L1 , and a suppressing signal S again representing the low-pass filtered signal of L1 ( 1st order , τ = 250 ms ) . The response was calculated as the product of E and D , divided by S . Signals had the following weights k: kE = 5 , kD = 5 , kS = 10 . To avoid division by zero and to account for flicker responses , a DC term of 1 . 0 was added to each signal . To simulate flicker and apparent motion stimuli ( Figure 5b ) , light pulses of 450 ms length and amplitude 1 . 0 were delivered to selected columns and the response of an individual model T4 cell was evaluated . To simulate the responses to sine-gratings ( Figure 5c , d ) , responses were calculated either as the summed responses of an array of such units ( ‘T4 response’ ) or after subtracting the signals of oppositely tuned T4 cells ( ‘VS response’ ) . Simulations without null direction suppression or preferred direction enhancement were performed by setting either kE or kS to zero . Software was written in the Python programming language and is available as source code .	The brain extracts information from signals delivered from the eyes and other sensory organs in order to direct behavior . Understanding how the interactions and wiring of a multitude of individual nerve cells process and transmit this critical information to the brain is a fundamental goal in the field of neuroscience . One question many neuroscientists have tried to understand is how nerve cells in an animal’s brain detect direction when an animal sees movement of some kind – so-called motion vision . The raw signal from the light receptors in the eye does not discriminate whether the light moves in one direction or the other . So , the nerve cells in the brain must somehow compute the direction of movement based on the information relayed by the eye . For more than half a century , major debates have revolved around two rival models that could explain how motion vision works . Both models could in principle lead to neurons that prefer images moving in one direction over images moving in the opposite direction – so-called direction selectivity . In both models , the information about the changing light levels hitting two light-sensitive cells at two points on the eye are compared across time . In one model , signals from images moving in a cell’s preferred direction become amplified . In the other model , signals moving in the unfavored direction become canceled out . However , neither model perfectly explains motion vision . Now , Haag , Arenz et al . show that both models are partially correct and that the two mechanisms work together to detect motion across the field of vision more accurately . In the experiments , both models were tested in tiny fruit flies by measuring the activity of the first nerve cells that respond to the direction of visual motion . While each mechanism alone only produces a fairly weak and error-prone signal of direction , together the two mechanisms produce a stronger and more precise directional signal . Further research is now needed to determine which individual neurons amplify or cancel the signals to achieve such a high degree of direction selectivity .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Complementary mechanisms create direction selectivity in the fly
Proteasomes are essential for protein homeostasis in eukaryotes . To preserve cellular function , transcription of proteasome subunit genes is induced in response to proteasome dysfunction caused by pathogen attacks or proteasome inhibitor drugs . In Caenorhabditis elegans , this response requires SKN-1 , a transcription factor related to mammalian Nrf1/2 . Here , we use comprehensive genetic analyses to identify the pathway required for C . elegans to detect proteasome dysfunction and activate SKN-1 . Genes required for SKN-1 activation encode regulators of ER traffic , a peptide N-glycanase , and DDI-1 , a conserved aspartic protease . DDI-1 expression is induced by proteasome dysfunction , and we show that DDI-1 is required to cleave and activate an ER-associated isoform of SKN-1 . Mammalian Nrf1 is also ER-associated and subject to proteolytic cleavage , suggesting a conserved mechanism of proteasome surveillance . Targeting mammalian DDI1 protease could mitigate effects of proteasome dysfunction in aging and protein aggregation disorders , or increase effectiveness of proteasome inhibitor cancer chemotherapies . The proteasome is a multi-protein complex responsible for the majority of protein degradation in eukaryotic cells ( Tomko and Hochstrasser , 2013 ) . The essential function of the proteasome , and its highly conserved structure and mechanism of proteolysis renders it an attractive target for bacteria and other competitors . Production of small molecule inhibitors and protein virulence factors that target the proteasome by some bacteria and fungi exploits this vulnerability to gain a growth advantage ( Fenteany et al . , 1995; Groll et al . , 2008; Meng et al . , 1999 ) . In addition , environmental stresses antagonize the proteasome by causing accumulation of unfolded and aggregated proteins that can form a non-productive inhibitory interaction with proteasomes ( Ayyadevara et al . , 2015; Deriziotis et al . , 2011; Kristiansen et al . , 2007; Snyder et al . , 2003 ) . Human diseases in which proteasome dysfunction is implicated highlight the importance of maintaining proteasome function in the face of these challenges ( Ciechanover and Kwon , 2015; Paul , 2008; Tomko and Hochstrasser , 2013 ) , and it follows that animal cells possess mechanisms to monitor and defend proteasome function . A conserved response to proteasome disruption is the transcriptional up-regulation of proteasome subunit genes ( Fleming , 2002; Meiners et al . , 2003; Wójcik and DeMartino , 2002 ) . In mammalian cells members of the Cap’ n’ Collar basic leucine zipper ( CnC-bZip ) family of stress responsive transcription factors mediate this transcriptional response . Two CnC-bZip franscription factors , Nrf1/NFE2L1 and Nrf2 , have similar DNA-binding domains and may regulate an overlapping set of downstream targets . However , only Nrf1 is required for upregulation of proteasome subunits following proteasome disruption , whereas Nrf2 may activate proteasome expression under other circumstances ( Arlt et al . , 2009; Radhakrishnan et al . , 2010; Steffen et al . , 2010 ) . The events leading to Nrf1 activation in response to proteasome disruption are complex . In vitro analyses in human and mouse cells indicate that Nrf1 is an endoplasmic reticulum ( ER ) membrane associated glycoprotein that is constitutively targeted for proteasomal degradation by the ER-associated degradation ( ERAD ) pathway . Upon proteasome inhibition Nrf1 is stabilized , undergoes deglycosylation and proteolytic cleavage , and localizes to the nucleus ( Radhakrishnan et al . , 2014; Sha and Goldberg , 2014; Wang , 2006; Zhang and Hayes , 2013; Zhang et al . , 2015 , 2007 , 2014 ) . How processing of Nrf1 is orchestrated , and its significance in responses to proteasome disruption in vivo are not understood . Upon proteasome disruption , C . elegans induces transcription of proteasome subunit , detoxification , and immune response genes , and animals alter their behavior to avoid their bacterial food source ( Li et al . , 2011; Melo and Ruvkun , 2012 ) . The transcriptional response to proteasome disruption involves skn-1 , which encodes multiple isoforms of a transcription factor with similarities to both Nrf1 and Nrf2 ( Blackwell et al . , 2015; Li et al . , 2011 ) . skn-1 was originally identified for its essential role in embryonic development ( Bowerman et al . , 1992 ) , but is also required after these early stages for stress responses in a manner analogous to mammalian Nrf1/2 ( An and Blackwell , 2003; Oliveira et al . , 2009; Paek et al . , 2012 ) . SKN-1 binds to the promoters of proteasome subunit genes and mediates their upregulation in response to proteasome disruption , and is required for survival of a mutant with attenuated proteasome function ( Keith et al . , 2016; Li et al . , 2011; Niu et al . , 2011 ) . The molecular mechanism that links SKN-1 activation to the detection of proteasome dysfunction has not been established . Here , we use genetic analysis to uncover the mechanism that couples detection of proteasome defects to these transcriptional responses in C . elegans . We find that an ER-associated isoform of SKN-1 ( SKN-1A ) , is essential for this response . Our genetic data show that the ER-association of this transcription factor normally targets it for poteasomal degradation via ERAD , but is also required for its correct post-translational processing and activation during proteasome dysfunction . After ER-trafficking , our data argues that the PNG-1 peptide N-glycanase removes glycosylation modifications that occur in the ER , and then the DDI-1 aspartic protease cleaves SKN-1A . Each of these steps in SKN-1A processing is essential for the normal response to proteasomal dysfunction . This pathway is essential for compensation of proteasome function under conditions that partially disrupt the proteasome; when compensation is disabled , mild inhibition of the proteasome causes lethal arrest of development . Thus we reveal a vital mechanism of proteasome surveillance and homeostasis in animals . The proteasome subunit gene rpt-3 is upregulated in a skn-1-dependent manner in response to proteasome disruption ( Li et al . , 2011 ) . We generated a chromosomally integrated transcriptional reporter in which the rpt-3 promoter drives expression of GFP ( rpt-3::gfp ) . This reporter gene is upregulated in response to drugs such as bortezomib or mutations that cause proteasome dysfunction . To identify the genetic pathways that sense proteasome dysfunction and trigger the activation of SKN-1 , we took advantage of a regulatory allele affecting the pbs-5 locus . pbs-5 encodes the C . elegans ortholog of the beta 5 subunit of the 20S proteasome . The pbs-5 ( mg502 ) mutation causes constitutive skn-1-dependent activation of rpt-3::gfp expression , but does not otherwise alter fertility or viability ( Figure 1—figure supplement 1 ) . Following EMS mutagenesis , we isolated a collection of recessive mutations that suppress the activation of rpt-3::gfp caused by pbs-5 ( mg502 ) , and identified the causative mutations by whole genome sequencing ( Table 1 ) . The collection includes multiple alleles of genes encoding factors required for ERAD . In ERAD , misfolded glycoproteins are retrotranslocated from the ER lumen to the cytoplasm , where they are degraded by the proteasome ( Smith et al . , 2011 ) . We isolated 3 alleles of sel-1 , a gene that encodes the C . elegans orthologue of HRD3/SEL1 . HRD3/SEL1 localizes to the ER membrane and recognizes ERAD substrates in the ER ( Carvalho et al . , 2006; Denic et al . , 2006; Gauss et al . , 2006 ) , and a single allele of sel-9 , which encodes the C . elegans orthologue of TMED2/EMP24 , which is also ER-localized and implicated in ER quality control ( Copic et al . , 2009; Wen and Greenwald , 1999 ) . We also found mutations in png-1 , which encodes the C . elegans orthologue of PNG1/NGLY1 . After ERAD substrates have been retrotranslocated to the cytoplasm , PNG1/NGLY1 removes N-linked glycans to allow their degradation by the proteasome ( Kim et al . , 2006; Suzuki et al . , 2016 ) . Most strikingly , we isolated six alleles of C01G5 . 6 ( hereafter ddi-1 ) , which encodes the C . elegans orthologue of DDI1 ( DNA damage inducible 1 ) . DDI-1 is an aspartic protease , highly conserved throughout eukaryotes ( Sirkis et al . , 2006 ) . DD1’s function is poorly understood , but it has been implicated in regulation of proteasome function and protein secretion ( Kaplun et al . , 2005; White et al . , 2011 ) . 10 . 7554/eLife . 17721 . 003Table 1 . EMS-induced mutations that disrupt rpt-3::gfp activation . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 003AlleleAffected geneEffectHomologuesmg563C01G5 . 6L245FDDI1 . Aspartic protease . mg555C01G5 . 6C277Smg544C01G5 . 6G293Rmg542C01G5 . 6R350STOPmg543C01G5 . 6M244Img557C01G5 . 6L334Fmg565sel-1G594EHRD3/SEL1 . ER membrane protein , required for ERAD substrate recognition . mg567sel-1A522Tmg547sel-1splice sitemg550sel-9S140FEMP24/TMED2 . ER membrane protein . mg561png-1G498RPNG1/NJLY1 . Peptide N-glycanase . Removes N-linked glycans during ERAD . mg564png-1splice site We examined activation of rpt-3::gfp in ERAD and ddi-1 mutant animals following disruption of proteasome function by RNAi of the essential proteasome subunit rpt-5 . rpt-5 ( RNAi ) caused larval arrest confirming that all genotypes are similarly susceptible to RNAi . While rpt-5 ( RNAi ) causes robust activation of rpt-3::gfp in wild-type animals , mutants lacking ERAD factors or ddi-1 failed to fully activate rpt-3::gfp ( Figure 1a ) . The requirement for sel-11 , which encodes an ER-resident ubiquitin ligase required for ERAD ( Smith et al . , 2011 ) , supports a general requirement for ERAD in activation of rpt-3::gfp expression . These genes are also required for upregulation of rpt-3::gfp following proteasome disruption by bortezomib ( data not shown ) . 10 . 7554/eLife . 17721 . 004Figure 1 . ER-associated degradation factors and the aspartic protease DDI-1 are required for responses to proteasome disruption . ( a ) rpt-3::gfp expression following disruption of proteasome function by rpt-5 ( RNAi ) in various mutant backgrounds . Scale bars 100 µm . ( b ) Table showing growth vs . arrest phenotypes of various mutants in the presence of bortezomib or upon rpn-10 RNAi . For RNAi experiments , L1 animals were incubated for 3 days on indicated RNAi plates , and scored for developmental arrest ( +; normal development , Lva; larval arrest ) . For bortezomib experiments , ~15 L1 animals were incubated for 4 days in liquid cultures containing varying concentrations of bortezomib , and scored for developmental progression . The number ( n ) of replicate bortezomib experiments performed for each genotype is shown on the right . Each colored rectangle is divided into equal parts to show results from each replicate . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 00410 . 7554/eLife . 17721 . 005Figure 1—figure supplement 1 . skn-1-dependent activation of rpt-3::gfp in pbs-5 ( mg502 ) mutants . Fluorescence images show rpt-3::gfp expression in animals grown for 3 days at 20°C . Scale bar 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 005 Unlike the wild type , mutants defective in ERAD , or lacking DDI-1 , arrested or delayed larval development in the presence of low doses of bortezomib ( Figure 1b ) . We analyzed independently derived alleles of png-1 and ddi-1 , indicating that hypersensitivity to proteasome inhibition is unlikely to be a consequence of linked background mutations . png-1 animals consistently showed the most severe defect , and were unable to grow in the presence of very low concentrations of bortezomib . Consistent with their drug sensitivity , mild disruption of proteasome function by RNAi-mediated depletion of the non-essential proteasome subunit RPN-10 causes a synthetic larval lethal phenotype in animals mutant for png-1 . The bortezomib sensitivity of ddi-1; sel-11 double mutants was not enhanced compared to that of ddi-1 single mutants , suggesting that ddi-1 and ERAD factors act in the same genetic pathway . We conclude that ERAD and DDI-1 are required for transcriptional upregulation of proteasome subunits and survival during proteasome dysfunction . Given the defective activation of rpt-3::gfp , a direct target of SKN-1 , it is likely that upon proteasome disruption , these factors are required to activate SKN-1 . The skn-1 gene generates 3 protein isoforms using alternative transcription start sites ( SKN-1A , SKN-1B , and SKN-1C; Figure 2a ) . The isoforms share an identical C-terminal DNA binding domain , but differ in their N-termini . skn-1 ( RNAi ) targets sequences common to all three transcripts . Both skn-1 ( RNAi ) , and the skn-1 ( zu67 ) mutation cause defective responses to proteasome dysfunction ( NL unpublished ) . skn-1 ( zu67 ) is a nonsense allele affecting an exon shared by SKN-1A and SKN-1C , but that does not affect SKN-1B , suggesting SKN-1A and/or SKN-1C are required . SKN-1C encodes a 61 kD protein expressed specifically in the intestine , and SKN-1A encodes a 71 kD protein expressed in most tissues ( An and Blackwell , 2003; Bishop and Guarente , 2007; Staab et al . , 2014 ) . SKN-1A differs from SKN-1C solely by the presence of 90 additional amino acids at the N-terminus that includes a predicted transmembrane domain ( Figure 2b ) , and SKN-1A has been found to associate with the ER ( Glover-Cutter et al . , 2013 ) . 10 . 7554/eLife . 17721 . 006Figure 2 . SKN-1A , a transmembrane-domain-containing isoform of SKN-1 mediates transcriptional responses to proteasome disruption . ( a , b ) Schematic of the ( a ) skn-1 locus and ( b ) SKN-1A protein . In ( a ) , the CRISPR-induced skn-1a-specific mutations are indicated . ( c , d ) rpt-3::gfp induction in wild type and isoform-specific skn-1 mutants in ( c ) the pbs-5 ( mg502 ) mutant , or ( d ) rpt-5 ( RNAi ) . Scale bars 100 µm . ( e ) Developmental arrest of isoform-specific skn-1 mutants exposed to mild proteasome disruption by rpn-10 ( RNAi ) but not on control RNAi . Scale bar 1 mm . ( f ) Expression and localization of functional SKN-1A::GFP fusion protein after proteasome disruption by rpt-5 ( RNAi ) . Apparent GFP signal in control treated animals is background auto-fluorescence . Scale bar 10 µm . ( g ) No developmental arrest of skn-1 mutants carrying an isoform-specific skn-1a::gfp transgene , and exposed to mild proteasome disruption by rpn-10 ( RNAi ) . Scale bar 1 mm . ( h ) Table showing growth vs . arrest phenotypes of skn-1a mutants in in the presence of bortezomib . All skn-1a alleles are identical in their effect on skn-1a coding sequence ( G2STOP ) . Experiments performed identically to those shown in Figure 1b , and data for the wild type from Figure 1 are shown for reference . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 00610 . 7554/eLife . 17721 . 007Figure 2—figure supplement 1 . SKN-1A[∆DBD]::GFP is stabilized upon proteasome disruption . ( a ) Fluorescence images comparing expression and localization of SKN-1A[∆DBD]::GFP under control conditions and following rpt-5 ( RNAi ) . SKN-1A[∆DBD]::GFP is not detected in the absence of proteasome disruption , but accumulates to high levels in both nucleus and cytoplasm following proteasome disruption . Scale bar shows 10 µm . ( b ) fluorescence images comparing levels of SKN-1A::GFP ( left panels ) and SKN-1A[∆DBD]::GFP ( right panels ) upon proteasome disruption . SKN-1A[∆DBD]::GFP reaches much higher levels than the full length fusion protein . Scale bar shows 10 µm . All images show animals raised for two days on the indicated RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 007 We used CRISPR/Cas9 to generate an isoform-specific genetic disruption of SKN-1A , by introducing premature stop codons to the skn-1a specific exons of the skn-1 locus ( hereafter referred to as skn-1a mutants ) . Homozygous skn-1a mutant animals are viable , and under standard conditions show a growth rate and fertility indistinguishable from the wild type . However , skn-1a mutant animals fail to activate rpt-3::gfp in the pbs-5 ( mg502 ) mutant background , or upon RNAi of essential proteasome subunit genes , or exposure to bortezomib ( Figure 2c , d , data not shown ) . We note that in these experiments skn-1a mutants failed to activate rpt-3::gfp in all tissues , including the intestine , where SKN-1C is expressed . Consistent with the failure to upregulate rpt-3::gfp , skn-1a mutants show larval lethality when proteasome dysfunction is induced by rpn-10 ( RNAi ) or treatment with a low dose of bortezomib ( Figure 2e , h ) . These skn-1a mutations specifically affect SKN-1A , but leave SKN-1B and SKN-1C unaltered , indicating that SKN-1A is essential for normal responses to proteasome disruption and in the absence of SKN-1A , the other isoforms are not sufficient . A number of stimuli that trigger stabilization and nuclear accumulation of a transgenic SKN-1C::GFP fusion protein are known , but relatively little is known about whether these stimuli also affect SKN-1A ( Blackwell et al . , 2015 ) . We used miniMos transgenesis ( Frøkjær-Jensen et al . , 2014 ) to generate genomically integrated single-copy transgenes that expresses C-terminally GFP-tagged full length SKN-1A ( SKN-1A::GFP ) , and a second C-terminally GFP tagged truncated SKN-1A that lacks the DNA binding domain ( SKN-1A[∆DBD]::GFP ) . When driven by the ubiquitously active rpl-28 promoter , we did not observe accumulation of SKN-1A::GFP or SKN-1A[∆DBD]::GFP , consistent with constitutive degradation of these fusion proteins . Upon disruption of proteasome function , we observed stabilization and nuclear localization of SKN-1A::GFP and SKN-1A[∆DBD]::GFP in many tissues ( Figure 2f , Figure 2—figure supplement 1a ) . For unknown reasons , the SKN-1A[∆DBD]::GFP transgene accumulated to higher levels than the full length transgene ( Figure 2—figure supplement 1b ) . We generated similar transgenes to express tagged full length and truncated SKN-1C , but did not observe any effect of proteasome disruption ( data not shown ) . These data suggest proteasome dysfunction triggers activation of SKN-1A , but not SKN-1C . We introduced the SKN-1A::GFP transgene into the skn-1a ( mg570 ) and skn-1 ( zu67 ) mutant backgrounds . SKN-1A::GFP rescued the maternal effect lethal phenotype of skn-1 ( zu67 ) . SKN-1A::GFP also restored wild-type resistance to proteasome disruption , as assayed by growth on rpn-10 ( RNAi ) ( Figure 3 ) , or growth in the presence of low concentrations of bortezomib ( data not shown ) . This indicates that the SKN-1A::GFP fusion protein is functional , and that SKN-1A::GFP is sufficient for normal responses to proteasome dysfunction even in the absence of SKN-1C ( which is disrupted by the zu67 allele ) . As such , the transmembrane-domain-bearing SKN-1A isoform is necessary and sufficient for responses to proteasome dysfunction . 10 . 7554/eLife . 17721 . 008Figure 3 . ERAD is required for constitutive SKN-1A degradation , and for activation of SKN-1A upon proteasome disruption . ( a ) Western blot showing expression and post-translational processing of SKN-1A[∆DBD]::GFP in ERAD mutant animals , treated with either solvent control ( DMSO ) or 5 ug/ml bortezomib . SKN-1A[∆DBD]::GFP is only detected in wild-type animals upon bortezomib exposure; a major band at ~70 kD and a minor band at ~90 kD are detected . In ERAD defective mutants , the ~90 kD band is strongly detected under all conditions , and the ~70 kD band appears only following bortezomib treatment . Actin is used as a loading control . ( b ) Expression and localization of SKN-1A::GFP in wild type and sel-1 and sel-11 ERAD defective mutants after proteasome disruption by rpt-5 ( RNAi ) . In ERAD defective mutants , SKN-1A::GFP fails to localize to the nucleus . Scale bar 10 µm . ( c ) Expression and localization of SKN-1A::GFP in wild type and png-1 mutants after proteasome disruption by rpt-5 ( RNAi ) . In png-1 mutants , SKN-1A::GFP is able to localize to the nucleus , although at reduced levels compared to the wild-type . Scale bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 008 Mutation of ddi-1 does not enhance the sensitivity of skn-1a mutants to bortezomib , suggesting that DDI-1 acts through SKN-1A to promote resistance to proteasome inhibitors ( Figure 2h ) . Additionally removing SEL-11 weakly enhanced the bortezomib sensitivity of ddi-1 skn-1a double mutants , and also caused occasional growth defects even in the absence of proteasome disruption , suggesting that ERAD promotes resistance to proteasome inhibitors largely , but not solely , through regulation of SKN-1A . We examined how ERAD factors regulate SKN-1A using the SKN-1A::GFP transgenes . sel-1 and sel-11 mutants accumulate high levels of SKN-1A[∆DBD]::GFP even in the absence of proteasome inhibitors , showing that SKN-1A is constitutively targeted for proteasomal degradation via ERAD ( Figure 3a ) . Upon proteasome disruption , sel-1 and sel-11 mutants show defects in SKN-1A::GFP nuclear localization consistent with defective release from the ER ( Figure 3b ) . Following proteasome disruption in png-1 mutants SKN-1A::GFP localizes to the nucleus , indicating PNG-1 acts downstream of release from the ER ( Figure 3b ) . Lower levels of SKN-1A::GFP accumulate in the nuclei of png-1 mutants than in the wild type , but this mild effect is unlikely to fully account for the severely defective responses to proteasome inhibition in png-1 mutant animals , suggesting retention of glycosylation modifications normally removed by PNG-1 likely disrupts SKN-1A’s nuclear function . These data suggest that activation of ER-associated and N-glycosylated SKN-1A is required for responses to proteasome dysfunction . To examine the expression and subcellular localization of the DDI-1 protease , we used miniMos to generate a single copy integrated transgene expressing full length DDI-1 fused to GFP at the N-terminus , under the control of the ddi-1 promoter . The GFP::DDI-1 fusion protein is expressed in most tissues and shows diffuse cytoplasmic and nuclear localization under control conditions , and can rescue a ddi-1 mutant ( see below ) . Following disruption of proteasome function by rpt-5 ( RNAi ) , GFP::DDI-1 expression is dramatically induced , and GFP::DDI-1 is enriched in nuclei ( Figure 4a ) . We used CRISPR/Cas9 to modify the ddi-1 locus to incorporate an HA epitope tag near the N-terminus of endogenous DDI-1 . Following bortezomib treatment of ddi-1 ( mg573[HA::ddi-1] ) animals , we observed strong upregulation ( greater than 10-fold , based on blotting of diluted samples ) of the HA-tagged endogenous DDI-1 ( Figure 4b ) . The ddi-1 promoter contains a SKN-1 binding site ( Niu et al . , 2011 ) . Upregulation of GFP::DDI-1 by rpt-5 ( RNAi ) is greatly reduced in skn-1a ( mg570 ) mutants ( Figure 4c ) , suggesting that DDI-1 upregulation is mostly mediated by SKN-1A . The remaining weaker ddi-1 upregulation in the skn-1a mutant may represent a second skn-1a-independent mechanism that couples DDI-1 levels to proteasome function . 10 . 7554/eLife . 17721 . 009Figure 4 . DD1-1 is upregulated upon proteasome disruption . ( a ) A functional GFP::DDI-1 fusion protein is strongly induced and localizes to the nucleus upon proteasome disruption by rpt-5 ( RNAi ) . Scale bar 20 µm . ( b ) Western blot showing induction of HA-tagged endogenous DDI-1 upon proteasome disruption by bortezomib . Actin is used as a loading control . ( c ) Induction of GFP::DDI-1 upon proteasome disruption by rpt-5 ( RNAi ) is lost in skn-1a mutants . Scale bar 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 009 The EMS-induced ddi-1 missense alleles that cause failure to activate rpt-3::gfp are clustered within the aspartic protease domain of DDI-1 , and affect conserved residues that are thought to form the substrate-binding pocket of the enzyme ( Figure 5a , b ) , suggesting that the protease activity of DDI-1 is required ( Sirkis et al . , 2006 ) . We used CRISPR/Cas9 mutagenesis to generate a protease dead mutant containing two amino acid substitutions at conserved residues of the catalytic motif , including the aspartic acid residue that forms the active site ( D261N , G263A ) . We additionally isolated a CRISPR-induced deletion that deletes most of the aspartic protease domain and introduces a frameshift , which we presume to be a null allele . Both mutations cause a similar , strong defect in rpt-3::gfp activation by the pbs-5 ( mg502 ) mutant , or upon proteasome RNAi , and cause a similar sensitivity to bortezomib ( Figure 5c , data not shown ) . 10 . 7554/eLife . 17721 . 010Figure 5 . The DDI-1 aspartic protease is required for proteolytic activation of SKN-1A . ( a ) Schematic of the DDI-1 protein showing residues affected by EMS-induced loss of function alleles . ( b ) Multiple alignment of the aspartic protease domain of DDI-1 . Red text indicates residues affected by mutations that disrupt rpt-3::gfp activation , blue text indicates the DS/TGAQ catalytic motif . ( c ) rpt-3::gfp induction in the pbs-5 ( mg502 ) mutant background ( left panel ) , compared to animals carrying a ddi-1 deletion ( middle panel ) , or a point mutation affecting the catalytic motif ( right panel ) . Scale bars 100 µm . ( d ) Localization of SKN-1A::GFP in wild-type and ddi-1 mutant animals following disruption of proteasome function by rpt-5 ( RNAi ) . SKN-1A::GFP nuclear localization is intact in the ddi-1 mutant . In some gut cells of ddi-1 mutant animals SKN-1A::GFP localizes to puncta that are not detected in the wild type . Scale bars 10 µm . ( e ) Localization of SKN-1A[∆DBD]::GFP in gut nuclei of wild-type and ddi-1 mutant animals . Nuclear foci of SKN-1A[∆DBD]::GFP are found in ddi-1 mutants , but not wild type . ( f ) Western blot showing expression and processing of SKN-1A[∆DBD]::GFP in ddi-1 mutant animals , treated with either solvent control ( DMSO ) or 5 ug/ml bortezomib , and blotted for GFP . In the ddi-1 mutant animals , the major band detected is ~30 kD larger than in the wild type . ( g ) Western blot showing expression and processing of HA::SKN-1A:GFP in ddi-1 mutants animals , treated with either solvent control ( DMSO ) or 5 ug/ml bortezomib , and blotted for HA . In the wild type , a ~20 kD band is detected in animals exposed to bortezomib . In ddi-1 mutants this low molecular weight fragment is absent , and a ~110 kD band is detected . In ( f ) and ( g ) ddi-1 mutations were ddi-1 ( mg571 ) [deletion] or ddi-1 ( mg572 ) [active site] and actin is used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 01010 . 7554/eLife . 17721 . 011Figure 5—figure supplement 1 . The aspartic protease , but not the N- or C-terminal domains of DDI-1 are essential for resistance to bortezomib . ( a ) Schematic showing wild type and mutant DDI-1 proteins tested . Each protein was expressed under the ddi-1 promoter and fused to GFP at the N-terminus . ( b ) Table showing growth vs . arrest phenotypes of ddi-1 ( mg571 ) mutants expressing wild-type or mutant GFP::DDI-1 in the presence of bortezomib . ~15 L1 animals were incubated for 4 days in liquid cultures containing presence of varying concentrations of bortezomib ( as shown in Figure 1 ) , and scored for developmental arrest . The number ( n ) of replicate bortezomib experiments performed for each genotype is shown on the right , for each condition the corresponding colored rectangle is divided into n equal parts and shows the result from each replicate . Data for the wild type and ddi-1 ( mg571 ) from Figure 1 are shown for reference . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 011 S . cervisiae Ddi1 contains an N-terminal ubiquitin-like ( UBL ) domain and a C-terminal ubiquitin-associated ( UBA ) domain , but these domains are not detected by standard protein sequence comparisons with C . elegans DDI-1 . To address the possibility that UBL or UBA domains with highly divergent sequence may be present in DDI-1 , we generated N-terminally truncated ( ∆N ) , and C-terminally truncated ( ∆C ) gfp::ddi-1 transgenes . We tested their ability to rescue the bortezomib sensitivity phenotype of ddi-1 ( mg571 ) alongside wild-type gfp::ddi-1 and an aspartic protease active site ( D261N ) mutant . The active site mutation abolished rescue by the gfp::ddi-1 transgene , whereas the ∆N and ∆C truncated transgenes restored bortezomib sensitivity to near wild-type levels ( Figure 5—figure supplement 1 ) . These data are consistent with the lack of conservation of the UBL and UBA domains of DDI-1 , and confirm the essential role of DDI-1 aspartic protease activity . In animals lacking DDI-1 , SKN-1A::GFP localizes at normal levels to the nucleus upon proteasome disruption by rpt-5 ( RNAi ) , suggesting that DDI-1 regulates SKN-1A function after nuclear localization of the transcription factor ( Figure 5d ) . We noticed that SKN-1A::GFP occasionally showed abnormal localization within gut nuclei of ddi-1 mutants , accumulating in highly fluorescent puncta . We observed this defect for both SKN-1A::GFP and SKN-1A[∆DBD]::GFP , indicating that the DBD of SKN-1A is not required for this mis-localization ( Figure 5e ) . As in the wild type , SKN-1A[∆DBD]::GFP does not accumulate in the absence of proteasome disruption in ddi-1 mutants , indicating that the DDI-1 peptidase does not participate in constitutive degradation of SKN-1A by the proteasome ( Figure 5f ) . SKN-1A[∆DBD]::GFP accumulates to similar levels upon proteasome disruption by bortezomib in wild-type and ddi-1 mutants , but in ddi-1 mutants is ~20 kD larger than in the wild type , and approximates the expected size of SKN-1A[∆DBD]::GFP . To test whether these differences reflect DDI-1-dependent proteolytic processing of SKN-1A , we generated a transgene that expresses full length SKN-1A with an N-terminal HA tag and a C-terminal GFP tag ( HA::SKN-1A::GFP ) . The expression , localization and rescue activity of the dually tagged fusion protein is indistinguishable from that of the full length SKN-1A::GFP transgene . In wild-type animals carrying the HA::SKN-1A::GFP transgene , Western blotting for the HA tag reveals a ~20 kD band that accumulates specifically upon proteasome disruption by bortezomib treatment . In ddi-1 deletion or active site mutants , a ~110 kD protein accumulates upon proteasome disruption , equivalent in size to full-length HA::SKN-1::GFP ( Figure 5g ) . As such , SKN-1A is cleaved at a position approximately 20 kD from the N-terminus , and the protease active site of DDI-1 is required for this cleavage . In sel-1 and sel-11 mutants , unprocessed HA::SKN-1A::GFP is present in both control and bortezomib treated animals , with some processed HA::SKN-1A::GFP appearing upon bortezomib treatment ( Figure 6a ) . In png-1 mutant animals HA::SKN-1A::GFP is incompletely processed following proteasome disruption ( Figure 6b ) , suggesting that DDI-1 dependent cleavage of SKN-1A normally occurs after ER trafficking and deglycosylation . Further , in wild-type animals , an HA::SKN-1A::GFP transgene that lacks the putative transmembrane domain is not targeted for constitutive degradation , and is not subject to detectable proteolytic cleavage upon proteasome disruption ( HA::SKN-1A[∆TM]::GFP; Figure 6c ) . Although expressed , HA::SKN-1A[∆TM]::GFP is unable to rescue the bortezomib sensitivity of skn-1a mutants ( data not shown ) . These results indicate that ER trafficking is needed to target SKN-1A for later cleavage and activation by ddi-1 . 10 . 7554/eLife . 17721 . 012Figure 6 . Proteolytic processing of SKN-1A occurs downstream of ER trafficking . ( a ) Western blot showing expression and processing of HA::SKN-1A::GFP upon proteasome disruption by bortezomib in sel-1 and sel-11 mutant animals . ERAD-defective mutants accumulate similar levels of uncleaved ( ~110 kD ) HA::SKN-1A::GFP in both absence and presence of bortezomib . A relatively small amount of the ~20 kD cleavage product accumulates only upon exposure to bortezomib . ( b ) Western blots comparing expression and processing of HA::SKN-1A::GFP upon proteasome disruption by bortezomib in wild-type and png-1 mutant animals . Both cleaved ( ~20 kD ) and uncleaved ( smear ~100-150 kD ) HA::SKN-1A::GFP accumulates upon bortezomib treatment in png-1 mutants . ( c ) Western blot showing expression and processing of HA::SKN-1A[∆TM]::GFP upon proteasome disruption by bortezomib , in otherwise wild-type animals . Uncleaved ( ~100 kD ) HA::SKN-1A[∆TM]::GFP is detected at similar levels under both conditions suggesting the protein is not subject to proteasomal degradation , and a low molecular weight cleavage product is not detected under either condition . For each experiment , mixed stage cultures were treated with either solvent control ( - ) , or 5 ug/ml bortezomib ( + ) prior to collection for SDS-PAGE . HA::SKN-1A::GFP/ HA::SKN-1A[∆TM]::GFP is detected by anti-HA antibodies , Actin is used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 01210 . 7554/eLife . 17721 . 013Figure 6—figure supplement 1 . Model showing post-translational processing of SKN-1A by DDI-1 . SKN-1A associates with the ER via its N-terminal transmembrane domain . In the ER , SKN-1A is subject to glycosylation , and possibly other post-translational modifications . ERAD-dependent release of SKN-1A from the ER targets SKN-1A for proteasome-mediated degradation . When the proteasome is functional ( shown on the left half of the model ) , SKN-1A is rapidly degraded by the proteasome . Under these conditions , the DDI-1 protease is expressed at low levels and localizes largely to the cytoplasm . During proteasome dysfunction ( shown on the right half of the model ) , SKN-1A is not efficiently degraded . Stabilized SKN-1A is deglycosylated by PNG-1 ( this deglycosylation might take place in the nucleus or cytoplasm ) and localizes to the nucleus . Proteasome dysfunction also leads to increased expression and nuclear localization of DDI-1 ( this is in part due to regulation of ddi-1 transcription by SKN-1A , but likely involves additional SKN-1A-independent mechanisms ) . Once in the nucleus , SKN-1A is cleaved by DDI-1 to produce the active truncated form of SKN-1A , allowing upregulation of proteasome subunits and other downstream genes . Cleavage of SKN-1A is disrupted by mutations that affect ERAD , or if SKN-1A lacks its transmembrane domain , suggesting that DDI-1-dependent cleavage occurs downstream of ER-trafficking and may require post-translational modifications to SKN-1A acquired during this process . DOI: http://dx . doi . org/10 . 7554/eLife . 17721 . 013 Our genetic screen for mutants that fail to activate SKN-1 and dissection of the isoform-specific role of SKN-1A reveals the molecular details of proteasome surveillance . We show that SKN-1A is an ER associated protein that is normally targeted by the ERAD pathway for proteasomal degradation in the cytoplasm . Mutations affecting ERAD genes sel-1/SEL1/HRD3 and sel-11/HRD1 , stabilize SKN-1A , but also disrupt localization of SKN-1A following proteasome disruption , due to a failure to efficiently release SKN-1A from the ER . Our data argues that following release from the ER , SKN-1A must be deglycosylated by PNG-1 and cleaved by DDI-1 to become fully active Figure 6—figure supplement 1 . The bortezomib sensitivity defect of png-1 mutants is similar in strength to skn-1a mutants , and both png-1 and skn-1a mutations are synthetic lethal with rpn-10 ( RNAi ) . These similarities suggest that SKN-1A activity may be completely abolished in the absence of PNG-1; likely as a result of failure to deglycosylate SKN-1A following its retrotranslocation from the ER , although we cannot rule out the possibility that deglycosylation of other proteins contributes indirectly to SKN-1A function . Surprisingly , we were unable to generate png-1; skn-1a double mutants , apparently due to lethality of double mutant embryos ( NL unpublished ) . This indicates that in png-1 mutant animals , SKN-1A ( likely in a glycosylated state ) retains a function that is essential for development . This suggests that SKN-1A is not only important for proteasome homeostasis , but is also important for cellular homeostasis upon disruption of glycoprotein metabolism . Consistently , skn-1 mutants are hypersensitive to tunicamycin , an inhibitor of protein glycosylation ( Glover-Cutter et al . , 2013 ) . NGLY1 , the human PNG-1 orthologue , plays important roles in human development; NGLY1 deficiency , a recently described genetic disorder of protein deglycosylation , is caused by mutations at the NGLY1 locus ( Enns et al . , 2014 ) . Failure to deglycosylate Nrf1 , and consequent defects in proteasome homeostasis , likely contributes to the symptoms associated with NGLY1 deficiency . As such our work identifies a pathway that may be targeted for treatment of NGLY1 deficiency - genetic screens for suppressors of the defective proteasome gene regulation of png-1 mutants could indicate targets for drug development . The expression of DDI-1 is dramatically responsive to proteasome inhibition , indicating that its synthesis or stability is coupled to surveillance of proteasome dysfunction . ChIP analysis of SKN-1 shows binding at the promoter element of the DDI-1 operon ( Niu et al . , 2011 ) , suggesting that DDI-1 upregulation may occur via SKN-1 mediated transcriptional regulation , and we observed a defect in the upregulation of GFP::DDI-1 following proteasome disruption in skn-1a mutants . This suggests a positive feedback loop , wherein SKN-1A upregulates DDI-1 , and DDI-1 promotes SKN-1A activation . Positive feedback may ensure a timely and robust response to proteasome inhibition . Mutation of ddi-1 causes defective regulation of rpt-3::gfp and increases sensitivity of C . elegans to proteasome inhibitors . SKN-1A is cleaved at a site ~20 kD from the N-terminus , and ddi-1 is required for this cleavage . The requirement for its catalytic function strongly suggests that DDI-1 is the enzyme directly responsible for cleavage of SKN-1A , although we cannot rule out the possibility that DDI-1 acts upstream of an as-yet unidentified second protease . There are precedents for cascades of proteases , for example caspases in apoptosis , complement cascades in immunology , and thrombin cascades in blood clotting . Our genetic screen for proteasome surveillance defective mutants isolated six independent ddi-1 alleles , but as yet no alleles of any other genes that encode proteases . This argues that either DDI-1 is the only protease in the pathway , or that any other proteases function redundantly or have other essential functions . Uncleaved SKN-1A localizes to the nucleus in ddi-1 mutants , so cleavage of SKN-1A is not essential for its nuclear localization , and SKN-1A cleavage may occur either in the nucleus or cytoplasm . Given that GFP::DDI-1 is largely ( but not exclusively ) nuclear under conditions of proteasome disruption , we speculate that DDI-1 cleavage of SKN-1A is nuclear . In either case , unusually for a membrane-associated transcription factor , SKN-1A is released from the ER by a mechanism that does not require proteolytic cleavage . DDI-1-dependent cleavage therefore activates SKN-1A by some other mechanism ( s ) downstream of ER release . Cleavage of SKN-1A may be required to remove domains in the N-terminus that interfere with its normal function in the nucleus . For example , retention of the hydrophobic transmembrane domain may be disruptive once the protein has been extracted from the ER membrane . Mutant SKN-1A lacking the transmembrane domain is not subject to proteasomal degradation , and is not cleaved by DDI-1 . So , in addition to serving to link SKN-1A levels to proteasome function via ERAD , ER trafficking of SKN-1A is important for subsequent DDI-1-dependent activation . The bortezomib sensitivity of skn-1a mutants ( or skn-1a mutants carrying the transmembrane domain-lacking transgene ) is more severe than that of ddi-1 and ERAD mutants , so ER-association must also promote SKN-1A activation by additional mechanisms . As well as proteasome disruption , skn-1 is implicated in responses to several endocrine and environmental stimuli ( Blackwell et al . , 2015 ) . Modifications such as glycosylation that SKN-1A acquires in the ER may tailor its activity to respond to proteasome dysfunction , identifying these modifications and how they are regulated will be of interest . S . cerevisiae Ddi1 contains an N-terminal UBL domain and a C-terminal UBA domain , This domain architecture is typical of extraproteasomal ubiquitin receptors , which play a role in recruiting ubiquinated proteins to the proteasome ( Tomko and Hochstrasser , 2013 ) . S . cerevisiae Ddi1 binds to both ubiquitin and the proteasome , and participates in the degradation of some proteasome substrates ( Bertolaet et al . , 2001; Gomez et al . , 2011; Kaplun et al . , 2005; Nowicka et al . , 2015 ) , and synthetic genetic interactions with extraproteasomal ubiquitin receptor and proteasome subunit mutants supports a role for Ddi1 in proteasome function ( Costanzo et al . , 2010; Díaz-Martínez et al . , 2006 ) . Although the aspartic protease domain is highly conserved , DDI-1 in C . elegans and related nematodes apparently lacks both UBL and UBA domains , and the UBA domain is not found in mammalian Ddi1 orthologues , so it remains unclear whether Ddi1 orthologues function as extraproteasomal ubiquitin receptors in animals ( Nowicka et al . , 2015 ) . The effect of ddi-1 on development upon proteasome disruption by bortezomib is entirely dependent on skn-1a , indicating that DDI-1 promotes resistance to proteasome disruption via SKN-1A , rather than a general effect on proteasome function . Regardless , it will be of interest to determine whether DDI-1 binds to proteasomes and/or ubiquitin , and whether this affects its function in SKN-1A activation . Activation of the mammalian SKN-1 homologue Nrf1 involves both deglycosylation and proteolytic cleavage , but the enzymes responsible are not known ( Radhakrishnan et al . , 2014; Zhang et al . , 2015 ) . However , a large scale screen for gene inactivations that render cells more sensitive to proteasome inhibitors supports a model that human DDI1 protease also processes Nrf1: DDI2 ( one of two human orthologues of DDI-1 ) and Nrf1 were highly ranked hits in this screen that identified hundreds of gene inactivations that increase sensitivity of multiple myeloma cells to proteasome inhibitors ( Acosta-Alvear et al . , 2015 ) . This suggests that DDI2 is required to cleave and activate Nrf1 in human cells . The site at which Nrf1 is cleaved during proteasome dysfunction has been identified ( Radhakrishnan et al . , 2014 ) , but the primary sequence of this site is not conserved in SKN-1A . Comparisons of SKN-1A with its nematode orthologues reveals conservation at positions consistent with the ~20 kD cleavage product we have observed ( NL unpublished ) . It is possible that DDI-1 and its substrate ( s ) have divergently co-evolved in different lineages . Thus , we suggest that DDI-1 and SKN-1A are core components of a conserved mechanism of proteasome surveillance in animals . Here we have shown that correct post-translational processing of SKN-1A is essential for development if proteasome function is disrupted . Deregulated proteasome function a feature of aging and age-related disease ( Saez and Vilchez , 2014; Taylor and Dillin , 2011 ) . skn-1 is a critical genetic regulator of longevity , and controls lifespan in part through regulation of proteasome function ( Blackwell et al . , 2015; Steinbaugh et al . , 2015 ) As such , the SKN-1A processing pathway described here suggests the mechanism that links SKN-1/Nrf to proteasome function and longevity . Proteasome inhibitors are important drugs in the treatment of multiple myeloma , but relapse and emergence of drug resistant tumors remains a challenge ( Dou and Zonder , 2014 ) . Nrf1 promotes survival of cancerous cells treated with proteasome inhibitors , and activation of this pathway might mediate resistance ( Acosta-Alvear et al . , 2015; Radhakrishnan et al . , 2010; Steffen et al . , 2010 ) . Blocking the activation of proteasome subunit gene expression by Nrf1 has been proposed as a potential strategy to improve effectiveness of proteasome inhibitors in cancer treatment . The conserved SKN-1A/Nrf1 processing factors we have identified , particularly DDI-1 , are ideal targets for such an approach . C . elegans were maintained on standard media at 20°C and fed E . coli OP50 . A list of strains used in this study is provided in Supplementary Table 1 . Mutagenesis was performed by treatment of L4 animals in 47 mM EMS for 4 hr at 20°C . RNAi was performed as described in Kamath and Ahringer ( 2003 ) . The mgIs72[rpt-3::gfp] integrated transgene was generated from sEx15003 ( Hunt-Newbury et al . , 2007 ) , using EMS mutagenesis to induce integration of the extrachromosomal array . Some strains were provided by the CGC , which is funded by NIH Office of Research Infrastructure Programs ( P40 OD010440 ) . sel-11 ( tm1743 ) was kindly provided by Shohei Mitani . png-1 ( ok1654 ) was generated by the C . elegans Gene Knockout Project at the Oklahoma Medical Research Foundation , part of the International C . elegans Gene Knockout Consortium . Genomic DNA was prepared using the Gentra Puregene Tissue kit ( Qiagen , #158689 ) according to the manufacturer’s instructions . Genomic DNA libraries were prepared using the NEBNext genomic DNA library construction kit ( New England Biololabs , #E6040 ) , and sequenced on a Illumina Hiseq instrument . Deep sequencing reads were analyzed using Cloudmap ( Minevich et al . , 2012 ) . Following deep sequencing analysis , a number of criteria were taken into account to identify the causative alleles , as shown in Supplemental file 1 . In many cases , the causative alleles were strongly suggested by the identification of multiple independent alleles for a given gene . Even for those genes only identified by a single allele , the strong functional connection with other independently mutated genes suggests that they are causative ( e . g . isolation of multiple alleles of the sel gene class ) . We also obtained genetic linkage data supporting these assignments for most alleles . For most of the mutants considered , deep sequencing was performed using a DNA from a pool of 20–50 mutant F2s generated by outcrossing the original mutant strain to the parental ( non-mutagenised ) background , which allowed us to use Cloudmap variant discovery mapping to identify the genetic linkage of the causative allele , or linkage was confirmed by testing linkage in crosses with strains carrying precisely mapped miniMos insertions . For ddi-1 and png-1 , we confirmed that disruption by an independent means ( with an independently derived allele ) has the same effect on rpt-3::gfp expression as the EMS-induced mutation . The mg502 allele was isolated in an EMS mutagenesis screen in which mgIs72[rpt-3::gfp] animals were screened for recessive mutations causing constitutive activation of GFP expression . The mutation was identified as described above . The pbs-5 ( mg502 ) lesion is a 122bp deletion in the promoter of CEOP1752 . This operon consists of K05C4 . 2 and pbs-5 . Animals carrying this mutation show constitutive activation of rpt-3::gfp , but have normal growth and fertility under control conditions . Guide RNAs were selected by searching the desired genomic interval for ‘NNNNNNNNNNNNNNNNNNRRNGG’ , using Ape DNA sequence editing software ( http://biologylabs . utah . edu/jorgensen/wayned/ape/ ) . All guide RNA constructs were generated by Q5 site directed mutagenesis as described ( Dickinson et al . , 2013 ) . Repair template oligos were designed as described ( Paix et al . , 2014; Ward , 2015 ) . Injections were performed using the editing of pha-1 ( to restore e2123ts ) or dpy-10 ( to generate cn64 rollers ) as phenotypic co-CRISPR markers ( Arribere et al . , 2014; Ward , 2015 ) . Injection mixes contained 60 ng/ul each of the co-CRISPR and gene of interest targeting Guide RNA/Cas9 construct , and 50 ng/ul each of the co-CRISPR and gene of interest repair oligos . Guide RNA and homologous repair template sequences are listed in Supplemental file 1 . Cloning was performed by isothermal/Gibson assembly ( Gibson et al . , 2009 ) . All plasmids used for transgenesis are listed in Supplemental file 1 . All miniMos constructs were assembled in pNL43 , a modified version of pCFJ909 containing the pBluescript MCS , and are described in more detail below . MiniMos transgenic animals were isolated as described , using unc-119 rescue to select transformants ( Frøkjaer-Jensen et al . , 2014 ) . Low magnification bright field and GFP fluorescence images ( those showing larval growth and rpt-3::gfp expression ) were collected using a Zeiss AxioZoom V16 , equipped with a Hammamatsu Orca flash 4 . 0 digital camera camera , and using Axiovision software . High magnification differential interference contrast ( DIC ) and GFP fluorescence images ( those showing SKN-1A::GFP and GFP::DDI-1 expression ) were collected using a Zeiss Axio Image Z1 microscope , equipped with a Zeiss AxioCam HRc digital camera , and using Axiovision software . Images were processed using ImageJ software . For all fluorescence images , any images shown within the same figure panel were collected together using the same exposure time and then processed identically in ImageJ . Bortezomib sensitivity was assessed by the ability of L1 animals to develop in the presence of a range of concentrations of bortezomib ( LC Laboratories , #B1408 ) . The assays were carried out in liquid culture in ½ area 96 well plates ( Corning , #3882 ) . Each well contained a total volume of 35uL . We mixed ~15 L1 larvae with concentrated E . coli OP50 suspended in S-basal ( equivalent to bacteria from ~200 uL of saturated LB culture ) , supplemented with 50 ug/ml Carbenicillin , and the desired concentration of bortezomib . All treatment conditions contained 0 . 01% DMSO . The plates were sealed with Breathe-Easy adhesive film ( Diversified Biotech , #9123–6100 ) . The liquid cultures were incubated for 4 days at 20°C and then C . elegans growth was manually scored under a dissecting microscope . Growth was scored into three categories: ( 1 ) Normal - indistinguishable from wild type grown in DMSO control , most animals reached adulthood; ( 2 ) Delayed development - most animals are L3 or L4 larvae; ( 3 ) Larval arrest/lethal - all animals are L1 or L2 larvae . For each genotype all conditions were tested in at least 2 replicate experiments . Drug treatments were performed in liquid culture in 6-well tissue culture plates . In each well we mixed C . elegans suspended in S-basal ( ~1000–2000 worms collected from a mixed stage culture grown at 20°C on NGM agar plates ) and E . coli OP50 in S-Basal ( equivalent to E . coli from ~4 mL saturated LB culture ) , supplemented with 50 ug/ml Carbenicillin , and the desired concentration of bortezomib , and made the mixture up to a final volume of 700 ul . All wells contained 0 . 01% DMSO . The tissue culture plates were sealed with Breathe-Easy adhesive film and incubated at 20°C for 7–9 hr . After the treatment the animals were collected to 1 . 5 ml microcentrifuge tubes , washed twice in PBS to remove bacteria and the worm pellet was snap frozen in liquid nitrogen and stored at −80°C . The worm pellet was briefly thawed on ice , mixed with an equal volume of 2x Sample buffer ( 20% glycerol , 120 mM Tris pH 6 . 8 , 4% SDS , 0 . 1 mg/ml bromophenol blue , 5% beta mercaptoethanol ) , heated to 95°C for 10 min , and centrifuged at 16 , 000g for 10 min to pellet debris . SDS-PAGE and western blotting was performed using NuPAGE apparatus , 4–12% polyacrylamide Bis-Tris pre-cast gels ( Invitrogen , #NP0321 ) and nitrocellulose membranes ( Invitrogen , #LC2000 ) according to the manufacturer’s instructions . The following antibodies were used: mouse anti-GFP ( Roche; #11814460001 ) ; HRP-conjugated mouse anti-HA ( Roche , # 12013819001 ) , mouse anti-Actin ( Abcam; #3280 ) . Multiple alignment was performed using Clustal Omega ( www . ebi . ac . uk/tools/clustalo ) .	Proteins perform many important roles in cells , but these molecules can become toxic if they are damaged or are no longer needed . A molecular machine called the proteasome destroys ‘unwanted’ proteins in animal and other eukaryotic cells . If the proteasome stops working properly , unwanted proteins start to accumulate and cells respond by increasing the activity of genes that make proteasomes . A protein called SKN-1 is involved in this response and activates the genes that encode proteasome proteins , but it is not understood how SKN-1 “senses” that proteasomes are not working properly . Here , Lehrbach and Ruvkun used a roundworm called Caenorhabditis elegans to search for new genes that activate SKN-1 when the proteasome’s activity is impaired . The roundworms were genetically engineered to produce a fluorescent protein that indicates when a particular gene needed to make proteasomes is active . Lehrbach and Ruvkun identified some roundworms with mutations that cause the levels of fluorescence to be lower , indicating that SKN-1 was less active in these animals . Further experiments showed that some of these mutations are in genes that encode enzymes called DDI-1 and PNG-1 . DDI-1 is able to cut certain proteins , while PNG-1 can remove sugars that are attached to proteins . Therefore , it is likely that these enzymes directly interact with SKN-1 and alter it to activate the genes that produce the proteasome . More work is now needed to understand the details of how modifying SKN-1 changes its activity in cells . In the future , drugs that target DDI-1 or PNG-1 might be used to treat diseases in which proteasome activity is too high or low , including certain cancers and neurodegenerative diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2016	Proteasome dysfunction triggers activation of SKN-1A/Nrf1 by the aspartic protease DDI-1
It was studied if early suPAR-guided anakinra treatment can prevent severe respiratory failure ( SRF ) of COVID-19 . A total of 130 patients with suPAR ≥6 ng/ml were assigned to subcutaneous anakinra 100 mg once daily for 10 days . Primary outcome was SRF incidence by day 14 defined as any respiratory ratio below 150 mmHg necessitating mechanical or non-invasive ventilation . Main secondary outcomes were 30-day mortality and inflammatory mediators; 28-day WHO-CPS was explored . Propensity-matched standard-of care comparators were studied . 22 . 3% with anakinra treatment and 59 . 2% comparators ( hazard ratio , 0 . 30; 95% CI , 0 . 20–0 . 46 ) progressed into SRF; 30-day mortality was 11 . 5% and 22 . 3% respectively ( hazard ratio 0 . 49; 95% CI 0 . 25–0 . 97 ) . Anakinra was associated with decrease in circulating interleukin ( IL ) −6 , sCD163 and sIL2-R; IL-10/IL-6 ratio on day 7 was inversely associated with SOFA score; patients were allocated to less severe WHO-CPS strata . Early suPAR-guided anakinra decreased SRF and restored the pro-/anti-inflammatory balance . This study was funded by the Hellenic Institute for the Study of Sepsis , Technomar Shipping Inc , Swedish Orphan Biovitrum , and the Horizon 2020 Framework Programme . NCT04357366 . Patients with severe infections caused by the novel coronavirus SARS-CoV-2 ( known as COVID-19 ) have high circulating concentrations of pro-inflammatory cytokines . Although these concentrations are often not as high as in patients with classical acute respiratory distress syndrome ( ARDS ) and septic shock ( Sinha et al . , 2020 ) , systemic inflammation is an important feature of severe COVID-19 . It has been shown that when severe respiratory failure ( SRF ) necessitating mechanical ventilation ( MV ) emerges , two separate immune phenomena predominate in the infected host; ( i ) macrophage activation syndrome involving 25% patients; or ( ii ) complex immune dysregulation with down-regulation of the human leukocyte antigen DR on circulating monocytes , lymphopenia , and over-production of interleukin ( IL ) −6 by monocytes involving 75% of the patients ( Giamarellos-Bourboulis et al . , 2020 ) . It is hypothesized that these immune reactions start early in patients with lower respiratory tract infection ( LRTI ) by SARS-CoV-2 and are progressively enhanced so as to lead to SRF . This has been suggested to be due to the early release of IL-1 from the lung epithelial cells that are infected by the virus; IL-1 stimulates further cytokine production from alveolar macrophages ( Conti et al . , 2020 ) . As a consequence , it is assumed that early start of anti-IL-1 anti-inflammatory treatment may prevent SRF ( van de Veerdonk and Netea , 2020 ) . However , most probably not all patients with LRTI by SARS-CoV-2 are in need of early treatment and a screening tool to capture those who are likely to progress to SRF would be an asset . Soluble urokinase plasminogen activator receptor ( suPAR ) seems to be such a screening tool . We and others recently demonstrated that suPAR concentrations above 6 ng/ml herald worsening to SRF 14 days earlier ( Rovina et al . , 2020; Azam et al . , 2020 ) . The positive predictive value for the early prediction of SRF was as high as 85 . 9% . uPAR is anchored to the cell membranes of the lung endothelial cells . As a result of the activation of kallikrein , uPAR is cleaved and enters the systemic circulation as the soluble counterpart suPAR ( Pixley et al . , 2011 ) . We conducted the SAVE trial ( suPAR-guided Anakinra treatment for Validation of the risk and Early management of SRF by COVID-19 ) to investigate whether the early administration of anakinra in patients with LRTI due to SARS-CoV-2 and suPAR equal to or greater than 6 μg/l may prevent the development of SRF . Anakinra is the recombinant soluble IL-1 receptor antagonist that blocks both IL-1α and IL-1β . We report herein the results of the interim analysis in the first 130 enrolled patients and compare the efficacy of anakinra with parallel patients receiving standard-of-care ( SOC ) treatment . Interim analysis was performed when the 30-day follow-up of the first 130 patients was completed . The first patient was enrolled on April 16 , 2020 and the last on September 12 , 2020 . Registration at the EU repository EudraCT was done on March 31 , 2020 before submission for regulatory approval . Although the study is on-going , we present here the results of the pre-planned interim analysis described in the amendment of the SAVE trial on October 15 , 2020 to the National Organization for Medicines of Greece . The inclusion of 179 parallel SOC comparators was done within the same time frame . The baseline values of Acute Physiology and Chronic Health Evaluation ( APACHE ) II score , Sequential Organ Failure Assessment ( SOFA ) score , pneumonia severity index ( PSI ) , white blood cells , and ferritin of all SOC comparators were different from the 130 participants in the SAVE trial . To match for these differences , propensity score-matching was done and 130 fully matched parallel SOC comparators were selected ( Table 1 ) . The study flow chart is shown in Appendix 1—figure 1 . In all , 211 patients were excluded because they had suPAR less than 6 ng/ml; patients were followed-up and SRF developed in seven ( 3 . 3% ) patients . Baseline demographics of patients receiving anakinra with SOC treatment and of patients receiving only SOC treatment are shown in Table 1; baseline demographics did not differ . The level of care offered in the two groups was also similar ( Appendix 1—table 1 ) . Twenty-nine patients ( 22 . 3%; 95% CI , 16 . 0–30 . 2% ) among the intent-to-treat ( ITT ) population receiving anakinra and SOC progressed to SRF until day 14 . The incidence of SRF among the 130 parallel SOC-treated comparators was 59 . 2% ( n = 77 ) ( 95% CI , 50 . 6–67 . 3% , p<0 . 0001 ) ( hazard ratio 0 . 30; 95% CI , 0 . 20–0 . 46 ) ( Figure 1A and Table 2 ) . The superiority of anakinra treatment was also documented over the total of 179 available SOC parallel comparators ( Figure 1—figure supplement 1 ) . However , the baseline differences between the total SOC parallel comparators and the participants of the SAVE trial led to limit the remaining analysis between the 130 anakinra-treated participants of the SAVE trial and the 130 fully matched SOC parallel comparators . Multivariate step-wise Cox regression analysis for variables showed that anakinra treatment was the only independent variable protective from SRF ( hazard ratio 0 . 28; 95% CI , 0 . 18–0 . 44; p<0 . 0001 ) ( Table 3 ) . The reported higher frequency of dexamethasone intake among patients who developed SRF should not be interpreted as causality; it does simply reflect that the prescription of dexamethasone was greater among patients who were considered more severe by the treating physicians . One separate multivariate step-wise Cox regression analysis among patients treated with dexamethasone showed anakinra to be the only independent variable protective from SRF ( Appendix 1—table 2 ) . Anakinra treatment was of benefit in most of the secondary study endpoints , that is , 30-day mortality; absolute change of SOFA score by day 14; and absolute change of the respiratory symptoms score by days 7 and 14 ( Table 2 ) . Mortality of participants in the SAVE trial receiving anakinra treatment after 30 days was 11 . 5% ( 95% CI , 7 . 1–18 . 2% ) ; this was 22 . 3% ( 95% CI , 16 . 0–30 . 2% ) in parallel comparators receiving SOC ( Figure 1B ) . Multivariate step-wise Cox regression analysis for variables showed that anakinra treatment was the only independent variable protective from 30-day mortality ( hazard ratio 0 . 49; 95% CI 0 . 25–0 . 97; p: 0 . 041 ) ( Appendix 1—table 3 ) . Two main secondary study outcomes were the effects of anakinra treatment on circulating inflammatory biomarkers and on the function of peripheral blood mononuclear cells ( PBMCs ) . Compared to SOC comparators , anakinra-treated subjects experienced increase of the absolute lymphocyte count and decreases of IL-6 , sCD163 , and sIL-2R ( Figure 2 ) . The IL-10/IL-6 ratio of serum ( an index of the anti-inflammatory/pro-inflammatory balance in severe COVID-19 [McElvaney et al . , 2020] ) was inversely associated with the absolute increase of the SOFA score on day 14 among anakinra-treated patients , compatible with the anti-inflammatory effect of anakinra . Remarkably , suPAR was increased among anakinra-treated patients on day 7 from baseline . Thus , anakinra is associated with protection against progression to SRF even when the suPAR signal indicates an unfavorable outcome . The function of PBMCs of patients was modulated among anakinra-treated patients . More precisely , the production of IL-1β and IL-10 on day 7 was greater among patients who did not develop SRF pointing toward a restoration of the ability of the PBMCs to adapt to balanced production of anti-inflammatory and pro-inflammatory cytokines . This was further corroborated with the positive association between the IL-10/IL-1β ratio of production from PBMCs on day 7 with the ratio of serum IL-10/IL-6 of the same day ( Figure 3 ) . The exploratory endpoints were ventilator-free days until day 28 , the 28-day World Health Organization clinical progression scale ( WHO-CPS ) , 90-day mortality , and the cost of hospitalization . The number of ventilator-free days until day 28 was increased with anakinra treatment ( Table 2 ) and 90-day mortality was decreased ( Table 2 and Appendix 1—figure 2 ) . Participants of the SAVE trial were allocated to strata of lower severity by day 28 compared to propensity-matched parallel SOC comparators ( Figure 4 ) . The overall cost of hospitalization decreased from median €2398 . 4 among SOC treated comparators to €1291 . 4 among anakinra-treated patients ( Appendix 1—figure 3 ) . The adverse events ( AEs ) and serious adverse events ( SAEs ) that were captured during the study period of 14 days are listed in Table 4 . The incidence of the same events was depicted among SOC treated comparators . As shown in Table 4 , the incidence of these events was not greater in the anakinra group than comparators , with the only exception of leukopenia having a trend to be higher in the anakinra group . SAEs were fewer among anakinra-treated patients . Anakinra treatment of COVID-19 patients admitted with LRTI and suPAR concentrations greater or equal than 6 μg/l was associated with a relative decrease of the incidence of SRF by 70% . Anakinra-treated patients who were eventually admitted to the ICU had a shorter stay than those who did not receive anakinra . So apparently the benefit of previous anakinra treatment remained . This is also reflected by the overall decrease of mortality by day 30 and by day 90 . An important point of the SAVE study is the strategy to select for early anakinra treatment using the predictive biomarker suPAR . In previous studies in COVID-19 and in sepsis , this marker turned out to be able to predict the likelihood for unfavorable outcome ( Hayek et al . , 2020; Rovina et al . , 2020; Azam et al . , 2020; Giamarellos-Bourboulis et al . , 2012 ) . Other studies have reported favorable effects of anakinra treatment in COVID-19 pneumonia . In a retrospective analysis by Cavalli et al . , 29 patients with respiratory failure and respiratory ratio below 100 mmHg were treated with high dose anakinra intravenously ( 5 mg/kg twice daily ) . Treatment was associated with clinical improvement in 72% and remarkably higher survival rate ( Cavalli et al . , 2020 ) . In the Ana-COVID prospective study , 52 patients with confirmed COVID-19 and bilateral lung infiltrates and oxygen saturation less than 93% were treated with 100 mg anakinra subcutaneously twice daily for 3 days followed by 100 mg subcutaneously once daily for another 7 days . The study had a composite endpoint much similar to the endpoint of the SAVE study; that is , MV and/or death . Anakinra treatment achieved a 78% decrease of this composite endpoint compared to 44 historical controls . However , studied comparators were not matched for co-administered medication such as azithromycin and hydroxychloroquine ( Huet et al . , 2020 ) . In a retrospective study , 12 patients with COVID pneumonia and increased C-reactive protein were intravenously treated with anakinra 300 mg/day; none died ( Cauchois et al . , 2020 ) . It is difficult to compare these three studies as enrolled patients had different stages of COVID and variable disease severity . Hence the clinical dilemma is which patients would benefit from anakinra , in what stage of COVID , and in what dosage regimen . In this respect , the SAVE study gives guidance: the best candidates are patients with high likelihood of SRF as defined by increased suPAR; anakinra presented one well-acceptable safety profile at the standard subcutaneous daily dose of 100 mg . An important additional finding of the SAVE trial is that it provides mechanistic insight into the biological effects of anakinra: the treatment was associated with a reset of the pro- versus anti-inflammatory balance of the host . The production capacity of PMBCs for the anti-inflammatory IL-10 was increased and this was reflected by the serum IL-10/IL-6 ratio . IL-10/IL-6 is an index of the anti-inflammatory/pro-inflammatory balance which is severely disturbed in severe COVID-19 , much more than in bacterial sepsis ( McElvaney et al . , 2020 ) . The described reset in the production capacity of the PBMCs was linked to clinical benefit since the serum IL-10/IL-6 ratio was inversely associated with the absolute change of SOFA score . Anakinra treatment also decreased the elevated serum concentrations of sCD163 and sIL2-R that are biomarkers of macrophage activation ( Bleesing et al . , 2007; Rubin , 1990 ) . It was noted before that patients with COVID-19 who deteriorate have characteristics of macrophage activation ( Zhou et al . , 2020 ) ; the decrease of these biomarkers indicates attenuation of macrophage activation among anakinra-treated patients . Recent data suggest early activation of the NLRP3 inflammasome in severe COVID-19; the formation of the end product caspase-1 was enhanced among patients who had unfavorable outcome . Stimulation of human monocytes with SARS-CoV-2 could not induce production of IL-1β and priming was needed . Findings suggest that over-produced IL-1β is derived from the cleavage of pro-IL-1β induced by SARS-CoV-2 ( Rodrigues et al . , 2021 ) . While anakinra treatment is meant to inhibit IL-1 bioactivity through blocking of its receptor , it may also inhibit the early production of IL-1β following NLRP3 activation due to the autocrine effects of IL-1 ( Dinarello et al . , 1987 ) . Anakinra also inhibits IL-1α . SARS-CoV-2 infection is suggested to cause massive release of IL-1α followed by sensing from monocytes and tissue macrophages and further activation of the NLRP3 inflammasome leading to perpetuation of the pro-inflammatory responses ( Cavalli et al . , 2021 ) . The lack of randomized design is acknowledged as a limitation in study design . The non-randomized design led to two more limitations: the use of SOC parallel comparators and the lack of availability of follow-up samplings on day 7 . The SAVE trial was designed in mid-March 2020 at the beginning of the pandemic in Greece . It was chosen to adapt an open-label and single-arm design in an attempt to help as many patients as possible since no SOC was framed at that time period . The SOC parallel comparators were optimally matched since matching was based on similarities in timeframe of enrolment , level of care , baseline severity , and co-administered treatment . When the study was started , dexamethasone treatment was not yet part of the SOC treatment but in the following months many patients received dexamethasone . Since dexamethasone acts as a cytokine-inhibiting agent , it is an important question how strong would the effect of anakinra be when patients also receive dexamethasone . Although this is not a preset endpoint , post hoc analysis shows that anakinra is also protective in patients receiving dexamethasone ( Appendix 1—table 2 ) . Several meetings held between the investigators addressed the need to adapt one placebo comparator arm of treatment; however , the available data about the efficacy of anakinra did not lead to take the decision of integration of placebo comparators . This interim analysis was presented to the Emergency Task Force for COVID-19 of the European Medicine Agency . Provided advice led to the design of the pivotal , confirmatory phase III randomized clinical trial with the acronym SAVE-MORE ( EudraCT number: 2020-005828-11; Clinicaltrial . gov NCT04680949 ) which is actually running in 40 study sites; 32 sites in Greece and eight sites in Italy . In conclusion , we propose a novel strategy using suPAR as an early biomarker that can effectively identify those patients at high risk for SRF . In these patients , prophylactic treatment with regular doses of anakinra is associated with prevention of the incidence of SRF . The restoration of the pro-inflammatory/anti-inflammatory balance is proposed as the mechanism of anakinra action . SAVE is an ongoing open-label non-randomized trial conducted in six study sites in Greece ( EudraCT number 2020-001466-11; National Ethics Committee approval 38/20; National Organization for Medicines approval ISO 28/20; ClinicalTrials . gov registration NCT04357366 ) . Parallel comparators receiving SOC treatment were hospitalized at the same time period in seven departments of Internal Medicine in tertiary hospitals of Greece who were participating in the registry of the Hellenic Sepsis Study Group ( HSSG ) without participating in the SAVE trial ( Appendix 1—table 1 ) ( http://www . sepsis . gr ) . All consecutive admissions in the six study sites where the SAVE trial was conducted and in the seven study sites where parallel SOC comparators were hospitalized were screened for eligibility . The trial was conducted by the Hellenic Institute for the Study of Sepsis ( HISS ) and funded in part by HISS , by Technomar Shipping Inc , by Swedish Orphan Biovitrum AB and by the Horizon 2020 RISKinCOVID grant . The funders had no role in the design , conduct , analysis and interpretation of data , and decision to publish . The laboratory of Immunology of Infectious Diseases of the 4th Department of Internal Medicine at ATTIKON University General Hospital served as a central laboratory for the study . The initial draft of the manuscript was written by the first and the last authors . All authors vouch for the adherence of the trial to the protocol and first and last authors vouch for the accuracy and completeness of the data and analysis . Enrolled patients were adults hospitalized with confirmed infection by SARS-CoV-2 virus by real-time PCR reaction of nasopharyngeal secretions; radiological findings compatible with LRTI; and plasma suPAR level ≥6 μg/l using the suPARnostic Quick Triage kit ( Virogates S/A , Blokken 45 , 3460 Birkerød , Denmark ) . Exclusion criteria were: any stage IV malignancy; any do not resuscitate decision; ratio of partial oxygen pressure to the fraction of inspired oxygen ( pO2/FiO2 ) less than 150 mmHg; need for MV or non-invasive ventilation under positive pressure ( NIV ) ; any primary immunodeficiency; neutropenia ( <1500/mm3 ) ; any intake of corticosteroids at a daily dose ≥0 . 4 mg/kg prednisone or equivalent the last 15 days; any anti-cytokine biological treatment the last month; and pregnancy or lactation . The procedure to identify appropriate comparators treated in parallel with SOC was as follows: ( a ) patients receiving SOC were hospitalized in seven medical departments that participate in the HSSG ( http://www . sepsis . gr ) . These departments are active in the research network of the HSSG since 2006 where they collaborate with the six medical departments participating in the SAVE trial . They are tertiary departments providing to severe patients SOC treatment regularly updated with the current guidelines; ( b ) the period of hospitalization was exactly the same as the period of hospitalization of the participants in the SAVE trial; ( c ) selection required that patients were meeting all the inclusion criteria of the SAVE trial; and ( d ) selection required that patients did not meet any of the exclusion criteria of the SAVE trial . Clinical data of these patients on the day of hospital admission and start of SOC were used for comparison to the baseline data of participants of the SAVE trial . Written informed consent was provided by the patient or legal representative before screening . Enrolled patients received 100 mg anakinra subcutaneously once daily for 10 days . All other drugs were allowed . Fifteen milliliters of whole blood was collected before start and 7 days after start of anakinra and collected into EDTA-coated tubes and sterile and pyrogen-free tubes for the isolation of PBMCs , serum , and plasma . PBMCs were stimulated for cytokine production . Blood was sampled from parallel SOC comparators on the day of hospital admission and repeated after 7 days; plasma and serum were prepared . Biomarkers and cytokines were measured in plasma , serum , and supernatants of PBMC cultures . PBMCs were isolated after gradient centrifugation over Ficoll ( Biochrom , Berlin , Germany ) for 20 min at 1400 g . After three washings in ice-cold PBS pH 7 . 2 , PBMCs were counted in a Neubauer plate with trypan blue exclusion of dead cells . They were then diluted in RPMI 1640 enriched with 2 mM of L-glutamine , 500 μg/ml of gentamicin , 100 U/ml of penicillin G , 10 mM of pyruvate , 10% fetal bovine serum ( Biochrom ) , and suspended in wells of a 96-well plate . The final volume per well was 200 μl with a density of 2 × 106 cells/ml . PBMCs were exposed in duplicate for 24 hr or 5 days at 37°C in 5% CO2 to different stimuli: 10 ng/ml of Escherichia coli O55:B5 lipopolysaccharide ( LPS , Sigma , St . Louis , USA ) or 5 × 105 colony forming units of heat-killed Candida albicans . F Concentrations of IL-1β , IL-6 , and IL-10 were measured in cell supernatants or serum in duplicate by an enzyme immunoassay ( Invitrogen , Carlsbad , California , USA ) . The lowest detections limits were: for IL-1β 10 pg/ml; for IL-6 10 pg/ml; and for IL-10 5 pg/ml . Concentrations of ferritin ( ORGENTEC Diagnostika GmbH , Mainz , Germany ) , sCD163 ( Affymetrix Inc , Santa Clara , CA ) , and sIL-2R ( ORGENTEC Diagnostika GmbH , Mainz , Germany ) were measured in serum by an enzyme-immunoassay; the lower limit of detection was 75 ng/ml for ferritin; 0 . 31 ng/ml for sCD163; and 0 . 5 ng/ml for sIL-2R . Hospitalization cost was calculated per patient in Euros as the sum of all administered medicines and the addition of the nominal cost of daily stay in the intensive care unit or in the general ward . The unit price for counted items derived from the official pricelist as defined by the Greek government ( KYA 4α/οικ . 13740/27 . 03 . 2012; government gajette 4898 τΒ/01 . 1 . 2018 ) . The cost of human resources ( salaries of nursing and medical personnel ) was not counted . The incidence of SRF by day 14 was the primary outcome . SRF was defined as any decrease of pO2/FiO2 below 150 necessitating MV or NIV ( Giamarellos-Bourboulis et al . , 2020 ) . Patients dying before day 14 were considered meeting the primary endpoint . Secondary outcomes were 30-day mortality; and the changes of respiratory symptoms score ( Ramirez et al . , 2019; Stets et al . , 2019; Barrera et al . , 2016 ) , of SOFA score , of PBMC cytokine stimulation and of circulating plasma inflammatory mediators between days 1 and 7 . Ventilator-free days by day 28 , the 28-day WHO-CPS , 90-day mortality , and the cost of hospitalization were exploratory endpoints . AEs ( Common Terminology Criteria for Adverse Events , version 4 . 03 ) and SAEs were captured . The sample size was calculated assuming the incidence of SRF would decrease from 60% to 45% with anakinra treatment . To achieve so with 90% power at the 5% level of significance , 260 patients were needed . An interim analysis was planned when the first 130 patients would be enrolled . Analysis of the primary endpoint was done by the ITT principle . Qualitative data were presented as percentages with confidence intervals ( CI ) and quantitative data as median with quartiles . Among all parallel comparators receiving SOC treatment , 130 comparators were selected by propensity score 1:1 matching with patients treated with anakinra with SOC . Matching criteria were: age; Charlson’s comorbidity index ( CCI ) ; admission severity scores namely PSI , APACHE II score , SOFA score , and WHO severity classification of COVID-19; and SOC treatment with azithromycin , hydroxychloroquine , and dexamethasone . Comparison with anakinra-treated patients was performed by the Fisher’s exact test using confirmatory forward stepwise Cox analysis ( IBM SPSS Statistics v . 25 . 0 ) . Comparisons of cytokines and mediators between groups were done by the Student’s ‘t-test’ for parametric variables; and by the Mann–Whitney U-test for non-parametric variables . Paired comparisons were done by the Wilcoxon’s rank-signed test . Non-parametric correlations were done according to Spearman . Cost comparisons were done by the Mann–Whitney U-test . Any two-sided p-value < 0 . 05 was statistically significant ( please refer also to Supplementary file 1: Protocol and statistical analysis plan ) .	People infected with the SARS-CoV-2 virus , which causes COVID-19 , can develop severe respiratory failure and require a ventilator to keep breathing , but this does not happen to every infected individual . Measuring a blood protein called suPAR ( soluble urokinase plasminogen activator receptor ) may help identify patients at the greatest risk of developing severe respiratory failure and requiring a ventilator . Previous investigations have suggested that measuring suPAR can identify pneumonia patients at highest risk for developing respiratory failure . The protein can be measured by taking a blood sample , and its levels provide a snapshot of how the body’s immune system is reacting to infection , and of how it may respond to treatment . Anakinra is a drug that forms part of a class of medications called interleukin antagonists . It is commonly prescribed alone or in combination with other medications to reduce pain and swelling associated with rheumatoid arthritis . Kyriazopoulou et al . investigated whether treating COVID-19 patients who had developed pneumonia with anakinra could prevent the use of a ventilator and lower the risk of death . The findings show that treating COVID-19 patients with an injection of 100 milligrams of anakinra for ten days may be an effective approach because the drug combats inflammation . Kyriazopoulou et al . examined various markers of the immune response and discovered that anakinra was able to improve immune function , protecting a significant number of patients from going on a ventilator . The drug was also found to be safe and cause no significant adverse side effects . Administering anakinra decreased of the risk of progression into severe respiratory failure by 70% , and reduced death rates significantly . These results suggest that it may be beneficial to use suPAR as an early biomarker for identifying those individuals at highest risk for severe respiratory failure , and then treat them with anakinra . While the findings are promising , they must be validated in larger studies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "immunology", "and", "inflammation" ]	2021	An open label trial of anakinra to prevent respiratory failure in COVID-19
Methanogenic archaea use a [NiFe]-hydrogenase , Frh , for oxidation/reduction of F420 , an important hydride carrier in the methanogenesis pathway from H2 and CO2 . Frh accounts for about 1% of the cytoplasmic protein and forms a huge complex consisting of FrhABG heterotrimers with each a [NiFe] center , four Fe-S clusters and an FAD . Here , we report the structure determined by near-atomic resolution cryo-EM of Frh with and without bound substrate F420 . The polypeptide chains of FrhB , for which there was no homolog , was traced de novo from the EM map . The 1 . 2-MDa complex contains 12 copies of the heterotrimer , which unexpectedly form a spherical protein shell with a hollow core . The cryo-EM map reveals strong electron density of the chains of metal clusters running parallel to the protein shell , and the F420-binding site is located at the end of the chain near the outside of the spherical structure . [NiFe]-hydrogenases are microbial enzymes that heterolytically cleave H2 , after which the electrons from a hydride ion are reversibly transferred to electron carriers . These enzymes are involved in many metabolic pathways in microbial ecosystems , notably in methanogenesis . The members of the [NiFe]-hydrogenase family are composed of at least two subunits , large ( ∼60 kDa ) and small ( ∼30 kDa ) , which contain a [NiFe] dinuclear metal center and three iron–sulfur ( Fe-S ) clusters , respectively . The family is classified into five distantly related groups based on a phylogenetic analysis of their primary structures ( Vignais and Billoud , 2007; Constant et al . , 2011 ) . The cytoplasmic F420-reducing [NiFe]-hydrogenase ( Frh ) belongs to the group 3 [NiFe]-hydrogenases , which catalyze the reversible reduction of the soluble hydride carrier , NAD ( P ) or F420 . F420 is a deazaflavin derivative that acts as an important hydride acceptor/donor in the central methanogenic pathway . The group 3 hydrogenases contain an additional subunit that interacts with the soluble coenzymes , for example , the iron–sulfur flavoprotein FrhB . To date , the crystal structures of [NiFe]-hydrogenases only from group 1 have been determined: six derived from sulfate-reducing bacteria , including two orthologs harboring a [NiFeSe]-center as active site ( Volbeda et al . , 1995; Higuchi et al . , 1997; Montet et al . , 1997; Garcin et al . , 1999; Matias et al . , 2001; Marques et al . , 2010 ) , one from a photosynthetic bacterium ( Ogata et al . , 2010 ) , and three from hydrogen-oxidizing bacteria and Escherichia coli , which in contrast to the other species are oxygen-tolerant and have an usual [4Fe-3S] proximal cluster ( Fritsch et al . , 2011; Shomura et al . , 2011; Volbeda et al . , 2012 ) . Frh is a key enzyme in the metabolism of methanogenic archaea ( Thauer et al . , 2010 ) . The reduction of carbon dioxide to methane is an overall eight-electron reduction process involving the oxidation of four molecules of H2 by several hydrogenases . Four electrons are provided through the reduced form of the coenzyme F420 , which is regenerated by Frh . Frh is a heterotrimeric enzyme composed of the large subunit FrhA ( 45 kDa ) with a binuclear [NiFe]-center , the small subunit FrhG ( 26 kDa ) with three [4Fe4S] clusters and the iron–sulfur flavoprotein FrhB ( 31 kDa ) with a [4Fe4S] cluster and FAD , and the F420-binding site . Frh accounts for about 1% of the cytoplasmic protein and has been shown by electron microscopy to form a huge complex of similar appearance in all species investigated: Methanococcus voltae ( Muth et al . , 1987 ) , Methanospirillum hungatei ( Sprott et al . , 1987 ) , Methanobacterium thermoautotrophicum ΔH ( Wackett et al . , 1987 ) , and Methanobacterium thermoautotrophicum Marburg ( Braks et al . , 1994 ) . From negative stain images , the complex was interpreted as a flat cylinder consisting of an octamer ( Wackett et al . , 1987 ) or hexamer ( Braks et al . , 1994 ) of the heterotrimer . Recently , developments in instrumentation and image processing software have made it possible to determine structures of large macromolecular complexes to near-atomic resolution by cryo-electron microscopy . Most successful have been studies of large icosahedral viruses ( Yu et al . , 2008 , 2011; Liu et al . , 2010; Wolf et al . , 2010; Zhang et al . , 2010a; Chen et al . , 2011; Settembre et al . , 2011 ) , but also a few smaller complexes in the 1 MDa size range of lower symmetry have yielded high-resolution structures , notably GroEL ( Ludtke et al . , 2008 ) and archaeal chaperonins ( Cong et al . , 2010 , 2012; Zhang et al . , 2010b ) . We present here the structure of the Frh complex from the hydrogenotrophic methanogenic archaeon , Methanothermobacter marburgensis , by cryo-electron microscopy . The Frh complex is a 1 . 2-MDa dodecamer with tetrahedral symmetry . The location of all cofactors were identified , and the backbone of the three proteins was traced , including FrhB that has a novel fold . The Frh complex was highly purified from M . marburgensis under strict anaerobic conditions in the presence of FAD . Its homogeneity was confirmed by electron microscopy with negative staining . We found that all projections of the Frh complex are ring-shaped with a diameter of ∼16 nm ( Figure 1A ) . Class averages showed twofold and threefold symmetry ( Figure 1B ) , which is consistent with tetrahedral symmetry . We conclude that the complexes consist of six dimers of FrhABG arranged into a shell around a hollow core , not in a cylindrical arrangement as in previous EM models ( Wackett et al . , 1987; Braks et al . , 1994 ) . The dodecamer has a total MW of 1 . 2 MDa . Iterative structural refinement of a 84 , 000-particle cryo-EM dataset , applying tetrahedral symmetry , resulted in a 3D reconstruction ( Figure 2 ) at a resolution where the pitch of α-helices and many side chains are recognizable in the map , and β-strands are clearly separated ( Figure 2E–G , 5B ) . The globular protein complex showed no preferred orientations in the ice , resulting in a homogeneous coverage of angles ( Figure 1C ) . The map shows six distinct dimers , all clearly separated from one another ( Figure 2C ) , sitting on the faces of a cube . Each monomer contains a series of five high-density features separated by ∼10 Å , forming an arc in the interior of the protein ( Figure 2D and Video 1 ) . These features are easily assigned to the four Fe-S clusters and the [NiFe] center; the latter is recognized by its smaller size . Although at first glance the position of the FrhABG heterotrimers in the dimer is not obvious , the path of the electron transfer chains provides definitive clues to the location of an FrhABG heterotrimer . FrhABG contains one approximately 35-Å-long helix on the outer surface and a bundle of four up to 40-Å-long helices at the interface of different dimers , running from the outside to the interior of the complex . Other helices are considerably shorter . 10 . 7554/eLife . 00218 . 003Figure 1 . Cryo-electron microscopy and image processing . ( A ) A representative area of an electron micrograph taken at 200 kV on an FEI Polara . The defocus was determined as 2 . 05 µm . The scale bar represents 25 nm . ( B ) Representative class averages and corresponding reprojections of the model for the final refinement iteration . The first and last images represent a view down the twofold and threefold axis of the tetrahedron , respectively . The scale bar represents 10 nm . ( C ) Euler angle distribution for the final reconstruction . Each cylinder represents one class average in the asymmetric triangle ( 1/12 of the tetrahedron ) ; the height of the cylinder is proportional to the number of particles in the class . The equal distribution shows that there are no preferred orientations for the Frh complex . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 00310 . 7554/eLife . 00218 . 004Figure 2 . High-resolution cryo-EM map of Frh . ( A ) view down the twofold axis . Each of the 12 FrhABG heterotrimers is shown in a different color . ( B ) The same view as ( A ) with the two trimers at the front removed , ( C ) view down the threefold axis . ( D ) Close-up of the front dimer in ( A ) shown as a transparent surface with the high densities of the metal centers in red . ( E ) A 10-Å thick slice of the map with a ribbon model for one FrhABG subunit ( the green one in A and D ) superimposed . In this and other figures , FrhA is green , FrhG purple , and FrhB slate-blue . At this level , a complete chain of cofactors can be seen: the [NiFe] cluster in FrhA ( green and brown spheres ) , three FeS clusters in FrhG and one in FrhB ( brown and yellow spheres ) , and the FAD in FrhB ( yellow sticks ) . Details of the map with a full-atom model superimposed: ( F ) Part of the α-helix bundle in FrhA , ( G ) Helix hairpin in FrhA coordinating the [NiFe] center ( green and brown balls ) , ( H ) Part of FrhG and FrhB showing an FeS cluster and FAD . In ( E–H ) , the map was filtered at 4 . 5-Å resolution and sharpened using a B-factor of -54 Å2 . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 00410 . 7554/eLife . 00218 . 005Video 1 . The Frh map at different density levels . The Frh map contains high densities corresponding to the metal clusters . The map , filtered to 5 . 5 Å for clarity , is shown at increasing isosurface levels . At density level 1 ( gray ) , the whole complex is seen; levels 2 ( yellow ) and 3 ( gold ) show decreasing protein density . At level 4 ( orange ) , only a string of five densities remains , corresponding to the four [4Fe4S] clusters and the [NiFe] cluster . The FeS clusters , but not the [NiFe] , are still visible at level 5 and 6 ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 005 The large subunit FrhA has considerable homology to the known large subunit structures ( Figure 3 ) , and is characterized by four long α-helices . Docking the large subunit of the group 1 enzyme into the map provides a good fit for the α-helix bundle , with the [NiFe] center correctly situated in its identified location and many other structural elements of FrhA matching closely , including two three-stranded antiparallel β-sheets . FrhA is about 15 kDa smaller than the large subunits of known structure; all missing elements are located on the periphery . A structural model for FrhA ( Figure 4A ) was created based on this fit and a sequence alignment with the protein from Desulfovibrio vulgaris Hildenborough ( Figure 3 ) . All but the first three amino acids in the sequence could be traced in our maps . Densities for large side chains confirm the structural assignment ( Figures 2 , 5 and Video 2 ) . The [NiFe] center is located near four cysteine residues , Cys 63 , 66 , 380 , and 383 , which are fully conserved in all [NiFe]-hydrogenases . The group 1 [NiFe]-hydrogenases contain another ion , typically either Mg2+ or Fe2+/3+ ( Higuchi et al . , 1997; Garcin et al . , 1999 ) . In our map , there is strong density at this position ( Figure 5D ) , which is part of the structurally conserved region of FrhA , and its surrounding residues , Glu44 , Glu229 , and His386 , are fully conserved in FrhA ( Figure 3A ) , so we interpret this density as an ion . One of the helices of the four-helix bundle has a pronounced kink . This helix was interpreted as residues 166–193 , and a proline residue ( Pro180 ) not present in the group I proteins was found exactly at the position of the kink , supporting the assignment . No density is present beyond His386 , and there is no room in the structure for the final 19 amino acids of the protein . The C-terminal residues after the equivalent histidine in [NiFe]-hydrogenases are enzymatically removed during maturation of the enzymes ( Menon et al . , 1993; Theodoratou et al . , 2005 ) . We can conclude that FrhA is posttranslationally modified in the same way , likely by FrhD , a homologous endopeptidase contained in the Frh operon FrhADGB ( Liesegang et al . , 2010; Thauer et al . , 2010 ) . 10 . 7554/eLife . 00218 . 006Figure 3 . Sequence and secondary structure of FrhA and a comparison with homologous proteins . ( A ) Alignment of the [NiFe]-hydrogenase large subunit from Desulfovibrio vulgaris Hildenborough ( first line ) with Methanothermobacter marburgensis FrhA ( third line ) . The second line shows the identical ( \| ) and similar ( : ) amino acids . The alignment was done with EMBOSS needle ( http://www . ebi . ac . uk/Tools/psa/emboss_needle/ ) and manually adjusted after fitting the FrhA structure . The fourth line shows the consensus sequence of 19 archaeal FrhA species . Identical amino acids in capitals , similar ones in lower case ( h: hydrophobic; s: small ( GAS ) ; l: large ( LIFYHW ) ; a: aromatic ( FYWH ) ; z: T or S; n: negative , D or E; p: positive , R or K ) . The α-helices are highlighted in green , β-strands in blue . The [NiFe] ligands are in orange in the ligands of the third ion in red . ( B ) Comparison of the FrhA model ( rainbow coloring from blue to red ) and the group 1 [NiFe]-hydrogenase large subunit from Desulfovibrio vulgaris Hildenborough ( gray ) ( pdb 2wpn ) ( Marques et al . , 2010 ) in two different orientations ( left and right ) . The [NiFe] center and another ion ( spheres ) overlap as do the structural elements around them , as well as two 3-stranded β-sheets and the lower part of the four-helix bundle . Differences are confined to the periphery . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 00610 . 7554/eLife . 00218 . 007Figure 4 . Models of Frh subunits . ( A ) FrhA , ( B ) FrhG , and ( C ) FrhB in rainbow colors from blue ( N-terminus ) to red ( C-terminus ) . ( D ) Stereo pair of the FrhABG heterotrimer with the string of four bound Fe-S clusters ( orange/yellow ) , the [NiFe] binuclear center ( green/orange ) at the top , and the FAD ( red ) and F420 cofactors ( green ) below . FrhA is green , FrhG is purple , and FrhB is slate-blue . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 00710 . 7554/eLife . 00218 . 008Figure 5 . Details of the model and map . ( A ) The ferredoxin domain of FrhG . The eight cysteine residues surrounding the two [Fe4S4] clusters are shown in green . ( B ) β-Sheet in FrhB . The strands are clearly separated . ( C ) The dimer interface of FrhA inside the particle with two salt bridges Asp312-Arg314 . The two FrhA molecules are viewed from inside the complex . ( D ) The C-terminal three-stranded β-sheet in FrhA with the [NiFe] center ( green and brown ) and another ion ( gray ) near the C-terminal His386 . The conserved residues Glu44 , Glu229 , and His386 coordinate the ion . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 00810 . 7554/eLife . 00218 . 009Video 2 . FrhA . The model of FrhA is shown superimposed on the map . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 009 The fit of the small subunit of the group 1 enzyme reveals structural homology to the N-terminal domain of FrhG including the proximal Fe-S cluster . Some peripheral helices present in the group 1 protein are missing in the FrhG structure , and the similarity is generally lower than for the large subunit ( Figure 6C ) . The N-terminal domain is characterized by a conserved parallel β-sheet flanked by short α-helices ( Figure 4B ) . In the group 1 [NiFe]-hydrogenases , the proximal Fe-S cluster is located at the C-terminal end of the β-sheet and is coordinated by four cysteines . In the primary structure of FrhG , one of the conserved cysteine residues for the coordination of this cluster is replaced by Asp60 ( Figure 6A ) . The bacterial hydrogenases as well as most of the archaeal species have a cysteine at this position , but aspartates have been identified before as ligands of [4Fe4S] clusters ( Calzolai et al . , 1995; Muraki et al . , 2010; Gruner et al . , 2011 ) . Accordingly , we positioned this aspartate near the proximal Fe-S cluster . 10 . 7554/eLife . 00218 . 010Figure 6 . Sequence and secondary structure of FrhG and a comparison with homologous proteins . ( A ) Alignment of the [NiFe]-hydrogenase small subunit from Desulfovibrio vulgaris Hildenborough ( first line ) with Methanothermobacter marburgensis FrhG ( third line ) and ferredoxin from Peptostreptococcus asaccharolyticus ( pdb 1dur ) ( last line ) . The second and fourth lines show the identical ( \| ) and similar ( : ) amino acids . The alignment was done with ClustalW . The α-helices are highlighted in green , β-strands in blue . Ligands of the proximal [4Fe4S] cluster are in yellow , of the medial cluster in orange , and of the distal cluster in red . Amino acids not seen in the Frh structure are in gray font . The first 38 amino acids of the FrhG sequence deduced from the M . marburgensis genome are probably not part of the protein ( see main text ) and are shown in light gray font . ( B ) Consensus sequence of archaeal FrhG species . Capitals: conserved residues; lower case: similar residues ( h: hydrophobic; s: small ( GAS ) ; l: large ( LIFYHW ) ; a: aromatic ( FYWH ) ; z: T or S; n: D or E; p: R or K ) . ( C ) Comparison of the FrhG model ( rainbow coloring from blue to red ) and the [NiFe]-hydrogenase small subunit from Desulfovibrio vulgaris Hildenborough ( pdb 2wpn ) ( gray ) in two different orientations ( left and right ) . The structure near the proximal [4Fe4S] cluster ( blue-green ) is conserved , but the periphery diverges . The ferredoxin domain ( yellow to red ) containing the medial and the distal iron-sulfur clusters is not homologous and the clusters do not overlap . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 010 The C-terminal domain of FrhG has no homology to the group 1 [NiFe]-hydrogenases . Instead , the sequence indicates a ferredoxin domain , with two CxxCxxCxxxC sequences coordinating two [4Fe4S] clusters ( Figure 6A ) ( Alex et al . , 1990 ) . Accordingly , we modeled the C-terminal domain of FrhG on ferredoxin ( Figures 4B , 5A and Video 3 ) . In our model , the residues 235–244 form a β-hairpin with Gly240 at the turn . The two β-strands were predicted and the hairpin is clearly visible in the density , confirming the assignment . Unlike FrhA , which could be traced over its full length , density of FrhG for a 12-residue stretch ( 188–199 ) is missing ( Figure 6A ) . This region is located at the surface and may be disordered . FrhG as deduced from the M . marburgensis genome sequence is 275 residues long ( Liesegang et al . , 2010 ) , but no density is present for the 45 N-terminal amino acids . The frhG gene most probably starts with the initiation codon GTG annotated as Val39 in the genome sequence ( Fox et al . , 1987; Alex et al . , 1990 ) , which means that density is missing in the map for only the first seven amino acids of FrhG . 10 . 7554/eLife . 00218 . 011Video 3 . FrhG . The model of FrhG is shown superimposed on the map . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 011 FrhB contains one Fe-S cluster , an FAD , and the F420-binding site . There is no homolog of known structure . FrhB was therefore traced ab initio , aided by secondary structure predictions ( Figure 7 ) and recognizable features in the map that included 10 α-helices , 11 β-strands , and the Fe-S cluster . The most prominent feature of FrhB is a ∼35 Å surface helix that corresponds to the predicted 24-residue C-terminal helix . 10 . 7554/eLife . 00218 . 012Figure 7 . Sequence and secondary structure of FrhB . Light gray font: No density; green highlight: α-helix; gray highlight: β-strand . Second line: Consensus sequence of FrhB species . Capitals: conserved residues; lower case: similar residues ( h: hydrophobic; s: small [GAS]; l: large [LIFYHW]; a: aromatic [FYWH]; z: T or S; n: negative [D or E]; p: positive [R or K] ) . In the consensus sequence , the residues for coordination of the iron–sulfur cluster , FAD , and F420 are highlighted in orange , cyan , and yellow , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 012 We examined the F420-binding proteins of known structure in the protein databank , and found no structural homology with FrhB . All these proteins have cofactors different from FAD , for example , methylene-H4MPT or NADP ( Ceh et al . , 2009 ) . Also no homologous FAD-binding proteins were found . We conclude that FrhB has a novel fold ( Figure 4C and Video 4 ) . The Fe-S cluster is most probably a 4Fe-4S cluster because it is surrounded by four cysteines ( Cys 104 , 134 , 192 , and 195 ) , which are all fully conserved ( Figure 7 ) . The core of FrhB is a mixed six-stranded β-sheet . A density located ∼8 Å from the Fe-S cluster could be interpreted as the isoalloxazine ring of the FAD cofactor . The pyrophosphate moiety is located at the N-terminal end of an α-helix , which is a common FAD-binding motif where the dipole of the helix compensates the negative charge ( Dym and Eisenberg , 2001 ) ( Figure 8 ) , and nearby density for the adenine moiety is visible . The FAD has a bent conformation , with the adenine and isoalloxazine ring only 4–5 Å apart ( Figure 8 ) near the highly conserved loop Lys75-Tyr76 . The FAD is completely surrounded by highly conserved protein regions ( Figure 7 and Video 5 ) , further verifying the backbone trace . 10 . 7554/eLife . 00218 . 013Video 4 . FrhB . The model of FrhB is shown superimposed on the map . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 01310 . 7554/eLife . 00218 . 014Figure 8 . Electron-transfer chain and F420 . Maps in the absence ( A ) and presence ( B ) of the substrate F420 differ in a region near a conserved loop between two β-strands near the FAD ( carbons in yellow ) . The isoalloxazine ring of F420 fitted into this density is shown in pink . ( C ) Electron transfer chain with minimal distances indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 01410 . 7554/eLife . 00218 . 015Video 5 . Conserved residues in FrhB . The FrhB model is shown with conserved residues in red , similar residues in orange and unconserved residues in gray ( see Figure 7 for sequence ) . FAD is yellow and F420 green . Highly conserved regions surround the FeS cluster and the FAD as well as the F420 isoalloxazine ring . An entrance pathway for F420 is suggested as well . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 015 To identify the substrate-binding site , we imaged the FrhABG complex in the presence of a large excess of F420 , and calculated maps , which were refined to a similar resolution . At all refinement stages , the map showed an additional density in a region that was empty in maps of Frh without substrate ( Figure 8A , B ) . This density runs approximately parallel to the FAD isoalloxazine ring , at a distance of ∼4 Å . We conclude that it represents the isoalloxazine ring of the F420 substrate . It is seated in a large pocket surrounded by highly conserved residues , with easy access to the surface of the complex . The F420 density is close to the small residues S209 and V210 in a loop between two β-strands , which are part of the longest conserved region of FrhB ( Figure 7 ) . A predicted three-stranded β-sheet ( 148–172 ) flanked by two short helices ( Figure 7 ) fits nicely in density . The only conserved region of this sheet , 163IGKGK , forms a turn close to the position of the isoalloxazine ring of F420 . The location between this ring and the surface of the complex suggests that one or both of the conserved lysines may interact with the phosphate group of F420 . The long α-helix on the surface that was interpreted as the C-terminal helix of FrhB is not very well resolved . The residues in the helix are mostly not conserved in FrhB , except at the N-terminal end ( Figure 7 ) . In our model , the conserved residues are located near the loop 164GKGK mentioned above . This region also contains several other conserved residues ( Video 5 ) and likely constitutes the access channel for the substrate F420 . The map of Frh in the presence of substrate is similar to the substrate-free map , indicating that F420 binding does not involve major conformational changes . The residues coordinating the Fe-S cluster , FAD , and F420 are conserved in the FrhB family , which includes subunits of F420-dependent sulfite reductase ( Fsr ) , F420H2:quinone oxidoreductase ( FqoF ) , F420H2:phenazine oxidoreductase ( FpoF ) , F420-dependent glutamate synthase and formate dehydrogenase ( FdhB ) ( Johnson and Mukhopadhyay , 2005 ) ( Figure 9 ) . Some of these proteins contain additional domains , like an N-terminal ferredoxin domain ( which in Frh is part of FrhG ) , or a C-terminal domain containing a binding site for the substrate of the electron transfer from F420 . Clearly , these enzymes share a common fold with FrhB . 10 . 7554/eLife . 00218 . 016Figure 9 . Alignment of the FrhB family of F420-binding proteins . FrhB: FrhB of Methanothermobacter marburgensis DSM 2133; GS: F420-dependent glutamate synthase from M . marburgensis , ADL58239; FqoF: F420H2:quinone oxidoreductase subunit F from Archaeoglobus fulgidus DSM 4304 , NP_070660; FpoF: F420H2:phenazine oxidoreductase subunit F from Methanosarcina barkeri str . Fusaro , YP_303819; Fsr: N-terminal domain of Fsr , F420-reducing sulfite reductase from Methanocaldococcus jannaschii DSM 2661 , Y870_METJA; FdhB: beta subunit of formate dehydrogenase from M . marburgensis , YP_003850414 . Cons: the consensus sequence of the FrhB family . Capitals: conserved residues; lower case: similar residues ( h: hydrophobic; s: small [GAS]; l: large [LIFYHW]; a: aromatic [FYWH]; z: T or S; n: negative [D or E] , p: positive [R or K] , c: charged [D , E , R , K] ) . In the consensus sequence , residues indentified in the FrhB structure for coordination of the iron–sulfur cluster , FAD , and F420 are highlighted in orange , cyan , and yellow , respectively . Secondary structure prediction for each of the proteins was done with PSIPRED ( http://bioinf . cs . ucl . ac . uk/psipred/ ) : green highlight: α-helix; gray highlight: β-strand . Highly conserved amino acids are indicated in bold font . Amino acid numbers are shown in red; note that the C-terminal 83 residues of glutamate synthase ( GS ) align with the N-terminus of FrhB and the N-terminal 182 residues with the C-terminus of FrhB . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 016 Recent advances in instrumentation and image processing procedures have made cryo-electron microscopy of isolated macromolecular complexes a powerful technique in structural biology . After the first virus structure at subnanometer resolution ( Böttcher et al . , 1997 ) showed the feasibility of the method , icosahedral viruses were reconstructed in recent years to better than 4 Å resolution ( Yu et al . , 2008 , 2011 , Zhang et al . , 2008 , 2010a; Liu et al . , 2010; Wolf et al . , 2010; Chen et al . , 2011; Settembre et al . , 2011; ) , taking advantage of the 60-fold symmetry and huge size ( ∼20 to 150 MDa ) , which together contribute to the relatively high signal-to-noise ratio for these particles that makes the determination of the orientation unambiguous . For smaller complexes , the techniques have been pioneered with the 800-kDa GroEL complex with D7 symmetry , reaching subnanometer resolution for the first time in 2004 ( Ludtke et al . , 2004 ) and ∼4 Å in 2008 ( Ludtke et al . , 2008 ) . Subsequent reconstructions to this resolution were achieved for archaeal ( Zhang et al . , 2010b ) and mammalian ( Cong et al . , 2010 ) chaperonins . The chaperonins have different conformational states , and usually only one of these gave a high-resolution map , whereas for the others the resolution was limited due to conformational flexibility . The appearance of our reconstructions is similar to the chaperonins; in agreement with this , Fourier shell correlation indicates a resolution of 3 . 9 and 4 . 0 Å for the two maps , respectively ( Figure 10A ) . Recently it has been noted that the resolution of such reconstructions are usually overestimated due to correlation of noise at high resolution ( Scheres and Chen , 2012 ) . Using a ‘gold standard' procedure to estimate resolution , where two half-data sets are independently refined to a reference not containing high-resolution information ( see ‘Materials and methods' ) , yielded 5 . 5 Å ( Figure 10B ) . However , an objective indication of ( local ) resolution is provided by the appearance of secondary structure: the α-helical pitch of 5 . 4 Å is very well defined in many regions ( Figure 2F , G ) , as is the separation of β-strands ( axial distance ∼4 . 8 Å ) ( Figure 5B ) . Comparison of these areas with maps calculated from the model low-pass filtered to different resolutions suggest a resolution ∼4 . 5 Å . The overall resolution estimate is an average of these well-resolved protein regions and other regions that are less well resolved , especially those at the periphery of the complex . An FSC curve between the experimental map and a map calculated from the model indicated a resolution of 5 . 8 Å ( Figure 10C ) ; this figure also constitutes an average over the whole structure , with the model probably best for internal α-helices and worst for loops at the periphery . Since the model was derived from the map , and not from an independent experiment , this measure does not properly reflect the resolution of the map , but rather describes how well our admittedly incomplete model explains the map densities . It should be noted that we used tight protein structure restraints during modeling , in order to prevent overfitting to noise while creating an unrealistic model that would have yielded a higher model-to-map correlation . 10 . 7554/eLife . 00218 . 017Figure 10 . Resolution estimations and B-factor determination . ( A ) Fourier shell correlation plot , showing the resolution for the substrate-free map ( black ) and the F420-containing map ( red ) . At the 0 . 5 FSC criterion , the resolution is 3 . 9 and 4 . 0 Å , respectively . ( B ) Gold standard FSC between two half-data sets independently refined from a low-resolution model . The resolution at the 0 . 143 FSC criterion is 5 . 5 Å . ( C ) FSC between the map and model . At 0 . 5 FSC , the indicated resolution is 5 . 8 Å . ( D ) Plots of the natural logarithm of the spherically averaged structure factor amplitude as a function of the resolution ( Å−2 ) . Blue: experimental amplitudes; red: Cref-weighted amplitudes; green: amplitudes sharpened with a B-factor of −54 Å2 , as determined by the program embfactor ( Fernández et al . , 2008 ) for the resolution zone 10–4 . 5 Å . The purple line shows the average scattering amplitude for the Frh complex ( √Natoms ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 017 Several factors contributed to the successful structure determination of Frh . The tetrahedral symmetry means that there are 12 asymmetric units in each particle , less than the 14 and 16 in the D7 and D8 symmetric chaperonins , but the resulting ball-like shape ensures that the complex has no preferential orientation in the ice ( Figure 1D ) , unlike the cylinder-shaped chaperonins that tend to be oriented with their symmetry axis parallel or perpendicular to the ice ( Ludtke et al . , 2001; Zhang et al . , 2011 ) . From the quality of the reconstructions , we conclude that the complex is very rigid . The maps of Frh with and without substrate are of similar resolution and quality , implying that any protein conformational changes due to F420 binding are too small to detect at the present resolution . The density for F420 is weak , but its position was inferred at an early stage of the analysis from the fact that maps of Frh without the substrate were consistently empty and maps with F420 always showed some density here . The later tracing of the amino acid chain of FrhB and the FAD confirmed the assignment: F420 is in van der Waals distance from its hydride donor , the FAD isoalloxazine ring , and the residues in the vicinity are highly conserved in FrhB and in the FrhB family of F420-binding proteins ( Figures 7 , 9 and Video 5 ) . While the chain trace for FrhA and FrhG was relatively straightforward using the group I hydrogenase homolog , FrhB was built ab initio . The visibility of the FeS cluster density ( Figure 2D and Video 1 ) made it possible to use the four Fe-S ligand cysteines together with a secondary structure prediction as an initial guide; large side chain densities helped to establish the correct sequence assignment . The secondary structure prediction was fully confirmed , and weak or discontinuous densities in the map were almost invariably found to represent glycine residues , further indicating a correct assignment . Density for the FAD cofactor only became clear when most of the protein chain was placed; density for the ribose and ribitol are missing ( Figure 8 ) . The pyrophosphate group is closely associated with the N-terminal α-helix 26–64 of FrhB . The N-terminus of this helix is one of the most conserved regions in the FrhB family ( Figures 7 , 9 ) , and its role in compensating the negative charge of the phosphates is a common motif in FAD-binding proteins ( Dym and Eisenberg , 2001 ) . Conserved regions of FrhB were found near the FAD and surrounding the F420-binding site and its putative entrance channel ( Video 5 ) . Of the three proteins , only FrhG is missing density for an internal stretch , residues 188–199 ( Figure 6A ) . The flanking residues of this stretch are located less than 10 Å apart on the surface of the complex and they probably form a flexible loop . FrhA and FrhB were traced over their full length . All three proteins follow the rule that hydrophobic residues form the core of the protein and charged residues face the outside . We found several pockets lined with hydrophobic side chains , not just in FrhA and FrhG , where these features are shared with the bacterial protein used as a template for modeling , but also in FrhB . FrhB also contains a highly hydrophobic helix ( 27GIVTGLLAYAL ) , which is buried completely inside the subunit . These features all indicate a correct chain trace . Our structure shows that each heterotrimer encompasses a complete electron transfer chain that runs from the [NiFe] cluster in FrhA via the three Fe-S clusters in FrhG and one Fe-S cluster and FAD in FrhB to F420 ( Figures 4D , 11 and Video 6 ) . The cofactors describe a curvilinear path in the protein interior , almost parallel to the surface of the complex , with distances of ∼10 Å between neighbors , except FAD and F420 , which are in van der Waals contact to enable hydride transfer ( Figure 8C ) . The predicted F420-binding site is located on the outside of the complex , which facilitates access of the cytosolic F420 . The two electron transfer chains in a dimer are separated by at least 17 Å and are thus independent of each other , indicating that the functional unit is one FrhABG heterotrimer and there is no cooperativity in the complex . The three proteins in the Frh heterotrimer form a tight complex with extensive subunit interfaces and a large number of subunit interactions , including two salt bridges between loops of the two FrhA molecules in a dimer , from Asp312 of one FrhA and Arg314 of the other ( Figure 5C ) . The heterotrimer is quite slender compared to the two-subunit [NiFe] hydrogenases of known structure , with a small contact surface especially between FrhB and the other two subunits ( Figure 12 ) , and complex formation may be necessary for stability of the complex and keeping the metal clusters at optimal distances . The same quaternary structure of Frh has been observed not only in the closely related thermophile M . thermautotrophicus ( Wackett et al . , 1987 ) but also in the phylogenetically distant mesophilic M . voltae ( Muth et al . , 1987 ) and M . hungatei ( Sprott et al . , 1987 ) , so it is clear that the formation of a large complex plays an essential role for the function of Frh . 10 . 7554/eLife . 00218 . 018Figure 11 . Model of the dodecameric Frh complex . Each FrhABG heterotrimer is colored differently with the same color scheme as the map in Figure 1 . ( A ) View down the twofold axis , ( B ) view of the FrhB trimer , and ( C ) view of the closely packed FrhA trimer as Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 01810 . 7554/eLife . 00218 . 019Video 6 . The Frh dodecamer . A model of the tetrameric Frh complex . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 01910 . 7554/eLife . 00218 . 020Figure 12 . A comparison of ( A ) Frh and ( B ) the group I [NiFe] hydrogenase from Desulfovibrio gigas ( 2wpn ) ( Marques et al . , 2010 ) shows a lower intersubunit contact area in the former . The large subunit/FrhA is shown in green , the small subunit/FrhG in magenta , and FrhB in blue . The left panel shows Frh in an orientation along the twofold axis of the complex , as seen from the outside , and the right panel a view 90° rotated , as seen from the dimer partner . DOI: http://dx . doi . org/10 . 7554/eLife . 00218 . 020 Methanothermobacter marburgensis ( DSM 2133 ) was obtained from the Deutsche Sammlung von Mikroorganismen ( DSMZ , Braunschweig , Germany ) . The archaeon was grown anaerobically at 65°C on 80% H2/20% CO2/0 . 1% H2S in a 12-L fermenter containing 10 L complete mineral salt medium ( Schönheit et al . , 1980 ) . Cells were harvested by the use of a continuous-flow centrifuge under anoxic conditions at the late exponential phase and stored at −80°C . Purification was performed under strictly anaerobic conditions at 18°C in an anaerobic camper ( Coy Laboratory Products , Grass Lake , MI ) . All buffers used contained 2 mM DTT and 25 µM FAD . Cell extracts were routinely prepared from 20 g ( wet mass ) of M . marburgensis cells . The cells were suspended in 35 ml 50 mM Tris/HCl pH 7 . 6 ( buffer A ) , and passed four times through a French pressure cell at 125 MPa . Cell debris were removed by centrifugation at 15 , 000×g for 30 min . The supernatant , designated cell extract and containing ∼1000 mg protein , was adjusted to 800 mM ammonium sulfate in buffer A and stirred for 20 min . The cell extract was applied to a Phenyl Sepharose 6 Fast Flow column ( 6 × 10 cm ) equilibrated with 800 mM ( NH4 ) 2SO4 in buffer A . Protein was eluted by a ( NH4 ) 2SO4 step gradient in buffer A: 800 mM ( NH4 ) 2SO4 for 250 ml , 200 mM ( NH4 ) 2SO4 for 250 ml , and 0 mM ( NH4 ) 2SO4 for 50 ml ( flow rate: 8 ml/min ) . The Frh activity was eluted in the 0 mM ( NH4 ) 2SO4 fractions . The protein solution was concentrated with ultrafiltration by Amicon filters ( 100-kDa cutoff ) to 4–5 ml , which were then applied to a Sephacryl S-400 HR column ( 2 . 6 × 60 cm ) equilibrated with buffer B ( buffer A + 150 mM NaCl ) . The Frh activity was eluted after washing the column with 180 ml buffer B ( flow rate: 1 ml/min ) . The pooled fractions were concentrated with ultrafiltration by Amicon filters ( 100-kDa cutoff ) to 20 mg/ml . CHAPS was added to the concentrate ( 48 mM final concentration ) , and the solution was incubated for 12 hr at room temperature with slow stirring . The protein solution was washed 5 times on Amicon filters ( 100-kDa cutoff ) with buffer A with 4 mM CHAPS ( buffer A2 ) and applied to a MonoQ column ( 1 × 8 cm ) equilibrated with buffer A2 . Protein was eluted by a NaCl linear gradient in buffer A2: 0–400 mM NaCl in 25 ml and then 400–600 mM NaCl in 20 ml ( flow rate: 0 . 8 ml/min ) . In the linear gradient of 400–600 mM NaCl , the active Frh was eluted from the column at a concentration of 540 mM NaCl . The protein solution was concentrated on Amicon filters ( 100-kDa cutoff ) to 2 ml and further purified on a Sephacryl S-400 HR column in order to remove remaining smaller particles and bigger aggregates . F420 was isolated from M . marburgensis by established methods ( Shima and Thauer , 2001 ) . 3 µl of a 0 . 7 mg/ml Frh sample was applied to freshly glow discharged Quantifoil R1/4 grids ( Quantifoil Micro Tools , Jena , Germany ) that had been pretreated for 15 s in chloroform . The \|grids were blotted in an FEI Vitrobot using a 2 . 5 s blotting time at 70% humidity and 10°C and plunge-frozen in liquid ethane . For collection of data of Frh in the presence of substrate , 0 . 5 , 1 . 0 , or 10 mM F420 was added to the sample under oxygen-free conditions just prior to grid preparation . Images were collected at liquid nitrogen temperature on an FEI Tecnai Polara operated at 200 kV . Before images were recorded , the microscope was carefully aligned in an iterative process to correct for objective astigmatism and beam tilt by coma-free alignment ( Glaeser et al . , 2011 ) . The corrections were carried out at a dose of 15 e−/Å2 and at half the defocus value used for collecting the images , and were repeated for each grid square from which images were collected . The alignments were done with a Gatan 4k × 4k CCD using unbinned images . Images were recorded on Kodak SO-163 film at a magnification of 59 , 000× with a dose of 10–15 e−/Å2 at a defocus of 1 . 5–2 . 8 µm . The film was developed for 12 min in full-strength Kodak D-19 developer and fixed for 8 min in Kodak Rapid Fix . Five hundred and six negatives were collected of substrate-free Frh and 512 of Frh with F420 . Films that showed obvious flaws ( too thin ice with particles just around the edge of the Quantifoil hole , too high particle concentration , broken ice , obvious drift ) were discarded . Four hundred and two negatives of substrate-free Frh were scanned and 277 of F420 with Frh . The latter preparation had a higher protein concentration , resulting in many films with too many overlapping particles . This was more than compensated by the higher number of particles per film . Films were digitized on a Zeiss Photoscan scanner with a pixel size of 7 µm , corresponding to 1 . 14 Å on the specimen as calibrated with fatty acid synthase ( Gipson et al . , 2010 ) . Particle selection was done semiautomatically with the Boxer module from EMAN ( Ludtke et al . , 1999 ) and data processing with EMAN2 ( Tang et al . , 2007 ) . The contrast transfer function ( CTF ) of the selected particles from each film was determined with EMAN2 and images that showed visible Thon rings in the power spectrum to high resolution and no indication of drift or astigmatism were selected for further processing and their CTF was corrected by phase flipping . A tetrahedral starting model was created by the EMAN2 program e2initialmodel from a number of class averages . This model was iteratively refined using the main EMAN2 program e2refine , which determines the 3D orientation of each particle by comparison to a set of projections of the current 3D reference map . Particles in the same orientation are aligned and averaged , and a new 3D map is constructed from the averages that are then reprojected to create references for the next refinement cycle ( Ludtke et al . , 1999; Tang et al . , 2007 ) . Tetrahedral ( T ) symmetry was applied throughout . In the initial refinement steps , the data were binned to a pixel size of 2 . 28 Å . After the resolution reached ∼10 Å , the unbinned data were used . At every refinement step , the 30% worst members of each class were discarded . Because of the high symmetry and globular shape of the particle , all class averages had similar quality ( see Figure 1B ) and were used for each reconstruction . The final data set of substrate-free Frh contained 84 , 000 particles from 101 negatives of the 402 scanned negatives . Between iterations , the projection angle was varied . In the last refinement step an angle of 0 . 9° between consecutive reference reprojections was used , yielding 2134 reference images . The F420 data were refined with a high-resolution map of substrate-free Frh as a starting reference map . Ninety seven thousand particles were selected from 80 negatives . Smaller datasets of Frh with 0 . 5 , 1 . 0 , and 10 mM F420 were used initially , but no significant differences were seen and the datasets were merged . The resolution of the maps was estimated using the command eotest in EMAN2 , which calculates the Fourier shell correlation ( FSC ) between reconstructions made by half-data sets from the odd and even numbered particles . This indicated 3 . 9 and 4 . 0 Å for the maps without and with F420 , respectively , using the 0 . 5 FSC criterion ( Figure 10A ) . The resolution estimate can be inflated by the overfitting of noise at high resolution ( Grigorieff , 2000; Scheres and Chen , 2012 ) . To test for this effect , we used the procedure e2refine-evenodd in EMAN2 , where the phase of the starting model are randomized from a cutoff resolution lower than the expected resolution of the map , and then two half-data sets are refined to convergence against this model . High-resolution noise will be uncorrelated between the two maps . Subsequently , the FSC between the two resulting maps is calculated , and the resolution from these two independent half-data sets determined at 0 . 143 FSC ( Rosenthal and Henderson , 2003 ) . This procedure yielded a resolution of 5 . 5 Å ( Figure 10B ) . B-factors of the reconstructions were estimated with the program embfactor ( Rosenthal and Henderson , 2003; Fernández et al . , 2008 ) in the resolution range 10–4 . 5 Å as 54 Å2 for the substrate-free map and 60 Å2 for the map with F420 ( see Figure 10D ) . Maps were visualized in Chimera ( Pettersen et al . , 2004 ) . Homologous protein structures were fitted manually in the map followed by a rigid body fit ( pdb 2wpn , Marques et al . , 2010 , for FrhA and the C-terminal domain of FrhG; pdb 1dur for the ferredoxin domain of FrhG ) . Models were then mutated to the correct amino acid and manually rebuilt in Coot ( Emsley and Cowtan , 2004 ) . Regions for which no homolog was available , including all of FrhB , was built ab initio in Coot , based on secondary structure predictions done using the PSIPRED server ( Bryson et al . , 2005 ) . The conserved cysteines serving as ligands for the Fe-S clusters and the visibility of the Fe-S clusters in the map gave clear initial hints to the identity of secondary structure elements . Helices were built using the command Place helix . Loops and β-sheets were built by first placing Cα atoms using the Baton build module . Full-atom models were built based on the visibility of side chains , which led in all cases to an unambiguous assignment . The models were refined using the Regularize Zone module . Where possible , side chains were fitted to available density ( overall about 55% of side chains were visible , including most aromatic residues and arginines ) . During refinement , torsion angle , planar peptide , and Ramachandran restraints were used in order not to create a well-fitting but unrealistic model . The final model contains 878 residues out of a possible 903 , a [NiFe] cluster , 4 [4Fe4S] clusters , an FAD and part of an F420 molecule; 88 . 1% of residues lie in the most favored regions of a Ramachandran plot , 8 . 0% in generously allowed regions , and 3 . 8% are outliers . A full dodecamer model was generated in Chimera ( Pettersen et al . , 2004 ) and a map was calculated from this model . An FSC curve between this map and the experimental map indicated an overall agreement of 5 . 8 Å at 0 . 5 FSC ( Figure 10C ) . Figures were made using Chimera ( Pettersen et al . , 2004 ) and the PyMOL Molecular Graphics System ( DeLano , 2002 ) .	Many microbes grow by producing methane gas from carbon dioxide and hydrogen gas , and enzymes known as hydrogenases play important roles in this metabolic process . The production of methane in these microbes depends on a nickel–iron hydrogenase called Frh adding electrons to a coenzyme called F420 . This hydrogenase cleaves a hydrogen molecule into two electrons , which are transferred to the F420 coenzyme , and two protons . The reduced form of F420 is then used for several reactions in the methane production process . This process , which is known as methanogenesis , provides the microbes with energy . Nickel–iron hydrogenases can be divided into five different groups , but researchers have been able to determine the detailed structures of the enzymes in just one of these groups . All nickel–iron hydrogenases contain at least two subunits: a large subunit with a catalytic center composed of both nickel and iron ions and a small subunit that contains three iron–sulfur clusters . Frh—which is short for F420-reducing nickel–iron hydrogenase—is known to have a third subunit comprising an extra iron–sulfur cluster and a coenzyme called FAD that allows it to interact with the F420 coenzyme . However , until now , little was known about the detailed structure of the Frh enzyme . Mills et al . have used electron cryo-microscopy ( cryo-EM ) to determine the structure of Frh when it is on its own , and also when it is bound to F420 . This technique involves freezing a solution of the enzyme in a thin layer of ice and recording an image of this layer in an electron microscope . By combining a large number of images , each of which contains many identical enzymes in different orientations , it is possible to determine the 3-dimensional structure of the enzyme . Mills et al . found that Frh forms a very large tetrahedral complex that contains six Frh dimers . And by comparing the structure with and without F420 , they identify a pocket near the FAD coenzyme that the F420 coenzyme binds to . They also identify a fold in the third subunit that allows proteins to bind both FAD and F420 . The work demonstrates the potential of cryo-EM to elucidate structures that cannot be determined by other approaches .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "Methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2013	De novo modeling of the F420-reducing [NiFe]-hydrogenase from a methanogenic archaeon by cryo-electron microscopy
Among coupled exchangers , CLCs uniquely catalyze the exchange of oppositely charged ions ( Cl– for H+ ) . Transport-cycle models to describe and explain this unusual mechanism have been proposed based on known CLC structures . While the proposed models harmonize with many experimental findings , gaps and inconsistencies in our understanding have remained . One limitation has been that global conformational change – which occurs in all conventional transporter mechanisms – has not been observed in any high-resolution structure . Here , we describe the 2 . 6 Å structure of a CLC mutant designed to mimic the fully H+-loaded transporter . This structure reveals a global conformational change to improve accessibility for the Cl– substrate from the extracellular side and new conformations for two key glutamate residues . Together with DEER measurements , MD simulations , and functional studies , this new structure provides evidence for a unified model of H+/Cl– transport that reconciles existing data on all CLC-type proteins . CLC transporter proteins are present in intracellular compartments throughout our bodies – in our hearts , brains , kidneys , livers , muscles , and guts – where they catalyze coupled exchange of chloride ( Cl– ) for protons ( H+ ) ( Jentsch and Pusch , 2018 ) . Their physiological importance is underscored by phenotypes observed in knockout animals , including severe neurodegeneration and osteopetrosis ( Sobacchi et al . , 2007; Stobrawa et al . , 2001; Hoopes et al . , 2005; Kasper et al . , 2005 ) , and by their links to human disease , including X-linked mental retardation , epileptic seizures , Dent’s disease , and osteopetrosis ( Lloyd et al . , 1996; Hoopes et al . , 2005; Veeramah et al . , 2013; Hu et al . , 2016 ) . CLC-ec1 is a prokaryotic homolog that has served as a paradigm for the family ( Estévez et al . , 2003; Lin and Chen , 2003; Engh and Maduke , 2005; Miller , 2006; Matulef and Maduke , 2007 ) . Its physiological function enables resistance to acidic conditions , such as those found in host stomachs ( Iyer et al . , 2003 ) . Like all CLC proteins , CLC-ec1 is a homodimer in which each subunit contains an independent anion-permeation pathway ( Miller and White , 1984; Ludewig et al . , 1996; Middleton et al . , 1996; Dutzler et al . , 2002 ) . Studies of CLC-ec1 revealed the importance of two key glutamate residues – ‘Gluex’ and ‘Gluin’ ( Figure 1A ) in the transport mechanism . Gluex is positioned at the extracellular entryway to the Cl–-permeation pathway , where it acts both as a ‘gate’ for the transport of Cl– and as a participant in the transport of H+ ( Dutzler et al . , 2003; Accardi and Miller , 2004 ) . Gluin is located towards the intracellular side of the protein and away from the Cl–-permeation pathway , where it appears to act as a H+ transfer site ( Accardi et al . , 2005; Lim and Miller , 2009 ) . In CLC transporter crystal structures , Gluex has been observed in three different positions relative to the Cl–-permeation pathway: ‘middle’ , ‘up’ , and ‘down’ . The ‘middle’ conformation is observed in the WT CLC-ec1 structure , where Gluex occupies the extracellular anion-binding site , ‘Sext’ ( Dutzler et al . , 2002; Figure 1B ) . The ‘up’ conformation is seen when Gluex is mutated to Gln , mimicking protonation of Gluex; here , the side chain moves upward and away from the permeation pathway , allowing a Cl– ion to bind at Sext ( Dutzler et al . , 2003; Figure 1B ) . The ‘down’ conformation is seen in the eukaryotic cmCLC structure , where Gluex plunges downwards into the central anion-binding site , ‘Scen’ ( Feng et al . , 2010; Figure 1B ) . The intracellular anion-binding site , ‘Sint’ , is a low-affinity site ( Picollo et al . , 2009 ) and is not depicted . The rotation of the Gluex side chain is the only conformational change that has been detected crystallographically in the CLC transporters . A central question , therefore , is whether and how other protein conformational changes contribute to the CLC transport mechanism . In previous work , we used a spectroscopic approach to evaluate conformational changes in CLC-ec1 , and we found that raising [H+] ( to protonate Gluex ) caused conformational change in regions of the protein outside of the permeation pathway , up to ~20 Å away from Gluex ( Elvington et al . , 2009; Abraham et al . , 2015 ) . Using a combination of biochemical crosslinking , double electron-electron resonance ( DEER ) spectroscopy , functional assays , and molecular dynamics ( MD ) simulations , we concluded that this H+-induced conformational state represents an ‘outward-facing open’ state , an intermediate in the transport cycle that facilitates anion transport to and from the extracellular side ( Khantwal et al . , 2016 ) . Here , to obtain a high-resolution structure of the H+-bound conformational state , we crystallized a triple mutant , ‘QQQ’ , in which glutamines replace three glutamates: Gluex , Gluin , and E113 . E113 is located within hydrogen bonding distance of Gluin and is computationally predicted to be protonated at neutral pH ( Faraldo-Gómez and Roux , 2004 ) . In contrast to the single-point mutants of Gluin and Gluex , which reveal either no conformational change ( Glnin ) ( Accardi et al . , 2005 ) or only a simple side-chain rotation ( Glnex ) ( Dutzler et al . , 2003 ) , the QQQ mutant structure reveals global conformational change , which generates the expected opening of the extracellular permeation pathway . Unexpectedly , this structure additionally reveals new side-chain conformations for both Glnex and Glnin . Based on this new structure , together with MD simulations , DEER spectroscopy , and functional studies , we propose an updated framework for modeling the CLC transport cycle . The QQQ mutant ( E148Q/E203Q/E113Q ) was crystallized in the lipidic cubic phase , without any antibody Fab fragment . The structure , determined at 2 . 6 Å resolution ( Table 1 ) , reveals an unanticipated change in the conformation of the mutated Gluex residue , Q148 ( Glnex ) . Instead of occupying the ‘up’ position , as seen in the structure of the E148Q protein ( Dutzler et al . , 2003 ) , the sidechain has moved away from the permeation pathway and into the hydrophobic core of the protein , a conformation we designate as ‘out’ ( Figure 1C , D ) . This conformation , which has not previously been observed in the CLC transporters , resembles the conformation of Gluex in the CLC-1 channel structure ( Park and MacKinnon , 2018 ) . Originally , it was suggested that this ‘out’ position may be relevant only to CLC channels , due to the steric clashes with conserved residues that the ‘out’ conformation would generate based on known CLC transporter structures ( Park and MacKinnon , 2018 ) . However , our new structure reveals that small adjustments in residues 186 , 190 , 199 , and 357 suffice to accommodate Glnex occupancy in the hydrophobic core ( Figure 1E ) . Analysis of the QQQ structure using HOLE , a program for analyzing the dimensions of pathways through molecular structures ( Smart et al . , 1996 ) , reveals an opening of the extracellular vestibule , increasing accessibility from the extracellular solution to the anion-permeation pathway , in contrast to previously described structures . In the WT protein , two sub-Angstrom bottlenecks occur between Scen and the extracellular side of the protein ( Figure 2A ) . In the QQQ protein , these bottlenecks are relieved , widening the pathway to roughly the size of a Cl– ion ( Figure 2A , Video 1 ) . In contrast , a single point mutation at the Gluex position ( E148Q ) relieves only one of the two bottlenecks ( Figure 2A ) . This observation is consistent with the QQQ structure representing the CLC-ec1 outward-facing open state . The extracellular bottleneck to anion permeation is formed in part by Helix N , which together with Helix F forms the anion-selectivity filter ( Dutzler et al . , 2002 ) . Previously , we proposed that generation of the outward-facing open state involves movement of Helix N in conjunction with its neighbor Helix P ( at the dimer interface ) to widen this bottleneck ( Khantwal et al . , 2016 ) . In addition , Helix N motions have been inferred from experiments on the mammalian antiporter CLC-4 ( Osteen and Mindell , 2008 ) and from the gating effects of Helix-N disease causing mutations in CLC-1 ( Wollnik et al . , 1997; Zhang et al . , 2000; Pusch , 2002; Tang and Chen , 2011 ) . Disease-causing mutations in Helix N are also found in CLC-Kb , CLC-5 , and CLC-7 ( Konrad et al . , 2000; Leisle et al . , 2011; Lourdel et al . , 2012 ) . Structural alignment of the QQQ mutant with either E148Q ( Figure 2B–D ) or WT ( Figure 2D ) confirms the movement of these helices . These structural changes involve shifts in highly conserved residues near the anion-permeation pathway , including F190 ( Helix G ) , F199 ( Helix H ) , and F357 ( Helix N ) . The side chains of all repositioned residues show good electron density ( Figure 2E , Figure 2—figure supplement 1 ) . Together , these motions widen the extracellular bottleneck ( Figure 2F , G ) . Ion channels have uninterrupted permeation pathways that extend from extracellular to intracellular sides of the membrane , facilitating ion movement down ( and only down ) the ion’s electrochemical gradient . In contrast , to facilitate ion pumping , secondary active transporters must have permeation pathways that are alternately exposed to the extracellular or intracellular sides of the membrane , but never to both simultaneously ( Jardetzky , 1966; Tanford , 1983; Forrest et al . , 2011 ) . Since Gluex in the ‘out’ position has only been observed in a channel structure ( CLC-1 ) before now , it is prudent to question whether this positioning is compatible with an alternating access mechanism . We therefore examined the QQQ structure along the intracellular aspect of the permeation pathway . In WT CLC-ec1 , Cl– permeation to and from the intracellular side is controlled by a constriction formed by conserved residues S107 and Y445 , which is thought to act either as a kinetic barrier ( Feng et al . , 2010; Feng et al . , 2012 ) or as a gate that opens and closes ( Accardi , 2015; Basilio and Accardi , 2015 ) . In QQQ , this intracellular constriction is unchanged compared to WT CLC-ec1 and is narrower than that observed in the CLC-1 channel ( Figure 3 , Figure 3—figure supplement 1 ) . This maintained intracellular constriction supports QQQ as a viable representative for a transporter intermediate . The widening of the extracellular Cl– entryway in QQQ is accompanied by subtle changes in the Sext Cl–-binding site ( Figure 2—figure supplement 2 ) . We therefore hypothesized that Cl– binding to this site might be altered . To test this hypothesis , we used isothermal titration calorimetry ( ITC ) to compare Cl– binding to QQQ and E148Q ( Gluex to Glnex ) , which both show the Sext site occupied by Cl– ( in contrast to WT CLC-ec1 , where the Sext site is occupied by Gluex ) . We found that QQQ and E148Q bind Cl– with similar affinities ( Kd = 138 ± 26 µM and 116 ± 6 µM , respectively ) ( Figure 4A ) . Since these measurements do not distinguish between binding to Scen versus binding to Sext , we also attempted to distinguish binding at the different sites crystallographically , making use of the anomalous signal obtained from Br– binding , as has been done previously ( Lobet and Dutzler , 2006 ) . However , we have not been successful in our attempts to produce well-diffracting QQQ crystals in the presence of Br– . Regardless , while we cannot make specific conclusions about the Sext site , we can conclude that overall binding of Cl– to QQQ and E148Q CLC-ec1 occurs with similar affinity . The widening of the extracellular Cl– entryway in QQQ ( Figure 2 ) predicts that Cl– transport through QQQ will be faster than through E148Q , if extracellular gate-opening is a rate-limiting step . On the other hand , if Cl– transport through E148Q and QQQ have the same rate-limiting step , then Cl– transport rates , like Cl– binding affinities , should be similar . Experimental measurements revealed that QQQ Cl– transport rates are ~2 fold faster than E148Q transport rates at pH 7 . 5 , the pH at which the binding studies were performed ( Figure 4B , C; Figure 4—source data 1 ) , indicating that the two mutants have different rate-limiting steps . If this difference in transport rates is due to the wider extracellular Cl– entryway in QQQ compared to E148Q , then lowering the pH – to allow E148Q to adopt the QQQ-like ( outward-facing open ) conformation – should increase the Cl– transport rate . Consistent with this prediction , transport rates of the E148Q mutant increase by 2-fold at pH 4 . 5 ( Figure 4D; Figure 4—source data 1 ) . These results support the conclusion that E148Q CLC-ec1 ( and by extension WT CLC-ec1 ) undergoes an opening of the extracellular vestibule at low pH . We note that the relatively slow Cl– transport by both E148Q and QQQ compared to WT CLC-ec1 ( ~300 s−1 versus 2200 s−1 ) is not surprising , given that these Gluex mutants lack a negatively charged carboxylate to compete Cl– out of the permeation pathway . To evaluate overall conformational changes in the QQQ structure , we generated difference distance matrices , which provide comparisons that are independent of the structural alignment method ( Nishikawa et al . , 1972 ) . Overall , comparison of QQQ to WT CLC-ec1 using the difference distance matrices confirms a hot spot of conformational change at Helices K-N ( as illustrated in Figure 2 ) and highlights additional changes at G , H , I , and Q ( Figure 5A; Figure 5—figure supplement 1; Figure 5—source data 1 ) . In contrast , comparison of single-Glu mutant structures to WT reveals only minor ( ≤0 . 8 Å ) changes ( Figure 5B ) . CLCs , like many secondary active transporters , are comprised of pairs of inverted structural repeats ( Duran and Meiler , 2013; Forrest , 2015 ) . These repeats are homologous domains that are inserted with inverted orientation into the membrane , related to one another by an axis of pseudosymmetry along the membrane plane ( Figure 5C ) . Conformational exchange of the repeats , with the first repeat adopting the conformation of the second and vice versa , has been shown in other transporters to convert the overall protein structure from outward- to inward-facing , thus facilitating the alternating access mechanism required to achieve secondary active transport ( Forrest et al . , 2008; Crisman et al . , 2009; Forrest and Rudnick , 2009; Palmieri and Pierri , 2010; Radestock and Forrest , 2011 ) . Since all previous CLC transporter structures appear ‘occluded’ ( neither inward- nor outward-facing ) , it is uncertain a priori whether conformational swap will occur upon transition to the outward-facing state in the CLCs . Within each repeat of the QQQ structure , significant changes occur compared to the equivalent repeat in WT CLC-ec1 , particularly in Helices H and Q ( Figure 5D ) . However , the repeats have not interconverted ( Figure 5—source data 2 ) . In addition to the changes within each repeat domain , there are substantial changes between the two repeats ( relative to one another ) , most strikingly between Helices G-I of Repeat 1 and Helices J-N of Repeat 2 ( Figure 5D ) . Within Repeat 2 , the major change occurs at Helices O-Q . This region is of mechanistic interest because a cross-link between residue 399 on Helix O and 432 on Helix Q is known to inhibit transport through coupling to the inner gate ( Basilio et al . , 2014 ) . Interestingly , although we find that Helices O and Q move with respect to the rest of Repeat 2 , residues 399 and 432 do not change position with respect to one another and do not couple to movement at the inner gate ( Figure 5—figure supplement 3A–D ) . Thus , our data suggest that the Helix-O motions involved in transition to the outward-facing state are distinct from those postulated to facilitate inner-gate opening . While the mechanistic details linking Helix O-Q movements between outward- and inward-facing states are currently unknown , it is of interest to note that many disease-causing mutations occur in this segment , in both CLC channels ( Saviane et al . , 1999; Pusch , 2002; Bignon et al . , 2020 ) and transporters ( Lourdel et al . , 2012; Veeramah et al . , 2013 ) . In Helix O , for example , mutation of a highly conserved glycine residue occurring mid-helix can cause Dent’s disease ( CLC-5 , Smith et al . , 2009 ) or Bartter syndrome ( CLC -Kb , Lin et al . , 2009 ) . In WT CLC-ec1 , the helix is kinked at this glycine; in QQQ , the helix is straight ( Figure 5—figure supplement 3E , F ) . Within Repeat 1 , the major change occurs at Helices G-I . This segment also changes position relative to Repeat 2 ( Figure 5D ) . Helix H is of mechanistic interest because it contains the Gluin residue that is thought to transfer H+ from the intracellular solution to Gluex ( Accardi et al . , 2005; Accardi , 2015 ) and because Helix H is one of the most highly conserved regions of the protein ( Dutzler et al . , 2002 ) . The movement of Helices G-I relative to Repeat two is illustrated with structural overlays in Figure 6A , B . This outward movement provides space for Glnex to move to the ‘out’ position ( Figure 6B ) . The movement of Helices G-I relative to other helices in Repeat one is illustrated in Figure 6C , D . This movement releases the interaction between residue 113 ( Helix D ) and Glnin ( Helix H ) ( Figure 6E ) . Strikingly , Glnin moves into the hydrophobic core of the protein , to within 6 Å of Glnex ( Figure 6F ) . This change is accompanied by a rearrangement of water molecules in the internal core ( Figure 6—figure supplement 1 ) . Our working hypothesis is that the QQQ mutant structure mimics the outward-facing open intermediate in the WT CLC transport cycle . When working with a mutant , however , one always wonders whether any conformational change observed is relevant to the WT protein . We therefore used DEER spectroscopy to evaluate conformational change in WT CLC-ec1 . DEER spectroscopy is advantageous because it can evaluate conformational change by site-directed spin labeling , without the constraints of crystallization . Accurate distance distributions can be obtained for spin labels separated by ~20–70 Å ( Jeschke , 2012; Mishra et al . , 2014; Stein et al . , 2015 ) . Since CLC-ec1 is a homodimer ~100 Å in diameter , a simple labeling strategy with one spin label per subunit can provide a sample with optimally spaced probes for distance-change measurements . For example , the extracellular sides of Helices N and O ( Figure 2B , C ) are separated by ~50 and 35 Å respectively from their correlates in the other subunit . To test the hypothesis that these helices move , we generated WT CLC-ec1 ( cysteine-less background ) with spin labels at positions 373 , 374 ( Helix N ) and 385 ( Helix O ) and performed DEER measurements under two conditions , pH 7 . 5 and pH 4 . 5 . The rationale for this experimental strategy is that pH 4 . 5 will promote protonation of Gluex and Gluin , thus favoring a global conformation comparable to that observed in the QQQ structure ( Figure 7A ) . Consistent with our hypothesis , spin labels on Helices N and O exhibited pH-dependent changes in distance distributions in the direction predicted by the QQQ structure ( Figure 7B–D ) . To provide a more direct comparison of WT to QQQ , we additionally made measurements on spin-labeled QQQ samples ( cysteine-less background ) . At all three positions , the DEER distance distributions showed little to no pH dependence , and they resembled the distributions observed with WT at pH 4 . 5 ( Figure 7B–D ) . Similar results were obtained for a spin label on Helix P ( Figures 2C and 7E ) , which had been shown by cross-linking experiments to move during the CLC-ec1 transport cycle ( Khantwal et al . , 2016 ) , thus providing further support for the relevance of the conformational changes observed in QQQ . Finally , to test the predicted conformational change at the intracellular side near the H+ permeation pathway ( Helix G-I movements , Figure 6C ) , we examined a spin label on Helix G . ( Helix G inter-subunit distances are better suited to DEER measurements than Helix H inter-subunit distances . ) Once again , the WT protein showed a pH-dependent shift in the direction predicted by the QQQ structure , and the QQQ protein showed distance distributions resembling those of WT at pH 4 . 5 ( Figure 7F ) . Taken together , the DEER distance distributions provide strong support for the conclusion that the QQQ structure represents a WT CLC-ec1 conformation . Previous computational studies indicated that water wires can transiently bridge the Gluex and Gluin residues separated by 12 . 8 Å in the CLC-ec1 WT structure , which may serve as the pathway for H+ transfer ( Wang and Voth , 2009; Han et al . , 2014; Jiang et al . , 2016 ) . The proximity of these residues in the QQQ structure ( Figure 6F , Figure 6—figure supplement 1 ) motivated us to re-evaluate this phenomenon . In our previous studies on WT CLC-ec1 , extended MD simulations revealed that water spontaneously enters the hydrophobic core of the protein and transiently and repeatedly forms water wires connecting Gluex and Gluin ( Han et al . , 2014; Jiang et al . , 2016 ) . Analogous simulation of the QQQ mutant revealed a dramatic and unanticipated result: water penetration into the hydrophobic core of the protein is greatly increased , and water pathways directly connect bulk water in the intracellular solution to Glnex , without requiring intermediate connection to Glnin ( Figure 8A–C ) . These water pathways were observed frequently during our 600-ns simulation ( Figure 8D , Video 2 , Figure 8—figure supplement 1—source data 1 ) ; in contrast , such water pathways were not observed in our previous 400-ns WT simulations ( Han et al . , 2014; Jiang et al . , 2016 ) . The number of water molecules needed to reach bulk water follows a normal distribution , with chains of 5 or six water molecules predominating ( Figure 8D ) . In contrast , the majority of water wires connecting Gluex to Gluin in the WT simulation involved seven or more water molecules ( Han et al . , 2014; Jiang et al . , 2016 ) . Moreover , the occurrence of water pathways in the QQQ simulation ( 36 . 5% ) is over an order of magnitude greater than the occurrence of water wires between Gluin and Gluex in the WT simulation ( 1 . 3% ) . The absence of water pathways in the WT simulation is likely due to steric hindrance by Gluin , E113 , and bulky side chains in the vicinity , which together block direct access of intracellular bulk water toward the protein interior ( despite the conformational flexibility of Gluin [Wang and Voth , 2009] ) . In the QQQ simulation , Glnin can equilibrate among five side-chain conformations ( Clusters 1–5 ) , all of which can support water pathways ( Figure 8—figure supplement 1A , B ) . Most of the water pathways ( 96% of pathways observed ) occur when Glnin is rotated away from its starting conformation ( Clusters 1–3 ) , allowing water to flow along a pathway near Q113 ( Figure 8—figure supplement 1C , D , Figure 8—figure supplement 1—source data 1; Video 3 ) . In these conformations , the Glnin side chain bends away from Q113 and from the bulky residues F199 and I109 , thus allowing intracellular bulk water to enter the protein interior without encountering steric occlusion ( Figure 8—figure supplement 2A ) . The predominant water pathway observed in our simulations is roughly parallel to the Cl– permeation pathway ( Figure 8A ) . This pathway for water ( and hence H+ ) entry into the protein is different from that previously suggested by us and others . Previously , it was proposed that H+ access to the interior of the protein occurs via an entry portal located near the interfacial side of the homodimer ( Lim et al . , 2012; Han et al . , 2014; Jiang et al . , 2016 ) rather than on the ‘inner’ pathway observed here . While we do see some water pathways occurring along the interfacial route , on a pathway that is lined by Glnin , these occur only rarely ( Figure 8—figure supplement 1D ) . Importantly , the previous mutagenesis studies supporting the interfacial route are also concordant with the inner water pathway observed here . In the previous studies , mutations that add steric bulk at either E202 ( Lim et al . , 2012 ) or the adjacent A404 ( Han et al . , 2014 ) were found to inhibit the H+ branch of the CLC-ec1 transport cycle . The observation that all water pathways involve rotation of E202 away from its starting position ( Figure 8B , Figure 8—figure supplement 2 , Figure 8—figure supplement 3 ) , can explain why bulky mutations at this position would interfere with H+ transport . Gluin has long been modeled as a H+-transfer site in the CLC Cl–/H+ mechanism ( Accardi et al . , 2005; Miller , 2006; Lim and Miller , 2009; Basilio et al . , 2014; Accardi , 2015; Khantwal et al . , 2016 ) . However , this modeling is contradicted by the observation that several CLC transporter homologs can pump H+ with robust stoichiometry in the absence of a titratable residue at the Gluin position ( Feng et al . , 2010; Phillips et al . , 2012; Stockbridge et al . , 2012 ) . In our analysis of water pathways in the MD simulations , we observed that these pathways are not always lined by the Glnin side chain ( Figure 8—figure supplement 1 ) . This finding strongly suggests that while Gluin facilitates water pathways , it is not required as a direct H+-transfer site . To test this hypothesis experimentally , we tested for Cl–-coupled H+ pumping in mutants with non-titratable residues at the Gluin position . We designed the experiment with a large Cl– gradient and a small H+ gradient , both favoring outward movement of the ions; thus , any H+ transport into the vesicles must occur via Cl–-driven H+ pumping , not leak ( Figure 9A ) . With an Ala residue at the Gluin position ( E203A ) , we observed clear H+ pumping above the background signal; by comparison , in line with conventional modeling , the Gluex mutant E148Q exhibited H+ signals similar to control vesicles ( Figure 9B , C; Figure 9—source data 1 ) . While the coupling stoichiometry of E203A is somewhat degraded compared to the WT protein ( Figure 9D ) , the thermodynamic fact arising from this experiment is that H+ pumping occurs with a non-titratable residue at Gluin . In the original pioneering study by Accardi et al . , Gluin was identified as a H+-transfer site following a mutagenesis scan of titratable residues at the intracellular side of CLC-ec1 ( Accardi et al . , 2005 ) . Of the 10 mutants tested , only the Gluin mutant E203Q completely abolished detectable H+ transport . To directly compare to this original study , we evaluated E203Q in our assay . As with E203A , we found that E203Q can pump H+; however , its coupling stoichiometry is substantially more degraded ( Cl–/H+~75 , Figure 9—figure supplement 1; Figure 9—source data 1 ) . Therefore , it is not too surprising that H+ pumping was not detected in the original study . We also examined double mutants , missing titratable residues at both E203 and its hydrogen-bonding partner E113 . When these positions are substituted with two alanines , or with the residues found in the cmCLC homolog ( Thr and Lys ) , H+ pumping is retained ( Figure 9—figure supplement 1 ) . Thus , the conventional thinking of Gluin as a H+-transport site must be re-evaluated . Based on the information gleaned from our study of the QQQ structural intermediate , we propose an updated framework for understanding 2:1 Cl–/H+ exchange by CLC transporters . This updated model is inspired by four key findings . First , the outward-facing state has improved accessibility for Cl– to exchange to the extracellular side ( Figure 2 ) . This state had been previously predicted ( Khantwal et al . , 2016 ) but is now seen in molecular detail . Second , the protonated Gluex can adopt an ‘out’ conformation , within the hydrophobic core of the protein ( Figure 1D , E ) . This novel conformation allows us to eliminate a disconcerting step that was part of all previous models: movement of a protonated ( neutral ) Gluex – in competition with Cl– – into the Scen anion-binding site ( Miller and Nguitragool , 2009; Feng et al . , 2012; Basilio et al . , 2014; Khantwal et al . , 2016 ) . Third , water pathways can connect Glnex ( and presumably Gluex ) directly to the intracellular solution ( Figure 8 ) . Finally , H+ pumping does not require a titratable residue at Gluin ( Figure 9 ) . Together , these findings allow us to propose a revised framework for the Cl–/H+ exchange model , which maintains consistency with previous studies and resolves lingering problems . In our revised model ( Figure 10A ) , the first three states are similar to those proposed previously ( Miller and Nguitragool , 2009; Feng et al . , 2012; Basilio et al . , 2014; Khantwal et al . , 2016 ) . State A reflects the structure seen in WT CLC-ec1 , with Gluex in the ‘middle’ conformation , occupying Sext , and a Cl– occupying Scen . Moving clockwise in the transport cycle , binding of Cl– from the intracellular side displaces Gluex by a ‘knock-on’ mechanism ( Miller and Nguitragool , 2009 ) , pushing it to the ‘up’ position and making it available for protonation from the extracellular side ( State B ) . Protonation generates state C , which reflects the structure seen in E148Q CLC-ec1 where Glnex mimics the protonated Gluex . This sequence of Cl– binding and protonation is consistent with the experimental finding that Cl– and H+ can bind simultaneously to the protein ( Picollo et al . , 2012 ) . Subsequently , a protein conformational change generates an ‘outward-facing open’ state ( D ) . While this state had previously been postulated ( Khantwal et al . , 2016 ) , the QQQ structure presented here provides critical molecular details . State D involves a widening of the extracellular vestibule , which will facilitate Cl– binding from and release to the extracellular side . In the QQQ structure ( our approximation of State D ) , the reorientation of Helix N results in subtle changes in Cl–-coordination at the Sext site ( Figure 2—figure supplement 2 ) , which suggests that binding at this site may be weakened , though we currently lack direct evidence for this conjecture . Regardless of the affinity at Sext , the opening of the extracellular permeation pathway in State D will promote Cl– exchange in both directions , which is essential to achieving reversible transport . In addition to involving a widening of the extracellular vestibule , state D has the protonated Gluex in an ‘out’ conformation and within ~5 Å of Gluin ( Figure 6F ) . At first glance , this positioning suggested to us that Gluin might be participating in an almost direct hand-off of H+ to and from Gluex , through an intervening water molecule . However , MD simulations revealed that Glnin is highly dynamic and most often is rotated away from its starting position , allowing the robust formation of water pathways from the intracellular bulk water directly to Glnex ( Figure 8 ) ( State E ) . Once such transfer occurs , the deprotonated Gluex will be disfavored in the hydrophobic core , and it will compete with Cl– for the Scen anion-binding site , generating State F . Although this conformational state has not been observed crystallographically for CLC-ec1 , computational studies found that Gluex favors the Scen position when there are no Cl– ions bound in the pathway ( as in State F ) , ( Picollo et al . , 2012 ) and that the ‘down’ position is in general the preferred orientation for Gluex ( Mayes et al . , 2018 ) . In addition , a recent structure of an Aspex CLC-ec1 mutant supports that the carboxylate likes to reach down towards Scen , in a ‘midlow’ position , excluding the presence of Cl– at both Scen and Sext ( Park et al . , 2019 ) , as depicted in State F . From this state , binding of Cl– from the intracellular side ( coordinated with inner-gate opening [Basilio et al . , 2014] ) knocks Gluex back up to Sext , generating the original state A . This transport cycle is fully reversible , allowing efficient transport in both directions , as is observed experimentally ( Matulef and Maduke , 2005 ) . Our proposed updated transport model ( Figure 10A ) , in addition to retaining key features based on previous models ( Miller and Nguitragool , 2009; Feng et al . , 2012; Basilio et al . , 2014; Khantwal et al . , 2016 ) , unifies our picture of both CLC transporter and channel mechanisms . First , it is compatible with transporters that have non-titratable residues at Gluin and E113 . Our simulations and experiments ( Figures 8 and 9 ) lead to the conclusion that these residues play a key role in regulating water pathways rather than in direct hand-off of H+ . From this perspective , the evolution of non-titratable residues in either ( Feng et al . , 2010 ) or both ( Phillips et al . , 2012; Stockbridge et al . , 2012 ) of these positions is perfectly sensible . In addition , previous mutagenesis experiments on CLC-ec1 and on mammalian transporters , which demonstrate a surprising tolerance for mutations at Gluin ( Zdebik et al . , 2008; Lim and Miller , 2009 ) now make more sense . Strikingly , the structural positioning of T269 in cmCLC , located at the Gluin sequence position , matches the structural positioning of Glnin in the QQQ mutant , such that side-chain dynamics could facilitate comparable water pathways ( Figure 10B ) . The second unifying feature of our model is that it attests to Gluex movements being conserved amongst every known type of CLC: 2:1 Cl–/H+ exchangers , 1:1 F–/H+ exchangers , and uncoupled Cl– channels . Previously , an ‘out’ position for Gluex had been proposed to be essential to the mechanism of F–/H+ exchangers ( Last et al . , 2018 ) , which allow bacteria to resist fluoride toxicity ( Stockbridge et al . , 2012 ) . However , such a conformation had not been directly observed , and it was postulated that it may be only relevant to the F–/H+ branch of the CLC family . Structurally , Gluex in the ‘out’ position has previously only been observed in a CLC channel . Thus , this conformation is a unifying feature of CLC channels and transporters . Moreover , this conclusion connotes that all CLC proteins act via a ‘windmill’ mechanism ( Last et al . , 2018 ) , in which the protonated Gluex favors the core of the protein while the deprotonated Gluex favors the anion-permeation pathway . Such a mechanism is preferable to previous ‘piston’-type mechanisms , with Gluex moving up and down within the anion-permeation pathway , which required a protonated ( neutral ) Gluex to compete with negatively charged Cl– ions . Elements of the transport cycle require future experiments to elucidate details . Prominently , the nature of the inward-facing conformational state remains uncertain . In our model , we indicated inward-opening with dotted lines ( Figure 10 , States F , A , B ) to reflect this uncertainty . One proposal is that the inner-gate area remains static and transport works via a kinetic barrier to Cl– movement to and from the intracellular side ( Feng et al . , 2010 ) . Consistent with this proposal , multiscale kinetic modeling revealed that 2:1 Cl–/H+ exchange can arise from kinetic coupling alone , without the need for large protein conformational change ( Mayes et al . , 2018 ) . An alternative proposal is that CLCs visit a conformationally distinct inward-open state , based on the finding that transport activity is inhibited by cross-links that restrict motion of Helix O , located adjacent to the inner gate ( Basilio et al . , 2014; Accardi , 2015 ) . This putative inward-open state appears distinct from the conformational change observed in the QQQ mutant , as the inter-residue distances for the cross-link pairs ( 399/432 and 399/259 ) are unchanged in QQQ relative to WT ( Figure 5—figure supplement 3D ) . The details of the kinetic-barrier and conformational-change models , and the need for additional experiments on this aspect of transport , have been clearly and comprehensively discussed ( Accardi , 2015; Jentsch and Pusch , 2018 ) . Simulations of the QQQ conformational state with the glutamine residues reverted to the native , protonatable glutamate side chains will be needed for full understanding of how protonation and deprotonation of these residues affect the conformational dynamics of side chains and water pathways . In our current model , we propose that Gluin needs to be in the protonated ( neutral ) state to adopt the position that allows water pathways . This proposal appears harmonious with the hydrophobic nature of the protein core explored by the Glnin side chain and the fact that other CLC homologs use neutral residues at this position ( Feng et al . , 2010; Phillips et al . , 2012; Stockbridge et al . , 2012 ) . In addition , the proposal is consistent with MD simulations that show the Gluex/Gluin doubly protonated state is highly populated ( Mayes et al . , 2018 ) and can favor formation of water pathways under certain conditions ( Ko and Jo , 2010 ) . Nevertheless , simulations with glutamate side chains in the QQQ conformational state , together with explicit evaluation of H+ transport ( Wang et al . , 2018; Duster et al . , 2019 ) , are needed to elaborate details of the H+-transfer steps . In addition , multiscale modeling can expand the picture to include multiple pathways that are likely to occur ( Mayes et al . , 2018 ) . Recognizing the importance of elaborating these details , the results reported here represent an essential and pivotal step toward a complete , molecularly detailed description of mechanism in the sui generis CLC transporters and channels . Mutations were inserted in the WT CLC-ec1 protein using Agilent QuikChange Lightning kit and were confirmed by sequencing . Protein purification was carried out as described ( Walden et al . , 2007 ) , with a few changes depending on the type of experiment . For ITC experiments , QQQ or E148Q were purified in buffer A ( 150 mM Na-isethionate , 10 mM HEPES , 5 mM anagrade decyl maltoside ( DM ) at pH 7 . 5 ) . For crystallization experiments , QQQ was extracted with DM . The detergent was gradually exchanged for lauryl maltose neopentyl glycol ( LMNG ) during the cobalt-affinity chromatography step . The final size-exclusion chromatography step was performed in a buffer containing LMNG . All detergents were purchased from Anatrace ( Maumee , OH ) . For DEER spectroscopy experiments , cysteine mutations were made on a WT or QQQ cysteine-less background ( C85A/C302A/C347S ) ( Nguitragool and Miller , 2007 ) . Proteins were purified under reducing conditions and then labeled with the spin label MTSSL ( 1-Oxyl-2 , 2 , 5 , 5 , -tetramethylpyrroline-3-methyl methanethio-sulfonate ) as described ( Khantwal et al . , 2016 ) . Purified QQQ protein was concentrated to at least 30 mg/mL . Concentrated protein was mixed with 1 . 5 parts ( w/w ) of monolein containing 10% ( w/w ) cholesterol using the syringe reconstitution method ( Caffrey and Cherezov , 2009 ) , to generate a lipidic cubic phase mixture . 25 nL droplets of the mixture were dispensed on glass plates and overlaid with 600 nL of precipitant using a Gryphon crystallization robot ( Art Robbins Instruments , Sunnyvale , CA ) . Crystallization trials were performed in 96-well glass sandwich plates incubated at 16°C . The best crystals were obtained using a precipitant solution consisting of 100 mM Tris ( pH 8 . 5 ) , 100 mM sodium malonate , 30% PEG 400% and 2 . 5% MPD Crystals were harvested after 3–4 weeks of incubation and flash-frozen in liquid nitrogen without further additives . Figures were prepared using PyMOL and Adobe Illustrator . X-ray diffraction data were collected at APS at GM/CA beamline 23ID-D and were processed using XDS ( Kabsch , 2010 ) and AIMLESS ( Evans , 2006 ) from the CCP4 suite ( Winn et al . , 2011 ) . Owing to radiation damage , a complete dataset was collected by merging data from three different crystals . Phases were obtained using PHASER ( McCoy et al . , 2007 ) with PDB ID 1ots as a search model . Iterative refinement was performed manually in Coot ( Emsley and Cowtan , 2004 ) and REFMAC ( Murshudov et al . , 1997 ) . The final model contained all residues except those of Helix A due to lack of density for this region of the protein . Helix A is observed in different conformations in the monomeric versus dimeric CLC-ec1 structures , and has no impact on function ( Robertson et al . , 2010 ) . Flux assay results presented in this paper required a variety of experimental conditions for reconstitutions and flux assays , summarized in Table 2 . For flux assays comparing activity at pH 7 . 5 and 4 . 5 , purified CLC-ec1 were first reconstituted at pH 6 . The samples were then aliquoted and pH-adjusted using a 9:1 ratio of sample and the adjustment buffer . This step was taken to eliminate variability from separate reconstitutions . For experiments testing H+ pumping in mutants , a pH gradient was used to ensure any measured H+ transport was from H+ pumping and not H+ leak . To measure the rate of H+ and Cl– transport in flux assays , purified CLC-ec1 proteins were reconstituted into phospholipid vesicles ( Walden et al . , 2007 ) . E coli polar lipids ( Avanti Polar Lipids , Alabaster , AL ) in chloroform were dried under argon in a round-bottomed flask . To ensure complete removal of chloroform , the lipids were subsequently dissolved in pentane and dried under vacuum on a rotator , followed by further drying ( 5 min ) under argon . The lipids were then solubilized at 20 mg/mL in buffer R ( Table 2 ) with 35 mM CHAPS on the rotator for 1 . 5–2 hr . Purified protein ( 0 . 4–5 µg per mg lipids ) samples or control buffer solution were then added to the prepared lipid-detergent mix and incubated for 10–20 min . Each protein or control reconstitution was divided into 2–4 samples for dialysis to remove detergent , with three buffer changes over 36–60 hr . Following dialysis , each sample was divided into 2–4 for replicate measurements . For experiments shown in Figures 4 and 9 , the replicate measurements were averaged to obtain a turnover rate value , and each of these averages was counted as one ‘n’ . For experiments shown in Figure 7—figure supplement 1 ( DEER samples ) , the replicate measurements from each sample are shown separately . Reconstituted vesicles were subjected to four freeze-thaw cycles and were then extruded with an Avanti Mini Extruder using a 0 . 4 µm-filter ( GE Healthcare , Chicago , IL ) 15 times . For each assay , 60–120 µL of extruded sample were buffer-exchanged through 1 . 5- to 3 . 0 mL Sephadex G-50 Fine resin ( GE Healthcare , Chicago , IL ) columns equilibrated with buffer F ( Table 2 ) . Exchange was accomplished by spinning the columns at ~1100 g for 90 s using a clinical centrifuge . The collected sample ( 80–200 µL ) was then added to buffer F ( 500–600 µL ) for flux-assay measurement . Extravesicular [Cl–] and [H+] were monitored using a Ag·AgCl electrode and a pH electrode , respectively . The electrodes were calibrated by known additions of KCl ( in 20–136 nmol steps ) and NaOH ( in 10–50 nmol steps ) . Sustained ion transport by CLC-ec1 was initiated by addition of 1 . 7–3 . 4 µg/mL of valinomycin ( from 0 . 5 mg/mL stock solution in ethanol ) . At the end of each transport experiment , detergent was added to release all Cl– from the vesicles . This step served as a quality check to confirm that a reasonable yield of vesicles was obtained following the spin-column step . Samples that exhibited a total Cl– release ( sum of Cl– released by transport and detergent release ) >30% than the average were excluded . Using this criterion , 3 out of the 210 assays performed as part of this study were excluded . Titration isotherms were obtained using a VP-ITC microcalorimeter ( MicroCal LLC , Northampton , MA ) at 25°C . For the experiment , QQQ or E148Q protein samples were purified in buffer A . Titrant used in the experiment was 30 mM KCl in buffer A . The starting concentration of protein was 15–20 µM , in a volume of 1 . 5 mL . KCl ( 30 mM ) was syringe-titrated into the sample cell in thirty 10 µL injections . The reference data were obtained by titrating buffer A into the protein-containing solution . Data were analyzed using Origin 7 . 0 software , with fitting using the ‘one set of sites’ model ( keeping n = 1 ) . The other thermodynamic parameters were obtained accordingly . Isethionate was chosen as the anion of choice for purification of proteins for the ITC experiments since the QQQ mutant shows aggregation upon purification in tartrate-containing solutions , which were previously used for ITC experiments with WT and mutant variants of CLC-ec1 ( Picollo et al . , 2009; Khantwal et al . , 2016 ) . The mutant is comparatively stable in isethionate and continues to remain stable throughout the ITC experiment . Figure 4A shows the gel filtration chromatograms of the mutants before and after the ITC experiments . CW-spectra were collected on a Bruker EMX at 10 mW power with a modulation amplitude of 1 . 6G . Determination of the spin concentration of the samples were obtained using Bruker’s built-in Spin Quantitation method . The spin concentration is divided by the protein concentration to obtain the labeling efficiency . DEER experiments were performed at 83 K on a Bruker 580 pulsed EPR spectrometer at Q-band frequency ( 33 . 5 GHz ) using either a standard four-pulse protocol ( Jeschke and Polyhach , 2007 ) or a five-pulse protocol ( Borbat et al . , 2013 ) . Analysis of the DEER data to determine P ( r ) distance distributions was carried out in homemade software running in MATLAB ( Brandon et al . , 2012; Stein et al . , 2015 ) . In the original five-pulse protocol paper the pure five-pulse signal was obtained by subtracting the artefact four-pulse data ( Borbat et al . , 2013 ) . This method requires the ability to discern clearly the extent of the artefact . For the data in this study , we chose to simultaneously fit the four- and five-pulse data with a single Gaussian component in order to improve accuracy of subtracting the four-pulse artefact ( Figure 7—figure supplement 1 ) . Confidence bands for the distance distributions were determined using the delta method ( Hustedt et al . , 2018 ) . The confidence bands define the 95% confidence interval that the best fit distance distribution will have . In the case of a Gaussian distribution , the shape of the confidence bands can be non-Gaussian . The structure of the CLC-ec1 QQQ mutant crystallized in this work at 2 . 6 Å resolution was used as the starting structure for the MD simulation . The 2 Cl– ions bound at Scen and Sint sites in each of the two subunits were preserved for the simulation . In our initial refinement of the QQQ structure , we had modeled water rather than Cl– at the Sext site , and therefore the simulation was performed without Cl– at this site . The pKa of each ionizable residue was estimated using PROPKA ( Olsson et al . , 2011; Rostkowski et al . , 2011 ) , and the protonation states were assigned based on the pKa analysis at pH 4 . 5 . Specifically , E111 , E202 , E235 , and E414 were protonated in the simulation . All other side chains are in their default protonation state . Missing hydrogen atoms were added using PSFGEN in VMD ( Humphrey et al . , 1996 ) . In addition to the crystallographically resolved water molecules , internal water molecules were placed in energetically favorable positions within the protein using DOWSER ( Zhang and Hermans , 1996; Morozenko et al . , 2014 ) , including in a bridging position between Glnex and Glnin . This water was not present in our initial structural model but was subsequently added based on experimental density ( Figure 6E , F ) . The QQQ protein was embedded in a POPE lipid bilayer using the CHARMM-GUI Membrane Builder ( Wu et al . , 2014 ) . The membrane/protein system was fully solvated with TIP3P water ( Jorgensen et al . , 1983 ) and buffered in 150 mM NaCl to keep the system neutral . The resulting systems consisting of ~155 , 000 atoms were contained in a 164 × 127×98 Å3 simulation box . MD simulation was carried out with NAMD2 . 12 ( Phillips et al . , 2005 ) using CHARMM36 force field ( Klauda et al . , 2010; Huang and MacKerell , 2013 ) and a time step of 2 fs . Periodic boundary conditions were used throughout the simulations . To evaluate long-range electrostatic interactions without truncation , the particle mesh Ewald method ( Darden et al . , 1993 ) was used . A smoothing function was employed for short-range nonbonded van der Waals forces starting at a distance of 10 Å with a cutoff of 12 Å . Bonded interactions and short-range nonbonded interactions were calculated every two fs . Pairs of atoms whose interactions were evaluated were updated every 20 fs . A cutoff ( 13 . 5 Å ) slightly longer than the nonbonded cutoff was applied to search for interacting atom pairs . Simulation systems were subjected to Langevin dynamics and the Nosé–Hoover Langevin piston method ( Nosé , 1984; Hoover , 1985 ) to maintain constant pressure ( p=1 atm ) and temperature ( T = 310 K ) ( NPT ensemble ) . The simulation system was energy-minimized for 10 , 000 steps , followed by two stages of 1-ns relaxation . Both the protein and the Cl– ions in the binding sites were positionally restrained ( k = 1 kcal⋅mol−1⋅Å−2 ) in the first 1-ns simulation to allow the membrane to relax . In the second 1-ns simulation , only the protein backbone and the bound Cl– ions were positionally restrained ( k = 1 kcal⋅mol−1⋅Å−2 ) to allow the protein side chains to relax . Then a 600-ns equilibrium simulation was performed for the system without any restraint applied . The water pathways between Q148 ( Glnex ) and the intracellular bulk water was searched using a breadth-first algorithm . In Subunit 2 of the homodimer , Glnex drifted up and away from the ‘out’ position at the beginning of the simulation ( within five ns ) , and it did not return to the ‘out’ position during the simulation . In Subunit 1 , Glnex remained near the ‘out’ position for the first 400 ns; we focused our analysis of water pathways on this subunit . A distance-based criterion of 2 . 5 Å for the hydrogen bonds , which was found to be useful and inexpensive in computational terms in a previous study ( Matsumoto , 2007 ) was used to determine whether water molecules are connected through continuous hydrogen-bonded network . The water pathway with the smallest number of O-H bonds in each frame was considered as the shortest hydrogen-bonded path . The first water molecule in each water pathway is searched using a distance cutoff of 3 . 5 Å for any water oxygen atoms near the OE1/NE2 atoms of Q148 . The water pathway is considered to reach the intracellular bulk once the oxygen atom of the newly found water molecules is at z < −15 Å ( membrane center is at z = 0 ) .	Cells are shielded from harmful molecules and other threats by a thin , flexible layer called the membrane . However , this barrier also prevents chloride , sodium , protons and other ions from moving in or out of the cell . Channels and transporters are two types of membrane proteins that form passageways for these charged particles . Channels let ions flow freely from one side of the membrane to the other . To do so , these proteins change their three-dimensional shape to open or close as needed . On the other hand , transporters actively pump ions across the membrane to allow the charged particles to accumulate on one side . The shape changes needed for that type of movement are different: the transporters have to open a passageway on one side of the membrane while closing it on the other side , alternating openings to one side or the other . In general , channels and transporters are not related to each other , but one exception is a group called CLCs proteins . Present in many organisms , this family contains a mixture of channels and transporters . For example , humans have nine CLC proteins: four are channels that allow chloride ions in and out , and five are ‘exchange transporters’ that make protons and chloride ions cross the membrane in opposite directions . These proteins let one type of charged particle move freely across the membrane , which generates energy that the transporter then uses to actively pump the other ion in the direction needed by the cell . Yet , the exact three-dimensional changes required for CLC transporters and channels to perform their roles are still unknown . To investigate this question , Chavan , Cheng et al . harnessed a technique called X-ray crystallography , which allows scientists to look at biological molecules at the level of the atom . This was paired with other methods to examine a CLC mutant that adopts the shape of a normal CLC transporter when it is loaded with a proton . The experiments revealed how various elements in the transporter move relative to each other to adopt a structure that allows protons and chloride ions to enter the protein from opposite sides of the membrane , using separate pathways . While obtained on a bacterial CLC , these results can be applied to other CLC channels and transporters ( including those in humans ) , shedding light on how this family transports charged particles across membranes . From bone diseases to certain types of seizures , many human conditions are associated with poorly functioning CLCs . Understanding the way these structures change their shapes to perform their roles could help to design new therapies for these health problems .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2020	A CLC-ec1 mutant reveals global conformational change and suggests a unifying mechanism for the CLC Cl–/H+ transport cycle
Stem cell properties change over time to match the changing growth and regeneration demands of tissues . We showed previously that adult forebrain stem cell function declines during aging because of increased expression of let-7 microRNAs , evolutionarily conserved heterochronic genes that reduce HMGA2 expression . Here we asked whether let-7 targets also regulate changes between fetal and adult stem cells . We found a second let-7 target , the RNA binding protein IMP1 , that is expressed by fetal , but not adult , neural stem cells . IMP1 expression was promoted by Wnt signaling and Lin28a expression and opposed by let-7 microRNAs . Imp1-deficient neural stem cells were prematurely depleted in the dorsal telencephalon due to accelerated differentiation , impairing pallial expansion . IMP1 post-transcriptionally inhibited the expression of differentiation-associated genes while promoting the expression of self-renewal genes , including Hmga2 . A network of heterochronic gene products including Lin28a , let-7 , IMP1 , and HMGA2 thus regulates temporal changes in stem cell properties . Stem cell properties change throughout life in many tissues in response to changing growth and regeneration demands ( He et al . , 2009 ) . These changes are particularly evident in the central nervous system ( CNS ) forebrain , where neural stem cells persist throughout life . During fetal development rapidly dividing neural stem cells expand in number before differentiating in precisely defined temporal windows , first to form neurons and then to form glia ( Salomoni and Calegari , 2010 ) . Largely quiescent neural stem cells persist into adulthood in the lateral wall of the lateral ventricle subventricular zone ( SVZ ) as well as in the dentate gyrus , where they give rise to new interneurons throughout adult life ( Alvarez-Buylla and Lim , 2004; Zhao et al . , 2008 ) . However , the rate of neurogenesis , the frequency of stem cells , and their rate of proliferation all decline with age ( Kuhn et al . , 1996; Enwere et al . , 2004; Maslov et al . , 2004; Molofsky et al . , 2006; Bonaguidi et al . , 2011; Encinas et al . , 2011 ) . A fundamental question concerns the mechanisms that control these temporal changes in stem cell properties . The declines in SVZ proliferation , stem cell self-renewal potential , and neurogenesis during aging are regulated by a pathway that includes let-7 microRNAs , the chromatin-associated HMGA2 high mobility group protein , and the p16Ink4a cyclin-dependent kinase inhibitor: let-7b expression increases with age , reducing Hmga2 expression and increasing p16Ink4a expression ( Nishino et al . , 2008 ) . p16Ink4a deficiency or overexpression of a let-7 insensitive form of Hmga2 partially rescues the declines in neural stem cell function and neurogenesis in aging mice ( Molofsky et al . , 2006; Nishino et al . , 2008 ) . This pathway appears to be conserved among multiple mammalian tissues as p16Ink4a deficiency also increases the function of hematopoietic stem cells and pancreatic beta cells during aging ( Janzen et al . , 2006; Krishnamurthy et al . , 2006 ) . HMGA2 also promotes hematopoietic stem cell self-renewal ( Cavazzana-Calvo et al . , 2010; Ikeda et al . , 2011 ) and myoblast proliferation ( Li et al . , 2012 ) . let-7 microRNAs are evolutionarily conserved heterochronic genes that regulate developmental timing ( Pasquinelli et al . , 2000 ) and aging ( Shen et al . , 2012 ) in Caenorhabditis elegans . In mammals , let-7 microRNAs are known to regulate embryonic stem cells ( Melton et al . , 2010 ) , primordial germ cells ( West et al . , 2009 ) , and adult neural stem cells ( Zhao et al . , 2010 ) but it is unclear to what extent let-7 targets regulate developmental changes in mammalian stem cell function over time . For example , it is unclear whether the let-7-regulated pathway we identified in aging stem cells only regulates stem cell aging or whether it is one branch of a larger network of heterochronic genes that regulates temporal changes in stem cell function throughout life . let-7 microRNAs negatively regulate the expression of a number of gene products , including Insulin-like growth factor two mRNA binding protein 1 ( IMP1; also known as CRD-BP and VICKZ1 ) ( Boyerinas et al . , 2008 ) . IMP1 binds to target RNAs , post-transcriptionally regulating their localization , turnover , and translation ( Doyle et al . , 1998; Nielsen et al . , 1999; Farina et al . , 2003; Atlas et al . , 2004 ) . Imp1 expression is widespread in fetal tissues but declines perinatally and is not detected in most adult tissues ( Hansen et al . , 2004; Hammer et al . , 2005 ) . Imp1 expression is elevated in several cancers ( Ioannidis et al . , 2004; Yisraeli , 2005 ) , partly as a consequence of Wnt signaling , which promotes Imp1 transcription ( Noubissi et al . , 2006; Gu et al . , 2008 ) . Over-expression of IMP1 can promote tumorigenesis ( Tessier et al . , 2004 ) . Imp1 deficient mice have a dwarf phenotype with some neonatal mortality ( Hansen et al . , 2004 ) . However , it is unknown if IMP1 regulates stem cells . Canonical Wnt signaling promotes a rapid expansion in the number of undifferentiated stem cells during forebrain development ( McMahon et al . , 1992; McMahon and Bradley , 1990; Ikeya et al . , 1997; Dickinson et al . , 1994; Wrobel et al . , 2007 ) . Wnt signaling prevents cell cycle exit and delays differentiation in these cells ( Megason and McMahon , 2002; Chenn and Walsh , 2002; Machon et al . , 2003; Zechner et al . , 2003; Zhou et al . , 2006; Woodhead et al . , 2006; Gulacsi and Anderson , 2008; Wrobel et al . , 2007 ) . Wnt signaling also promotes the maintenance of stem cells in the adult forebrain ( Kuwabara et al . , 2009; Qu et al . , 2010 ) . However , it is unclear why Wnt signaling expands the number of neural stem cells during development but only maintains declining numbers of stem cells during adulthood . Here we report that Imp1 is expressed in fetal neural stem/progenitor cells as a consequence of Wnt signaling but that its expression declines in late fetal development , partly as a consequence of increasing let-7 microRNA expression . Imp1 promoted the expansion of fetal neural stem cells and Imp1 deficiency reduced brain mass . IMP1 bound to a number of mRNAs , post-transcriptionally promoting the expression of gene products that promote self-renewal , including Hmga2 , and inhibiting the expression of gene products involved in differentiation . Our findings demonstrate a novel role for IMP1 in the expansion of fetal neural stem cells and suggest that the perinatal loss of IMP1 expression is part of the mechanism that allows Wnt signaling to promote the expansion of fetal stem cells . More broadly , our results demonstrate that a network of heterochronic genes regulates temporal changes in stem cell function throughout life . We examined Imp1 expression by quantitative RT-PCR ( qPCR ) in CNS stem/progenitor cells from the embryonic day ( E ) 12 . 5 dorsal telencephalon , E14 . 5 dorsal telencephalon , postnatal day ( P ) 0 lateral ventricle ventricular zone ( VZ ) , and P30 lateral ventricle subventricular zone ( SVZ ) . Imp1 expression was high at E12 . 5 but declined over 100-fold by P0 and was no longer detected in the P30 SVZ ( Figure 1A ) . The family members Imp2 and Imp3 were expressed in patterns very similar to Imp1: high in the telencephalon VZ at E12 . 5 but declining sharply throughout fetal development ( Figure 1—figure supplement 1H , I ) . This raises the possibility of redundancy among IMP family members during CNS development . In contrast to the Imp1 expression pattern , let-7b expression was very low at E12 . 5 but increased approximatly 40-fold by P0 and continued to increase into adulthood ( Figure 1B ) . To confirm that let-7b can regulate Imp1 , we overexpressed let-7b in neural stem/progenitor cells cultured from E14 . 5 dorsal telencephalon . IMP1 protein levels were reduced in neurospheres that overexpressed let-7b ( Figure 1C ) . This suggests that Imp1 expression declines as let-7b expression increases during fetal development and that let-7b can inhibit Imp1 expression . 10 . 7554/eLife . 00924 . 003Figure 1 . Imp1 expression declines over time in neural stem/progenitor cells in the dorsal telencephalon and is extinguished postnatally . ( A and B ) qPCR for Imp1 ( A ) and let-7b ( B ) in E12 . 5 dorsal telencephalon , E14 . 5 dorsal telencephalon , P0 lateral ventricle VZ/SVZ , and P30 lateral ventricle VZ/SVZ ( fold change mean±SD for 3–4 mice/stage; U , not detectable above background; p<0 . 01 ) . ( C ) Western blot of E14 . 5 wild-type neurospheres infected with either GFP-only control lentivirus ( − ) or with let-7b+GFP lentivirus ( + ) . Let-7b overexpression reduced IMP1 expression . ( D ) X-gal staining of sections from E12 . 5 and P0 Imp1β-geo/+ forebrain . Imp1 was expressed in a medial-high/lateral-low gradient in the E12 . 5 dorsal telencephalon and confined to undifferentiated cells in the VZ/SVZ ( solid line ) . At P0 , no X-gal staining was detectable . A high magnification image is shown for the boxed area on the low magnification image to the left . See also Figure 1—figure supplement 1A , B for X-gal staining in E10 . 5 , E14 . 5 , and E16 . 5 brains . ( E ) Immunostaining for LacZ in the dorsal telencephalon from E13 . 5 Imp1β-geo/+ and Imp1+/+ mice . Imp1 was expressed by the Pax6+ neural progenitors in the VZ but not by the TuJ1+ neurons at the cortical plate ( white bars ) . Nuclei were visualized using 4′6-diamino-2-phenylindole dihydrocloride ( DAPI ) staining . ( F ) Virtually all neurospheres cultured from E12 . 5 Imp1β-geo/+ dorsal telencephalon , but not from P0 Imp1β-geo/+ neocortical VZ , stained with X-gal ( mean ± SD % X-gal+ , three experiments ) . ( G ) LacZ and Nestin immunostaining overlapped in sections through neurospheres cultured from Imp1β-geo/+ E12 . 5 dorsal telencephalon . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 00310 . 7554/eLife . 00924 . 004Figure 1—figure supplement 1 . Imp1 expression in the fetal brain is spatially restricted over time and extinguished in adult brain . ( A–B ) X-gal staining of sections from the E10 . 5 Imp1β-geo/+ or wild-type telencephalon ( A ) and E14 . 5 or E16 . 5 Imp1β-geo/+ forebrains ( B ) . Imp1 was expressed throughout the dorsal and ventral telencephalon at E10 . 5 except for the floor plate and roof plate ( arrowheads ) . At E14 . 5 and E16 . 5 , Imp1 was expressed in a medial-high/lateral-low gradient in the dorsal telencephalon and confined to the undifferentiated cells in the VZ/SVZ ( solid lines ) . Higher magnification images of the boxed areas are shown to the right of each low magnification image . ( C and D ) X-gal staining was maintained in most secondary neurospheres upon passaging of E12 . 5 Imp1β-geo/+ dorsal telencephalon-derived primary neurospheres ( C ) . No X-gal staining was detected in primary neurospheres cultured from P60 Imp1β-geo/+ SVZ ( D ) . ( mean ± SD % X-gal+ , three experiments ) . ( E–G ) In situ hybridization for Imp1 on coronal sections of E14 . 5 telencephalon ( E ) , P60 hippocampus ( F ) , or sagittal sections of P60 olfactory bulb ( G ) . Imp1 transcripts were detected in the dorsal region of the telencephalon but not in neurogenic regions of the adult brain . Higher magnification images reflect the boxed regions in each low magnification image . ( H ) qPCR for Imp2 and Imp3 in E12 . 5 dorsal telencephalon , E14 . 5 dorsal telencephalon , P0 lateral ventricle VZ/SVZ , and P30 lateral ventricle VZ/SVZ ( fold-change mean ± SD for 3–4 mice/stage; U , not detectable above background; p<0 . 01 ) . ( I ) In situ hybridization for Imp2 and Imp3 in E13 . 5 dorsal telencephalon sections . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 00410 . 7554/eLife . 00924 . 005Figure 1—figure supplement 2 . Schematic showing fetal telencephalon development and Imp1 expression . ( A ) Anatomy during fetal telencephalon development ( D: dorsal , V:ventral , L:left , and R:right ) . At E10 . 5 , two signalling centers , floor plate ( FP ) and roof plate ( RP ) exist in the ventral and dorsal telencephalon . At E12 . 5 and later , dorsal telencephalon can be divided into dorsomedial telencephalon ( DMT , light blue ) and dorsolateral telencephalon ( DLT , pink ) . Choloid plexus ( CP ) and cortical hem ( CH ) constitute the dorsal end of telencephalon . ( B ) Regions where Imp1 was expressed are indicated in green . At E10 . 5 , Imp1 was expressed throughout the forebrain , except in the floor plate and roof plate . At E12 . 5 , Imp1 was expressed typically in undifferentiated cells in the dorsal telencephalon but not in ventral telencephalon or in basal neurons . At E14 . 5 and later , Imp1 expression was gradually confined to the apical region of the dorsomedial telencephalon . ( C and D ) Schematics show a close up of the boxed regions in panel B to illustrate the cellular layers during cortical development and the cells that expressed Imp1 in those layers . At E10 . 5 , undifferentiated Pax6+ neural stem cells ( blue ovals in D ) dominate all layers of the developing telencephalon and all cells express Imp1 ( green in C ) . At E12 . 5 , neural stem cells start to differentiate into Tbr2+ intermediate neural progenitors ( yellow ovals in D ) and Tuj1+ neurons ( red ovals in D ) that form the preplate ( PP in panel C ) . At E14 . 5 and E18 . 5 , newly formed neurons constitute the marginal zone ( MZ ) , cortical plate ( CP ) and subplate ( SP ) on the basal side of the developing cortex . The intermediate zone ( IZ ) is a cell sparse region that lies between the SP and VZ . Imp1 is expressed in undifferentiated cells at these stages ( green ovals in C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 005 To systematically examine Imp1 expression we analysed a gene-trap mouse ( Imp1β-geo/+ ) in which β-galactosidase-neomycin ( β-geo ) was inserted into the second intron of Imp1 ( Hansen et al . , 2004 ) . This led to the expression of an IMP1-β-geo fusion protein that contained the IMP1 RNA recognition motif encoded by exons 1 and 2 but lacked the second RNA recognition motif and the four hnRNP K homology domains , which are essential for biological activity ( Nielsen et al . , 2002 ) . This mouse therefore provided a loss of function allele that allowed us to monitor Imp1 expression by β-galactosidase activity ( Hansen et al . , 2004 ) . At E10 . 5 , Imp1 was expressed throughout the VZ of the entire developing brain with the exception of the floor plate and roof plate ( Figure 1—figure supplement 1A ) . The anatomy of the developing forebrain and the Imp1 expression pattern are schematically summarized in Figure 1—figure supplement 2 . At later stages ( E12 . 5-E16 . 5 ) Imp1 expression was gradually restricted , mainly to the dorsomedial telencephalon ( DMT ) , where it continued to be expressed by Pax6+ undifferentiated neural stem/progenitor cells in the VZ/SVZ ( Figure 1D , E , Figure1—figure supplement 1B ) . There was little or no Imp1 expression by the differentiated neurons that accumulated at the cortical plate ( Figure 1E ) . At birth , there was little or no Imp1 expression in the cerebral cortex ( Figure 1D ) . In situ hybridization to endogenous Imp1 transcripts revealed a similar expression pattern as observed with X-gal staining of Imp1β-geo/+ mice: Imp1 was mainly expressed in the VZ/SVZ of the dorsal telencephalon at E14 . 5 and was not detected in P60 hippocampus or olfactory bulb ( Figure 1—figure supplement 1E–G ) . We cultured cells from E12 . 5 or P0 cerebral cortex or P60 lateral ventricle SVZ in non-adherent cultures at clonal density . Almost all neurospheres formed by E12 . 5 Imp1β-geo/+ telencephalon cells , but not littermate control cells , stained with X-gal and this staining was maintained upon passaging of neurospheres ( Figure 1F , Figure 1—figure supplement 1C ) . Immunostaining for β-galactosidase and Nestin co-localized in sections from E12 . 5 Imp1β-geo/+ neurospheres ( Figure 1G ) . We could not detect X-gal staining in neurospheres cultured from P0 or P60 Imp1β-geo/+ lateral ventricle VZ cells ( Figure 1F , Figure 1—figure supplement 1D ) . Imp1 is therefore expressed in neural stem/progenitor cells in the fetal telencephalon but it’s expression is extinguished postnatally . Consistent with an earlier report ( Hansen et al . , 2004 ) , Imp1 deficiency led to growth retardation in mice that was evident by late fetal development and persisted into adulthood ( Figure 2—figure supplement 1A , C ) . The brains of Imp1β-geo/β-geo mice were also significantly ( p<0 . 01 ) smaller than the brains of littermate controls at P0 and P30 ( Figure 2—figure supplement 1B , D ) . Histological analysis of E18 . 5 Imp1β-geo/β-geo and littermate control brains showed that pallial expansion was impaired in Imp1β-geo/β-geo brains ( Figure 2A ) . The pial surface from the pallial/subpallial boundary to the retrosplenial cortex , was significantly ( p<0 . 05 ) shortened in Imp1β-geo/β-geo brains as compared to littermate controls ( L in Figure 2A , D ) . The lateral ventricle was collapsed in Imp1β-geo/β-geo brains , in contrast to littermate controls ( Figure 2A ) . Cortical thickness was not significantly affected by Imp1 deficiency ( T in Figure 2A , C ) . These morphological abnormalities in the cortex first became apparent around E14 . 5 , with a shortened pial surface length , and became increasingly severe throughout the rest of development ( Figure 2A , D ) . At E16 . 5 and E18 . 5 , the pial surface length was significantly shorter in the Imp1β-geo/β-geo forebrain ( Figure 2D ) and the lateral ventricle collapsed ( Figure 2A ) . 10 . 7554/eLife . 00924 . 006Figure 2 . Imp1 deficiency reduces brain size and pallial expansion due to reduced proliferation of fetal neural stem/progenitor cells . ( A ) Coronal sections of Imp1+/+ and Imp1β-geo/ β-geo telencephalons . The lateral ventricle is indicated with an asterisk in the Imp1+/+ brain . Morphological abnormalities were visible in the Imp1β-geo/β-geo telencephalon as early as at E14 . 5 . Pallial expansion was impaired and the lateral ventricle collapsed in the Imp1β-geo/ β-geo brain at E18 . 5 . The length ( L ) and thickness ( T ) of the pallial regions are indicated with yellow arrows . ( B ) Dorsomedial telencephalon ( DMT ) sections from E12 . 5 , E14 . 5 , E16 . 5 or E18 . 5 Imp1+/+ Imp1β-geo/+ or Imp1β-geo/β-geo embryos were stained with an antibody against BrdU . Reduction of cell proliferation was apparent in Imp1β-geo/ β-geo telencephalon at E16 . 5 and E18 . 5 . ( C–D ) The length ( L ) of the pial surface from the pallial/subpallial boundary to the retrosplenial cortex ( L in panel A ) was significantly shortened in the E16 . 5 and E18 . 5 Imp1β-geo/β-geo telencephalon ( p<0 . 05 , p<0 . 01; four brains/genotype ) , but cortical thickness ( T ) was not significantly affected . ( E ) BrdU immunostaining revealed a significant reduction in the frequency of proliferating cells in E16 . 5 and E18 . 5 Imp1β-geo/β-geo telencephalon but not at E14 . 5 or E12 . 5 ( p<0 . 05; mean ± SD for 3–5 brains/genotype/stage and 6–8 sections/brain ) . The reduction was more prominent in DMT , where Imp1 was more strongly expressed , relative to dorsolateral telencephalon ( DLT ) . ( F ) Typical neurospheres after 8 days culture from E12 . 5 or E18 . 5 dorsal telencephalon cells dissociated from wild-type or Imp1β-geo/β-geo mice . ( G–H ) Imp1 deficiency significantly reduced the percentage of cells that formed multilineage neurospheres and their self-renewal potential ( the number of cells from individual primary neurospheres that formed multilineage secondary neurospheres upon subcloning ) , at E18 . 5 but not at E12 . 5 ( p<0 . 05; mean ± SD for 5–6 experiments/stage ) . We observed lower frequency and self-renewal potential of Imp1 deficient multipotent neurospheres in two experiments performed at E15 . 5 . ( I ) At E18 . 5 , but not at E12 . 5 , the percentage of cells within Imp1β-geo/β-geo multilineage colonies that incorporated a 20 min pulse of BrdU was significantly lower than in Imp1+/+ colonies ( three independent experiments/stage; *p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 00610 . 7554/eLife . 00924 . 007Figure 2—figure supplement 1 . Imp1-deficient mice exhibit growth retardation and reduced brain mass but not increased cell death . ( A ) Imp1β-geo/β-geo ( β/β ) mice were smaller than wild-type littermates ( +/+ ) at E16 . 5 , P0 , and P30 . ( B ) Imp1β-geo/β-geo ( β/β ) brains were smaller than wild-type brains ( +/+ ) at P0 and P30 . ( C ) Imp1β-geo/ β-geo mice had significantly reduced body mass relative to wild-type controls ( 8–14 embryos at E16 . 5 , 7–12 mice at P0 , and 6–10 mice at P30; p<0 . 01; error bars represent SD ) . ( D ) Imp1β-geo/β-geo brains ( β/β ) were significantly smaller than wild-type controls ( +/+ ) at P0 and at P30 ( p<0 . 01; 13–14 brains/genotype at P0 , and 6–10 at P30 ) . ( E–G ) To monitor apoptotic cell death , sections from the dorsal telencephalon of wild-type or Imp1β-geo/β-geo mice at E13 . 5 ( E ) or E17 . 5 ( F ) were subjected to TdT-mediated dUTP nick end labeling ( TUNEL ) . Nuclei were visualized with DAPI staining . TUNEL positive cells ( green ) were rare in dorsal telencephalon irrespective of genotype , in contrast to E13 . 5 dorsal root ganglia ( DRG ) or E17 . 5 corpus callosum ( CC; positive controls ) where apoptotic cells were common . Activated caspase-3 positive dying cells were also rare within E18 . 5 neural stem cell colonies irrespective of genotype ( lower two images ) ( G ) . Some cultures were treated with camptothecin to induce apoptosis ( positive controls , upper two images ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 00710 . 7554/eLife . 00924 . 008Figure 2—figure supplement 2 . Imp1 deficiency , but not Imp1-βgeo overexpression , reduced fetal ( but not adult ) neural stem cell self-renewal . ( A-C ) E18 . 5 wild-type ( +/+ ) or Imp1β-geo/β-geo ( β/β ) dorsal telencephalon cells were infected with GFP-only control retrovirus ( GFP ) or 3XFLAG-Imp1-βgeo-GFP retrovirus ( Imp1-βgeo ) . Imp1 deficiency , but not the over-expression of the 3XFLAG-βgeo fusion protein , significantly reduced neurosphere size ( B ) and self-renewal ( C ) . ( p<0 . 05; mean ± SD for three experiments ) . Note that the Imp1-βgeo fusion construct did not contain the Imp1 3′ UTR that has the let-7 binding sites and therefore would not be expected to be influenced by let-7 expression . ( D–F ) Imp1 deficiency did not affect the percentage of cells cultured from P60 SVZ that formed multilineage neurospheres ( D ) , neurosphere size ( E ) , or self-renewal potential ( the number of cells from individual primary neurospheres that formed multilineage secondary neurospheres upon subcloning ) ( F; mean ± SD for three experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 008 It is not clear whether the overall growth retardation ( Figure 2—figure supplement 1A–D ) reflects cell autonomous effects of Imp1 deficiency in stem cells from multiple tissues or whether it reflects non-cell autonomous effects . However , the impaired pallial expansion in the dorsal cortex ( Figure 2A ) coincided precisely with the domain of Imp1 expression that persisted throughout fetal development ( Figure 1D , E; Figure 1—figure supplement 1B , Figure 1—figure supplement 2 ) . We therefore hypothesized that Imp1 is required cell autonomously within stem cells in the dorsal telencephalon to promote pallial expansion , a key event in cortical development . We examined cells within the dorsal telencephalon in more detail to better understand why pallial expansion was impaired in the absence of Imp1 . The number of proliferating cells that incorporated a pulse of BrdU in the VZ did not significantly differ between Imp1β-geo/β-geo and control sections at E12 . 5 or E14 . 5 ( Figure 2B , E ) . However , we detected significantly ( p<0 . 05 ) lower numbers of BrdU+ cells in the VZ of the Imp1β-geo/β-geo dorsomedial telencephalon at E16 . 5 and E18 . 5 , and in the VZ of Imp1β-geo/β-geo dorsolateral telencephalon at E18 . 5 ( Figure 2B , E ) . The reduction in proliferating VZ cells was more prominent in dorsomedial telencephalon , where Imp1 is normally strongly expressed , as compared to the dorsolateral telencephalon where Imp1 is more weakly expressed ( Figure 2E ) . We did not observe any of these effects in Imp1β-geo/+ heterozygous mice ( Figure 2E ) , demonstrating that these effects reflect a loss of IMP1 function rather than a gain-of-function associated with the mutant allele . Moreover , retroviral over-expression of the Imp1-βgeo fusion construct did not significantly affect the size or self-renewal of neurospheres ( Figure 2—figure supplement 2A–C ) . Apoptotic cells were rare and their numbers were not affected by Imp1 deficiency in the E13 . 5 or E17 . 5 telencephalon ( Figure 2—figure supplement 1E , F ) . The reduced proliferation in the dorsal telencephalon appears to reduce pallial expansion in Imp1β-geo/β-geo mice . To assess whether this prenatal reduction in VZ cell proliferation in the Imp1β-geo/β-geo telencephalon affected the self-renewal potential of individual neural stem cells we cultured cells from E12 . 5 , E15 . 5 , and E18 . 5 dorsomedial telencephalon from Imp1β-geo/β-geo mice and littermate controls . We cultured the cells at low density in nonadherent cultures and then transferred individual neurospheres to adherent secondary cultures to determine the percentage of telencephalon cells that formed neurospheres that underwent multilineage differentiation . The percentage of cells that formed multipotent neurospheres did not differ between Imp1β-geo/β-geo and wild-type telencephalon at E12 . 5 , but was reduced in Imp1β-geo/β-geo telencephalon at E15 . 5 and at E18 . 5 ( Figure 2G ) . Imp1β-geo/β-geo neurospheres did not significantly differ from wild-type neurospheres at E12 . 5 , but were smaller than wild-type neurospheres at E15 . 5 and E18 . 5 and formed significantly ( p<0 . 05 ) fewer multipotent secondary neurospheres upon subcloning ( Figure 2F , H ) . This reduced self-renewal potential was associated with reduced proliferation within Imp1β-geo/β-geo stem cell colonies at E15 . 5 and E18 . 5 ( Figure 2I ) . As in vivo , cell death was rare within colonies of both genotypes ( Figure 2—figure supplement 1G ) . The observation that IMP1 promotes the self-renewal of individual neural stem cells in culture demonstrates that IMP1 acts autonomously within neural stem/progenitor cells from the dorsal telencephalon to promote self-renewal . We also cultured cells from the lateral ventricle SVZ of adult ( P60 ) Imp1β-geo/β-geo mice and littermate controls to assess whether Imp1 deficiency affected the self-renewal of adult neural stem cells . Consistent with our inability to detect Imp1 expression in these cells ( Figure 1—figure supplement 1D , F , G ) , the percentage of cells that formed multipotent neurospheres , neurosphere size , and self-renewal potential did not significantly differ between Imp1β-geo/β-geo and wild-type cells ( Figure 2—figure supplement 2D–F ) . To assess whether neural stem cell depletion was evident in the Imp1β-geo/β-geo telencephalon in vivo , we examined the number of Pax6+ neural stem cells in the VZ of Imp1β-geo/β-geo mice and littermate controls . The number of Pax6+ cells did not differ among genotypes at E12 . 5 , but was significantly ( p<0 . 05 ) reduced in the Imp1β-geo/β-geo dorsomedial telencephalon from E14 . 5 to E18 . 5 and in the Imp1β-geo/β-geo dorsolateral telencephalon at E18 . 5 ( Figure 3A , B ) . The percentage of Pax6+ cells that were also BrdU+ or phospho-Histone H3+ ( pH3+ ) did not differ between Imp1β-geo/β-geo and control telencephalon at E12 . 5 or E14 . 5 , but was significantly ( p<0 . 05 ) reduced in the Imp1β-geo/β-geo dorsomedial telencephalon at E18 . 5 ( Figure 3A , C , Figure 3—figure supplement 1A , B ) . Similar to the overall reduction of cell proliferation ( Figure 2B , E ) , the frequencies of Pax6+ cells and Pax6+BrdU+ cells were more strongly reduced in the dorsomedial telencephalon where Imp1 is strongly expressed , as compared to the dorsolateral telencephalon ( Figure 3B , C ) . We did not observe these effects in Imp1β-geo/+ heterozygous mice ( Figure 3B , C , E , F ) , demonstrating that they do not reflect a gain-of-function associated with the mutant allele . Pax6+ neural stem cells therefore become depleted in the Imp1β-geo/β-geo dorsal telencephalon in vivo as a consequence of a loss of IMP1 function . 10 . 7554/eLife . 00924 . 009Figure 3 . Imp1 deficiency leads to precocious maturation of Pax6+ stem cells into Tbr2+ intermediate neuronal progenitors in the dorsal telencephalon . ( A–C ) Imp1 deficiency significantly reduced the number of Pax6+ neural stem cells in the dorsomedial telencephalon ( DMT ) at E14 . 5 and E18 . 5 , and in the dorsolateral telencephalon ( DLT ) at E18 . 5 ( p<0 . 05; mean ± SD for 3–4 mice/genotype at each stage with 6–8 sections/brain ) . Imp1 deficiency significantly reduced the percentage of Pax6+ neural stem cells that were BrdU+ in E18 . 5 DMT ( p<0 . 05; mean ± SD for 3–4 mice/genotype at each stage with 6–7 sections/brain ) . ( D–F ) Imp1 deficiency transiently increased the number of Tbr2+ intermediate progenitors in the DMT at E12 . 5 and E14 . 5 , and in the DLT at E12 . 5 ( p<0 . 01; mean ± SD for 3–5 brains/genotype at each stage with 6–8 sections/brain ) . Imp1 deficiency did not significantly affect the percentage of Tbr2+ cells that were also BrdU+ ( mean ± SD for 3–4 mice/genotype with six sections/brain ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 00910 . 7554/eLife . 00924 . 010Figure 3—figure supplement 1 . Imp1 deficiency significantly reduced the percentage of proliferating Pax6+ neural stem cells in E18 . 5 dorsomedial telencephalon ( DMT ) . ( A–B ) DMT sections stained with antibodies against Pax6 and phospho-Histone H3 ( pH3 ) , a marker of mitotic cells . ( B ) Imp1 deficiency significantly reduced the percentage of Pax6+ neural stem cells that were pH3+ in E18 . 5 DMT but not at E12 . 5 or E14 . 5 , or in DLT ( p<0 . 05; mean ± SD for 3–4 brains/genotype at each stage with 4–7 sections/brain ) . ( C–D ) DMT sections stained with antibodies against Tbr2 and phospho-Histone H3 . ( D ) Imp1 deficiency did not significantly affect the percentage of Tbr2+ cells that were also pH3+ ( mean ± SD for 3–4 brains/genotype with six sections/brain ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 010 Complementary to the depletion of Pax6+ stem cells , the number of Tbr2+ intermediate neuronal progenitors was significantly ( p<0 . 01 ) increased in E12 . 5 and E14 . 5 Imp1β-geo/β-geo dorsomedial telencephalon and in E12 . 5 Imp1β-geo/β-geo dorsolateral telencephalon as compared to littermate controls ( Figure 3D , E ) . This increase in the number of Tbr2+ intermediate neuronal progenitors was transient as it was no longer statistically significant at E18 . 5 ( Figure 3D , E ) . The increase in the number of Tbr2+ cells was not attributable to increased proliferation by these cells as the percentage of Tbr2+ cells that were also BrdU+ or pH3+ was not significantly different between Imp1β-geo/β-geo and control dorsal telencephalon at E12 . 5 , E14 . 5 , or E18 . 5 ( Figure 3D , F , Figure3—figure supplement 1C , D ) . This suggests that Pax6+ stem cells are depleted in the absence of IMP1 by precocious maturation into Tbr2+ intermediate progenitors , transiently expanding the number of Tbr2+ cells from E12 . 5 to E14 . 5 . The number of Tuj1+ neurons was significantly ( p<0 . 05 ) increased in the dorsomedial telencephalon of Imp1β-geo/β-geo mice at E12 . 5 and E14 . 5 relative to littermate controls ( Figure 4A , B ) . The number of TAG-1+ corticofugal projection neurons was also significantly ( p<0 . 05 ) increased in the dorsal telencephalon of Imp1β-geo/β-geo mice at E12 . 5 and E14 . 5 ( not shown ) . We also dissociated cells from the telencephalons of Imp1β-geo/β-geo mice and littermate controls at E12 . 5 and cultured them adherently at clonal density . When colonies were stained after 9 days culture , we observed elevated numbers of Tuj1+ neurons within Imp1β-geo/β-geo multilineage colonies as compared to control multilineage colonies ( Figure 4C ) . The frequency of neuron-only colonies formed by Imp1β-geo/β-geo telencephalon cells was also significantly ( p<0 . 05 ) increased ( Figure 4D ) but the number of cells within these colonies was not affected ( Figure 4E ) . 10 . 7554/eLife . 00924 . 011Figure 4 . IMP1 prevents premature neuronal and glial differentiation by stem cells in the dorsal telencephalon . ( A ) Dorsomedial telencephalon sections from E12 . 5 or E14 . 5 control ( +/+ ) or Imp1β-geo/β-geo ( β/β ) embryos were stained with an antibody against the neuronal marker Tuj1 . ( B ) The number of Tuj1+ neurons per section was significantly increased in Imp1β-geo/β-geo ( β/β ) DMT as compared to littermate controls ( +/+: Imp1+/+ or β/+: Imp1β-geo/+ ) at E12 . 5 and E14 . 5 ( p<0 . 05; mean ± SD for four brains/genotype with 4–6 sections/brain ) . ( C–E ) E12 . 5 dorsal telencephalon cells were cultured adherently for 9 days at clonal density . ( C ) Tuj1+ neurons were significantly more common in multipotent colonies from Imp1β-geo/β-geo ( β/β ) as compared to Imp1+/+ ( +/+ ) mice . ( D ) Significantly more neuron-only colonies were formed by Imp1β-geo/β-geo ( β/β ) as compared to Imp1+/+ ( +/+ ) telencephalon cells; however , the number of cells within control ( +/+ ) or Imp1β-geo/β-geo ( β/β ) neuron-only colonies did not significantly differ ( E; p<0 . 05; mean ± SD for three independent experiments ) . ( F ) Dorsal telencephalon sections from E18 . 5 control ( +/+ ) or Imp1β-geo/β-geo ( β/β ) embryos were stained with an antibody against GFAP . ( G ) The number of GFAP+ astrocytes per section was significantly increased in the Imp1β-geo/β-geo ( β/β ) dorsal telencephalon as compared to littermate controls ( +/+ ) at E18 . 5 ( p<0 . 05; mean ± SD for four brains/genotype with 6–8 sections/brain ) . ( H–J ) E18 . 5 dorsal telencephalon cells were cultured adherently for 9 days at clonal density . ( H ) GFAP+ astrocytes were more common in multipotent colonies from Imp1β-geo/β-geo ( β/β ) as compared to Imp1+/+ ( +/+ ) mice . ( I ) Significantly more glia-only colonies were formed by Imp1β-geo/β-geo ( β/β ) as compared to Imp1+/+ ( +/+ ) telencephalon cells; however , the number of cells within control ( +/+ ) or Imp1β-geo/β-geo ( β/β ) glia-only colonies did not significantly differ ( J; p<0 . 05; mean ± SD for three independent experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 011 Although significant increases in the numbers of Tbr2+ progenitors and Tuj1+ neurons were observed beginning at E12 . 5 we did not observe a significant depletion of Pax6+ neural stem cells until E14 . 5 . We believe this is because Pax6+ stem cells are much more numerous than Tbr2+ cells and Tuj1+ cells at E12 . 5 . In wild-type mice we counted 617 Pax6+ cells as compared to only 55 Tbr2+ cells and 74 Tuj1+ cells in the same sections ( Figure 3B , E and 4B ) . In Imp1-mutant mice we counted 633 Pax6+ cells as compared to 121 Tbr2+ cells and 117 Tuj1+ cells in the same sections . Therefore , the average increases in Tbr2+ and Tuj1+ cells totalled only 109 cells per section . Given that the standard deviation in Pax6+ cells per section was approximately 100 , we were not able to detect such small changes in the number of Pax6+ cells at E12 . 5 . At later stages of development when larger increases in Tbr2+ cells and Tuj1+ cells were observed we did detect significant declines in the numbers of Pax6+ cells . To test whether Imp1 deficiency also leads to premature gliogenesis , we examined telencephalon sections from Imp1β-geo/β-geo mice and littermate controls at E18 . 5 . The number of GFAP+ astrocytes was significantly increased in the Imp1β-geo/β-geo telencephalon ( Figure 4F , G ) , suggesting that gliogenesis is also precocious in Imp1β-geo/β-geo mice . We also adherently cultured cells from the E18 . 5 telencephalon of Imp1β-geo/β-geo mice and littermate controls at clonal density . When colonies were stained after 9 days of culture we observed increased numbers of GFAP+ cells within Imp1β-geo/β-geo neural stem cell colonies ( Figure 4H ) . The frequency of glia-only colonies formed by Imp1β-geo/β-geo telencephalon cells was also significantly ( p<0 . 05 ) increased ( Figure 4I ) but the number of cells within these colonies was not affected ( Figure 4J ) . The observation that the effects of Imp1 deficiency in vivo are greatest in the dorsal telencephalon where Imp1 expression is highest , and the observation that these effects are also observed when cells are cultured at clonal density , suggest that IMP1 acts autonomously within neural stem cells in the dorsal telencephalon to prevent premature differentiation . To directly test whether IMP1 acts autonomously within neural stem cells , we injected virus bearing Imp1 shRNA or scrambled control RNA into the telencephalic ventricles of wild-type mice in utero . Infection with the Imp1 shRNA virus , but not the control virus , efficiently reduced IMP1 expression in neurospheres cultured from E14 . 5 telencephalon ( Figure 5A ) . When these viruses were injected in the telencephalic ventricles of E14 . 5 wild-type mice in utero and analysed three days later , we distinguished infected from non-infected cells based on GFP expression ( which was also carried in the viral vectors ) in the dorsal telencephalon . GFP+ cells infected by the control shRNA tended to localize apically , among dividing Pax6+ cells in the VZ/SVZ ( Figure 5B , Figure 5—figure supplement 1 ) . In contrast , most GFP+ cells infected by the Imp1 shRNA tended to localize basally , among Tbr2+ and Tuj1+ cells ( Figure 5B , Figure 5—figure supplement 1 ) . When cell proliferation was assessed by Ki67+ immunostaining , the percentage of GFP+ cells that were also Ki67+ was significantly ( p<0 . 05 ) reduced in mice infected with Imp1 shRNA as compared to control shRNA ( Figure 5C , Figure5—figure supplment 1 ) . This indicates that Imp1 acts cell autonomously to maintain the proliferation of stem/progenitor cells in the dorsal telencephalon . 10 . 7554/eLife . 00924 . 012Figure 5 . In utero knockdown of Imp1 cell-autonomously reduces cell proliferation and accelerates differentiation of neural stem/progenitor cells . ( A ) Western blot of neurospheres cultured from E14 . 5 wild-type telencephalon cells infected with lentivirus bearing either control ( scrambled ) shRNA or Imp1 shRNA . Imp1 shRNA reduced IMP1 expression . ( B and C ) Viruses expressing either control or Imp1 shRNA were injected into the telencephalic ventricles of E14 . 5 wild-type mice , infecting a small percentage of cells that could be identified based on GFP expression . Brains were fixed at E17 . 5 and dorsomedial telencephalon sections were immunostained with antibodies against GFP and Ki67 . ( B ) Low magnification view of sections through the dorsal telencephalon including VZ ( apical; bottom ) and differentiated cell layers ( basal; top ) . In Imp1 shRNA infected telencephalon , GFP+ cells were more likely to be found basally as compared to control shRNA infected cells , suggesting that Imp1 shRNA promoted the differentiation of infected cells . ( C ) The percentages of GFP+ ( infected ) cells that were Ki67+ ( dividing ) or Ki67- at E17 . 5 . Imp1 shRNA infection significantly reduced the percentage of GFP+ cells that were Ki67+ ( p<0 . 05; mean ± SD for four experiments ) . ( D–F ) Viruses expressing either control or Imp1 shRNA were injected into the telencephalic ventricles of E12 . 5 wild-type mice . Brains were fixed at E15 . 5 and dorsomedial telencephalon sections were immunostained to assess the differentiation of GFP+ cells . ( D ) Imp1 shRNA significantly increased the percentage of GFP+ cells that were Tbr2+ or Tuj1+ and significantly reduced the percentage that were Pax6+ ( *p<0 . 05; mean ± SD for five experiments ) . ( E ) Triple immunostaining with antibodies against GFP , Pax6 , and Tuj1 . GFP+/Pax6+ cells ( yellow cells in merged image ) are marked with arrowheads , and GFP+/Pax6- cells are indicated with arrows . ( F ) Triple immunostaining with antibodies against GFP , Tbr2 , and Tuj1 . GFP+/Tbr2+ cells ( yellow cells in merged image ) are marked with arrowheads and GFP+/Tbr2- cells are indicated with arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 01210 . 7554/eLife . 00924 . 013Figure 5—figure supplement 1 . Imp1 knockdown in neural stem cells reduces cellular proliferation . Viruses expressing either control ( scrambled ) shRNA or Imp1 shRNA were injected into the telencephalic ventricles of E14 . 5 wild-type mice . Brains were fixed at E17 . 5 and dorsomedial telencephalon sections were immunostained with antibodies against GFP and Ki67 . GFP+/Ki67+ dividing , virally infected cells ( yellow cells in merged image ) are indicated with arrowheads , and GFP+/ Ki67- non-dividing virally infected cells ( green cells in merged image ) are indicated with arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 013 To assess whether Imp1 knockdown affects the differentiation of neural stem cells , we injected viruses into the telencephalic ventricles of E12 . 5 wild-type mice in utero and 3 days later sections were immunostained with antibodies against Pax6 , Tbr2 and Tuj1 . In mice injected with Imp1 shRNA , significantly fewer GFP+ cells were Pax6+ and significantly more GFP+ cells were Tbr2+ or Tuj1+ as compared to control shRNA ( Figure 5D–F ) . These data indicate that Imp1 acts cell autonomously to maintain Pax6+ stem cells in the dorsal telencephalon by opposing their maturation into Tbr2+ intermediate progenitors and their differentiation into neurons . The timing of neuronal differentiation by stem/progenitor cells in the telencephalon is regulated by the timing of cell cycle exit , such that over-expression of cyclin D prolongs the proliferation of undifferentiated cells and delays the onset of neurogenesis ( Dehay and Kennedy , 2007; Lange et al . , 2009 ) . We therefore examined the expression of cell cycle regulators in the dorsal telencephalon of Imp1β-geo/β-geo mice . We observed a significant ( p<0 . 01; *p<0 . 05 ) decline in the levels of all cyclin D family transcripts by both antibody staining ( Figure 6A ) and qPCR ( Figure 6B ) in the dorsomedial telencephalon of Imp1β-geo/β-geo mice as compared to control mice ( Figure 6A ) . The decline in cyclin D1 staining was most pronounced in the dorsomedial telencephalon ( Figure 6A , arrow ) where Imp1 expression was strongest ( Figure 6A , arrowhead ) . 10 . 7554/eLife . 00924 . 014Figure 6 . Imp1 deficiency reduced cyclin D expression and accelerated cell cycle exit in the dorsal telencephalon . ( A ) Sections of E13 . 5 wild-type or Imp1β-geo/β-geo dorsal telencephalon were immunostained with antibodies against Cyclin D1 and LacZ . Cyclin D1 expression was reduced relative to control in Imp1β-geo/β-geo DMT ( see arrow ) where strong LacZ immunostaining indicated the highest levels of Imp1 expression ( see arrowhead ) . In contrast , Cyclin D1 immunostaining was retained in Imp1β-geo/β-geo DLT , where lacZ immunostaining was weak . Higher magnification images on the right show boxed areas from low magnification images . ( B ) qPCR analysis of cyclin D and cyclin E transcripts in dorsomedial ( green bar ) and dorsolateral ( purple bar ) telencephalon from E13 . 5 mice . Each bar represents the fold change in Imp1β-geo/β-geo/wild-type ( error bars represent SD , four brains/genotype; p<0 . 01 , p<0 . 05 ) . ( C ) Dorsal telencephalon sections from E14 . 5 Imp1+/+ or Imp1β-geo/β-geo mice that had been administered a single pulse of BrdU at E13 . 5 were stained with anti-BrdU and anti-Ki67 antibodies . Cells that exited the cell cycle after BrdU incorporation were BrdU+Ki67- ( green; arrows ) while cells that continued to divide were BrdU+Ki67+ cells ( yellow; arrowheads ) . The Imp1β-geo/β-geo telencephalon had a significantly higher percentage of BrdU+Ki67- cells ( 22 ± 4%; p<0 . 05; mean ± SD for four brains/genotype; 4–6 sections/brain ) . ( D ) Most BrdU+Ki67+ cells expressed Pax6 and most BrdU+Ki67- cells were either Tbr2+ or Tuj1+ . Single channel images are presented in Figure 6—figure supplement 1 . ( E–G ) E18 . 5 wild-type ( +/+ ) or Imp1β-geo/β-geo ( β/β ) neural stem cells were infected with either GFP-only control retrovirus , cyclin D1-GFP retrovirus , or cyclin D2-GFP retrovirus and cultured . Within the resulting neural stem cell colonies , cell proliferation was assessed by BrdU incorporation ( E ) , glial differentiation was assessed based on levels of GFAP staining ( F ) , and Cyclin D1 or Cyclin D2 expression was examined by western blot ( G ) . Imp1 deficiency reduced Cyclin D1 or Cyclin D2 expression and neural stem cell proliferation and increased gliogenesis . These proliferation and premature gliogenesis phenotypes were partially rescued by cyclin D1 or cyclin D2 over-expression ( three experiments; p<0 . 01 , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 01410 . 7554/eLife . 00924 . 015Figure 6—figure supplement 1 . BrdU+/Ki67+ cells were Pax6+ while BrdU+/Ki67- cells were Tbr2+ or Tuj1+ in E14 . 5 dorsomedial telencephalon . Brains were dissected from E14 . 5 wild-type or Imp1β-geo/β-geo mice after a 24 hr pulse of BrdU . Coronal sections were immunostained with antibodies against BrdU , Ki67 , and Pax6 ( A ) , Tbr2 ( B ) , or Tuj1 ( C ) . Most BrdU+Ki67+ cells ( yellow cells in Ki67/BrdU images ) expressed Pax6 and most BrdU+Ki67- cells ( red cells in Ki67/BrdU images ) were either Tbr2+ or Tuj1+ . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 01510 . 7554/eLife . 00924 . 016Figure 6—figure supplement 2 . Canonical Wnt signaling promotes Imp1 expression . ( A ) In E12 . 5 dorsolateral telencephalon explants cultured for 12 hr , Imp1 transcript levels were significantly increased by Wnt-3a , decreased by Dkk-1 , and unaffected by BMP4 . Each bar represents the ratio of Imp1 transcript levels in the presence/absence of the recombinant proteins ( error bar represent SD , 3–5 independent experiments; p<0 . 01; p<0 . 05 ) . ( B ) Within 3 kb upstream of the transcription starting site ( TSS , designated as +1 ) , two sequences are conserved across species and match consensus binding sites for TCF/LEF transcription factors ( A and B; green ovals ) . The 5′-untranslated region of Imp1 is shown as a blue box and exons are shown as red boxes . We used exon 6 sequence as an internal negative control during chromatin immunoprecipitation . ( C ) Chromatin immunoprecipitation of TCF4 from E12 . 5 wild-type CNS neurospheres . Two sites in the promoters of Imp1 ( sites A and B in panel B ) , and Axin-2 ( a positive control ) were significantly enriched by TCF4 immunoprecipitation ( blue bars ) compared to IgG immunoprecipitation ( orange bars ) . No enrichment was detected for exon 6 of Imp1 or Lgi4 ( negative controls ) . ( D ) Luciferase assay performed in P19 embryonal carcinoma cells transfected with plasmids either containing intact Imp1 enhancer/promotor ( A+/B+ ) , site A eliminated ( A mt /B+ ) , site B eliminated ( A+/B mt ) , or both site A/site B eliminated ( A mt/B mt ) . TOP-flash or empty vector are included as positive or negative controls . Site B elimination significantly reduced luciferase activity whereas site A elimination had little effect ( error bar represent SD of three independent experiments; p<0 . 05 ) . ( E ) Imp1 deficiency did not affect the levels of ß-catenin ( Ctnnb1 ) transcripts in E13 . 5 dorsomedial ( green bar ) or dorsolateral telencephalon ( purple bar ) , or in E13 . 5 neurospheres ( blue bar ) . Error bars represent SD; three brains/genotype , three independent experiments . ( F ) Imp1 deficiency did not affect ß-catenin protein levels in E13 . 5 dorsal telencephalon or neurospheres . ( G ) Imp1 deficiency did not significantly affect the stability of ß-catenin mRNA after treatment with Actinomycin D in neurospheres cultured from E13 . 5 control ( blue ) or Imp1β-geo/β-geo ( red ) dorsal telencephalons ( error bars represent SD; three independent experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 016 To assess whether increased neurogenesis in the Imp1-deficient telencephalon was associated with accelerated cell cycle exit , E14 . 5 Imp1β-geo/β-geo and littermate control mice were pulse labeled with BrdU for 24 hr then sacrificed and sections were stained with antibodies against BrdU and Ki67 . The fraction of cells that exited the cell cycle during the BrdU pulse was estimated based on the frequency of BrdU+Ki67- cells and the fraction that remained in cycle was estimated based on the frequency of BrdU+Ki67+ cells , as described previously ( Chenn and Walsh , 2002 ) . The frequency of BrdU+Ki67- cells was significantly ( p<0 . 05 ) higher in the Imp1β-geo/β-geo as compared to control telencephalon ( Figure 6C ) , suggesting that cell cycle exit is accelerated in neural stem/progenitor cells from the Imp1β-geo/β-geo telencephalon . We could not detect a significant difference in the number of BrdU+Ki67+ cells . Most BrdU+Ki67+ cells were Pax6+ and most BrdU+Ki67- cells were either Tbr2+ or Tuj1+ ( Figure 6D , Figure 6—figure supplement 1 ) . These data suggest that most BrdU+Ki67+ cells were stem/progenitor cells while BrdU+Ki67- cells were a mixture of newborn neurons and neuronal progenitors . Over-expression of cyclin D1 or cyclin D2 in neural stem/progenitor cells from E18 . 5 Imp1β-geo/β-geo telencephalon significantly increased Cyclin D1 or Cyclin D2 protein levels and proliferation , and reduced premature glial differentiation by Imp1β-geo/β-geo cells ( Figure 6E–G ) . This suggests that reduced Cyclin D1 or Cyclin D2 expression contributes to the defects in Imp1 deficient neural stem/progenitor cells . IMP1 is therefore required to maintain the expression of proteins that promote progression through G1 phase of the cell cycle such that Imp1 deficiency accelerates cell cycle exit and differentiation . During corticogenesis , a medial-lateral gradient of canonical Wnt signaling maintains the proliferation of undifferentiated Pax6+ stem cells in the pallial region of the telencephalon , preventing the premature generation of Tbr2+ intermediate progenitors and differentiated neurons ( Chenn and Walsh , 2002; Machon et al . , 2007; Wrobel et al . , 2007; Mutch et al . , 2010 ) . The Imp1 expression pattern we observed ( Figure 1D and 6A , Figure 1—figure supplement 1B ) was reminiscent of the gradient of Wnt signaling observed in the telencephalon ( Machon et al . , 2007; Mutch et al . , 2010 ) ; furthermore , the Imp1 loss-of-function phenotype in the telencephalon ( Figures 2–6 ) was reminiscent of the phenotype observed in mutants with reduced Wnt signaling ( Mutch et al . , 2010 ) . We therefore cultured E12 . 5 wild-type lateral telencephalon explants for 12 hr with or without recombinant Wnt3a , and examined Imp1 transcript levels by qPCR . Imp1 expression significantly ( p<0 . 01 ) increased in the presence of Wnt3a ( Figure 6—figure supplement 2A ) . Recombinant Dkk-1 , which inhibits Wnt signaling , significantly ( p<0 . 05 ) reduced Imp1 expression ( Figure 6—figure supplement 2A ) . Addition of BMP4 , which also regulates dorsoventral patterning in the telencephalon ( Hebert et al . , 2003 ) , did not affect Imp1 expression ( Figure 6—figure supplement 2A ) . These observations indicate that Wnt signaling can promote Imp1 expression in neural stem/progenitor cells . Next we assessed whether TCF4 , a transcriptional mediator of canonical Wnt signaling , can directly bind to the Imp1 enhancer/promotor . We did not detect TCF4 binding to exon 6 of Imp1 or to the promoter/enhancer of Lgi4 ( negative controls ) but did detect TCF4 binding to the first intron of Axin ( a positive control [Jho et al . , 2002] ) ( Figure 6—figure supplement 2C ) . We also examined two putative TCF/Lef binding sites that are conserved across species and located within 3 kb upstream of the Imp1 translational start site ( sites A and B in Figure 6—figure supplement 2B ) . We detected significant enrichment of TCF4 binding at both sites by chromatin immunoprecipitation ( Figure 6—figure supplement 2C ) . Relative luciferase activity was significantly reduced when one of these TCF/Lef binding sites ( site B in Figure 6—figure supplement 2B ) was eliminated ( Figure 6—figure supplement 2D ) . These observations suggest that canonical Wnt signaling might directly regulate Imp1 expression . To assess whether Imp1 expression is regulated by canonical Wnt signaling under physiological conditions we examined the telencephalons of Apc mutant mice ( hGFAP-Cre; Apcflox/flox ) to assess the consequences of increased Wnt signaling and ß-catenin mutant mice ( Nestin-Cre; Ctnnb1flox/flox ) to assess the consequences of decreased Wnt signaling . Imp1 expression was increased in the dorsal telencephalon of Apc deficient mice and decreased in the dorsal telencephalon of ß-catenin deficient mice relative to littermate controls ( Figure 7A , B ) . Wnt signaling therefore promotes Imp1 expression in the telencephalon in vivo . 10 . 7554/eLife . 00924 . 017Figure 7 . Canonical Wnt signaling promotes , and let-7 inhibits , Imp1 expression . ( A–B ) Imp1 transcript levels were elevated in E14 . 5 dorsal telencephalon of Apc-deficient mice , and reduced in ß-catenin ( Ctnnb1 ) -deficient mice by both in situ hybridization ( A ) and qPCR ( B ) . Bars represent fold change in Imp1 transcript levels in dorsomedial ( green bars ) and dorsolateral ( purple bars ) telencephalon of the indicated mutant mice/wild-type controls ( p<0 . 01 , p<0 . 05; error bar represents SD , 3–4 brains/genotype ) . ( C ) Western blot for IMP1 or α-tubulin in E13 . 5 telencephalon cells isolated from doxycycline administered wild-type ( +/+ ) or let-7 inducible transgenic mice ( ilet-7 Tg ) . ( D–E ) Dorsomedial telencephalon sections were prepared from doxycycline administered E13 . 5 wild-type ( +/+ ) or let-7 inducible transgenic mice ( ilet-7 Tg ) . Induction of let-7 transgene expression significantly reduced the number of proliferating cells ( assessed by a 1 hr pulse of BrdU ) and the frequency of Pax6+ cells , and increased the numbers of Tbr2+ intermediate progenitors and Tuj1+ neurons ( p<0 . 01 , *p<0 . 05; error bar represents SD , 3–4 mice/genotype ) . ( F–I ) E18 . 5 dorsal telencephalon cells or P60 SVZ cells were isolated from doxycycline administered wild-type ( +/+ ) or Lin28a inducible transgenic mice ( Lin28a Tg ) and cultured as neurospheres . ( F ) Western blot for IMP1 or α-tubulin . IMP1 protein expression was elevated in Lin28a transgenic cells relative to control at E18 . 5 and not detected in either cells at P60 . ( G–I ) Lin28a induction significantly increased the size of E18 . 5 neurospheres ( G and H ) and their self-renewal potential ( I; p<0 . 01; mean ± SD for four experiments ) . ( J–K ) E12 . 5 wild-type dorsal telencephalon or P60 wild-type SVZ cells were infected with GFP-only control retrovirus ( GFP ) , 3′UTR truncated 3XFLAG-Imp1-GFP retrovirus ( 3Δ ) , or full length 3XFLAG Imp1-GFP retrovirus ( full ) . Truncated Imp1 lacked let-7 binding sites in the 3′ UTR . ( J ) Over-expression of truncated 3XFLAG-Imp1-GFP increased IMP1 protein expression more efficiently than full length 3XFLAG Imp1-GFP over-expression at P60 when let-7 expression is high . The smaller band corresponds to endogenous IMP1 while the larger band corresponds to FLAG-tagged IMP1 . ( K ) Only 3′ UTR truncated 3XFLAG Imp1-GFP over-expression significantly increased the self-renewal of neurospheres relative to uninfected cells at P60 . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 01710 . 7554/eLife . 00924 . 018Figure 7—figure supplement 1 . let-7 over-expression inhibits neural stem cell self-renewal while Imp1 over-expression inhibits neurogenesis . Conditional deletion of let-7b/c2 did not affect neural stem cell function or Imp1 or Hmga2 expression . ( A–D ) Neurospheres were cultured from E13 . 5 dorsal telencephalons isolated from doxycycline administered wild-type ( +/+ ) or let-7 inducible transgenic mice ( ilet-7 ) . Induction of let-7 significantly reduced the percentage of cells that formed multilineage neurospheres ( B ) , neurosphere size ( C ) , and self-renewal potential ( the number of cells from individual primary neurospheres that formed multilineage secondary neurospheres upon subcloning ) ( D; mean ± SD for three experiments ) . ( E ) In E18 . 5 or P60 Lin28a Tg neurospheres , let-7b expression was significantly reduced and Hmga2 transcript expression was significantly increased compared to control . Hmga2 transcript expression was significantly reduced in E13 . 5 ilet-7 dorsal telecephalon compared to control . ( F–G ) P60 wild-type SVZ cells were infected with GFP-only control retrovirus ( GFP ) , 3’-UTR truncated 3XFLAG-Imp1-GFP retrovirus ( 3Δ ) , or 3′-UTR containing 3XFLAG Imp1-GFP retrovirus ( full-length ) . Only 3′-UTR truncated Imp1 over-expression significantly reduced the number of Tuj1+ neurons per section ( mean ± SD for four experiments ) . ( H–L ) Dorsal telencephalon cells from E18 . 5 wild-type ( +/+ ) or Nestin-Cre; let-7b/c2 conditional mutant ( let-7b/c2 cKO ) mice were cultured non-adherently . let-7b/c2 deficiency did not affect the percentage of cells that formed multipotent neurospheres ( H ) , neurosphere size ( I ) , or neural stem cell self-renewal ( the number or percentage of cells from individual primary neurospheres that formed multilineage secondary neurospheres upon subcloning; J and K ) . Expression levels of Imp1 or Hmga2 transcripts assessed by qPCR were also not significantly altered by let-7b/c2 deficiency ( L ) ( mean ± SD for three experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 01810 . 7554/eLife . 00924 . 019Figure 7—figure supplement 2 . Generation of a conditional mutant allele of let-7b and let-7c2 ( let-7b/c2fl ) by gene targeting . ( A ) In the wild-type genome sequences that encode primary let-7b microRNA ( pri-let-7b ) and primary let-7c2 microRNA ( pri-let-7c2 ) are closely linked ( black boxes ) . A targeting vector was designed to flank these sequences with loxP elements ( green triangles ) , flippase recognition sites ( FRT , purple triangles ) , and a neomycin resistance cassette ( Neo , orange hexagon ) . For negative selection of ES cells that randomly inserted the targeting construct , thymidine kinase ( TK ) and diphtheria toxin ( DT ) were included at both ends of the targeting vector . Correct targeting in ES cells was confirmed by Southern blot using 5′ and 3′ probes ( blue ovals ) . After mice were generated from correctly targeted ES clones , the neomycin resistance cassette was eliminated by mating the mice with Actin-Flpe mice ( Rodriguez et al . , 2000 ) , and confirmed by PCR with primer pairs ( blue arrows ) . S: ScaI endonuclease sites , B: BamHI endonuclease sites . ( B and C ) Correctly targeted ES cell clones were confirmed by Southern blot . ( B ) With 5′-probe , the wild-type allele ( + , arrow ) gave a band of 8 . 2 kb and the mutant allele ( neo , arrowhead ) gave a band of 6 . 6 kb . ( C ) With 3′-probe , the wild-type allele ( + , arrow ) gave a band of 11 . 8 kb and the mutant allele ( neo , arrowhead ) gave a band of 6 . 6 kb . ( D ) PCR genotyping of the progeny from let-7b/c2neo and Actin-Flpe mice . The wild-type allele ( + , arrow ) yielded a 317 bp band and the floxed allele ( flox , arrowhead ) yielded a 408 bp band . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 019 To test whether there is a positive feedback between Imp1 and canonical Wnt signaling ( Gu et al . , 2008 ) in neural stem/progenitor cells , we assessed ß-catenin expression in uncultured E13 . 5 telencephalon and in cultured neurospheres by qPCR and by western blot . However , we did not detect a significant difference in ß-catenin expression between Imp1β-geo/β-geo and control cells ( Figure 6—figure supplement 2E , F ) . We also did not detect a statistically significant difference in the stability of ß-catenin transcripts between Imp1β-geo/β-geo and control neurospheres ( Figure 6—figure supplement 2G ) , suggesting that there is no positive feedback from IMP1 to ß-catenin in neural stem/progenitor cells . IMP1 therefore appears to act downstream of Wnt signaling . The similarity between the Imp1 expression pattern ( Figure 1D ) and the Wnt signaling gradient in the dorsal telencephalon ( Machon et al . , 2007; Mutch et al . , 2010 ) , the observation that Wnt signaling promotes Imp1 expression in the dorsal telencephalon in vivo ( Figure 7A ) , and the similarity in the phenotypic consequences of reduced Imp1 function ( Figures 2–6 ) and reduced Wnt signaling ( Mutch et al . , 2010 ) all suggest that IMP1 acts autonomously within stem cells in the dorsal telencephalon to potentiate the effects of Wnt signaling during pallial expansion . To test whether let-7 regulates Imp1 expression in vivo we administered doxycycline for 5 days to wild-type and tetracyclin-inducible let-7 transgenic ( ilet-7 ) mice ( Zhu et al . , 2011 ) . IMP1 protein expression was reduced in telencephalon cells freshly isolated from E13 . 5 ilet-7 mice relative to controls , indicating that elevated expression of let-7 negatively regulates IMP1 expression ( Figure 7C ) . In E13 . 5 ilet-7 telencephalon , the number of cells that incoporated a 1 hr pulse of BrdU and the number of Pax6+ neural progenitors were significantly reduced relative to controls , whereas the number of Tbr2+ neuronal progenitors and Tuj1+ neurons were significantly increased ( Figure 7D , E ) . let-7 induction thus phenocopied the IMP1 loss-of-function , consistent with the conclusion that let-7 negatively regulates IMP1 expression . To assess whether the elevated expression of let-7 affects neural stem cell function we cultured dorsal telencephalon cells from doxycyclin-treated E13 . 5 wild-type and ilet-7 mice . The percentage of cells that formed multipotent neurospheres , neurosphere size , and self-renewal potential were all significantly ( p<0 . 01 ) reduced in ilet-7+ cells relative to control cells ( Figure 7—figure supplement 1A–D ) . These data indicate that increased let-7 expression negatively regulates Imp1 expression and fetal neural stem cell self-renewal . Since there are 10 let-7 family members that are thought to act redundantly to repress target gene expression ( Roush and Slack , 2008 ) , we used inducible Lin28a transgenic ( Lin28a Tg ) mice ( Zhu et al . , 2010 ) to test whether reduced expression of endogenous let-7s would increase IMP1 expression . Lin28a negatively regulates the expression of all mature let-7 microRNAs ( Heo et al . , 2008; Newman et al . , 2008; Rybak et al . , 2008; Viswanathan et al . , 2008 ) . We cultured E18 . 5 dorsal telencephalon cells or P60 lateral ventricle SVZ cells from doxycycline-treated wild-type or inducible Lin28a Tg mice and observed a significant reduction in endogenous let-7b expression in Lin28a Tg as compared to control neurospheres ( p<0 . 01; sevenfold-reduction in Lin28a Tg E18 . 5 neurospheres and 14-fold reduction in P60 Lin28a Tg neurospheres; Figure 7—figure supplement 1E ) . IMP1 protein expression was elevated in Lin28a Tg neurospheres relative to control neurospheres at E18 . 5 but not at P60 ( Figure 7F ) . Neurosphere size and the self-renewal of multipotent neurospheres were significantly ( p<0 . 01 ) increased in E18 . 5 Lin28a Tg compared to control neurospheres ( Figure 7G–I ) . These data suggest that physiological let-7 expression reduces Imp1 expression and self-renewal potential in fetal but not adult neural stem cells . Since we did not detect Imp1 transcription or Imp1β-geo reporter expression in the postnatal forebrain ( Figure 1D , F , Figure 1—figure supplement 1D , F , G ) , Imp1 expression is silenced postnatally by mechanisms independent of let-7 ( Figure 7F ) . We also generated a floxed allele of the linked microRNAslet-7b and let-7c2 ( let-7b/c2fl; Figure 7—figure supplement 2 ) and conditionally deleted it from fetal neural stem/progenitor cells using Nestin-Cre ( Nestin-Cre; let-7b/c2flox/flox ) . The percentage of cells that formed multipotent neurospheres , neurosphere size , and self-renewal potential did not significantly differ between Nestin-Cre; let-7b/c2flox/flox mice and littermate controls at E18 . 5 or P60 ( Figure 7—figure supplement 1H–K , and data not shown ) . Expression of Imp1 and Hmga2 transcripts also did not significantly differ between telencephalon cells obtained from Nestin-Cre; let-7b/c2flox/flox mice and littermate controls ( Figure 7—figure supplement 1L ) . This is consistent with the expectation that let-7 family members act redundantly to regulate gene expression ( Roush and Slack , 2008 ) such that deletion of let-7b/c2 is not sufficient to change Imp1 expression or neural stem cell function . Although the major effects of Lin28a are mediated by let-7 , there are also let-7-independent effects ( Cho et al . , 2012; Wilbert et al . , 2012 ) . To ensure that the effects of Lin28a on Imp1 expression ( Figure 7F ) are really mediated by changes in the physiological levels of let-7 family microRNAs we independently addressed this issue by testing whether Imp1 expression is regulated by the let-7 binding sites in the 3′ untranslated region ( UTR ) . We overexpressed either full-lengh Imp1 cDNA that contains all let-7 binding sites ( the let-7 sensitive form ) or a truncated Imp1 cDNA that lacks the 3′-UTR ( the let-7 insensitive form ) in E12 . 5 telencephalon cells or P60 SVZ cells . We were able to over-express either form of IMP1 in the E12 . 5 cells , when let-7 expression is low , but in P60 SVZ cells higher levels of IMP1 were expressed from the truncated Imp1 cDNA that lacks the let-7 binding sites ( Figure 7J ) . Ectopic expression of the let-7 insensitive form of Imp1 , but not the let-7 sensitive form , significantly increased self-renewal and significantly reduced the number of differentiated neurons that arose in culture from neurospheres cultured from P60 SVZ ( Figure 7K , Figure 7—figure supplement 1F , G ) . Over-expression of either form of Imp1 did not significantly affect self-renewal in E12 . 5 telencephalon cells ( Figure 7K ) . Regulation of IMP1 expression and neural stem cell function requires let-7 binding sites in the Imp1 3′-UTR when let-7 microRNA expression levels are elevated in vivo . To investigate the mechanism by which IMP1 promotes stem cell self-renewal we identified target RNAs bound by IMP1 in neural stem/progenitor cells . To do this we overexpressed FLAG-tagged IMP1 in neurospheres cultured from E13 . 5 Imp1β-geo/β-geo telencephalon then immunoprecipitated IMP1 and its target RNAs using an anti-FLAG antibody . When these RNAs were identified by deep-sequencing we found transcripts from 103 genes that were significantly enriched in the immunoprecipitated fraction compared with total RNA ( three samples per genotype , >twofold-enrichment , p<0 . 05 ) ( Supplementary file 1A ) . Many of these gene products are associated with differentiated cells and their expression increased in the telencephalon between E13 . 5 and E18 . 5 ( as IMP1 expression declined ) ( Figure 8—figure supplement 1A ) and in cultured neurospheres as differentiation progressed ( Figure 8—figure supplement 1B ) . We could not detect a significant change in the expression of these genes by qPCR in Imp1-deficient neurospheres ( Figure 8A ) , but did observe increased expression of several proteins against which effective antibodies were available ( Figure 8B ) . We also detected a significant shift of these transcripts to the polysomal fraction in Imp1-deficient neurospheres ( Figure 8C , Figure 8—figure supplement 1C ) . These data suggest that IMP1 acts post-transcriptionally to negatively regulate the expression of some proteins associated with neural differentiation . 10 . 7554/eLife . 00924 . 020Figure 8 . Imp1 acts post-transcriptionally and cell autonomously to negatively regulate the expression of gene products associated with differentiation and to promote the expression of self-renewal genes , including Hmga2 . ( A–B ) Imp1 deficiency did not affect the levels of synaptotagmin1 ( Syt1 ) , Leucine rich repeat transmembrane neuronal 2 ( Lrrtm2 ) , Unc5d , or Oligodendrocyte myelin glycoprotein ( Omg ) transcripts by qPCR in E13 . 5 dorsal telencephalon-derived neurospheres; however , Imp1 deficiency did increase the levels of SYT1 , LRRTM2 , UNC5d , and OMG proteins . ( C ) Neurosphere lysates from E13 . 5 wild-type ( +/+ ) or Imp1β-geo/β-geo ( β/β ) dorsal telencephalon were fractionated in a 10–50% sucrose gradient ( Figure 8—figure supplement 8 ) and the transcripts in the polysome fraction were assessed by qPCR . Synaptotagmin1 ( Syt1 ) , Leucine rich repeat transmembrane neuronal 2 ( Lrrtm2 ) , and Oligodendrocyte myelin glycoprotein ( Omg ) transcripts were significantly enriched in the polysome fractions in Imp1 deficient cells relative to control cells ( p<0 . 05; error bars represent SD from three independent experiments ) . ( D ) Hmga2 , but not Hmga1 , transcript levels were significantly reduced in Imp1β-geo/β-geo dorsomedial ( green bar ) and dorsolateral telencephalon ( purple bar ) ( p<0 . 01 , p<0 . 05; error bars represent SD , 3–4 brains/genotype ) . ( E ) Imp1 deficiency reduced HMGA2 levels in dorsomedial , and dorsolateral telencephalon . ( F ) Imp1 deficiency reduced Hmga2 transcript levels ( by in situ hybridization; purple ) in dorsomedial and dorsolateral telencephalon from E13 . 5 mice . Higher magnification views of the boxed regions of dorsomedial ( upper ) or dorsolateral ( lower ) telencephalons are shown to the right of lower magnification images . Note that Hmga2 expression declined to a greater extent in the dorsomedial telencephalon ( arrow ) where Imp1 expression is highest and did not decline in the ventral telencephalon where Imp1 is not expressed at this stage ( see Figure 1D ) . Imp1 is thus required cell autonomously in the dorsal telencephalon to maintain Hmga2 expression . ( G ) Hmga2 , but not Hmga1 , transcript levels were significantly ( p<0 . 05 ) reduced after Actinomycin D treatment in Imp1β-geo/β-geo ( red ) relative to wild-type ( blue ) neurospheres cultured from E13 . 5 dorsal telencephalon . Error bars represent SD in three experiments . ( H–I ) Hmga2 deficiency did not affect Imp1 transcript levels in E13 . 5 telencephalon cells by qPCR ( H ) or in situ hybridization ( I ) ( three independent experiments ) . ( J ) Ink4a transcript levels were significantly elevated by Imp1 deficiency in neurospheres cultured from E18 . 5 wild-type or Imp1β-geo/β-geo dorsal telencephalon ( *p<0 . 05; error bars represent SD , 3–4 mice/genotype ) . ( K and L ) E18 . 5 wild-type ( +/+ ) or Imp1β-geo/β-geo ( β/β ) dorsal telencephalon cells were infected with GFP-only control retrovirus ( GFP ) or Hmga2-GFP retrovirus . Imp1 deficiency reduced HMGA2 expression and neural stem cell self-renewal but both were restored by Hmga2 over-expression . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 02010 . 7554/eLife . 00924 . 021Figure 8—figure supplement 1 . Multiple mRNAs bound by IMP1 increased their expression during brain development and neural differentiation . ( A ) A number of mRNAs bound by IMP1 encode gene products associated with differentiated neurons and glia . The levels of these mRNAs were compared by qPCR in dorsal telencephalons from E13 . 5 and E18 . 5 wild-type mice . Bars represent the ratio of transcript levels in E18 . 5/E13 . 5 telencephalon ( p<0 . 01 , *p<0 . 05; three mice/time point; error bars always represent SD ) . The expression levels of these mRNAs generally increased during development as would be expected for gene products expressed by differentiated cells . ( B ) The levels of these mRNAs were also compared by qPCR in neurospheres before and after differentiation . Neurospheres were cultured from E13 . 5 wild-type dorsal telencephalons . Bars represent the ratio of transcript levels in differentiated/undifferentiated neurospheres ( p<0 . 01 , **p<0 . 05; three independent experiments ) . ( C ) Cell lysates from E13 . 5 wild-type or Imp1β-geo/β-geo neurospheres were fractionated in a 10–50% sucrose gradient and the distribution of ribosomal proteins was assessed by immunoblotting against ribosomal P0 , P1 , and P2 antigens . RNAs were isolated from the input or polysomal fraction in the cycloheximide treated gradient ( +Chx , fractions #10–13 ) and subjected to qPCR in Figure 7C to quantitate the relative abundance of mRNAs in the polysomal fraction . In parallel , EDTA was included in the gradient to dissociate ribosomal complexes to assess the position of the polysomal fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 021 Two let-7 microRNA targets that promote neural stem cell self-renewal , Hmga2 and cyclin D2 ( Ccnd2 ) , were also significantly enriched in the fraction bound by IMP1 ( Supplementary file 1A ) . Ccnd2 expression was reduced in the dorsomedial telencephalon of Imp1 deficient mice ( Figure 6B ) . To assess whether IMP1 regulates Hmga2 expression we compared the levels of Hmga2 and it’s family member Hmga1 by qPCR in E13 . 5 and E18 . 5 dorsomedial and dorsolateral telencephalon cells from Imp1β-geo/β-geo mice and littermate controls . Hmga2 , but not Hmga1 , transcript levels were significantly ( p<0 . 05 ) reduced in Imp1β-geo/β-geo cells by qPCR ( Figure 8D ) . This reduction in Hmga2 expression was also confirmed at the protein level by western blot ( Figure 8E ) . In sections from Imp1β-geo/β-geo mice , the reduction in Hmga2 expression was confirmed by in situ hybridization in the dorsomedial and dorsolateral telencephalon ( where Imp1 is normally expressed ) but not in the ventral telencephalon ( where Imp1 is not normally expressed; compare Figure 8F to Figure 1—figure supplement 1B , E ) . Indeed , the region of the dorsal telencephalon in which Hmga2 expression declined ( Figure 8F ) corresponded precisely with the region of Imp1 expression ( Figure 1D ) . This demonstrates that IMP1 acts autonomously within stem cells in the dorsal telencephalon to maintain HMGA2 expression . IMP1 appears to promote HMGA2 expression by increasing the stability of its mRNA as we detected a significantly ( p<0 . 05 ) accelerated decay of mRNA for Hmga2 , but not Hmga1 , in neurospheres cultured from E13 . 5 Imp1β-geo/β-geo mice as compared to wild-type controls ( Figure 8G ) . HMGA2 did not appear to regulate Imp1 expression because Imp1 transcript levels were not affected by Hmga2 deficiency by either qPCR ( Figure 8H ) or in situ hybridization ( Figure 8I ) . We previously demonstrated that HMGA2 promotes neural stem cell self-renewal in the fetal telencephalon , partly by negatively regulating the expression of Ink4a , which encodes a cyclin-dependent kinase inhibitor ( Nishino et al . , 2008 ) . Consistent with this , we detected significantly ( p<0 . 05 ) increased Ink4a expression in neurospheres cultured from E18 . 5 Imp1β-geo/β-geo mice as compared to wild-type controls ( Figure 8J ) . Over-expression of Hmga2 restored normal levels of Hmga2 protein in E18 . 5 Imp1β-geo/β-geo neurospheres ( Figure 8K ) and significantly increased their self-renewal potential ( Figure 8L; p<0 . 05 ) . IMP1 therefore promotes the self-renewal of fetal neural stem cells partly by promoting the expression of HMGA2 . This suggests that part of the mechanism by which neural stem cells expand in the dorsal telencephalon in response to Wnt signaling is through IMP1-promoted HMGA2 expression , but that HMGA2 expression and neural stem cell expansion decline postnatally as a consequence of increased let-7 expression ( Nishino et al . , 2008 ) and a loss of IMP1 expression ( Figure 1A ) . IMP1 is one element of a network of heterochronic genes that regulates temporal changes in neural stem cell function throughout life ( see Figure 9 for a graphical summary ) . Expression of the let-7 target , Imp1 , by stem cells in the dorsal telencephalon ( Figure 1 , Figure 1—figure supplements 1B , E and 2 ) promoted the expansion of undifferentiated stem cells in response to Wnt signaling during fetal development . Wnt signaling promoted Imp1 expression in the dorsal telencephalon in a medial-lateral gradient ( Figure 7A–B , Figure 6—figure supplement 2A–D ) similar to the gradient in canonical Wnt signaling ( Machon et al . , 2007; Mutch et al . , 2010 ) . Imp1 deficiency reduced stem cell self-renewal potential ( Figure 2F–I ) and caused premature neuronal and glial differentiation ( Figure 4 ) , leading to stem cell depletion ( Figures 2G , 3A–C ) and reduced pallial expansion ( Figure 2A , D ) . Therefore , IMP1 expression by stem/progenitor cells during forebrain development regulates the timing of neuronal and glial differentiation . The postnatal loss of IMP1 expression may contribute to the decline in neural stem cell function during adulthood . 10 . 7554/eLife . 00924 . 022Figure 9 . Schematic showing a network of heterochronic genes that regulate temporal changes in CNS stem cell properties from fetal development throughout adulthood . A network of heterochronic genes changes with age , leading to temporal changes in stem cell properties from fetal development throughout adulthood . The promotion of expression/function is indicated as blue arrows and negative regulation is indicated as red bars . In fetal neural stem cells , Imp1 expression is promoted transcriptionally by canonical Wnt signaling and inhibited post-transcriptionally by let-7 . IMP1 expression is reduced by increasing let-7 expression in late fetal development and transcriptionally silenced postnatally . HMGA2 expression is high in early development but declines with time in response to declining IMP1 and increasing let-7 . Increasing let-7 expression contributes to the reduction in HMGA2 expression over time during adulthood . HMGA2 negatively regulates the expression of Ink4a/Arf . Declining expression of HMGA2 during adulthood allows the expression of these gene products to increase during aging . Overall , neural stem cell function declines over time . A decline in Wnt signaling during aging also contributes to these effects ( Seib et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00924 . 022 As let-7 expression increases during late fetal development ( Figure 1B ) , Imp1 expression declines ( Figure 1D ) . Increasing let-7 expression in neural stem cells reduces IMP1 expression ( Figures 1C and 7C ) , the number of neural stem cells , and their self-renewal potential ( Figure 7D–E , Figure 7—figure supplement 1A–D ) . Reducing the expression of mature let-7 microRNAs by inducing Lin28a expression increases IMP1 expression and neural stem cell self-renewal ( Figure 7F–I ) . The let-7 binding sites in the Imp1 3′UTR impede IMP1 expression in adult neural stem cells ( Figure 7J–K ) . Nonetheless , let-7 is not solely responsible for the perinatal loss of Imp1 expression as β-geo expression in Imp1β-geo/+ mice is also lost perinatally even though β-geo does not carry the Imp1 3′ UTR that contains the let-7 binding sites . Increasing expression of let-7 microRNAs during fetal development reduces IMP1 expression before other mechanisms silence Imp1 postnatally . IMP1 is not detectably expressed in the adult forebrain ( Figure 1A , Figure 1—figure supplement 1 ) and reduced let-7 microRNA expression does not restore IMP1 expression in adult neural stem cells ( Figure 7F ) . IMP1 is also dispensable for the regulation of self-renewal in adult neural stem cells ( Figure 2—figure supplement 2D–F ) . IMP1 is therefore required for the regulation of fetal but not adult neural stem/progenitor cells and contributes to the increased proliferation of undifferentiated cells in the fetal as compared to the adult forebrain . let-7 regulates the function of stem cells in the fetal and adult forebrains ( Zhao et al . , 2010 ) ; however , let-7 microRNAs appear to regulate different targets at different ages in neural stem cells . let-7 microRNAs , IMP1 , and HMGA2 are three important elements of a network of heterochronic genes that regulates temporal changes in stem cell function throughout life . Imp2 and Imp3 exhibit similar expression patterns as Imp1 ( Figure 1—figure supplement 1H , I ) . The overlap in expression among IMP family members suggests potential redundancy among family members in stem cells from the dorsal telencephalon . This raises the question of whether compound deletion of multiple family members would further accelerate stem cell depletion . IMP1 is known to increase the levels of some proteins by increasing the stability of their mRNAs ( Noubissi et al . , 2006 ) and to reduce the levels of other proteins by inhibiting translation ( Hansen et al . , 2004; Atlas et al . , 2007 ) . Consistent with this , IMP1 post-transcriptionally reduced the levels of some proteins associated with differentiation ( Figure 8A–C , Figure 8—figure supplement 1A–C ) while promoting the expression of the self-renewal factor , HMGA2 ( Figure 8D–F ) . IMP1 appeared to promote HMGA2 expression by increasing the half-life of Hmga2 mRNA ( Figure 8G ) , leading to higher levels of HMGA2 ( Figure 8E ) and lower levels of p16Ink4a ( Figure 8J ) . We showed previously that HMGA2 promotes , and p16Ink4a inhibits , stem cell self-renewal in the telencephalon ( Nishino et al . , 2008 ) . The postnatal lack of IMP1 expression in forebrain neural stem cells ( Figure 1D , F , Figure 1—figure supplement 1D , F , G ) is therefore likely to contribute to their reduced HMGA2 expression and self-renewal potential . Recently , two genome-wide studies in HEK293 cells suggested the existence of putative IMP1 recognition sequences: 5′-CAUH-3′ ( H = A , U , or C ) ( Hafner et al . , 2010 ) and 5′-CCYHHCC-3′ ( Y = C , U and H = A , C , U ) ( Jønson et al . , 2007 ) . We found these sequences at least once in all of our IMP1 pulled-downed mRNAs , but since these sequences would be expected to occur by chance in the mRNAs , no conclusion can be drawn regarding the functional relevance of these binding sites in the mRNAs we observed . Recently , Imp ( a fly ortholog of mammalian Imp1 ) was shown to non-cell-autonomously regulate the aging of germline stem cells in the fly testis ( Toledano et al . , 2012 ) . This raises the possibility that IMP1 might have non-cell-autonomous effects on mammalian stem cells , at least in certain contexts . Indeed , the overall growth retardation observed in Imp1-mutant mice ( Figure 2—figure supplement 1A ) could reflect non-cell-autonomous effects of Imp1 deficiency . Nonetheless , our data indicate that IMP1 cell-autonomously regulates neural stem cell function in the dorsal telencephalon . First , neural stem cell specific knockdown of Imp1 in a small percentage of cells in the developing telencephalon reduced cell proliferation and accelerated differentiation in a cell autonomous way ( Figure 5 ) . Second , all of our experiments demonstrating neural stem/progenitor cell phenotypes in Imp1 deficient ( Figures 2F–I and 4 ) , Lin28 transgenic ( Figure 7G–I ) or ilet-7 transgenic mice ( Figure 7—figure supplement 1A–D ) involved studies of individual isolated neural stem cells in culture . Third , in all of our experiments that involved viral infection ( Figure 2—figure supplement 2B , C , Figures 6E–F , 7K , 8L , Figure 7—figure supplement 1F–G ) , we compared the growth of infected neurospheres to non-infected neurospheres growing in the same cultures , or in control cultures . We never observed non-cell autonomous effects of infected neurospheres on non-infected neurospheres within the same cultures . Finally , we observed a correlation between Imp1 expression and function within the telencephalon: the phenotypes observed in Imp1 deficient mice were consistently more profound in the dorsomedial telencephalon , where Imp1 expression was high , than in the dorsolateral telencephalon , where Imp1 expression was low ( Figures 2E and 3 , Figure 4B ) . IMP1 therefore cell-autonomously regulates the function of neural stem/progenitor cells in the dorsal telencephalon , but likely has non-cell-autonomous effects in other contexts . We have thus demonstrated a novel function of the let-7 microRNA target , IMP1 , to promote the expansion of neural stem cells during cortical development . let-7b also negatively regulates the self-renewal of adult neural stem cells by reducing the expression of Hmga2 ( Nishino et al . , 2008 ) and TLX ( Zhao et al . , 2010 ) . This is consistent with our conclusion that a network of let-7 gene targets regulates stem cell function and that the regulation of different targets at different ages by let-7 contributes to temporal changes in stem cell properties . Other RNA binding proteins also regulate stem cell function . FBF proteins and GLD proteins control germline stem cell maintenance and the timing of meiosis in C . elegans ( Kimble and Crittenden , 2007 ) . The evolutionarily conserved Piwi family proteins that bind to Piwi-interacting RNAs are also required for the maintenance of germline stem cells ( Juliano et al . , 2011 ) . Musashi proteins appear to promote the self-renewal of fetal and adult stem cells from multiple tissues by translational repression of target RNAs ( Sakakibara et al . , 2002; Okano et al . , 2005; Kharas et al . , 2010 ) . Lin28 is preferentially expressed by embryonic cells and essential for the development of primordial germ cells ( West et al . , 2009; Viswanathan and Daley , 2010 ) . Indeed , the decline in Lin28 expression with time during fetal development may contribute to the increase in let-7 expression and the decline in IMP1 expression . A network of heterochronic genes including let-7 microRNAs , Imp1 , Hmga2 , and the cell cycle regulators cyclin D and p16Ink4a , regulates temporal changes in stem cell function . While we have demonstrated the functional importance of this network in neural stem cells , this network is also likely to regulate temporal changes in stem cells from other tissues . Beyond the network components we identified , many additional let-7 target genes are likely to regulate developmental changes in stem cells , integrating stem cell properties with changing tissue growth and regeneration demands throughout life . Imp1β-geo/+ ( Hansen et al . , 2004 ) ( MMRRC stock number 011720-UCD ) , APCflox/+ ( Shibata et al . , 1997 ) , β-catenin flox/+ ( Brault et al . , 2001 ) , human-GFAP Cre ( Malatesta et al . , 2003 ) , Lin28a transgenic ( Zhu et al . , 2010 ) , ilet-7 transgenic ( Zhu et al . , 2011 ) , Nestin-Cre ( Tronche et al . , 1999 ) , Hmga2+/− mice ( Zhou et al . , 1995 ) , and let-7b/c2 fl/+ mice were each backcrossed at least six times onto a C57BL/Ka background and housed at the University of Michigan Unit for Laboratory Animal Medicine or the University of Texas Southwestern Medical Center Animal Resource Center . Mice were genotyped by PCR . For let-7 or Lin28 induction , mice were administered water containing 2 μg/ml doxycycline ( Research Products International Co . , Mount Prospect , IL ) for 4 to 5 days . CNS progenitors were isolated as described in prior studies ( Molofsky et al . , 2005; Nishino et al . , 2008 ) . For adherent cultures , CNS progenitors were plated at a clonal density of 0 . 66 cells/μl ( 1000 cells per 35 mm well ) in 6-well plates ( Corning , Tewksbury , MA ) that had been sequentially coated with 150 μg/ml poly-d-lysine ( Biomedical Technologies , Stoughton , MA ) and 20 μg/ml laminin ( Sigma , St . Louis , MI ) . For the non-adherent culture of neurospheres , CNS progenitors were plated at a density of 1 . 33–2 . 67 cells/μl ( 2000–4000 cells per 35 mm well ) in ultra-low binding 6-well plates ( Corning ) . Cells were initially cultured for 7 to 9 days in ‘self-renewal medium’ to promote the formation of undifferentiated colonies . This medium was a 5:3 mixture of DMEM-low:neurobasal medium , supplemented with 20 ng/ml recombinant human bFGF ( R&D Systems , Minneapolis , MN ) , 20 ng/ml epidermal growth factor ( EGF ) ( R&D Systems ) , 10% chick embryo extract ( CEE; made as described [Stemple and Anderson , 1992] ) , 1% N2 supplement ( GIBCO , Grand Island , NY ) , 2% B27 supplement ( GIBCO ) , 50 mM 2-mercaptoethanol , and penicillin/streptomycin ( BioWhittaker , Walkersville , MD ) . After 7–9 days in self-renewal medium , cultures were fed with ‘differentiation medium’ and allowed to grow for an additional 4 to 6 days . Differentiation medium contained 10 ng/ml of bFGF ( instead of 20 ng/ml ) , 5% fetal bovine serum ( GIBCO ) , no EGF , and no CEE . After being grown in self-renewal medium , neurospheres were transferred to adherent cultures containing differentiation medium before being stained to assess multilineage differentiation . When Lin28a Tg neurospheres were cultured , 2 µg/ml doxycycline were added to the culture medium to sustain Lin28 transgene expression . All cultures were maintained at 37°C in 6% CO2 incubators . To quantify self-renewal potential , individual CNS neurospheres were dissociated by trituration then replated at clonal density ( 1 . 33 cells/μl ) in nonadherent secondary cultures . Secondary neurospheres were counted 7 to 9 days later , then transferred to adherent cultures containing differentiation medium to measure the percentage of secondary neurospheres that could undergo multilineage differentiation . For viral infection ( sometimes lentivirus and sometimes retrovirus , depending on the experiment ) , CNS progenitors were plated at a high density of 10 cells/μl and cultured adherently in self-renewal medium . After 48 hr , viral supernatant was added for 24 hr then switched to fresh self-renewal medium for a further 24 hr . Cells were harvested by incubating for 1 . 5 min at 37°C in trypsin/EDTA and transferred to nonadherent cultures to form neurospheres for 7 days . To measure mRNA decay , neurospheres formed by cells dissected from E13 . 5 dorsal telencephalon were plated adherently and cultured for 2 days in self-renewal medium . Then fresh self-renewal medium with 10 μg/ml Actinomycin D was added and the cells were harvested at the indicated time points to examine mRNA levels by quantitative RT-PCR . For explant cultures , pieces of E12 . 5 wild-type dorsolateral telencephalon were placed onto transwell plates ( 6 . 5 mm with 8 . 0 µm Pore Polycarbonate Membrane Insert , Corning ) and cultured for 12 hr with ‘explant culture medium’ ( a 5:3 mixture of DMEM-low:neurobasal medium , 1% N2 supplement , 2% B27 supplement , 50 mM 2-mercaptoethanol , and penicillin/streptomycin ) supplemented with or without recombinant mouse Wnt-3a ( 100 ng/ml , R&D Systems ) , recombinant mouse Dkk-1 ( 200 ng/ml , R&D Systems ) , or recombinant mouse BMP-4 ( 100 ng/ml , R&D Systems ) . To generate let-7b flox/flox mice , bacterial artificial chromosome ( BAC ) clones containing the let-7b/c2 genomic locus were purchased ( Invitrogen , Grand Island , NY ) and a targeting vector was constructed using bacterial recombineering ( Copeland et al . , 2001; Liu et al . , 2003 ) . In this construct , the let-7b/c2 genomic locus was flanked by loxP elements ( see Figure 7—figure supplement 2 ) . Neomycin resistance , diphtheria toxin fragment A ( DT-A ) , and thymidine kinase cassettes were included for positive and negative selection . Bruce 4 . G9 ES cells were electropolated with the targeting construct , positively selected with G418 ( Gibco ) , and negatively selected with gancyclovir ( cytovene from Syntex ) . Correctly targeted ES cells were identified by Southern blot and their karyotypes were assessed . Three independent euploid ES cell clones were injected into blastcysts obtained from B6 ( Cg ) -Tyrc-2j/J mice ( Jackson Laboratory , Bar Harbor , ME ) . The resulting male ES cell/mouse chimeras were bred with B6 ( Cg ) -Tyrc-2j/J mice to obtain germline transmission . After germline transmission , the neo cassette was removed by crossing with B6 . Cg-Tg ( ACTFLPe ) 9205Dym/J mice ( Rodriguez et al . , 2000 ) . Conditional let-7b/c2 mutant mice ( Nestin-Cre; let-7b/c2flox/flox ) were obtained by breeding Nestin-Cre mice ( Tronche et al . , 1999 ) with let-7b/c2 flox/+ mice . CNS neurospheres formed by cells dissected from E13 . 5 Imp1β-geo/β-geo dorsal telencephalon were infected with pMIG-3XFLAG-Imp1 retrovirus . Immunoprecipitation of Imp1-FLAG fusion protein along with the mRNAs it bound was done using an RIP-Assay kit ( MBL , Woburn , MA ) according to the manufacturer’s instructions . Briefly , neurospheres were washed with ice cold-PBS and lysed in buffer containing proteinase inhibitor cocktail ( Sigma ) and RNAse inhibitor ( Roche , Madison , WI ) on ice for 15 min . After clearing cell debris by centrifugation , supernatants were incubated with Protein G-Agarose for 60 min to absorb non-specific binding . After eliminating Protein G-Agarose by centrifugation , small fractions of supernatants were saved as input , and the rest was incubated for 3 hr with anti-FLAG M2 Affinity Gel ( Sigma ) . Immunoprecipitated fractions were washed with washing buffer and RNAs were eluted with solution Ⅰ-ⅠⅤ , precipitated with ice-cold 2-propanol , washed with ice-cold ethanol , and suspend in RNAse free water according to the RIP-Assay kit manufacturer’s instructions ( MBL ) . cDNA synthesis and library construction were performed using Ovation RNA-Seq System V2 and Encore NGS Library System Ⅰ ( Part Numbers 7102 and 300; NuGEN Technologies , Inc . , San Carlos , CA ) following the manufacturer’s instructions . We used 100 ng total RNA from the FLAG immunoprecipitated fraction or input fraction from three independent experiments . Briefly , first strand cDNAs were synthesized by incubating RNAs with first strand reagents in the Ovation RNA-Seq System V2 kit by incubating at 65°C for 5 min , 4°C for 1 min , 25°C for 10 min , 42°C for 10 min , 70°C for 15 min . Second strand cDNAs were generated by incubating with second strand reagents in the Ovation RNA-Seq System V2 kit at 4°C for 1 min , 25°C for 10 min , 50°C for 30 min , 80°C for 20 min . Primers and nucleotides were eliminated from purified cDNAs by mixing with Agencourt RNAclean XP purification beads ( Beckman coulter ) followed by magnetic separation on a 96-well magnetic plate ( ALPAQUA ) for 5 min cDNA was amplified by incubating with SPIA reagents in the Ovation RNA-Seq System V2 kit at 4°C for 1 min , 47°C for 60 min , and 80°C for 20 min . Amplified cDNAs were quantified with Quant-iT PicoGreen dsDNA reagent and kits ( Molecular probes Inc . , Eugene , OR ) , and sheared to 150–200 bp fragments using a Covaris S220 ultra-sonicator ( Covaris , Woburn , MA ) at 10% duty cycle , intensity set at 5 , 100 cycles/burst for 5 min . Sheared cDNAs were then endo-repaired by incubating with endo-repair reagents in the Encore NGS Library System Ⅰ kit at 25°C for 30 min , 70°C for 10 min . Primers and nucleotides were eliminated from cDNAs by mixing with Agencourt RNAclean XP purification beads ( Beckman coulter , Brea , CA ) followed by magnetic separation on a 96-well magnetic plate ( ALPAQUA , Beverly , MA ) for 5 min . Next adaptors were ligated by incubating with ligation adaptor reagents in the Encore NGS Library System Ⅰ kit at 25°C for 10 min , then cDNAs were amplified by incubating with amplification reagents in the same kit at 72°C for 2 min , followed five cycles at 94°C for 30 s , 55°C for 30 s , 72°C for 1 min , 10 cycles at 94°C for 30 s , 63°C for 30 s , 72°C for 1 min , and 72°C for 5 min . cDNA libraries were sequenced to 50-fold coverage using a Hiseq 2000 Sequencing System ( Illumina , Inc . , San Diego , CA ) . The raw sequence data were assessed for quality using FASTQC software ( http://www . bioinformatics . bbsrc . ac . uk/projects/fastqc/ ) and aligned to the mouse reference genome MM9 ( build 37 ) using TopHat ( Langmead et al . , 2009; Trapnell et al . , 2009 ) . Differences in RNA abundance were assessed using the CuffDiff algorithm in the CuffLinks software ( Roberts et al . , 2009; Trapnell et al . , 2010 ) . Parameter settings were: fragment-bias-correct ( mm9 . fa ) , compatible-hits-norm , multi-read-correct , and upper-quartile-norm . Using a locally derived perl script , we selected genes as being differentially expressed if they showed test status ‘OK’ , FDR <= 0 . 05 , and fold-change of at least 2 . 0 . E12 . 5-E18 . 5 dorsal telencephalons were removed and dissociated by incubating for 4 min at 37°C in 0 . 5 ml/ml DNAse1 ( Sigma ) in Ca , Mg-free HBSS , and resuspended in staining medium: L15 medium ( Gibco ) containing 1 mg/ml BSA ( Sigma A-3912 , St . Louis , MO ) , 10 mM HEPES ( pH7 . 4 ) and penicillin/streptomycin ( BioWhittaker , Walkersville , MD ) . After centrifuging ( 200×g for 4 min ) , the cells were gently triturated , filtered through nylon screen ( 45 nm , Sefar America , Kansas City , MO ) , counted by hemocytometer , and plated . E12 . 5 or E14 . 5 timed-pregnant female C57BL/6 mice were anesthetized with isofluorane and their uteri were exposed . 0 . 5–1 μl of lentiviral supernatant , including 10 μg/ml Polybrene ( Millipore ) and 0 . 05% fast green ( Sigma F-7258 , St . Louis , MO ) , were delivered into the lateral ventricle of telencephalons of each embryos using glass capillaries . After injection , uteri were placed back into the abdomen and the wounds were closed with surgical sutures ( Tevdek Ⅱ 3-0 , DEKNATEL , Gurnee , IL ) . Three days later , embryos were fixed and sectioned for immunostaining . CNS neurospheres were tested for multipotency by replating one neurosphere per well of 48-well plates coated with poly-d-lysine and laminin . The adherent neurospheres were allowed to differentiate for 4 to 6 days , then incubated first in anti-O4 antibody ( 1:800 ascites , Developmental Study Hybridoma Bank , Iowa city , IA ) , and then fixed in acid ethanol ( 5% glacial acetic acid in 100% ethanol ) for 20 min at −20°C . After blocking and washing , the cultures were stained in donkey anti-mouse-IgM secondary antibody conjugated to horse radish peroxidase ( Jackson Immunoresearch , West Grove , PA ) , followed by Nickel diaminobenzidine staining . Then cultures were stained with Tuj1 ( 1:500 Covance , Princeton NJ ) and anti-GFAP ( 1:200 , Sigma G-3893 ) primary antibodies followed by Alexa-Fluor 488 or 555 conjugated goat anti-mouse IgG1 and goat anti-mouse IgG2a secondary antibodies ( 1:1000 each , Molecular Probes Inc . , Eugene , OR ) . For analyses of cell proliferation in culture , cells were pulsed with 10 μM BrdU ( Sigma ) for 20 min , fixed in 70% ethanol for 30 min at −20°C , and stained with an anti-BrdU antibody ( 1:200 Caltag , Burlingame , CA ) overnight at 4°C . Alexa-Fluor 488 conjugated goat anti-mouse IgG2a secondary antibody ( Molecular Probes; 1:1000 ) was then stained for 1 hr at room temperature . For caspase-3 staining , cultures were fixed for 10 min at room temperature in 4% paraformaldehyde , blocked , then stained with an anti-activated caspase 3 antibody ( 1:1000 , Pharmingen , San Diego , CA ) overnight at 4°C . Alexa-Fluor 555 conjugated goat anti-rabbit IgG secondary antibody ( Molecular Probes; 1:1000 ) was then stained for 1 hr at room temperature . In all cases , cell were counter stained for 10 min at room temperature with 10 μg/ml DAPI ( Sigma D-8417 ) . For X-gal staining of neurospheres , E12 . 5 or E18 . 5 CNS neurospheres were fixed with 1% paraformaldehyde plus 0 . 2% glutaraldehyde for 5 min at 4°C , and incubated for 1 hr at 37°C in staining solution: PBS containing 2 mM 5-bromo-4-chloro-3-indolyl-beta-D-galactosidase ( X-dgal; Molecular Probes , Eugene OR , USA ) , 2 mM MgCl2 , 5 mM potassium ferrocyanide , 5 mM potassium ferricyanide , and 0 . 02% NP-40 . In some cases , neurospheres were fixed with 4% paraformaldehyde for 10 min at 4°C , cryoprotected in 30% sucrose , embedded in OCT compound ( Sakura Fineteck Inc . , Torrance , CA ) and frozen . Then 10 μm sections were cut and stained with chick anti-beta-galactosidase antibody ( 1:2000 , BGL-1040 , Aves Labs Inc . , Tigard , OR ) and anti-nestin antibody ( 1:400 , MAB353 , Millipore , Billerica , CA ) . Alexa-Fluor 488 conjugated goat anti-mouse IgG2a secondary antibody ( Molecular Probes; 1:1000 ) was then stained for 1 hr at room temperature . Brains were fixed in 4% paraformaldehyde at 4°C overnight , cryoprotected in 30% sucrose , embedded in OCT compound , and frozen . 12 μm sections were cut , then pre-blocked for at least 1 hr at room temperature in blocking solution ( PBS containing 5% goat serum , 0 . 2% bovine serum albumin , and 0 . 5% Triton X-100 ) , incubated with primary antibody at 4°C overnight , followed by washing , and incubation in secondary antibody for 1 hr at room temperature . For some antigens ( Ki67 , BrdU ) , sections were boiled before blocking in 10 mM sodium citrate ( pH 6 . 0 ) for 10 min to retrieve antigens . Sections were counter stained in 2 . 5 μg/ml DAPI for 10 min at room temperature , then mounted using ProLong antifade solution ( Molecular Probes Inc . , Eugene , OR ) . Primary antibodies included those against beta-galactosidase ( 1:2000 ) , Tuj1 ( 1:1000 ) , phospho-Histone H3 ( Cell Signaling Technology Inc . , Danvers , MA , 1:200 ) , Pax6 ( 1:1000 , Millipore , Billerica , MA ) , Tbr2 ( 1:200 , Abcam , Cambridge , MA ) , GFAP ( 1:100 , 0 , DAKO , Carpinteria , CA ) , Cyclin D1 ( Thermo Scientific , Fremont , CA , 1:200 ) , BrdU ( 1:200 , Accurate Chemical , Westbury , NY ) , Ki67 ( 1:200 , clone B56 , BD Biosciences , San Jose , CA ) , and TAG-1 ( 1:400 ascites , Developmental Study Hybridoma Bank , University of Iowa , Iowa city , IA ) . For secondary antibodies , Alexa-Fluor 488 or 555 or 647 conjugated antibodies were used ( 1:1000 each , Molecular Probes Inc . , Eugene , OR ) . TUNEL staining was performed using the Apoptag fluorescein In Situ Apoptosis Detection kit ( Millipore ) . For cell cycle exit analysis , E13 . 5 pregnant dams were injected intraperitoneally with 50 mg BrdU/kg body mass . 24 hr later , E14 . 5 pups were dissected and brains were fixed and processed as described above . For X-gal staining , E10 . 5-E12 . 5 mouse embryos or E14 . 5-P0 brains were fixed with 1% paraformaldehyde plus 0 . 2% glutaraldehyde for 15 min at 4°C . Then whole brains or cryosections were incubated in staining solution at 37°C for 4 to 16 hr as described above . For in situ hybridization to Imp1 and Hmga2 transcripts in tissues , brains were fixed in 4% paraformaldehyde at 4°C overnight , cryoprotected in 30% sucrose , embedded in OCT , and frozen . 12 μm sections were cut , pretreated with 2 μg/ml Proteinase K at 37°C for 20 min , with 0 . 2N HCl for 10 min at room temperature , with 0 . 1M triethanolamine-HCl for 10 min at room temperature and Digoxigenin-labeled antisense probe at 55°C overnight . The next day , sections were washed with 2 × SSC for 30 min at 55°C , with 0 . 2 × SSC for 40 min at 55°C , blocked with 20% goat serum for 1 hr , and incubated with anti-Digoxigenin-labeled-AP ( Alkaline phosphatase ) Fab fragment ( 1:2000 , Roche ) for 60 min at room temperature . Sections were washed with Tris buffered saline ( pH 9 . 5 ) with 0 . 1% Tween-20 for 30 min at room temperature , and incubated with 0 . 5 μl/ml NBT ( nitro-blue tetrazolium chloride ) plus 3 . 5 μl/ml BCIP ( 5-Bromo-4-Chloro-3′-Indolylphosphatase p-Toluidine salt ) ( Roche ) . Cells or tissues were resuspended in ice-cold cell lysis buffer ( Cell Signaling Technology , Danvers , MA ) with protease inhibitor cocktail ( Sigma ) , and incubated for 20 min on ice . SDS PAGE was done in 4–20% Tris-Glycine Gels ( Invitrogen ) and transferred to PDVF membranes ( Millipore ) . The membranes were blocked in Tris buffered saline with 0 . 05% Tween-20 and 5% milk powder , incubated with primary and secondary antibodies , and washed following standard procedures . Horse radish peroxidase conjugated secondary antibodies were detected by Supersignal West Femto Chemiluminescent Sustrate ( Pierce ) . Primary antibodies were rabbit anti-IGF2BP/IMP1 ( MBL , 1:2000 ) , mouse anti-β-Catenin ( 1:2500 , BD Biosciences ) , mouse anti-Synaptotagmin 1 ( Abcam , 1: 2000 ) , rabbit anti-LRRTM2 ( 1:2000 ) , mouse anti-UNC5D ( Abcam , Cambridge , MA , 1:1000 ) , rat anti-OMgp ( 1:2500 , R&D Systems ) , rabbit anti-HMGA2 ( 1:2000 , a generous gift from M Narita and S Lowe ) , mouse anti-FLAG ( Sigma , M2 1:5000 ) , human anti-ribosomal P antigen ( Immunovision , Springdale , AR , 1: 20000 ) , and mouse anti-α-tubulin ( 1:10000 , Sigma ) . Quantitative RT-PCR was performed as described previously ( Nishino et al . , 2008 ) . Primers used for amplification are listed in Supplementary file 1B . For let-7b , small RNAs ( <200 nt ) were extracted with mirVana miRNA isolation kit ( Ambion , Grand Island , NY ) , and RT-PCR was performed with specific primers and probes supplied in Taqman MicroRNA Assay kits ( Applied Biosystems , Grand Island , NY ) . The Imp1-GFP vector was constructed by subcloning mouse Imp1 cDNA ( corresponding to NCBI NM_009951 from 312 to 7455 , including the Imp1 ORF and let-7 binding sites in the 3′-UTR but lacking the polyadenylation signal ) with N-terminal 3XFLAG into Bgl2-XhoI sites of the retroviral vector pMIG ( MSCV-IRES-GFP ) . For Imp1 ( 3′-UTR del ) -GFP vector construction , the Imp1 ORF ( NCBI NM_009951 from 312 to 2045 ) was used instead . For Imp1-β-geo fusion protein+GFP vector construction , the 5′-fragment of Imp1 ORF ( NCBI NM_009951 from 312 to 547 ) and β-geo ORF were subcloned with N-terminal 3XFLAG into Bgl2-XhoI sites of retroviral vector pMIG . The CyclinD1-GFP vector was constructed by subcloning mouse cyclin D1 ORF ( NCBI NM_007631 from 233 to 1120 ) into the EcoRl site of the retroviral vector pMIG . Constructs for let-7b or Hmga2 were described previously ( Nishino et al . , 2008 ) . Viral supernatants were prepared by co-transfecting proviral plasmids and packaging vectors ( pCL-Eco and pC1-VSVG ) into 293T cells by standard calcium phosphate precipitation methods . The supernatant was collected after 72 hr and incubated with Retro-X Concentrator ( Clontech , Mountain View , CA ) at 4°C overnight . The next day , a viral pellet was obtained by centrifugation at 1 , 500×g in a Beckman JS 5 . 3 rotor for 45 min at 4°C and resuspended in a 5:3 mixture of DMEM-low:neurobasal medium for addition to culture medium . CNS neurospheres were formed by non-adherently culturing E12 . 5 wild-type dorsal telencephalon cells . The neurospheres were then plated adherently on 100 mm dishes coated with poly-d-lysine and laminin , and cultured in self-renewal medium ( see above for composition ) for additional 2 days . ChIP was done using the EZ ChIP Chromatin Immunoprecipitation kit ( Upstate , Billeria , CA ) according to the manufacturer’s instructions . Briefly , cells were fixed with 1% paraformamide for 10 min at room temperature , washed with ice cold PBS , and lysed in SDS lysis buffer supplemented with protease inhibitor cocktail . Then DNA was sheared using a sonicator ( Virsonic 100 , VirTis Inc . , Warminster , PA ) , and incubated with Protein G-Agarose for 60 min at 4°C to absorb non-specific binding . After eliminating Protein G-Agarose by centrifugation , small fractions of supernatants were saved as input , and the rest were incubated overnight at 4°C with either monoclonal anti-TCF4 antibody ( clone 6H5-3 , Millipore ) or normal mouse IgG ( supplied in EZ ChIP kit ) . The next day , immunoprecipitated fractions were collected by incubating with Protein G-Agarose for 60 min at 4°C , washed with washing buffer , eluted with elution buffer , and reverse-crosslinked in 0 . 2 M NaCl at 65°C overnight . Then DNA was purified using an affinity column , and subjected to PCR . PCR primers are listed in Supplementary file 1B . An Imp1 genomic DNA fragment ( corresponding to −1980 to +1 base pairs in Figure 5—figure supplement 1 ) was subcloned into pBluesript SK ( + ) ( Stratagene , La Jolla , CA ) , then site A ( from −1906 to −1901 bp ) , site B ( from −487 to −481 bp ) , or both site A and site B were eliminated by PCR . Intact or mutated fragments were subcloned into pGL3-Vector ( Promega , Madison , WI ) to generate luciferase reporter plasmids . These reporter plasmids , TOPflash ( Millipore ) , or empty pGL3-Vector were mixed with pRL-TK vector ( Promega ) at a 10:1 ratio , and transfected to P19 embryonic carcinoma cells ( ATCC ) using lipofectamine 2000 ( Invirogen ) . After transfection for 36 hr , cells were exposed to medium supplemented with 20 mM LiCl and cultured for an additional 12 hr . A dual luciferase assay was conducted using Dual-Luciferase Reporter Assay System ( Promega ) according to the manufacturer’s instructions . Briefly , cells were washed with PBS , and lysed in 1X passive lysis buffer at ambient temprature for 20 min . Cell lysates were mixed with Luciferase Assay Reagent Ⅱ and firefly luciferase activity was measured by microplate reader ( FLUOstar Omega , BMG LABTECH , Cary , NC ) . Next , 1X Stop & Glo Reagent was added and mixed , and Renilla luciferase activity was measured . CNS neurospheres were formed by non-adherently culturing E13 . 5 wild-type dorsal telencephalon cells then plated adherently on 100 mm dishes coated with poly-d-lysine and laminin , and cultured in self-renewal medium for an additional 2 days . Cells were treated with 0 . 1 μg/ml cycloheximide ( Sigma ) for 3 min , washed three times with PBS plus 0 . 1 μg/ml cycloheximide , and then lysed on ice for 15 min in polysome extraction solution ( 10 mM Tris [7 . 4] , 15 mM MgCl2 , 0 . 3M NaCl , 1% Triton X-100 , and 10 μg/ml heparin [Sigma] ) plus 0 . 1 μg/ml cycloheximide . After clearing cell debris by centrifugation , small aliquots of lysates were saved as input controls and the rest were loaded on top of a 10–50% sucrose gradient in polysome extraction buffer plus 0 . 1 μg/ml cycloheximide , and spun down at 4°C at 35 , 000 rpm for 190 min in an SW41 rotor ( Beckman Coulter ) . After ultracentrifugation , fractions were taken serially from the top ( 10% sucrose ) to the bottom ( 50% sucrose ) . Each fraction was divided into half and either saved in Trizol ( Invitrogen ) for RNA extraction or subjected to methanol/chloroform protein extraction . To assess the shift of ribosomal proteins , parallel cultures were processed in solutions that contain 30 mM EDTA instead of cycloheximide . Genotyping was performed by PCR following the manufacturer’s instructions using Go Taq Flexi DNA polymerase ( Promega ) for Imp1 , Lin28a , let-7 , Hmga2 , and let-7b/c2 mutant mice and Choice-Taq DNA Polymerase ( Denville Scientific Inc . Metuchen , NJ ) for Apc , β-catenin , and Nestin-Cre mice . The PCR conditions for Imp1 , Hmga2 , and let-7b/c2 mutant mice were 94°C for 2 min , then 33 cycles of 94°C for 30 s followed by 60°C for 30 s and 72°C for 1 min , with 72°C for 2 min at the end . For Lin28a and let-7 inducible transgenic mice PCR genotyping conditions were 94°C for 2 min , then 35 cycles of 94°C for 30 s followed by 55°C for 30 s and 72°C for 1 min , with 72°C for 2 min at the end . For Apc mutant mice , 94°C for 3 min , then 34 cycles of 94°C for 1 min followed by 55°C for 1 min and 72°C for 2 min , with 72°C for 10 min at the end . For β-catenin mutant mice , 94°C for 3 min , then 34 cycles of 94°C for 1 min followed by 59°C for 1 min and 72°C for 2 min , with 72°C for 10 min at the end . For Nestin-cre , 94°C for 3 min , then 34 cycles of 94°C for 1 min followed by 63°C for 1 min and 72°C for 2 min , with 72°C for 10 min at the end . Primers used for genotyping are listed in Supplementary file 1B .	Stem cells are found throughout the body , and play key roles in promoting tissue growth during fetal development , and in maintaining tissues in the adult . When stem cells divide , they can either give rise to more stem cells , or they can generate specialized cells required for tissue function . However , the properties of stem cells must change over time to match the changing growth and regeneration demands of tissues . A previous study by Nishino et al . has shown that expression of a micro RNA molecule called let-7 increases throughout adulthood , and this reduces the activity of stem cells in older animals . Now , Nishino et al . report that let-7 , and other genes it regulates , also control the dramatic changes that occur in the properties of stem cells between fetal development and adulthood . Whereas stem cells in the fetal forebrain undergo rapid division and are capable of generating many different cell types , stem cells in the adult forebrain divide less often and can generate only a few specific types of cell . While Nishino et al . performed their study on stem cells in the brain , their results are likely to apply also to stem cells in other tissues . Nishino et al . show that let-7 regulates the production of an RNA binding protein called IMP1 . Mice with stem cells that lack IMP1 have a smaller cerebral cortex than normal mice because their stem cells undergo fewer rounds of division before committing to become brain cells . Additional experiments revealed that IMP1 inhibits the expression of genes that trigger stem cells to commit to specific fates and promotes the expression of genes related to self-renewal . These results indicate that the gene that encodes IMP1 is expressed in fetal neural stem cells , but not in adult neural stem cells , and that the reduced production of this protein contributes to the developmental switch from highly proliferative neural stem cells in the fetus to the more quiescent stem cells found in adults . Further studies are likely to identify many more targets of let-7 that enable stem cells to adapt their properties to the changing needs of the organism over time . These results are interesting because let-7-regulated networks were first discovered based on their ability to regulate the timing of developmental transitions in worms . This suggests that the mechanisms employed by mammalian tissue stem cells to regulate changes in their properties over time , are at least partly evolutionarily conserved mechanisms inherited from invertebrates .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2013	A network of heterochronic genes including Imp1 regulates temporal changes in stem cell properties
Many developmental processes break left–right ( LR ) symmetry with a consistent handedness . LR asymmetry emerges early in development , and in many species the primary determinant of this asymmetry has been linked to the cytoskeleton . However , the nature of the underlying chirally asymmetric cytoskeletal processes has remained elusive . In this study , we combine thin-film active chiral fluid theory with experimental analysis of the C . elegans embryo to show that the actomyosin cortex generates active chiral torques to facilitate chiral symmetry breaking . Active torques drive chiral counter-rotating cortical flow in the zygote , depend on myosin activity , and can be altered through mild changes in Rho signaling . Notably , they also execute the chiral skew event at the 4-cell stage to establish the C . elegans LR body axis . Taken together , our results uncover a novel , large-scale physical activity of the actomyosin cytoskeleton that provides a fundamental mechanism for chiral morphogenesis in development . Most organisms are bilaterally asymmetric with morphologically distinct left and right hand sides . Bilateral asymmetry of organisms , organs , and tissues emerges early in development and is dependent on chiral symmetry breaking of cells and subcellular structures ( Hayashi and Murakami , 2001; Shibazaki et al . , 2004; Danilchik et al . , 2006; Xu et al . , 2007; Hejnol , 2010; Tamada et al . , 2010; Vandenberg and Levin , 2010; Savin et al . , 2011; Taniguchi et al . , 2011; Wan et al . , 2011; Huang et al . , 2012 ) . In many species the primary determinant of chirality has been linked to the cytoskeleton with both the microtubule ( Nonaka et al . , 1998; Ishida et al . , 2007 ) and the actomyosin cytoskeleton ( Danilchik et al . , 2006; Hozumi et al . , 2006; Spéder et al . , 2006 ) ( AD Bershadsky , personal communication , November 2013 ) playing prominent roles . Generally , how chiral molecules and chiral molecular interactions generate chiral morphologies on larger scales remains to be a fundamental problem ( Turing , 1952; Brown and Wolpert , 1990; Henley , 2012 ) . For example , it has been observed that myosin motors can rotate actin filaments in motility assays ( Sase et al . , 1997; Beausang et al . , 2008 ) . Yet , it remains unknown which types of large-scale mechanical activities arise from such types of chiral molecular interactions . In this study , we describe that the actomyosin cytoskeleton can generate active torques at cellular scales , and that the cell uses active torques to break chiral symmetry . We investigated chiral behaviours of the actomyosin cell cortex in the context of polarizing cortical flow in the 1-cell Caenorhabditis elegans embryo ( Munro et al . , 2004; Mayer et al . , 2010 ) . The cell cortex , sandwiched between the membrane and cytoplasm , is a thin actin gel containing myosin motors and actin binding proteins ( Pollard and Cooper , 1986; Clark et al . , 2013 ) . Given the chirality of cortical constituents , we first asked if cortical flow breaks chiral symmetry . We quantified the cortical flow velocity field v using particle image velocimetry in C . elegans zygotes containing GFP-tagged non-muscle myosin II ( NMY-2 ) ( Mayer et al . , 2010 ) . Flow proceeds primarily along the anteroposterior ( AP ) axis ( x-direction ) , however , we also observed flow vectors to have a small component in the direction orthogonal to the AP axis ( y-direction ) . Notably , the posterior and anterior halves of the cortex counter-rotate relative to each other ( Figure 1A , B , Figure 1—figure supplement 1 , Video 1 ) , with y-velocities of ∼–2 . 5 μm/min and ∼1 μm/min respectively ( Figure 1D ) . We define the chiral counter-rotation velocity vc as the difference between spatially averaged y-velocities in the posterior and the anterior region ( Figure 1B ) and measured vc at 858 time points during flow in 25 embryos . We find that the distribution of vc is shifted towards negative values , with a mean of −2 . 9 ± 0 . 3 μm/min ( mean ± error of mean at 99% confidence unless stated otherwise , Figure 1C ) . Thus , counter-rotating cortical flow breaks chiral symmetry at the 1-cell stage , with the posterior half undergoing a counterclockwise rotation when viewed from the posterior pole ( Figure 1A ) . Notably , chiral counter-rotating flow precedes the previously reported chiral whole-cell rotation of the zygote during cell division ( Schonegg et al . , 2014 ) . 10 . 7554/eLife . 04165 . 003Figure 1 . Chiral flow depends on myosin activity . ( A ) Sketch of a C . elegans embryo . Curved arrows illustrate chiral counter-rotating flow in the anterior ( A , red ) and posterior ( P , green ) half of the embryo , respectively . ( B ) Time-averaged cortical flow field ( arrows ) at the bottom surface of a representative C . elegans embryo viewed from the outside of the embryo in this and all other images . Arrow colors indicate y-velocity . Scale bar , 5 μm . Velocity scale arrow , 20 μm/min . ( C ) Histogram of instantaneous chiral counter-rotation velocity vc=〈vy〉P−〈vy〉A , where 〈vy〉A ( 〈vy〉P ) is the average of the y-component of the velocity v over the left ( right ) shaded area in ( B ) , for non-RNAi ( 858 frames from 25 embryos; gray ) and mlc-4 ( RNAi ) ( 8 hrs; 223 frames from 7 embryos; beige ) . Dashed vertical lines indicate mean vc . ( D ) y-velocity vy along the AP axis averaged over 18 vertical stripes as indicated , for non-RNAi ( black , averaged over 25 embryos ) and mlc-4 ( RNAi ) ( beige , averaged over 7 embryos ) . Error bars , SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 00310 . 7554/eLife . 04165 . 004Figure 1—figure supplement 1 . Similar flow fields were obtained when imaging actin as compared to imaging myosin Video 8 . Quantification of chiral flow for myosin and actin using a dual-colored transgenic line ( SWG003 ) . ( A ) Histogram of instantaneous chiral counter-rotation velocity vc=〈vy〉P−〈vy〉A , where 〈vy〉A ( 〈vy〉P ) is the average of the y-component of the velocity v over the left ( right ) shaded area in Figure 1B , for NMY-2::GFP ( 245 frames from 6 embryos; gray ) and Lifeact::tagRFP-T ( 245 frames from 6 embryos; beige ) . Dashed vertical lines indicate mean vc . ( B ) y-velocity vy along the AP axis averaged over 18 vertical stripes , as indicated in Figure 1B , for NMY-2::GFP ( black , averaged over 6 embryos ) and Lifeact::tagRFP-T ( beige , averaged over 6 embryos ) . Error bars , SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 00410 . 7554/eLife . 04165 . 005Video 1 . Cortical flow breaks chiral symmetry . Cortical flow during AP polarization of the C . elegans zygote exhibits chiral behaviors with the posterior and the anterior halves of the cortex counter-rotating relative to each other . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 005 Since AP flow depends on myosin activity ( Mayer et al . , 2010 ) , we asked if chiral flow does so as well . We tested if reducing myosin activity through RNAi of the myosin regulatory light-chain mlc-4 reduces the chiral counter-rotation velocity vc . We found that 8 hrs of mlc-4 ( RNAi ) not only reduces the AP flow velocity ( Munro et al . , 2004 ) but also significantly reduces vc ( Wilcoxon rank sum test at 99% confidence; mean: −1 . 1 ± 0 . 4 μm/min , Figure 1C , Video 2 ) when compared to non-RNAi embryos . We conclude that chiral flow depends on myosin activity . 10 . 7554/eLife . 04165 . 006Video 2 . Chiral flow depends on myosin activity . 8 hrs of mlc-4 ( RNAi ) leads to a substantial reduction of both AP and chiral flow . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 006 We next sought to understand how myosin activity can drive both AP and chiral flow . We pursue the idea that molecular-scale torque generation ( Sase et al . , 1997; Beausang et al . , 2008 ) leads to the emergence of active torques on larger scales and make use of a physical description of the cell cortex as a thin film of an active chiral fluid ( Fürthauer et al . , 2012 , 2013 ) . In our description , force and torque generation at the molecular scale give rise to both an active contractile tension T and an active torque density τ ( Figure 2A ) to drive cortical flow . Under conditions of azimuthal symmetry ( Figure 2—figure supplement 1 , appendix ) , the AP flow velocity ( vx ) and the y-velocity ( vy ) obey the equations of motion , ∂xT=η∂x2vx−γvx ( 1 ) ∂xτ=12η∂x2vy−γvy , where η is the 2D viscosity of the cortical layer and γ quantifies friction with membrane and cytoplasm ( Mayer et al . , 2010 ) . From the structure of Equation 1 , we see that gradients in active tension T along the AP axis drive AP flow , while gradients in active torque density τ along the AP axis drive chiral flow orthogonal to the AP axis ( Figure 2B , bottom sketch ) . We introduce the chirality index c = τ/T , which quantifies their relative magnitude . We assume that both active tension T and active torque density τ are proportional to the local myosin concentration , leading to a single value of the chirality index c that is constant over the embryo . This remains a useful approximation even for cases where T and τ exhibit more complex dependencies on myosin concentration or where they are independently regulated ( see below ) . In such cases the single value of c we determine corresponds to an average , capturing the overall chirality index of the embryo ( see appendix ) . Accordingly , we calculated the theoretical AP and chiral flow profiles from the experimentally determined myosin distribution and found a best match with the experimental profiles for a hydrodynamic length of ℓ=η/γ=16±0 . 6 μm ( Mayer et al . , 2010 ) and an overall chirality index of c = 0 . 58 ± 0 . 09 ( Figure 2B; see appendix ) . We conclude that a significant part of myosin activity is utilized for generating active torques . The handedness of active torques is clockwise when viewed from the outside of the embryo as indicated by the positive sign of the chirality index c . When considering the observed AP myosin gradients , active torques of this handedness give rise to counterclockwise flow in the posterior domain when viewed from the posterior tip , see Figure 2B for an illustration . 10 . 7554/eLife . 04165 . 007Figure 2 . The cortex actively generates torques . ( A ) Left , myosin heads consume ATP to pull ( Kron and Spudich , 1986 ) and twist ( Sase et al . , 1997; Beausang et al . , 2008 ) actin filaments , leading to the generation of a force dipole ( top , magenta ) and a torque dipole ( bottom , beige ) . Right , these can generate an active tension and an active torque density at larger scales , causing an isolated piece of cortex to contract ( top ) and rotate ( bottom ) . Gray surface , membrane; cube with wire frames , non-contracted ( non-rotated ) piece of cortex; magenta ( beige ) cubes , contracted ( rotated ) piece of cortex . The gray arrow points from the outside to the inside of the cell and the rotation is clockwise when viewed from the outside . ( B ) Top , myosin intensity ( blue markers ) and velocity profiles ( magenta markers , AP flow velocity vx; beige markers , y-velocity vy ) along the AP axis ( Figure 1B , D ) for the non-RNAi condition ( averaged over 25 embryos ) . Error bars , SEM . Magenta and beige curves , respective theoretical velocity profiles ( c = 0 . 58 ± 0 . 09 ) . Bottom , sketch of a C . elegans embryo with clockwise active torques in beige ( as viewed from the outside of the embryo ) . A gradient in myosin concentration along the AP axis ( see plot above ) leads to a gradient in active torques ( shown here with varying sizes of the clockwise torques ) , resulting in a chiral flow ( red and green arrows ) orthogonal to the gradient . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 00710 . 7554/eLife . 04165 . 008Figure 2—figure supplement 1 . Myosin distribution is azimuthally symmetric . ( A ) Left , representative image of the actomyosin cortex labeled with GFP-tagged NMY-2 ( gray , myosin ) at 30% cortical retraction . Scale bar , 5 μm . Right , histogram of the difference in spatially averaged myosin fluorescence intensity between the posterior and anterior halves of the embryo , quantified from the respective shaded regions for non-RNAi embryos ( N = 250 frames from 25 embryos; gray ) . Dashed line , mean of the difference in myosin intensity . Only the last 50 s of cortical flow was utilized from each video for generating this histogram . ( B ) Left , representative image of the actomyosin cortex labeled with GFP-tagged NMY-2 ( gray , myosin ) . Right , histogram of the difference in spatially averaged myosin fluorescence intensity between the top and bottom halves of the embryo in the posterior from the respective shaded regions for non-RNAi embryos ( N = 858 frames from 25 embryos; gray ) . Dashed line , mean of the difference in myosin intensity . The entire cortical flow period was utilized from each video for generating this histogram . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 008 We next sought to investigate if changing myosin activity affects the overall chirality index . To this end , we performed a series of mild-to-stronger ( Baggs et al . , 2009 ) mlc-4 ( RNAi ) experiments with feeding times of 4 , 6 , and 8 hrs , respectively , and determined c for each condition . We refer to this as weak perturbation RNAi experiments as we aim to identify principle phenotypical alterations upon a mild deviation from non-RNAi conditions , similar to determining the linear response to a small perturbation . While AP flow velocity vx and the chiral flow velocity vc were generally reduced at 4 and 6 hrs of mlc-4 ( RNAi ) ( Figure 3A , Figure 3—figure supplement 1–3 , Video 3 ) , c remained unchanged from non-RNAi conditions ( c , 0 . 61 ± 0 . 07 at 4 hrs and 0 . 52 ± 0 . 06 at 6 hrs of RNAi , compared to 0 . 58 ± 0 . 09 for non-RNAi; Figure 3A , Figure 3—figure supplement 3 ) . However , 8 hrs of mlc-4 ( RNAi ) not only resulted in a large reduction of both AP and chiral flow velocities but also led to a significant reduction of c ( 0 . 14 ± 0 . 04 , Figure 3A , Figure 3—figure supplement 3 ) . This indicates that the overall ratio of active torque density to active tension is not changed by weak reduction of mlc-4 activity but is altered at stronger RNAi conditions when cortical structure is affected ( Figure 3—figure supplement 4A and Video 2 ) . 10 . 7554/eLife . 04165 . 009Figure 3 . Ratio of active torque to active tension is modulated by Rho . ( A ) Chiral counter-rotation velocity vc ( top ) , AP velocity vx ( middle ) , and chirality index c ( bottom ) for non-RNAi ( gray ) , mlc-4 ( 4 , 6 , 8 hrs RNAi ) , ect-2 ( 4 , 6 , 8 hrs RNAi ) , and rga-3 ( 3 , 5 , 40 hrs RNAi ) . Error bars , error of the mean with 99% confidence . Yellow bars , significant difference to non-RNAi condition; brown bars , no significant difference . ( B ) Histogram of instantaneous chiral counter-rotation velocity vc for mlc-4 ( left; 6 hrs; 235 frames from 7 embryos ) , ect-2 ( middle; 6 hrs; 338 frames from 9 embryos ) , and rga-3 ( right; 5 hrs; 402 frames from 10 embryos ) RNAi . Gray histograms , non-RNAi condition . Dashed lines , mean vc . ( C ) Respective time-averaged cortical flow field ( arrows ) of representative embryos ( gray , myosin ) . Arrow colors indicate y-velocity vy . Scale bar , 5 μm . Velocity scale arrow , 20 μm/min . ( D ) Respective average myosin intensity ( blue markers ) and velocity profiles ( magenta markers , AP flow velocity vx; beige markers , y-velocity vy ) along the AP axis for each RNAi condition . Error bars , SEM . Magenta and beige curves , respective theoretical velocity profiles . Dashed lines , non-RNAi theoretical velocity profiles . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 00910 . 7554/eLife . 04165 . 010Figure 3—figure supplement 1 . Chiral counter-rotation velocity vc for mlc-4 , ect-2 , and rga-3 RNAi . Each graph presents the instantaneous chiral counter-rotation velocity vc histogram for the RNAi condition specified ( beige ) . The histogram from the non-RNAi condition is shown in gray . Downward arrows indicate a significant decrease and upward arrows indicate a significant increase in vc compared to the non-RNAi condition ( Wilcoxon rank sum test with 99% confidence ) . The number of hours of RNAi is as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 01010 . 7554/eLife . 04165 . 011Figure 3—figure supplement 2 . AP velocity vx for mlc-4 , ect-2 , and rga-3 RNAi . Each graph presents the instantaneous AP velocity vx histogram for the RNAi condition specified ( magenta ) . The histogram from the non-RNAi condition is shown in gray . Downward arrows indicate a significant decrease and upward arrows indicate a significant increase in vx compared to the non-RNAi condition ( Wilcoxon rank sum test with 99% confidence ) . The number of hours of RNAi is as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 01110 . 7554/eLife . 04165 . 012Figure 3—figure supplement 3 . Theoretical velocity profiles for mlc-4 , ect-2 , and rga-3 RNAi . Each graph presents the respective average myosin intensity ( blue markers ) and velocity profiles ( magenta markers , AP flow velocity vx; beige markers , y-velocity vy ) along the AP axis for each RNAi condition specified . Error bars , SEM . Magenta and beige curves , respective theoretical velocity profiles . Dashed lines , non-RNAi theoretical velocity profiles . The number of hours of RNAi is as indicated . See respective Videos 2–5 . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 01210 . 7554/eLife . 04165 . 013Figure 3—figure supplement 4 . Comparison of cortical structure through imaging GFP-tagged NMY-2 . ( A ) Representative images for non-RNAi , 4 , 6 , 8 , and 11 hrs of mlc-4 ( RNAi ) are shown . Cortical structure is disrupted by 8 hrs of mlc-4 ( RNAi ) ( magnified view with characteristic foci size obtained from spatial myosin fluorescence intensity–intensity correlation , compare to non-RNAi ) . ( B ) Representative images for non-RNAi , 4 , 6 , 8 , and 11 hrs of ect-2 ( RNAi ) . Cortical structure is disrupted by 8 hrs of ect-2 ( RNAi ) ( magnified view with characteristic foci size , compare to non-RNAi ) . ( C ) Representative images for non-RNAi , 3 , 5 , and 40 hrs of rga-3 ( RNAi ) . Cortical structure is disrupted by 5 hrs of rga-3 ( RNAi ) ( magnified view with characteristic foci size , compare to non-RNAi ) . Scale bars , 10 μm . Error bars , error of the mean with 99% confidence . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 01310 . 7554/eLife . 04165 . 014Video 3 . Chirality of the cortex is unaffected under weak perturbation of myosin activity . 4 and 6 hrs of mlc-4 ( RNAi ) leads to a proportional change of AP and chiral flow , with the chirality index remaining unchanged under these conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 014 We next asked whether there are conditions that modify active torque generation without affecting active tension . To this end , we tested if small changes in Rho signaling , which regulates myosin activity as well as actin dynamics ( Maekawa et al . , 1999 ) , have a different impact on AP and chiral flow and thus change c . We performed a series of mild-to-stronger RNAi of the Rho GEF ect-2 and the Rho GAP rga-3 . We found that weak perturbation RNAi of ect-2 led to a substantial decrease in chiral but not AP flow and thus a decrease in the overall chirality index c when compared to non-RNAi conditions ( Figure 3A , D; see also Figure 3—figure supplement 4B , Video 4 ) . Conversely , weak perturbation RNAi of rga-3 led to a substantial increase in chiral but not AP flow and thus an increase in the overall chirality index c ( Figure 3A , D; see also Figure 3—figure supplement 4C , Video 5 ) . Thus , a weak perturbation of ect-2 and rga-3 affects chiral but not AP flow , unlike a weak perturbation of mlc-4 which affects both . We conclude that a principle phenotypical alteration upon mild modifications of Rho pathway activity is a change of the chirality index , or , in other words , mild modifications of Rho pathway activity change active torque generation without affecting active tension . 10 . 7554/eLife . 04165 . 015Video 4 . Chiral flow decreases with decreasing Rho activity . ect-2 ( RNAi ) leads to a substantial reduction in chiral flow with a minimal change in AP flow . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 01510 . 7554/eLife . 04165 . 016Video 5 . Chiral flow increases with increasing Rho activity . rga-3 ( RNAi ) leads to a substantial increase in chiral flow with a minimal change in AP flow . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 016 We next asked whether actomyosin active torques participate in bilateral symmetry breaking , since this requires a chiral process . In C . elegans , embryonic handedness is determined at the 4-cell stage when the ABa and ABp cells skew clockwise by ∼20° ( as viewed dorsally in the AP–LR plane , Figure 4A ) ( Wood , 1991; Bergmann et al . , 2003 ) . We first tested if the clockwise skew in ABa is accompanied by chiral cortical flow . Strikingly , we observed chiral cortical flow in ABa , with the cortex in both future daughter cells counter-rotating ( vc = −5 . 2 ± 1 . 1 μm/min , mean ± error of mean at 95% confidence , Video 6 ) . The handedness of chiral flow is identical to that at the 1-cell stage , indicative of a presence of active torques with the same sign of c . If these chiral counter-rotating flows participate in the clockwise skew of both daughter cells , we would expect that changing active torque generation should affect the chiral skew at the 4-cell stage . To this end , we performed weak perturbation RNAi of the Rho pathway members , ect-2 and rga-3 , to specifically modify active torques . We first tested whether chiral flows are affected at the 4-cell stage under these conditions . We found that 4 . 5 hrs of ect-2 ( RNAi ) led to a significant decrease in chiral flow velocity , vc ( −3 . 4 ± 1 . 4 μm/min ) , while 4 . 5 hrs of rga-3 ( RNAi ) led to a significant increase in vc in the ABa cell ( −6 . 7 ± 0 . 7 μm/min , Figure 4B , Video 6 ) , similar to our observations at the 1-cell stage . We next tested whether changing chiral flow velocity at the 4-cell stage is concomitant with a change in the degree of clockwise skew . Indeed , we found that 4 . 5 hrs of ect-2 ( RNAi ) led to a significant decrease in skew ( 15 . 8° ± 4 . 9° ) in the ABa cell measured in the AP–LR plane , while 4 . 5 hrs of rga-3 ( RNAi ) led to a significant increase in skew ( 37 . 8° ± 6 . 1° ) when compared to non-RNAi conditions ( 23 . 6° ± 3 . 7°; Figure 4A and Figure 4—figure supplement 1 ) . Similar results were obtained in ABp ( Figure 4—figure supplement 1 ) . Thus , changing counter-rotating chiral flow velocity in these cells by weak perturbation of the Rho pathway leads to a change in the degree of skew . This suggests that active torque generation and chiral counter-rotating flow participate in the execution of the LR symmetry breaking chiral skew event at the 4-cell stage . 10 . 7554/eLife . 04165 . 017Figure 4 . Active torques participate in L/R body axis establishment . ( A ) A schematic of the skew angle measurement in the AP–LR plane . Gray dashed line , initial nuclei position; black dashed line , skewed nuclei position; beige arrows , direction of cortical flow on the dorsal surface ( Video 6 ) . To the right are the chiral skew angles of ABa for non-RNAi ( gray ) , ect-2 ( RNAi ) ( 4 . 5 hrs ) and rga-3 ( RNAi ) ( 4 . 5 hrs ) in the AP–LR plane . Gray circles , skew angle in individual videos; shaded areas , SEM; green horizontal lines , mean skew angle; red horizontal lines , median skew angle; yellow shaded areas , knockdown conditions with a significant difference ( 95% confidence with the Wilcoxon rank sum test ) from the non-RNAi condition . ( B ) Chiral counter-rotation velocity vc for non-RNAi ( gray ) , ect-2 ( RNAi ) ( 4 . 5 hrs ) and rga-3 ( RNAi ) ( 4 . 5 hrs ) quantified at the 4-cell stage during ABa cytokinesis . Note that one outlier was removed for computing mean vc for rga-3 ( RNAi ) . The expected flow profiles from our theoretical description , given a stripe of high myosin activity ( corresponding to the cleavage plane ) , is shown in Figure 4—figure supplement 5 . ( C ) Overall chirality index c , for non-RNAi ( gray ) and for Wnt signaling genes ( 40 hrs RNAi ) that impact the establishment of the L/R body axis . Interestingly , gsk-3 not only results in a reduced chiral counter-rotation velocity but also in an increased AP velocity ( Figure 4—figure supplements 2–4 ) . Error bars , error of the mean with 99% confidence . Yellow bars , significant difference to non-RNAi condition; brown bars , no significant difference . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 01710 . 7554/eLife . 04165 . 018Figure 4—figure supplement 1 . Chiral skew quantifications during bilateral symmetry breaking of the organism . ( A ) A schematic of the skew angle measurement in the AP–LR plane . Gray dashed line , initial nuclei position; black dashed line , skewed nuclei position; beige arrows , direction of cortical flow on the dorsal surface ( ABa counter-rotating flow shown in Video 6 ) . Below , skew angles of ABa and ABp for non-RNAi ( gray ) , ect-2 ( RNAi ) ( 4 . 5 hrs ) and rga-3 ( RNAi ) ( 4 . 5 hrs ) in the AP–LR plane . Gray circles , skew angle in individual videos; shaded areas , SEM; green horizontal lines , mean skew angle; red horizontal lines , median skew angle; yellow shaded areas , knockdown conditions with a significant difference ( 95% confidence with the Wilcoxon rank sum test ) from the non-RNAi condition . ( B ) Same as ( A ) with the skew determined in the DV–LR plane . ( C ) Skew calculated without projections on to a particular plane . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 01810 . 7554/eLife . 04165 . 019Figure 4—figure supplement 2 . Chiral counter-rotation velocity vc for RNAi of Wnt signaling genes . Each graph presents the instantaneous chiral counter-rotation velocity vc histogram for the RNAi condition specified ( beige ) . The histogram from the non-RNAi condition is shown in gray . Downward arrows indicate a significant decrease and upward arrows indicate a significant increase in vc compared to the non-RNAi condition ( Wilcoxon rank sum test with 99% confidence ) . The number of hours of RNAi is as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 01910 . 7554/eLife . 04165 . 020Figure 4—figure supplement 3 . AP velocity vx for RNAi of Wnt signaling genes . Each graph presents the instantaneous AP velocity vx histogram for the RNAi condition specified ( magenta ) . The histogram from the non-RNAi condition is shown in gray . Downward arrows indicate a significant decrease and upward arrows indicate a significant increase in vx compared to the non-RNAi condition ( Wilcoxon rank sum test with 99% confidence ) . The number of hours of RNAi is as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 02010 . 7554/eLife . 04165 . 021Figure 4—figure supplement 4 . Theoretical velocity profiles for RNAi of Wnt signaling genes . Each graph presents the respective average myosin intensity ( blue markers ) and velocity profiles ( magenta markers , AP flow velocity vx; beige markers , y-velocity vy ) along the AP axis for each RNAi condition specified . Error bars , SEM . Magenta and beige curves , respective theoretical velocity profiles . Dashed lines , non-RNAi theoretical velocity profiles . The number of hours of RNAi is as indicated . See Video 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 02110 . 7554/eLife . 04165 . 022Figure 4—figure supplement 5 . Theoretical velocity profiles for a stripe of high myosin activity . The graph presents theoretical axial velocity vx ( magenta ) and chiral velocity vy ( beige ) profiles given a stripe of high myosin activity ( blue ) . The gradient in myosin activity will then lead to axial flows along the gradient and directed towards the stripe as well as chiral flows orthogonal to the gradient leading to counter-rotations . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 02210 . 7554/eLife . 04165 . 023Video 6 . Chiral flow accompanies the LR symmetry breaking skew event at the 4-cell stage . Dorsal view of a representative 4-cell stage embryo , with ABa and ABp cells exhibiting counter-rotating cortical flow during cytokinesis , for 4 . 5 hrs of ect-2 ( RNAi ) , non-RNAi and 4 . 5 hrs of rga-3 ( RNAi ) . Anterior view of these Videos is shown at the bottom for visualizing counter-rotation in ABa cell . Flashing cyan arrows indicate the direction of counter-rotating cortical flow . Note that counter-rotation of AB cells is significantly reduced in ect-2 ( RNAi ) and significantly increased in rga-3 ( RNAi ) compared to the non-RNAi condition . Quantification of chiral flow velocities ( Figure 4B ) was performed in the ABa cell ( marked in red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 023 Finally , we tested whether genes that affect establishment of the L/R body axis impacts chiral flow . To investigate this , we quantified chiral flow velocities and the overall chirality index c at the 1-cell stage under conditions of RNAi of the Wnt signaling genes dsh-2 , gsk-3 , mig-5 , mom-2 , and mom-5 , which are known to regulate aspects of bilateral symmetry breaking ( Walston et al . , 2004; Pohl and Bao , 2010 ) . Strikingly , we found that all these conditions ( except mom-5 ) led to reduced chiral flow and a significant reduction of the overall chirality index c at the 1-cell stage ( Figure 4C , Figure 4—figure supplement 2–4 , Video 7 ) . These results are indicative of a fundamental link between genes that affect LR symmetry breaking and chiral counter-rotating flow . Since Wnt-induced signals in many systems propagate through Rho GTPases to promote morphological changes ( Schlessinger et al . , 2009 ) , we speculate that these effects are propagated through Rho signaling ( Figure 3 ) . Taken together , our results indicate that active torque generation and chiral counter-rotating flows participate in the establishment of the L/R body axis of C . elegans . 10 . 7554/eLife . 04165 . 024Video 7 . Wnt signaling genes regulate chiral flow . RNAi of Wnt signaling genes leads to a substantial reduction in chiral flow . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 02410 . 7554/eLife . 04165 . 025Video 8 . Chiral flow observed with an actin probe . Cortical flow visualized through Lifeact::tagRFP-T exhibits similar chiral behaviors . DOI: http://dx . doi . org/10 . 7554/eLife . 04165 . 025 To conclude , the actomyosin cytoskeleton in C . elegans generates active chiral torques with clockwise handedness when viewed from the outside of the cell . They drive a specific pattern of chiral flows which can be understood quantitatively based on the physics of active gels with chiral asymmetries ( Kruse et al . , 2005; Fürthauer et al . , 2012 , 2013 ) . Furthermore , our weak perturbation RNAi experiments indicate that Rho activity affects cortical chirality in a way that does not depend on its role in activating myosin . On the one hand , this raises an interesting question whether active chiral torques arise directly from chiral interactions between actin and myosin ( Figure 2A ) or whether they rather emerge through myosin molecular force generation and non-trivial tension–torque coupling ( Gore et al . , 2006; De La Cruz et al . , 2010 ) in the actomyosin network . On the other hand , through these weak perturbation RNAi experiments we have identified specific conditions under which cortical chirality and active torques can be selectively modified . Bilateral symmetry breaking requires a chiral process , and we used these specific conditions to demonstrate that in C . elegans , this chiral process could be provided by active chiral torque generation of the actomyosin cortical layer for driving the spindle skew at the 4-cell stage . We note that a plausible scenario for driving spindle skew by counter-rotating flows is similar to the rotation that a crawler excavator or a digger can execute on the spot . This is done by such a machine rotating its two chains in opposite directions . In our context , the chain rotations correspond to the counter-rotating flows and the rotation of the machine corresponds to the spindle rotation giving rise to the skew . Our results imply that active torques are generated at multiple stages during development , in the zygote during polarity establishment , without immediate consequences with respect to LR symmetry breaking , and again at the 4-cell stage , but here as an instructional and mechanistic event that helps to break left/right symmetry . Chiral morphogenetic rearrangements have been observed at other stages in C . elegans development ( Pohl and Bao , 2010 ) and during the first cleavage ( Schonegg et al . , 2014; Singh and Pohl , 2014 ) , as well as in other systems ( Shibazaki et al . , 2004; Danilchik et al . , 2006; Géminard et al . , 2014 ) . It is interesting to speculate that all these events might be driven by active torque generation in the actomyosin layer . As such , our work paves the way for a mechanistic understanding of chiral morphogenesis of cells , tissues , and organisms . The following transgenic lines were used in this study: TH455 ( unc-119 ( ed3 ) III; zuIs45[nmy-2::NMY-2::GFP + unc-119 ( + ) ] V; ddIs249[TH0566 ( pie1::Lifeact::mCherry:pie1 ) ] ) for imaging cortical flow , LP133 ( nmy-2 ( cp8[NMY-2::GFP + unc-119 ( + ) ] ) I; unc-119 ( ed3 ) III ) ( Dickinson et al . , 2013 ) for imaging counter-rotation of AB cells , and SWG003 ( nmy-2 ( cp8[NMY-2::GFP + unc-119 ( + ) ] ) I; unc-119 ( ed3 ) III; gesIs002[unc-119 ( ed3 ) III; ( pie-1::Lifeact::tagRFP-T::pie-1 + unc-119 ( + ) ) ] ) for quantifying chiral flow fields with an actin probe . For imaging the chiral skew event at the 4-cell stage , a mCherry::Histone; mCherry::PH-PLC1δ1 transgenic line was generated by crossing OD70 ( Kachur et al . , 2008 ) to a line expressing Moesin::GFP and mCherry::Histone obtained from the Piano lab ( New York University , New York , USA ) . C . elegans worms were cultured on OP50-seeeded NGM agar plates as described ( Brenner , 1974 ) . RNAi experiments were performed by feeding ( Timmons et al . , 2001 ) . Worms were placed on feeding plates ( NGM agar containing 1 mM isopropyl-β-D-thiogalactoside and 50 μg ml−1 ampicillin ) and incubated for the specified number of hours at 25°C . We defined feeding time ( number of hours of RNAi ) as the time between transfer of worms to the feeding plate and putative fertilization of the egg . Worms were dissected in M9 buffer and the embryos were mounted on 2% agarose pads for image acquisition . rga-3 feeding clone was obtained from Ahringer lab ( Gurdon institute , Cambridge , United Kingdom ) , ect-2 and mlc-4 from Hyman lab ( MPI-CBG , Dresden , Germany ) . Feeding clones dsh-2 , gsk-3 , mig-5 , mom-2 , and mom-5 were obtained from Source Bioscience ( Nottingham , United Kingdom ) . For performing weak perturbation RNAi experiments ( from 3–12 hrs of RNAi ) , L4 staged worms were first incubated overnight on OP50 plates at 25°C . Young adults were then transferred to respective RNAi feeding plates . For performing 40 hr RNAi experiments , early L4 staged worms were directly transferred to respective RNAi feeding plates and incubated at 25°C . All videos were acquired at 23–24°C , with a spinning disc confocal microscope using a Zeiss C-Apochromat 63X/1 . 2 NA objective lens and a Yokogawa CSU-X1 scan head . The following emission filter was used for all acquisitions unless specifically stated: 525/50 nm bandpass filter from Semrock ( Rochester , New York ) . Micromanager software ( Vale lab , UCSF ) was used to acquire videos using the Hamamatsu ORCA-flash camera . Confocal videos of cortical NMY-2::GFP for non-RNAi , mlc-4 , ect-2 , and rga-3 ( RNAi ) were acquired using an Andor iXon EMCCD camera ( 512 by 512 pixels ) . A stack consisting of three z-planes ( 0 . 5 μm apart ) with a 488 nm laser and an exposure of 150 ms was acquired at an interval of 5 s from the onset of cortical flow until the first cell division . The maximum intensity projection of the stack at each time point was then subjected for further analysis . Confocal videos of cortical NMY-2::GFP for dsh-2 , gsk-3 , mig-5 , mom-2 , and mom-5 ( RNAi ) were acquired using an Andor Neo sCMOS camera ( 2560 by 2160 pixels ) . A stack consisting of two z-planes ( 0 . 5 μm apart ) with a 488 nm laser and an exposure of 150 ms was acquired at an interval of 5 s from the onset of cortical flow until the first cell division . The maximum intensity projection of the stack at each time point was then subjected for further analysis . Chiral skew at the 4-cell stage was imaged by using mCherry::Histone; mCherry::PH-PLC1δ1 dual transgenic line . Confocal videos were acquired using a Hamamatsu ORCA-flash 4 . 0 camera ( 2048 by 2048 pixels ) . A stack consisting of 25–30 z-planes ( 1 μm apart ) with a 561 nm laser and an exposure of 300 ms ( emission filter – 641/75 nm bandpass filter from Semrock ) was acquired at an interval of 30 s from metaphase of the AB lineage at the 4-cell stage until telophase of the AB lineage at the 8-cell stage . Counter-rotation of the AB cells was imaged using the LP133 strain . Embryos at the 4-cell stage were first identified and an eye-lash tool was then used to rotate the embryo to obtain a dorsal view . Confocal videos from the dorsal side of the embryo were then acquired using a Hamamatsu ORCA-flash 4 . 0 camera ( 2048 by 2048 pixels ) . A stack consisting of 25 z-planes ( 0 . 5 μm apart ) with a 488 nm laser and an exposure of 100 ms was acquired at an interval of 5 s from the start of telophase of the AB lineage at the 4-cell stage until cytokinesis . 2D cortical flow velocity fields were obtained by performing Particle Image Velocimetry ( PIV ) ( Raffel et al . , 2007 ) using the freely available PIVlab MATLAB algorithm ( pivlab . blogspot . de ) . PIVlab was employed by performing a 3-step multi pass ( with linear window deformation ) , where the final interrogation area was 16 pixels with a step of 8 pixels . To obtain the flow profiles , 2D velocity fields were projected to the AP axis by dividing the embryo into 18 bins along the AP axis ( Figure 1B ) , and by spatially averaging the x-component or the y-component of velocity along each bin in a single frame . The average velocity in each bin was then averaged over time across the entire flow period ( from start of the flow till pseudocleavage ) . These time-averaged flow profiles were then averaged across all embryos for one experimental condition . Bin extent in the y direction was restricted to a stripe of about 13 μm ( Figure 1B ) . An accurate quantification of chiral flow fields is only possible when the flow axis is approximately aligned with the long axis of the embryo , and we removed from our analysis a small number of embryos ( 3/28 embryos for the non-RNAi condition ) which in the bottom plane analysis appeared to clearly polarize from the side . Chiral counter-rotation velocity vc was quantified in each frame by subtracting the y-component of velocity in the anterior ( spatially averaged across bins 3 to 6 , Figure 1B ) from the y-component of velocity in the posterior ( spatially averaged across bins 13 to 16 , Figure 1B ) . vc from each frame was computed across the entire flow period from all embryos of one experimental condition and a histogram was plotted . For quantification of chiral counter-rotation velocity vc in the ABa cell , the cleavage plane as viewed from the dorsal side of the embryo was first manually identified using FIJI . PIV was then performed and the component of velocity vectors parallel to the cleavage plane was calculated . vc was then computed by subtracting velocity components in the right daughter cell ( spatially averaged across a box of width 5 μm close to the cleavage plane ) from the velocity components in the left daughter cell ( spatially averaged across a box of size 5 μm close to the cleavage plane ) . vc from each frame was computed across the counter-rotation flow period and a time-averaged vc was reported . We removed from our analysis one video each from non-RNAi and ect-2 ( RNAi ) conditions and 2 videos from rga-3 ( RNAi ) condition that resulted in a marked whole body rotation along the DV-LR plane . Chiral skew analysis was performed by analyzing the multi-stack videos using Imaris v7 . 6 . 5 and v7 . 7 software ( Bitplane , Zurich , Switzerland ) . The anterior and posterior pole positions were first manually identified and used to define the AP-vector . The DV-vector was then obtained utilizing the nuclear position of EMS and the AP-vector . The initial and skewed nuclei vectors of ABa and ABp cells at the beginning and end of telophase , respectively , were determined by identifying the corresponding nuclei positions . We used MATLAB to determine the angle between the initial and skewed nuclei vectors of ABa and ABp . We first determined the skew angle by projecting all vectors on to the AP–LR plane ( dorsal view , Figure 4—figure supplement 1A ) . This is the plane in which the skew has been reported previously ( Wood , 1991; Bergmann et al . , 2003 ) . Next , we determined the skew angle in the DV–LR plane ( anterior view , Figure 4—figure supplement 1B ) by projecting all vectors on to the DV–LR plane . Finally , we also determined the full ( 3D ) angle between the respective initial and skewed vectors . As described in the main text ( Figure 4A ) , in the ABa cell , ect-2 ( RNAi ) ( 4 . 5 hrs ) led to a significantly reduced skew in the AP–LR plane , whereas rga-3 ( RNAi ) ( 4 . 5 hrs ) led to a significantly increased skew . Similar to the ABa cell , the ABp skew was also significantly reduced under ect-2 ( RNAi ) condition ( Figure 4—figure supplement 1A ) . However , the skew was unchanged in the ABp cell for rga-3 ( RNAi ) . We next determined the unprojected full ( 3D ) skew angle for each condition ( Figure 4—figure supplement 1C ) . Intriguingly , we observed that the unprojected full ABp cell skew was marginally increased under rga-3 ( RNAi ) condition ( Figure 4—figure supplement 1C ) even though the projected AP–LR plane skew remained unchanged ( Figure 4—figure supplement 1A ) . Similarly , we observed that the unprojected full ABa cell skew was not significantly different for ect-2 ( RNAi ) compared to the non-RNAi condition ( Figure 4—figure supplement 1C ) . This difference in skew between the AP–LR projected and the unprojected angles for the same conditions indicated that there could be an additional skew in a different plane . To this end , we determined the skew in the DV–LR plane . Interestingly , we detected an ∼20° skew in this plane even for the non-RNAi condition ( Figure 4—figure supplement 1B ) . Generally , the difference between the full ( 3D ) skew angles and the AP–LR projected angles is due to this additional skew in the DV–LR plane . This illustrates that the chiral skew at the 4-cell stage is more complex than previously reported , with a chiral rotation in the DV–LR plane in addition to the AP–LR plane . A characteristic myosin foci size was determined by performing spatial myosin fluorescence intensity autocorrelation in MATLAB . The autocorrelation was performed in a stripe of about 27 μm wide and 13 μm high in the anterior of the embryo and the analysis was carried out in each frame during the first 75 s of cortical flow . The spatial autocorrelation decay determined in each analysis frame was then fitted with a single exponential function and we define the decay length of this fit as the characteristic foci size . The foci size thus determined in each frame was then averaged over all analysis frames for a single embryo and an ensemble average was reported .	Most living things have left and right sides that are not identical . A well-known example of this ‘left–right asymmetry’ is the position of the human heart within the human body . While the human heart is always on the left , in other situations it is possible for either the left side or the right side to be preferred: for example , some people prefer to write with their right hand , while others prefer to write with their left hand . In animals , left–right asymmetry starts early in the development of the embryo . A structure in cells called the cytoskeleton is known to be responsible for generating the asymmetry in many species . The cytoskeleton is mostly made of two types of proteins—rod-like proteins called microtubules and filaments of a protein called actin—but it is not clear how it is involved in establishing left–right asymmetry . The cytoskeleton has many functions in the cell: for example , it maintains the shape of the cell , it splits the contents of the cell during cell division , and it transports various things around inside the cell . The cytoskeleton is constantly moving and changing shape: all this activity involves another protein called myosin that binds to the actin filaments and moves along them to generate pulling forces . Naganathan et al . studied newly fertilized embryos of the nematode worm Caenorhabditis elegans when they contained just one cell . The experiments showed that myosin can generate turning forces that twist the actin cortical layer , leading to local rotations in the cytoskeleton that make the cell asymmetrical . This is controlled by a group of proteins called Rho proteins . Next , Naganathan et al . studied embryos that contained four cells . Again , myosin generates local rotations in the cytoskeleton , which are involved in setting up left–right body direction in this stage of development . These experiments show that changes in the cytoskeleton of individual cells can drive asymmetry in the whole embryo . The next challenge will be to understand how myosin is controlled so that rotations only occur during specific cell divisions .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems" ]	2014	Active torque generation by the actomyosin cell cortex drives left–right symmetry breaking
Although Rad51 is the key protein in homologous recombination ( HR ) , a major DNA double-strand break repair pathway , several auxiliary factors interact with Rad51 to promote productive HR . We present an interdisciplinary characterization of the interaction between Rad51 and Swi5-Sfr1 , a conserved auxiliary factor . Two distinct sites within the intrinsically disordered N-terminus of Sfr1 ( Sfr1N ) were found to cooperatively bind Rad51 . Deletion of this domain impaired Rad51 stimulation in vitro and rendered cells sensitive to DNA damage . By contrast , amino acid-substitution mutants , which had comparable biochemical defects , could promote DNA repair , suggesting that Sfr1N has another role in addition to Rad51 binding . Unexpectedly , the DNA repair observed in these mutants was dependent on Rad55-Rad57 , another auxiliary factor complex hitherto thought to function independently of Swi5-Sfr1 . When combined with the finding that they form a higher-order complex , our results imply that Swi5-Sfr1 and Rad55-Rad57 can collaboratively stimulate Rad51 in Schizosaccharomyces pombe . DNA double-strand breaks ( DSBs ) are a particularly toxic form of DNA damage in which a DNA molecule is broken into two ( or more ) fragments . A major DSB repair pathway is homologous recombination ( HR ) . During HR , an intact stretch of DNA that shares sequence similarity to the DSB site is identified and utilized as a template for synthesis-dependent repair . Dysregulation of HR results in misrepair of DSBs , resulting in genomic instability , a potent driver of tumorigenesis ( Jeggo et al . , 2016 ) . HR is initiated by the formation of 3’ single-stranded DNA ( ssDNA ) at the DSB site . This ssDNA is first bound by RPA , then by the ubiquitous RecA-family recombinase Rad51 , which forms a right-handed nucleoprotein filament . The Rad51 filament is able to capture intact double-stranded DNA ( dsDNA ) and—by assessing the extent of base-pairing with the filamentous ssDNA—identify regions of DNA that share substantial sequence similarity to the DSB site ( Prentiss et al . , 2015 ) . After initial pairing with the complementary strand of the dsDNA , the Rad51 filament further displaces the noncomplementary strand by driving strand transfer , resulting in the formation of an intermediate structure known as a displacement loop . The 3’ end of the invading strand is then utilized as a primer for DNA synthesis , leading to its extension and the recovery of lost genetic information . In the simplest case , ejection of this extended strand allows it to anneal with the complementary DNA on the other side of the DSB ( Mehta and Haber , 2014 ) . Following gap filling by further DNA synthesis and ligation of resultant nicks , recombinational DNA repair is complete . Efforts to elucidate the underlying biochemistry of HR have typically involved measuring the ability of purified Rad51 to drive pairing and subsequent strand transfer of homologous DNA substrates , a process known as DNA strand exchange . Such experiments established RPA as a critical component of the DNA strand exchange reaction ( Sung , 1994 ) . However , when RPA was added to the reaction concomitantly with Rad51 , which more closely reflects the situation in vivo , the stimulatory effect of RPA was abolished . This paradoxical finding led to the discovery that other proteins known to be involved in HR serve as auxiliary factors that interact directly with Rad51 and can negate the inhibitory effect of RPA ( Hays et al . , 1995; Johnson and Symington , 1995; Sung , 1997a; Sung , 1997b; Tsutsui et al . , 2000; Tsutsui et al . , 2001; Jensen et al . , 2010 ) . Numerous distinct families of recombination auxiliary factors have been identified throughout eukaryotes , including Rad52 , BRCA2 , Rad54 , Rad51 paralogs , Swi5-Sfr1 and the Shu complex ( Zelensky et al . , 2014 ) . Each group is thought to have non-overlapping roles in HR , although the mechanistic differences are yet to be elucidated ( Akamatsu et al . , 2007; Khasanov et al . , 2004; Martín et al . , 2006; Shor et al . , 2005; Sugawara et al . , 2003 ) . Aside from the Shu complex , all auxiliary factors are capable of binding directly to Rad51 , which is thought to be essential for their respective roles in HR ( Zelensky et al . , 2014 ) . Sfr1 was first discovered in the fission yeast Schizosaccharomyces pombe as an interactor of Rad51 , and along with Swi5 , was shown to comprise an HR sub-pathway that functions independently of and in parallel to the Rad51 paralogs Rad55-Rad57 ( Akamatsu et al . , 2003; Akamatsu et al . , 2007 ) . Subsequent biochemical reconstitutions demonstrated that substoichiometric concentrations of Swi5-Sfr1 were able to efficiently stimulate the strand exchange activity of Rad51 and Dmc1 , the meiosis-specific RecA-family recombinase ( Haruta et al . , 2006 ) . This enhancement of strand exchange was attributed to stabilization of the nucleoprotein filaments and stimulation of the recombinases’ ATPase activity ( Haruta et al . , 2006; Kurokawa et al . , 2008; Murayama et al . , 2013 ) . Rad51-driven DNA strand exchange was recently shown to fit a three-step kinetic model with two reaction intermediates ( Ito et al . , 2018 ) . Swi5-Sfr1 enhanced transitioning of the first three-strand intermediate ( corresponding to a paranemic joint ) into the second three-strand intermediate ( corresponding to a plectonemic joint ) , and conversion of the second intermediate into reaction products , thus making it the only auxiliary factor known to potentiate Rad51 in both the presynaptic and synaptic phases of DNA strand exchange . These findings highlight the unique role of Swi5-Sfr1 as an HR regulator . Limited proteolysis of Swi5-Sfr1 yielded a stable C-terminal Sfr1 fragment in complex with Swi5 ( Swi5-Sfr1C , residues 181–299 of Sfr1; Kuwabara et al . , 2010 ) , and this , along with the N-terminal half of Sfr1 ( Sfr1N , residues 1–176 of Sfr1 ) , could be stably expressed and purified ( Kuwabara et al . , 2012 ) . Whereas Sfr1N was predicted to be intrinsically disordered ( Kokabu et al . , 2011; Saikusa et al . , 2013 ) , crystallographic analyses demonstrated that Swi5-Sfr1C forms a kinked structure that is able to stimulate Rad51-driven strand exchange by stabilizing the presynaptic filament and enhancing the ATPase activity of Rad51 ( Kuwabara et al . , 2012 ) . By contrast , Sfr1N had no direct effect on these activities but was seen to co-immunoprecipitate ( co-IP ) with Rad51 . Such complex formation was not detected between Rad51 and Swi5-Sfr1C , despite the stimulatory effect of Swi5-Sfr1C on Rad51 . Taken together with the observation that Swi5-Sfr1C was only able to stimulate Rad51 activity when present at much higher concentrations than full-length Swi5-Sfr1 , these results led to a model in which Sfr1N keeps Swi5-Sfr1C anchored in close proximity to Rad51 ( Kuwabara et al . , 2012 ) . Due to the use of truncated proteins in which entire domains were deleted ( Kuwabara et al . , 2012 ) , it was not possible to determine whether Sfr1N has any function other than anchoring Swi5-Sfr1 to Rad51 . To explore this , we employed an interdisciplinary approach to further characterize Sfr1N . We provide direct evidence that Sfr1N is intrinsically disordered and contains two sites that interact cooperatively with Rad51 . Mutation of critical residues within these two sites rendered Rad51 refractory to the stimulatory effects of full-length Swi5-Sfr1 , mimicking the results obtained with Swi5-Sfr1C ( i . e . , when the N-terminus of Sfr1 is absent ) , indicating that the primary function of Sfr1N is to facilitate the interaction between Swi5-Sfr1 with Rad51 . Unexpectedly , and in contrast to the severely impaired Rad51 stimulation observed in vitro , these interaction mutants only showed defects in Rad51-mediated DNA repair in the absence of Rad55-Rad57 , implying that these Rad51 paralogs can facilitate Swi5-Sfr1-dependent DNA repair . Consistent with this possibility , purified Swi5-Sfr1 was found to interact with partially purified Rad55-Rad57 . Collectively , these results provide a molecular basis for Rad51 stimulation by Swi5-Sfr1 and reveal a novel interplay between recombination auxiliary factors . Since Sfr1N binds to Rad51 but does not stimulate DNA strand exchange , and Swi5-Sfr1C stimulates DNA strand exchange despite not forming a detectable complex with Rad51 , it was proposed that Sfr1N functions exclusively to facilitate the interaction between Swi5-Sfr1C and Rad51 ( Kuwabara et al . , 2012 ) . However , it remained possible that Sfr1N only exerts a stimulatory effect when in the presence of Swi5-Sfr1C . To test this , strand exchange reactions containing purified Rad51 and plasmid-sized DNA substrates ( Figure 1A ) were supplemented with equimolar concentrations of both Sfr1N and Swi5-Sfr1C . Even in this setting , Sfr1N did not have any stimulatory effect on DNA strand exchange ( Figure 1B , C ) , raising the possibility that it is dispensable for the physiological function of Swi5-Sfr1 . To determine the requirement for these two modules in Rad51-dependent DNA repair , strains lacking the C-terminal or N-terminal half of Sfr1 ( sfr1N and sfr1C , respectively ) were constructed . Both strains showed the same sensitivity to DNA damage as a strain in which Sfr1 was completely absent ( sfr1Δ; Figure 1D ) . Furthermore , combining these truncations with rad55Δ sensitized cells to DNA damaging agents to the same degree as the rad55Δ sfr1Δ strain , which displays a complete loss of Rad51-dependent DNA repair ( Figure 1—figure supplement 1A; Akamatsu et al . , 2003; Akamatsu et al . , 2007 ) . Sfr1N and Sfr1C were detected at comparable levels to full-length Sfr1 by immunoblotting , indicating that the sensitivity of the sfr1N and sfr1C strains is not due to a reduction in protein levels ( Figure 1—figure supplement 1B ) . Furthermore , this sensitivity was not rescued by fusing Sfr1N or Sfr1C to the SV40 large T antigen nuclear localization signal , suggesting that the observed phenotype is not caused by a failure to localize to the nucleus ( Figure 1—figure supplement 1C ) . Thus , although not essential for stimulation of Rad51 in vitro , Sfr1N is essential for the function of Swi5-Sfr1 in promoting Rad51-dependent DNA repair . Having confirmed the physiological importance of Sfr1N , a structural approach was employed to glean insights into its molecular function . Primary sequence analysis and ion mobility mass spectrometry of Sfr1N suggested that this domain is intrinsically disordered ( Kokabu et al . , 2011; Saikusa et al . , 2013 ) . To directly test this , Sfr1N was analyzed by circular dichroism ( CD ) and nuclear magnetic resonance ( NMR ) spectroscopy . The far-UV CD spectrum of Sfr1N lacked local minima above 210 nm and showed a negative peak at ~200 nm ( Figure 2A ) , implying a lack of secondary structural units such as α-helices and β-sheets ( Greenfield , 2006 ) . Furthermore , examination of the 1H-15N heteronuclear single quantum coherence ( HSQC ) spectrum of Sfr1N revealed that most of the main chain amide protons resonated in a narrow chemical shift range between 7 . 7 and 8 . 7 ppm ( Figure 2B ) , which is a characteristic feature of disordered proteins ( Konrat , 2014 ) . To extract structural information for each residue , NMR signals from main-chain 1HN , 13Cα , 13CO , and 15NH atoms as well as 13Cβ resonances were assigned by analyzing a set of triple resonance spectra . This was assisted by the 1H-15N HSQC spectra of selectively-15N labeled versions of Sfr1N and several Sfr1N variants ( Figure 2—figure supplement 1 ) . The chemical shifts obtained for the main-chain and 13Cβ atoms enabled secondary structure prediction . The secondary chemical shift of 13Cα , 13Cβ , and 13CO atoms , which is the chemical shift difference between the measured values and the corresponding amino acids in random coil peptides , was determined ( Figure 2C ) . The majority of Sfr1N residues showed 13C secondary chemical shift values within a limited range , suggesting that random coil structures are present in these regions . Nonetheless , a few groups of residues exhibited more than three consecutive secondary chemical shift values outside of this range , raising the possibility that some structures resembling α-helices ( E27 to D29 , D32 to Q34 ) or β-strands ( L103 to K105 , R161 to K164 ) may form within Sfr1N ( Wishart and Sykes , 1994 ) . Further secondary structure analysis was performed using the 1HN , 13Cα , 13Cβ , 13CO , and 15NH chemical shift values and the program TALOS+ ( Shen et al . , 2009 ) , which predicted Sfr1N to be entirely disordered , with a low probability for α-helix formation from E27 to S33 ( Figure 2D ) . To analyze the dynamical features of Sfr1N , the steady state heteronuclear nuclear Overhauser effect ( NOE ) for the main-chain amide groups of the protein was analyzed ( Farrow et al . , 1994; Kay et al . , 1989 ) . NOE values for all residues was less than 0 . 44 , indicating that the entire protein is flexible with pico-to-nanosecond timescale motions ( Figure 2E ) . Such fast motions are typically observed for unstructured proteins . The NOE values were not completely uniform . While the average NOE values for all Sfr1N residues was 0 . 15 , residues E27 to S33 showed slightly increased values with an average of 0 . 32 . Other regions within Sfr1N also showed similarly increased NOE values ( e . g . , residues F91 to A99 , average of 0 . 32 ) . However , unlike these other regions , several additional lines of evidence pointed toward the possibility that residues E27 to S33 may form an α-helix ( see above ) . Collectively , these results demonstrate that , unlike the structured Swi5-Sfr1C complex ( Kuwabara et al . , 2012 ) , the N-terminal half of Sfr1 is intrinsically disordered and flexible . Disorder predictions using DISOPRED3 ( Jones and Cozzetto , 2015 ) suggest that the disordered state of Sfr1N may have been conserved throughout evolution ( Figure 2—figure supplement 2 , see Discussion for more details ) . Kuwabara et al . ( 2012 ) suggested that Sfr1N facilitates the interaction between Swi5-Sfr1 and Rad51 . To identify the site ( s ) within Sfr1N that binds to Rad51 , NMR spectra of 15N-labeled Sfr1N were analyzed in the absence and presence of Rad51 . Superimposed 1H-15N HSQC spectra of 15N-labeled Sfr1N with increasing amounts of Rad51 were constructed ( Figure 3A ) . The most prominent spectral changes , defined as a reduction in signal intensity of >80% ( Sfr1N:Rad51 ratio of 1:0 . 25 ) , were observed for 19 out of 142 non-overlapped residues ( Figure 3B , E ) . In addition to these marked changes , 18 and 20 residues experienced signal intensity reductions of 60–80% and 40–60% , respectively ( Figure 3E ) . The signal intensity of these residues was further attenuated upon incremental addition of Rad51 ( Figure 3—figure supplement 1A–C ) . Most of these attenuated signals did not display obvious chemical shift changes following Rad51 binding . However , five residues ( A71 , T73 , D75 , L76 , and T146 ) displayed incremental chemical shift changes and reductions in signal intensity upon addition of increasing amounts of Rad51 ( Figure 3C , Figure 3—figure supplement 1D ) . The remaining 60 residues that were analyzed experienced minimal effects upon addition of Rad51 ( <40% reduction in signal intensity; Figure 3D , E ) . These findings implicate two sites in Sfr1N , Site 1 ( S84 to T114 ) and Site 2 ( T152 to S168 ) , where the most significantly attenuated signal intensities are sandwiched by moderately attenuated signal intensities ( Figure 3E ) , as being important for the Sfr1N-Rad51 interaction . Site 1 is highly basic and hydrophobic compared to other regions of Sfr1N . Positively charged residues are also a prominent feature of Site 2 , but this site is not especially hydrophobic . These results suggest that , while both Sites 1 and 2 are involved in electrostatic interactions with Rad51 , Site 1 may also participate in hydrophobic interactions with Rad51 . We note that , like all RecA-family recombinases , S . pombe Rad51 exists as a multimer in solution , with a size corresponding to ~160 kDa ( Figure 3—figure supplement 2 , see Discussion for more details ) . To provide further support that Sites 1 and 2 within Sfr1N interact with Rad51 , site-specific crosslinking experiments were conducted ( Figure 4A ) . By utilizing Escherichia coli with an expanded genetic code , synthetic amino acids can be introduced at a site of interest via suppression of the amber ( UAG ) stop codon ( Young et al . , 2010 ) . Translation is terminated prematurely without amber suppression ( e . g . , when the synthetic amino acid is omitted from the growth media ) , hence ensuring that all full-length protein products contain the synthetic amino acid . Several residues within Sites 1 and 2 were replaced with the photoreactive amino acid p-benzoyl-L-phenylalanine ( pBPA ) . Following exposure to UV light , proteins within ~3 Å of pBPA become crosslinked to it ( Tanaka et al . , 2008 ) . Such crosslinked proteins can be detected as slow-migrating species by immunoblotting and are implicated in forming part of the interface in a protein-protein interaction ( Miyazaki et al . , 2016 ) . Rad51 was co-expressed in E . coli with Sfr1N and cells were irradiated with UV . Proteins were then analyzed by immunoblotting with anti-Rad51 and anti-Sfr1 antibodies . Some non-specific crosslinking was observed when cells were treated with UV ( Figure 4—figure supplement 1A ) . In addition to these non-specific crosslinks , numerous instances of specific crosslinking—defined as being dependent on a TAG mutation , the inclusion of pBPA in the media , and UV treatment—were observed ( Figure 4B ) . Whereas several positions within Site 1 showed robust crosslinking to Rad51 , positions within Site 2 showed little crosslinking ( Figure 4C , Figure 4—figure supplement 1B ) , perhaps because the associations between Site 2 and Rad51 are more transient than those involving Site 1 ( see Discussion ) . Taken together , the results obtained from the NMR interaction analysis and Site-specific crosslinking experiments indicate that two sites within Sfr1N , designated as Sites 1 and 2 , interact with Rad51 . A sequence alignment of Sfr1 orthologs within the genus Schizosaccharomyces highlighted conserved patches of positively charged residues in Sites 1 and 2 ( bold residues in blue typeface , Figure 5—figure supplement 1A ) . Combined with the knowledge that Sfr1N only co-IPs with Rad51 under low-salt conditions ( Kuwabara et al . , 2012 ) , it seemed plausible that these residues might be important for electrostatic interactions with Rad51 . Hence , three residues in Site 1 were mutated ( 3A ) and four residues in Site 2 were mutated ( 4A ) . Additionally , these mutants were combined to generate the 7A mutant ( Figure 5A ) . To directly assess whether these mutations disrupt the interaction with Rad51 , Swi5 and full-length Sfr1 were co-purified to homogeneity ( Figure 5—figure supplement 1B ) . Next , purified Rad51 was crosslinked to Affi-gel matrix and mixed with Swi5-Sfr1 . A substantial fraction of wild-type Swi5-Sfr1 was recovered in the eluate ( Rad51-bound fraction ) , although some of the protein remained in the flow-through ( unbound fraction; Figure 5B ) . By contrast , the amount of 3A and 4A mutant proteins detected in the eluate was reduced , with much of the protein remaining in the flow-through . The 7A mutant protein was barely detected in the eluate , indicating that the binding seen in the 3A mutant was dependent on Site 2 and the binding seen in the 4A mutant was dependent on Site 1 . Comparable trends were observed in a co-IP assay that did not involve crosslinking of Rad51 ( Figure 5—figure supplement 1C ) . Taken together , these results suggest that Sites 1 and 2 facilitate the binding of Swi5-Sfr1 to Rad51 in a cooperative manner . In the 7A mutant , both Sites 1 and 2 are mutated but the remainder of the N-terminus is intact . Thus , the 7A mutant can be employed to test whether the N-terminal domain of Sfr1 has any significant role other than to facilitate binding to Rad51 . We therefore proceeded to characterize the biochemical activities of the 7A mutant . The 3A and 4A mutants were included to glean further insights into the nature of Rad51 stimulation by Swi5-Sfr1 . While Swi5-Sfr1C can stimulate Rad51 activity despite the absence of Sfr1N , 5-to-10-fold more of the complex is required to achieve the same level of stimulation as full-length Swi5-Sfr1 ( Figure 1B lanes 3 and 12; Kuwabara et al . , 2012 ) , suggesting that the interaction between Sfr1N and Rad51 is important for efficient stimulation of strand exchange . Consistent with the observed binding defect , substoichiometric concentrations of the 7A mutant failed to efficiently stimulate Rad51-driven strand exchange , with a higher concentration of mutant protein required to achieve a wild-type level of joint molecules ( JM , reaction intermediates ) and nicked-circles ( NCs , reaction products; Figure 5C ) . At 0 . 25 µM , the defect of the 7A protein was more pronounced for NC ( ~15 fold reduction ) than JM ( ~5 fold reduction ) , suggesting that the ability of Swi5-Sfr1 to stimulate both the initial pairing of homologous DNA and the subsequent strand transfer by Rad51 are defective when the interaction with Sites 1 and 2 is ablated ( Figure 5D–F ) . By contrast , the 3A and 4A mutants were able to promote efficient JM and NC formation at substoichiometric concentrations . Nevertheless , the loss of Rad51 stimulation observed at higher concentrations of wild-type Swi5-Sfr1 was attenuated in the 3A and 4A mutants ( Figure 5C lanes 6 , 21 and 28 ) , suggesting that this loss of stimulation occurs due to unproductive interactions with Rad51 or sequestration of DNA substrates by Swi5-Sfr1 ( Figure 6—figure supplement 1 and see Discussion ) . Consistent with this notion , efficient stimulation of Rad51 was maintained at higher concentrations of the 7A mutant ( Figure 5C lanes 8 and 15 ) and Swi5-Sfr1C ( Figure 1B , C; Kuwabara et al . , 2012 ) . Collectively , these results indicate that interactions between Rad51 and both Sites 1 and 2 are important for efficient stimulation of strand exchange . To determine why stimulation of Rad51-driven strand exchange is inefficient when Sites 1 and 2 are mutated , the molecular roles of Swi5-Sfr1 were considered . At substoichiometric concentrations , Swi5-Sfr1 effectively stabilizes Rad51 presynaptic filaments ( Kurokawa et al . , 2008 ) . Thus , it seemed feasible that the observed impairment in strand exchange might be explained by defects in Rad51 filament stabilization . To test this possibility , filament stability was examined by fluorescence anisotropy . When Rad51 binds to a fluorescently-labeled oligonucleotide and forms a filament , fluorescence anisotropy increases due to a retardation in the motion of the labeled oligonucleotide ( Figure 6A ) . The dissociation of Rad51 is accompanied by a reduction in anisotropy , with the rate of decline reflective of Rad51 filament stability . Rad51-ssDNA filaments were formed in the presence of ATP and filament collapse was induced via dilution into reaction buffer containing ATP but lacking DNA and protein . In the absence of Swi5-Sfr1 , the decrement in anisotropy was sharp and reached a value that was observed in the absence of protein ( ~0 . 1 ) within ~500 s . Inclusion of wild-type Swi5-Sfr1 resulted in a slower reduction in anisotropy , indicating that Rad51 filaments had been stabilized ( Figure 6B ) . Strikingly , inclusion of the 7A mutant did not result in any obvious filament stabilization ( Figure 6C ) . Furthermore , although both 3A and 4A mutants showed some stabilization of Rad51 filaments , the magnitude of this stabilization was less than that observed for wild-type protein ( Figure 6D , E ) . Consistent with these observations , the reaction rate constants for dissociation of Rad51-ssDNA complexes ( koff ) showed a substantial decline in the presence of Swi5-Sfr1 , a lesser decline for the 3A and 4A mutants , and only a marginal decline for the 7A mutant ( Figure 6F ) . Taken together , these results indicate that Sites 1 and 2 within Sfr1N interact cooperatively with Rad51 to facilitate filament stabilization by Swi5-Sfr1 . In addition to stabilizing Rad51 filaments , Swi5-Sfr1 has been shown to stimulate the ATPase activity of Rad51 , which is also important for efficient strand exchange ( Haruta et al . , 2006; Kurokawa et al . , 2008; Ito et al . , 2018 ) . Since substoichiometric concentrations of Swi5-Sfr1C failed to efficiently stimulate the ATPase activity of Rad51 ( Kuwabara et al . , 2012 ) , we sought to determine whether Rad51-dependent ATP hydrolysis was potentiated by the 7A mutant . As expected , wild-type Swi5-Sfr1 was able to efficiently enhance ATP hydrolysis by Rad51 at substoichiometric concentrations ( Swi5-Sfr1:Rad51 ratio of 1:20 ) , with a 1 . 85-fold increase in ATP turnover ( Figure 6G ) . By contrast , the 7A mutant only managed to enhance the ATPase activity of Rad51 1 . 28-fold , which is similar to the 1 . 27-fold stimulation observed with Swi5-Sfr1C . The 3A and 4A mutants stimulated ATP hydrolysis like wild type . These results suggest that interaction of either Site 1 or 2 with Rad51 is sufficient to promote efficient stimulation of ATP hydrolysis . To analyze the in vivo DNA repair activity of the Sfr1-Rad51 interaction mutants , strains were constructed in which the native sfr1+ gene was replaced with either sfr1-7A , sfr1-3A or sfr1-4A . Unexpectedly , the interaction mutants did not show any obvious sensitivity to DNA damage ( Figure 7A , Figure 7—figure supplement 1A ) , which is in sharp contrast to the DNA repair defect of the N-terminal deletion strain ( Figure 1D , Figure 1—figure supplement 1A , C ) . A marginal sensitivity was observed for the sfr1-7A strain but this was not statistically significant ( Figure 7B ) . Previous genetic studies suggested that there are two HR sub-pathways in S . pombe: one dependent on Swi5-Sfr1 and the other dependent on the Rad51 paralogs Rad55-Rad57 ( Akamatsu et al . , 2003; Akamatsu et al . , 2007 ) . Thus , in the absence of Rad55/Rad57 , it is possible to evaluate the Rad51-dependent DNA repair that is mediated solely by Swi5-Sfr1 . For this purpose , the interaction mutants were introduced into the rad55Δ background . Strikingly , in the absence of Rad55 , the sfr1-7A mutant showed the same DNA damage sensitivity as the sfr1Δ mutant ( Figure 7C ) . Furthermore , both the sfr1-3A and sfr1-4A mutants were more sensitive to DNA damage than sfr1+ in the rad55Δ background , although this sensitivity was not as severe as that observed for the rad55Δ sfr1Δ and rad55Δ sfr1-7A strains ( Figure 7C , D ) . Similar results were obtained in the rad57Δ background ( Figure 7—figure supplement 1B ) . These results indicate that the residues mutated in sfr1-7A are indeed important for DNA repair , while also suggesting that the DNA repair defects of Sfr1 amino acid-substitution mutants are suppressed by a Rad55-Rad57-dependent mechanism . One possible explanation for the above results is that Rad55-Rad57 plays a role in facilitating the recruitment of Swi5-Sfr1 to Rad51 , perhaps by acting as a molecular bridge . A prerequisite of this model is that Rad55-Rad57 can bind to both Rad51 and Swi5-Sfr1 . While an interaction between Rad57 and Rad51 has been reported by yeast two-hybrid analysis ( Tsutsui et al . , 2001 ) , a possible interaction between Rad55-Rad57 and Swi5-Sfr1 was never examined due to the genetic evidence suggesting that they comprise independent sub-pathways of HR ( Akamatsu et al . , 2003; Akamatsu et al . , 2007 ) . To test if Rad55-Rad57 and Swi5-Sfr1 interact with each other , we sought to partially purify Rad55-Rad57 . Protein A-tagged Rad55 and untagged Rad57 were co-expressed in S . pombe ( Figure 7—figure supplement 1C ) . As a negative control , the same experiment was conducted in parallel using a strain transformed with empty vectors . Cleared cell lysates were incubated with IgG-agarose resin to enrich Protein A-tagged proteins . Rad55-Rad57 was then specifically eluted via the addition of 3C protease , which cleaves between Rad55 and resin-bound Protein A . Rad55 and Rad57 constituted the two major bands in this partially purified protein preparation and were present in seemingly stoichiometric amounts ( Figure 7—figure supplement 1D ) , consistent with previous reports in budding yeast ( Sung , 1997a; Liu et al . , 2011 ) . Partially purified Rad55-Rad57 was incubated with purified Swi5-Sfr1 ( wild type or 7A ) and Rad57 was immunoprecipitated . The contents of these immunoprecipitates were then examined by immunoblotting . As expected , Rad55 was found to co-IP with Rad57 ( Figure 7E ) . Moreover , both Swi5-Sfr1 and Swi5-Sfr1-7A coIP’d with Rad57 . While partially purified Rad55-Rad57 appeared to be reasonably pure ( Figure 7—figure supplement 1D ) , it remained possible that Rad51 had co-purified with Rad55-Rad57 throughout the purification process . If so , Rad51 bound to Rad55-Rad57 could interact with Swi5-Sfr1 , leading to indirect co-IP of Sfr1 with Rad57 . To test this possibility , the anti-Rad57 immunoprecipitates were probed with an anti-Rad51 antibody . Importantly , Rad51 was completely undetectable in these immunoprecipitates ( Figure 7F ) , suggesting that Swi5-Sfr1 and Rad55-Rad57 directly interact in a Rad51-independent manner . Although Sfr1N is not essential for stimulation of Rad51-driven DNA strand exchange ( Figure 1B , C; Kuwabara et al . , 2012 ) , it is essential for the promotion of Rad51-dependent DNA repair by Swi5-Sfr1 ( Figure 1D , Figure 1—figure supplement 1A , C ) . The fact that Swi5-Sfr1C can exert a stimulatory effect on Rad51 in vitro points towards the existence of a physical interaction between the two . However , the inability to detect such a complex , combined with the relative inefficiency with which Swi5-Sfr1C potentiates Rad51 , strongly suggests that the interaction is too weak to observe by conventional means . This binding is likely augmented by the N-terminal fragment of Sfr1 , which we posit functions as an anchor to keep Swi5-Sfr1C in close proximity to Rad51 . Consistent with such a role for Sfr1N , NMR interaction analysis revealed that two domains within Sfr1N interact with Rad51 ( Figure 3 , Figure 3—figure supplement 1 ) . Mutation of Site 1 or 2 weakened the interaction with Rad51 , while mutation of both sites resulted in a near-complete loss of interaction ( Figure 5B , Figure 5—figure supplement 1C ) , indicating that two sites within Sfr1N bind cooperatively to Rad51 . Interestingly , Rad51-dependent strand exchange and ATP hydrolysis were significantly impaired only when both sites were mutated ( Figures 5C–F and 6G ) . These results indicate that the reduced interaction in the single site mutants is sufficient for Swi5-Sfr1 to fully stimulate Rad51 in these assays . Although this points toward functional redundancy between Sites 1 and 2 , it is possible that these assays are not sensitive enough to detect marginal defects in the stimulation of Rad51 . Indeed , the fluorescence anisotropy assay revealed a severe defect in Rad51 filament stabilization for the 7A mutant and a modest defect for the 3A and 4A mutants ( Figure 6B–F ) , arguing that interaction of both Sites 1 and 2 with Rad51 is important for efficient filament stabilization . Swi5-Sfr1 has also been shown to stabilize Rad51 filaments against the F-box helicase Fbh1 ( Tsutsui et al . , 2014 ) . It would be interesting to test whether the interaction mutants can function in a similar capacity . The reduction in NMR signals from Sites 1 and 2 in the presence of Rad51 ( Figure 3E ) , combined with the gradual chemical shift changes observed for some residues ( Figure 3C , Figure 3—figure supplement 1D ) , indicated that the association and dissociation of Sfr1N and Rad51 is fast on the NMR timescale , suggesting that the Sfr1N-Rad51 interaction is relatively weak . While monomeric Rad51 ( ~40 kDa ) is too small to cause severe line-broadening of NMR signals , the interaction of multimeric Rad51 ( Figure 3—figure supplement 2 ) with Sfr1N could explain the drastic reduction in NMR signals for Sites 1 and 2 . These results were largely substantiated by site-specific crosslinking of residues within Site 1 , and to a much lesser extent Site 2 , to Rad51 ( Figure 4 , Figure 4—figure supplement 1B ) . Since both sites are involved in electrostatic interactions with Rad51 , the more robust crosslinking of Site 1 may be due to the added contribution of hydrophobic interactions between Site 1 and Rad51 . One interpretation of these results is that the Site 1-Rad51 interaction is of higher affinity than the Site 2-Rad51 interaction . However , the results of co-IP experiments suggested that mutating Site 2 disrupts the Rad51 interaction to a greater extent than mutating Site 1 ( compare 3A and 4A in Figure 5B , Figure 5—figure supplement 1C ) . When making such comparisons , it is important to note that the NMR and crosslinking experiments only involved Sfr1N , whereas the co-IP analysis involved full-length Sfr1 in complex with Swi5 . It is possible that Site 2 plays a more prominent role in the interaction with Rad51 in the context of the full-length complex . The reduced crosslinking of Site 2 residues could also be explained by the fact that replacement of a given residue with the aromatic pBPA could itself disrupt the interaction with Rad51 , especially since Site 2 is less hydrophobic than Site 1 . Previous structural analysis indicated that the self-association of Rad51 involves the conserved FxxA motif ( where x can be any residue ) and that the BRC3 and BRC4 repeats of BRCA2 bind to this motif through Rad51 mimicry , leading to destabilization of the Rad51 filament ( Shin et al . , 2003 ) . This identified destabilization/stabilization of the interactions between Rad51 monomers as one way of regulating filament stability . A distinct method of stabilizing Rad51 filaments was uncovered for budding yeast Rad55-Rad57 , which was shown to cap the end of Rad51 filaments and antagonize the filament destabilizing activity of the Srs2 anti-recombinase ( Liu et al . , 2011 ) . Interestingly , neither Swi5 nor Sfr1 contains an FxxA motif . This is consistent with the identification of Rad51 interaction sites within the intrinsically disordered N-terminal domain of Sfr1 ( Figure 2 , Figure 2—figure supplement 1 ) , which has undergone substantial sequence divergence ( Figure 5—figure supplement 1A ) . We therefore favor the previously proposed model whereby Swi5-Sfr1C inserts into the wide grooves of the Rad51 nucleoprotein filament , locking the filament in an active conformation , with Sfr1N plastering along the outside of the filament to maintain Swi5-Sfr1C within the groove ( Fornander et al . , 2014; Kokabu et al . , 2011; Kuwabara et al . , 2012 ) . While this model could explain how Swi5-Sfr1 stabilizes the filament , it does not explain how Swi5-Sfr1 stimulates ATP hydrolysis and extensive strand transfer by Rad51 , which are highly dynamic processes thought to involve dissociation of Rad51 from DNA ( Ito et al . , 2018 ) . It is possible that the flexibility of Sfr1N allows Swi5-Sfr1 to remain bound to Rad51 despite conformational changes in the filament . This could also entail release of dissociating Rad51 molecules and re-binding of Sfr1N to molecules incorporated in the filament , thus preventing diffusion of Swi5-Sfr1 from the Rad51 filament . NMR interaction analysis suggested that binding of Sfr1N to Rad51 is relatively short-lived ( see above ) , consistent with a model in which the dynamic association and dissociation of Swi5-Sfr1 from Rad51 plays a role in stimulation of DNA strand exchange . Sites 1 and 2 may comprise a single Rad51 interacting unit in 3D space , although they are unlikely to share the same interface on Rad51 , given that Site 1 is more hydrophobic than Site 2 . Alternatively , Site 1 may bind to Rad51 with Site 2 acting as a scaffold , consistent with the results of our crosslinking ( Figure 4C , Figure 4—figure supplement 1B ) and co-IP ( Figure 5B , Figure 5—figure supplement 1C ) experiments . We note that the 7A mutant complex displayed a near-loss of DNA binding and the 4A mutant showed a clear defect in DNA binding ( Figure 6—figure supplement 1A ) . A mild impairment in DNA binding became apparent upon closer inspection of the 3A mutant ( Figure 6—figure supplement 1B ) . It is likely that these defects were caused by mutation of the positively charged Lys/Arg residues to Ala residues , leading to neutralization of the electrostatic attraction to DNA . Despite showing different levels of DNA binding , 3A and 4A were indistinguishable in all other aspects ( Figures 5–7 ) . Although it is formally possible that our observations with the 7A mutant are related to a defect in DNA binding , the high similarity of the results obtained with 3A and 4A argues that DNA binding by Swi5-Sfr1 is impertinent to its role in stimulating Rad51 activity or promoting Rad51-dependent DNA repair . Notably , mouse Swi5-Sfr1 ( mSwi5-Sfr1 ) stimulates Rad51 through very similar mechanisms despite being unable to bind DNA ( Tsai et al . , 2012 ) . This is in stark contrast to other recombination auxiliary factors such as Rad52 , Rad54 and Hop2-Mnd1 , whose ability to bind DNA is integral for recombinase stimulation ( Seong et al . , 2008; Wright and Heyer , 2014; Zhao et al . , 2014 ) . In addition to its involvement in DNA repair , human Sfr1 ( hSfr1 ) has been implicated in transcriptional regulation ( Feng et al . , 2013; Yuan and Chen , 2011 ) , raising the possibility that S . pombe Sfr1 may also have functions unrelated to DNA repair . In agreement with this , Cipak et al . ( 2009 ) reported that Swi5-Sfr1 forms a complex with the XPG-family RNA nuclease Mkt1 ( SPAC139 . 01c ) , which was recently shown to be involved in RNAi-mediated silencing and establishment of heterochromatin ( Taglini et al . , 2019 ) . We speculate that the ability of Sfr1 to bind DNA may be related to an as yet uncharacterized function . While the sfr1Δ and rad55Δ single mutants are sensitive to DNA damage , neither is as sensitive as rad51Δ , which is epistatic to both ( Akamatsu et al . , 2003; Khasanov et al . , 1999 ) . However , because the sfr1Δ rad55Δ double mutant shows the same sensitivity as rad51Δ ( e . g . , Figure 7A ) , it was concluded that two independent sub-pathways of HR exist in S . pombe ( Akamatsu et al . , 2003 ) . Despite the numerous defects observed in vitro , the sfr1-7A mutant strain was proficient for DNA repair , but this repair was dependent on Rad55-Rad57 ( Figure 7A–D ) , indicating that a Rad55-Rad57-dependent mechanism overcomes defects in the binding of Swi5-Sfr1 to Rad51 . To explain these results , we propose that the interaction of Swi5-Sfr1 with Rad51 is enabled by two redundant mechanisms: one through a direct interaction involving Sites 1 and 2 in Sfr1N and the other through Rad55-Rad57 , which interacts with Rad51 ( Tsutsui et al . , 2001 ) and acts as a molecular bridge to facilitate the recruitment of Swi5-Sfr1 to Rad51 . Hence , although Swi5-Sfr1-7A cannot interact directly with Rad51 , Rad55-Rad57 aids the recruitment of Swi5-Sfr1-7A to Rad51 , allowing it to exert a stimulatory effect on Rad51; this would explain why sfr1-7A is proficient for DNA repair . However , in the absence of Rad55/Rad57 , this tethering is lost but Swi5-Sfr1 can nevertheless promote some DNA repair via its direct interaction with Rad51 , thus explaining why rad55Δ is not as sensitive as rad51Δ . It is only when both interaction mechanisms are defective , as in the rad55Δ sfr1-7A strain , that the promotion of Rad51-mediated DNA repair by Swi5-Sfr1 is completely lost . Consistent with this possibility , Swi5-Sfr1 was found to form a Rad51-independent complex with partially purified Rad55-Rad57 ( Figure 7E , F ) . Critically , Swi5-Sfr1-7A was also able to form a complex with Rad55-Rad57 . The apparent absence of Rad51 ( Figure 7F ) —the major interacting partner of Rad55-Rad57—combined with the reasonably high purity achieved by the one-step purification process ( Figure 7—figure supplement 1D ) implies that this complex formation involves direct binding of Swi5-Sfr1 to Rad55-Rad57 , in support of the proposed model . We therefore surmise that , while the Swi5-Sfr1 and Rad55-Rad57 sub-pathways are capable of operating independently of each other , as observed in the rad55Δ and sfr1Δ backgrounds , Swi5-Sfr1 and Rad55-Rad57 likely collaborate to promote Rad51-dependent DNA repair in wild-type cells . Rad55-Rad57 facilitates recruitment of the Shu complex to Rad51 by binding to both and acting as a molecular bridge ( Gaines et al . , 2015; Khasanov et al . , 2004 ) , so it could plausibly fulfill a similar role for Swi5-Sfr1 . However , unlike the Shu complex , Swi5-Sfr1 can interact directly with Rad51 , so any contribution made by Rad55-Rad57 to this interaction would enhance rather than enable complex formation with Rad51 . The requirement for such a mechanism may stem from the fact that the direct interaction between Swi5-Sfr1 and Rad51 is relatively weak ( see above ) . Indeed , previous attempts by us and others to co-IP Swi5-Sfr1 and Rad51 from yeast extracts has been unsuccessful ( Akamatsu et al . , 2003; Cipak et al . , 2009 ) , suggesting that the cellular interaction is too weak to capture . It is tempting to speculate that Rad55-Rad57 , Swi5-Sfr1 , and the Shu complex exist as a higher-order auxiliary factor complex , perhaps as part of a Rad52-containing DNA repair center ( Lisby et al . , 2001; Lisby et al . , 2003 ) . Evidence for the existence of such a complex , along with the delineation of the relationship between Swi5-Sfr1 and Rad55-Rad57 , will be the focus of future research . Interestingly , unlike sfr1-7A , sfr1C showed the same DNA damage sensitivity as sfr1Δ even in the presence of Rad55-Rad57 ( Figure 1D , Figure 1—figure supplement 1A , C ) . This suggests that , in addition to its primary role in anchoring Swi5-Sfr1 to Rad51 , Sfr1N may have a secondary role in coordinating the collaboration between Swi5-Sfr1 and Rad55-Rad57 . Although the role of Swi5-Sfr1 in promoting Rad51-dependent DNA repair is conserved in mammals ( Akamatsu and Jasin , 2010; Argunhan et al . , 2017a; Lu et al . , 2018; Su et al . , 2014; Su et al . , 2016; Yuan and Chen , 2011 ) , the precise mode of interaction with Rad51 appears to have undergone some divergence ( Tsai et al . , 2012; Yuan and Chen , 2011 ) . In Saccharomyces cerevisiae , the Swi5-Sfr1 homolog Sae3-Mei5 is produced only during meiosis and functions exclusively in the Dmc1 branch of meiotic HR ( Hayase et al . , 2004; Tsubouchi and Roeder , 2004 ) . Both Swi5-Sfr1 and Sae3-Mei5 stimulate the strand exchange activity of Dmc1 via similar mechanisms ( Ferrari et al . , 2009; Haruta et al . , 2006; Murayama et al . , 2013 ) , and although Sae3-Mei5 does interact directly with Rad51 , it does not stimulate the activity of Rad51 ( Cloud et al . , 2012; Say et al . , 2011 ) , unlike Swi5-Sfr1 . A consistent trend across all examined species is that Sfr1 plays some role in facilitating the interaction with the recombinase partner . Since the amino acid sequence of the N-terminal half of Sfr1 shows little conservation compared to the C-terminal half ( Figure 5—figure supplement 1A ) , it is tempting to ascribe the similarities among species to the C-terminus . However , sequence divergence across large evolutionary distances does not necessarily reflect a lack of functional conservation for intrinsically disordered regions , which accumulate mutations at a higher rate than structured domains ( Brown et al . , 2002 ) . Notably , the large subunit of RPA contains an intrinsically disordered region whose function is conserved despite significant sequence divergence ( Daughdrill et al . , 2007 ) , raising the possibility that the structure and/or function of Sfr1N is conserved . While empirical evidence is lacking , disorder predictions ( Jones and Cozzetto , 2015 ) for the N-terminal half of S . pombe Sfr1 agree with the data presented here ( Figure 2—figure supplement 2A ) , and similar profiles were generated for Sfr1 from Schizosaccharomyces japonicus and Schizosaccharomyces octosporus ( Figure 2—figure supplement 2B , C ) . Furthermore , the N-terminal halves of mSfr1 , hSfr1 and Mei5 are predicted to be enriched in intrinsically disordered regions ( Figure 2—figure supplement 2D–F ) . This analysis also highlighted potential protein binding sites within the N-terminal halves of S . japonicus Sfr1 , hSfr1 and Mei5 . In support of this , the N-terminal half of Mei5 has already been shown to interact with Dmc1 ( Hayase et al . , 2004; Say et al . , 2011 ) . Thus , in addition to the conserved function of Swi5-Sfr1 in promoting HR , the intrinsically disordered nature of Sfr1’s N-terminus and its role in facilitating interactions with recombinases may be evolutionarily conserved . Further studies will be required to test the validity of this prediction . For the Key Resources Table , please see Supplementary file 1 . All strains are listed in the Key Resources Table ( Supplementary file 1 ) . Except for BA1 ( Msmt-0 leu-1–32 ura4-D18 arg3-D1 isp6::hphMX4 psp3::kanMX4 ) , all S . pombe strains are isogenic derivatives of strain YA119 ( Akamatsu et al . , 2003 ) ; Msmt-0 leu-1–32 ura4-D18 his3-D1 arg3-D1 ) . Standard media was used for growth ( YES ) , selection ( YES with drugs or EMM ) , and sporulation ( SPA ) , as described previously ( Hentges et al . , 2005 ) . All strains are listed in the Key Resources Table ( Supplementary file 1 ) . E . coli strains were constructed by transforming BL21 ( DE3 ) containing the pEVOL-pBpF plasmid ( Young et al . , 2010 ) with a pBKN220 plasmid ( Haruta et al . , 2006 ) encoding Sfr1N ( with or without a TAG mutation ) and a pET28a plasmid encoding Rad51 . Thus , strains only differ in the expression of Sfr1N or Rad51 and only these differences are indicated in the Key Resources Table ( Supplementary file 1 ) . A ‘–” sign indicates that cells were transformed with an empty vector , whereas a ‘+” sign signals the presence of Sfr1N or Rad51 on that vector . If a codon in Sf1N was mutated to TAG , the mutated residue is listed instead of a ‘+” sign . Strains are listed in order of appearance . Standard media was used for growth ( LB ) and selection ( LB with antibiotics ) , unless otherwise indicated . A single colony was resuspended in 2 mL of YES and grown for 24 hr ( rad+ ) or 48 hr ( rad– ) . Cells from these cultures were then seeded into 2 mL of fresh YES and grown for ~14 hr until they reached log phase ( ~0 . 8×107 cells/mL ) . Cell density was adjusted to 2 × 107 cells/mL , 10-fold serial dilutions were made , and 5 µL of each dilution was spotted onto YES plates ( no treatment control ) or YES plates containing the indicated genotoxins . For UV irradiation , cells were spotted onto YES plates and treated with acute UV exposure of the indicated dose . The leftmost spot on each plate contains 1 × 105 cells . Cells were photographed with a digital camera after growth at 30 °C ( 2–4 days , as indicated ) . For clonogenic assays , cells were grown as described above and spread onto several YES plates and irradiated with the indicated dose of UV . After 3 ( rad+ ) or 4 ( rad– ) days of growth , colonies were counted . Cells ( 1 × 108 ) were harvested and processed exactly as previously described ( Argunhan et al . , 2017b ) . Briefly , harvested cells were resuspended in 1 mL of ice-cold water . 150 µL of lysing solution ( 1 . 85 M NaOH 7 . 5% beta-mercaptoethanol ) was added and mixed with the cells , followed by a 15 min incubation on ice . 150 µL of 55% TCA was added , followed by a further 10 min incubation on ice . Precipitated proteins were pelleted by centrifugation ( 16 , 000 g 10 min 4 °C ) and dissolved with mixing ( 65 °C 10 min ) in 100 µL of urea loading buffer ( 8 M urea , 5% SDS , 200 mM Tris-Cl pH 6 . 8 , 1 mM EDTA , 0 . 01% BPB , freshly supplemented with 10% vol each of 1 M DTT and 2 M Tris ) . Proteins were separated by SDS-PAGE , transferred to PVDF membranes , and detected with the indicated antibodies . CD measurements were made using a Jasco J-720W spectrometer with a Peltier temperature controller . Sfr1N ( 4 µM ) in buffer N ( 20 mM sodium phosphate [pH 6] , 25 mM NaCl , 1 mM DTT ) was placed in a 0 . 1 cm path length quartz cuvette . The CD spectrum was acquired from 180 nm to 260 nm at 25°C with a 1 . 0 nm bandwidth , 0 . 5 nm resolution , 50 nm/min scan speed , with 1 s averaging at each wavelength . Three spectra were averaged to give the spectrum of the protein and blank spectrum measured for buffer N alone was subtracted to produce the final spectrum . Sfr1N ( residues 1–176 ) was subcloned into pBKN220 ( Haruta et al . , 2006 ) , which was transformed into the E . coli strain BL21 ( DE3 ) RIPL . Plasmids containing Sfr1N variants were prepared by using the protocol in the QuikChange Site-Directed Mutagenesis Kit ( Agilent ) . For the production of uniformly 15N-labeled or 13C and 15N-labeled proteins , E . coli cells were grown in M9 media supplemented with 15NH4Cl ( 1 g/L , Cambridge Isotope Laboratories ) or with 15NH4Cl and 13C6-glucose ( 3 g/L , SHOKO Science ) , respectively . For the production of amino acid selectively labeled proteins , modified M9 media supplemented with a 15N-labeled amino acid ( either Ala [200 mg/L] ) , Arg [200 mg/L] , Ile [100 mg/L] , Leu [100 mg/L] , Lys [100 mg/L] or Phe [50 mg/L] , all from Cambridge Isotope Laboratories ) and other non-labeled amino acids ( 400 mg/L Ala , 400 mg/L Arg , 250 mg/L Asp , 50 mg/L Cys , 400 mg/L Glu , 400 mg/L Gly , 100 mg/L His , 100 mg/L Ile , 100 mg/L Leu , 150 mg/L Lys , 50 mg/L Met , 50 mg/L Phe , 150 mg/L Pro , 1000 mg/L Ser , 100 mg/L Thr , 50 mg/L Trp , 100 mg/L Tyr , and 100 mg/L Val ) was used for culturing cells . Cultures were shaken in baffled flasks at 37°C until the OD600 reached 0 . 8 ~ 1 . 0 . Protein expression was then induced by the addition of 0 . 5 mM isopropyl-β-D-thiogalactoside ( IPTG ) for 20 hr at 20°C . For cultures with labeled amino acids , the induction was limited to ≤4 hr to minimize isotope scrambling . Sfr1N was then purified as described ( Kuwabara et al . , 2010 ) , except the storage buffer used was buffer N ( 20 mM sodium phosphate [pH 6] , 25 mM NaCl , 1 mM DTT ) . A truncated version of Sfr1N ( 127-176 ) was purified as a fusion protein with an N-terminal maltose-binding protein tag . Following purification with amylose resin ( NEB ) , the tag was cleaved with Factor Xa and the cleaved peptide was isolated by ultrafiltration ( Amicon Ultra-15 MWCO 10K , Merck ) and subsequent solid phase extraction using a Sep-Pak C8 plus short cartridge ( Waters ) . NMR experiments were carried out using Bruker Avance III HD 500 and 800 spectrometers equipped with TCI cryoprobes at 25°C . The spectra were processed using the program NMRPipe ( Delaglio et al . , 1995 ) and analyzed using the program Sparky ( Goddard , T . D . and Kneller , D . G . University of California , San Francisco ) . For the main-chain resonance assignments of Sfr1N , 13C and 15N doubly labeled samples at concentrations of 0 . 15 ~ 0 . 20 mM in buffer N mixed with 5% D2O were prepared and placed in 5 mm symmetrical microtubes ( Shigemi ) . The 1H-15N HSQC spectrum ( Kay et al . , 1992; Grzesiek and Bax , 1993 ) was acquired at a 1H frequency of 800 MHz with a scan number of 8 , 1024 complex points and an acquisition time of 91 . 8 ms in the observed dimension , and 256 complex points and an acquisition time of 132 ms in the indirect dimension . The HNCA ( Grzesiek and Bax , 1992; Kay et al . , 1994 ) , HN ( CO ) CA ( Grzesiek and Bax , 1992; Kay et al . , 1994 ) , HNCACB ( Muhandiram and Kay , 1994 ) , HN ( CO ) CACB ( Yamazaki et al . , 1994 ) , HNCO ( Grzesiek and Bax , 1992; Kay et al . , 1994 ) , and HN ( CA ) CO ( Werner-Allen et al . , 2006 ) spectra were acquired at 800 MHz with a scan number of 8 . All spectra were acquired with 512 complex points and an acquisition time of 45 . 9 m in the observed dimension . In the 15N dimension , all spectra except HNCACB were acquired with 45 complex points and an acquisition time of 23 . 1 ms , while the HNCACB spectrum was measured with 47 complex points and an acquisition time of 24 . 1 ms . The experiments with 13Cα , 13Cα/13Cβ , and 13CO evolution were acquired with 64 , 128 , and 43 complex points and acquisition times of 13 . 3 , 10 . 6 , and 8 . 90 ms in the 13C dimension , respectively . To verify the signal assignments , the following samples were prepared and their 1H-15N HSQC spectra were obtained at 500 MHz; amino-acid selectively 15N-labeled ( Ala , Arg , Ile , Leu , Lys , or Phe ) Sfr1N ( 1-176 ) , Arg-selectively 15N-labeled R97A variant of Sfr1N ( 1-176 ) , Lys-selectively 15N-labeled K93A variant of Sfr1N ( 1-176 ) , and uniformly 15N-labeled Sfr1 ( 127-176 ) . From the 160 expected main-chain amide NH signals , 157 were detected ( 98% ) and assigned to specific residues in Sfr1N . The remaining three signals from Q2 , S3 , and H51 could not be assigned due to line-broadening . NMR data were deposited in the Biological Magnetic Resonance Bank ( BMRB ) repository with accession number 27885 . The secondary structural elements were analyzed by calculating deviations of the observed 13Cα , 13Cβ , and 13CO chemical shifts from their residue-dependent random coil values ( Wishart and Sykes , 1994; Wishart et al . , 1995 ) . Residues were deemed to form random coils if they displayed secondary chemical shift values within a limited range ( between −0 . 7 and 0 . 7 for 13Cα and 13Cβ atoms , and between −0 . 5 and 0 . 5 for 13CO atoms of non-proline residues , and −4 to 4 for all three 13C atoms of proline residues ) . The program TALOS+ ( Shen et al . , 2009 ) was also used to predict the secondary structural units where 1HN , 13Cα , 13Cβ , 13CO , and 15NH chemical shifts were used as input data . Predictions of disorder and protein binding sites for Sfr1 orthologs were generated by DISOPRED3 ( Jones and Cozzetto , 2015 ) . The heteronuclear {1H}-15N NOE experiment ( Kay et al . , 1989; Farrow et al . , 1994 ) was carried out for uniformly 15N-labeled Sfr1N at 800 MHz . The NOE values were determined from the ratio INOE/Iref , where INOE and Iref indicate the signal intensities in the spectra acquired with and without 3 s 1H presaturation , respectively . The NOE and reference spectra were acquired in an interleaved manner with a scan number of 32 , 1024 complex points and an acquisition time of 91 . 8 ms in the observed dimension , and 256 complex points and an acquisition time of 132 ms in the indirect dimension . 250 μL of uniformly 15N-labeled Sfr1N at a concentration of 0 . 1 mM in buffer N was mixed with a 0 . 1 mM Rad51 solution in buffer N at Sfr1N:Rad51 molar ratios of 1:0 , 1:0 . 25 , 1:0 . 5 , 1:0 . 75 , and 1:1 . These Sfr1N-Rad51 mixtures were concentrated to 250 μl ( Amicon Ultra-4 MWCO 10K , Merck ) and used for NMR measurements . For each mixture , 1H-15N HSQC spectra were acquired at 500 MHz with a scan number of 16 , 1024 complex points and an acquisition time of 146 ms in the observed dimension , and 256 complex points and an acquisition time of 126 ms in the indirect dimension . E . coli strains used in this study are listed in the Key Resources Table ( Supplementary file 1 ) . Experiments were performed essentially as described ( Miyazaki et al . , 2016 ) . Co-expression of Sfr1N ( residues 1–176 , with or without a specific amber codon ) and Rad51 was induced with 1 mM IPTG at an OD600 of ~0 . 35 in E . coli strain BL21 ( DE3 ) containing the pEVOL-pBpF plasmid either in the presence or absence of 1 mM pBPA . After 1 hr at 30 °C , cultures were pre-chilled on ice for 5 min before 250 µL of cells were spotted in a radial manner onto petri dishes at 4 °C and UV irradiated for 10 min at a distance of 4 cm with a B-100AP UV lamp ( 365 nm; UVP , LLC ) . 200 µL of cells was recovered from the plate and pelleted by centrifugation ( 20 , 000 g 5 min 4 °C ) . This pellet was dissolved in 70 µL of urea loading buffer ( Argunhan et al . , 2017b ) and subjected to SDS-PAGE followed by immunoblotting . Previously published protocols were followed to purify Rad51 ( Kurokawa et al . , 2008 ) , RPA ( Haruta et al . , 2006 ) , and Swi5-Sfr1 ( wild type ( Haruta et al . , 2006 ) , Sfr1N ( Kuwabara et al . , 2010 ) , and Swi5-Sfr1C ( Kuwabara et al . , 2010 ) . Swi5-Sfr1 mutants ( 3A , 4A , 7A ) were purified by the same method as wild-type Swi5-Sfr1 except they were diluted to 25 mM NaCl instead of 100 mM NaCl before being applied to the HiTrap Heparin column . For the partial purification of Rad55-Rad57 , Rad55 with a C-terminal dual tag ( 7xHis-Protein A ) was co-expressed with untagged Rad57 from the REP1 and REP2 plasmids , respectively , in a protease-deficient S . pombe strain ( BA1 ) . As a negative control , the same base strain was transformed with the REP1 and REP2 plasmids and treated exactly the same . After 20 hr in EMM without thymine , 400 mL of culture was harvested , washed with yeast wash solution ( 25 mM HEPES-KOH [pH 7 . 5] , 150 mM NaCl , 1 mM PMSF ) and stored as aliquots of 2 × 109 cells in −80 °C until required . An aliquot of cells was resuspended in 400 µL of yeast lysis buffer ( YLB; 50 mM HEPES-KOH [pH 7 . 5] , 500 mM NaCl , 5 mM Mg ( OAc ) 2 , 5% glycerol , 0 . 05% igepal CA630 , 1 mM ATP , 0 . 25 mM TCEP , 1 mM PMSF , 1x protease inhibitor cocktail [Roche] ) and lysis was performed with ~500 µm glass beads using a Yasui Kikai Multi-Beads shocker ( 12 cycles , 30 s on , 30 s off , 2700 rpm ) . Following sequential clarification ( 20 , 000 g 10 min 2 °C , 20 , 000 g 5 min 2 °C ) , the soluble cell extract was incubated with IgG-agarose resin ( Sigma ) for 3 hr at 4 °C . Resin-bound proteins were washed with YLB ( 500 µL x3 ) then buffer H200 ( 25 mM HEPES-KOH [7 . 5] , 200 mM NaCl , 5 mM Mg ( OAc ) 2 , 5% glycerol , 1 mM ATP , 0 . 25 mM TCEP; 500 µL x2 ) . The resin was then resuspended in 200 µL of H200 and Rad55-Rad57 was eluted by incubating with PreScission protease ( GE Healthcare ) , which cleaves between the 7xHis and Protein A tag on Rad55 . 3 . 75 µL or 7 . 50 µL of this IgG-agarose eluate was mixed with an equal volume of 2x SDS loading buffer and subjected to purity analysis by SDS-PAGE and staining with CBB ( Figure 7—figure supplement 1D ) . All proteins were free of contaminating nuclease , protease and ATPase activity for the duration of the relevant assays . In all assays where comparisons were made between reactions with and without protein , the equivalent volume of protein storage buffer was added instead of the protein . Strand exchange buffer ( 30 mM Tris-Cl [pH 7 . 5] , 100 mM NaCl , 20 mM KCl , 3 . 5 mM MgCl2 , 2 mM ATP , 1 mM DTT , 5% glycerol , 8 mM phosphocreatine , 8 units/mL creatine phosphokinase ) containing 10 µMnt PhiX174 virion DNA ( NEB ) was supplemented with 5 µM Rad51 and incubated at 37 °C for 10 min . The indicated concentration and variant of Swi5-Sfr1 was then added and the reaction was incubated for 10 min at 37 °C . Next , 1 µM of RPA was added and the reaction was incubated for 7 min at 37 °C . The reaction was initiated through the addition of 10 µMnt PhiX RF I DNA ( NEB ) linearized with ApaLI and incubated for a further 2 hr at 37°C . The 10 µL reactions were supplemented with 1 µL of psoralen ( 200 mg/mL ) and subjected to psoralen-UV crosslinking to capture labile DNA structures . 1 . 95 µL of stop solution was then added ( 30 mM Tris-Cl [pH 7 . 5] , 40 mM EDTA , 3% SDS , 5 mg/mL proteinase K ) . Following a 15 min incubation at 37°C , DNA was resolved in 1% agarose gels and stained with SYBR Gold ( Thermo Fisher Scientific ) . For the Affi-gel interaction assay ( Figure 5B ) , BSA or Rad51 was covalently attached to Affi-gel 15 ( 2 µg protein/µL gel ) according to the manufacturer’s instructions . 2 µg of the indicated Swi5-Sfr1 variant was diluted into 300 µL of Affi-gel buffer ( 25 mM HEPES-KOH [pH 7 . 5] , 150 mM NaCl , 3 . 5 mM MgCl2 , 0 . 5 mM DTT , 0 . 05 µg/µL BSA , 10% glycerol , 0 . 05% igepal CA-630 ) , input sample was taken , and 140 µL of this solution was mixed with 10 µL of Affi-BSA or Affi-Rad51 . Reactions were then incubated with gentle mixing ( 30 °C 30 min ) . Following a brief centrifugation , flow-through samples were taken and the resin was washed with Affi-gel buffer ( 200 µL , x2 ) . Bound proteins were eluted in 50 µL of SDS loading buffer with gentle mixing ( 37 °C 15 min ) . 9 µL ( input , flow-through ) or 3 µL ( eluate ) of sample was separated by SDS-PAGE , proteins were transferred to PVDF membranes and Sfr1 was detected with an anti-Sfr1 antibody ( Haruta et al . , 2006 ) . For the IP experiment in Figure 5—figure supplement 1C , 250 nM of a Swi5-Sfr1 variant and 250 nM of Rad51 were mixed on ice in 120 µL of IP buffer ( 30 mM Tris-Cl [pH 7 . 5] , 150 mM NaCl , 3 . 5 mM MgCl2 , 5% glycerol , 0 . 1% Igepal CA-630 ) . Input sample was taken and 100 µL of the solution was incubated at 30 °C for 15 min . Dynabeads Protein A ( ThermoFisher ) preincubated with anti-Rad51 antibody ( Haruta et al . , 2006 ) was added and mixtures were incubated with gentle mixing ( 4 °C 2 hr ) . Beads were washed with IP buffer ( 500 µL , x3 ) and bound proteins were eluted in 50 µL of SDS loading buffer with gentle mixing ( 65 °C 10 min ) . Proteins were separated by SDS-PAGE , transferred to PVDF membranes and detected with the indicated antibodies . For the IP experiments in Figures 7E , F and 50 µL of the IgG-agarose eluate from a strain transformed with empty vectors or Rad55-Rad57 containing vectors was mixed on-ice with 45 µL of buffer H200 ( 25 mM HEPES-KOH [7 . 5] , 200 mM NaCl , 5 mM Mg ( OAc ) 2 , 5% glycerol , 1 mM ATP , 0 . 25 mM TCEP ) and 5 µL of 20 µM Swi5-Sfr1 ( wild type or 7A ) . 5 µL of input sample was taken , diluted with 20 µL of milliQ and mixed with an equal volume of 2x SDS loading buffer . The remaining 95 µL of reaction was incubated at 30 °C for 10 min then 4 °C for 5 min . Dynabeads Protein A ( ThermoFisher ) preincubated with anti-Rad57 antibody ( Tsutsui et al . , 2001 ) was added and mixtures were incubated with gentle mixing ( 4 °C 1 hr ) . Beads were washed with buffer H200 ( 300 µL , x3 ) and bound proteins were eluted in 50 µL of SDS loading buffer with gentle mixing ( 65 °C 10 min ) . Either 1 µL ( Rad57 ) , 3 µL ( Rad55 ) or 5 µL ( Sfr1 and Rad51 ) of eluate was separated by SDS-PAGE , transferred to PVDF membranes and detected with the indicated antibodies . For the immunoblots of site-specific crosslinking experiments: anti-Rad51 ( Rb 1:10 , 000; provided by Hiroshi Iwasaki ) ; anti-Sfr1 ( Rb 1:5 , 000; provided by Hiroshi Iwasaki ) . For the immunoblots of cellular proteins: anti-MYC ( Rb 1:1 , 000; Sigma Aldrich C3956 ) , anti-Rad51 ( Rat 1:10 , 000; provided by Hiroshi Iwasaki ) , and anti-tubulin ( Mu 1:10 , 000; Sigma Aldrich T5168 ) . For the detection of Sfr1 in the Affi-gel assay: anti-Sfr1 ( Rb 1:5 , 000; provided by Hiroshi Iwasaki ) . For IP of Rad51 complexes , anti-Rad51 ( Rb; provided by Hiroshi Iwasaki ) was used , and for detection by immunoblotting: anti-Rad51 ( Rat 1:10 , 000; provided by Hiroshi Iwasaki ) , anti-Sfr1 ( Mu 1–5 and 76–80 , 1:1000 each; provided by Hiroshi Iwasaki ) , anti-Rad55 ( 1:5 , 000; provided by Hiroshi Iwasaki ) , and anti-Rad57 ( 1:5 , 000; provided by Hiroshi Iwasaki ) . For IP of Rad57 complexes , anti-Rad57 ( Rb; provided by Hiroshi Iwasaki ) antibody was used . Anisotropy buffer ( 30 mM HEPES-KOH [pH 7 . 5] , 100 mM KCl , 10 mM NaCl , 3 mM MgCl2 , 1 mM ATP , 1 mM DTT , 5% glycerol ) containing 1 . 5 µMnt of oligo dT ( 72 mer ) with a 5’ TAMRA label was supplemented with 0 . 5 µM Rad51 and incubated at 25°C for 5 min . Next , a Swi5-Sfr1 variant was added at the indicated concentration and the reaction was incubated at 25°C for a further 5 min . This solution was transferred into a 0 . 3 × 0 . 3 cm quartz cuvette and the fluorescence anisotropy was monitored once per second for 60 s ( 25°C , excitation 546 nm , emission 575 nm ) to confirm filament formation . Next , a 1 . 0 × 1 . 0 cm quartz cuvette containing 2 mL of anisotropy buffer was placed into the spectrofluorometer with constant stirring ( 450 r . p . m ) and , after 60 s of measurement , 50 µL of the solution containing Rad51 filaments with or without Swi5-Sfr1 was injected into this cuvette . Fluorescence anisotropy was then monitored once per second for the indicated time . Dissociation rate constants ( koff ) were calculated in KaleidaGraph . Cuvettes were purchased from Hellma Analytics . ATPase buffer ( 30 mM Tris-Cl [pH 7 . 5] , 100 mM KCl , 20 mM NaCl , 3 . 5 mM MgCl2 , 5% glycerol ) containing 10 µMnt PhiX virion DNA was mixed on ice with 5 µM Rad51 and 0 . 25 µM of a Swi5-Sfr1 variant . Reactions were initiated through the addition of 0 . 5 mM ATP . Time zero was immediately withdrawn ( 10 µL ) and mixed with 2 µL of 120 mM EDTA to terminate the reaction . Following incubation at 37°C , aliquots were withdrawn at the indicated timepoints and processed as above . Upon completion of the time course , aliquots were diluted two-fold with water to reduce the concentration of ATP to 0 . 25 mM . Inorganic phosphate generated by ATP hydrolysis was then detected using a commercial malachite green phosphate detection kit ( BioAssay Systems ) . DNA binding buffer ( 25 mM HEPES-KOH [pH 7 . 5] , 100 mM NaCl , 3 . 5 mM MgCl2 , 1 mM DTT , 5% glycerol ) containing 5 µMnt PhiX174 virion DNA or 5 µMnt ApaLI-linearized PhiX RF I DNA ( both NEB ) was supplemented with the indicated concentration of Swi5-Sfr1 ( wild type or mutants ) in a 10 µL reaction . Following incubation at 37°C for 15 mins , 2 µL of loading dye was added and 8 µL of the reaction mixture was separated by agarose electrophoresis ( 0 . 8% gel in TAE buffer , 2 . 5 hr 4°C ) . The gel was stained with SYBR Gold ( Thermo Fisher Scientific ) .	The DNA within cells contains the instructions necessary for life and it must be carefully maintained . DNA is constantly being damaged by radiation and other factors so cells have evolved an arsenal of mechanisms that repair this damage . An enzyme called Rad51 drives one such DNA repair process known as homologous recombination . A pair of regulatory proteins known as the Swi5-Sfr1 complex binds to Rad51 and activates it . The complex can be thought of as containing two modules with distinct roles: one comprising the first half of the Sfr1 protein and that is capable of binding to Rad51 , and a second consisting of the rest of Sfr1 bound to Swi5 , which is responsible for activating Rad51 . Here , Argunhan , Sakakura et al . used genetic and biochemical approaches to study how this first module , known as “Sfr1N” , interacts with Rad51 in a microbe known as fission yeast . The experiments showed that both modules of Swi5-Sfr1 were important for Rad51 to drive homologous recombination . Swi5-Sfr1 complexes carrying mutations in the region of Sfr1N that binds to Rad51 were unable to activate Rad51 in a test tube . However , fission yeast cells containing the same mutations were able to repair their DNA without problems . This was due to the presence of another pair of proteins known as the Rad55-Rad57 complex that also bound to Swi5-Sfr1 . The findings of Argunhan , Sakakura et al . suggest that the Swi5-Sfr1 and Rad55-Rad57 complexes work together to activate Rad51 . Many genetically inherited diseases and cancers have been linked to mutations in DNA repair proteins . The fundamental mechanisms of DNA repair are very similar from yeast to humans and other animals , therefore , understanding the details of DNA repair in yeast may ultimately benefit human health in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2020	Cooperative interactions facilitate stimulation of Rad51 by the Swi5-Sfr1 auxiliary factor complex
Munc13–1 acts as a master regulator of neurotransmitter release , mediating docking-priming of synaptic vesicles and diverse presynaptic plasticity processes . It is unclear how the functions of the multiple domains of Munc13–1 are coordinated . The crystal structure of a Munc13–1 fragment including its C1 , C2B and MUN domains ( C1C2BMUN ) reveals a 19 . 5 nm-long multi-helical structure with the C1 and C2B domains packed at one end . The similar orientations of the respective diacyglycerol- and Ca2+-binding sites of the C1 and C2B domains suggest that the two domains cooperate in plasma-membrane binding and that activation of Munc13–1 by Ca2+ and diacylglycerol during short-term presynaptic plasticity are closely interrelated . Electrophysiological experiments in mouse neurons support the functional importance of the domain interfaces observed in C1C2BMUN . The structure imposes key constraints for models of neurotransmitter release and suggests that Munc13–1 bridges the vesicle and plasma membranes from the periphery of the membrane-membrane interface . The release of neurotransmitters by synaptic vesicle exocytosis is critical for neuronal communication . This exquisitely regulated process involves several steps , including tethering of synaptic vesicles to specialized sites of the presynaptic plasma membrane called active zones , a priming step ( s ) that leaves the vesicles ready to release , and Ca2+-triggered membrane fusion ( Südhof , 2013 ) . Each of these steps can be modulated in a variety of presynaptic plasticity processes that underlie multiple forms of information processing in the brain ( Regehr , 2012 ) . The protein machinery that controls release ( Rizo and Xu , 2015; Jahn and Fasshauer , 2012; Südhof and Rothman , 2009 ) includes the neuronal soluble N-ethylmaleimide-sensitive factor attachment protein receptors ( SNAREs ) synaptobrevin , syntaxin-1 and SNAP-25 , which play a key role in membrane fusion by forming a tight four-helix bundle ( the SNARE complex ) that brings the vesicle and plasma membranes into close proximity ( Söllner et al . , 1993; Hanson et al . , 1997; Poirier et al . , 1998; Sutton et al . , 1998 ) . This complex is disassembled by N-ethylmaleimide-sensitive factor ( NSF ) and soluble NSF attachment proteins ( SNAPs; no relation to SNAP-25 ) ( Söllner et al . , 1993 ) to recycle the SNAREs for another round of fusion ( Mayer et al . , 1996; Banerjee et al . , 1996 ) . The Sec1/Munc18 protein Munc18–1 and the large ( 200 kDa ) active zone proteins called Munc13s are also crucial for release . Munc18–1 binds to a self-inhibited ‘closed’ conformation of syntaxin-1 ( Dulubova et al . , 1999; Misura et al . , 2000 ) and orchestrates SNARE complex assembly in an NSF-SNAP-resistant manner together with Munc13 , which helps to open syntaxin-1 ( Richmond et al . , 2001; Ma et al . , 2011 , 2013 ) . Because of their multiple functions , Munc13s have emerged as particularly central regulators of neurotransmitter release that link the core membrane fusion apparatus to diverse forms of presynaptic plasticity through their multidomain architecture ( illustrated in Figure 1A for Munc13–1 , the most abundant mammalian isoform ) . Thus , neurotransmitter release is completely abrogated in the absence of Munc13s ( Augustin et al . , 1999; Richmond et al . , 1999; Aravamudan et al . , 1999; Varoqueaux et al . , 2002 ) . This phenotype most likely arises because Munc13s play key roles in docking and priming ( Varoqueaux et al . , 2002; Weimer et al . , 2006; Hammarlund et al . , 2007; Imig et al . , 2014 ) that are associated at least in part to their activity in opening syntaxin-1 and thus stimulating SNARE complex formation through their MUN domain ( Richmond et al . , 2001; Basu et al . , 2005; Yang et al . , 2015 ) . However , the C1-C2B region and the C-terminal C2C domain are also critical for release , which may arise because these domains help bridge the vesicle and plasma membranes ( Liu et al . , 2016 ) . There is also evidence for a role of Munc13s in events downstream of priming [e . g . ( Hammarlund et al . , 2007; Shin et al . , 2010; Liu et al . , 2016 ) . Moreover , Munc13s are involved in multiple presynaptic plasticity processes , including isoform-specific depression and augmentation ( Rosenmund et al . , 2002 ) , diacyglycerol ( DAG ) -phorbol ester-dependent potentiation of release via the C1 domain ( Rhee et al . , 2002 ) and Ca2+-dependent short-term plasticity through the C2B domain and a calmodulin-binding region ( Shin et al . , 2010; Junge et al . , 2004 ) . Binding of the Munc13–1 C2A domain to RIMs also provides a connection to RIM-dependent forms of short- and long-term presynaptic plasticity ( Betz et al . , 2001; Dulubova et al . , 2005 ) . The crucial physiological importance of these modulatory processes was emphatically illustrated by the finding that knock-in mice bearing a point mutation in the Munc13–1 C1 domain that disrupts phorbol-ester-dependent potentiation of release die 2–3 hr after birth even though the mutation causes no impairment of evoked release ( Rhee et al . , 2002 ) . 10 . 7554/eLife . 22567 . 003Figure 1 . Crystal structure of Munc13–1 C1C2BMUN . ( A ) Domain diagram of rat Munc13–1 , with A-D corresponding to the four subdomains of the MUN domain . CaMb = calmodulin-binding sequence . ( B ) Ribbon representation of the structure of Munc13–1 C1C2BMUN color-coded as in the domain diagram . Helices are numbered and labeled . The position of the NF sequence involved in opening syntaxin-1 is indicated . The Zn2+ ions bound to the C1 domain are shown as yellow spheres . ( C ) Superimposition of the structures of the Munc13–1 isolated C1 domain ( PDB code 1Y8F ) , Ca2+-bound C2B domain ( PDB code 3 KWU ) and our refined structure of the almost complete MUN domain with the structures of these domains in the crystal structure of C1C2BMUN [r . m . s . d . between equivalent Cα atoms are 1 . 15 ( 45 Cα atoms ) , 0 . 35 ( 85 Cα atoms ) and 0 . 96 ( 465 Cα atoms ) , respectively] . Ca2+ ions of the C2B domain structure are shown as green spheres . D-F . Interfaces of helix H1 with the linker helices and with the MUN domain ( D ) , of the C1 domain with the C2B domain ( E ) and of the C2B domain with the MUN domain ( F ) . Selected sides chains in the interfaces , including those that were mutated , are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 00310 . 7554/eLife . 22567 . 004Figure 1—figure supplement 1 . Zn2+ ions in the C1 domain of C1C2BMUN . Superimposed on the atoms is the anomalous difference map colored in magenta mesh and contoured at the 3σ level . The map was calculated from data with resolution limits of 49 . 9–6 . 0 Å , collected at the zinc K-absorption edge . ( A ) C1 domain of chain A . ( B ) C1 domain of chain B . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 00410 . 7554/eLife . 22567 . 005Figure 1—figure supplement 2 . Comparison of chains A ( green ) and B ( red ) of C1C2BMUN . Superposition of 809 C-α carbon atoms resulted in an r . m . s . d . of 1 . 51 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 00510 . 7554/eLife . 22567 . 006Figure 1—figure supplement 3 . Lattice contacts for C1C2BMUN domains , C2 symmetry . Chain A is shown in green , chain B in cyan . In all panels , monomers in the foreground were removed for clarity . ( A , B ) Shown in dark grey are residues 1255–1517 of chain A . The view shown in B is rotated 90 degrees and zoomed in from that shown in A . ( C , D ) Shown in dark grey are residues 1255–1514 of chain B . The view shown in D is rotated 90 degrees and zoomed in from that shown in C . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 00610 . 7554/eLife . 22567 . 007Figure 1—figure supplement 4 . Plot of mean atomic displacement parameters ( B-factors ) versus residue number for the C1C2BMUN fragment . Chain A is shown in blue and chain B in green . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 00710 . 7554/eLife . 22567 . 008Figure 1—figure supplement 5 . Omit maps for domain interfaces where mutations were made . ( A ) Superimposed on the refined coordinates for the C1C2BMUN structure is the omit map ( mFo-DFc , green ) contoured at the 3σ level . Residues A748-A752 ( Ser748 , Asp749 , Arg750 , Ile751 , Lys752 ) were omitted from the C1C2BMUN structure in order to calculate the map . Shown in blue is the 2mFo-DFc map ( blue ) contoured at the 1σ level . ( B ) Superimposed on the refined coordinates for the C1C2BMUN structure is the omit map ( mFo-DFc , green ) contoured at the 3σ level . Residues Trp795 , Glu885 and Asn940 of chain A were omitted from the C1C2BMUN structure in order to calculate the map . Shown in blue is the 2mFo-DFc map ( blue ) contoured at the 1σ level . ( C ) Superimposed on the refined coordinates for the C1C2BMUN structure is the omit map ( mFo-DFc , green ) contoured at the 3σ level . Residues Val549 , Tyr550 , Lys551 , Lys552 , Thr 553 and Leu554 of chain A were omitted from the C1C2BMUN structure in order to calculate the map . Shown in blue is the 2mFo-DFc map ( blue ) contoured at the 1σ level . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 008 Despite these advances and the availability of three-dimensional structures for most of the Munc13 domains except C2C and part of the MUN domain ( Shen et al . , 2005; Dulubova et al . , 2005; Rodríguez-Castañeda et al . , 2010; Shin et al . , 2010; Li et al . , 2011; Yang et al . , 2015 ) , it is still unclear how the functions of these domains are related and coordinated , in part because no structure of a fragment containing multiple Munc13 domains has been described . Here , we report the crystal structure of a Munc13–1 fragment spanning its C1 , C2B and MUN domains ( C1C2BMUN ) , revealing a long , 195 Å rod formed by 26 α-helices that packs at one end against the C1 and C2B domains . The DAG-binding region of the C1 domain and the Ca2+-binding region of the C2B domain are near each other and point in the same direction , which is expected to facilitate cooperation between the two domains in membrane binding and thus enable synergy between the effects of DAG and Ca2+ in neurotransmitter release during repetitive stimulation . Electrophysiological experiments show that mutations designed to disrupt interfaces between different domains of C1C2BMUN impair evoked release and vesicle priming to different extents , and also have differential effects on Ca2+-dependent short-term plasticity as well as phorbol ester-induced potentiation . These results suggest that the highly elongated nature of C1C2BMUN and the relative disposition of its domains in our crystal structure are critical for the normal functions of Munc13–1 in neurotransmitter release and presynaptic plasticity , placing important structural constraints on possible models for the mechanisms of neurotransmitter release and presynaptic plasticity . The work presented herein culminates 12 years of efforts dedicated to determine the three-dimensional structures of fragments encompassing part of or the entire highly conserved C-terminal region of Munc13–1 , which includes the C1 , C2B , MUN and C2C domains ( Figure 1A ) . While crystals were obtained for several of the fragments that we prepared , they tended to diffract poorly . Key for the success in obtaining crystals of C1C2BMUN of sufficient quality to diffract in the 3–3 . 5 Å range were the choice of N- and C-termini ( residues 529 and 1531 , respectively ) , as well as the removal of residues 1408–1452 , which correspond to a long loop within the MUN domain that is poorly conserved and is subject to alternative splicing ( Brose et al . , 1995 ) . Removal of this loop generally increases the solubility of C-terminal Munc13–1 fragments ( Ma et al . , 2011; Li et al . , 2011 ) . Even with the best diffraction data obtained with C1C2BMUN , structure determination was hindered by low resolution , significant anisotropy , non-isomorphism and an inability to obtain selenomethionyl-derivatized protein . These problems were overcome by the use of single wavelength anomalous dispersion phases obtained from a dataset collected at the tantalum LIII edge on native crystals soaked with a tantalum bromide cluster , coupled with molecular replacement phases obtained from the previously determined structures of the C1 domain ( Shen et al . , 2005 ) , the C2B domain ( Shin et al . , 2010 ) and the nearly complete MUN domain of Munc13–1 . For this purpose , it was necessary to first re-refine the structure of the nearly complete MUN domain ( Yang et al . , 2015 ) using the deposited structure factors ( PDB code 4Y21 ) to reduce the level of side chain outliers and of steric clashes , as well as to correct the sequence numbering ( see Materials and methods ) . Placement of the C1 domains in the cell of the C1C2BMUN crystals ( Figure 1—figure supplement 1 ) was verified by an anomalous difference map calculated from data collected on native crystals at the zinc K-edge energy . Data collection and refinement statistics for the final structure of C1C2BMUN , as well as for the re-refined structure of the nearly complete MUN domain , are described in Table 1 . 10 . 7554/eLife . 22567 . 009Table 1 . Data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 009Data collectionCrystalTa LIII-edge peakZn K-edge peakMUN domainNativeSpace groupC2C2P21212C2Cell constants ( Å , ° ) 171 . 70 Å , 82 . 93 Å , 201 . 59 Å , 90 . 0° , 115 . 32° , 90 . 0°174 . 72 Å , 84 . 55 Å , 202 . 10 Å , 90 . 0° , 115 . 11° , 90 . 0°114 . 1 Å , 270 . 9 Å , 47 . 7 Å , 90 . 0° , 90 . 0° , 90 . 0°176 . 13 Å , 86 . 34 Å , 202 . 13 Å , 90 . 0° , 115 . 54° , 90 . 0°Wavelength ( Å ) 1 . 254891 . 282180 . 9790 . 97931Resolution range ( Å ) 42 . 73–4 . 50 ( 4 . 64–4 . 50 ) 49 . 81–4 . 00 ( 4 . 07–4 . 00 ) 39 . 02–2 . 90 ( 2 . 97–2 . 90 ) 45 . 60–3 . 35 ( 3 . 41–3 . 35 ) Unique reflections12 , 839 ( 620 ) 21 , 415 ( 961 ) 33 , 802 ( 2 , 770 ) 37 , 636 ( 1 , 363 ) Multiplicity3 . 6 ( 2 . 8 ) 3 . 8 ( 2 . 5 ) 5 . 4 ( 5 . 1 ) 3 . 9 ( 3 . 3 ) Data completeness ( % ) 93 . 1 ( 64 . 4 ) 94 . 3 ( 86 . 4 ) 99 . 3 ( 98 . 7 ) 93 . 9 ( 68 . 9 ) Rmerge ( % ) †3 . 1 ( 40 . 0 ) 10 . 1 ( 100 . 0 ) 6 . 9 ( 63 . 4 ) 5 . 6 ( 69 . 2 ) Rpim ( % ) ‡1 . 8 ( 25 . 7 ) 5 . 7 ( 76 . 5 ) NA3 . 1 ( 38 . 8 ) CC1/2 ( outermost resolution shell ) 0 . 9740 . 490NA0 . 809I/σ ( I ) 25 . 0 ( 0 . 8 ) 30 . 0 ( 2 . 1 ) 27 . 3 ( 1 . 95 ) 20 . 7 ( 1 . 3 ) Wilson B-value ( Å2 ) 85 . 4104 . 7Wilson B-value , sharpened ( Å2 ) §32 . 6209 . 2NA49 . 6Refinement statisticsResolution range ( Å ) 39 . 02–2 . 90 ( 2 . 98–2 . 90 ) 45 . 60–3 . 35 ( 3 . 46–3 . 35 ) No . of reflections Rwork/Rfree33 , 801/1 , 712 ( 2 , 634/136 ) 29 , 935/1 , 490 ( 752/34 ) Data completeness ( % ) 99 . 2 ( 98 . 0 ) 75 . 3 ( 22 . 0 ) Atoms ( non-H protein/Zn2+/Cl− ) 4286/NA/NA13 , 597/4/2Rwork ( % ) 22 . 8 ( 34 . 5 ) 25 . 4 ( 32 . 7 ) Rfree ( % ) 25 . 3 ( 35 . 2 ) 29 . 0 ( 44 . 5 ) R . m . s . d . bond length ( Å ) 0 . 0020 . 003R . m . s . d . bond angle ( ° ) 0 . 4990 . 610Mean B-value ( Å2 ) ( chain A/chain B/ Zn2+/Cl− ) 101 . 6/NA/NA78 . 8/53 . 2/44 . 4/8 . 2Ramachandran plot ( % ) ( favored/additional/disallowed ) #95 . 3/3 . 9/0 . 892 . 1/6 . 7/1 . 2Clashscore/Overall score#2 . 57/1 . 373 . 51/1 . 62Maximum likelihood coordinate error0 . 380 . 41Missing residues933–941 , 1041–1049 , 1524–1531A: 529–540 , 704–707 , 759–773 , 801–807 , 821–823 , 923–928 , 1038–1052 , 1191–1196 , 1342–1352 , 1404–1469 , 1518–1531 . B: 529–541 , 626-630 , 703–708 , 743–745 , 759–774 , 802–806 , 820–824 , 925–929 , 1038–1050 , 1338–1352 , 1405–1467 , 1516–1531 . Data for the outermost shell are given in parentheses . Bijvoet-pairs were kept separate for data processing . †Rmerge=100∑h∑i\|Ih , i−⟨Ih⟩\|/∑h∑i⟨Ih , i⟩ , where the outer sum ( h ) is over the unique reflections and the inner sum ( i ) is over the set of independent observations of each unique reflection . ‡Rpim=100∑h∑i[1/ ( nh−1 ) 1/2]\|Ih , i−⟨Ih⟩\|/∑h∑i⟨Ih , i⟩ , where nh is the number of observations of reflections h . §B-factor sharpening was performed in the autocorrection mode of HKL-3000 ( Borek et al . , 2013 ) . #As defined by the validation suite MolProbity ( Chen et al . , 2010 ) . The structures of the two molecules of C1C2BMUN present in the asymmetric unit of its crystals are very similar ( Figure 1—figure supplement 2 ) . The structure of C1C2BMUN can be viewed as a highly elongated rod spanning ca . 195 Å and is formed mostly of α-helical bundles , with a total of 26 helices; the C1 and C2B domains pack at the N-terminal end of the rod ( Figure 1B ) . Much of the long rod is formed by the MUN domain , which is homologous to subunits of complexes that mediate tethering in diverse membrane compartments ( Pei et al . , 2009; Yu and Hughson , 2010 ) and was previously shown to be formed by four subdomains of about five helices each ( named A-D and colored in blue , green , yellow and orange on Figure 1 ) ( Li et al . , 2011; Yang et al . , 2015 ) . Comparison of our C1C2BMUN structure with the crystal structure of the almost complete MUN domain ( with the N-terminus at residue 933 ) , which was described in Yang et al . ( 2015 ) and we re-refined , shows that the MUN domain is very similar in both structures ( Figure 1C ) , as expected because this structure was used in crystallographic phase determination by the molecular replacement method . However , the structure of C1C2BMUN shows that the MUN domain actually starts at residue 828 and contains five additional helices ( helices H6-H10 ) , resulting in a total of seven helices for subdomain A ( Figure 1C ) . In addition , there are five more helices preceding the MUN domain ( helices H1-H5; Figure 1B ) . Four of these helices correspond to the linker sequence between the C1 and C2B domains , whereas one helix ( H1 ) is spanned by a sequence preceding the C1 domain ( colored in red ) . These observations nicely explain a few findings made during our crystallization efforts . For instance , Munc13–1 fragments starting right at the beginning of the C1 domain were unstable and we were not able to express fragments containing only the C1 domain , the C2B domain and the linker between them in soluble form . As the structure of C1C2BMUN now reveals , there are extensive contacts between helix H1 , the four helices of the linker , the MUN domain , the C1 domain and the C2B domain ( Figure 1B , D–F ) , and hence it is not surprising that elimination of some of these interfaces impairs proper folding . We also note that the structures of the C1 and C2B domains in C1C2BMUN are similar to those determined for the isolated domains ( Shen et al . , 2005; Shin et al . , 2010 ) , although the Ca2+-binding loops are not visible in the C2B domain of C1C2BMUN ( Figure 1C ) . This correlates with the previous finding that these loops were observable in the crystal structure of the C2B domain bound to Ca2+ but not in its crystal structure without Ca2+ ( Shin et al . , 2010 ) , as our C1C2BMUN structure was determined in the absence of Ca2+ . The architecture of C1C2BMUN has important implications to understand the functions of Munc13–1 in docking and priming , as well as the regulatory roles of its various domains . First , the elongated structure and the overall packing of the different domains at the N-terminal end of the structure suggest that C1C2BMUN may function as a rigid or semi-rigid unit that bridges the synaptic vesicle and plasma membranes . Note in this context that the similarity in the structures of the two molecules from the asymmetric unit of the C1C2BMUN crystals ( Figure 1—figure supplement 2 ) and between these structures and that of the nearly complete MUN domain ( Figure 1C ) suggest that the overall architecture of C1C2BMUN has limited flexibility . Second , a model of the structure of C1C2BMUN incorporating the Ca2+-binding loops of the C2B domain , which mediate its interactions with PIP2-containing membranes in a Ca2+-dependent manner ( Shin et al . , 2010 ) , shows that these loops are proximal to the DAG-phorbol ester-binding region of the C1 domain ( Figure 2A , B ) . This arrangement is expected to promote cooperation between the C1 and C2B domains in membrane binding , thus suggesting a natural mechanism for synergy between increases in DAG and intracellular Ca2+ concentrations to enhance release probability upon repetitive stimulation . However , it is noteworthy that there are abundant basic residues in the C1-C2B region ( Figure 2A ) that could potentially mediate membrane binding in alternative orientations in the absence of DAG and Ca2+-bound to the C2B domain ( e . g . Figure 2C ) . These observations suggest that increased DAG and intracellular Ca2+ concentrations may alter release probability by promoting a different orientation of the Munc13–1 C-terminal region that is more efficient in promoting priming and/or fusion ( see Discussion and models described therein ) . Third , the finding that helix H1 is packed against the MUN domain , with the N-terminus pointing toward the center of the long rod ( Figure 1B ) , suggests that N-terminal sequences of Munc13–1 not included in this structure could perform their regulatory functions by influencing MUN domain activity . For instance , the calmodulin-binding region of Munc13–1 might inhibit release by binding to the middle of the MUN domain , where an NF sequence ( Figure 1B ) is known to be critical for opening syntaxin-1 ( Yang et al . , 2015 ) , and binding of Ca2+-calmodulin to this region may enhance release by relieving this inhibition . Interactions of homodimerized Munc13–1 C2A domain ( Lu et al . , 2006 ) with the MUN domain might also underlie the inhibition caused by homodimerization , which is relieved by RIM binding ( Deng et al . , 2011 ) . 10 . 7554/eLife . 22567 . 010Figure 2 . Distribution of basic residues clusters in C1C2BMUN . ( A ) Ribbon diagram of C1C2BMUN where the C1 and C2B domains were replaced with the structures of the isolated C1 domain ( cyan; PDB code 1Y8F ) and the isolated Ca2+-bound C2B domain ( salmon; PDB code 3 KWU ) ( Shin et al . , 2010 ) to include basic residues that are not observed in the structure of C1C2BMUN , likely because they are disordered . Arginine and lysine side chains are shown as blue spheres . The Zn2+ ions in the C1 domain are shown as yellow spheres and two Ca2+ ions bound to the C2B domain are shown as green spheres . Note that the two Ca2+ ions are not observed in the C1C2BMUN structure , which was crystallized in the absence of Ca2+ . ( B ) Close up of the structure of C1C2BMUN as shown in A , but including a phorbol ester drawn as brown spheres . The position of the phorbol ester is based on superimposing the C1 domain of C1C2BMUN with the phorbol-ester bound structure of the PKC-δ C1B domain ( PDB code 1PTR ) ( Zhang et al . , 1995 ) . The dashed line indicates the expected approximate location of the plasma membrane bound to the C1 domain through DAG and to the Ca2+-bound C2B domain through PIP2 . ( C ) Close up of the same region of C1C2BMUN but shown in another orientation with a region containing multiple basic side chains ( most from the C1 and C2B domains ) at the bottom . The Ca2+-binding sites of the C2B domain are shown in gray to symbolize that the sites are not occupied . The dashed line indicates a potential localization of the plasma membrane resulting from binding of this basic face to PS , which is sufficient for binding of C1C2BMUN to liposomes even in the absence of Ca2+ , DAG and PIP2 ( Liu et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 010 A key question that arises from the crystal structure of C1C2BMUN is whether the arrangement of the different domains at the N-terminal end of the structure is caused by crystal packing or reflects the native structure of Munc13–1 in neurons , and hence is important for its functions . To address this question , we designed three mutations to strongly disrupt domain interfaces observed in the C1C2BMUN structure: ( i ) a V549E , L554E double mutation designed to disrupt the hydrophobic packing of helix H1 against the linker region and against the MUN domain ( Figure 1D ) ; ( ii ) an R750E , K752E double mutation to disrupt salt bridges formed between the C1 and C2B domains ( Figure 1E ) ; and ( iii ) an N940W mutation designed to perturb the interface between the C2B and MUN domains ( Figure 1F ) . We tested the physiological consequences of these mutations by performing rescue experiments in autaptic hippocampal neuronal cultures from Munc13-1/2 double KO mice ( Varoqueaux et al . , 2002 ) that express WT or mutant Munc13–1s through a lentiviral vector . To dissect which aspects of neurotransmitter release and its regulation might be affected by the mutations , we evaluated spontaneous release , Ca2+-dependent release triggered by a single action potential ( AP ) , vesicle priming , high-frequency stimulation and PDBu-induced facilitation . Western blot analyses showed that the WT and mutant proteins were detectable at comparable levels ( Figure 3—figure supplement 1 ) . Although we cannot completely rule out the possibility that the differences observed in our electrophysiological measurements arise in part from distinct expression levels , this possibility is unlikely because no significant changes in synaptic properties are observed in heterozygous Munc13–1 ( +/- ) neurons and WT neurons mildly overexpressing WT munc13–1 in a Munc13–2 KO background ( MC and CR , unpublished results ) . We first assessed the effects on spontaneous release measuring the frequency and the amplitude of miniature excitatory postsynaptic currents ( mEPSCs ) . Only the V549E , L554E mutation in the N-terminal H1 helix showed an increase in mEPSC frequency , whereas the mEPSC charge and amplitude were unchanged ( Figure 3 ) . We next recorded excitatory postsynaptic currents ( EPSCs ) induced by a single AP to characterize evoked release . Analysis of the EPSC amplitudes revealed that the R750E , K752E mutation in the C1-C2B interface and the N940W mutation in the C2B-MUN interface impaired Ca2+-triggered release ( Figure 4A , B ) . However , expression of the V549E , L554E mutant where helix H1 is perturbed fully rescued evoked release . 10 . 7554/eLife . 22567 . 011Figure 3 . Effect on spontaneous release in synapses expressing Munc13–1 mutants that disrupt the C1C2BMUN domain interfaces . ( A ) Representatives traces of spontaneous release on synapses from Munc13-1/2 DKO rescued with the respective Munc13–1 WT and mutants indicated above . ( B ) Plots of mEPSC frequency , charge and amplitudes of Munc13–1 mutants normalized to corresponding Munc13–1 WT . Numbers in plots are n values for each group . Error bars represent SEM . Significance and p values were determined by comparison with the corresponding WT using the unpaired Student's t test: Mann-Whitney . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 01110 . 7554/eLife . 22567 . 012Figure 3—figure supplement 1 . Detection of protein expression from Munc13-1/2 DKO hippocampal neurons . Sample immunoblot of Munc13-1/2 DKO neurons rescued with Munc13–1 WT or Munc13–1 point mutants indicated in the top . Signal at 250 kDa corresponds to the expected Munc13–1-flag , and the signal at around 30 kDa indicates the cleaved product NLS-GFP . Lane 1 shows the lack of expression of the protein Munc13–1-flag or NLS-GFP in untransfected Munc13-1/2 DKO neurons as a negative control . Lanes 2–5 show the protein expression of all Munc13–1 proteins used . 30 µg of proteins from the lysates were used . Molecular weights ( kDa ) are indicated on the left side . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 01210 . 7554/eLife . 22567 . 013Figure 4 . Single action potential evoked synaptic transmission , ready releasable pool and release probability consequences by the disruption of the C1C2BMUN domain interfaces . ( A ) Representatives traces of AP-evoked EPSC amplitudes recorded form Munc13-1/2 DKO and DKO neurons rescued with Munc13–1 WT and mutants as indicated above . ( B ) Plot of AP-evoked EPSC amplitudes of Munc13–1 disrupting C1C2BMUN domain interfaces mutants normalized to corresponding Munc13–1 WT . ( C ) Representative traces of synaptic responses induced by 500 mM sucrose from Munc13-1/2 DKO neurons rescued with the respective Munc13–1 WT and mutants indicated above . ( D ) Plot of RRP charge of Munc13–1 mutants normalized to corresponding Munc13–1 WT data . ( E ) Plot of pvr for WT mutant Munc13–1s . ( F ) Plot showing the average paired-pulse ratios calculated from 2 AP-evoked EPSC amplitudes with a interstimulus interval of 25 ms . ( G ) Correlation between pvr and paired-pulse ratios from DKO neurons rescued with Munc13–1 WT and mutants . Numbers in plots are n values for each group . Error bars represent SEM . Significance and p values were determined by comparison with the corresponding WT using the unpaired Student's t test: Mann-Whitney . *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 013 To dissect putative changes in the parameters that underlie evoked release , we first quantified the size of the readily releasable pool ( RRP ) by measuring the responses induced by hypertonic solution ( Rosenmund and Stevens , 1996 ) . We found that the N940W mutation in the C2B-MUN interface caused a decrease in RRP size of close to 40% , similar to the change seen in the evoked response . The H1 helix mutation V549E , L554E decreased the RRP size by nearly 50% ( Figure 4C , D ) , which contrasts with the lack of an effect of this mutation on evoked release . Furthermore , the disruption of the C1-C2B interface by the R750E , K752E mutation led to normal vesicle priming despite reducing the evoked response . These experiments demonstrate that all three mutations impact Munc13–1 function but , interestingly , each mutation displays distinct effects on evoked release and vesicle priming . This finding strongly suggests that the domain interfaces differentially regulate the efficiency of the vesicle fusion process , which may lead to differences in vesicle release probability . We calculated the vesicular release probability ( Pvr ) by dividing the EPSC charge by the charge of the RRP . The V549E , L554E mutation in the H1 helix increased the vesicular release probability significantly , as expected , while the small differences in Pvr observed for the R750E , K752E and N940W mutants with respect to WT were not statistically significant ( Figure 4E ) . The impact on release probability in these mutants was further corroborated by analyzing the degree of facilitation or depression quantified by the paired pulse ratio of two consecutive AP-induced EPSC amplitudes ( Figure 4F , G ) . Compared to WT , the V549E , L554E mutant showed more depression , in correlation with our finding of enhanced release probability , while the other two mutants showed similar paired-pulse behavior to WT rescues , in agreement with the conclusion that these mutations did not disrupt basal release probability substantially . In general , vesicular release probability is also a major determinant for release during sustained AP trains . Initial high vesicular release probability predicts more depression of synaptic responses due to more prominent depletion of primed vesicles . However , structure function studies on Munc13 have identified changes in sustained release that are not explained by the initial release probability , and this has been interpreted as Munc13 playing an additional role in synaptic augmentation ( Rosenmund et al . , 2002 ) and in modifying vesicle replenishment through activity-dependent vesicle re-priming ( Shin et al . , 2010 ) . To test how disruption of the interfaces in Munc13–1 can affect the activity-dependent vesicle re-priming , we monitored synaptic responses during 10 Hz AP trains ( Figure 5 ) . We first defined the function of steady-state depression versus release probability in WT neurons by applying AP trains at various external Ca2+ concentrations , and found as expected a robust correlation between release probability and steady state depression ( Figure 5B ) . Interestingly , the V549E , L554E mutation in the H1 helix and the R750E , K752E mutation in the C1-C2B interface caused more pronounced depression of sustained release than expected from the initial release probability , while disrupting the C2B-MUN interface with the N940W mutation caused less depression than expected from the initial release probability ( Figure 5B , C ) . These results suggest that each one of the interfaces disrupted by the mutations plays a role in activity-dependent vesicle re-priming , and reinforce the notion that Munc13 exerts multiple functions in shaping release properties at synapses . 10 . 7554/eLife . 22567 . 014Figure 5 . Difference in sustained release during a high-frequency action potential train upon disruption of C1C2BMUN domain interfaces . ( A ) Exemplary traces of EPSCs evoked by 10 Hz stimulation trains of Munc13-1/2 DKO neurons rescued with WT and mutant Munc13–1s . ( B ) Correlation between the amount of steady-state EPSC amplitudes at the end of the 10 Hz train and vesicular release probability . Dotted gray curve provides the correlation of steady-state depression and the release probability by applying AP trains at various external Ca2+ concentrations in WT neurons . ( C ) Analysis of 50 EPSC amplitudes evoked at 10 Hz , which were normalized to the first EPSCs and plotted over time . WT control ( black circles ) is identical in all three groups . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 014 The regulatory function of Munc13 can also be probed by applying phorbol esters that act as exogenous agonists of the C1 domain and increase the vesicle release probability . To test how the mutations in the domain interfaces of Munc13–1 alter the coupling of phorbol-ester binding to changes in release probability , we examined the effects of acute application of 1 µM PDBu , which causes a 60% increase in the evoked postsynaptic responses in WT neurons ( Figure 6 ) . The N940W mutation in the C2B-MUN interface showed more pronounced potentiation than WT Munc13–1 ( Figure 6B ) , which correlates with the reduction in steady-state depression during 10 Hz AP trains observed for this mutant ( Figure 5C ) and indicates an easier transition for Munc13–1 to a potentiated state . Conversely , the V549E , L554E in helix H1 led to reduced potentiation by PDBu compared to WT , which likely arises because of the initial high vesicular release probability observed for this mutant ( Figure 4E ) . Remarkably , the R750E , K752E mutation in the C1-C2B interface caused an even more pronounced decrease in PDBu potentiation ( Figure 6B ) , which contrasts with the limited effect of this mutation on vesicular release probability ( Figure 4E ) but correlates with the strong depression observed in the 10 Hz AP trains ( Figure 5C ) . These findings strongly support the notion that the interactions between the C1 and C2B domain observed in our crystal structure are critical for the interplay between the two domains in Ca2+- and DAG-dependent presynaptic plasticity . 10 . 7554/eLife . 22567 . 015Figure 6 . Effect of mutations that disrupt the C1C2BMUN domain interfaces on the potentiation of release caused by the activation of the C1 domain by phorbol ester . ( A ) Exemplary EPSC traces ( dark colors ) from WT and mutant groups and their corresponding EPSCs after PDBu application ( light colors ) . ( B ) Potentiation of EPSC amplitudes by 1 μM PDBu evoked at 0 . 2 Hz in Munc13-1/2 DKO neurons expressing WT or mutants . The relative PDBu potentiation was calculated by normalizing the EPSC amplitude in PDBu with the presiding EPSC recorded in control extracellular solution . The solid symbols represent the normalized EPSC values in each time point ( ± SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 015 Because we observed changes in basal release probability and/or RRP size in the three mutants , we next examined how these release parameters are modulated during short-term plasticity induced by action potential trains . We recorded EPSC and sucrose-evoked responses before and 2 s after a 10 Hz action potential train to determine which of those parameters are individually affected ( Figure 7 ) . For rescues with WT Munc13–1 and the three mutants , the RRP size after the train tended to decrease compared to the RRP size before the train , but the difference was not significant ( Figure 7B ) . On the other hand , the EPSC amplitude increased for the WT rescue after the train ( Figure 7D , F ) , thus leading to an increased release probability with respect to the Pvr observed before the train ( Figure 7E ) . These results indicate that in the WT rescues the train does not have a substantial effect on the RRP , but the efficiency of evoked release increases due to some alteration ( s ) of the release machinery . 10 . 7554/eLife . 22567 . 016Figure 7 . Effect of mutations that disrupt the C1C2BMUN domain interfaces on the potentiation of release caused by high-frequency stimulation . ( A ) Schematic diagram illustrating experimental design and example traces of synaptic responses induced by 500 mM sucrose from Munc13-1/2 DKO rescued with Munc13–1 N940W mutant before and after a 10 Hz action potential train . ( B ) Normalized summary plot of RRP charge of Munc13-1/2 DKO neurons rescued with the respective Munc13–1 WT and mutants indicated below . The RRP charges from the different groups were normalized to corresponding RRP charge recorded 1 min before the 10 Hz train ( dashed red line ) . ( C ) Schematic diagram illustrating experimental design and example traces of EPSCs from Munc13-1/2 DKO rescued with Munc13–1 N940W mutant before and after a 10 Hz action potential train . ( D ) Normalized plot of AP-evoked EPSC amplitudes of Munc13-1/2 DKO neurons rescued with the respective Munc13–1 WT and mutants indicated below . The EPSC amplitudes were normalized to the corresponding EPSC recorded before the 10 Hz train ( doted red line ) . ( E ) Plot of the estimated Pvr of Munc13-1/2 DKO neurons rescued with the respective Munc13–1 WT and mutants indicated below normalized to the corresponding Pvr before high-frequency stimulation ( preHFS ) ( doted red line ) . ( F ) Normalized EPSC amplitudes of Munc13-1/2 DKO neurons rescued with the respective Munc13–1 WT and mutants indicated in each graph , in response to a low-frequency stimulus train ( 0 . 2 Hz ) that is interrupted by a 5 s 10 Hz stimulus train , as outlined in panel ( C ) . Numbers in plots are n values for each group . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 016 In rescues with the V549E , L554E mutation in helix H1 , comparison of RRP size and release probability before and after a 10 Hz train showed a failure to potentiate the RRP size , as observed in the WT rescue , but also a failure to increase the vesicular release probability ( Figure 7B , E ) . These results contrast with the impairment of vesicle priming and increment of vesicular release probability observed for the V549E , L554E mutant in naive synapses ( Figure 4D , E ) . Thus , the enhanced depression during the AP train observed for this mutant compared to WT Munc13–1 ( Figure 5C ) was due to a failure to potentiate release probability and not due to a failure in vesicle replenishment . Similar results were obtained for the rescue with Munc13–1 bearing the R750E , K752E mutation in the C1-C2B interface ( Figure 7B , D–F ) , which in naïve synapses caused no impairment of RRP size and vesicular release probability ( Figure 4D , E ) , but an enhanced depression of EPSC responses during AP trains ( Figure 5C ) . Thus , this enhanced depression was not due to a failure in vesicle replenishment but to a failure to potentiate release probability , as observed for the V549E , L554E mutant . Finally , the N940W mutation in the C2B-MUN interface , which leads to a 40% reduction in RRP size ( Figure 4D ) but no change in release probability ( Figure 4E ) in naive synapses , caused a selective increase in vesicular release probability after the AP train , similar to what is seen in WT rescues but with larger degree of potentiation ( Figure 7B , D–F ) . Thus , the finding that this mutant exhibits much less depression than WT during the AP train ( Figure 5C ) arises because this mutant is more likely to potentiate release probability during AP trains than WT Munc13–1 . Overall , our electrophysiological data provide compelling evidence for functional and structural interactions among the domains of the Munc13–1 C1C2BMUN fragment . These interactions play distinct roles in regulating vesicle priming , vesicle release probability and activity-dependent changes in release probability , thus having different impact on the ability of synapses to dynamically respond to incoming AP patterns . Our reconstitution experiments , which monitor fusion between synaptobrevin liposomes ( V-liposomes ) and syntaxin-1-SNAP-25 liposomes ( T-liposomes ) in the presence of Munc18–1 , NSF , αSNAP and a Munc13–1 fragment spanning its C1 , C2B , MUN and C2C domains ( C1C2BMUNC2C ) , recapitulate multiple features of neurotransmitter release , in particular the absolute requirement of Munc18–1 and Munc13–1 for membrane fusion ( Liu et al . , 2016 ) . Hence , we used these experiments to examine the effects of disrupting Munc13–1 domain interfaces on membrane fusion in vitro and compare them to those observed in our electrophysiological studies . Unfortunately , multiple attempts to express C1C2BMUNC2C bearing the V549E , L554E mutation failed to yield sufficient amounts of soluble protein , and we were also unable to obtain C1C2BMUNC2C bearing a single V549E or L554E mutation , suggesting that both substitutions strongly destabilize the fragment . Since the V549E , L554E mutant was expressed at levels comparable to WT Munc13–1 in our rescue experiments , it is plausible that this mutation is less destabilizing in the context of these experiments because of interactions of that region with N-terminal sequences of Munc13–1 , with other components of the release machinery , or with molecular chaperones . In our reconstitution experiments , we use an assay that simultaneously measures lipid mixing and content mixing ( Zucchi and Zick , 2011; Liu et al . , 2016 ) to ensure that true membrane fusion is observed , as lipid mixing can occur without content mixing [e . g . ( Chan et al . , 2009; Zick and Wickner , 2014 ) ] . We start the experiments in the absence of Ca2+ , and Ca2+ is added after 300 s to examine its effect on liposome fusion . Under the standard conditions of our experiments , in which we use 500 nM C1C2BMUNC2C and we include DAG and PIP2 in the T-liposomes , fusion is highly efficient upon Ca2+ addition and the Ca2+ dependence arises from Ca2+ binding to the C2B domain ( Liu et al . , 2016 ) . In initial experiments , we observed only mild effects of the R750E , K752E and N940W mutations in Munc13–1 C1C2BMUNC2C on fusion . To allow better discrimination , we lowered the C1C2BMUNC2C concentration to 100 nM , and we compared results obtained with T-liposomes that contained DAG+PIP2 , DAG only , or PIP2 only ( Figure 8 ) . WT C1C2BMUNC2C was still highly active under all these conditions , supporting Ca2+-dependent membrane fusion that strictly required Munc13–1 C1C2BMUNC2C and was optimal when the T-liposomes included both DAG and PIP2 ( Figure 8 and Figure 8—figure supplement 1 ) . When the T-liposomes contained DAG and PIP2 , the N940W mutation in the C2B-MUN interface slightly impaired fusion while the R750E , K752E mutation in the C1-C2B interface led to a considerably stronger impairment ( Figure 8A , B; see Figure 8—figure supplement 2 for quantification ) . This impairment became more overt with T-liposomes containing only DAG or only PIP2 , while the effects of the N940W mutation remained mild ( Figure 8C–F ) . 10 . 7554/eLife . 22567 . 017Figure 8 . Effects of mutations that disrupt C1C2BMUN interfaces on membrane fusion in reconstitution assays . Lipid mixing ( A , C , E ) between V- and T-liposomes was measured from the fluorescence de-quenching of Marina Blue-labeled lipids and content mixing ( B , D , F ) was monitored from the development of FRET between PhycoE-Biotin trapped in the T-liposomes and Cy5-Streptavidin trapped in the V-liposomes . The assays were performed in the presence of Munc18–1 , NSF-αSNAP and WT or mutant Munc13–1 C1C2BMUNC2C fragments as indicated . Experiments were started in the presence of 100 μM EGTA and 5 μM streptavidin , and Ca2+ ( 600 μM ) was added after 300 s . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 01710 . 7554/eLife . 22567 . 018Figure 8—figure supplement 1 . Dependence of liposome fusion on Munc13–1 C1C2BMUNC2C , DAG and PIP2 . Lipid mixing ( A ) between V- and T-liposomes was measured from the fluorescence de-quenching of Marina Blue-labeled lipids and content mixing ( B ) was monitored from the development of FRET between PhycoE-Biotin trapped in the T-liposomes and Cy5-Streptavidin trapped in the V-liposomes . The T-liposomes contained DAG and PIP2 , or DAG , or PIP2 , as indicated . The assays were performed in the presence of Munc18–1 , NSF-αSNAP and Munc13–1 C1C2BMUNC2C , except for a control lacking Munc13–1 C1C2BMUNC2C to show its requirement for membrane fusion . Experiments were started in the presence of 100 μM EGTA and 5 μM streptavidin , and Ca2+ ( 600 μM ) was added after 300 s . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 01810 . 7554/eLife . 22567 . 019Figure 8—figure supplement 2 . Quantification of the lipid and content mixing experiments of Figure 8 . Panels ( A–F ) correspond to panels ( A–F ) of Figure 8 , respectively . Bars represent averages of the normalized fluorescence observed after 500 s ( 200 s after Ca2+ addition ) in experiments performed at least in triplicate . Error bars represent standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 01910 . 7554/eLife . 22567 . 020Figure 8—figure supplement 3 . Effects of mutations that disrupt C1C2BMUN interfaces on membrane fusion in reconstitution assays including synaptotagmin-1 . Lipid mixing ( A ) between V-liposomes containing synaptotagmin-1 and T-liposomes was measured from the fluorescence de-quenching of Marina Blue-labeled lipids and content mixing ( B ) was monitored from the development of FRET between PhycoE-Biotin trapped in the T-liposomes and Cy5-Streptavidin trapped in the V-liposomes . The assays were performed in the presence of Munc18–1 , NSF-αSNAP and WT or mutant Munc13–1 C1C2BMUNC2C fragments as indicated . Experiments were started in the presence of 100 μM EGTA and 5 μM streptavidin , and Ca2+ ( 600 μM ) was added after 300 s . ( C–D ) Quantification of the lipid and content mixing experiments of panels ( A–B ) . Bars represent averages of the normalized fluorescence observed after 500 s ( 200 s after Ca2+ addition ) in experiments performed at least in triplicate . Error bars represent standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 020 To test whether the effects of the R750E , K752E mutation in these experiments are indeed due to disruption of the C1-C2B interface , we made a ‘reversal’ quadruple mutant where we mutated two of the acidic residues from the C1 domain in its interface with the C2B domain ( Figure 1E ) to basic residues ( R750E , K752E , E582R , E584K ) . These two additional substitutions partially compensated for the effects of the R750E , K752E mutation , as the quadruple mutation recovered some of the activity lost in the R750E , K752E mutant ( Figure 8 ) . The recovery was modest , but this result is not unexpected because the orientation of the mutated side chains in the quadruple mutant may not be as favorable to form salt bridges as in the WT protein . Hence , the observed functional recovery in the quadruple mutant supports the notion that the effects of the R750E , K752E mutation in the fusion assays are due to disruption of C1-C2B interactions . We note that all these reconstitution experiments were performed in the absence of synaptotagmin-1 because this Ca2+ sensor does not have a marked influence in the results obtained in these bulk assays under the conditions used , although it may cause an acceleration of Ca2+-dependent fusion that can only be detected at faster time scales than those used in our measurements ( Liu et al . , 2016 ) . Correspondingly , the relative activities of the WT and mutant C1C2BMUNC2C Munc13–1 fragments in reconstitution experiments that incorporated synaptotagmin-1 into the V-liposomes yielded similar results ( Figure 8—figure supplement 3 ) to those obtained in the absence of synaptotagmin-1 ( Figure 8A , B ) . The impairment of fusion caused by the R750E , K752E in our reconstitution assays correlates with the impairment of evoked release observed in the rescue experiments , as well as with the strong depression observed for this mutation in the 10 Hz trains and the impairment of PDBu-induced augmentation ( Figures 4–6 ) . Altogether , our results strongly support the notion that the C1-C2B interface observed in the crystal structure of C1C2BMUN is important for evoked release and for cooperation between the C1 and C2B domains in membrane binding during repetitive stimulation to generate an activated state that is stabilized by binding of the C1 domain to DAG and of the C2B domain to Ca2+-PIP2 . Thus , as we noted previously ( Liu et al . , 2016 ) , our reconstitutions including DAG and PIP2 in the T-liposomes may recapitulate more closely this activated state than the normal state that leads to release evoked by a single action potential . Interpreting the results of the N940W mutation in the reconstitutions is hindered by the fact that this mutation impaired evoked and sucrose-induced release but had less depression than WT in the 10 Hz trains and a slightly higher potentiation by PDBu ( Figures 4–6 ) . The small impairment of fusion observed in the reconstitutions for this mutant may reflect a balance between the inhibitory and stimulatory effects observed in the rescue experiments but , overall , the mild nature of the effect on fusion appears to be more reminiscent of the small ( albeit contrary ) effects observed in the PDBu treatment , thus supporting also the notion that the reconstitutions emulate an activated state of the release machinery . Munc13–1 acts as a master regulator of neurotransmitter release , playing a crucial role in release itself and multiple functions in integrating the effects of diverse signals that alter release probability during presynaptic plasticity . Thus , elucidating how the functions of the multiple domains of Munc13–1 are coordinated to control release and plasticity constitutes a major , fundamental challenge in neuroscience . The crystal structure of Munc13–1 C1C2BMUN and the functional data presented here represent critical steps to meet this challenge , revealing a highly elongated structure that is key to understanding how Munc13–1 bridges the vesicle and plasma membranes to control SNARE complex formation , and how the C1 and C2B domains mediate presynaptic plasticity processes that depend on Ca2+ and DAG . Notable features of the structure of C1C2BMUN include its length , which is close to 20 nm and is thus comparable to the radius of a synaptic vesicle , and the interfaces between its different domains . The observation that the three mutations designed to disrupt domain interfaces observed in the structure have distinct effects on evoked release , vesicle priming , vesicle release probability , high-frequency stimulation and PDBu-induced facilitation demonstrates the functional relevance of the overall structure and shows that the different domain interfaces perturbed by the mutations play differential roles in release and short-term presynaptic plasticity . This conclusion is further supported by the observation that each of the three mutations uniquely alters the normal relation between vesicle release probability and steady-state EPSC amplitude in 10 Hz AP trains in neurons expressing WT Munc13–1 ( Figure 5B ) , a relation that arises from the natural increase in RRP depletion rate as vesicle release probability increases . Similarly , the three mutations also change the normal inverse relation between the extent of PDBu-induced facilitation and the vesicle release probability . This finding is not surprising , as the effects of PDBu are likely to be at least partially related to those observed in 10 Hz trains . Thus , phorbol ester activation of the Munc13 C1 domain is believed to mimic the effects of increased DAG levels during repetitive stimulation , which result from accumulation of intracellular Ca2+ and activation of PLCs ( Rhee et al . , 2002 ) . Accumulation of Ca2+ also activates Munc13s by binding to the C2B domain ( Shin et al . , 2010 ) . The finding that the DAG/phorbol ester-binding region of the C1 domain and the Ca2+-binding loops of the C2B domain are close to each other and point in the same direction in our C1C2BMUN structure ( Figure 2 ) suggest that Ca2+ , DAG and PIP2 can cooperate in inducing binding of Munc13–1 to the plasma membrane , indicating that these different forms of activation are closely interrelated . These conclusions are supported by the physiological effects of the R750E , K752E mutation in the C1-C2B interface . This mutation leads to a modest impairment of evoked release and to no significant change in the RRP , resulting also in no significant change in vesicular release probability ( Figure 4 ) . However , this mutation causes a considerably stronger depression during 10 Hz trains than observed for WT Munc13–1 and to severe impairment of PDBu-dependent facilitation ( Figures 5 and 6 ) . These results suggest that the interaction between the C1 and C2B domains observed in our crystal structure is important for evoked release but is particularly critical for activation of Munc13–1 by phorbol esters as well by Ca2+ and DAG during repetitive stimulation . Note also that this mutation strongly disrupts the ability to increase the release probability after repetitive stimulation observed for WT Munc13–1 ( Figure 7E ) . Because the C1 and C2B domains pack at the N-terminal end of the long helical structure , far from the middle region of the MUN domain that contains the NF residues involved in opening syntaxin-1 ( Yang et al . , 2015 ) ( Figure 1B ) , our structure does not support models whereby the C1 and C2B domains impair the activity of the MUN domain by direct intramolecular interactions , and binding to DAG or Ca2+ releases these inhibitory interactions [e . g . ( Rizo and Rosenmund , 2008 ) ] . Instead , our structure suggests that activation of Munc13–1 by PDBu or by repetitive stimulation involves cooperative binding of the C1 and C2B domains to the plasma membrane in a defined orientation that is promoted by increases in the levels of DAG and intracellular Ca2+; this activated state may be more effective in mediating priming and/or downstream events leading to fusion than the state existing under resting conditions , which may involve a different orientation of Munc13–1 with respect to the plasma membrane favored by Ca2+- and DAG-independent interactions involving multiple basic residues of the C1 and C2B domains ( Figure 2C; see also Figure 9 and discussion below ) . 10 . 7554/eLife . 22567 . 021Figure 9 . Models of neurotransmitter release inspired by the structure of C1C2BMUN and our functional data . ( A ) C1C2BMUNC2C is represented by the structure of C1C2BMUN , with the C2B domain replaced by the structure of the isolated Ca2+-bound C2B domain ( colored in salmon ) , and by a blue ellipse corresponding to the C2C domain . C1C2BMUNC2C is shown bridging the two membranes through interactions of the vesicle membrane with the C2C domain and the plasma membrane with the basic surface formed by the C1 and C2B domains ( left panel ) or with the DAG-binding region of the C1 domain and the Ca2+-binding region of the C2B domain ( right panel ) ( see Liu et al . , 2016 ) . Arginine and lysine side chains are shown as blue spheres . The C1 domain ribbon is in cyan and Zn2+ ions are shown as yellow spheres . The two Ca2+ ions bound to the C2B domain are shown as green spheres on the right; on the left , they are shown as gray spheres to represent that the sites are not occupied . The NF sequence involved in opening syntaxin-1 ( Yang et al . , 2015 ) is represented by orange spheres . The closed syntaxin-1-Munc18–1 complex ( PDB code 3C98 ) is shown to scale to allow comparison of its size with that of C1C2BMUN . The C-terminus of the syntaxin-1 SNARE motif and the cytoplasmic region of synaptobrevin are shown as dashed curved lines , and their transmembrane regions are represented by cylinders . B . C1C2BMUNC2C is represented as in A but , instead of being located between the two membranes , is bridging the two membranes from a peripheral location . The model must be visualized in three dimensions , such that the C2C domain is bound to the outer surface of the vesicle membrane rather than inserted into the vesicle lumen . On the right side , the C2C domain is in the front , whereas in the left side the C2C domain is in the back of the vesicle , which is represented by a semitransparent gray surface . In both panels , C1C2BMUNC2C is shown in the orientation proposed for the activated state induced during repetitive stimulation , where binding to the plasma membrane is mediated by the DAG-binding region of the C1 domain and the Ca2+-binding region of the C2B domain . The SNARE complex ( PDB code 1N7S ) ( Ernst and Brunger , 2003 ) , with the SNARE motifs in green for SNAP-25 , in red for synaptobrevin and in yellow for syntaxin-1 , is shown partially assembled at the top and fully assembled at the bottom . The arrows in the top panel are meant to illustrate that complete assembly of the C-terminus of the SNARE complex might pull the membranes radially outwards , which could create strong membrane tension to trigger membrane fusion . See text for further details . DOI: http://dx . doi . org/10 . 7554/eLife . 22567 . 021 The V549E , L554E mutation designed to disrupt the packing of helix H1 against the linker sequence and the MUN domain decreases the RRP but not the amplitude of the EPSCs induced by a single action potential , resulting in an increased vesicle release probability ( Figure 4 ) that mirrors an enhancement in spontaneous release ( Figure 3 ) . Correspondingly , this mutant exhibits more depression during 10 Hz trains than WT neurons , but the depression is stronger than expected from the observed vesicle release probability ( Figure 5B ) . The V549E , L554E mutation also leads to a strong impairment in PDBu-induced facilitation ( Figure 6 ) that likely arises from the increased Pvr . Overall , the effects of this mutation resemble those caused by the H567K mutation that is expected to unfold the C1 domain ( Rhee et al . , 2002; Basu et al . , 2007 ) . The decreases in the RRP observed for both the V549E , L554E and the H567K mutants may result from disruption of coupling between the C-terminal region and the N-terminus containing the C2A domain , which contributes to the docking-priming activity of Munc13–1 ( MC and CR , unpublished results ) . The increased release probability observed for both mutants suggests that the disruptive structural effects of the two mutations may mimic to some extent the changes involved in activating Munc13–1 during repetitive stimulation , which may involve in part the release of inhibitory effects caused by N-terminal sequences . However , note that the V549E , L554E mutation prevented the increase in Pvr observed after repetitive stimulation for WT Munc13–1 ( Figure 7E ) , indicating that different features govern this increase in Pvr and the release probability in naïve synapses . The N940W mutation in the C2B-MUN interface impairs evoked release and decreases the RRP , leading to a release probability similar to that of WT Munc13–1 ( Figure 4 ) . However , this mutant exhibits smaller depression during 10 Hz trains than WT Munc13–1 and a slightly larger PDBu-induced facilitation ( Figures 5 and 6 ) . These results suggest that the packing of the C2B-MUN interface observed in our structure is important for normal vesicle priming and release caused by a single action potential , but the alteration of this interface caused by the N940W mutation favors the activated state that is generated during repetitive stimulation . This view is reinforced by the finding that the N940W mutation causes a larger increase in release probability after repetitive stimulation than that observed with WT Munc13–1 ( Figure 7E ) . The neurotransmitter release machinery is highly complex , including several large proteins of hundreds of kDa that form the active zone in addition to the SNARES , Munc18–1 , Munc13–1 , NSF , αSNAP , synaptotagmin-1 and multiple additional components ( Südhof , 2013 ) . Without knowing how Munc13–1 interacts with other proteins , interpretation of the structure of Munc13–1 C1C2BMUN in terms of specific models of neurotransmitter release is necessarily speculative . Nevertheless , the structure poses critical constraints on working models of release and its regulation . For instance , the model of Figure 9A is based on the finding that the Munc13–1 C1C2BMUNC2C fragment bridges V-liposomes to T-liposomes in the absence of Ca2+ ( Liu et al . , 2016 ) . Such bridging might facilitate the activity of the MUN domain in opening syntaxin-1 to initiate SNARE complex formation and may involve interactions of the polybasic face of the C1-C2B region ( Figure 2C ) with the T-liposomes and of C2C with the V-liposomes . Binding of the C1 domain to DAG and of the C2B domain to Ca2+ during repetitive stimulation could change the orientation of Munc13–1 with respect to the plasma membrane ( Figure 2B ) in such a way that the two membranes are brought into closer proximity ( Figure 9A ) and the SNARE complex is formed more efficiently . Importantly , if Munc13–1 is located between the two membranes , progress toward complete SNARE complex assembly and fusion would require release of Munc13–1 from the fusion complex because otherwise its large size would hinder further membrane proximity . In this scenario , Munc13–1 could not underlie changes in release probability directly , but its enhanced activity could lead to an increase in the number of preassembled SNARE complexes that might underlie the increases in release probability caused by up-regulation of DAG , by PDBu treatment or by the K630W mutation in a Ca2+ binding loop of the Munc13–2 C2B domain ( Rhee et al . , 2002; Shin et al . , 2010 ) . However , it is difficult to explain with this type of model why the H567K mutation in the C1 domain does not alter the vesicle refilling rate under resting conditions while increasing the Pvr ( Rhee et al . , 2002 ) , and why PDBu changes the vesicle release probability without altering the RRP ( Basu et al . , 2007 ) . The distinct effects of Munc13 mutations on evoked release , RRP , vesicular release probability , synaptic responses to repetitive stimulation and PDBu-dependent facilitation [Figures 3–7 and ( Rhee et al . , 2002; Basu et al . , 2007; Shin et al . , 2010 ) are more readily explained by models postulating that Munc13–1 remains bound to the SNARE complex and/or other components of the release apparatus after SNARE complex assembly , forming part of the fusion complex . Such models predict at least two major actions for Munc13s , one in docking-priming ( i . e . orchestrating SNARE complex assembly ) and another in fusion . The length of the C1C2BMUN structure suggests that the Munc13–1 C-terminal region cannot remain in the central space between two membranes ( as in Figure 9A ) after the SNARE complex is even partially assembled . However , the activities of Munc13–1 in bridging the two membranes and mediating SNARE complex assembly , as well as potential downstream functions , could be performed with Munc13–1 located in the periphery of the membrane-membrane interface , with an arrangement where the large size of Munc13–1 does not hinder completion of SNARE complex assembly and membrane fusion ( e . g . Figure 9B ) . With such an arrangement , reorientation of Munc13–1 upon binding to DAG and Ca2+-PIP2 could still underlie an increased activity in facilitating SNARE complex formation , but such reorientation could also lead to an optimization of the primed state that can lead more readily to membrane fusion upon Ca2+ influx than that observed without Munc13–1 activation . This separation of the roles of Munc13–1 into at least two steps can easily explain the differential effects of the V549E , L554E , R750E , K752E and N940W mutations in the different functional assays , including the finding that N940W impairs priming and evoked release , but enhances Munc13–1 function during 10 Hz trains or PDBu treatment ( Figures 4–7 ) . Clearly , the primed fusion complex depicted in Figure 9B ( top ) might contain additional components such as Munc18–1 , αSNAP , NSF , synaptotagmin-1 and complexins , among others . An attractive feature of this model is that it places the SNAREs on the periphery of the membrane-membrane interface rather than between the membranes , consistent with electron microscopy images of SNARE-mediated liposome fusion intermediates ( Hernandez et al . , 2012 ) and of presynaptic terminals showing practically direct contact between docked-primed vesicles and the plasma membrane [e . g . ( Harlow et al . , 2001 ) ] . With this architecture , the SNARE complex could not induce fusion by bringing the vesicle and plasma membranes into proximity , as they are already in contact before Ca2+ influx , but could pull the two membranes radially outwards as assembly of the SNARE complex is completed ( Figure 9B ) . In this case , fusion would result from the tension created in both membranes , which would favor transient formation of non-bilayer intermediates that might or might not resemble a stalk . Components of the fusion complex that are bound to the SNARE complex and bridge the membranes ( e . g . Munc13–1 in Figure 9B ) would play an important role to establish support points from which the SNAREs can exert their force on the membranes more efficiently . As discussed in the results section , the strong effect of the R750E , K752E mutation in our fusion assays ( Figure 8 ) suggests that our reconstitutions recapitulate better the mechanism of release during repetitive stimulation than release evoked by a single action potential . This proposal is supported by the findings that the tight Ca2+ dependence of membrane fusion in our reconstitution assays ( Figure 8 ) arises largely from Ca2+ binding to the Munc13–1 C2B domain ( Liu et al . , 2016 ) , and that disrupting the Ca2+-binding sites of the Munc13–2 C2B domain did not impair evoked release in rescue experiments but did impair release during repetitive stimulation ( Shin et al . , 2010 ) . However , it is plausible that functional redundancy with other proteins and/or compensatory effects may have masked an actual role for the Munc13–2 C2B domain in Ca2+ sensing during evoked release , in cooperation with synaptotagmin-1 . A primed state that contains Munc13 as key component bodes well for such a Ca2+-sensing role of the C2B domain , which is also suggested by the increased release probability caused by a K630W mutation in the Ca2+-binding loops of the Munc13–2 C2B domain ( Shin et al . , 2010 ) and may be reflected in the tight Ca2+-dependence of our fusion assays . Clearly , further research is needed to test all these ideas , and it will be particularly critical to assess whether Munc13–1 is indeed part of the primed fusion complex and , if this is the case , how Munc13–1 binds to the SNAREs and other potential components of the complex . Regardless of these possibilities , the structure of C1C2BMUN presented here provides a fundamental framework to design and interpret future studies . To express rat Munc13–1 fragments encoding the C1C2BMUN and C1C2BMUNC2C regions ( residues 529–1531 and 529–1735 , respectively; both constructs have residues 1408–1452 from a flexible loop deleted ) , the corresponding DNA sequences were originated from full-length rat Munc13–1 ( NM_022861 , L756W , Δ1415–1437 , Δ1533–1551 , E1666G ) . C1C2BMUN was cloned into pFASTBacHTb ( EcoRI , HindIII ) ; C1C2BMUNC2C was cloned into pFASTBac ( EcoRI , HindIII ) , which was modified by adding a GST tag and a TEV cleavage site in front of the EcoRI cloning site . The constructs were used to generate a baculovirus using the Bac-to-Bac system ( Invitrogen; Waltham , MA ) . Insect cells ( sf9 ) were infected with the baculovirus , harvested about 72–96 hr post-infection , and re-suspended in lysis buffers C1C2BMUN ( 50 mM Tris pH8 . 0 , 250 mM NaCl , 10 mM imidazole ) ; C1C2BMUNC2C ( 50 mM Tris pH8 . 0 , 250 mM NaCl , 1 mM TCEP ) . Cells were lysed and centrifuged at 18 , 000 rpm for 45 min . The clear supernatant of C1C2BMUN was incubated with Ni-NTA resin at 4°C for 2 hr , then the beads were washed with: ( i ) lysis buffer; ( ii ) lysis buffer containing 1% Triton X-100; ( iii ) lysis buffer containing 1M NaCl; and ( iv ) lysis buffer . The protein was eluted in lysis buffer with 150 mM imidazole , incubated with TEV to remove the His-tag , and purified by ion exchange chromatography . The clear supernatant of C1C2BMUNC2C was incubated with GST agarose at room temperature for 2 hr . The beads were washed with: ( i ) lysis buffer; ( ii ) lysis buffer containing 1% Triton X-100; ( iii ) lysis buffer containing 1M NaCl; and ( iv ) lysis buffer . The protein was treated with TEV protease on the GST agarose at 22°C for 2 hr . Both C1C2BMUN and C1C2BMUNC2C were further purified via gel filtration chromatography and were concentrated to 3–4 mg/ml for storage in 10 mM Tris buffer ( pH 8 . 0 ) containing 10% glycerol , 5 mM TCEP and 250 mM NaCl . The C1C2BMUNC2C mutants were generated by site-directed mutagenesis , and purified as the WT fragment . The finding that the mutants eluted at the same volumes as WT C1C2BMUNC2C in gel filtration and did not exhibit a higher tendency to degradation strongly suggest that the mutations do not cause folding problems . Rat Munc13–1 C1C2BMUN ( 529-1407 , 1453-1531 ) in 0 . 01 M Tris ( pH 8 . 0 ) , 0 . 25 M NaCl , 10% ( v/v ) glycerol and 5 mM TCEP was concentrated to 4–6 mg/ml for crystallization using the sitting drop vapor diffusion method . Drops in a ratio of 1 μl protein to 1 μl well solution were equilibrated against 200 μl 0 . 1 M Tris ( pH 8 . 0–8 . 5 ) , 0 . 2 M LiCl , 12% ( v/v ) PEG 10 , 000 at 20°C . Multiple crystals appeared spontaneously in 3 days and were used for dilution microseeding . Drops in a ratio of 1 μl protein to 1–2 μl well solution premixed with crystal seeds were equilibrated against 200 μl 0 . 1 M Tris ( pH 8 . 0–8 . 5 ) , 0 . 2 M LiCl , 10–12% ( v/v ) PEG 10 , 000 at 20°C , and crystals were harvested within 10 days . Tantalum derivatized crystals were obtained by overnight incubation with solid Na2Ta6Br12 in mother liquor drops that contained pre-grown single C1C2BMUN crystals . Crystals were cryoprotected by successive transfer in increasing steps of 5% ethylene glycol to a final solution of 20–25% ( v/v ) ethylene glycol , 0 . 1 M Tris ( pH 8 . 0 ) , 0 . 2 M LiCl , 0 . 15 M NaCl , 10% ( v/v ) glycerol , 5 mM TCEP and flash-cooled in liquid nitrogen . Native C1C2BMUN crystals exhibited the symmetry of space group C2 with unit-cell parameters of a = 176 . 1 Å , b = 86 . 4 Å , c = 202 . 1 Å and β = 115 . 5° and contained two molecules of C1C2BMUN per asymmetric unit . C1C2BMUN crystals displayed strong anisotropy and were highly non-isomorphous . Native C1C2BMUN crystals diffracted to a dmin of 3 . 35 Å when exposed to synchrotron radiation . All diffraction data were collected at beamline 19-ID ( SBC-CAT ) at the Advanced Photon Source ( Argonne National Laboratory , Argonne , IL , USA ) and processed with HKL3000 ( Minor et al . , 2006 ) , with applied corrections for effects resulting from absorption in a crystal and for radiation damage ( Borek et al . , 2003; Otwinowski et al . , 2003 ) , the calculation of an optimal error model , and corrections to compensate the phasing signal for a radiation-induced increase of non-isomorphism within the crystal ( Borek et al . , 2010 , 2013 ) . These corrections were crucial for successful phasing . Initial phases for C1C2BMUN were obtained by molecular replacement ( MR ) with Phaser ( McCoy et al . , 2007 ) using the crystal structure of the N-terminally truncated MUN BCD domain ( PDB code: 4Y21 ) ( Yang et al . , 2015 ) with the coordinates for several loops and residues at the N- and C-termini removed ( residues 942–947 , 1038–1041 , 1193–1196 , 1342–1353 , 1404–1409 , 1452–1469 , 1515–1523 ) as the search model . One C2B domain was located by MR using the crystal structure of the calcium-free C2B domain ( PDB code: 3 KWT ) as a search model ( Shin et al . , 2010 ) . Phasing using MR versus the native dataset stalled at this point , and an iterative MR-SAD/rigid body refinement procedure was then adopted . MR-SAD phases calculated in Phaser from a tantalum bromide dataset collected at the tantalum LIII edge revealed interpretable density for missing helices H7-H9 , which were modeled initially with a polyalanine sequence and rigid body refined in Phenix ( Adams et al . , 2010 ) . The updated model was placed in the native cell via MR in Phaser and rigid-body refined in Phenix . Density modification using two-fold non-crystallographic symmetry was performed in Parrot ( Cowtan , 2010 ) , and automated model rebuilding of only the newly added four helices of each MUN domain was performed in Buccaneer ( Cowtan , 2006 ) . Multiple cycles of alternating this procedure between the tantalum and native datasets allowed the modeling and sequence assignment to helices H1-H9 , and the placement of the second C2B domain by MR . Subsequently , both C1 domains were located by MR using the NMR structure ( PDB code: 1Y8F ) as a search model ( Shen et al . , 2005 ) , and the locations were confirmed by an anomalous difference map calculated from a dataset collected at the zinc K-edge ( Figure 1—figure supplement 1 ) . Iterative model building and refinement were performed with COOT and Phenix , respectively ( Emsley and Cowtan , 2004 ) . Restraints used in the initial cycles of model refinement included non-crystallographic symmetry , secondary structure and reference models for the C1 , C2B and MUN domains . As the coordinates for the MUN domain deposited in the PDB ( ID 4Y21 ) had incorrect sequence numbering and exhibited a high level of side chain outliers , a high clashscore and high overall score in MolProbity ( Chen et al . , 2010 ) , the coordinates for the MUN domain were re-refined versus the deposited structure factors and the sequence numbering was corrected prior to use as a reference model for restrained refinement of the C1C2BMUN model . The reference model restraints were removed for the final cycles of refinement . A superposition of chains A and B yield a root mean square deviation ( r . m . s . d . ) of 1 . 51 Å for 809 aligned Cα carbons ( Figure 1—figure supplement 2 ) . The final model for C1C2BMUN ( Rwork = 25 . 4% , Rfree = 29 . 0% ) contained 1685 residues in two monomers , 4 Zn2+ and 2 Cl- ions . The higher-than-average Rfree value is probably due to the relative dearth of lattice contacts ( Figure 1—figure supplement 3 ) for the CD subdomains of MUN chain A of C1C2BMUN , as evidenced by weak electron density and high average thermal displacement factors ( ADP ) ( 151 . 2 Å2 ) for those subdomains ( residues 1255–1517 ) ( Figure 1—figure supplement 4 ) . Due to the high ADP values for the CD subdomain of chain A , the authors recommend that interpretation of the CD subdomain should be performed on residues in chain B . The density for the remaining domains of chain A as well as chain B of C1C2BMUN is strong and well connected; in fact , the average ADPs for residues 541–950 are lower for chain A ( 41 . 0 Å2 ) than chain B ( 62 . 4 Å2 ) . Omit maps for regions where site directed mutations were made are shown in Figure 1—figure supplement 5 . A Ramachandran plot generated with MolProbity ( Chen et al . , 2010 ) indicated that 92 . 6% of all protein residues are in the most favored regions and 1 . 2% in disallowed regions . The majority of the outliers in the Ramachandran plot are located in surface loops with weak electron density that connect domains or secondary structural elements . Data collection and structure refinement statistics are summarized in Table 1 . The coordinates of the C1C2BMUN structure and of the refined MUN domain structure have been deposited in the Protein Data Bank with accession numbers 5UE8 and 5UF7 , respectively . The cDNAs of Munc13–1 full length and Munc13–1 V549E , L554E , Munc13–1 R750E , K752E and Munc13–1 N940W were generated from rat Munc13–1 ( Basu et al . , 2005 ) by PCR amplification . The reverse primer harbors a 3xFLAG sequence ( Sigma-Aldrich , Hamburg , Germany ) to allow expression analysis . The corresponding PCR products were fused to a P2A linker ( Kim et al . , 2011 ) after a nuclear localized GFP sequence into the lentiviral shuttle vector , which allows a bicistronic expression of NLS-GFP and the Munc13–1-Flag protein under the control of a human synapsin-1 promotor . Concentrated lentiviral particles were prepared as described ( Lois et al . , 2002 ) . Animal welfare committees of Charité Medical University and the Berlin state government Agency for Health and Social Services approved all protocols for animal maintenance and experiments ( license no . T 0220/09 ) . Hippocampi were dissected from embryonic day 18 . 5 Munc13 1/2 DKO mouse and enzymatically treated with 25 units/ml of papain for 45 min at 37°C . After enzyme digestion , hippocampi were mechanically dissociated and the neuron suspension was plated onto astrocytes microislands at a final density of 300 cells cm−2 . Neurons were infected 24 hr after plating with the lentiviral rescue constructs and incubated at 37°C and 5% CO2 for 13–16 days . Whole-cell voltage clamp recordings were done at room temperature in 13–16 days in vitro ( DIV ) autaptic hippocampal Munc13- 1/2 DKO neurons expressing Munc13–1 WT , Munc13–1 V549E , L554E , Munc13–1 R750E , K752E or Munc13–1 N940W . Synaptic currents were monitored using a Multiclamp 700B amplifier ( Molecular Devices ) . The series resistance was compensated by 70% and only cells with series resistances <10 MΩ were analyzed . Data were acquired using Clampex 10 software ( Molecular Devices , Sunnyvale , CA ) at 10 kHz and filtered using a low-pass Bessel filter at 3 kHz . Borosilicate glass pipettes with a resistance between 2 and 3 . 5 MΩ were used . Pipettes were filled with internal recording solution contained the following ( in mM ) : 136 KCl , 17 . 8 HEPES , 1 EGTA , 4 . 6 MgCl2 , 4 Na2ATP , 0 . 3 Na2GTP , 12 creatine phosphate , and 50 U/ml phosphocreatine kinase; 300 mOsm; pH 7 . 4 . During recordings , neurons were continuously perfused with standard extracellular solution including the following ( in mM ) : 140 NaCl , 2 . 4 KCl , 10 HEPES , 10 glucose , 2 CaCl2 , 4 MgCl2; 300 mOsm; pH 7 . 4 . Spontaneous release was measured by recording miniature EPSCs for 30 s at −70 mV . To detect false-positive events 3 mM of kynurenic acid was applied for an equal amount of time . Action potential-evoked EPSCs were triggered by 2 ms somatic depolarization from −70 to 0 mV . The size of the readily-releasable pool ( RRP ) was determined by the application with a fast flow system of 500 mM sucrose added to the standard extracellular solution for 5 s . Evoked sucrose responses are characterized by a transient inward current followed by a steady state current . The steady state component represents refilling of primed vesicles and was used to define the baseline . The area under the baseline in the transient curve component was quantified to determine the total charge released by the RRP ( Rosenmund and Stevens , 1996 ) . The vesicular release probability ( pvr ) was calculated by dividing the EPSC charge by the RRP charge . The paired-pulse stimulation protocol contained two inductions of an AP at an interval of 25 ms ( 40 Hz ) . The paired-pulse ratio was calculated by dividing the amplitude of the second EPSC by the amplitude of the first . To analyze release induced by a high frequency stimulation train , EPSCs were evoked at a frequency of 10 Hz for 5 s . To define the quantitative relationship between Pvr and the steady state EPSC amplitudes during a 10 Hz action potential train , we recorded EPSCs and sucrose evoked responses from wildtype neurons . EPSCs and EPSC trains were recorded in external solutions containing 0 . 5 , 1 , 2 and 4 mM calcium and 4 mM magnesium . The hypertonic sucrose solution for all responses contained 2 mM calcium and 4 mM magnesium . The Pvr was computed and correlated with the mean amplitude of the last 10 EPSCs of the 10 Hz AP train . Phorbol ester experiments were monitored in the same cell , EPSC amplitudes were recorded at 0 . 2 Hz for 30 s in the absence of 1 µM PDBu and the following 30 s in its presence . To dissect how an AP train modulates synaptic output , we examined RRP and release probability separately by probing EPSC amplitude and RRP size 2 s following the 10 Hz train . We then compared the relative changes in RRP and EPSC to the corresponding EPSC and RRP values preceding the AP train . An approximately equal number of cells were recorded from control and experimental groups per day from 3 to 4 consecutive days . The present dataset was acquired from three separate cultures . To minimize variability between cultures the values of mEPSC , EPSC and RRP from each experimental groups of recording were normalized to the mean value of the WT group for each culture . Data were analyzed offline using Axograph X ( Axograph Scientific , Sidney , Australia ) and the values were normalized to the WT group . Data summation and statistical analyses were performed using Prism 7 ( GraphPad ) . Significance and p values were determined by comparison of each mutant group with the Munc13–1 WT using the unpaired Student's t test: Mann-Whitney . Hippocampal neurons from Munc13-1/2 DKO expressing Munc13–1 full length , Munc13–1 V549E , L554E , Munc13–1 R750E , K752E and Munc13–1 N940W mutants were lysed after 15 DIV at 4°C with RIPA lysis buffer including protease inhibitor cocktail-complete mini ( Roche Diagnostics , Berlin , Germany ) . Equal amounts of proteins from the lysates of the four different groups were mixed with Laemmli sample buffer containing 0 . 1 M DTT , and boiled 5 min at 99°C . Protein lysates were separated on SDS polyacrylamide gels ( 4–8%% SDS-PAGE ) and transferred to a polyvinyl difluoride ( PVDF ) membrane . Membranes were blocked for 1 hr with 5% skim milk in TBST and incubated at 4°C over night with primary antibodies: anti-Flag M2 ( F1804; Sigma-Aldrich ) , and anti-Living Colors GFP ( 632375; Clontech , Mountain View , CA ) . Secondary antibodies were horseradish peroxidase-conjugated ( Jackson ImmunoResearch , West Grove , PA ) . The immunoreactive proteins were detected by ECL Plus Western Blotting Detection Reagents ( GE Healthcare Biosciences , Pittsuburgh , PA ) in a Fusion FX7 detection system ( Vilber Lourmat , Eberhardzell , Germany ) . Data were collected from three separate Munc13-1/2 DKO cultures and analyzed offline using ImageJ .	The human brain contains billions of cells called neurons that communicate with each other using molecules called neurotransmitters . An electrical signal in one neuron triggers the release of neurotransmitters from the cell , which then activate or inhibit electrical signals in neighboring neurons . Inside the cell , neurotransmitters are stored in small bubble-like structures called synaptic vesicles . The vesicles fuse with the membrane that surrounds the cell to release the neurotransmitters . This process must be tightly controlled to ensure that neurotransmitters are released rapidly and at the right time . A protein called Munc13 is a key component of the machinery that regulates the fusion of synaptic vesicles . It helps the synaptic vesicle to dock onto the cell membrane and get ready for fusion . Munc13 is a large protein and contains several different regions , including three domains called C1 , C2B and MUN . These three domains control the release of neurotransmitters , but how they do so is poorly understood . Xu , Camacho et al . used a technique called X-ray crystallography to analyse the three-dimensional shape of the part of Munc13 that contains the three domains . The experiments reveal that the MUN domain forms a long rod-like shape with the C1 and C2B domains packed at one end . Several mutations that reduce the ability of the domains to interact with each other altered the release of neurotransmitters from mouse neurons to different extents . These findings suggest that the overall architecture of the region containing the C1 , C2B and MUN domains is important for the normal activity of Munc13 . The structure revealed by Xu , Camacho et al . sets a framework for understanding how Munc13 controls neurotransmitter release , and thus mediates diverse forms of information processing in the brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2017	Mechanistic insights into neurotransmitter release and presynaptic plasticity from the crystal structure of Munc13-1 C1C2BMUN
Recent epidemiological and laboratory-based studies suggest that the anti-diabetic drug metformin prevents cancer progression . How metformin diminishes tumor growth is not fully understood . In this study , we report that in human cancer cells , metformin inhibits mitochondrial complex I ( NADH dehydrogenase ) activity and cellular respiration . Metformin inhibited cellular proliferation in the presence of glucose , but induced cell death upon glucose deprivation , indicating that cancer cells rely exclusively on glycolysis for survival in the presence of metformin . Metformin also reduced hypoxic activation of hypoxia-inducible factor 1 ( HIF-1 ) . All of these effects of metformin were reversed when the metformin-resistant Saccharomyces cerevisiae NADH dehydrogenase NDI1 was overexpressed . In vivo , the administration of metformin to mice inhibited the growth of control human cancer cells but not those expressing NDI1 . Thus , we have demonstrated that metformin's inhibitory effects on cancer progression are cancer cell autonomous and depend on its ability to inhibit mitochondrial complex I . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 001 Metformin is widely used to treat patients with type II diabetes mellitus who have high levels of circulating insulin ( Nathan et al . , 2009 ) . Metformin suppresses liver gluconeogenesis thereby reducing glucose release from the liver ( Inzucchi et al . , 1998; Viollet et al . , 2012 ) . In several recent retrospective studies , investigators have observed an association between metformin use and diminished tumor progression in patients suffering from different types of cancers ( Evans et al . , 2005; Bowker et al . , 2006; Dowling et al . , 2012 ) . These data have prompted several prospective clinical trials to determine the efficacy of metformin as an anti-cancer agent . However , the underlying mechanism by which metformin diminishes tumor growth is not fully understood . There are two postulated mechanisms by which metformin reduces tumor growth . Metformin may act at the organismal level , reducing levels of circulating insulin , a known mitogen for cancer cells . Alternatively , metformin may act in a cancer cell autonomous manner . Metformin is known to inhibit mitochondrial complex I in vitro ( Ota et al . , 2009; El-Mir et al . , 2000; Owen et al . , 2000 ) and it is thus possible that this targeting of the electron transport chain could inhibit tumor cell growth ( Birsoy et al . , 2012 ) . This latter hypothesis has been questioned as cancer cells have the ability to survive on ATP produced exclusively by glycolysis . Furthermore , cancer cells have been shown to conduct glutamine-dependent reductive carboxylation to generate the TCA cycle intermediates required for cell proliferation when the electron transport chain is inhibited ( Mullen et al . , 2012; Fendt et al . , 2013 ) . Thus , it is not clear whether inhibition of complex I by metformin would result in decreasing tumor growth . In the present study , we directly tested whether inhibition of cancer cell mitochondrial complex I by metformin was required to decrease cell proliferation in vitro and tumor progression in vivo . Human HCT116 p53−/− colon cancer cells have previously shown to be sensitive to metformin ( Buzzai et al . , 2007 ) . To determine if metformin treatment inhibited cellular oxygen consumption in these cells , we treated HCT116 p53−/− cells with increasing concentrations of metformin in media containing the metabolic substrates glucose , pyruvate , and glutamine for 24 hr . Subsequently , we measured cellular oxygen consumption . Metformin inhibited cellular oxygen consumption of HCT 116 p53−/− cells at concentrations ( 0 . 25–1 . 0 mM ) similar to those reported to affect bioenergetics and gluconeogenesis in primary hepatocytes in vitro ( Figure 1A; Foretz et al . , 2010; Miller et al . , 2013 ) . 10 . 7554/eLife . 02242 . 003Figure 1 . Metformin inhibits mitochondrial complex I function . ( A ) Relative mitochondrial oxygen consumption rate ( OCR ) of intact Control-HCT 116 p53−/− and ( B ) NDI1-HCT 116 p53−/− cells treated with metformin in complete media for 24 hr . ( C ) Relative complex I ( 2 mM malate , 10 mM pyruvate , 10 mM ADP ) -driven oxygen consumption rate of saponin permeabilized Control-HCT 116 p53−/− cells and ( D ) NDI1-HCT 116 p53−/− cells treated with metformin for 20 min in mitochondrial assay buffer . ( E ) Relative complex II-driven oxygen consumption rate of saponin permeabilized Control-HCT 116 p53−/− cells treated with 10 mM succinate and 10 mM ADP in the presence of 1 mM metformin or the complex II inhibitor 3-Nitropropionic acid ( 3-NPA ) . ( F ) Representative western blot and quantification of levels of OCT1 protein in Control BFP-HCT 116 p53−/− and NDI1-HCT 116 p53−/− cells . Error bars are SEM ( OCR: n = 4; OCT1: n = 4 ) . * indicate significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 00310 . 7554/eLife . 02242 . 004Figure 1—figure supplement 1 . NDI1 expression slightly increases oxygen consumption . ( A ) Basal mitochondrial oxygen consumption rate of Control-HCT 116 p53−/− ( +BFP ) and NDI1-HCT 116 p53−/− cells normalized to Control-HCT 116 p53−/− cells . ( B ) Coupled mitochondrial oxygen consumption rate of Control-HCT 116 p53−/− and NDI1-HCT 116 p53−/− cells normalized to Control-HCT 116 p53−/− cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 004 To determine whether metformin's inhibition of cellular oxygen consumption depended on mitochondrial complex I , we stably overexpressed the Saccharomyces cerevisiae protein NDI1 in HCT 116 p53−/− cells ( hereon referred to as NDI1-HCT 116 p53−/− cells ) . NDI1 is a single-subunit NADH dehydrogenase , which oxidizes NADH in a process similar to the multi-subunit mammalian complex I; however without proton pumping or ROS generation ( Seo et al . , 1998 ) . By contrast , mammalian complex I contains 45 subunits that pumps protons and generates ROS . NDI1-HCT 116 p53−/− cells demonstrated a slight , non-significant elevation in basal cellular oxygen consumption compared to control cells and were completely resistant to the effects of metformin on cellular oxygen consumption ( Figure 1—figure supplement 1 , Figure 1B ) . To ensure that the inhibition of cellular oxygen consumption by metformin was a direct effect of metformin on complex I , we examined mitochondrial respiratory function in saponin-permeabilized cells . Saponin removes cholesterol from plasma membranes , allowing the entry of metabolic substrates directly to mitochondria ( Jamur and Oliver , 2010 ) . In the presence of ADP and the complex I substrates pyruvate and malate , metformin fully inhibited oxygen consumption in permeabilized Control-HCT 116 p53−/− cells ( Figure 1C ) . By contrast , metformin had no effect on pyruvate/malate-driven oxygen consumption in NDI1-HCT 116 p53−/− cells ( Figure 1D ) . Metformin also had no effect on oxygen consumption in saponin-permeabilized cells respiring on the complex II substrate succinate in the presence of ADP ( Figure 1E ) . Interestingly , in saponin-permeabilized cells , metformin significantly inhibited complex I-dependent respiration at a much lower concentration than that required to inhibit oxygen consumption of intact cells , suggesting that transport across the plasma membrane is a barrier to metformin's inhibition of complex I . Metformin is known to slowly accumulate in cells in which its uptake is mediated by organic cation transporters ( OCTs ) ( Emami Riedmaier et al . , 2013 ) . To ensure that NDI1-HCT 116 p53−/− cells are not refractory to metformin because of a change in metformin uptake , we analyzed the expression of OCT 1 in both control and NDI1-HCT 116 p53−/− cells . Expression of OCT1 protein did not change with the presence of NDI1 ( Figure 1F ) . We next sought to determine if metformin-dependent inhibition of complex I resulted in changes in proliferation and survival of HCT116 p53−/− cells . Metformin did not induce cell death in Control-HCT 116 p53−/− or NDI1-HCT 116 p53−/− cells in the presence of glucose ( Figure 2A , B ) , however , in the absence of glucose , metformin induced cell death in Control-HCT 116 p53−/− but not in NDI1-HCT 116 p53−/− cells ( Figure 2C , D ) . Metformin diminished cell proliferation in Control-HCT 116 p53−/− cells but not in NDI1-HCT 116 p53−/− cells in media containing glucose ( Figure 2E , F ) . 10 . 7554/eLife . 02242 . 005Figure 2 . Metformin decreases cell proliferation by inhibiting mitochondrial complex I . ( A ) Percentage of live Control-HCT 116 p53−/− or ( B ) NDI1-HCT 116 p53−/− treated with metformin for 72 hr in media containing 10 mM glucose . ( C ) Percentage of live Control-HCT116 p53−/− or ( D ) NDI1-HCT 116 p53−/− treated with metformin for 24 hr followed by glucose withdrawal for 16 hr . ( E ) Cell number of Control-HCT 116 p53−/− cells and ( F ) NDI1-HCT 116 p53−/− cells 24 , 48 , and 72 hr post treatment with 0 . 5 mM or 1 mM metformin in complete media . Error bars are SEM ( n = 4 ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 00510 . 7554/eLife . 02242 . 006Figure 2—figure supplement 1 . Metformin decreases cellular proliferation through inhibition of mitochondrial complex I function in HCT 116 p53+/+ cells . ( A ) Relative mitochondrial oxygen consumption rate of Control-HCT 116 p53+/+ cells and ( B ) NDI1-HCT 116 p53+/+ cells treated with metformin in complete media for 24 hr . ( C ) Cell number of Control-A549 cells and ( D ) NDI1-A549 cells 24 , 48 , and 72 hr post treatment with 0 . 5 mM or 1 mM metformin in complete media . Error bars are SEM ( Relative OCR n = 3 , Cell number n = 4 ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 00610 . 7554/eLife . 02242 . 007Figure 2—figure supplement 2 . Metformin decreases cellular proliferation through inhibition of mitochondrial complex I function in A549 cells . ( A ) Relative mitochondrial oxygen consumption rate of Control-A549 cells and ( B ) NDI1-A549 cells treated with metformin in complete media for 24 hr . ( C ) Cell number of Control-A549 cells and ( D ) NDI1-A549 cells 24 , 48 , and 72 hr post treatment with 0 . 5 mM or 1 mM metformin in complete media . Error bars are SEM ( Relative OCR n = 3 , Cell number n = 4 ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 007 These results indicate that the metformin-dependent inhibition of complex I decreases cell proliferation in the presence of glucose and increases cell death under glucose deprivation . These inhibitory effects of metformin were not specific to HCT116 p53−/− cells as metformin inhibited oxygen consumption and cellular proliferation of Control-HCT 116 p53+/+ cells and Control-A549 human lung cancer cells but not NDI1-HCT 116 p53+/+ or NDI1-A549 cells ( Figure 2—figure supplement 1 and 2 ) . Taken together , these results indicate that the anti-proliferative and cell death promoting effects of metformin require mitochondrial complex I inhibition . We also examined whether phenformin , a more lipophilic biguanide , also exert its anti-proliferative effects on cancer cells through inhibition of complex I . Phenformin inhibited oxygen consumption in Control-HCT 116 p53−/− cells and saponin-permeabilized Control HCT 116 p53−/− cells at 100-fold lower concentration compared to metformin ( Figure 3A , C ) . Expression of NDI1 rescued the phenformin-mediated decrease in oxygen consumption ( Figure 3B , D ) . Phenformin diminished cell proliferation in the control but not NDI1 expressing HCT116 p53−/− cells ( Figure 3E , F ) , and did not induce cell death in media containing glucose , similar to metformin ( Figure 3G , H ) . Collectively , these results indicate that phenformin also exerts its biological effects through inhibition of mitochondrial complex I . 10 . 7554/eLife . 02242 . 008Figure 3 . Phenformin decreases cell proliferation by inhibiting mitochondrial complex I . ( A ) Relative mitochondrial oxygen consumption rate ( OCR ) of intact Control-HCT 116 p53−/− and ( B ) NDI1-HCT 116 p53−/− cells treated with phenformin in complete media for 24 hr . ( C ) Relative complex I ( 2 mM malate , 10 mM pyruvate , 10 mM ADP ) -driven oxygen consumption rate of saponin permeabilized Control-HCT 116 p53−/− cells and ( D ) NDI1-HCT 116 p53−/− cells treated with phenformin for 20 min in mitochondrial assay buffer . ( E ) Cell number of Control-HCT 116 p53−/− cells and ( F ) NDI1-HCT 116 p53−/− cells 24 , 48 , and 72 hr post treatment with 0 or 5 µM phenformin in complete media . ( G ) Percentage of live Control-HCT 116 p53−/− or ( H ) NDI1-HCT 116 p53−/− treated with metformin for 72 hr followed in complete media . Error bars are SEM ( Relative OCR n = 5; Cell number n = 4 ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 00810 . 7554/eLife . 02242 . 009Figure 3—figure supplement 1 . Metformin inhibits mitochondrial complex I of CCL16 cells . ( A ) Relative mitochondrial oxygen consumption rate of Control-CCL16 cells and ( B ) NDI1-CCL16 cells treated with metformin in complete media for 20 min . ( C ) Percentage of live Control-CCL16 , ( D ) B2-CCL16 , or ( E ) NDI1-CCL16 treated with metformin for 16 hr in media with galactose was substituted for glucose . ( F ) Cell number of Control-CCL16 cells , ( G ) B2-CCL16 and ( H ) NDI1-CCL16 cells 48 and 72 hr post treatment with 2 mM or 4 mM metformin in complete media . Error bars are SEM ( n = 4 ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 00910 . 7554/eLife . 02242 . 010Figure 3—figure supplement 2 . Phenformin inhibits mitochondrial complex I of CCL16 cells . ( A ) Relative mitochondrial oxygen consumption rate of Control-CCL16 cells and ( B ) NDI1-CCL16 cells treated with metformin in complete media for 20 min . ( C ) Percentage of live Control-CCL16 or ( D ) NDI1-CCL16 treated with metformin for 16 hr in media with galactose was substituted for glucose . ( E ) Cell number of Control-CCL16 cells , ( F ) B2-CCL16 and ( G ) NDI1-CCL16 cells 48 and 72 hr post treatment with 50 μM or 100 μM phenformin in complete media . Error bars are SEM ( n = 4 ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 010 To determine whether metformin and phenformin diminish proliferation and survival of cells lacking endogenous complex I activity , we utilized a variant of CCL16 hamster fibroblasts that harbors a mutation in complex I ( B2-CCL16 ) ( Seo et al . , 1998 ) . Metformin inhibited proliferation of wild-type Control-CCL16 hamster fibroblasts but not that of B2-CCL16 cells or of B2-CCL16 cells reconstituted with NDI1 ( NDI1-CCL16 ) . Metformin and phenformin inhibited cellular oxygen consumption in wild-type CCL16 but not in CCL16-NDI1 cells ( Figure 3—figure supplement 1 and 2 ) . When these cells were cultured in galactose-substituted media , both metformin and phenformin induced cell death only in wild-type CCL16 cells . Survival of NDI1-CCL16 cells was not affected by metformin or phenformin ( Figure 3—figure supplements 1 and 2 ) . The B2-CCL16 cells die in galactose in the absence of metformin or phenformin since they harbor a mutation in complex I . Taken together , these results confirm that the anti-proliferative effects of metformin and phenformin require mitochondrial complex I inhibition . The positive charge of metformin has been proposed to account for it is accumulation within the matrix of mitochondria that exhibit a robust inner mitochondria membrane potential ( Owen et al . , 2000 ) . Alternatively , the non-polar hydrocarbon-side chain of the drug could promote binding to complexes within mitochondrial membranes . We tested whether the mitochondrial membrane potential is necessary for metformin-dependent inhibition of complex I . Saponin-permeabilized Control-HCT 116 p53−/− cells were induced to respire on pyruvate/malate in the presence of either ADP or CCCP . Although both ADP and CCCP induce mitochondrial respiration , only CCCP depolarizes mitochondrial inner membrane potential . Metformin inhibited ADP but not CCCP stimulated oxygen consumption indicating that the metformin-mediated inhibition of mitochondrial complex I required polarized mitochondria ( Figure 4A , B ) . Rotenone , an irreversible inhibitor of complex I , does not require polarized mitochondria to inhibit mitochondrial oxygen consumption ( Figure 4C , D ) . Our results suggest that metformin would not be effective in suppressing complex I activity of intact cells if mitochondrial inner membrane potential was disrupted . Metformin-mediated inhibition of the electron transport chain diminishes proton pumping , which might depolarize the mitochondrial membrane , thus limiting accumulation of the drug . However , we did not observe a reduction in the mitochondrial inner membrane potential measured using TMRE fluorescent dye in Control and NDI1 HCT116 p53−/− cells after metformin treatment ( Figure 4E , F ) . When electron transport function is inhibited , the ATP synthase can function in reverse such that it uses ATP generated by glycolysis to pump protons across the inner mitochondrial membrane , maintaining membrane potential ( Appleby et al . , 1999 ) . The ATP synthase inhibitor , Oligomycin A , diminished TMRE fluorescence in Control-HCT 116 p53−/− cells treated with metformin suggesting that in the presence of metformin , intact cells maintain their mitochondrial membrane potential by reversal of the ATP synthase ( Figure 4E ) . 10 . 7554/eLife . 02242 . 011Figure 4 . Metformin inhibition of complex I requires an intact mitochondrial inner membrane potential . ( A ) Complex I ( 2 mM malate , 10 mM pyruvate ) -driven oxygen consumption rate of saponin permeabilized Control-HCT 116 p53−/− cells over time . At t = 5 min permeabilized cells were treated with either 10 mM ADP to induce respiration with an intact mitochondrial membrane potential or ( B ) 10 µM CCCP to induce respiration in absence of mitochondrial membrane potential . At t = 12 min 1 mM metformin was added to cells . At t = 48 min antimycin A was added . ( C ) Complex I ( 2 mM malate , 10 mM pyruvate ) -driven oxygen consumption rate of saponin-permeabilized Control-HCT 116 p53−/− cells . At t = 5 min permeabilized cells were treated with either 10 mM ADP to induce respiration with an intact mitochondrial membrane potential or ( D ) 10 µM CCCP to induce respiration in absence of mitochondrial membrane potential . At t = 15 min , 1 μM rotenone was added to cells . At t = 25 min antimycin A was added . ( E ) Mitochondrial membrane potential measured by TMRE staining of Control-HCT116 p53−/− cells or ( F ) NDI1-HCT 116 p53−/− in the presence of 1 mM Metformin , 10 µM CCCP or 2 . 5 µM Oligomycin A . Error bars are SEM ( n = 4 ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 011 Rotenone irreversibly inhibits complex I , which contributes to its high toxicity in vivo . Because metformin is well tolerated , we sought to determine if metformin might reversibly bind to complex I . Saponin-permeabilized Control-HCT 116 p53−/− cells were treated with pyruvate and malate to maintain the mitochondrial inner membrane potential . Metformin was then added , followed by injection of either ADP or CCCP . Metformin inhibited ADP , but not CCCP-stimulated oxygen consumption ( Figure 5A , B ) . As metformin accumulation requires mitochondrial membrane polarization ( Figure 4A , B ) , these results indicate that metformin reversibly inhibits mitochondrial complex I . If the metformin that accumulated in the mitochondrial matrix irreversibly inhibited complex I , then oxygen consumption would have remained attenuated in CCCP-treated cells after metformin treatment . 10 . 7554/eLife . 02242 . 012Figure 5 . Metformin reversibly inhibits mitochondrial complex I . ( A ) Complex I ( 2 mM malate , 10 mM pyruvate ) -driven oxygen consumption rate of saponin permeabilized Control-HCT 116 p53−/− cells over time . At t = 5 min permeabilized cells were exposed to 1 mM metformin . At t = 25 min respiration was stimulated with either 10 mM ADP to induce respiration with an intact mitochondrial membrane potential or ( B ) 10 µM CCCP to induce respiration lacking membrane potential with 10 mM ADP . At t = 42 min antimycin A was added . For mitochondrial membrane potential error bars are SEM ( n = 4 ) . For oxygen consumption rates , error bars are standard deviation ( n = 6 ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 012 An emerging function of mitochondria distinct from their ability to perform biosynthetic and bioenergetic reactions is the generation of H2O2 , which promotes signaling in normal and cancer cells ( Hamanaka and Chandel , 2010; Sena and Chandel , 2012 ) . Mitochondrial complexes I and III produce superoxide into the mitochondrial matrix , where it is converted quickly to H2O2 by SOD2 ( Brand , 2010 ) . Mitochondrial complex III also generates superoxide into the mitochondrial intermembrane space where it escapes through VDACs to cytosol and is converted into H2O2 by SOD1 ( Han et al . , 2003; Muller et al . , 2004 ) . We measured production and subsequent release of H2O2 from isolated mitochondria in the presence of metformin , rotenone , or antimycin A ( complex III inhibitor ) using pyruvate and malate as substrates . Consistent with previous reports , rotenone and antimycin increased the release of H2O2 from mitochondria isolated from Control-HCT 116 p53−/− cells ( Figure 6A , B; St-Pierre et al . , 2002; Muller et al . , 2004 ) . In contrast , metformin did not substantially increase H2O2 release , suggesting that metformin and rotenone act on different sites of complex I ( Figure 6A ) . When mitochondria were isolated from NDI1–HCT 116 p53−/− cells , only antimycin lead to a significant increase in H2O2 release ( Figure 6B ) . Previous reports have shown that metformin does not substantially increase H2O2 production in isolated liver mitochondria and that metformin diminishes mitochondrial H2O2 production in response to paraquat , which induces mitochondrial ROS production ( Batandier et al . , 2006; Algire et al . , 2012 ) . 10 . 7554/eLife . 02242 . 013Figure 6 . Metformin reduces HIF-1 activation through inhibition of mitochondrial complex I . ( A and B ) H2O2 levels emitted by mitochondria isolated from Control-HCT 116 p53−/− and NDI1-HCT 116 p53−/− cells respiring on 2 mM malate and 10 mM pyruvate . Mitochondria were treated with 1 mM Metformin , 500 nM rotenone , 500 nM Antimycin , or left untreated . H2O2 levels were measured using Amplex Red . ( C ) Levels of HIF1α protein in Control-HCT 116 p53−/− and NDI1-HCT 116 p53−/− cells treated with 0 or 1 mM metformin for 24 hr , then placed in normoxia ( 21% O2 ) , hypoxia ( 1 . 5% O2 ) or treated with Deferoxamine ( DFO ) for 8 hr . ( D ) Quantification of HIF1α protein levels from panel C . ( E ) Hypoxic-induced expression of HIF target genes in Control-HCT 116 p53−/− and NDI1-HCT 116 p53−/− treated with 0 , 0 . 5 mM or 1 mM metformin for 24 hr following treatment with normoxia or hypoxia for 16 hr . Error bars are SEM ( n = 3 for Amplex Red; Blot is representative of four independent blots quantified in D , n = 4 for gene expression ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 013 One biological consequence of mitochondrial-generated H2O2 is hypoxic stabilization of the hypoxia-inducible factors ( HIFs ) ( Chandel et al . , 2000 ) . HIFs are involved in metabolic adaptation of tumor cells to hypoxia ( Semenza , 2012 ) . Metformin reduced hypoxic stabilization of HIF-1α in Control-HCT 116 p53−/− but not in NDI1-HCT 116 p53−/− ( Figure 6C , D ) . Metformin did not decrease deferoxamine ( DFO ) stabilization of HIF-1α protein . DFO is an iron chelator known to directly stabilize HIF-1α protein independent of upstream signaling events . Metformin also significantly diminished hypoxic activation of HIF-dependent target genes , vascular endothelial growth factor ( VEGF ) , and carbonic anhydrase 9 ( CA9 ) in Control-HCT 116 p53−/− but not in NDI1-HCT 116 p53−/− ( Figure 6E ) . Thus , metformin is an effective agent to reduce hypoxic activation of HIF-1 . Finally , we directly tested whether tumor cell autonomous inhibition of mitochondrial complex I by metformin was required to decrease tumor progression in vivo . As our NDI1-HCT 116 p53−/− cells are refractory to multiple effects of metformin in vitro , we reasoned that if metformin acted directly on mitochondrial complex I within the tumor cells to reduce tumorigenesis then NDI1-HCT 116 p53−/− xenograft tumors would not be inhibited in their growth . However , if metformin acts at the organismal level to diminish tumorigenesis then NDI1-HCT 116 p53−/− xenograft tumor growth would be suppressed similar to control tumors . Control-HCT 116 p53−/− cells subcutaneously injected into the left flank of nude mice rapidly grew in vivo , while tumors from mice fed metformin through drinking water ad libitum starting 4 days post-implantation exhibited a marked reduction in growth ( Figure 7A , B ) . NDI1-HCT 116 p53−/− xenograft growth was resistant to metformin therapy ( Figure 7A , B ) , suggesting that the metformin carries out its tumor inhibitory effects in a cancer cell autonomous manner through inhibition of mitochondrial complex I . Importantly , the consumption of water containing metformin was similar between control and NDI1 tumor barring mice ( Figure 7C ) . Transcripts for the HIF target genes CA9 and VEGF were diminished in control tumors treated with metformin but not in NDI1 expressing tumors ( Figure 7D , E ) . The blood glucose , plasma lactate , insulin , and IGF-1 level displayed no differences between the metformin-treated animals and control animals at the end of the study ( Figure 7—figure supplement 1 ) , consistent with previous reports ( Tomimoto et al . , 2008 ) . 10 . 7554/eLife . 02242 . 014Figure 7 . Metformin inhibits mitochondrial complex I to diminish tumor growth . ( A ) Average tumor volume in mice injected with 3 × 106 Control-HCT 116 p53−/− or NDI1-HCT 116 p53−/− cells injected into the left flank of J:Nu mice . Mice were given ad libitum , water free of metformin ( squares ) or were treated with 250 mg/kg of metformin in the drinking water starting 4 days post tumor injection ( triangles ) . ( B ) Average tumor mass from mice injected with 3 × 106 Control-HCT 116 p53−/− or NDI1-HCT 116 p53−/− cells injected into the left flank of J:Nu mice after 32 days . ( C ) Average daily water consumption of mice treated with metformin ( 1 . 25 mg/ml ) . ( D ) HIF target genes expression measured in Control-HCT 116 p53−/− or NDI1-HCT 116 p53−/− tumors treated with metformin . Error bars are SEM ( n = 8 per group for tumor study , n = 8 for H2O consumption , error bars represent standard deviation of two cages with four mice house in each cage , n = 3 for gene expression ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 01410 . 7554/eLife . 02242 . 015Figure 7—figure supplement 1 . Metformin treatment of tumor bearing mice does not alter blood glucose , plasma lactate , IGF-1 , and insulin levels . ( A ) Blood glucose levels of mice injected with 3 × 106 Control-HCT 116 p53−/− or NDI1HCT 116 p53−/− cells into the left flank of J:Nu mice . Mice were treated with water free of metformin or were treated with 250 mg/kg of metformin ad libitum in the drinking water starting 4 days post tumor injection for 27 days ( Figure 4 ) . ( B ) Plasma insulin , ( C ) IGF-1 , and ( D ) lactate levels from tumor bearing mice from Figure 4 27 days after beginning metformin treatment . Error bars are SEM ( n = 8 per group ) . * indicates significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 01510 . 7554/eLife . 02242 . 016Figure 7—figure supplement 2 . Metformin inhibits cellular proliferation and pro- proliferative signaling via complex I inhibition . ( A ) Immunoblot for the complex I protein NDUFS3 in Control-A549 cells expressing control shRNA and NDI-A549 cells expressing a shRNA specific for the NDUFS3 subunit . ( B ) Relative mitochondrial oxygen consumption rate ( OCR ) of intact Control-A549 and ( C ) NDI1-NDUFS3-A549 cells treated with metformin in complete media for 20 min . ( D ) Relative complex I ( 2 mM malate , 10 mM pyruvate , 10 mM ADP ) -driven oxygen consumption rate of saponin permeabilized Control-A549 cells and ( E ) NDI1-NDUFS3-A549 cells treated with metformin for 20 min in mitochondrial assay buffer . ( F ) Cell number of Control-A549 and ( G ) NDI1-NDUFS3-A549 cells 24 , 48 , and 72 hr post treatment with 2 mM metformin in complete media . Error bars are SEM ( OCR n = 4; Cell proliferation n = 3 ) . * indicate significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 01610 . 7554/eLife . 02242 . 017Figure 7—figure supplement 3 . NDI1 expressing A549 cells are refractory to metformin treatment in a xenograft model of tumor growth . ( A ) Hypoxic induction of HIF1a in Control-A549 and NDI1-A549 cells treated with 0 or 2 mM metformin for 24 hr then placed in normoxia ( 21% O2 ) , hypoxia ( 1 . 5% O2 ) or treated with Deferoxamine ( DFO ) for 8 hr . ( B ) Average tumor volume in mice injected with 3 × 106 Control-A549 cells or ( C ) 3 × 106 NDI1-NDUFS3-A549 cells injected into the left flank of nu/nu mice . Mice were treated with water free of metformin ( squares ) or were treated with 300 mg/kg of metformin in the drinking water ( triangles ) 1 week prior to tumor and injection and continued throughout the duration of the study . Error bars are SEM ( n = 10 per group ) . * indicate significance p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02242 . 017 To further bolster our conclusions , we examined tumor growth of A549 cells expressing NDI1 and shRNA targeting mammalian NDUFS3 , a subunit of the human mitochondrial complex I ( Vogel et al . , 2007 ) . A549 cells are null for the tumor suppressor LKB1 and are known to be responsive to metformin therapy ( Rocha et al . , 2011 ) . Furthermore , LKB1-deficient tumors are more susceptible to the related biguanide phenformin ( Shackelford et al . , 2013 ) . We replaced the endogenous mitochondrial complex I by expressing the NDI1 protein in A549 cells stably expressing shRNA against NDUFS3 ( Figure 7—figure supplement 2A ) , referred to as NDI1-shNDUFS3-A549 cells . The Control-A549 cells contain the empty vectors with selection markers BFP and puromycin for NDI1 and shRNA , respectively . NDI1-NDUFS3-A549 cells were resistant to the metformin-mediated reduction in cellular and mitochondrial oxygen consumption and cell proliferation ( Figure 7—figure supplement 2B–G ) . Metformin also decreased HIF-1a protein levels in control but not in NDI1-NDUFS3-A549 cells ( Figure 7—figure supplement 3A ) . Control-A549 cells subcutaneously injected into the left flank of nude mice rapidly grew in vivo , while tumors from mice fed metformin through drinking water ad libitum starting 2 weeks prior to tumor induction exhibited a marked reduction in growth over 45 days ( Figure 7—figure supplement 3B ) . By contrast , NDI1-NDUFS3-A549 xenografts were completely resistant to metformin therapy , suggesting that metformin carries out its tumor inhibitory effects in a cell autonomous manner through inhibition of mitochondrial complex I in these cells ( Figure 7—figure supplement 3C ) . NDI1-NDUFS3-A549 xenografts grew slower than control xenografts in untreated mice . An alternative explanation for the resistance of NDI1-NDUFS3-A549 cells to metformin could be that effects of metformin are blunted in slower-growing cells . However , based on our results from HCT116 cells in vivo and extensive analysis of A549 cells in vitro , we find that it is likely that NDI1 expressing A549 cells are also resistant to metformin in vivo due to rescue of complex I activity by the NDI1 protein . The mechanisms by which metformin inhibits cancer growth is not fully understood . Metformin has been previously shown to inhibit mitochondrial complex I , yet it is not known whether metformin exhibits its anti-tumor effects through inhibition of complex I . It is important to note that many commonly utilized drugs have been shown to inhibit mitochondrial function in vitro; however , for many of these drugs it is not clear whether the in vivo mechanism of action is through modulating mitochondrial metabolism ( Gohil et al . , 2010 ) . In the present study , we genetically demonstrate that metformin acts as a complex I inhibitor in vitro and in vivo to inhibit tumorigenesis . Our current findings bolster our previous findings highlighting the importance of mitochondrial metabolism as essential for cancer proliferation ( Weinberg et al . , 2010; Sullivan et al . , 2013 ) . Our results are consistent with a recent report demonstrating another biguanide phenformin exerts its anti-tumor effects by inhibiting complex I ( Birsoy et al . , 2014 ) . However , unlike metformin , phenformin has been discontinued because of frequent occurrence of lactic acidosis ( Bailey and Turner , 1996 ) . While our study highlights the importance of mitochondrial complex I inhibition within cancer cells as a major mechanism by which metformin reduces tumor burden , it does not necessarily preclude any additional organismal effects of metformin that might reduce tumor progression in certain cancers . Specifically , cancer cells that express insulin receptors might be affected by metformin's inhibition on hepatic gluconeogenesis to reduce circulating insulin levels ( Goodwin et al . , 2012 ) . The levels of metformin within cells is regulated by a balance between uptake mechanisms , which are dependent on expression of organic cation transporters ( OCT 1 , 2 , and 3 ) , and expulsion mechanisms , which are dependent on expression on multidrug and toxin extrusion proteins ( MATE1 and MATE 2 ) ( Emami Riedmaier et al . , 2013 ) . Cancer cells are likely to have a wide range in expression of uptake and extrusion proteins allowing for metformin accumulation . The uptake of metformin in mitochondria is also dependent on mitochondrial membrane potential . Metformin exists as a cation at physiological pH and thus its accumulation within mitochondria is predicted to increase as a function of the mitochondrial membrane potential . The inhibition of electron transport at complex I by metformin should reduce the mitochondrial membrane potential as proton pumping is linked to electron transport . However , we observed that metformin-treated cells maintain their mitochondrial membrane potential in part due to the reversal of the ATP synthase thus allowing accumulation of metformin in the mitochondrial matrix ( Figure 3 ) . It is well known that cells with compromised ETC can sustain mitochondrial membrane potential through this mechanism ( Appleby et al . , 1999 ) . Collectively , our results indicate that will inhibit mitochondrial complex I and decrease tumorigenesis in cancer cells that have transporters across the plasma membrane , as well as a robust inner mitochondrial membrane potential to allow metformin to reach the mitochondrial matrix . It will be of interest to observe whether metformin's efficacy as an anti-cancer agent is dependent on the tumor expression of OCTs . Our study also reveals that metformin's inhibition on mitochondrial complex I is distinct from the classical complex I inhibitor rotenone . Metformin requires a robust mitochondrial membrane potential to accumulate in the mitochondrial matrix and reversibly inhibits complex I . By contrast , rotenone is an irreversible inhibitor of complex I that does not require mitochondrial membrane potential . Rotenone accumulation is not dependent on specific transporters expressed in the plasma membrane and readily accumulates in all cells and thus is highly toxic . Furthermore , metformin does not promote generation of ROS from complex I , while rotenone stimulates ROS production from complex I ( Figure 5A ) . The flavin site within Complex I produces ROS and previous studies indicate that rotenone inhibits complex I downstream of the flavin site , stimulating ROS generation ( Pryde and Hirst , 2011; St-Pierre et al . , 2002 ) . It is likely that metformin acts upstream of this site , inhibiting complex I activity while also inhibiting ROS generation . Mitochondrial complex III also produces ROS ( Muller et al . , 2004 ) . Therefore , metformin by inhibiting the complex I would limit electron flow to complex III thereby reducing ROS generation from complex III . A consequence of mitochondrial complex III ROS generation is the hypoxic activation of HIF-1 ( Bell et al . , 2007 ) . Indeed , we found that metformin reduced hypoxic stabilization of HIF-1α protein and HIF-dependent target genes . This result is consistent with our previous study in which we identified multiple mitochondrial inhibitors in a large-scale chemical screen for inhibitors of hypoxic activation of HIF ( Lin et al . , 2008 ) . Beyond cancer , metformin might be an effective treatment for diseases associated with hyperactivation of HIF such as pulmonary hypertension ( Shimoda and Semenza , 2011 ) . The concentration of metformin we administered to mice in this report is predicted to achieve plasma concentrations of metformin similar to those in humans receiving metformin therapy . In our study , metformin was administered to drinking water at a concentration of 1 . 25 mg/ml , approximately 250 mg/kg . Conversion of doses from mice to human is achieved using a formula based on body surface area normalization ( Reagan-Shaw et al . , 2008 ) . The human equivalent dose of 250 mg/kg metformin in mice is 20 . 27 mg/kg or 1418 mg for a 70 kg adult . Patients typically receive 500–2500 mg metformin per day ( Scarpello and Howlett , 2008 ) . In these patients receiving metformin therapy , plasma levels range between 0 . 5–2 µg/ml; approximately 4 µM–15 µM ( Graham et al . , 2011 ) . Previous studies have shown that mice receiving drinking water containing 1–5 mg/ml of metformin have a plasma steady-state metformin concentration of 0 . 45–1 . 7 µg/ml ( Memmott et al . , 2010 ) , indicating that our metformin dosage falls within a clinically relevant range . Metformin concentrations ( 250–1000 µM ) used for in vitro experiments in this report are in the range utilized by investigators in the diabetes field to examine the effects of metformin on primary murine and human hepatocytes ( Foretz et al . , 2010; Stephenne et al . , 2011; Miller et al . , 2013 ) . A perplexing observation revealed in our study and previous studies is that the metformin concentrations required to induce biological effects in vitro are a magnitude of order higher than plasma level concentrations of metformin in vivo . A likely cause for this difference is the accumulation of metformin in liver or tumors to local concentrations that are much higher than in the circulating plasma . Indeed , the gut has been shown to accumulate metformin in the mM range ( Bailey et al . , 2008; Proctor et al . , 2008 ) . In summary , our results indicate that metformin reversibly inhibits mitochondrial complex I within cancer cells to reduce tumorigenesis . Metformin inhibits tumorigenesis through multiple mechanisms including the induction of cancer cell death in conditions , when glucose is limited and through inhibition of mitochondrial ROS-dependent signaling pathways that promote tumorigenesis ( i . e . , HIF ) . These results indicate that metformin would be most effective in low glucose and oxygen conditions . It will be of interest to determine whether metformin treatment might provide a useful adjunct to therapies that limit glucose uptake ( e . g . , PI3K inhibitors ) or drive tumors to low glucose and oxygen levels ( e . g . , anti-angiogenic inhibitors ) . NDI1 ( +NDI1 ) and control pWPI vectors containing BFP ( +BFP ) were transfected into 293FT cells using lipofectamine 2000 ( Invitrogen , Carlsbad , CA , USA ) along with pMD2 . G and psPAX2 packaging vectors to produce Control-BFP or NDI-BFP lentivirus . Control-HCT116 p53−/− , NDI1-HCT 116 p53−/− , Control-A549 , and NDI1-A549 were created by infecting parental HCT116 p53−/− or A549s with either Control-BFP lentivirus or NDI1-BFP lentivirus . Selection of the BFP expressing Control-HCT116 p53−/− , NDI1-HCT 116 p53−/− , Control-A549 , and NDI1-A549 was done by periodic fluorescence-activated cell sorting ( FACS ) for BFP-positive cells with a MoFlo ( Beckman–Coulter , Brea , CA , USA ) . Control-HCT 116 p53−/− , NDI1-HCT 116 p53−/− , Control-HCT116 p53+/+ , NDI1-HCT 116 p53+/+ , Control-A549 , NDI1-A549 cells , NDI1-NDUFS3-A549 , CCL16 , CCL16-B2 , and CCL16-NDI1 were cultured in DMEM supplemented with 10% fetal bovine serum , 1% HEPES , and 1% penicillin-streptomycin . Cells were routinely checked for BFP expression using FACS to ensure high NDI1 expression . TRC consortium validated pLKO . 1 shRNA clones against control or NDUFS3 were obtained from Sigma and the lentivirus was produced in COS1 cells ( TRCN0000218593 ) . Control-A549 and NDI1-A549 cells infected the pLKO . 1 lentivirus were additionally selected under continuous puromycin selection ( 1 μg/ml ) . 1 . 5 × 105 HCT 116 wild-type or p53 null cells ( 1 × 105 A549 and CCL16 cells ) were plated on 35-mm dishes . At 24 , 48 , and 72 hr after plating , cells were trypsinized and counted using a Vi-Cell ( Beckman–Coulter , Brea , CA , USA ) . Cell viability percentage was determined by trypan blue exclusion after 24 hr of treatment with metformin in DMEM with 10% dialyzed fetal bovine serum , 1% HEPES and 1% penicillin-streptomycin and lacking glucose ( HCT 116 ) or with galactose substituted for glucose ( CCL16 ) . Oxygen consumption rates ( OCR ) were measured utilizing the XF24 Seahorse Biosciences Extracellular Flux Analyzer . 2 × 104 HCT 116 s ( 1 . 5 × 104 A549 and CCL16 cells ) were plated in the seahorse cell plate and incubated overnight . HCT 116 cells were then incubated for 24 hr in metformin hydrochloride ( Sigma-Aldrich , St . Louis , CA , USA ) . 30 min before assay , the media were changed to 500 μl of fresh media containing metformin . For A549 and CCL16 cells , metformin was injected onto the cells using the Flux Analyzer and OCR was measured following a 20-min incubation . Basal mitochondrial oxygen consumption rate was determined by subtracting the antimycin A ( Sigma , 1 μM ) sensitive OCR from the basal OCR following normalization for cell number . The coupled mitochondrial oxygen consumption rate was determined by subtracting the mitochondrial respiration following Oligomycin A ( Sigma ) treatment from the basal mitochondrial oxygen consumption rate . The complex I and complex II specific contributions to oxygen consumption were measured by changing 5 × 104 HCT 116 cells ( 2 × 104 A549 cells ) plated overnight in complete DMEM into mitochondrial assay buffer ( 70 mM sucrose , 220 mM mannitol , 10 mM KH2PO4 , 5 mM MgCl2 , 2 mM HEPES , 1 . 0 mM EGTA , and 0 . 2% ( wt/vol ) fatty-acid free BSA , pH 7 . 2 ) , supplemented with 10 mM ADP ( Sigma ) and permeabilized by saponin injection ( 120 μg/ml , Sigma ) . The OCR was observed after saponin injection for loss of O2 usage until stabilization at baseline where intracellular substrates have diffused out of the permeabilized cells . The complex I substrates ( pyruvate 10 mM , malate 2 mM ) or the complex II substrate ( succinate 10 mM ) was then added and the increase in OCR was measured . For complex II OCR , rotenone ( Sigma , 1 μM ) was added to inhibit complex I oxygen consumption . Post-injection of mitochondrial substrates , metformin was added and the OCR was measured . Complex I or II OCR was determined as the difference between substrate-free and substrate-added OCR . To examine whether mitochondrial inhibitors depend on membrane potential to inhibit complex I , cells were treated with saponin ( 120 μg/ml ) and the complex I substrates ( pyruvate 10 mM , malate 2 mM ) . Next , the cells were treated with either Carbonyl cyanide m-chlorophenyl hydrazone ( CCCP , Sigma −10 μM ) or ADP ( 10 mM ) to induce respiration . The complex I inhibitors metformin ( 1 mM ) or rotenone ( 1 μM ) were then added to assay inhibitory capacity in the presence or absence of membrane potential . Finally , the cells were treated with antimycin ( 1 μM ) to completely inhibit mitochondrial oxygen consumption . To determine the reversibility of metformin inhibition of complex I , cells were permeabilized with saponin and concurrently treated with complex I substrates ( pyruvate 10 mM , malate 2 mM ) . The cells were then treated with 0 or 1 mM metformin for 20 min . Next , the cells were treated with CCCP ( 10 μM ) or ADP ( 10 mM ) to induce respiration in the presence or absence of mitochondrial membrane potential . Finally , the cells were treated with antimycin A ( 1 μM ) . 1 × 106 Control-HCT 116 p53−/− or NDI1-HCT 116 p53−/− cells were plated in 35-mm dishes in complete media and incubated overnight . The next day the cells were switched to media containing metformin and incubated for an additional 24 hr . After 24 hr , 50 nM tetramethylrhodamine ( TMRE , Life Technologies ) was added to the media for 30 min . The cells were washed , isolated , and resuspended in 1 ml of PBS . 10 μM CCCP , 2 . 5 μM Oligomycin A or no treatment was added and cells were incubated for an additional 30 min . The cells were then analyzed on BD LSRFortessa ( San Jose , CA , USA ) machine . The cells were gated for singlets and then the mean TMRE fluorescence was obtained using FlowJo analysis software . The cells were scraped into mitochondrial isolation buffer ( 250 mM Sucrose , 1 mM EGTA , 20 mM Tris pH 7 . 4 ) and disrupted by 10 strokes with a Dounce homogenizer and 5 expulsions through a 28-gauge needle . The lysates were centrifuged at 500×g for 10 min to remove nuclei , followed by centrifugation at 18 , 000×g for 20 min to pellet mitochondria . Mitochondrial fractions were suspended in a reaction buffer containing 120 mM KCl , 5 mM KH2PO4 , 3 mM HEPES , 1 mM EGTA , and 0 . 3% BSA , pH 7 . 2 . Mitochondria respired on 2 . 5 mM pyruvate and 1 mM malate in the presence or absence of 500 nm Antimycin A or 1 mM Metformin . Superoxide dismutase ( 200 U/ml; Sigma ) was used to convert mitochondria-produced superoxide to hydrogen peroxide ( H2O2 ) . In the presence of horseradish peroxidase ( 10 U/ml , Thermo Scientific ) , Amplex Red ( 100 µM , Molecular Probes ) reacts with H2O2 , producing the fluorescent oxidation product , resorufin . Fluorescence was measured at excitation 544 nm , emission 590 nm . Protein was extracted using cell lysis buffer ( Cell Signaling , Danvers , MA , USA ) plus PMSF ( 600 μM ) . Protein concentration was quantified using the BCA Protein Assay ( Pierce , Rockford , IL , USA ) . Protein samples were resolved on SDS polyacrylamide gels ( Bio-Rad ) and subsequently transferred to nitrocellulose membranes by semi-dry transfer using the Trans-Blot Turbo ( Bio-Rad ) . To determine levels of HIF1α ( BD Transduction Laboratories , Clone 54/HIF-1a ) protein levels , HCT 116 cells were pretreated with 1 mM metformin or left untreated for 24 hr followed by treatment with 100 µM DFO , or incubation in 1 . 5% O2 for 8 hr . For A549s , cells were pretreated with 2 mM metformin or mock treatment for 1 hr and treated with mock treatment , 100 μM DFO , or incubated in a hypoxia chamber at 1 . 5% O2 for 4 hr . OCT1 ( Novus Biologicals , Littleton , CO , USA ) and NDUFS3 ( MitoSciences ) protein expression were assessed using the same technique above without any treatment conditions . α-tubulin ( Sigma ) was used as a loading control for protein blots assessing HIF1a and NDUFS3 protein levels while β-actin ( Sigma ) was used as a control for OCT1 protein quantification . Image Studio Lite version 3 . 1 ( Licor ) was used for analysis and quantification of protein levels . In brief , blots were scanned at high resolution and imported into Image Studio . Individual lanes of HIF1α and OCT1 protein levels were normalized to tubulin and actin respectively . Relative levels of HIF1a were normalized to an untreated sample incubated at 21% O2 . 1 . 5 × 105 Control-HCT 116 p53−/− or NDI1-HCT 116 p53−/− cells were plated on 24-well plates in complete media and 24 hr later treated with media without serum and metformin for 24 hr . Fresh media with or without metformin were added just before cells were exposed to either normoxia ( 21% O2 ) or hypoxia ( 1 . 5% O2 ) for 16 hr . Cells were washed once with ice-cold PBS and RNA was extracted using commercial kit ( Aurum Total RNA mini Kit , Bio-Rad Hercules , CA ) . 1 µg of RNA was transcribed to cDNA using the iScript cDNA synthesis kit ( Bio-Rad ) and the relative concentrations of cDNA were analyzed by qPCR on a Bio-Rad CFX384 Touch Real-Time PCR Detection System using iQ SYBR Green Supermix ( Bio-Rad ) and the following primer sequences: CA9 sense: CCGAGCGACGCAGCCTTTGA CA9 antisense: GGCTCCAGTCTCGGCTACCT VEGF sense: TACCTCCACCATGCCAAGTG VEGF antisense: GATGATTCTGCCCTCCTCCTT 18s sense: CGTTGATTAAGTCCCTGC CCTT 18s antisense: TCAAGTTCGACCGTCTTCTCAG . HIF target gene expression was analyzed from tumor RNA isolated using TRIzol ( Life Technologies , Grand Island , NY , USA ) . RNA was isolated and immediately transcribed to cDNA using the iScript cDNA synthesis kit , and the relative concentrations of cDNA were analyzed as described above . Blood was extracted weekly from female outbread athymic nude mice ( J:Nu , Jackson Labs ) . The mice were injected with 1 × 106 A549 cells and treated with either water free of metformin or water containing 250 mg/kg metformin . Blood glucose levels were determined weekly using a glucometer ( One Touch ) . Blood was collected into heparinized tubes at the end of the xenographic tumor study described below . Plasma was then isolated from whole blood by spinning at 21 , 000 × g for 5 min . IGF-1 levels were determined using an ELISA ( Abcam , Cambridge , MA , USA ) , as were insulin levels ( Millipore , Billerica , MA , USA ) . Finally , lactate levels were measured using a colorimetric kit ( Abcam ) . Female J:Nu mice were injected with 3 × 106 Control-HCT 116 p53−/− or NDI1-HCT 116 p53−/− using a 27-gauge needle . 4 days post-injection mice were treated with water supplemented with metformin ( Sigma ) or water alone and fluid intake was monitored daily to ensure an effective dose of 1 . 25 mg/ml or 250 mg/kg ( Buzzai et al . , 2007 ) of metformin . In mice receiving drinking water containing 1–5 mg/ml of metformin , the plasma steady-state metformin concentration is reported to be in the range of 0 . 45–1 . 7 µg/ml ( Memmott et al . , 2010 ) . In patients receiving metformin therapy , plasma levels are between 0 . 5 µg/ml and 2 µg/ml ( Graham et al . , 2011 ) . This treatment created four experimental groups: Control-HCT 116 p53−/− with H2O ( n = 8 ) , Control-HCT 116 p53−/− with Metformin ( n = 8 ) , NDI1-HCT 116 p53−/− with H2O ( n = 8 ) , NDI1-HCT 116 p53−/− with Metformin ( n = 8 ) . Experiments are from two independent cohorts of four mice each . Tumors were measured three times per week using calipers and tumor volume was determined using the equation ( 3 . 14/6 × L × W2 ) . At the completion of the study , mice were euthanized and the tumors were extracted and weighed . For A549s , 3 × 106 Control-A549 or NDI1-NDUFS3-A549 cells were injected into the left flank of female nude mice ( Nu:Nu ) administered metformin ( 250 mg/kg ) in their drinking water for 2 weeks prior to cell injection and continuously administered throughout the experiment . All mouse work was done in accordance with Northwestern University Institutional Animal Care and Use Committee . Data are presented as the mean ± SEM . Statistical significance was determined using 1-way ANOVA with a Bonferroni posttest correction , 2-way ANOVA when two variables were present , or the students t test comparing control to experimental conditions for p<0 . 05 . For all differences uncovered , a student t test was performed to verify differences between control and experimental groups .	Metformin is widely used to reduce the high blood sugar levels caused by diabetes . Recently , several studies have suggested that patients taking metformin who also develop cancer have tumors that grow more slowly than average . As clinical trials have already started to investigate if metformin is an effective anti-cancer treatment , it is important to understand how it might restrict tumor growth . Researchers have proposed two ways that metformin could affect tumors . First , insulin is known to prompt cancer cells to divide , so the slower rate of tumor growth could just be a side-effect of the metformin reducing the amount of insulin in the blood . Alternatively , metformin could target cancer cells more directly by cutting the energy supply produced by their mitochondria . Metformin has been shown to disrupt complex I of the electron transport chain that is used by cells to generate energy . However , it is not known if disrupting complex I would actually stop cells dividing because they can generate energy in other ways . Wheaton , Weinberg et al . have now demonstrated that metformin does target complex I in cancer cells , and that its effects depend on the amount of glucose available for cells to convert , without involving mitochondria , into energy . When there is plenty of glucose , metformin slows down the rate at which cancer cells divide , which slows down tumor growth . When the cells are deprived of glucose , metformin kills the cells instead . Metformin also inhibits the pathways that regulate hypoxia inducible factors ( HIFs ) , which are part of a system that helps cells to survive low-oxygen conditions , a prominent feature of many tumors . This means that metformin may combat cancer more effectively if used alongside other treatments that reduce the availability of both oxygen and glucose inside cells . Metformin could also potentially treat conditions that are linked to overactive HIFs , such as pulmonary hypertension .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2014	Metformin inhibits mitochondrial complex I of cancer cells to reduce tumorigenesis
Neuropeptides released from dense-core vesicles ( DCVs ) modulate neuronal activity , but the molecules driving DCV secretion in mammalian neurons are largely unknown . We studied the role of calcium-activator protein for secretion ( CAPS ) proteins in neuronal DCV secretion at single vesicle resolution . Endogenous CAPS-1 co-localized with synaptic markers but was not enriched at every synapse . Deletion of CAPS-1 and CAPS-2 did not affect DCV biogenesis , loading , transport or docking , but DCV secretion was reduced by 70% in CAPS-1/CAPS-2 double null mutant ( DKO ) neurons and remaining fusion events required prolonged stimulation . CAPS deletion specifically reduced secretion of stationary DCVs . CAPS-1-EYFP expression in DKO neurons restored DCV secretion , but CAPS-1-EYFP and DCVs rarely traveled together . Synaptic localization of CAPS-1-EYFP in DKO neurons was calcium dependent and DCV fusion probability correlated with synaptic CAPS-1-EYFP expression . These data indicate that CAPS-1 promotes fusion competence of immobile ( tethered ) DCVs in presynaptic terminals and that CAPS-1 localization to DCVs is probably not essential for this role . Neuropeptides and neurotrophic factors are essential for brain development and synaptic plasticity ( McAllister et al . , 1996; Huang and Reichardt , 2001; Poo , 2001; Samson and Medcalf , 2006; van den Pol , 2012 ) . These neuromodulators are transported in dense-core vesicles ( DCVs ) . Dysregulation of DCV transport and fusion is associated with cognitive and post-traumatic stress disorders ( Sadakata et al . , 2007b; Meyer-Lindenberg et al . , 2011; Sah and Geracioti , 2013 ) . DCVs bud off at the Golgi network ( Kim et al . , 2006 ) and are transported via microtubule-based motor proteins ( Hirokawa et al . , 2009; Schlager and Hoogenraad , 2009 ) . High-frequency firing facilitates DCV fusion and the resultant calcium influx triggers SNARE complex-dependent DCV secretion ( Bartfai et al . , 1988; Hartmann et al . , 2001; de Wit et al . , 2009; van de Bospoort et al . , 2012 ) . Unlike synaptic vesicles ( SVs ) , DCVs lack a local recycling mechanism . To ensure a constant and uniform supply of DCVs at release sites , DCVs are generally very dynamic although some DCVs are stationary . Stimulation triggers arrest of moving DCVs ( de Wit et al . , 2006; Shakiryanova et al . , 2006; Matsuda et al . , 2009; Wong et al . , 2012 ) , probably promoting their local availability for secretion . DCV fusion sites can be located in the entire neuron but DCVs preferentially fuse at presynaptic terminals and release at extra-synaptic sites requires more robust stimulation ( van de Bospoort et al . , 2012 ) . Recently , we have shown that Munc13 is an important regulator of DCV fusion at synapses ( van de Bospoort et al . , 2012 ) . However , in contrast to SV release , a comprehensive insight in the molecular mechanisms of DCV secretion is still lacking . Previous studies in Caenorhabditis elegans and Drosophila have implicated calcium-activator protein for secretion ( CAPS ) proteins in DCV secretion . Mutants of the C . elegans CAPS ortholog UNC-31 show reduced peptide release without affecting synaptic vesicle fusion ( Speese et al . , 2007 ) . In C . elegans neurons , UNC-31 is required for the docking of DCVs at the plasma membrane ( Zhou et al . , 2007; Hammarlund et al . , 2008; Lin et al . , 2010 ) . In Drosophila , deletion of dCAPS also affects DCV release but in contrast to C . elegans , leads to an increased DCV presence in terminals ( Renden et al . , 2001 ) . Mammals express two CAPS genes , CAPS-1 and CAPS-2 , which are complementarily expressed in brain ( Speidel et al . , 2003 ) and are essential for synaptic transmission ( Jockusch et al . , 2007 ) . In adrenal chromaffin cells , CAPS-1 deletion affects catecholamine uptake in chromaffin granules ( Speidel et al . , 2005; Brunk et al . , 2009 ) and deletion of CAPS-1 and CAPS-2 abolishes their fusion without affecting docking ( Liu et al . , 2010 ) . CAPS-2 is important for cerebellar development and neuron survival ( Sadakata et al . , 2004 , 2007a ) , and deletion of CAPS-1 in cerebellar neurons perturbs DCV trafficking ( Sadakata et al . , 2010 , 2013 ) . Hence , these studies suggest that CAPS proteins are involved in several aspects of DCV trafficking and release in invertebrates and in mammalian chromaffin cells , but their role in mammalian versus invertebrate systems appears to differ considerably , especially regarding synaptic transmission and cell survival . In this study , we analyzed the distribution and function of CAPS proteins in DCV trafficking and fusion in mammalian neurons using wild type ( WT ) , CAPS-2 null mutant and CAPS-1/2 null mutant mice . Endogenous CAPS-1 was present in puncta that partially overlapped with synaptic markers and also co-localized with DCV markers . CAPS deletion did not affect DCV biogenesis , neuropeptide loading or average DCV transport velocity . In CAPS double null mutant ( DKO ) neurons , DCV fusion was strongly reduced at synapses and at extra-synaptic sites . We developed a novel release assay to track single DCVs prior to fusion and found that CAPS deletion strongly affected DCV release of stationary , presumably tethered vesicles . We provide evidence that CAPS-1 localization at synapses is calcium dependent and that DCV release probability correlates with synaptic CAPS-1 expression levels . To understand CAPS-1 function in neuronal DCV release , we first investigated its sub-cellular localization in cultured neurons . Hippocampal neurons at 14 days in vitro ( DIV 14 ) were stained with a novel , CAPS-1-specific antibody ( Figure 1—figure supplement 1 ) . CAPS-1 was present in the cytosol and in dendritic and axonal puncta , probably membrane domains ( Figure 1A ) . Approximately 45% of these CAPS-1 puncta co-localized with the synaptic marker VGLUT1 in the entire neuron ( Figure 1B , C , Pearson's coefficient: 0 . 42±0 . 04 , n=7 ) . CAPS-1 immunoreactivity was detectable in approximately 60% of VGLUT1 positive synapses and vice versa approximately 60% of the CAPS-1 puncta co-localized with VGLUT1 ( Figure 1B , D , Manders' coefficients CAPS-1 in VGLUT1: 0 . 67±0 . 08 , and VGLUT1 in CAPS-1: 0 . 64±0 . 05 , n=8 ) . CAPS-1 domains were also found at extra-synaptic sites ( Figure 1B , E , white arrowheads ) . The CAPS-1 expression pattern differed from the sub-cellular localization of the DCV priming protein Munc13-1 , which was much more restricted to the synapse ( Figure 1E , F , Pearson's coefficient in the entire neuron: 0 . 64±0 . 04 , n=10 ) with 96% of synapses containing M13-1 ( Figure 1G , number of synapses containing M13-1: 96 . 3±0 . 7% , n=4 , number of synapses containing CAPS-1: 58 . 1±10 . 9% , n=7 ) . Since CAPS proteins have been initially identified as DCV resident proteins ( Berwin et al . , 1998 ) , we tested the co-localization of CAPS-1 with DCVs in hippocampal neurons . Using the endogenous DCV protein chromogranin B ( ChrB ) we found that the majority of DCVs are located in the axon ( Figure 1—figure supplement 2 ) . Antibody incompatibility precluded co-staining of CAPS-1 antibody with antibodies against ChrB . Therefore , we used the DCV cargo neuropeptide Y ( NPY ) fused to Venus ( Nagai et al . , 2002 ) , which showed more than 80% co-localization with this endogenous marker ( Figure 1H , I , Manders' coefficients for chromogranin B in NPY-Venus puncta: 0 . 97±0 . 02 , and NPY-Venus in ChrB puncta: 0 . 84±0 . 01 , n=14 ) . Approximately 35% of NPY-Venus labeled DCVs co-localized with endogenous CAPS-1 ( Figure 1J–L , Pearson's coefficient: 0 . 45±0 . 05 , n=21 , number of NPY-labeled DCVs co-expressing CAPS-1: 34 . 71±3 . 03% , n=6 ) . These data show that endogenous CAPS-1 is present at many but not all synapses . In addition , CAPS-1 domains are found at extra-synaptic regions and CAPS-1 partly co-localizes with DCV markers . 10 . 7554/eLife . 05438 . 003Figure 1 . CAPS-1 clusters are present at synaptic and extra-synaptic sites and partly co-localize with DCVs . ( A ) Example image of a hippocampal neuron ( DIV 14 ) stained for endogenous CAPS-1 ( green ) , dendrite marker MAP2 ( blue ) and synapse marker VGLUT1 ( red ) . Scale bar 10 µm . ( B ) Zoom of a neurite stained for CAPS-1 and VGLUT1 . CAPS-1 rich domains not overlapping with VGLUT1 ( filled arrowhead ) , VGLUT1 punctum not enriched for CAPS-1 ( open arrowhead ) , VGLUT1 puncta overlapping with a CAPS-1 rich domain ( stars ) . Scale bar 2 µm . ( C ) Co-localization of CAPS-1 with VGLUT1 in the entire neuron quantified by Pearson's correlation . Co-localization of VAMP2 with VGLUT1 was used as a positive control ( VGLUT1-VAMP2: 0 . 8±0 . 02 , n=7 neurons; VGLUT1-CAPS-1: 0 . 4±0 . 05 , n=7 neurons , **p<0 . 0001 ) . ( D ) Mander's coefficients for the proportion of CAPS-1 immuno-reactivity in VGLUT1 positive locations: 0 . 67±0 . 08 , n=8 neurons or proportion of VGLUT1 immunoreactivity in CAPS-1 positive locations: 0 . 64±0 . 05 , n=8 neurons . ( E ) Example images of neurites from hippocampal neurons ( DIV 14 ) stained for endogenous CAPS-1 ( green , left panel ) and VGLUT1 ( red ) or for Munc13-1 ( M13-1 , green , right panel ) and VGLUT1 ( red ) . CAPS-1 domains not overlapping with VGLUT1 ( filled arrowheads ) , VGLUT1 puncta not enriched for CAPS-1 ( open arrowheads ) . Synapses ( VGLUT1 puncta ) overlapping with CAPS-1 rich domains ( stars ) . Scale bar 5 µm . ( F ) CAPS-1 co-localization with VGLUT1 in the entire neuron is lower compared to co-localization of Munc13-1 and VGLUT1 . Pearson's correlation M13-1-VGLUT1: 0 . 6±0 . 04 , n=10; CAPS-1-VGLUT1: 0 . 4±0 . 08 , n=12 , p<0 . 05 . ( G ) Percentage of VGLUT1 labeled synapses expressing CAPS-1 is lower than VGLUT1 labeled synapses expressing ( VGLUT1/CAPS-1: 58 . 5±10 . 9% , n=7 neurons , number of synapses = 239; VGLUT1/M13-1: 96 . 2±0 . 7% , n=4 neurons , number of synapses = 350 ) . ( H ) Example images of neurites from hippocampal neurons ( DIV 14 ) infected with lentivirus encoding NPY-Venus ( green ) and stained for chromogranin B ( ChgB , red ) . Scale bar 2 µm . ( I ) Mander's coefficients for the proportion of endogenous ChrB immuno-reactivity in NPY-Venus puncta: 0 . 97±0 . 02 , n=14 neurons or proportion of NPY-Venus immunoreactivity in ChrB puncta: 0 . 84±0 . 01 , n=14 neurons . ( J ) Example images of neurites from hippocampal neurons ( DIV 14 ) infected with lentivirus encoding NPY-Venus and stained for CAPS-1 ( red ) and MAP2 ( blue ) . Scale bar 2 µm . ( K ) Quantification of co-localization of CAPS-1 and NPY-Venus in the entire neuron . Pearson's coefficient: 0 . 45±0 . 05; n=21 neurons . ( L ) Percentage of NPY-Venus labeled DCVs co-localizing with CAPS-1: 34 . 71±3 . 03% , n=6 , number of DCVs = 414 . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 00310 . 7554/eLife . 05438 . 004Figure 1—figure supplement 1 . Specificity of CAPS-1 antibody . ( A ) CAPS-1 immunoreactivity ( green ) and co-localization with synaptobrevin ( VAMP , red ) in CAPS-2 KO neurons stained for the dendritic marker MAP2 ( blue ) ( top panel ) . Bottom panel: absence of immunoreactivity in CAPS DKO neurons . Scale bar 10 μm . ( B ) CAPS-1 immunoreactivity ( green ) and co-localization with VGLUT1 ( red ) . Bottom panel: zoom indicated by the box in top panel . Scale bar 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 00410 . 7554/eLife . 05438 . 005Figure 1—figure supplement 2 . Endogenous DCV marker ( chromogranin B ) distribution in isolated single neurons . ( A ) Confocal image of a WT hippocampal neuron ( DIV 14 ) labeled with antibodies for the dendritic marker MAP2 ( green ) , endogenous DCV cargo protein ChgB ( red ) and the synaptic marker VGLUT1 ( blue ) . A punctate distribution of ChgB is seen in dendritic ( arrow ) and axonal regions ( MAP2-negative; arrowheads ) and accumulations of ChgB are found in growth cones ( asterisks ) . ( B ) Higher magnification of boxed area 1 ( zoom 1 ) shows ChgB puncta in axonal regions . Zoom 2 shows co-localization of ChgB and VGLUT1 , indicating the presence of DCVs in synapses . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 005 CAPS proteins are implicated in catecholamine uptake into secretory vesicles in chromaffin cells ( Speidel et al . , 2005 ) . As a defect in vesicle loading could influence our analysis of release events , we first analyzed possible effects of CAPS deletion on the loading of proteins into DCVs . The fluorescence intensity distribution of single DCV release events measured with the DCV cargo Semaphorin-3A coupled to pH-sensitve GFP ( SemapHluorin , see below ) in WT and CAPS DKO cells was similar; indicating that SemapHluorin loading of fusing DCVs was not affected in CAPS DKO neurons ( Figure 2A ) . To analyze SemapHluorin loading of all DCVs in the cell , we quantified the fluorescence intensity of SemapHluorin labeled DCVs upon application of NH4+ ( which instantly de-quenches all intra-vesicular pHluorin , Figure 2B ) and of antibody-labeled endogenous DCV cargo protein , secretogranin II , in WT and CAPS DKO neurons . NH4+ application showed that loading of SemapHluorin was comparable between WT and CAPS DKO neurons ( Figure 2C , D ) . Also , fluorescence intensity levels of secretogranin II were unchanged between WT and CAPS DKO neurons ( Figure 2E ) . As neuronal viability is affected in CAPS-2 null mutant cerebellum ( Sadakata et al . , 2004 , 2007a ) , we analyzed neuronal morphology and DCV numbers in CAPS DKO neurons and did not find differences between WT and DKO neurons ( Figure 2F–H ) . Hence , neuronal morphology , DCV biogenesis and protein loading in hippocampal neurons are not affected by deletion of CAPS-1 and CAPS-2 . 10 . 7554/eLife . 05438 . 006Figure 2 . CAPS deletion does not influence DCV peptide loading or DCV biogenesis in hippocampal neurons . ( A ) Frequency distribution of fluorescence intensity increase ( ΔF ) of individual DCV fusion events in WT and CAPS DKO neurons . No major differences are observed in ΔF of fusion events between WT and CAPS DKO neurons indicating similar Semaphluorin loading per DCV in WT and CAPS DKO neurons . The number of DCVs per bin is normalized to the total number of DCVs released . ( B ) Inverted wide-field image of a neuron expressing Semaphluorin upon NH4+ application to reveal all DCVs present in the cell . Zoom shows the effect of vesicle de-acidification upon NH4+ application ( −NH4+ before application , +NH4+ during application ) . ( C ) Average number of DCV puncta per neuron quantified from the NH4+ response is similar in WT ( n=16 neurons ) and CAPS DKO ( n=24 ) neurons . ( D ) Average intensity ( in arbitrary units , AU ) of single DCV puncta quantified from the NH4+ response in WT and CAPS DKO neurons is similar ( WT n=16 neurons and 4435 puncta , CAPS DKO n=24 neurons and 4892 puncta ) . ( E ) Average intensity ( in AU ) of single DCV puncta in the field of view of confocal images is similar in non-transfected WT ( n=10 neurons ) and CAPS DKO ( n=10 ) neurons stained for the endogenous DCV cargo secretogranin II and the dendritic marker MAP2 . ( F ) Average number of DCV puncta per field of view . ( G ) Average dendritic length per field of view . ( H ) Number of DCV puncta per dendritic length . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 006 To examine the function of CAPS proteins in DCV secretion we used two different fluorescent DCV cargo proteins in two secretion assays , hippocampal mass cultures and isolated single neuron cultures . First , DCVs were labeled with the secreted axon-guidance protein Semaphorin 3A coupled to the pH-sensitive enhanced green fluorescent protein ( EGFP ) variant , pHluorin ( SemapHluorin , Figure 3A , B ) and release was measured in CAPS-1/CAPS-2 DKO ( Jockusch et al . , 2007 ) and WT hippocampal neurons in DIV 14 mass cultures ( Figure 3A ) . We have previously shown that SemapHluorin labels all DCVs in cultured neurons and reports single DCV fusion events as a sudden increase in fluorescence upon opening of the fusion pore ( de Wit et al . , 2009; van de Bospoort et al . , 2012 ) . DCV release was triggered by electrical stimulation using a protocol known to elicit robust neuropeptide release from neurons ( 16 bursts of 50 action potentials at 50 Hz , Hartmann et al . , 2001; de Wit et al . , 2009; van de Bospoort et al . , 2012 ) . Figure 3B top panel shows a typical fusion event reported by SemapHluorin: upon fusion pore opening intravesicular pH rises sharply unquenching pHluorin , which results in a strong increase in fluorescence . Fluorescence increase of two standard deviations above initial fluorescence ( Figure 3B , grey dotted line ) was scored as fusion event in panels C and E . After this increase , fluorescence may remain high ( left trace , persistent event ) or may decay ( middle and right trace , transient events ) . Both persistent and transient events were counted as fusion events in panels C and E . Figure 3—figure supplement 1 explains the different fusion modes reported by SemapHluorin in more detail ( see also de Wit et al . , 2009 ) . 10 . 7554/eLife . 05438 . 007Figure 3 . CAPS deletion reduces DCV fusion in hippocampal neurons . ( A ) Schematic drawing of a neuronal mass culture used in these experiments in which one neuron expresses a fluorescently tagged morphology marker ( green ) and a DCV marker ( red ) . Non-labeled neurons are indicated in light blue . ( B ) Top panel: two stills showing a typical example of a fusion event of a SemapHluorin labeled DCV with the corresponding schematic drawing . Fusion pore opening causes a sudden increase of fluorescence intensity corresponding to the increase in intravesicular pH . Bottom panel: example traces of SemapHluorin labeled DCV fusion events . Fluorescence increase of two standard deviations above initial fluorescence ( grey dotted line ) was scored as fusion event in C and E . After which , fluorescence may decrease ( transient events ) or remain high ( persistent events ) . Figure 3—figure supplement 1 explains this behavior in detail . Scale bar 1 μm . ( C ) Average number of DCV fusion events per field of view during electrical stimulation with 16 bursts of 50 AP at 50 Hz with 0 . 5 s interval ( WT: 23 . 7±4 . 5 , n=24 neurons; CAPS DKO: 9 . 6±1 . 5 , n=32 neurons , N=4 , independent experiments , *p<0 . 01 ) . ( D ) Cumulative frequency plot of DCV fusion events during stimulation ( average vesicle fusion rate per cell during stimulation WT: 1 . 2±0 . 4 vesicles/s; CAPS-1/2 DKO: 0 . 15±0 . 1 vesicles/s ) . Blue bars represent stimulation of 16×50 AP at 50 Hz with 0 . 5 s interval . ( E ) Number of DCV fusion events per field of view during the first four bursts of the stimulation in C , ( WT: 4 . 8±1 . 5 , n=24; CAPS DKO: 1 . 4±0 . 3 , n=32 , p<0 . 05 ) . ( F ) Cumulative frequency plot of DCV fusion events during the first four bursts of the stimulation , showing that the initial release rate is slower in CAPS DKO neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 00710 . 7554/eLife . 05438 . 008Figure 3—figure supplement 1 . Different fusion events reported by SemapHluorin . Cartoon depicting the different fusion events reported by SemapHluorin and analyzed in depth in de Wit et al . , 2009 . Electrical stimulation triggers calcium and SNARE protein-dependent membrane fusion . Opening of the fusion pore results in a sudden dequenching of vesicular SemapHluorin and increase in fluorescence ( asterisk ) . This is scored as fusion event in Figure 3C , vE . Upon the sudden increase in fluorescence the signal either remains high ( A: persistent event ) or dims to baseline ( B: transient event ) . Transient events represent incomplete release followed by vesicle retrieval and re-acidification ( B1 ) or full release followed by cargo diffusion ( B2 ) . In B2 the vesicle may integrate into the plasma membrane or re-seal without SemapHluorin cargo ( B2 , grey box top and bottom panel , respectively ) . Persistent events reflect the continuous presence of cargo at the cell surface and may either represent stable deposits of SemapHluorin or vesicles with a permanently open fusion pore ( A , grey box top and bottom panel , respectively ) . Persistent events are typical for cargo that interacts with the intra-vesicular matrix like Semaphorin . These events are very rare when using NPY as cargo . Our previous analysis of fusion events reported by SemapHluorin showed that 60% of fusion events are transient events of which half result in full release and 40% of events are persistent ( de Wit et al . , 2009 ) . The typical example traces show persistent and transient fusion events . Both types of events were scored as fusion events in Figure 3C , E when fluorescence increased above two times the standard deviation of the initial fluorescence ( grey dotted line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 008 Upon stimulation , CAPS DKO neurons showed a more than 60% reduction in the number of DCV fusion events compared to WT ( Figure 3C , WT: 23 . 7±4 . 5 events/field of view , n=24; CAPS DKO: 9 . 6±1 . 5 events/field of view , n=32 , N=4 , p<0 . 01 ) . The average vesicle fusion rate during stimulation , calculated from the cumulative release plots ( Figure 3D ) , was 1 . 2±0 . 4 vesicles/s for WT compared to 0 . 15±0 . 1 vesicles/s for CAPS DKO neurons . Also when we zoomed in on the first four bursts of 50 action potentials , the number of fusion events was significantly lower in CAPS DKO neurons compared to WT ( Figure 3E , F ) . Together , these findings show that deletion of both CAPS isoforms strongly reduces activity-dependent secretion of neuronal DCVs by decreasing the total number of fusing vesicles and vesicle release rate during stimulation . Mass cultures can be used to assess the average number of release events per field of view , not per neuron . Therefore , we adapted a single neuron culture protocol using micro-islands of astrocytes ( Bekkers and Stevens , 1991; Wierda et al . , 2007 ) , which enables imaging of the entire axonal and dendritic arbor of a single neuron to perform quantitative single-cell DCV release measurements ( Figure 4A ) . DCVs were labeled with NPY-pHluorin , and vesicle fusion was triggered by electrical stimulation ( as in Figure 3 ) . As CAPS-2 expression in hippocampal neurons is very low and deletion of CAPS-2 does not affect SV release ( Jockusch et al . , 2007 ) , we used neurons from CAPS-2 null mutant ( CAPS-2KO ) littermates as controls . The number of DCV fusion events per cell reported by the sudden increase in fluorescence upon fusion pore opening was strongly decreased in CAPS DKO compared to controls ( Figure 4B , C , CAPS-2KO: 49 . 6±16 . 3 events/cell , n=9; CAPS DKO: 10 . 6±7 . 9 events/cell , n=5 , N=3; p<0 . 05 ) . Hence , CAPS-1 deletion strongly inhibits DCV fusion in isolated hippocampal neurons . 10 . 7554/eLife . 05438 . 009Figure 4 . CAPS-1 deletion impairs DCV fusion in isolated neurons . ( A ) Schematic drawing of a single isolated neuron grown on a micro island of astrocytes and expressing fluorescently tagged morphology marker ( green ) and a synapse marker ( red ) in addition to either NPY-pHluorin or NPY-mCherry ( not shown ) . Average island diameter is 375 μm . ( B ) Average number of DCV fusion events per cell upon electrical stimulation of 16 bursts of 50 AP at 50 Hz using NPY-pHluorin as DCV marker ( CAPS-2KO: 49 . 6±16 . 3 , n=9; CAPS DKO: 10 . 6±7 . 9 , n=5 , number of independent experiments ( N ) =3 , p<0 . 05 ) . Inset shows a typical example of a fusion event reported by NPY-pHluorin . Fluorescence increase of two standard deviations above initial fluorescence ( grey dotted lines ) was scored as fusion event in B . ( C ) Cumulative frequency plot of DCV fusion events in B , showing that DCV release is triggered by electrical stimulation paradigm ( blue bars represent 16 bursts of 50 AP at 50 Hz , 16×50 AP at 50 Hz ) . ( D ) White arrowhead: Typical example of a fusion event with complete cargo release reported by the sudden and complete disappearance of NPY-mCherry fluorescence intensity . These fusion events were in measured in E and K . Time in seconds ( s ) after start of stimulation . Neurite marker is ECFP ( blue ) . Scale bar 1 μm . ( E ) Average number of fusion events with complete cargo release per cell upon electrical stimulation using NPY-mCherry as DCV marker ( CAPS-2KO: 14 . 6±3 . 3 , n=14; CAPS DKO: 3 . 4±1 . 4 , n=8 , N=3 , p<0 . 01 ) . ( F ) Cumulative frequency plot of DCV fusion events with complete cargo release in E , ( blue bars represent 16 bursts of 50 AP at 50 Hz , 16×50 AP at 50 Hz ) . ( G ) Percentage of non-secreting cells is increased in CAPS DKO neurons ( CAPS-2KO: 15±12 . 4% , n=26; CAPS DKO 53 . 6±12 . 6% , n=26 , p<0 . 05 ) . Non-secreting cells were excluded from the analyses in B and E . ( H ) Example images showing NPY-pHluorin labeled DCVs fusing at synapsin-mCherry labeled synapses , indicated by the red dashed lines and DCV fusion events at extra-synaptic sites , indicated by the green dashed lines . Bar is 1 μm . ( I ) Percentage of synaptic and extra-synaptic DCV release events measured with NPY-pHluorin shows a similar distribution in CAPS DKO compared to CAPS-2KO ( CAPS-2KO synaptic: 67 . 0±3 . 6 , CAPS-2KO extra-synaptic: 0 . 32±1 . 4 , n=20 , p<0 . 01; CAPS DKO synaptic: 64 . 7±2 . 3 , CAPS DKO extrasynaptic: 35 . 6±0 . 9 , n=28 , p<0 . 01 ) . ( J ) CAPS-1 expression levels assessed by semi-quantitative immunofluorescence in CAPS-2KO , CAPS DKO , and CAPS DKO expressing CAPS-1 ( Rescue ) ( CAPS-2KO: 2 . 8±0 . 7 AU , n=5; CAPS DKO: 0 . 1±0 . 0 , n=4; Rescue: 2 . 4±0 . 7 , n=3 , **p<0 . 0001 ) . ( K ) Average number of fusion events leading to complete release per cell upon electrical stimulation with NPY-mCherry as DCV marker in CAPS-2KO , CAPS DKO and Rescue ( CAPS-2KO: 8 . 5±1 . 7 , n=25; CAPS DKO: 0 . 7±0 . 3 , n=16; Rescue: 6 . 4±1 . 7 , n=23 , N=6 ) . ( L ) Cumulative frequency plot of complete release DCV fusion events in K , ( blue bars represent 16 bursts of 50 AP at 50 Hz , 16×50 AP at 50 Hz ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 009 DCV fusion pore opening can progress to complete release of cargo or to partial release and fusion pore resealing ( Alabi and Tsien , 2013; Wu et al . , 2014 ) . To investigate these release modes in CAPS-2 KO and CAPS DKO neurons , we labeled DCVs with NPY-mCherry , which reports full cargo release events as complete disappearance of fluorescent puncta ( Figure 4D ) . The average number of such fusion events in control cells was much lower compared to events reported by NPY-pHluorin , which reports all fusion events irrespective of complete or incomplete release of cargo ( compare Figure 4E and Figure 4B , NPY-mCherry CAPS-2KO: 14 . 6±3 . 2 events/cell; NPY-pHluorin CAPS-2KO: 49 . 6±16 . 3 events/cell ) . However , also with this reporter a more than 50% decrease in the number of fusion events per cell was observed in CAPS DKO neurons compared to CAPS-2KO ( Figure 4E , F , CAPS-2KO: 14 . 6±3 . 2 events/cell , n=14; CAPS DKO: 3 . 4±1 . 3 events/cell , n=8 , N=3 , p<0 . 01 ) . In addition to the strong decrease in DCV fusion events in CAPS DKO neurons , in almost 50% of CAPS DKO cells expressing either NPY-pHluorin or NPY-mCherry electrical stimulation failed to induce vesicle fusion ( Figure 4G , these cells were excluded from analysis in Figure 4B–F ) . We previously showed that DCVs preferentially fuse at synapses and that Munc13 plays a crucial role in synaptic preference of DCV release ( van de Bospoort et al . , 2012 ) . To test whether CAPS deletion affected synaptic preference of DCV fusion we labeled presynaptic terminals with the live cell marker synapsin-mCherry and analyzed NPY-pHluorin fusion events ( Figure 4H ) . Unlike Munc13 deletion , deletion of CAPS affected extra-synaptic and synaptic DCV release to the same extent ( Figure 4I ) . To confirm that the secretion defect in CAPS DKO neurons was due to lack of CAPS-1 , CAPS-1 was re-introduced in CAPS DKO cells at 10 DIV via lenti-virus infection ( ‘rescue’ ) and DCV release was tested 4 days later . CAPS-1 expression levels in rescued neurons were comparable to endogenous levels ( Figure 4J ) . Introduction of CAPS-1 in CAPS DKO cells rescued the secretion defect ( Figure 4K ) and release kinetics were similar to control cells ( Figure 4L ) . Thus , deletion of CAPS-1 resulted in a strong reduction of DCV fusion events reported by NPY-PHluorin or NPY-mCherry and fusion at synapses and extra-synaptic sites was affected to the same extent . DCVs can be highly mobile or stationary ( de Wit et al . , 2006; Wong et al . , 2012; Goodwin and Juo , 2013 ) . Both stationary and mobile vesicles fused upon electrical stimulation ( Figure 5A ) . In control neurons the majority of DCV fusion events were stationary DCVs ( Figure 5B , CAPS-2KO: stationary 12 . 2±2 . 8 fusion events per cell , moving 4 . 0±1 . 0 fusion events per cell , n=10 , N=3 , p<0 . 05 ) . However , in CAPS DKO neurons , release of stationary vesicles was strongly decreased while release of mobile vesicles was similar to CAPS-2KO ( Figure 5B , CAPS DKO: stationary 2 . 0±0 . 8 fusion events per cell , moving 3 . 0±1 . 1 events per cell , n=5 , N=2 ) . 10 . 7554/eLife . 05438 . 010Figure 5 . Deletion of CAPS-1 affects fusion of stationary DCVs . ( A ) Stills from DCV fusion assay . Electrical stimulation starts at second 10 . Yellow arrowheads indicate stationary vesicles that fuse , red arrowhead indicates a stationary vesicle that does not fuse and white arrow shows a moving DCV that fuses . Kymograph shows the trajectories of the stationary and moving vesicle over time . ( B ) Fusing DCVs classified as stationary or moving showing that CAPS deletion strongly affects fusion from stationary vesicles ( CAPS-2KO stationary prior to fusion: 12 . 2±2 . 8 , moving: 4±1 . 0 , n=10 , p<0 . 05; CAPS DKO stationary: 2 . 3±1 . 3 , moving: 4 . 0±1 . 7 , n=3 , ns , N=3 ) . ( C ) Non-fusing DCVs classified as stationary or moving showing that CAPS deletion does not affect general trafficking behavior of non-fusing vesicles ( CAPS-2KO stationary: 33 . 8±1 . 2 , moving: 17 . 6±0 . 7 , n=5 , p<0 . 05; CAPS DKO stationary: 33 . 7±3 . 0 , moving: 19 . 3±2 . 4 , n=3 , p<0 . 05 ) . ( D ) Average velocity of DCVs classified as moving prior to fusion in B , before stimulation ( PreStim = 30 s; CAPS-2KO: 193 . 6±9 . 9 nm/s , n=169 vesicles , CAPS DKO: 180 . 4±9 . 2 nm/s , n=18 ) and during electrical stimulation ( Stim: from second 30 to the onset of fusion; CAPS-2KO = 180 . 9±6 . 6 nm/s , n=169; CAPS DKO: 169 . 2±7 . 5 , n=18 ) . ( E ) Average velocity of non-fusing DCVs in C , before stimulation ( PreStim = 30 s; CAPS-2KO: 190 . 8±13 . 7 nm/s , n=168 , CAPS DKO 186 . 8±13 . 5 nm/s , n=91 ) , during stimulation ( Stim = 24 s; CAPS-2KO: 175 . 3±6 . 7 nm/s , n=168; CAPS DKO 201 . 8±11 . 8 nm/s , n=91 ) and after stimulation ( PostStim = 36 s CAPS-2KO: 182 . 2±6 . 2 nm/s , n=168 , CAPS DKO 196 . 2±8 . 5 nm/s , n=91 ) . ( F ) Typical examples of electron micrographs of neuronal DCVs in synapses . Scale bar 50 nm . ( G ) Number of synapses containing one or more DCVs ( CAPS-2KO: 53 . 9±12 . 1% , n=198 synapses; CAPS DKO 47 . 6±5 . 4% , n=152 synapses , N=4 ) . ( H ) Percentage of docked DCVs per synapse ( CAPS-2KO: 10 . 3±1 . 7% , DCVs = 285 , CAPS DKO: 11 . 1±3 . 7% , DCVs = 201 ) . ( I ) Average distance of DCVs to the closest plasma membrane ( CAPS-2KO: 81 . 2±4 . 8; CAPS DKO: 84 . 9±10 . 4 ) . ( J ) Average number of DCVs per synapse ( CAPS-2KO: 1 . 6±0 . 2; CAPS DKO: 1 . 5±0 . 2 ) . ( K ) Frequency distribution of number of DCVs per synapse ( % synapses , normalized to the total number of DCVs per group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 01010 . 7554/eLife . 05438 . 011Figure 5—figure supplement 1 . DCV docking definition and zooms of Figure 5F . Zooms of three example DCVs showing the classification of docked versus undocked vesicles . We define a docked DCV as a vesicle containing a dense core that has no detectable distance between the vesicular membrane and the plasma membrane . The minimal distance between membranes we can visualize is 0 . 52 nm . ( A ) Electron micrograph showing an example of docked DCV . ( B ) Electron micrograph showing an example of DCV close but not docked to the plasma membrane . ( C ) Electron micrograph showing an example DCV far away to the plasma membrane . Scale bars 100 nm . ( D ) Enlarged high-resolution electron microscopy images of Figure 5F . Scale bar 50 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 011 To test if CAPS deletion affected general DCV trafficking , we analyzed the dynamics of DCVs that did not fuse during the 90s imaging protocol . No differences between CAPS-2KO and CAPS DKO neurons were observed for these vesicles ( Figure 5C , CAPS-2KO: stationary 33 . 8±1 . 2 , moving: 17 . 6±0 . 7 , n=5 , p<0 . 05; CAPS DKO: stationary 33 . 66±3 . 0 , moving 19 . 33±2 . 4 , n=3 , p<0 . 05 ) . We also did not detect differences in average velocity of fusing and non-fusing DCVs prior to stimulation , during stimulation or in the period after stimulation ( Figure 5D , E ) . CAPS/UNC-31 deletion reduces the number of docked DCVs in C . elegans neuromuscular junctions ( Hammarlund et al . , 2008 ) while the total number of DCVs in mammalian synapses is not different between control and CAPS DKO ( Jockusch et al . , 2007 ) . We examined electron micrographs of synapses from CAPS DKO and control neurons to test if DCV localization was affected by deletion of CAPS . No morphological differences between the two groups were observed ( Figure 5F and Figure 5—figure supplement 1D ) . The number of synapses containing DCVs ( Figure 5G ) , the percentage of docked DCVs ( Figure 5H , see Figure 5—figure supplement 1A–C for our definition of docked vesicles ) , the distance of DCVs to the closest plasma membrane ( Figure 5I ) , and the average number of DCVs per synapse ( Figure 5J , K ) were similar between CAPS DKO and control synapses . Thus , deletion of CAPS-1 does not affect transport of DCVs or DCV localization at synapses but specifically affects fusing vesicles , by reducing the number of fusion-competent stationary vesicles . To investigate the intracellular dynamics of CAPS-1 , we introduced CAPS-1 fused to enhanced yellow fluorescent protein ( EYFP ) ( CAPS-1-EYFP ) in CAPS DKO neurons ( Figure 6A ) . As for endogenous CAPS-1 ( Figure 1 ) , the majority of CAPS-1-EYFP co-localized with the live-synaptic marker synapsin-ECFP ( Figure 6A , bottom panel , stars ) . In addition , we observed , again similar to endogenous CAPS-1 , synapses without detectable CAPS-1-EYFP ( Figure 6A , bottom panel , open arrowhead ) and CAPS-1 rich domains that did not co-localize with synapses ( Figure 6A , bottom panel , filled arrowhead ) . 10 . 7554/eLife . 05438 . 012Figure 6 . CAPS-1-EYFP fusion protein replaces CAPS-1 in DCV secretion and synaptic transmission . ( A ) Isolated single CAPS DKO neuron grown on glia micro-island expressing CAPS-1-EYFP ( green ) and synapsin-ECFP ( blue ) . Top panels: maximum projection of the entire neuron , Scale bar 10 µm . Bottom panels: zooms of top panels showing synapses without detectable CAPS-1-EYFP ( open arrowhead ) , CAPS-1 rich domain not overlapping with synapsin-ECFP ( filled arrowhead ) and a CAPS-1 rich synapse ( star ) . Scale bar 5 µm . ( B ) Average number of DCV fusion events per cell ( events/cell ) upon electrical stimulation ( 16×50 AP at 50 Hz ) using NPY-pHluorin as DCV marker in CAPS DKO neurons with ( +CAPS-1-EYFP ) and without ( CAPS DKO ) CAPS-1-EYFP ( +CAPS-1-EYFP: 23 . 0±6 . 8 , n=8 , CAPS DKO: 0 . 7±0 . 3 , n=16 , *p<0 . 0001 , Mann–Whitney test ) . ( C ) Example traces of evoked EPSCs in CAPS-2KO ( black ) , CAPS DKO ( dark green ) and CAPS-1-EYFP ( light green ) rescued CAPS DKO neurons . ( D ) Expression of CAPS-EYFP rescues EPSC amplitude in CAPS DKO neurons ( CAPS-2KO: 2 . 4±0 . 3 nA , n=4; CAPS DKO: 0 . 48±0 . 15 nA , n=3; CAPS-1-EYFP: 2 . 0±0 . 4 nA , n=3 , p<0 . 01 ) . ( E ) Example traces of spontaneous release ( mEPSCs ) . ( F ) Mean mEPSC frequency . ( CAPS-2KO: 7 . 5±1 . 1 Hz , n=3; CAPS DKO: 1 . 7±1 . 0 Hz , n=3; CAPS-1-EYFP: 6 . 9±1 . 3 Hz , n=2 ) . ( G ) Mean mEPSC amplitude . ( CAPS-2KO: 29 . 0±4 . 1 pA , n=3; CAPS DKO: 27 . 1±5 . 0 pA , n=3; CAPS-1-EYFP: 26 . 8±7 . 2 pA , n=2 ) . ( H ) Changes in EPSC amplitude induced by 100-pulse train at 40 Hz during low frequency ( 0 . 1 Hz ) stimulation . The interval between low- and high-frequency stimulation is 3 s ( I ) 100-pulses at 40 Hz induced rundown of normalized EPSC amplitude ( zoom of H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 012 CAPS-1-EYFP efficiently rescued DCV secretion in CAPS DKO neurons ( Figure 6B ) and also rescued synaptic transmission . Whole cell patch-clamp recordings of single neurons on micro-dot astrocyte islands revealed no differences between control and CAPS-1-EYFP rescued CAPS DKO neurons . Evoked postsynaptic current amplitude ( EPSC , Figure 6C , D ) and spontaneous release characteristics ( mEPSC frequency and amplitude , Figure 6E–G ) were similar to controls . Also EPSC rundown during high-frequency stimulation ( 40 Hz ) and its recovery ( Figure 6H , I ) were completely rescued . Hence , CAPS-1-EYFP mimics endogenous CAPS-1 and was used for further experiments to study the dynamics of this protein in living cells . We studied the dynamics of CAPS-1-EYFP upon electrical stimulation using the same stimulation paradigm used to elicit DCV release ( 16×50 AP at 50 Hz ) . We infected hippocampal neurons with lentiviral particles encoding CAPS-1-EYFP and synapsin-ECFP and imaged the two fluorophores simultaneously ( Figure 7A , B ) . During stimulation , the fluorescence intensity of CAPS-1-EYFP and synapsin-ECFP at synapses strongly decreased while the extra-synaptic intensities of both proteins increased ( Figure 7A–F ) . A large variability of intensity changes upon stimulation was observed between cells . In some neurons fluorescence intensity of synapsin-ECFP and CAPS-1-EYFP returned to baseline within 3 min ( Figure 7C , D ) , while in other cells the intensities of both proteins remained below their initial fluorescence ( Figure 7E , F ) . Fluorescence intensities of membrane-bound EYFP did not change at synapses during stimulation showing that the decrease of CAPS-1-EYFP fluorescence at synapses reported dispersion of the protein into neurites ( Figure 7D , F , insets ) . The average response of synaptic CAPS-1-EYFP showed a similar dynamic profile as synapsin-ECFP , dispersing from synapses during and up until 3 min after calcium influx , while extra-synaptic CAPS-1-EYFP fluorescence increased during stimulation ( Figure 7G , H ) . Thus , during stimulation a fraction of synaptic CAPS-1 redistributes from synapses into neurites similar to the synaptic vesicle protein synapsin 1 . 10 . 7554/eLife . 05438 . 013Figure 7 . Localization of CAPS-1 at synapses is calcium dependent . ( A and B ) Grey scale images of synapsin-ECFP ( left ) and CAPS-1-EYFP ( right ) labeled synapses showing the same region before ( pre-15s ) during ( stim-22s ) and after ( post-150s ) electrical stimulation ( 16×50 AP at 50 Hz ) . Synapsin-ECFP and CAPS-1-EYFP were imaged simultaneously at 0 . 5 Hz . ( C and E ) Traces of relative intensity changes ( ΔF/F0 ) of synapsin-ECFP at synapses and extra-synaptic locations ( 79 synaptic and 31 extra-synaptic ) showing increased extra-synaptic and decreased synaptic fluorescence upon stimulation ( 16×50 AP at 50 Hz ) . Open arrowheads in C indicate the pre- and post stimulations time points ( same for D–F ) . ( D and F ) Example traces of relative intensity changes ( ΔF/F0 ) of CAPS-1-EYFP at synapsin-ECFP labeled synapses from C . showing increased extra-synaptic and decreased synaptic fluorescence upon stimulation ( 16×50 AP at 50 Hz ) . Inset: synaptic fluorescence of membrane associated EYFP ( EYFP control ) as control . ( G ) Average relative intensity profiles of synapsin-ECFP , and CAPS-1-EYFP at synapses and CAPS-1-EYFP extra-synaptic , ( 395 synaptic regions , 155 extrasynaptic regions , n=5 cells ) . ( H ) Maximum relative intensity changes ( max ΔF/F0 ) of synapsin-ECFP , and CAPS-1 at synapses and CAPS-1 at extra synaptic regions at t=160 s calculated from G ( p<0 . 01 , n=5 cells each ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 013 We showed that on a cellular level , removing CAPS-1 strongly impairs DCV fusion ( Figure 3 and Figure 4 ) . To test whether at the single synapse level CAPS expression levels correlated with DCV release probability we simultaneously imaged NPY-mCherry labeled DCVs and CAPS-1-EYFP ( Figure 8A ) . Like endogenous CAPS-1 , CAPS-1-EYFP was present in puncta ( Figure 8A , left top panel ) . The majority of these were stationary throughout the imaging experiment ( Figure 8A , B ) . Only 5% of all CAPS-1-EYFP puncta were mobile during our imaging paradigm ( Figure 8B ) . Furthermore , the number of mobile CAPS-1-EYFP puncta that co-trafficked with dynamic NPY-mCherry labeled DCVs was very low ( seven out of 106 mobile DCVs showed co-trafficking of CAPS-1-EYFP , Figure 8C ) . 10 . 7554/eLife . 05438 . 014Figure 8 . Presence of CAPS-1 increases DCV release probability at single synapses . ( A ) Left top and bottom panels: kymographs of CAPS-1-EYFP and NPY-mCherry imaged simultaneously ( acquisition frequency 2 Hz ) for 90 s and stimulated at second 30 ( dashed box , 16×50 AP at 50 Hz ) . Note the dispersion of the majority of CAPS-1-EYFP puncta upon stimulation and the disappearance of NPY-mCherry puncta upon stimulation . Top right panel shows the merge of the two channels . The bottom right panel shows a schematic drawing of the merged CAPS-1-EYFP and NPY-mCherry channels to aid in the interpretation of the kymographs . Open arrowhead indicates DCV not co-localizing with CAPS-1-EYFP . Filled arrowhead indicates a CAPS-1-EYFP punctum co-trafficking with a mobile DCV . Scale bar 5 µm . ( B ) Percentage of stationary and moving CAPS-1-EYFP puncta during image acquisition as described in ( A ) . ( Stationary: 96 . 6±1 . 7 , moving: 3 . 3±1 . 7 , number of cells = 9; number of kymographs per cell = 3 , total number of puncta = 116 ) . ( C ) Percentage of mobile DCVs co-trafficking with CAPS-1-EYFP ( mobile DCVs not co-trafficking with CAPS-1-EYFP ( green bar ) : 99 or 93 . 4%; mobile DCVs co-trafficking with CAPS-1-EYFP: 7 or 6 . 3% , total number of cells = 20 , moving DCVs analyzed = 106 ) . ( D ) Typical examples of synapsin-ECFP labeled synapses with high expression levels of CAPS-1EYFP ( star ) or low expression levels of CAPS-1-EYFP ( open arrowhead ) . ( E ) Percentage of DCV release events occurring at synapses enriched for CAPS-1 ( black bar ) or depleted for CAPS-1 ( green bar ) , ( synapses + CAPS-1: 84 . 0±3 . 3% , synapses − CAPS-1: 16 . 0±3 . 3% , total DCVs released = 84 , n=5 , p<0 . 0001 ) . ( F ) Cumulative frequency plot of the DCV release events in C . showing that release at CAPS-1 deficient synapses is reduced and delayed . Blue box represents 16×50 AP at 50 Hz stimulation . ( G ) Released DCVs categorized in stationary or moving before fusion ( synapses + CAPS-1: stationary DCVs = 46 , moving DCVs = 23 , total DCVs released = 69; synapse − CAPS-1: stationary DCVs = 6 moving DCVs = 8 , total DCVs released = 14 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05438 . 014 As CAPS-1 co-localizes with synaptic markers but is not enriched at every synapse ( Figure 1A–C , Figure 8D ) , we tested whether CAPS-1 positive synapses ( Figure 8D , star ) have a higher DCV release probability than CAPS-1-EYFP deficient synapses ( Figure 8D , open arrowhead ) . Synapses expressing CAPS-1-EFYP secreted more DCVs upon stimulation than synapses with no detectable CAPS-1-EYFP expression ( Figure 8E , percentage of DCVs released at CAPS-1-EYFP positive synapses: 84 . 0±3 . 3% , −CAPS-1: 16 . 0±3 . 1% , n=83 DCVs from five cells , **p<0 . 0001 ) . Furthermore , DCV fusion in synapses with no detectable CAPS-1-EYFP expression required more prolonged stimulation than in CAPS-1-EYFP positive synapses ( Figure 8F ) . Also , CAPS-1 enriched synapses released more DCVs from a stationary pool than from a mobile pool compared to CAPS-1 deficient synapses ( Figure 8G , CAPS-1-EYFP positive synapses: stationary DCVs: 46; moving DCVs: 23; total DCVs counted: 69; CAPS-1-EYFP negative synapses: stationary DCVs: 6; moving DCVs: 8; DCVs counted: 14 ) . Together these results show that DCVs preferentially reside and fuse at CAPS-1 rich domains and that CAPS-1-EYFP and DCVs rarely travel together in DIV 14 neurons . Furthermore , DCV release probability correlated with synaptic CAPS-1 expression levels . CAPS-1 was identified as a brain protein ( Walent et al . , 1992 ) that binds membranes associated with DCV secretion ( Berwin et al . , 1998 ) and stimulates calcium-dependent fusion of secretory vesicles in PC12 cells ( Mennerick et al . , 1995; Loyet et al . , 1998; Grishanin et al . , 2002 ) . This suggests that CAPS-1 might be a DCV resident protein that is delivered to sites of exocytosis on DCVs . However , we observed that in mammalian neurons endogenous CAPS-1 localized in axonal and dendritic domains that often , but not always , co-localized with synaptic and DCV markers . Cell-wide CAPS-1 co-localization with the canonical DCV marker NPY yielded an average co-localization of only 35% . Furthermore , in CAPS DKO neurons rescued with CAPS-1-EYFP , DCVs and CAPS-1-EYFP only rarely traveled together . Finally , the increased dynamics of synaptic CAPS-1-EYFP with CAPS-1 redistribution upon stimulation are best explained by diffusion ( see below ) but not with a redistribution of vesicular CAPS-1 . Thus , in mature mammalian neurons , CAPS-1 does not appear to be a general DCV-resident protein , but a dynamic , cytosolic factor translocating to synapses and extra-synaptic sites in an activity-dependent manner to promote fusion of pre-docked ( tethered ) DCVs . Figure 3 and Figure 4 show that in the absence of CAPS-1 , DCV fusion in hippocampal neurons is severely impaired , without defects in vesicle biogenesis and loading of endogenous or exogenous DCV cargo proteins . In addition , analysis of DCV dynamics in our live cell imaging experiments and intra-synaptic localization on electron micrographs of CAPS DKO neurons revealed that CAPS-1 deletion does not affect DCV trafficking along microtubules , synaptic accumulation and membrane docking of DCVs . Together , this suggests that in mature neurons most CAPS-1 molecules interact with DCVs after these vesicles arrived at the plasma membrane . This conclusion is consistent with observations in Drosophila neuromuscular junctions where deletion of dCaps results in an increase in the number of DCVs present at synapses ( Renden et al . , 2001 ) and the lack of a docking defect upon deletion of CAPS-1/-2 in adrenal chromaffin cells ( Liu et al . , 2010 ) . However , chemical fixation used in our study might produce artifacts that change the precise distance between vesicles and the membrane . As a consequence , vesicles touching the membrane in fixed tissue may in fact have been at a short distance from the membrane . Recent excellent work using cryo-fixation and EM tomography to analyse docking of SVs and DCVs in central synapses of several mutant mouse strains including CAPS DKO showed that SV docking is impaired in CAPS DKO neurons . In line with our findings , DCVs were present in similar numbers in CAPS DKO synapses . However , although not statistically significant , the distribution of DCVs within 200 nm of the active zone appeared to be reduced ( Imig et al . , 2014 ) . Hence , general consensus exists on a post-synapse delivery role for CAPS but future studies using cryo-fixation might unmask ( subtle ) docking defects upon CAPS loss . CAPS-1 likely functions at the final stages of DCV fusion , interacting with proteins and lipids that function in DCV docking , priming and fusion ( Grishanin et al . , 2004; Sadakata et al . , 2007b; James et al . , 2009; Parsaud et al . , 2013; Sah and Geracioti , 2013 ) . CAPS-1 changes the conformation of syntaxin-1 from a state incompatible with SNARE-complex formation ( ‘closed-state’ ) , to a state that allows formation of functional SNARE complexes ( ‘open-state’ ) , as release in CAPS DKO chromaffin cells is rescued by expression of an ‘open’-variant of syntaxin-1 ( Liu et al . , 2010 ) and overexpression of open syntaxin can bypass the requirement for CAPS in DCV docking in C . elegans ( Hammarlund et al . , 2008 ) . CAPS-1 shares this characteristic with Munc13-1 ( Augustin et al . , 1999; van de Bospoort et al . , 2012 ) . CAPS and Munc13 proteins may operate in the same molecular priming pathway ( Richmond et al . , 2001; Jockusch et al . , 2007; Zhou et al . , 2007 ) in a non-redundant manner . The fact that both proteins appear to bind syntaxin-1 via different binding modes may account for this non-redundancy ( Parsaud et al . , 2013 ) . A recent paper indeed showed that CAPS-2 and Munc13 use different mechanisms to prime vesicles , whereas Munc13-dependent priming requires its MUN domain this domain in CAPS-2 is dispensable for priming . Instead CAPS-2 appears to require its PIP2 binding pleckstrin homology domain ( Nguyen Truong et al . , 2014 ) . We found that both proteins play important stimulatory roles in DCV release from mammalian neurons . However , functional differences were evident: CAPS-1 deletion resulted in a larger reduction of DCV release events compared to deletion of Munc13-1 with a ±70% reduction in CAPS DKO ( this study ) and ±60% reduction in Munc13 DKO neurons ( van de Bospoort et al . , 2012 ) . For synaptic vesicle fusion the situation is opposite: Munc13 DKO neurons clearly show larger defects in synaptic transmission than CAPS DKO neurons ( Varoqueaux et al . , 2002; Jockusch et al . , 2007 ) . CAPS-1 displayed a different expression pattern than Munc13-1 . Munc13-1 expression is strictly synaptic and specifically supports DCV secretion from synapses , while being dispensable for extra-synaptic secretion ( van de Bospoort et al . , 2012 ) . In contrast , CAPS-1 accumulations are found at many synapses , but not all , and at extra-synaptic sites along neurites . Also , CAPS-1 deletion equally affects DCV release from synapses and from extra-synaptic sites . Finally , CAPS-1 redistributes from synapses during activity while Munc13-1 does not ( Kalla et al . , 2006 ) . These differences in ( re ) distribution suggest that ( 1 ) CAPS-1 domains at extra-synaptic sites promote DCV release independently of Munc13-1 . However , CAPS-1-dependent extra-synaptic release is less efficient than synaptic DCV release , which indicates that the concerted action of CAPS-1 and Munc13-1 is most efficient in priming DCVs for release . ( 2 ) At synapses , CAPS-1 and Munc13-1 have non-redundant roles in promoting DCV release , similar to their non-redundancy in SV release ( Jockusch et al . , 2007 ) . The larger effect of CAPS deletion on DCV release may be explained by the additional reduction of extra-synaptic release events compared to Munc13-1 deletion . Finally , some extra-synaptic DCV release remained in CAPS DKO neurons indicating that extra-synaptic DCV release can occur in the absence of CAPS and Munc13 albeit with very low release probability . Strong stimulation triggered redistribution of synaptic CAPS-1 into the axonal shaft . This redistribution was very similar to synapsin-ECFP ( our work and Richmond et al . , 2001; Parsaud et al . , 2013 ) and is also reported for other synaptic proteins like Rab3a ( Tsuriel et al . , 2009 ) , Syntaxin-1 ( Ribrault et al . , 2011 ) and Munc18 ( Cijsouw et al . , 2014 ) and for SVs ( Cheung and Cousin , 2011 ) . In contrast , the mobility of active zone proteins like Munc13-1 and Bassoon is not affected by acute stimulation ( Kalla et al . , 2006; Tsuriel et al . , 2009 ) . Hence , synapses rapidly exchange part of their components during high frequency stimulation . This allows synapses to adapt their release probability during and directly after stimulation as we showed for Munc18-1 ( Cijsouw et al . , 2014 ) . This is also an attractive explanation for the re-distribution of CAPS-1 as we found that synapses with increased CAPS-1 expression levels had a higher release probability than synapses with low/no CAPS-1 . CAPS-1 binds to PI 4 , 5-P2 ( PIP2 ) ( Loyet et al . , 1998 ) and localizes to PIP2 clusters in the plasma membrane via its PIP2-binding pleckstrin homology ( PH ) domain ( James et al . , 2008 ) . The PH domain is also required to prime secretory vesicles ( Kabachinski et al . , 2014; Nguyen Truong et al . , 2014 ) . Robust Ca2+ influx in our experiments likely activates phospholipase C , which hydrolysis PIP2 and may trigger CAPS-1 dispersion from synapses . Hence , calcium-dependent PIP2 hydrolysis may act as negative feedback mechanism reducing CAPS-1 availability at the synapse after robust stimulation . In addition to direct membrane interaction , CAPS-1 also binds syntaxin via its MUN domain . Syntaxin also disperses from synapses resulting in CAPS-1 co-dispersion . As Munc13-1 does not show activity-dependent redistribution among synapses but instead increases its membrane-bound fraction upon calcium influx , synapses appear to utilize two distinct mechanisms to control release probabilities during/after high frequency stimulation . SemapHluorin was generated by replacing EGFP in Sema3A-EGFP ( De Wit et al . , 2005 ) with super-ecliptic pHluorin ( SpH ) ( de Wit et al . , 2009 ) . NPY-Venus was previously described ( Nagai et al . , 2002 ) and NPY-SpH was generated by replacing Venus with SpH ( de Wit et al . , 2009 ) . Synapsin-mCherry was a kind gift of Dr A Jeromin ( Allen Brain Institute , Seattle , USA ) and synapsin-ECFP was obtained by replacing mCherry with ECFP . CAPS-1 ( KIAA1121-Kazusa DNA ) was sequence verified and cloned as CAPS-1-ires-EGFP and EYFP-CAPS-1 . All constructs but SemapHluorin were subcloned into pLenti vectors that were produced as described ( Naldini et al . , 1996 ) . Transduction efficiencies were tested on HEK cells . CAPS-1/2 double knockout mice have been described before ( Jockusch et al . , 2007 ) . Mouse embryos were obtained by caesarean section of pregnant females from timed mating . Animals were housed and bred according to institutional , Dutch and US governmental guidelines . Dissociated hippocampal neurons were prepared from embryonic day 18 mice as described ( de Wit et al . , 2009 ) . Hippocampi were dissected in Hanks buffered salts solution ( HBSS , Sigma , The Netherlands ) and digested with 0 . 25% trypsin ( Invitrogen , The Netherlands ) for 20 min at 37°C . Hippocampi were washed and triturated with fire-polished Pasteur pipettes , counted and plated in neurobasal medium ( Invitrogen ) supplemented with 2% B-27 ( Invitrogen ) , 1 . 8% HEPES , 1% glutamax ( Invitrogen ) and 1% Pen-Strep ( Invitrogen ) . High-density cultures ( 25 , 000 neurons/well ) were seeded on pre-grown cultures of rat glia cells ( 37 , 500 cells/well ) on 18 mm glass coverslips in 12-well plates . For micro-island culture originally described by ( Mennerick et al . , 1995 ) , hippocampal neurons were plated at a density of 2000 neurons/well of a 12-well plate on micro-islands of rat glia as in Wierda et al . , 2007 . These micro-islands were generated by plating 8000/well rat glia on UV-sterilized agarose-coated etched glass coverslips stamped with a 0 . 1 mg/ml poly-d-lysine ( Sigma ) and 0 . 2 mg/ml rat tail collagen ( BD Biosciences , The Netherlands ) solution . At 10 DIV , neuronal cultures were infected with a combination of lentiviruses encoding NPY-pHluorin , NPY-mCherry , synapsin-mCherry , CAPS-ires-EGFP , CAPS-YFP or synapsin-ECFP . Alternatively , neurons were transfected using calcium phosphate and expression plasmids for SemapHluorin and synapsin-mCherry . Neurons were imaged at DIV 14–DIV 18 . Coverslips were placed in an imaging chamber perfused with Tyrode's solution ( 2 mM CaCl2 , 2 . 5 mM KCl , 119 mM NaCl , 2 mM MgCl2 , 20 mM glucose and 25 mM HEPES , pH 7 . 4 ) . All live imaging experiments were performed on a custom-built tandem illumination microscope ( TIM; Olympus , The Netherlands ) consisting of an inverted imaging microscope ( IX81; Olympus ) and an upright laser-scanning microscope . The inverted microscope part was used for imaging fluorescence using an MT20 light source ( Olympus ) , appropriate filter sets ( Semrock , Rochester , NY ) , and a 40× oil objective ( NA 1 . 3 ) , or 60× ( NA 1 . 49 ) for experiments in Figure 7 , on an EM charge-coupled device camera ( C9100-02; Hamamatsu Photonics , Japan ) . Xcellence RT imaging software ( Olympus ) was used to control the microscope and record the images . In pHluorin experiments intracellular pH was neutralized with Tyrode's solution containing 50 mM ammonium chloride ( NH4Cl ) , which replaced sodium chloride ( NaCl ) on an equimolar basis . Ammonium ion ( NH4+ ) solution was delivered by gravity flow through a capillary placed onto the cells . To stimulate the cells electrically , parallel platinum electrodes placed close to the cell soma delivered 30 mA , 1 ms pulses controlled by a Master 8 system ( AMPI , Germany ) and a stimulus generator ( A385RC , World Precision Instruments , Germany ) . The stimulus used was 16 trains of 50 action potentials at 50 Hz with 0 . 5 s interval . All imaging experiments were performed at room temperature ( RT; 21–24°C ) . For DCV fusion assays imaging frequency used was 2 Hz . For SemapHluorin experiments in continental cultures fields of view were selected for presence of SemapHluorin-positive somata , which were placed in the center of the field of view . For protein dispersion experiments , CAPS-EYFP and Synapsin-ECFP were imaged at 0 . 5 Hz simultaneously for 3 min and stimulated with field electrodes ( 16×50 AP at 50 Hz ) . In Figure 8 , NPY-mCherry and CAPS-EYFP were imaged simultaneously at 2 Hz . Stacks from time-lapse recordings acquired with 0 . 5 s intervals were used to analyze DCV release . A 2×2 pixel region ( 0 . 4×0 . 4 μm ) was analyzed according to the experiment as follows . Sema and NPY-pHluorin: fluorescent traces were expressed as fluorescence change ( ∆F ) compared to initial fluorescence ( F0 ) , obtained by averaging the first four frames of the time-lapse recording . A fusion event was counted when fluorescence showed a sudden increase two standard deviations above F0 . Onset of fusion was defined as the first frame with an increase of fluorescence of two standard deviations above F0 . A cargo-pHl release event or punctum was scored as synaptic when the fluorescence center of such a release event/punctum was within 200 nm ( ±1 pixel , the approximate minimal point spread function of our system ) of the Synapsin-mCherry fluorescence centroid . Extra-synaptic events were all events that did not meet this criterion . We only measured release events from neurites and excluded somatic release events . Somatic release events cannot be reliably measured using wide-field fluorescence microcopy due to the bright fluorescence from vesicles in/near the Golgi apparatus in which the intraluminal pH is not yet acidic . The total number of vesicles was automatically analyzed from the NH4+ application time lapse using SynD software ( Schmitz et al . , 2011 ) . When using NPY-mCherry: only the fusion events were scored in which NPY-mCherry fluorescence completely disappeared from a 2×2 pixel punctum after bleaching correction ( ImageJ Bleaching correction plug-in ) . DCVs were categorized as stationary or moving based on the slope of the Kymopgraph ( ImageJ , MultipleKymograph ) , if the slope of the line over the kymograph was different from 0 at any point of the movie , the DCV was considered moving . CAPS-1-ires-EGFP was used to rescue CAPS DKO neurons in combination with NPY-mCherry for DCV fusion assays in Figure 4 . Protein dispersion was analyzed by placing regions of interest ( ROIs ) at the synapses and analyzing the ΔF over F0 ( average of the first four frames ) of Synapsin-ECFP and CAPS-1-EYFP over time . ROIs not overlapping with Synapsin-ECFP were chosen for analyzing CAPS-1-EYFP dispersion at extra-synaptic sites ( ΔF as above ) . These analyses were performed after bleaching correction ( ΔF of the soma over time was used as bleaching and subtracted to the measurements ) . Membrane associated myristoylated EYFP was used as negative control . For co-localization analysis we used ImagJ software ( National Institute of Health , USA , Plug-in JACoP ) . Pearson's coefficients were calculated to obtain cell wide correlation of fluorescent intensities and Mander's coefficients to obtain co-occurrence in VGLUT positive synapses , CAPS-1 puncta or NPY-puncta ( Figure 1 ) . Electrophysiological recordings were performed on single isolated glutamatergic hippocampal neurons between 14 and 18 DIV at RT ( 21–24°C ) . The patch-pipette was filled with a solution containing 135 mM potassium gluconate , 10 mM HEPES , 1 mM ethylene glycol tetra acetic acid ( EGTA ) , 4 . 6 mM magnesium chloride ( MgCl2 ) , 4 mM sodium-Adenosine 5′-triphosphate ( Na-ATP ) , 15 mM creatine phosphate , 50 U/ml phosphocreatine kinase , and 300 milliosmole ( mOsm ) , pH 7 . 3 . The standard extracellular medium consisted of 140 mM NaCl , 2 . 4 mM potassium chloride ( KCl ) , 10 mM HEPES , 10 mM glucose , 4 mM calcium chloride ( CaCl2 ) , 4 mM MgCl2 , and 300 mOsm , pH 7 . 3 . Recordings were performed with an Axopatch 200A amplifier ( Molecular Devices , Sunnyvale , CA ) . Digidata 1322A and Clampex 9 . 0 ( Molecular Devices ) were used for signal acquisition . After whole-cell mode , only cells with access resistance of <12 MΩ and leak current of <500 pA were accepted for analysis . Pipette resistance ranged from 4 to 6 MΩ . EPSCs were evoked by depolarizing the cell from −70 to +30 mV for 0 . 5 ms . Cells were fixed in 4% formaldehyde ( Electron Microscopies Sciences , Germany ) in phosphate-buffered saline ( PBS ) , pH 7 . 4 , for 20 min at RT and washed in PBS . First cells were permeabilized for 5 min in PBS containing 0 . 5% Triton X-100 ( Sigma–Aldrich ) then incubated for 30 min with PBS ( Gibco , The Netherlands ) containing 2% normal goat serum and 0 . 1% Triton X-100 . Incubations with primary and secondary antibodies were done for 1–2 hr at RT . Primary antibodies used were: polyclonal MAP2 ( Abcam , United Kingdom , 1:500 ) , monoclonal VAMP2 ( SySy , Germany , 1:2000 ) and polyclonal Munc13 ( SySy , 1:1000 ) , polyclonal chromogranin B ( SySy , 1:500 ) , VGLUT1 ( SySy , 1:5000 ) , CAPS-1 ( SySy , 1:200 ) , and polyclonal secretogranin II ( kind gift from P Rosa , Institute of Neuroscience , Milan , Italy ) . Alexa Fluor conjugated secondary antibodies were from Invitrogen . Coverslips were mounted in Mowiol and examined on a Zeiss LSM 510 confocal laser-scanning microscope with a 40× objective ( NA 1 . 3 ) or 60× ( NA 1 . 4 ) . Neurons were fixed at DIV 14 for 1–2 hr at RT with 0 . 1 M cacodylate buffer , 0 . 25 mM CaCl2 , 0 . 5 mM MgCl2 ( pH 7 . 4 ) and processed as described ( Wierda et al . , 2007 ) . Cells were washed three times for 5 min with 0 . 1 M cacodylate buffer ( pH 7 . 4 ) , post-fixed for 2 hr at RT with 1% osmium tetroxide/1% potassium ferro-cyanide , washed and stained with 1% uranyl acetate for 40 min in the dark . Cells were dehydrated with a series of increasing ethanol concentration steps and embedded in Epon and polymerized for 24 hr at 60°C . Cells of interest were selected by observing the flat Epon embedded cell monolayer under the light microscope , and mounted on pre-polymerized Epon blocks for thin sectioning . Ultrathin sections ( ∼90 nm ) were cut parallel to the cell monolayer and collected on single-slot , formvar-coated copper grids , and stained in uranyl acetate and lead citrate . Synapses were selected at low magnification using a JEOL 1010 electron microscope . All analyses were performed on single ultrathin sections of randomly selected synapses . The distribution of DCVs was measured with ImageJ on digital images of synapses taken at 100 , 000× magnification using analysis software ( Soft Imaging System , Gmbh , Germany ) . The observer was blinded for the genotype . For all morphological analyses we selected only synapses with intact synaptic plasma membranes with a recognizable pre and postsynaptic density . Docked DCVs had a distance of 0 nm from the vesicle membrane to the plasma membrane . Student's t tests for unpaired data were used , throughout the paper , unless otherwise specified . If deviations differed significantly , t tests were Welch corrected . The Mann–Whitney test was used to compare two groups when one or both groups did not pass the normality test . To test more than two groups , Kruskal–Wallis , Bonferroni corrected , analysis of variance was used . Kolmogorov–Smirnov test was used to test whether distributions were normally distributed . Data are plotted as mean with standard error of the mean; n represents number of neurons , N the number of independent experiments .	Our ability to think and act is due to the remarkable capacity of the brain to process complex information . This involves nerve cells ( or neurons ) communicating with each other in a rapid and precise manner by releasing synaptic vesicles containing neurotransmitters across the gaps—called synapses—between neurons . In addition to this fast neurotransmitter signalling , neurons can transmit signals by releasing chemical signals called neuropeptides . Neuropeptides are major regulators of human brain function , including mood , anxiety , and social interactions . Neuropeptides and other neuromodulators such as serotonin and dopamine are normally packaged into bubble-like compartments called dense-core vesicles . Compared to synaptic vesicles we know much less about how dense-core vesicles are trafficked and released . Dense-core vesicles are generally mobile and move around the inside of cells to release neuropeptides where and when they are needed . However , some vesicles are stationary and may even be loosely tethered to the cell membrane . Most of the sites where dense-core vesicles can fuse with the cell membrane are at synapses . Previous work has suggested that the protein CAPS-1 is important for moving dense-core vesicles to the correct sites on the cell membrane , and for releasing neuropeptides across the synapses of worms and flies . However , detailed insights into this process in mammalian neurons are lacking . By examining neurons from both normal mice and mice lacking the CAPS-1 protein , Farina et al . have now analyzed the role CAPS-1 plays in releasing neuropeptides . In cells lacking CAPS-1 fewer dense-core vesicles merged with the cell membrane than in cells containing the protein . However , a new technique that tracks the movement of individual vesicles revealed that only stationary dense-core vesicles had difficulties fusing; mobile vesicles continued to fuse with the cell membrane in the normal manner . Introducing CAPS-1 into cells lacking this protein corrected the fusion defect experienced by the stationary vesicles . Farina et al . also showed that CAPS-1 was present at most—but not all—synapses , and synapses that had more CAPS-1 released more neuropeptides . This work shows that CAPS proteins strongly influence the probability of dense-core vesicle release and that neurons can tune this probability at individual synapses by controlling the expression of CAPS . Future work will be aimed at understanding how neurons can achieve this and which protein domains in CAPS are required .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	CAPS-1 promotes fusion competence of stationary dense-core vesicles in presynaptic terminals of mammalian neurons
Innate immune cellular effectors are actively consumed during systemic inflammation , but the systemic traffic and the mechanisms that support their replenishment remain unknown . Here , we demonstrate that acute systemic inflammation induces the emergent activation of a previously unrecognized system of rapid migration of granulocyte-macrophage progenitors and committed macrophage-dendritic progenitors , but not other progenitors or stem cells , from bone marrow ( BM ) to regional lymphatic capillaries . The progenitor traffic to the systemic lymphatic circulation is mediated by Ccl19/Ccr7 and is NF-κB independent , Traf6/IκB-kinase/SNAP23 activation dependent , and is responsible for the secretion of pre-stored Ccl19 by a subpopulation of CD205+/CD172a+ conventional dendritic cells type 2 and upregulation of BM myeloid progenitor Ccr7 signaling . Mature myeloid Traf6 signaling is anti-inflammatory and necessary for lymph node myeloid cell development . This report unveils the existence and the mechanistic basis of a very early direct traffic of myeloid progenitors from BM to lymphatics during inflammation . Bacterial infections represent one of the major threats for the human immune system and can lead to sepsis and death ( Martin et al . , 2003 ) . A functional immune response is a key factor to control the outcome of bacterial infections . Therefore , the human immune system has evolved several effector mechanisms to fight bacterial infections that involve the innate and the adaptive arms of the immune system . Antigen presentation is an essential mechanism of activation that requires crosstalk between the innate and adaptive immune system to fight bacterial infections . Dendritic cells are short-lived professional antigen-presenting cells ( APCs ) , and their life span is further reduced during the inflammatory response to pathogens ( Kamath et al . , 2002 ) . Upon inflammation , primed APC thus need to be replaced . During inflammation , systemic signals alert and activate bone marrow ( BM ) hematopoietic stem cells and progenitors ( HSC/P ) ( Chavakis et al . , 2019; Mitroulis et al . , 2018; Nagai et al . , 2006; Ueda et al . , 2005 ) . Inflamed secondary lymphoid organs such as lymph nodes ( LN ) recruit antigen-presenting dendritic cells ( DCs ) ( Legler et al . , 1998; Luther et al . , 2000; Saeki et al . , 1999 ) , while pathogen-associated molecular pattern signals ( PAMPs ) trigger migration of tissue-resident DC to the LN ( Kaisho and Akira , 2001; Sallusto and Lanzavecchia , 2000 ) . Circulation of HSC/P that enter the lymphatic vessels from the peripheral blood ( PB ) with ability to amplify APCs has been described ( Massberg et al . , 2007 ) . However , the circuits used by these HSC/P populations , their characterization , and the cellular and molecular mechanisms that regulate this traffic in inflammatory conditions have not been addressed in detail . Lymphatics form part of an open circulatory system that drains cells and interstitial fluid from tissues . Recently , bone lymphatic endothelial cells have been shown to arise rapidly from pre-existing regional lymphatics in inducible bone-expressing Vegfc transgenic mice through Vegfr3 , osteoclast activation , and bone loss ( Hominick et al . , 2018; Monroy et al . , 2020 ) . Acute endotoxemia is associated with osteoclast activation and bone loss ( Hardy and Cooper , 2009; Nason et al . , 2009 ) . We postulated the pre-existence of an anatomical and functional patent circuit that communicates BM and lymphatic tissues that can be induced upon severe inflammatory conditions like endotoxemia . Our work identifies an emergent traffic of DC-biased myeloid progenitors through direct transit from BM to bone lymphatic capillaries . This traffic is highly activated in endotoxic inflammation . In human reactive lymphadenitis or just after a single immune endotoxic challenge , such as following lipopolysaccharide ( LPS ) stimulation in mice , a massive mobilization of myeloid progenitors from the BM to lymph and retention in the LN takes place . The mobilization is rapid prior to their appearance in PB . LPS simultaneously induces cell-autonomous Ccr7 expression on granulocyte-macrophage progenitors ( GMPs ) and macrophage-dendritic progenitors ( MDPs ) , and a non-cell-autonomous myeloid cell-dependent secretion of Ccl19 in the LN . In vivo blockade of LPS signaling in mature myeloid cells , deletion of hematopoietic Ccl19 , or neutralization of Ccr7 completely abrogated the GMP/MDP migration from the BM to the LN . Moreover , genetic and pharmacological approaches revealed that Traf6-Irak1/4-Ubc13-IκB kinase ( IKK ) signaling mediates NF-κB-independent-SNAP23 phosphorylation and secretion of pre-formed Ccl19 from a specific population of conventional dendritic cells ( cDCs ) , and mature myeloid cell Traf6-dependent signaling is of anti-inflammatory nature . These findings indicate that inflammation results in mobilization of cDC-forming cells directly from the BM to the lymph and LN . As such , emergent myeloid lineage mobilization from the BM to lymph may be important in inflammation by acutely differentiating into antigen-presenting precursors in lymph tissues and associate with an anti-inflammatory response in endotoxemia . To determine whether there is a circulation of HSC/P to human LN , we prospectively analyzed the presence of side population ( SP ) cells in LN biopsies ( Figure 1—figure supplement 1A ) obtained from lymphadenitis and lymphoma patients at diagnosis . Human and murine SP cells , with ability to extrude the dye Hoechst 33342 through upregulated activity of multidrug resistance protein complexes ( Zhou et al . , 2001 ) in BM and other tissues ( Brusnahan et al . , 2010; Challen and Little , 2006; Goodell et al . , 1996 ) , are enriched in long-term reconstituting HSC and other more committed populations of progenitors ( Matsuzaki et al . , 2004; Weksberg et al . , 2008 ) . We found an SP population at a frequency higher than 0 . 01% in 36 out of 64 LN biopsies ( 53 . 12% ) . However , the content of SP cells in the LN did correlate with the LN histological diagnosis . The elevated frequency of SP cells in LN did correlate with the LN histological diagnosis ( Figure 1A ) but not to the anatomical location of the lymphadenopathy ( Figure 1—figure supplement 1B and Supplementary file 1 ) . The accumulation of SP cells was significantly higher in LN from lymphadenitis patients than in lymphoma patients . Further dissection based on histological classifications by independent pathology analysis resulted in the lymphadenitis specimens being sorted into distinct histological categories that corresponded to follicular lymphadenitis with paracortical predominance ( FL ) , granulomatous lymphadenitis ( GL ) , and lymphadenopathies with histological or molecular evidence of viral etiology ( viral lymphadenitis [VL] ) . Interestingly , FL and GL LN contained a median of 0 . 2% SP cells with a range from <0 . 01% to ~40% , which was significantly higher than the content of SP cells in VL , Hodgkin’s lymphoma , and non-Hodgkin’s lymphoma LN ( Figure 1A , Figure 1—figure supplement 1A and Supplementary file 1 ) . The existence of myeloid-committed hematopoietic progenitors was confirmed in myeloid colony-forming cell unit ( CFU ) assays ( Figure 1—figure supplement 1C ) performed on samples from patients with FL . These data show that non-viral inflammatory lymphadenitis results in a significantly increased frequency of primitive hematopoietic cells in LN , while it does not reveal the type of progenitor cells . To confirm whether LN SP cells indeed contained HSC/P , we first sorted LN SP cells from patients with reactive lymphadenitis and plated them in methylcellulose cultures containing rhIL-3 , rhIL-6 , and rhSCF cytokines . CFU analysis demonstrated that SP cells were indeed capable of producing myeloid colonies ( Figure 1—figure supplement 1D ) , while non-SP cells were devoid of measurable CFU-forming ability ( data not shown ) . Immunophenotypic analysis of SP-derived progenitors was also consistent with enrichment of a heterocellular population of CD34 and CD133 expressing granulocyte- , granulocyte-macrophage- , and cDC-biased progenitors ( Bornhäuser et al . , 2005; Görgens et al . , 2013; Figure 1—figure supplement 1D ) . The vast majority of CD45+/CD34+ cells co-expressed CD133+ , and the CD45+/CD34- population was split ~50:50 into CD133+ and CD133- cells ( Figure 1—figure supplement 1D ) . In combination , these data show the accumulation of a myeloid-committed HSC/P population in human lymphadenitis . Adult inflammatory LN tissues therefore contain an increased number of myeloid-committed HSC/P . This increase can result from either the recruitment of these cells to LN via the bloodstream or the expansion of otherwise rare and already resident myeloid-committed HSC/P in these LN . The release of HSC/P from BM into the bloodstream follows circadian cycles ( Méndez-Ferrer et al . , 2008 ) controlled by the activity and fate of inflammatory cells ( Casanova-Acebes et al . , 2013; Chang et al . , 2014 ) . We postulated that if inflammation is responsible for the recruitment of HSC/P to the LN and possibly other organs , we should be able to recapitulate the process of BM egression , migration , and organ retention in an inflammatory murine model wherein the HSC/P migration process is highly conserved . Since the largest content of HSC/P in human LN was found in biopsies from patients with lymphadenitis , we generated a mouse model of Gram-negative sepsis by injection of Escherichia coli LPS into C57Bl/6 mice at the early time point of the circadian HSC/P mobilization cycle ( zeitgeber time [ZT] ) ( Bellet et al . , 2013; Figure 1B ) and analyzed the rapid migration of HSC/P to different organs . This model has the advantage of requiring one single dose of LPS and provides a relevant approach to the analysis of rapid migration mechanisms deprived of confounding effects of proliferation and/or survival induced by LPS at later time points ( Nagai et al . , 2006; Zhao et al . , 2014 ) . E . coli LPS is able to activate a large number of Toll-like receptors ( TLRs ) , which result in high-level activation of the inflammatory signaling cascade ( Beutler and Rietschel , 2003 ) . LPS is also a well-known inducer of HSC/P mobilization to PB ( Cline and Golde , 1977; Velders et al . , 2004; Vos et al . , 1972; Vos and Wilschut , 1979 ) . In our experiments , the circadian mobilization pattern of HSC/P in the PB was severely modified by the administration of LPS , with the increase in HSC/P appearing later and peaking at ZT10 , 6 hr post-administration ( Figure 1C ) , coincident with an increased neutrophil count in the PB ( Figure 1D ) . We observed similar kinetics of an increased number of HSC/P in the highly vascularized kidney and liver tissues after LPS administration ( Figure 1E , F ) , suggesting that the presence of HSC/P in these tissues closely paralleled their presence in the PB . Interestingly , LPS did not elicit a significant change in the level of splenic HSC/P within the first 12 hr after inflammation ( Esplin et al . , 2011; Wright et al . , 2001; Figure 1G ) . Notably , when the HSC/P content was reduced in the BM , the kinetics of their subsequent mobilization to the PB was discordant . The BM HSC/P content decreased , which supports the migratory nature of the increased HSC/P in the PB ( Figure 1H ) ; yet the nadir of the BM HSC/P content occurred as early as 3 hr after LPS administration ( at ZT7 ) , returning to normal values by 6 hr ( ZT10 , Figure 1H ) . The time lapse between the loss of retention of HSC/P in the BM and their presence in the PB circulation suggested that the migration of HSC/P from the BM to the PB required an intermediate step of circulation through other tissues . Based on an earlier description of a lymphatic circulation of HSC/P ( Massberg et al . , 2007 ) , we hypothesized that this delay in the appearance of HSC/P in the PB was due to an intermediate transit of HSC/P through the lymphatic circulation . Indeed , the lymphatic circulation in LPS-treated animals did show a significant increase in the levels of circulating HSC/P in the LN and the thoracic duct compared to controls that closely mirrored the decline of HSC/P in the BM ( Figure 1I , J ) . We next characterized the type of primitive cell populations migrating into the LN via the lymphatic circulation . We first analyzed whether the content of immunophenotypically identifiable BM HSC populations changed concomitantly with the progenitor population changes previously described . LPS induced expansion of BM Lin-/c-kit+/Sca-1+ ( LSK ) and immunophenotypically identified long-term ( LT ) -HSC , short-term ( ST ) -HSC , and multipotential progenitors ( MPPs ) populations at later time points ( Z10–ZT16 ) ( Figure 1—figure supplement 1E-K ) with no changes in the BM HSC content by ZT7 , suggesting a differential effect of LPS signaling on the HSC population . Interestingly , the reduction in the BM content of progenitors was not homogenous throughout the hematopoietic progenitor populations . Confirming the egress of BM CFU-GM described above , the GMP population was significantly decreased by ZT7 ( Figure 1—figure supplement 1F , K ) , while the content of immunophenotypically defined common myeloid progenitors ( CMPs ) only declined by ZT10 ( Figure 1—figure supplement 1F , L ) , and the megakaryocyte-erythroid progenitor ( MEP ) content was increased ( Figure 1—figure supplement 1F , M ) , resulting in no significant net changes in the content of BM Lin-/c-kit+/Sca1- ( LK ) cells ( Figure 1—figure supplement 1F , N ) . Functional in vivo assays of LN cell suspensions obtained at ZT7 demonstrated that the accumulation of progenitors in LN did not contain any significant numbers of long-term or medium-term repopulating HSC . The analysis of competitive repopulating units ( CRUs ) in the LN ( Figure 1—figure supplement 2A ) demonstrated that inflamed LN did not contain increased levels of repopulating cells by ZT7 ( Figure 1—figure supplement 2B ) . LN contained a transient , ST-myelopoietic progenitor population without medium- or long-term multilineage repopulation ability ( Figure 1—figure supplement 2C ) . Lineage analysis of donor-derived circulating cells demonstrated no significant change in T-cell transfer ( Figure 1—figure supplement 2D ) and a diminished transfer of B-cells into the lethally irradiated recipients ( Figure 1 , Figure 1—figure supplement 2E ) , indicating the presence of adoptively transferred lymphoid cells and the absence of mobilization of competent lymphoid progenitors to the inflamed LN . Furthermore , LN SP cells from LPS-treated mice are enriched in LK cells and depleted in LSK cells ( Figure 1—figure supplement 2F ) . LN SP cells contain exclusively ST-repopulating progenitors with the ability to differentiate into myeloid cells ( data not shown ) and are depleted from any significant 8–16 weeks engrafting HSC , as assessed using CRU assays ( Figure 1 , Figure 1—figure supplement 2F , G ) , unlike their BM SP counterparts that are enriched in LSK cells and LT-repopulating activity ( Figure 1 , Figure 1—figure supplement 2F , H ) . These results confirmed that , similar to human inflammatory LN , the LN myeloid progenitors from mice treated with E . coli LPS accumulate in inflamed LN and are depleted of stem cell activity . Altogether , these data indicate that LPS induces a selective lymphatic circulation of myeloid committed progenitors , but not other types of HSC/P populations . To explore the nature of the circuit of the myeloid progenitor migration to LN , we first analyzed the ability of HSC/P to seed LN in non-myeloablated mice . For this experiment , we labeled C57Bl/6-BM-derived lineage negative ( Lin- ) cells containing the HSC/P fraction with the lipophilic dye 1 , 1'-dioctadecyl-3 , 3 , 3' , 3'-tetramethylindocarbocyanine perchlorate ( DiI ) and adoptively transferred them into unmanipulated lymphatic vessel reporter Lyve1eGFP mice . Lyve1eGFP knock-in mice display enhanced green fluorescent protein ( eGFP ) fluorescence driven by the promoter/enhancer of the lymphatic vessel hyaluronan receptor 1 Lyve-1 identifying lymphatic endothelial cells ( Pham et al . , 2010 ) . We analyzed the 17 hr homing of Lin- cells to the BM and LN ( Figure 2A ) . The homing of cells to the BM in mice treated with LPS was reduced by ~65% compared with their vehicle-treated counterparts ( Figure 2B ) . To determine whether Lin-/DiI+ homed cells leave BM in response to LPS , we first determined the existence of transcortical vessels ( Figure 2C; Grüneboom et al . , 2019 ) and the presence of lymphatic vasculature inside the bone by two-photon microscopy in mice either treated with PBS or LPS ( Figure 2D , E and Videos 1 and 2 ) . Lyve1eGFP knock-in mice revealed that the bone of LPS-treated animals contains a Lyve1+network , which was only rarely identified in PBS-treated mice ( Figure 2D ) , suggesting that LPS-induced inflammation may upregulate the expression of the hyaluronan receptor Lyve1 and render patent a pre-existing network of Lyve1+bone cells . Interestingly , Lin-/DiI+ homed cells were located closer to the endosteum in response to LPS at as early as 1 . 5 hr after administration of LPS ( Figure 2E ) . Quantification of the distance of Lin-/DiI+ homed cells to endosteum area showed significant differences between PBS and LPS treatment , indicating increased proximity to the endosteum area after LPS ( Figure 2F ) . We found , albeit at a very low frequency , tiny lymphatics scattered and projected inside the bone ( Figure 2G and Video 3 ) . On the other hand , the seeding of BM-derived Lin-/DiI+ cells into LN increased approximately threefold , which mirrored the decline in BM homing ( Figure 2H–J ) . Histological analysis of BM-derived Lin-/DiI+ cells within the LN by confocal microscopy showed that the migrated HSP/C are spatially positioned in the cortex area surrounding primary follicles ( Figure 2I ) , consistent with localization in T-cell zone for antigen presentation . These findings strongly suggest that the rapid egress of hematopoietic progenitors from BM during inflammation may indeed occur through bone lymphatics draining into LN . To determine the potential of the myeloid progenitors mobilized to the LN , we further determined their in vitro and in vivo differentiation profile . To this end , we analyzed the differentiation capabilities of LN myeloid progenitors in specific cytokine-driven clonal assays in methylcellulose assays . The majority of the differentially accumulated myeloid progenitors in LN by 3 hr post-administration of LPS were granulocyte-macrophage progenitors ( CFU-GM ) and in a much lesser degree unipotent granulocyte progenitors ( GPs , CFU-G ) with no differential accumulation of unipotent macrophage progenitors ( MPs , CFU-M ) ( Figure 3A ) . Next , we investigated whether GMP were able to home and migrate to LN after in vivo administration of LPS . For this purpose , we adoptively transferred sorted β-actin/eGFP transgenic GMPs into congenic mice . Transgenic GMPs were allowed to home to the BM and after 17 hr recipient mice were treated with a single dose of LPS or vehicle control . On day 7 after PBS or LPS administration , murine BM and LN were analyzed for donor-derived granulocytes ( Gr1++/CD11b+/CD11cneg ) , macrophages ( Gr1dim/CD11b+/CD11cneg ) , and lymphoid tissue cDC ( Gr1neg/CD11b+/CD11c+ ) by flow cytometry ( Figure 3B ) . We found that LPS induced differential donor-derived specific GMP differentiation toward the formation and retention of cDC in LN ( Figure 3C ) , but not in the BM ( Figure 3D ) . The content of macrophages and granulocytes did not significantly change with LPS in either LN or BM ( Figure 3C , D ) , confirming the specific nature of the lymphoid-tissue cDC differentiation of mobilized GMP in LPS-treated mice . To elucidate whether the LN cDC content was dependent on migration of committed cDC precursors opposed to local specification of migrated macrophages or MPs , we analyzed the migration of macrophage-dendritic progenitors ( MDPs ) . BM MDPs represent a progenitor population that can differentiate into monocytes/macrophages or directly into cDCs without intermediate macrophage specification ( Fogg et al . , 2006; Geissmann et al . , 2003; Waskow et al . , 2008 ) . MDPs are characterized by high expression of the chemokine receptor Cx3cr1 ( in CX3CR-1GFP reporter mice ) , c-fms ( CD115 ) and Flt3 , and intermediate expression of c-Kit . Serial gating of lineage-/Cx3cr1-GFP++/c-Kitint cells ( Figure 3E , P2 ) showed an approximately threefold accumulation of CD115++/Flt3++ cells in LN and a concomitant 65% depletion in the BM as early as 3 hr after LPS administration ( Figure 3F , G , Figure 3—figure supplement 1A , B ) . In the absence of significant changes in the LN content of macrophages , these data demonstrate that LPS-induced systemic inflammation results in robust and specific recruitment of phenotypic GMP that are BM cDC-committed progenitors to the LN . Immune cells and HSC/P express TLR ( Beutler and Rietschel , 2003; Nagai et al . , 2006; Takeuchi and Akira , 2010 ) , which act as microorganism sensors . LPS stimulation of TLR recruits MyD88 and TRIF through the canonical and endosomal pathways , respectively . Both adaptors subsequently recruit TRAF6 , which acts as the molecular hub of both signaling branches ( Akira et al . , 2001; Kawai and Akira , 2006 ) . To determine whether Traf6 deficiency might affect the migration of HSC/P in response to LPS , we exploited an animal model in which Traf6 is deleted only in hematopoietic cells ( Kobayashi et al . , 2003; Figure 4A , Figure 4—figure supplement 1A ) . LN from Mx1Cre;WT chimeric mice after LPS administration ( at ZT7 , corresponding with the peak of progenitor content in LN , Figure 1H ) revealed a threefold increase in the frequency of CFU-GM . This increase was completely abrogated by the deficiency of Traf6 in hematopoietic cells ( Mx1Cre;Traf6Δ/Δ animals , Figure 4B ) , indicating that the signals that result in CFU-GM mobilization to LN are mediated by hematopoietic Traf6 . Although HSC/P respond directly to PAMPs such as LPS ( Nagai et al . , 2006; Zhao et al . , 2014 ) , direct Gram-negative infection-derived LPS sensing by HSC/P does not play an essential role in emergency granulopoiesis , but rather requires TLR4-dependent signals within the microenvironment ( Boettcher et al . , 2012; Kwak et al . , 2015 ) . To further test whether the hematopoietic Traf6-dependent response to LPS resulting in mobilization of GMP from the BM to the LN by ZT7 is indeed cell autonomous and not determined by the microenvironment , we used the conditional Traf6-deficiency model and analyzed the in vitro migration of myeloid progenitors toward chemoattractant gradients generated by LPS-stimulated BM or LN cells in assays designed to identify the hematopoietic cell population affected by LPS ( Figure 4C , E , G ) . We found that the cell-autonomous deficiency of Traf6 resulted in a relative decrease in migration of ~40% ( Figure 4D ) of BM myeloid progenitors in the presence of LPS , indicating that Traf6 is required for LPS-dependent cell-autonomous BM myeloid progenitor migration . Interestingly , analysis of non-cell-autonomous migration of BM myeloid progenitors demonstrated that LN-derived cells generated more potent chemoattractant signals , resulting in a much larger migration of WT BM myeloid progenitors ( fivefold higher , ~30% ) during the same period ( Figure 4E , F ) , which was drastically diminished ( ~50% reduction ) by Traf6-deficiency in LN cells , but not when using control BM cells as chemoattractant source ( Figure 4E , F ) . These data indicate that although LPS-mobilized myeloid progenitors depend on both cell-autonomous and non-cell-autonomous Traf6-dependent signals , the chemoattractant gradient generated by LPS on LN cells is the predominant effect responsible for Traf6-dependent myeloid progenitor migration . Interestingly , the non-cell-autonomous effect does not seem to be due to secretion of the chemokine inducer IFN-γ , which was not detectable in LN-derived supernatant ( Figure 4—figure supplement 1B ) , strongly suggesting that activated LN NK cells may not be responsible for the migration of BM myeloid progenitors . And , similarly , the effect of Traf6 expression on non-cell-autonomous , LN-mediated migration did not depend on the secretion of IL-1α , IL-2 , IL-13 , IL-4 , TNF-α , or IL-10 regulatory cytokines of inflammatory processes ( Figure 4—figure supplement 1C–H ) . To delineate the resident LN cell population responsible for the migration of GMPs into the LN , we isolated putative effector cell populations representing 1% or more of the cellularity of either BM or LN tissues . We isolated T-cells ( CD3e+ ) , B-cells ( B220+ ) , and myeloid cells ( CD11b+ ) from LN of WT or Traf6Δ/Δ mice and layered input cell equivalents on the bottom of the chamber with LPS , as in the previous experiments ( Figure 4G ) . Although only ~1% of LN cells are myeloid , we observed that LN CD11b+ cells , but not B- or T-cells , from Traf6Δ/Δ mice can recapitulate the same reduction of progenitor migration achieved by complete LN tissue ( Figure 4H ) . To confirm that a Traf6-dependent signaling in LN CD11b + cells is responsible for myeloid progenitor mobilization and eliminate the possible inflammatory effect of previous treatment with polyI:C in Mx1-Cre transgenic mice , we crossed Traf6flox/flox mice with Lyz2Cre transgenic mice ( Clausen et al . , 1999; Cross et al . , 1988 ) and analyzed the migration to LN after LPS administration in mature myeloid lineage-specific Traf6-deficient ( Lyz2Cre;Traf6flox/flox ) mice . Mature myeloid lineage-specific deletion of Traf6 abrogated the migration of myeloid progenitors to LN in response to LPS ( Figure 4I ) . A major consequence of the deficiency of Traf6 in mature myeloid lineage-specific cells was an increase in the endotoxemia-dependent mortality ( Figure 4J ) , indicating that Traf6 expression in mature myeloid cells is required for both migration of myeloid progenitors to LN and protection of LPS-induced mortality . Altogether , these data indicate that LPS/Traf6 signaling is required for migration of myeloid progenitors through predominantly long-range acting , mature myeloid lineage-dependent chemoattractant signals , and that LPS/Traf6 signaling in Lyz2-expressing cells is protective against endotoxin-induced inflammation . Activation of TLRs conserves inflammatory pathways that culminate in the activation of the NF-κB transcription factors ( Karin and Greten , 2005 ) . The LPS binds TLR4/MD2 complexes on the cell surface , and through a series of adaptors and kinases recruits Traf6 . By an E3 ligase-dependent mechanism , Traf6 activates the IκB kinase ( IKK ) complex , which initiates IκBα degradation , and our group has demonstrated that its constitutive deficiency results in reduced HSC quiescence and increased progenitor proliferation ( Fang et al . , 2018 ) . Subsequent nuclear translocation of NF-κB transcription factors results in the expression of cytokine and chemokine genes . To determine whether emergent NF-κB signaling is responsible downstream of LPS/Traf6 for the LPS-induced LN migratory effect of myeloid progenitors , we overexpressed a degradation-resistant mutant of IκBα ( IκBα super-repressor [IκBαSR] ) , in primary murine progenitors , which were then differentiated into macrophages/cDC by macrophage colony-stimulating factor ( M-CSF ) ( O'Keeffe et al . , 2010; Figure 4—figure supplement 1I ) . Analysis of LPS-driven migration in vitro ( Figure 4—figure supplement 1J ) demonstrated that the expression of IκBαSR does not reduce the effect of LPS on the migration of myeloid progenitors toward LPS-stimulated macrophages/cDC ( Figure 4—figure supplement 1K , L ) , indicating that NF-κB transcription factors are dispensable for myeloid progenitor migration . In contrast , inhibition of intracellular protein traffic using monensin dramatically decreased myeloid precursor migration ( Figure 4—figure supplement 1K , L ) , suggesting that intracellular protein trafficking is necessary for the migration phenotype . Collectively , these data indicate that LPS-induced myeloid progenitor migration occurs through an NF-κB-independent , intracellular protein traffic-dependent pathway , and suggests that the progenitor-mobilizing effect of LPS may not require transcriptional activation , depending rather on the intracellular traffic of secreted proteins . The secretome of myeloid cells includes multiple cytokines/chemokines with short- and long-range activities on activation , proliferation , survival , differentiation , and migration of target cells . Specifically , secreted chemokines stimulate migration of target cells following chemokines to the areas of highest concentration . It has been described that hematopoietic progenitor migration is dependent on Cxcl12 gradients ( Greenbaum et al . , 2013; Méndez-Ferrer et al . , 2010 ) . However , by ZT7 , LPS induced upregulation of Cxcl12 expression in BM , but not in LN , indicating that Cxcl12 tissue concentrations per se could not explain the mechanism of migration to LN ( Figure 4—figure supplement 2A ) . An array of tests on secreted chemokines and cytokines demonstrated distinct secretome signatures between BM and LN tissues after LPS administration ( Figure 4—figure supplement 2B–N ) . As expected , LPS induced upregulation after administration of several myeloid cell cytokines and chemokines with ability to recruit and differentiate macrophages and cDC in the extracellular fluid of LN rather than BM as early as 1 hr after LPS challenge ( Figure 4—figure supplement 2C–J ) . However , none of these candidate cytokines/chemokines were found to consistently generate a differential tissue concentration in vivo between LN and BM at both ZT5 and ZT7 ( Figure 4—figure supplement 2B ) . Similar to Cxcl12 , some cytokines/chemokines with potential chemoattractant ability were also found to be upregulated in BM rather than in LN or in both tissues similarly ( Figure 4—figure supplement 2K–N ) . The lack of in vivo tissue differential levels strongly suggested that these BM-derived cytokines or chemokines , although likely to play a role in the LPS-mediated inflammatory response , were unlikely to be responsible for the attraction of BM myeloid progenitors to the LN . The C-C chemokine receptor type 7 ( Ccr7 ) ligand macrophage-inflammatory protein ( MIP ) -3b/Ccl19 has been reported as a chemoattractant for BM and cord blood CD34+ cells in vitro , mainly CFU-GM ( Kim et al . , 1998 ) , and to direct DCs to LNs and elicit an adaptative immune response ( Förster et al . , 1999 ) . Analysis of Ccl19 in the extracellular fluid of the femoral cavity , LN , and plasma demonstrated that in vivo administration of LPS promotes a secretion of Ccl19 in LN when compared with BM and PB ( Figure 5A ) . This differential secretion is specific to Ccl19 since Ccl21 , a highly related chemokine , did not show the formation of similar differential tissue concentrations in LN after LPS administration ( Figure 5—figure supplement 1A ) . Ccl19 is secreted by LN myeloid cells after LPS stimulation and depends on Traf6 expression ( Figure 5—figure supplement 1B ) . Ccr7-mediated signals control the migration of immune cells to secondary lymphoid organs such as LN , facilitating efficient surveillance and targeted cellular response ( Förster et al . , 2008 ) . Also , LPS upregulates membrane Ccr7 expression on cDC and their committed progenitors ( Schmid et al . , 2011 ) . We therefore hypothesized that LN trafficking of phenotypic GMP/MDP is regulated by Ccr7 , and that therefore the Ccl19/Ccr7 axis might explain the coexistence of cell-autonomous and non-cell-autonomous mechanisms required for GMP migration from the BM to LN in response to LPS . To test our hypothesis , we first analyzed whether the specific deficiency of either Ccl19 or Ccr7 modified the level of progenitor migration to LN . To prevent the interference of long-term deficiencies of Ccl19 and Ccr7 expression described in deficient murine models ( Förster et al . , 1999; Mori et al . , 2001 ) , we performed short-term in vivo neutralization of Ccl19 ligand or the Ccr7 receptor by using specific antibodies or isotype controls ( Figure 5B ) and determined the content of CFU-GM in LN after LPS challenge or PBS control . We administered an anti-Ccl19 and an anti-Ccr7 neutralizing antibody ( or their controls ) twice within 15 hr before LPS administration . We found a dramatic reduction ( >90% ) in the number of CFU-GM in the LN of LPS- , anti-Ccl19-treated animals by ZT7 ( Figure 5C ) . Also , we confirmed that Ccr7-expressing GMP in BM rapidly decreased in response to LPS and increased in regional draining LN chains ( Figure 5—figure supplement 1C–E ) . The abrogation of accumulation of progenitors in LN was reproduced by the administration of anti-Ccr7 ( Figure 5D ) . Second , we analyzed the membrane expression of Ccr7 on BM-derived GMP , CMP , and MEP from Mx1Cre;WT or Mx1Cre;Traf6Δ/Δ mice , with or without LPS stimulation . Membrane Ccr7 levels were significantly upregulated as early as 1 hr after LPS administration on GMP in LPS-treated WT mice . Such upregulation was abrogated in LPS-treated Traf6-deficient GMP ( Figure 5—figure supplement 1F-G ) . Finally , we confirmed that hematopoietic chimeric Ccl19-/- animals did not mount a migratory response of myeloid progenitors from BM to LN in response to LPS ( Figure 5E , F ) . Together , these data indicate that the rapid migration of BM myeloid progenitors to LN depends on the expression of the chemokine Ccl19 and its receptor Ccr7 . Traffic of myeloid progenitors to regional LNs was recapitulated in mice receiving intrafemoral adoptive transfer of GMP ( Figure 5G , H ) . In these mice , in vivo L-selectin blockade did not abrogate GMP migration to regional LN while sinusoidal-dependent B-lymphocyte mobilization into regional LN was significantly impaired ( Figure 5H , I ) , indicating that the migration of BM myeloid progenitors , unlike B-cells , into the regional lymphatic circulation is L-selectin independent and therefore unlikely to be mediated by LN high endothelial venules ( Rosen , 2004 ) . Altogether , these data strongly indicated that Ccl19/Ccr7 chemokine signaling is required for the rapid migration of myeloid progenitors to LN upon LPS administration . Given the strong time association of these events , these data support a role for the Ccr7-dependent early traffic of myeloid progenitors in the amelioration or delay of the endotoxic shock induced by LPS . Chemokine secretion requires endosomal fusion with the membrane , which can be detected by exposure of the phosphatidylserine ( PS ) -rich inner leaflet of the endosomes to the external surface of the cell membrane , providing a venue to determine what cell types were responsible for the secretion of Ccl19 . We found an increase in the levels of PS residues on the outer membrane leaflet of WT LN cDC2 ( defined as CD11b+/CD11c+ ) , but not in the CD11b-/CD11c+ population , which comprises cDC1 and plasmacytoid DC ( Figure 6A ) . Interestingly , the exposure of PS residues was abrogated in Traf6∆∕∆ LN cDC ( Figure 6A ) . Given that Traf6 deficiency does not modify the survival of hematopoietic cells or progenitors ( Fang et al . , 2018; Figure 6—figure supplement 1A ) , our observations on annexin-V binding indicate that Traf6 mediates the process of vesicle exocytosis . Having demonstrated that LPS/Traf6 signaling is required for chemokine traffic/secretion in LN mature myeloid lineage cells , and NF-κB transcriptional activation is dispensable for LPS-dependent myeloid progenitor migration , we hypothesized that Traf6 acts through non-NF-κB-dependent IKK activity . To determine whether canonical LPS/TLR downstream effectors were involved in the process of myeloid progenitor migration , we analyzed the chemotaxis of BM myeloid progenitors toward a gradient generated by LN cells in the presence of LPS and specific inhibitors for interleukin receptor-associated kinase 1/4 ( Irak1/4 ) , ubiquitin-conjugating enzyme 13 ( Ubc13 ) , and IKKβ ( Figure 6B ) . Increased myeloid progenitor migration was reversed by all three specific inhibitors ( Figure 6B ) , indicating that the integrity of canonical signaling pathway upstream of NF-κB might be required to attract myeloid progenitors from the BM to the LN . The Traf6/IKK-dependent rapid response to LPS strongly suggests that LPS induces secretion of Ccl19 through a mechanism of rapid release from pre-stored pools . The release of pre-formed cytokines in pre-pooled , stored late endosomes depends on IKK activity through the phosphorylation of mediators of cell membrane fusion . SNAP23 is an essential component of the high-affinity receptor that is part of the general membrane fusion machinery and an important regulator of transport vesicle docking and fusion ( Karim et al . , 2013; Suzuki and Verma , 2008 ) . Phospho-SNAP23 ( Ser95 ) is significantly upregulated by LPS in LN cDC ( Figure 6C ) . Each of the inhibitors for Irak1/4 , Ubc13 , and IKK abrogated the activation of SNAP23 ( Figure 6C , Figure 6—figure supplement 1B ) . Altogether , this set of data indicates that the activation through Traf6/Irak1/4/Ubc13 induced by LPS activates vesicular fusion and vesicular cargo release of pre-formed Ccl19 accumulated in late endosomes of LN myeloid cells . Analysis of steady-state LN myeloid cell populations identified a subpopulation of cDC but not pDC or macrophages , containing most of the cytoplasmic expression of Ccl19 . Further analysis of the subpopulations of LN cDC2 demonstrated that B220-/CD8- cDC that expressed the endocytic receptor DEC-205 ( CD205+ ) and the mannose receptor signal regulatory protein α ( SIRPα , CD172a+ ) distinctly stored high levels of cytosolic Ccl19 ( Figure 6D , Figure 6—figure supplement 1C ) unlike other DC populations , which expressed low levels of intracellular Ccl19 ( Figure 6—figure supplement 1D ) . These data indicate that a subpopulation of cDC2 stores intracellular Ccl19 and is potentially able to self-regulate the migration of its own progenitors in inflammation . This study describes a previously unrecognized , rapid , emergent traffic of myeloid progenitor cells from the BM via lymphatic vessels directly to lymphatic tissues that bypass the peripheral blood stream . Careful analysis of serial femoral sections has not unveiled the existence of a communication between lymphatic and blood vessels in BM further , suggesting the lack of communication between both circuits within the BM cavity and thus likely functional regional independence of each circuit . Our data thus also supports the recently described existence of functional lymphatic vessels in the bone . High-resolution confocal and multiphoton microscopy demonstrated the existence of Lyve1 + cells in which their transgenic reporter illuminated upon exposure to high-dose LPS in vivo along with tiny projections of lymphatics penetrating into the bone . Probably , bone processing and cleaning before fixation and decalcification may have deprived us ( and other investigators ) from a better identification of notable , anatomically identifiable lymphatic vessels within the network of transcortical capillaries ( Grüneboom et al . , 2019 ) . Although the exact anatomical and functional nature of this lymphatic network in the bone remains poorly defined , our data demonstrate that the lymphatic-mediated traffic from BM is highly activated in endotoxic inflammation and drains into regional LN chains . Bone is a dynamic organ in constant remodeling . Upon inflammation , for example , cytokines and microbial LPS are capable of initiating bone absorption by activating osteoclasts ( Hardy and Cooper , 2009; Nason et al . , 2009 ) . Systemic inflammation has been associated with osteoclast activation and osteoblast thinning ( Hardy and Cooper , 2009; Nason et al . , 2009 ) , and bone lymphatic endothelial cells have been shown to arise rapidly from pre-existing regional lymphatics upon osteoclast activation ( Hominick et al . , 2018; Monroy et al . , 2020 ) . Osteoclast activation and osteoblast thinning are likely to facilitate transcortical migration of cells and fluid through existing transcortical vessels . Our data showed that as early as 90 min after LPS administration myeloid progenitors to or are in closer proximity to the lymphatic endothelium in BM while 1 . 5 hr later there is an ~70% reduction of myeloid progenitors within the BM and a marked increase of myeloid progenitors in the LN tissue . This observation , along with the need of a longer period of time to detect an increase in the frequency of myeloid progenitors in peripheral blood , suggests that two temporally distinct waves of progenitors take place , a fast one to the lymphatic circulation followed by a slower one into the blood stream . In our study , by using transgenic animals , we demonstrated that the administration of a single dose of LPS suffices to induce migration of GMP/MDP while no other types of progenitors or stem cells migrate to lymphatics in this first wave of egression before any significant contribution from or to PB and differentiate into short-lived LN cDCs in a murine acute model of inflammatory signaling by LPS , and that therefore they may modulate the course of infectious diseases and other inflammatory conditions . This traffic is also likely to happen in homeostatic conditions as previously shown ( Waskow et al . , 2008 ) while our analysis provides compelling evidence on its striking activation upon inflammation/LPS administration . Our data support the migration of a distinctly immature progenitor population composed of GMP/MDP with ability to generate cDC in LN upon traffic from BM to LN . This traffic of myeloid progenitors from BM to LN is rapid , before LPS induces proliferation and/or apoptosis ( Nagai et al . , 2006; Zhao et al . , 2014 ) , and can be recapitulated in mice receiving intrafemoral adoptive transfer of GMPs . These GMP/MDP tend to localize in the T-cell areas of LN . Cheong et al . reported that migratory monocyte-derived cDC2 can also localize in T-cell areas of the LN and acquire an inflammatory phenotype DC-SIGN/CD209a+ ( Cheong et al . , 2010 ) . No significant mobilization of M-CSF-responding monocyte progenitors ( CFU-M ) can be found as early as 3 hr after LPS administration before any significant effect of LPS on proliferation may take place . Interestingly , the interference of this traffic by blocking Traf6-dependent signaling in Lyz2-expressing myeloid cells results in increased animal death . Ccr7+ GMP/MDP , but not other myeloid or lymphoid progenitors , egress BM . Such egress follows differential tissue levels of Ccl19 resulting from activation of the secretion of the pre-formed chemokine by LN Lyz2-expressing mature myeloid cells , specifically a subpopulation of cDC expressing CD205 and CD172a . This process seems to be independent of Cxcl12 levels since no changes in Cxcl12 levels in LN , PB , or BM were observed , and this effect seems to be exclusively dependent on activation of non-canonical Traft6/IKK activity without need for transcriptional activation . Schmid et al . , 2011 demonstrated that a population of common dendritic progenitors , a non-GMP-derived population of progenitors , can also migrate from the BM to lymphoid and non-lymphoid tissues in response to TLR agonists and generate both cDCs and pDCs ( pDCs ) . The type of migration though depended on combined downregulation of Cxcr4 and upregulation of Ccr7 , which seem to imitate the mechanism of GMP migration . Interestingly , Ccl21 , which is expressed by lymphatic endothelial and stromal cells but not by myeloid cells ( Eberlein et al . , 2010 ) , does not induce any differential gradient of secretion between LN and BM or blood , suggesting that Ccl21 may not be a primary mediator of the myeloid progenitor migration from BM to LN upon LPS challenge , while the hematopoietic deficiency of Ccl19 suffices to completely abrogate the mobilization of myeloid progenitors to LN induced by LPS . Our data support the existence of a steady-state LN population of cDC that co-expresses the maturation antigens CD205 and CD172a and stores high levels of Ccl19 in their cytoplasm . An interesting possibility is , as our data indicate , that upon bacterial antigen challenge , differentiated myeloid cells of LN like cDCs , which respond to LPS by secreting chemokine-containing pre-formed exosomes , accelerate a positive feedback activation loop to recruit cDC progenitors to the lymphatic tissue . cDC in LNs might thus act as sensors for the presence of bacterial products and release Ccl19 within minutes . Individual DCs have a short half-life ( 1 . 5–2 . 9 days ) ( Kamath et al . , 2002 ) , and DC precursors have a short half-life in blood circulation ( Breton et al . , 2015 ) . We posit that the migration of DC progenitors through the lymph tissues provides a direct afferent communication between the LN mature cDC population responsible for the secretion of the chemokine Ccl19 and at the same time allows the emergent migration of functional cDC progenitors from the BM to differentiate into lymphatic APCs . Finally , our data also support the key role of an alternative inflammatory signaling pathway elicited by coordination of Traf6/IKK responsible for SNAP23 phosphorylation and Ccl19 secretion , before resulting in transcriptional regulation by their downstream effector NF-κB . Traf6 has been identified as a signaling molecule that can regulate splicing of downstream targets without affecting NF-κB in hematopoietic stem cells and progenitors ( Fang et al . , 2017 ) . Our data further identifies non-canonical signaling pathways elicited by Traf6 in differentiated myeloid cells to modulate the migration of hematopoietic progenitors . Traf6-dependent , cytosolic-mediated IKK signaling allows a fast Ccl19 secretory response before inflammatory transcriptional and post-transcriptional signatures are expressed . In summary , we describe , upon inflammation , a rapid trafficking of cDC-biased myeloid progenitors from the BM , via lymphatic vessels , directly to lymphatic tissues that bypasses the blood stream . This GMP/MDP migration represents a mechanism for fast replenishment of cDCs in lymphatic tissues . Rapid replenishment of a reserve of cDC-biased progenitors in LN may represent a major homeostatic function of this novel lymphatic circuit and may explain why the circulation of myeloid progenitors is conserved during the postnatal life . CD57Bl/6 ( CD45 . 2+ ) mice were used between 8–10 weeks of age and were purchased from Jackson Laboratory , Bar Harbor , ME; Harlan Laboratories , Frederick , MD . Mx1Cre;Traf6-floxed mice were generated by breeding Mx1-Cre transgenic mice ( Mikkola et al . , 2003 ) with biallelic TRAF6 floxed mice ( kindly provided by Dr . Yongwon Choi , University of Pennsylvania ) ( Kobayashi et al . , 2003 ) . Full chimeric mice were generated by non-competitive transplantation of Mx1Cre;WT or Mx1Cre;Traf6flox/flox whole BM cells into lethally irradiated B6 . SJLPtprca Pepcb/BoyJ ( CD45 . 1+ ) mice obtained from the CCHMC Animal Core . Traf6 was deleted upon induced expression of Cre recombinase after 3–6 intraperitoneal injections ( 10 mg/kg/b . w . Poly ( I:C ) ; Amersham Pharmacia Biotech , Piscataway , NJ ) every other day at 6 weeks after transplantation generating WT and Traf6∆/∆ mice . Lyz2Cre;Traf6flox/flox mice were generated by non-competitive transplantation of Lyz2Cre;WT or Lyz2Cre;Traf6flox/flox whole BM cells into lethally irradiated B6 . SJLPtprca Pepcb/BoyJ ( CD45 . 1+ ) mice obtained from the Division of Experimental Hematology/Cancer Biology of Cincinnati Children’s Hospital Research Foundation ( CCHRF ) . Vav-Cre;Traf6flox/flox mice and their control counterparts were generated as previously described ( Fang et al . , 2018 ) . Lyve1eGFP ( Pham et al . , 2010 ) and β-actin-eGFP ( Okabe et al . , 1997 ) and CX3CR-1GFP ( Jung et al . , 2000 ) transgenic mice were purchased from Jackson Laboratories . C57BL/6 mice for circadian cycle analysis of CFU-C were maintained on a 14 hr light/10 hr darkness lighting schedule . This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health . All of the animals were handled according to approved Institutional Animal Care and Use Committee ( IACUC ) protocol #2019-0041 of Cincinnati Children's Hospital . Lymphadenopathies from patients were obtained through Institutional Review Board-approved protocols of the Hospital Reina Sofia ( Cordoba , Spain ) , donor-informed consent , and legal tutor approval in the case of patients younger than 18 years old . Diagnostic lymphadenopathy biopsies from 64 consecutive patients from 2009 until 2013 were analyzed in this study . The median age of patients was 34 years old ( range: 3–89 ) . Diagnosis and histological classification of the type of lymphadenopathy and tumors were based on previously published criteria ( Campo et al . , 2011; Weiss and O'Malley , 2013 ) . Anatomical location of lymphadenopathies is described in Supplementary file 1 . Specimens were blindly analyzed through adjudication of unique identifiers . Mice received a single intraperitoneal injection of 30 mg/kg of E . coli LPS ( Sigma-Aldrich , St Louis , MO ) or PBS as vehicle control and were executed always at ZT4 or 4 hr after the initiation of light into the animal room . At different time points after PBS or LPS administration , BM cells from femurs , tibias , and pelvis were harvested by crushing in PBS containing 2% of FBS and erythrocytes were lysed using a hypotonic buffer from BD Biosciences . Blood was collected by retro-orbital bleeding or cardiac injection . Liver , kidney , and thoracic duct cells were harvested by enzymatic digestion solution with collagenase II ( 1 mg/mL , Thermo Fisher–Gibco , Waltham , MA ) and dispase ( 5 mg/mL , Gibco , Life Technologies ) in a shaking water bath at 37°C for 1 hr . Spleen cells were isolated by scraping with slides in sterile PBS following red blood cell ( RBC ) lysis ( Pharm Lyse; BD Biosciences , San Jose , CA ) . Extracted LN were derived from the cervical and axillary chains exclusively . Cells from BM , LNs , thoracic duct , blood , spleen , liver , and kidneys were depleted of RBC by 2 min incubation in Pharm Lyse ( BD Biosciences ) , washed , counted , and plated in SDisolid methylcellulose media ( Methocult 3434; StemCell Technology , Vancouver , Canada ) and cultured in an incubator ( 37°C , 5% CO2/>95% humidity ) , and the number of CFU-C was scored on day 7 or 8 of culture using an inverted microscope . To examine the type of myeloid progenitors migrating into LN , we used base methylcellulose medium ( Methocult 3134; StemCell Technology ) supplemented with 30% FCS , 1% protease-free , deionized BSA ( Roche ) , 100 mM b-mercaptoethanol , 100 IU/mL penicillin , 0 . 1 mg/mL streptomycin , and any of the following: for CFU-GM , rm-GM-CSF ( 100 ng/mL , PeproTech , Rocky Hill , NJ ) for specific analysis of CFU-GM content , rm-M-CSF ( 100 ng/mL , PeproTech ) for specific analysis of CFU-M content , or rh-G-CSF ( 100 ng/mL , Neupogen ) for specific analysis of CFU-G content . To analyze the long-term reconstitution capacity of HSC/Ps mobilized into LNs after LPS administration at different time of periods , 4−5 × 106 of erythrocyte-depleted CD45 . 2+ LN suspension cells were prepared in sterile conditions and transplanted together with 2 . 5 × 105 CD45 . 1+ BM competitor cells into lethally irradiated CD45 . 1+ recipient mice . In some experiments , 104 LN SP cells or 103 BM SP cells from CD45 . 2+ mice were competitively transplanted into CD45 . 1+ recipient mice . Competitive repopulating units ( CRU ) analysis was performed by flow cytometry analysis ( BD Biosciences ) at different time points post-transplantation ( Harrison , 1980 ) . For immunophenotype analysis of HSC/P populations by fluorescence-activated cell sorter ( FACS ) , erythrocyte-depleted BM cells were stained first for lineage markers with biotin-labeled mouse lineage panel ( BD Biosciences , Pharmingen ) containing anti-CD3e ( CD3ε chain ) , anti-TER-119/erythroid cells ( Ly-76 ) , anti-Gr1 ( Ly6G and Ly-6C ) , anti-CD45R ( B220 ) , anti-CD11b ( integrin α chain , Mac1α ) , followed by allophycocyanin and cyanine dye Cy7- ( APC-Cy7 ) -conjugated streptavidin , allophycocyanin ( APC ) -conjugated anti-c-Kit ( clone 2B8 ) , R-phycoerythrin and cyanine dye Cy7 ( PECy7 ) -conjugated anti-Sca1 ( clone D7 ) , eFluor 450-conjugated anti-CD34 ( clone RAM34 ) ( Affymetrix eBioscience , San Diego , CA ) , and PerCP and cyanine dye Cy5 . 5 ( PerCP Cy5 . 5 ) -conjugated anti-Fcγ-RII/III ( clone 2 . 4G2 ) ( BD Biosciences ) . FACS sequential discrimination on a lineage negative gated population was used to identified LK myeloid progenitor subpopulations: Lin- , c-Kit+Sca1-CD34+ FcγRII/III+ ( GMP ) ; Lin- , c-Kit+Sca1-CD34+FcγRII/IIIlo ( CMP ) ; Lin- cKit+Sca1- CD34+ FcγRII/III- . LSK ( Lin-Sca1 +c Kit- ) subpopulations were distinguished as Lin- , c-Kit-Sca1+ CD34-Flt3- for LT-HSC Lin- , c-Kit-Sca1+CD34+Flt3- for ST-HSC , and Lin- , c-Kit-Sca1+CD34+Flt3+ for MPPs . For chimera analysis in repopulated animals , 20 µL of red cell-depleted blood was stained with fluorescein isothiocyanate ( FITC ) -conjugated anti-CD45 . 1 ( clone A20 ) , R-phycoerythrin and cyanine dye Cy7 ( PECy7 ) -conjugated anti-CD45 . 2 ( clone 104 ) , allophycocyanin ( APC ) -conjugated anti-CD11b ( clone M1/70 ) , allophycocyanin and cyanine dye Cy7- ( APC-Cy7 ) -conjugated anti-B220 ( clone RA3-6B2 ) , R-phycoerythrin ( PE ) -conjugated anti-CD3e ( clone 145-2 C11 ) and BD Horizon V450-conjugated anti-Gr1 ( clone RB6-8C5 ) , PerCPefluor710 anti-CD115 ( clone AFS98 ) , and R-phycoerythrin-conjugated anti-CD135 or anti-Flt3 ( clone A2F0 . 1 ) . All monoclonal antibodies were purchased from BD Biosciences , Pharmingen . Cell acquisition was performed by flow cytometry ( LSRFortessa I , BD Biosciences ) equipped with FACSDIVA software ( BD Biosciences ) for multiparameter analysis of the data . FACS strategies were CD45 . 1-CD45 . 2+CD3ε+B220-CD11b- for LN-T cells , CD45 . 1- CD45 . 2+CD3ε-B220+/CD11b- for LN-B cells , and CD45 . 1- CD45 . 2+CD3ε-/B220-/CD11b+ for LN myeloid cells in a FACSAria II cell sorter ( BD Biosciences ) . For BM and LN -SP cells analysis and sorting , 2 × 106 cells/mL were stained with Hoescht 3342 ( 5 µg/mL ) as described previously ( Cheong et al . , 2010 ) . For intracellular analysis of the phosphorylated state of SNAP23 protein , surface antigen-labeled cells were fixed with Cytofix buffer ( BD Biosciences ) for 20 min and then permeabilized using Cytofix/Cytoperm buffer ( BD Biosciences ) for 20 min . After washing , cells were stained intracellularly using a rabbit non-conjugated monoclonal anti-phospho-SNAP23 ( Ser95 ) ( Karim et al . , 2013 ) for 40 min in Perm/Wash Buffer 1x ( BD Biosciences ) with 0 . 5% of rabbit serum . Cells were then incubated with a secondary Alexa Fluor 488-conjugated ( Thermo Fisher–Invitrogen ) goat anti-rabbit antibody for 40 min in Perm/Wash Buffer 1x with 0 . 5% of goat serum . All incubations after cell stimulation were done on ice and in darkness . Single-cell analysis was performed by flow cytometry and the histogram-overlay graphed ( LSRFortessa I; FlowJo xV0 . 7 software; BD Biosciences ) . The mean fluorescence intensity ( MFI ) ratio was calculated as the ratio of the fluorescence intensities of LPS-stimulated to PBS-stimulated ( control ) . LN suspension cells from Mx1Cre;WT and Mx1Cre;Traf6∆/∆ mice were obtained to performed LPS stimulation . 106 cells were plated into 24-well plates and treated with PBS or LPS for 1 hr . After 15 min labeling with surface antibodies against CD45 . 2 ( clone 104 ) , CD11c ( clone HL3 ) , CD11b ( clone M1/70 ) , and B220 ( clone RA3-6B2 ) , the samples were washed twice and then stained for annexin-V for 15 min and in darkness . All antibodies were purchased from BD Biosciences , Pharmingen . Single-cell analysis was performed using flow cytometry and the histogram-overlay graphed ( LSRFortessa I; FlowJo xV0 . 7 software; BD Biosciences ) . The MFI ratio between LPS MFI and PBS MFI was calculated . To determine HSPC cell death , BM cells were isolated and incubated with biotin-conjugated lineage markers as described above , followed by staining with streptavidin eFluor450 , Sca-1-PE-Cy7 , c-Kit-APC-eFluor780 . The cells were then fixed , permeabilized , and stained with TUNEL TMR red ( 12156792910 , Roche ) . Analysis was performed using FACSCanto and/or LSRII flow cytometers and with either Diva or FlowJo software . For homing assays , 2 × 106 of Lin- cells , previously depleted by immunomagnetic selection ( Lineage Cell Depletion kit , Miltenyi Biotec , Auburn CA ) , were stained by 1 , 1′-dioctadecyl-3 , 3 , 3′ , 3′ , tetramethylindocarbocyanine perchlorate;CILC18 ( 3 ) ( 5 µM/mL DiI , Thermo Fisher–Invitrogen ) and adoptively transferred intravenously into non-myeloablated Lyve1eGFP mice . 17 hr later , one single LPS dose ( 3 mg/mL ) or vehicle control ( PBS ) was administered intraperitoneally . 3 and 6 hr later ( ZT7 or ZT10 ) , mice were euthanized with pentobarbital ( 60–80 mg/kg ) and the whole body was fixed using a freshly made solution of PBS plus 2% of paraformaldehyde and 0 . 05% of glutaraldehyde infused by perfusion pump through left ventricle of the animal . 15–20 min later , the BM cells and LN organs were harvested and the percentage of labeled Lin- cells that had homed into BM was determined by FACS analysis . The homing calculation was done as previously reported ( Boggs , 1984 ) . Fixed LN organs were permeabilized for 15 min with PBS containing 0 . 2% of Triton X-100 . To detect GFP on the lymphatic endothelium , LN were incubated overnight with a primary antibody anti-GFP+ ( Thermo Fisher–Invitrogen ) . LNs were scanned by confocal microscopy ( Nikon A1R GaAsP ) through multidimensional acquisition to construct 3D representations of the whole organ at ×10 magnification . The merged images of GFP/DiI or DAPI/DiI are presented , and the total cell number of labeled Lin- cells was counted manually . Finally , harvested femurs were decalcified for 14 days with 10% of EDTA ( Sigma-Aldrich ) in PBS and embedded in paraffin . Longitudinal sections of bone were cut to 4 μm thickness and were then de-paraffinized and broke the protein crosslink before stain by antigen retrieval treatment with citrate buffer pH 6 ( Cancelas et al . , 2005 ) . Then bone sections were permeabilized with 0 . 2% of Triton X-100 for 15 min and blocked with 5% of BSA for 1 hr . Slides were stained with primary antibodies anti-GFP ( chicken polyclonal , Abcam Inc , Cambridge , MA ) and rat anti-mouse panendothelial cell antigen ( clone , MECA-32 , BD Biosciences , Pharmingen ) at 4°C overnight . Then we stained with secondary antibodies , goat anti-rat Alexa Fluor-488 and goat anti-chicken Alexa Fluor-568 , all from Invitrogen at 1:1000 v/v concentration for 1 hr at room temperature . Blood and lymphatic vessels were scanned by confocal microscopy ( Nikon A1R GaAsP ) through multidimensional acquisition to construct 3D representation . To further characterize lymphatic system in bone tissue and BM cavity and to image the close proximity of homed Lin-/DiI+ to lymphatic vessels into Lyve1eGFP mice , we utilized multiphoton IVM as previously described ( Gonzalez-Nieto et al . , 2012; Köhler et al . , 2009 ) . After LPS/PBS injections , long bones were harvested and muscle were carefully cleaned . Further , bone tissues were cautiously trimmed with an electric drill ( Dremel ) to get better access of the BM cavity for imaging by leaving a very thin ( ~30–40 μm ) layer of bone tissue . Bones were mounted in 2% low- melting agarose to minimize movements during imaging and covered with PBS . Multiphoton microscopy on the long bones ( femur and tibia ) was subsequently performed using a Nikon A1R Multiphoton Upright Confocal Microscope equipped with Coherent Chameleon II TiSapphire IR laser , tunable from 700 to 1000 nm , and signal was detected by low-noise Hamamatsu photomultiplier ( PMT ) tubes . Bone tissue was identified as second-harmonic ( SHG ) signal ( PMT ) . Bones were images in PBS using a 25× Apo 1 . 1 NA LWD water Immersion objective and NIS image software . For initial standardization , bones were scanned at wavelength of 800 , 850 , and 900 nm detecting GFP ( 530 nm ) and DiI red ( 580 nm ) . For imaging , a 500 × 500 μm area was scanned in ~35 steps of 4 μm down to 120–150 μm depth using an illumination wavelength of 800 nm detecting SHG signal ( 480 nm ) , green ( 530 nm ) , and red ( 580 nm ) fluorescence . Control C57BL/6 mice were used as a negative control for Lyve1eGFP mice to detect specific signal for GFP-lymphatic system in bone tissue and BM cavity . Lymphatic vessels were well detected in the bone tissue using Lyve1eGFP mice with DiI-labeled Lin- cells in the BM cavity . For quantification of proximity of Lin-/DiI+ with lymphatic system , Imaris software was used to measure distance between DiI-labeled cells and GFP-positive lymphatic vessels using 3D images . C57Bl6 mice received single intraperitoneal injections of MEL14 ( CD62L ) antibody ( BioXcell ) 200 μg . Control mice received the same amount of Rat IgG2a . Post 3 days of MEL14 antibody treatment , LDBM cells from ß-actin eGFP mice were injected intrafemorally . LPS ( 30 mg/kg , BW ) was injected post 1 hr of interfemoral injections . Mice were sacrificed at 3 hr post-administration , and ipsilateral and contralateral regional and distant LNs ( inguinal , popliteal , axillary , and cervical ) were isolated for analysis . LN cells were stained for granulocyte-macrophage progenitor ( GMP ) markers and anti-CD19-PECy7 ( Cat# 552854 , BD Biosciences ) and analyzed by flow cytometry and quantified the GFP+ GMP and B lymphocyte populations migrating to LN . Hematopoietic chimeras of WT and Ccl19-/- ( Link et al . , 2007 ) BM cells were generated by transplantation into CD45 . 1 + mice , similarly to Mx1Cre;d hematopoietic chimeras . Mice were followed for 8 weeks and found to have >95% chimera of CD45 . 2+ cells in peripheral blood . After 8 weeks , femoral LDBM cells from donor congenic-actin transgenic , CD45 . 2+ mice were injected ( 5 × 105 per mouse ) intrafemorally to both WT and Ccl19 hematopoietic chimeric mice . PBS or LPS ( 30 mg/kg , b . w . ) were injected at 1 hr post-intrafemoral injections and sacrificed at 3 hr post-administration of LPS . At that time , ipsilateral LN from inguinal and popliteal regions was isolated . Suspension of LN cells was counted and stained with specific antibodies for GMP and MDP characterization , and the frequency of different GFP+ GMP and MDP populations was analyzed by flow cytometry as mentioned above . For non-cell-autonomous effect analysis , 5 × 105 of BM or LN nucleated cells from Mx1Cre;WT or Mx1Cre;Traf6∆/∆ CD45 . 2+ were layered on bottom wells of 24-well transwell plate ( Corning Inc , Lowell , MA ) together with 100 ng/mL of LPS , and 1 × 105 WT CD45 . 1+ LDBM cells were layered on upper chamber at 37°C , 5% CO2 . For cell-autonomous effect analysis , 5 × 105 of BM or LN nucleated cells from WT CD45 . 1+ mice were layered in the lower chamber with 100 ng/mL of LPS , and 1 × 105 Mx1Cre;WT or Mx1Cre;Traf6Δ/Δ CD45 . 2+ LDBM cells were layered in the upper chamber at 37°C , 5% CO2 . After 4 hr , cells were resuspended and those adhered to the bottom layer were collected using an enzyme-free cell dissociation buffer ( Cell Dissociation Buffer , enzyme free , PBS , Thermo Fisher–Gibco ) . Progenitor responses toward migratory gradient were analyzed by flow cytometry analysis of LK cells . The percentage of migration was calculated by dividing the number of LK in the outputs by the number of LK in the inputs and multiplied by 100 . PBS was included as negative control . All assays were performed in triplicate . To analyze the NF-κB-dependent or -independent mechanism of myeloid progenitor migration , BM Lin- were transduced with pMSCVpuro-eGFP bicistronic retroviral vector encoding the full length of IκBα mutant ( super-repressor ) in the presence of the recombinant fragment of fibronectin , CH296 ( Takara Bio Inc , Madison , WI ) for 16 hr at 37°C . 24 hr later , GFP+ cells were sorted and macrophages were generated ( Chang et al . , 2014 ) . To characterize the expanded population , R-phycoerythrin ( PE ) -conjugated anti-CD169 ( clone 3D6 . 112 ) , PerCP-efluor 710-conjugated anti CD115 ( clone AF598 ) ( Affymetrix eBioscience ) , allophycocyanin and cyanine dye Cy7- ( APC-Cy7 ) -conjugated anti-CD11b , efluor 450-conjugated anti-F4/80 ( clone BM8 ) ( Affymetrix eBioscience ) , and Alexa Fluor 647-conjugated anti-CD68 ( clone FA-11 ) ( BD Biosciences ) were used for FACS analysis . 50 × 103 differentiated and transduced macrophages with empty or IκBα super-repressor were layered on bottom wells of 24-well transwell plate in the presence of 100 ng/mL of LPS , and 1 × 105 WT LDBM cells were layered on upper chamber at 37°C , 5% CO2 . 4 hr later , migrated LK cells were determined by flow cytometry as described above . All assays were performed in triplicate . BM , plasma , and LN cells were isolated in PBS containing a protease inhibitor cocktail ( Roche Diagnostics , Chicago , IL ) , and Ccl19/Ccl21 levels were determined by indirect sandwich of enzyme-linked immunosorbent assay ( ELISA ) following manufacturer’s instructions ( R&D Systems , Minneapolis , MN ) . Multi-analytic profiling beads using Milliplex Multiplex mouse cytokine/chemokine panels ( EMD Millipore , Billerica , MA ) according to manufacturer’s instructions were used to analyze chemokines and cytokines profile into BM and LN tissues at different time periods after LPS or PBS administration into WT mice . Monoclonal rat IgG2a antibody specific for Ccr7 ( clone 4B12 ) or polyclonal goat IgG antibody for Ccl19 ( AF880 ) and control rat IgG2a or control goat purified IgG were obtained from R&D Systems . 50 μg of antibodies were injected twice into C56BL/6 mice within 15 hr ( first dose i . v . and the second dose i . p . ) . The Irak1/4 inhibitor I , ubiquitin-conjugating enzyme E2 N ( UBE2N ) inhibitor , Ubc13 inhibitor ( Rhyasen et al . , 2013 ) , and IκB kinase inhibitor ( PS-1145 dihydrochloride ) were purchased from Sigma-Aldrich . LN cells from C57BL6 mice were treated with 10 μM of IRAK-Inh , 0 . 2 μM of Ubc13-Inh , and 10 μM of IKK-Inh for 45 min and compared with the vehicle dimethylsulfoxide ( DMSO ) at 0 . 1% in PBS . Monensin ( eBioscience ) was used at 2 µM . Quantitative data is given as mean ± standard error of the mean ( SEM ) or standard deviation ( SD ) . Statistical comparisons were determined using an unpaired Student's t-test , non-parametric Mann–Whitney test , and one-way or two-way ANOVA with Bonferroni corrections . A value of p<0 . 05 was considered to be statistically significant .	When the body becomes infected with disease-causing pathogens , such as bacteria , the immune system activates various mechanisms which help to fight off the infection . One of the immune system’s first lines of defense is to launch an inflammatory response that helps remove the pathogen and recruit other immune cells . However , this response can become overactivated , leading to severe inflammatory conditions that damage healthy cells and tissues . A second group of cells counteract this over inflammation and are different to the ones involved in the early inflammatory response . Both types of cells – inflammatory and anti-inflammatory – develop from committed progenitors , which , unlike stem cells , are already destined to become a certain type of cell . These committed progenitors reside in the bone marrow and then rapidly travel to secondary lymphoid organs , such as the lymph nodes , where they mature into functioning immune cells . During this journey , committed progenitors pass from the bone marrow to the lymphatic vessels that connect up the different secondary lymphoid organs , and then spread to all tissues in the body . Yet , it is not fully understood what exact route these cells take and what guides them towards these lymphatic tissues during inflammation . To investigate this , Serrano-Lopez , Hegde et al . used a combination of techniques to examine the migration of progenitor cells in mice that had been treated with lethal doses of a bacterial product that triggers inflammation . This revealed that as early as one to three hours after the onset of infection , progenitor cells were already starting to travel from the bone marrow towards lymphatic vessels . Serrano-Lopez , Hegde et al . found that a chemical released by an “alarm” immune cell already residing in secondary lymphoid organs attracted these progenitor cells towards the lymphatic tissue . Further experiments showed that the progenitor cells travelling to secondary lymphoid organs were already activated by bacterial products . They then follow the chemical released by alarm immune cells ready to respond to the immune challenge and suppress inflammation . These committed progenitors were also found in the inflamed lymph nodes of patients . These findings suggest this rapid circulation of progenitors is a mechanism of defense that contributes to the fight against severe inflammation . Altering how these cells migrate from the bone marrow to secondary lymphoid organs could provide a more effective treatment for inflammatory conditions and severe infections . However , these approaches would need to be tested further in the laboratory and in clinical trials .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "immunology", "and", "inflammation" ]	2021	Inflammation rapidly recruits mammalian GMP and MDP from bone marrow into regional lymphatics
The thermodynamics of protein folding in bulk solution have been thoroughly investigated for decades . By contrast , measurements of protein substrate stability inside the GroEL/ES chaperonin cage have not been reported . Such measurements require stable encapsulation , that is no escape of the substrate into bulk solution during experiments , and a way to perturb protein stability without affecting the chaperonin system itself . Here , by establishing such conditions , we show that protein stability in the chaperonin cage is reduced dramatically by more than 5 kcal mol−1 compared to that in bulk solution . Given that steric confinement alone is stabilizing , our results indicate that hydrophobic and/or electrostatic effects in the cavity are strongly destabilizing . Our findings are consistent with the iterative annealing mechanism of action proposed for the chaperonin GroEL . The Escherichia coli GroE chaperonin system , which comprises GroEL and its co-factor GroES , assists protein folding in vivo and in vitro in an ATP-dependent manner ( Thirumalai and Lorimer , 2001; Saibil et al . , 2013; Hayer-Hartl et al . , 2016; Gruber and Horovitz , 2016 ) . Binding of GroES to GroEL forms a cage in which encapsulated substrate proteins can fold in isolation from bulk solution . The GroE system has been studied intensively for more than three decades , but it is still unclear and controversial whether its cavity is a ‘passive cage’ in which protein substrate aggregation is prevented but the folding pathway is unchanged or a chamber in which the folding process is altered in some manner . It is also unclear whether encapsulation in the GroE cavity is thermodynamically stabilizing , for example because of confinement , or destabilizing owing , for example , to a diminished hydrophobic effect . The effects of encapsulation on folding kinetics have been examined in different studies but it has been difficult to generalize its impact , in part , because it may depend on the properties of the protein substrate studied such as its size and charge . According to some studies , encapsulation prevents aggregation but does not alter the protein substrate’s folding pathway and kinetics ( Horst et al . , 2007; Tyagi et al . , 2011 ) . In the case of rhodanese , for example , it was reported that encapsulation retards the folding of its C-terminal domain but has no effect on its N-terminal domain ( Hofmann et al . , 2010 ) . By contrast , other studies have suggested more ‘active’ models according to which folding is accelerated in the GroEL cavity compared to bulk solution . Recent work showed , for example , that GroEL repairs a folding defect of a mutant of maltose-binding protein by accelerating the formation of an on-pathway intermediate ( Ye et al . , 2018 ) . Accelerated folding upon encapsulation has been attributed to steric confinement and the negative charges of the cavity walls ( Tang et al . , 2006 ) . It has also been attributed to substrate interactions with the cavity-protruding C-terminal tails of GroEL , which comprise Gly-Gly-Met repeats ( Tang et al . , 2006; Weaver and Rye , 2014; Weaver et al . , 2017 ) . In addition , GroEL-mediated folding can be affected by the cavity-confined water , which can enhance the hydrophobic effect ( and thus accelerate folding ) by accumulating near the cavity walls ( England et al . , 2008 ) or diminish it ( and thus slow folding ) by being more ordered . Finally , an even more ‘active’ mechanism has been proposed according to which encapsulated misfolded protein substrates undergo ATP-promoted forced-unfolding ( Weaver and Rye , 2014 ) , in accordance with the iterative annealing model ( Todd et al . , 1996 ) , thereby giving them further opportunity to fold correctly . The iterative annealing model predicts higher folding yields but it has been suggested that annealing can also lead to accelerated folding ( Tang et al . , 2006; Gupta et al . , 2014; Weaver and Rye , 2014; Weaver et al . , 2017 ) . In contrast with the many and often conflicting studies on the kinetic effects of encapsulation , there have been virtually no reports on the thermodynamic effects of encapsulation . One complication in understanding the thermodynamic effects of encapsulation has been that protein substrates spend variable amounts of time folding in the cavity and in bulk solution because of GroEL-GroES cycling and the leakiness of its cavity ( Motojima and Yoshida , 2010 ) . As shown below , such leakiness is even greater in the complex of single-ring GroEL with GroES , which was used in some studies ( Hofmann et al . , 2010 ) and has been suggested to form during GroEL’s normal reaction cycle ( Yan et al . , 2018 ) . Another complication has been that measuring protein stability usually involves perturbations such as temperature or solution changes , which could also affect the stability of the chaperonin itself . The goal of the work described here was , therefore , to establish a system that would allow measuring the stability of an encapsulated protein in a non-leaky and unperturbed GroE complex . An appropriate system was found to be a chimera of dihydrofolate reductase from Moritella profunda ( DHFRMp ) fused to eGFP , which is encapsulated in the football-shaped and non-cycling BeFx-stabilized GroEL-GroES2 complex ( Figure 1 ) . Consequently , we were able to isolate the thermodynamic effect of encapsulation from other effects associated with GroEL-GroES cycling . Our results show that protein stability in the GroEL cavity is reduced dramatically in comparison with bulk solution . We chose to use dihydrofolate reductase from Moritella profunda , a psychrophilic bacterium with an optimal growth temperature at 2°C , as a model substrate since it is a relatively unstable protein at room temperature Xu et al . , 2003 whose folding can be monitored easily by measuring the regain in enzyme activity . DHFRMp was fused to the C-terminus of enhanced green fluorescent protein ( eGFP ) in order to further destabilize it as observed before for other proteins ( Sokolovski et al . , 2015; Dave et al . , 2016 ) . The fusion to eGFP also facilitated easy and sensitive determination of the location of the substrate , that is whether it is in the cavity or has leaked outside of it . Under our conditions , the apparent melting temperature , Tm , in bulk solution of DHFRMp alone was found to be about 41 . 2 ( ±1 . 8 ) °C whereas DHFRMp fused to eGFP was found to be strongly destabilized with a Tm of 22 . 8 ( ±1 . 1 ) °C ( Figure 2 ) . eGFP in the chimera was also found to be destabilized in bulk solution relative to eGFP alone , but to a lesser extent , with Tm values of 76 . 3 ( ±0 . 3 ) and 81 . 3 ( ±1 . 3 ) °C , respectively . GroEL and GroES can form either GroEL-GroES bullet-shaped or GroEL-GroES2 football-shaped complexes . It has been reported that the GroEL-GroES2 complex is stabilized in the presence of BeFx ( Taguchi et al . , 2004 ) . We , therefore , tested whether eGFP remains encapsulated in the BeFx-stabilized ‘football’ complex over a sufficiently long period of time . BeFx-stabilized and eGFP-containing ‘football’ complexes were prepared , purified by gel-filtration and then allowed to stand overnight at room temperature . The samples were then analyzed by gel-filtration in order to determine the extent , if any , of substrate escape . The results show that all the eGFP co-eluted with GroEL and GroES , thereby indicating that the complex remained intact and no escape occurred ( Figure 1—figure supplement 1 ) . Two other proteins , the p53 core domain and a chimera of the engrailed homeodomain transcription factor with eGFP with respective masses of 22 . 4 and 42 . 9 kDa , were also tested in this manner and found to not escape ( data not shown ) . Previously , it was believed that substrates encapsulated in the cage formed by single-ring GroEL in complex with GroES cannot escape since dissociation of GroES from one ring of GroEL is triggered by ATP binding to the opposite ring ( Rye et al . , 1997 ) . Experiments carried out with eGFP encapsulated in the cavity of single-ring GroEL in complex with GroES showed , however , that massive substrate leakage took place ( Figure 1—figure supplement 2 ) . BeFx-stabilized ‘football’ complexes containing the chimera were purified and the DHFR activity of the encapsulated chimera was monitored at 23°C ( i . e . near the apparent Tm of the fused DHFRMp in bulk solution ) by following the change in absorbance at 340 nm , upon addition of the substrates NADPH and dihydrofolate ( DHF ) . The data show a lag phase followed by a linear phase ( Figure 3 ) before activity starts to diminish owing to substrate depletion . Strikingly , the lag phase is absent when the chimera is free in bulk solution ( Figure 3 ) . One possible reason for the presence of the lag phase , in the case of the encapsulated chimera , is that the added substrates ( DHF and NADPH ) need to diffuse into the cavity . In such a case , the rate constant associated with the lag phase should increase with increasing substrate concentrations . Alternatively , the lag phase may reflect substrate ( DHF and/or NADPH ) -promoted folding of the DHFRMp part of the chimera , which is destabilized in the cavity relative to bulk solution . Substrate ( DHF and/or NADPH ) -promoted folding of the DHFRMp part of the chimera can take place via a mechanism of conformational selection , that is the substrates bind only to the folded state , thereby shifting the equilibrium in its favor . In such a case , the rate constant associated with the lag phase should decrease with increasing substrate concentrations ( Vogt and Di Cera , 2012 ) . Alternatively , substrate-promoted folding of the DHFRMp part of the chimera can also occur via a mechanism of induced fit in which case the rate constant associated with the lag phase should increase with increasing substrate concentrations . In other words , an increase in the rate constant of the lag phase with increasing substrate concentration can be due to substrate penetration or induced fit , whereas a decrease is evidence for conformational selection . Discriminating between these mechanisms can , therefore , be achieved by measuring the substrate concentration dependence of the rate constant of the lag phase . The change in the absorbance at 340 nm due to the enzyme activity of the encapsulated or free chimera was , therefore , monitored in the presence of different concentrations of DHF and a fixed concentration of NADPH . The data were fitted to: ( 1 ) [P]=Vt+A ( e−λt−1 ) where [P] is the product concentration ( or the absorbance at 340 nm which is proportional to it ) , V is the slope that corresponds to the linear steady-state velocity of the reaction and A and λ are the respective amplitude and rate constant of the lag phase . Equation 1 can be derived for the reaction scheme EU⇄EF⇄ES , where EU , EF and ES designate the respective unfolded , folded and substrate-bound folded states of the protein ( E ) , assuming that ligand binding is fast relative to the folding and unfolding steps . In such a case , the rate constant , λ , is given by Vogt and Di Cera , 2012: ( 2 ) λ=k1+k-1k-2k2[S]+k-2=k1+k-1Ka[S]+1where k1 and k-1 are the respective folding and unfolding rate constants , k2 and k-2 are the respective substrate association and dissociation rate constants and Ka = k2/k-2 is the substrate association constant . Inspection of Equation 2 shows that the value of the observed rate constant , λ , decreases with increasing substrate concentration as observed here in the case of the encapsulated chimera ( Figure 4A ) . This finding indicates that the DHFRMp part of the chimera is significantly destabilized in the cavity , but not in bulk solution ( where the lag phase is absent ) , and that the substrates DHF and NADPH shift its equilibrium toward the folded state via a conformational selection mechanism . In agreement with this finding , temperature melts of the chimera in bulk solution show that both DHF and NADPH stabilize its DHFRMp part ( Figure 4—figure supplement 1 ) . It should be noted that , in principle , a similar kinetic behavior would be observed for a scheme E'⇄EF⇄ES , that is when the equilibrium is between the folded protein and a mis-folded inactive species , E’ ( instead of between the folded and unfolded species ) . This possibility is unlikely , however , because it was found that only 20% of human DHFR mis-folds inside the cavity of the non-cycling complex of single-ring GroEL with GroES ( Horst et al . , 2007 ) . In such a case , the expected lag phase due to the mis-folded population would not be observed owing to the activity of the remaining DHFR , which does not mis-fold . Moreover , according to this model , the mis-folded state is much more stable than the folded state , which is in violation of Anfinsen’s dogma that the native state is at the minimum free energy if the conditions in the cavity are assumed to favor folding . Values of the folding and unfolding rate constants were obtained by fitting the plot of λ as a function of substrate ( DHF ) concentration to Equation 2 and found to be 1 . 4 ( ±0 . 4 ) x 10−4 and 7 . 3 ( ±0 . 8 ) x 10−3 sec , respectively ( Figure 4A , B ) . Consequently , the stability of the cavity-confined fused DHFRMp , in the absence of the substrates DHF and NADPH , is 2 . 4 ( ±0 . 2 ) kcal mol−1 , that is the folded state is extremely destabilized so that it is less stable than the unfolded state . By contrast , the stability in bulk solution of fused DHFRMp , in the absence of the substrates DHF and NADPH , is −3 . 45 ( ±0 . 09 ) kcal mol−1 ( Figure 4—figure supplement 2 ) . This value was determined by measuring the stability of fused DHFRMp , in the presence of different concentrations of NADPH , in order to minimize aggregation that takes place in bulk solution , and then extrapolating to zero concentration of NADPH . Taken together , the results obtained here show that encapsulation in the GroEL cavity destabilizes the DHFRMp part of the chimera by more than five kcal mol−1 relative to bulk solution . Interestingly , analysis of the linear phase in the progress curves ( e . g . in Figure 3 ) shows that the steady-state enzyme activity of DHFRMp is also affected by encapsulation . The Km value of the encapsulated fused DHFRMp for DHF is about 0 . 28 ( ±0 . 09 ) mM and , thus , about 10-fold higher than the value of 33 ( ±2 ) µM measured in bulk solution ( Figure 4C ) , in agreement with a previous determination ( Xu et al . , 2003 ) . The value of kcat may also be affected but these values are difficult to determine because of uncertainties regarding the active protein concentrations . In order to test whether the destabilization of the cavity-confined fused DHFRMp is due to its interaction with the cavity walls , we carried out fluorescence anisotropy decay measurements . Fits of the decay curves for the free and caged chimeras yielded similar rotational times of 22 . 85 ( ±0 . 30 ) ns and 25 . 40 ( ±0 . 45 ) ns , respectively ( Figure 5 ) . Likewise , the fits of the decay curves for free and caged eGFPs yielded similar rotational times of 14 . 38 ( ±0 . 20 ) ns and 15 . 90 ( ±0 . 25 ) ns , respectively ( Figure 5—figure supplement 1 ) , in agreement with previous work ( Striker et al . , 1999 ) . The results indicate free mobility of the chimera and eGFP inside the GroEL football complex . It is , therefore , unlikely that the cavity-induced thermodynamic destabilization of the fused DHFRMp is due to interactions with the cavity walls . Our results show that protein stability in the chaperonin cage is reduced dramatically by more than 5 kcal mol−1 compared to that in bulk solution . Given that steric confinement alone is expected to be stabilizing , our results indicate that protein destabilization in the cavity is likely to be due to hydrophobic and/or electrostatic effects . One possibility is that water in the cavity becomes more ordered in the presence of a substrate protein , thereby leading to a diminished hydrophobic effect . All-atom molecular dynamics simulations have shown that pairwise hydrophobic interactions are destabilized in hydrophobic nanopores when they contain water at bulk density ( Vaitheeswaran and Thirumalai , 2008 ) , as indicated by some evidence in the case of the GroEL cavity ( Franck et al . , 2014 ) . The cavity walls of the ‘football’ complexes in our experiments are , however , charged and the extent of destabilization would , therefore , depend on many factors including the distribution of charges on the cavity walls , the cavity geometry and the presence of regions where the water structure is disrupted . Future simulations should test how these various factors affect pairwise interactions under confinement . Similar to man-made machines , biological machines such as GroEL cycle between states that differ in their structural and functional properties . Substrate proteins first bind to GroEL in its apo state and can become destabilized through binding to its hydrophobic cavity walls ( see , for example , Libich et al . , 2015 ) . ATP binding to the GroEL-protein substrate complex , which follows , can cause further unfolding due to a stretching force applied to the substrate protein upon ATP-binding-promoted conformational changes ( Weaver et al . , 2017 ) . Next , protein substrates are encapsulated in the GroES-capped cavity for 1–10 s ( Bigman and Horovitz , 2019 ) before being discharged into bulk solution . Here , we have succeeded in isolating the effect of this key step on substrate stability and found that it can also lead to protein destabilization . It is important to point out , however , that the magnitude ( and possibly direction ) of the effect is likely to depend on the protein substrate’s size and other properties . Hence , future studies should employ the strategy developed here to examine other proteins . Finally , it is important to note that our finding that the cavity environment is destabilizing supports the iterative annealing mechanism of action proposed for the chaperonin GroEL ( Todd et al . , 1996; Thirumalai et al . , 2020 ) . According to this mechanism , protein substrates undergo kinetic partitioning , during each reaction cycle of GroEL , between productive folding to the native state and mis-folding . The remaining fraction of mis-folded protein substrates are then rebound to GroEL and ( partially ) unfolded , thereby giving them new opportunity to fold correctly . Our results show that ( partial ) unfolding is achieved not only because of binding to the hydrophobic cavity of the apo state and ATP-promoted forced unfolding , as shown before by others , but also because it is strongly favored thermodynamically under the conditions in the GroES-capped cavity . The Pet21a plasmid containing the gene for eGFP ( with the mutation A206K that stabilizes its monomeric state ) and an N-terminal His7-tag ( Bandyopadhyay et al . , 2017 ) was amplified by PCR using the following respective forward and reverse primers: 5’-ACTCGAGCACCACCACCACCACCACTGA-3’ 5’-GCTGTATAAGGGCAACCTGTATTTTCAGGGCACTCGAGCACCACC-3’ . This amplification resulted in introducing a TEV protease cleavage site at the C-terminus of eGFP . The gene coding for DHFRMp was then introduced at the 3’ end of the TEV coding sequence by restriction-free cloning using the forward and reverse primers: 5’-CCTGTATTTTCAGGGCATGATCGTAAGCATGATTGCCGCACTGGCG-3’ 5’-GGTGGTGGTGGTGCTCGAGTTCACTCGAGTTTGACTCTTTCAAGTAGAC-3’ , respectively . The final gene product codes for eGFP with an N-terminal His7-tag fused at its C-terminus to DHFRMp via a linker sequence containing a TEV protease cleavage site . E . coli BL21 cells ( Studier and Moffatt , 1986 ) harboring the Pet21d plasmid containing the gene for the chimera were inoculated into 250 ml 2 x TY with 100 µg/ml ampicillin and grown at 37°C until an O . D . of 0 . 6 was reached . Expression was induced by addition of 0 . 5 mM IPTG and growth was continued overnight at 16°C . The cells were then spun at 11 , 970 g for 20 min at 4°C , resuspended in 10% sucrose solution , spun again at 3 , 452 g for 30 min at 4°C and the pellet was stored at −80°C until purification . The pellet was resuspended in 30 ml of 50 mM Tris-HCl buffer ( pH 7 . 5 ) containing 10 mM NaCl and 1 mM DTT ( buffer A ) with 8 M urea to prevent proteolysis . The cells were lysed using a French press ( three passes ) and sonication ( 4 cycles of 20 s sonication at 70% intensity and 60 s intervals ) . The lysate was then centrifuged at room temperature for 10 min at 24 , 610 g and again at 38 , 720 g for 30 min . The supernatant was diluted with buffer A to 6 M urea and loaded on a 140 ml Q-Sepharose column pre-equilibrated with buffer A containing 6 M urea . The column was then washed with 200 ml buffer A containing 6 M urea and with 150 ml of this buffer containing also 150 mM NaCl . A 5 ml HisTrap column , which was pre-equilibrated with buffer A containing 6 M urea , was connected downstream to the Q-Sepharose column and both columns were washed with 200 ml of buffer A with 1 M NaCl . The Q-Sepharose column was then removed and the HisTrap column was washed at room temperature with 75 ml of 50 mM Tris-HCl buffer ( pH 7 . 5 ) containing 100 mM NaCl , 15 mM imidazole and 1 mM DTT ( buffer B ) with 5 M urea . Elution was carried out with a 75 ml linear gradient of 15 to 500 mM imidazole in buffer B with 5 M urea . Fractions were analyzed with SDS-PAGE and those containing the chimera were combined and concentrated to ~8 ml . The protein was then refolded at 4°C by mixing it with buffer B at a ratio of 1 . 5:98 . 5 , respectively , during loading on a HisTrap 5 ml column . After loading , the column was washed with 25 ml buffer B and the protein was eluted with a 75 ml linear gradient from 15 to 500 mM imidazole in buffer B . Fractions containing the pure chimera were identified by SDS-PAGE and pooled and the buffer was then exchanged to 20 mM Hepes ( pH 8 . 0 ) containing 100 mM KCl , 50 mM MgCl2 , 50 mM Na2SO4 , 10 mM NaF , 1 mM BeSO4 and 1 mM DTT ( working buffer ) using a PD 10 desalting column at 4°C . The concentration was determined from the absorption at 280 nm using an extinction coefficient of 50435 M−1 cm−1 and the protein was aliquoted and stored at −80°C for further use . Expression and purification of DHFRMp were achieved as before ( Xu et al . , 2003 ) with the following changes . E . coli Rosetta cells harboring the Pet21d plasmid containing the DHFRMp gene were inoculated into 7 L 2 x TY medium and grown at 37°C until an O . D . of 1 was reached . IPTG ( 0 . 5 mM ) was then added to induce protein expression and the cells were grown at 37°C for another 5 hr . Harvesting was carried out by spinning the cells at 11 , 970 g for 20 min at room temperature and then resuspending them in 100 ml buffer B containing 10 µg/ml aprotinin , 5 µg/ml antipain , 5 µg/ml pepstatin , 5 µg/ml chymostatin , 10 µg/ml leupeptin , 50 μl EDTA-free protease inhibitor cocktail ( Calbiochem ) , 1 . 5 μM PMSF , 6 μg/ml RNase A , 30 µg/ml DNase and 0 . 22 mg/ml folate . The cells were lysed by sonication ( 4 cycles of 20 s sonication at 70% intensity and 60 s intervals ) and one passage through a French press at 1500 atm . The lysate was centrifuged at 4°C for 30 min at 24 , 610 g and then for 30 min at 38 , 720 g . The supernatant was applied to a 5 ml His-Trap column , which was then washed with 25 ml buffer B , 25 ml buffer B containing 1 mM ATP and 50 mM MgCl2 and then with 25 ml buffer B again . DHFRMp was eluted with a 50 ml linear gradient of 15 mM to 500 mM imidazole in buffer B . Fractions were analyzed with SDS-PAGE and those containing DHFRMp were combined , concentrated and the buffer was exchanged to 50 mM Tris-HCl ( pH 8 . 0 ) containing 250 mM NaCl , 5 mM EDTA , 5 mM β-mercaptoethanol ( buffer C ) using a PD-10 desalting column at 4°C . The protein was applied to a 5 ml methotrexate column , which was then washed with 20 ml buffer C . DHFRMp was eluted with a linear gradient of 70 ml from 0 to 1 mM folate and 0 . 25 to 1 M NaCl in buffer C . Fractions were analyzed with SDS-PAGE and those containing DHFRMp were combined , concentrated , mixed with 1 mM ATP and 50 mM MgCl2 and then applied to a Superdex 75 10/300 gel-filtration column equilibrated with working buffer . Fractions were analyzed by SDS-PAGE and those containing pure DHFRMp were combined , aliquoted and stored at −80°C . The protein concentration was determined using the Bradford assay and a BSA-based calibration curve . The melting temperatures of eGFP and DHFRMp alone and in the chimera were measured by differential scanning fluorimetry using a StepOne real-time PCR instrument ( Applied Biosystems ) . The protein ( 5 µM ) alone or in the presence of 500 µM NADPH or dihydrofolate was mixed with 1X SYPRO Orange reagent ( Sigma ) in a 20 µl reaction volume of working buffer . The sample was heated from 4°C to 94°C with a temperature increment of 0 . 5°C every 45 s . The fluorescence of SYPRO Orange was measured after each temperature increment by exciting at 492 nm and measuring the emission at 610 nm . Analysis of the data was carried out using the StepOne software V2 . 3 . The stability of the DHFRMp part of the chimera was determined by measuring the fluorescence emission at 330 nm upon excitation at 280 nm ( using an ISS PC1 fluorimeter with excitation and emission bandwidths of 16 nm ) as a function of GuHCl concentration at different fixed concentrations of NADPH . Measurements were made at 23°C for samples containing about 1 . 4 μM of the chimera in working buffer after incubation for 10 min at the same temperature . The concentration of the stock solution of GuHCl was determined by measuring the refraction index at 23°C . The data were fitted using the following equation: ( 3 ) F=FU0+a[D]+ ( FN0+b[D] ) e−ΔG0+m[D]RT1+e−ΔG0+m[D]RTwhere ΔG0 is the free energy of unfolding in the absence of denaturant , [D] is the GuHCl concentration , m is the GuHCl-concentration dependence of the free energy of unfolding ( m=∂ΔG∂[D] ) , T is the temperature and R is the gas constant . The fluorescence of the native ( N ) and denatured ( U ) states are expressed as linear functions of the GuHCl concentration with slopes of a and b , respectively . This analysis is based on the assumption that the melting curves reflect the denaturation of only the DHFRMp part of the chimera , which is justified since eGFP is very stable under our conditions and the fluorescence of its single tryptophan residue changes very little ( and in a linear fashion ) as a function of GuHCl concentration . The values of ΔG0 obtained from the fits to Equation 3 were then plotted as a function of the concentration of NADPH and the data were fitted using the following equation: ( 4 ) ΔG=ΔG0−RTln ( 1+[S]/Km ) where [S] is the concentration of NADPH and Km is its Michaelis-Menten constant of 13 µM ( Evans et al . , 2010 ) . Here , ΔG0 is the free energy of unfolding in the absence of both GuHCl and NADPH . Encapsulation of eGFP and the chimera in the GroEL-GroES2 football was performed using different methods of denaturation . In the case of eGFP , the substrate ( 7 . 5 nmol ) was denatured in 200 µl of 60 mM HCl in a siliconized test-tube . The denatured eGFP was then added slowly to 20 ml of 50 mM Tris-HCl buffer ( pH 7 . 5 ) containing 100 mM KCl , 50 mM MgCl2 , 1 mM DTT , 0 . 05 µM GroEL oligomer and 0 . 125 µM GroES oligomer ( folding buffer ) and left for 40 min at room temperature with mild stirring . Football assembly was initiated by adding 2 ml containing 550 mM Na2SO4 , 110 mM NaF , 11 mM BeSO4 and 11 mM ATP ( activation mix ) to the folding buffer . The mixture was then stirred for 15 min at room temperature , concentrated and fractionated using a Sepharose 6 ( 10/300 ) gel-filtration column in working buffer . Fractions were analyzed by SDS-PAGE and those containing GroE-encapsulated eGFP were kept for further work . In the case of the chimera , a binary complex was first formed by incubating 9 nmol of the native chimera with ~5 nmol of GroEL oligomer in 300 µl of working buffer overnight at room temperature . The mixture was then added to a 200 µl solution containing 135 mM Na2SO4 , 25 mM NaF , 2 . 5 mM BeSO4 , 2 . 5 mM ATP and ~10 nmol of GroES oligomer , stirred for 15 min at room temperature and applied to a Superose 6 10/300 gel-filtration column . Activity assays were performed at 23°C in working buffer using an Infinite M200pro ( TECA Group Ltd . ) plate reader . Reactions were initiated by mixing 190 µl of the protein with 10 µl of NADPH ( 350 µM final concentration ) and different concentrations of DHF . The reaction progress was followed by measuring the decrease in absorption at 340 nm as a function of time . The chimera concentrations were verified from measurements of the fluorescence emission at 509 nm upon excitation at 488 nm ( with bandwidths of 4 , 8 or 16 nm depending on the protein concentration ) and using an eGFP-based calibration curve . Analysis and fitting of the data were performed using MatLab 2015b software . Fluorescence anisotropy decay curves of the target proteins were measured using a MicroTime200 fluorescence microscope ( PicoQuant ) . Samples of eGFP and the chimera were diluted to 50–100 nM in the working buffer containing 0 . 01% TWEEN 20 and then loaded into a flow cell pre-coated with a lipid bilayer ( Mazal et al . , 2019 ) . Molecules were excited with a 485 nm diode laser pulsed at a repetition rate of 20 MHz and with a power of 10 µW . Emitted photons were divided based on their polarization using a polarizing beam splitter cube , followed by filtration using band-pass filters ( 520/35 nm , BrightLine ) . Photon arrival times relative to the excitation pulse were registered with a resolution of 16 ps using two single-photon avalanche photo-diodes detectors ( Excelitas SPCM-AQR-14-TR ) coupled to a time-correlated single-photon counting module ( HydraHarp 400 , PicoQuant ) . The parallel and perpendicular fluorescence decays were constructed from the data and background corrected . Fluorescence anisotropy decays were then calculated using the following relation: rt=I∥ ( t ) -GI⊥ ( t ) I∥ ( t ) +2GI⊥ ( t ) , where I∥ ( t ) and I⊥ ( t ) are the time-dependent fluorescence intensities of the parallel and perpendicular components , and G is the polarization sensitivity factor ( whose value was determined to be 1 ) . Fluorescence anisotropy decays were fitted to a single exponential function to obtain the rotational correlation times of the proteins .	All cells contain molecules known as proteins that perform many essential roles . Proteins are made of chains of building blocks called amino acids that fold to form the proteins’ three-dimensional structures . Many proteins fold spontaneously into their well-defined and correct structures . However , some proteins fold incorrectly , which prevents them from working properly , and can lead to formation of aggregates that may harm the cell . To prevent such damage , cells have evolved proteins known as molecular chaperones that assist in the folding of other proteins . For example , a molecular chaperone called GroEL is found in a bacterium known as Escherichia coli . This molecular chaperone contains a cavity which prevents target proteins from forming clumps by keeping them away from other proteins . However , it remained unclear precisely how GroEL works and whether enclosing target proteins in its cavity has other effects . Moritella profunda is a bacterium that thrives in cold environments and , as a result , many of its proteins are unstable at room temperature and tend to unfold or fold incorrectly . To study how GroEL works , Korobko et al . used a protein from M . profunda called dihydrofolate reductase as a target protein for the chaperone . A clever trick was then used to determine the folding state of dihydrofolate reductase when inside the chaperone cavity . The experiments revealed that the environment within the cavity of GroEL strongly favors dihydrofolate reductase adopting its unfolded state instead of its folded state . This suggests that GroEL helps dihydrofolate reductase and other incorrectly folded target proteins to unfold , thus providing the proteins another opportunity to fold again correctly . Parkinson’s disease , Alzheimer’s disease and many other diseases are caused by proteins folding incorrectly and forming aggregates . A better understanding of how proteins fold may , therefore , assist in developing new therapies for such diseases . These findings may also help biotechnology researchers develop methods for producing difficult-to-fold proteins on a large scale .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2020	Measuring protein stability in the GroEL chaperonin cage reveals massive destabilization
Although numerous extracellular phosphoproteins have been identified , the protein kinases within the secretory pathway have only recently been discovered , and their regulation is virtually unexplored . Fam20C is the physiological Golgi casein kinase , which phosphorylates many secreted proteins and is critical for proper biomineralization . Fam20A , a Fam20C paralog , is essential for enamel formation , but the biochemical function of Fam20A is unknown . Here we show that Fam20A potentiates Fam20C kinase activity and promotes the phosphorylation of enamel matrix proteins in vitro and in cells . Mechanistically , Fam20A is a pseudokinase that forms a functional complex with Fam20C , and this complex enhances extracellular protein phosphorylation within the secretory pathway . Our findings shed light on the molecular mechanism by which Fam20C and Fam20A collaborate to control enamel formation , and provide the first insight into the regulation of secretory pathway phosphorylation . Reversible phosphorylation is a fundamental mechanism used to regulate cellular signaling and protein function . In the past several decades , efforts have been focused on understanding phosphorylation events in the cytoplasm and nucleus . However , little is known about the function and regulation of protein phosphorylation within the secretory pathway , despite the fact that numerous secreted proteins that carry out important biological functions are phosphorylated ( Bahl et al . , 2008; Zhou et al . , 2009; Carrascal et al . , 2010; Salih et al . , 2010; Stone et al . , 2011 ) . This is likely because the kinases in the secretory pathway have only recently been discovered ( Ishikawa et al . , 2008; Tagliabracci et al . , 2012 , 2013; Bordoli et al . , 2014 ) . Our laboratory recently identified a family of secretory pathway kinases , which includes the ortholog of Drosophila Four-jointed ( Fj ) as well as Fam20A , Fam20B , Fam20C , Fam198A and Fam198B ( Ishikawa et al . , 2008; Tagliabracci et al . , 2012 ) . Fam20C was identified as the long-sought Golgi casein kinase , which phosphorylates secreted proteins within Ser-x-Glu/phospho-Ser ( SxE/pS ) motifs , including casein and members of the secretory calcium binding phosphoprotein ( SCPP ) family ( Kawasaki and Weiss , 2008; Tagliabracci et al . , 2012 ) . Furthermore , some 75% of phosphoproteins identified in human serum and cerebrospinal fluid contain phosphate within the Fam20C consensus sequence , suggesting that Fam20C may have a broad spectrum of substrates and therefore play a major role in establishing the secreted phosphoproteome ( Bahl et al . , 2008; Zhou et al . , 2009; Salvi et al . , 2010; Tagliabracci et al . , 2013 ) . Nonetheless , how Fam20C is regulated is unknown . In addition to Fam20C , Fj , the founding member of the family , phosphorylates extracellular cadherin domains ( Ishikawa et al . , 2008 ) . Fam20B phosphorylates xylose within a tetrasaccharide linkage region of O-linked proteoglycans and functions to promote glycosaminoglycan chain elongation ( Koike et al . , 2009; Wen et al . , 2014 ) . The biochemical functions of the other members of this protein family are unknown . Here we address the function of Fam20A , the closest paralog of Fam20C . Both Fam20C and Fam20A are implicated in biomineralization . Fam20C deficiency in humans is associated with a severe and often lethal osteosclerotic bone dysplasia known as Raine syndrome ( Simpson et al . , 2007 ) . Patients with non-lethal forms of Raine syndrome exhibit hypophosphatemia , ectopic calcifications and dental anomalies ( Simpson et al . , 2009; Fradin et al . , 2011; Rafaelsen et al . , 2013 ) . These patients have misformed dentin and enamel , suggesting that Fam20C is essential for dentinogenesis and amelogenesis ( Fradin et al . , 2011; Rafaelsen et al . , 2013 ) . Furthermore , patients with FAM20A gene mutations also develop disorders of enamel formation referred to as Amelogenesis Imperfecta ( AI ) , which is often accompanied by ectopic calcification , such as nephrocalcinosis ( O'Sullivan et al . , 2011; Cho et al . , 2012; Jaureguiberry et al . , 2012; Cabral et al . , 2013; Wang et al . , 2013a , 2014; Kantaputra et al . , 2014a , 2014b ) . Notably , both Fam20A-knockout ( KO ) and Fam20C-KO mouse models have been generated and they exhibit similar enamel phenotypes , suggesting that Fam20A and Fam20C might function in the same pathway to control enamel formation ( Vogel et al . , 2012; Wang et al . , 2012 , 2013c ) . Further evidence to support a role for extracellular protein phosphorylation in the regulation of enamel formation comes from the discovery that AI can be caused by a missense mutation in the secreted protein enamelin ( ENAM ) that disrupts phosphorylation within an SxE motif ( Chan et al . , 2010 ) . ENAM is an enamel matrix protein that , together with amelogenin X ( AMELX ) , ameloblastin ( AMBN ) and amelotin ( AMTN ) , makes up a subfamily of the SCPPs ( Kawasaki and Weiss , 2008 ) . These proteins are secreted from specialized cells known as ameloblasts and act as a scaffold for enamel calcification ( Hu et al . , 2007; Moradian-Oldak , 2012 ) . ENAM , AMELX , AMBN and AMTN are all phosphorylated within SxE motifs and are therefore potential Fam20C substrates ( Kawasaki and Weiss , 2008 ) . In addition to ENAM phosphorylation , phosphorylation within a conserved SxE motif in AMELX also appears to be important for enamel mineralization ( Kwak et al . , 2009 ) . Here we show that Fam20C phosphorylates enamel matrix proteins in vitro and in cells . Even though Fam20A lacks a residue critical for catalysis and appears to be a pseudokinase , it forms a functional complex with Fam20C and allosterically activates Fam20C to efficiently phosphorylate secreted proteins . Our results reveal a novel mechanism to regulate secretory protein phosphorylation , and provide a molecular link between an observed biochemical function of Fam20A and the phenotype of patients with AI . To determine whether enamel matrix proteins are substrates of Fam20C , we expressed and purified recombinant human AMELX , AMTN , and the conserved proteolytic fragment of ENAM ( aa173–277 ) found in developing enamel as 6× His tag-fusion proteins in Escherichia coli ( Al-Hashimi et al . , 2009 ) . Recombinant Fam20C phosphorylated each of these proteins in a time-dependent manner in vitro , whereas an inactive Fam20C D478A mutant did not ( Figure 1A ) . ENAM ( 173–277 ) contains two SxE motifs ( S191 and S216 ) , both of which are highly conserved and known to be phosphorylated ( Fukae et al . , 1996; Al-Hashimi et al . , 2009 ) . As mentioned above , a S216L mutation causes AI ( Chan et al . , 2010 ) . We generated ENAM S191A and S216L mutants and tested them as substrates for Fam20C . As shown in Figure 1B , both the S191A and S216L ENAM mutants exhibited markedly reduced phosphorylation by Fam20C . Further , mutation of both Ser residues completely abolished Fam20C-dependent phosphorylation ( Figure 1B ) . Thus , Fam20C phosphorylates ENAM ( 173–277 ) on S191 and S216 . 10 . 7554/eLife . 06120 . 003Figure 1 . Fam20C phosphorylates the enamel matrix proteins within SxE motifs . ( A ) Phosphorylation of enamel matrix proteins by Fam20C in vitro . 250 μg/ml purified recombinant human ENAM ( 173–277 ) , AMTN or AMELX was incubated with recombinant Fam20C , Fam20C-D478A ( DA ) or Fam20A ( 5 μg/ml ) at 30°C in the presence of [γ-32P]-ATP . After reaction , proteins were separated by SDS-PAGE and visualized by Coomassie staining . 32P incorporation was detected by autoradiography . ( B ) Effect of SxE motif mutations on ENAM phosphorylation by Fam20C . Upper: the schematic of ENAM protein . Dark green , ENAM ( 173–277 ) . SP , signal peptide; AI , Amelogenesis Imperfecta . S216L mutation was reported to cause AI . Lower: phosphorylation of different forms of ENAM ( 173–277 ) by Fam20C . SA/SL , S191A and S216L . ( C ) Effect of Fam20C KO on the phosphorylation of AMBN . C-terminal V5-tagged AMBN ( AMBN-V5 ) was expressed in WT ALCs or two Fam20C-KO ALC clones ( C421 and C346 ) that were metabolically labeled with 32P orthophosphate . AMBN-V5 was immunoprecipitated from the conditioned media . Total protein and 32P incorporation were detected by western blot and autoradiography , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 00310 . 7554/eLife . 06120 . 004Figure 1—figure supplement 1 . CRISPR-mediated Fam20C KO in ALCs and its effect on osteopontin ( OPN ) phosphorylation . ( A ) The schematic of the single-guide ( sg ) RNA-targeting sites in the mouse Fam20c gene . Two targeting sequences were used in this study: m20c #3 and #4 . The first exon of the mouse Fam20c gene is shown as a box . White , 5′ UTR; Orange , the coding sequence . Targeting sites and protospacer adjacent motifs ( PAMs ) are indicated as blue and red bars , respectively . ( B and C ) Sequence alignment of WT Fam20c gene and the disrupted alleles from Fam20C-KO ALC clone C346 ( B ) or C421 ( C ) . Targeting sites and PAMs are indicated as blue and red bars , respectively . Inserted nucleotides are highlighted in red . Deleted regions are indicated with red dashes . Del X , deletion of X nucleotides; ins X , insertion of X nucleotides . ( D ) Effect of Fam20C KO on OPN phosphorylation . OPN is a model substrate of Fam20C . C-terminal V5-tagged OPN ( OPN-V5 ) was expressed in WT ALC or two Fam20C-KO ALC clones ( C346 or C421 ) that were metabolically labeled with 32P orthophosphate . OPN-V5 was immunoprecipitated from the conditioned media . Total protein and 32P incorporation were detected by western blot and autoradiography , respectively . No phosphorylation of OPN-V5 was detected in the KO cells , indicating that Fam20C activity was completely abolished . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 004 To confirm that Fam20C is responsible for phosphorylating enamel proteins in cells , we disrupted the Fam20c gene in mouse ameloblast-like cells ( ALCs; [Nakata et al . , 2003] ) by means of Clustered Regularly Interspaced Short Palindromic Repeats ( CRISPR ) /CRISPR-associated 9 ( Cas9 ) ( Figure 1—figure supplement 1 and Figure 3—figure supplement 1C; [Ran et al . , 2013] ) . We expressed V5-tagged AMBN in wild-type ( WT ) and Fam20C-KO ALCs metabolically labeled with 32P orthophosphate and subsequently analyzed V5-immunoprecipitates from conditioned medium . Disruption of Fam20C completely eliminated AMBN phosphorylation , suggesting that Fam20C is the kinase that phosphorylates enamel matrix proteins in vivo ( Figure 1C ) . Because Fam20A and Fam20C share a high degree of sequence similarity , are expressed in ameloblast cells , and mutations cause similar enamel phenotypes , we anticipated that Fam20A might also phosphorylate enamel matrix proteins ( Vogel et al . , 2012; Wang et al . , 2013b ) . However , recombinant Fam20A was unable to phosphorylate the enamel proteins in vitro ( Figure 1A ) . In fact , Fam20A was unable to phosphorylate any protein or carbohydrate substrates we tested , raising the question of whether Fam20A is an active kinase . A number of kinases exhibit intrinsic ATPase activities in the absence of their physiological substrates . Incubation of recombinant Fam20C with ATP resulted in the time-dependent release of phosphate , which correlated with the kinase activity of Fam20C ( Figure 2A ) . In contrast , Fam20A was unable to hydrolyze ATP ( Figure 2A ) . 10 . 7554/eLife . 06120 . 005Figure 2 . Fam20A lacks a residue critical for kinase activity . ( A ) Intrinsic ATPase activity of Fam20C , Fam20C D478A ( DA ) and Fam20A . Recombinant proteins were incubated with 0 . 5 mM ATP and 10 mM MnCl2 at 30°C for the indicated time . The amount of phosphate released during incubation was determined using malachite green reagent . ( B ) Thermal stability shift assay of MBP , Fam20A and Fam20C . Protein thermal stability was monitored by the fluorescence generated from binding of the dye SYPRO Orange to the hydrophobic region exposed upon protein denaturation . MBP purified the same way as Fam20A and Fam20C was used as a negative control . Reaction buffer contained 10 mM MnCl2 . ( C ) Sequence alignment of Fam20A and Fam20C protein orthologs ( Hs , Homo sapiens; Gg , Gallus gallus; Xt , Xenopus tropicalis; Dr , Danio rerio; Dm , Drosophila melanogaster; Ce , Caenorhabditis elegans ) . Catalytically important residues are highlighted and numbered according to human Fam20C . See Figure 2—figure supplement 1 for an extended alignment . ( D ) Fam20C active site from PDB:4kqb bound to ADP and Mn2+ ions . Conserved canonical kinase ion pair , ion interacting , and catalytic residues are highlighted ( cyan sticks ) and labeled according to human Fam20C . A Fam20-specific loop is colored in green and contributes a unique active site residue E306 . ( E ) Intrinsic ATPase activities of Fam20A Q258E and Fam20C E306Q . Reactions were carried out for 1 hr . ( F ) Effect of Q258E mutation on Fam20A kinase activity . Recombinant OPN was used as the substrate . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 00510 . 7554/eLife . 06120 . 006Figure 2—figure supplement 1 . Structure-guided sequence alignment of Fam20C-related atypical kinases . Labels above the line: P , polar salt bridge pair; I , ion coordinating; C , catalysis . Color codes: gray , mainly small residues; light yellow , mainly hydrophobic residues; light blue , mainly positive residues; light red , mainly negative residues; dark yellow , mainly aromatic residues . The NCBI gi number of each sequence is shown on the left . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 00610 . 7554/eLife . 06120 . 007Figure 2—figure supplement 2 . Effect of E306Q mutation on Fam20C kinase activity . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 007 ATP binding is known to enhance the thermal stability of kinases , thus increasing the melting temperature ( Tm ) ( Niesen et al . , 2007; Murphy et al . , 2014 ) . As expected , the Tm of Fam20C significantly increased in the presence of ATP , whereas the Tm of maltose-binding protein ( MBP ) was unchanged ( Figure 2B ) . We also observed an increase in the thermal stability of Fam20A in the presence of ATP , suggesting that Fam20A binds ATP ( Figure 2B ) . Collectively , these results support that Fam20A can bind , but not hydrolyze ATP . To gain insight into the catalytic differences between Fam20A and Fam20C , we looked for cognate residues missing in Fam20A that are critical for Fam20C kinase activity . Most of the residues important for Fam20C activity are conserved in Fam20A , including the metal-binding Asp ( D478 in human Fam20C ) , the catalytic Asp ( D458 in human Fam20C ) , and the ion pair ( K285 and E311 in human Fam20C ) ( Figure 2C ) . However , a conserved Glu in Fam20C ( E306 in human ) , which coordinates the Mn2+ ion and the ion-pair Lys and is indispensable for Fam20C kinase activity , is replaced by a Gln ( Q258 in human ) in all Fam20A orthologs ( Figure 2C , D , Figure 2—figure supplement 1 and [Xiao et al . , 2013] ) . Because Gln in Fam20A lacks a negatively charged side chain , it may not functionally substitute the Glu in Fam20C . To assess the effect of this evolutionarily conserved amino acid substitution on the biochemical properties of Fam20A and Fam20C , we generated Fam20A Q258E and Fam20C E306Q mutants . Recombinant Fam20C E306Q exhibited greatly reduced in vitro kinase activity towards casein , ENAM ( 173–277 ) , as well as osteopontin ( OPN ) , an SCPP and model substrate of Fam20C , as compared to WT Fam20C ( Figure 2—figure supplement 2 ) . Accordingly , the intrinsic ATPase activity of Fam20C E306Q was abolished ( Figure 2E ) . Conversely , Fam20A Q258E was able to hydrolyze ATP and phosphorylate OPN , albeit less robustly than Fam20C ( Figure 2E , F ) . These data suggest that Fam20A does not have all the residues essential for catalytic activity and that Fam20A is not an intrinsically active kinase . In order to elucidate the function of Fam20A , we generated Fam20A-KO ALCs by means of CRISPR/Cas9 ( Figure 3—figure supplement 1A ) . Depletion of Fam20A had no effect on the expression level of Fam20C ( Figure 3—figure supplement 1C ) . Although Fam20A is likely catalytically inactive , we observed that V5-tagged AMBN phosphorylation was greatly diminished in Fam20A-KO ALCs ( Figure 3A ) . This effect was not solely restricted to enamel proteins because phosphorylation of V5-tagged OPN was also dramatically reduced in Fam20A-KO ALCs ( Figure 3B ) . Reintroduction of Flag-tagged Fam20A in Fam20A-KO ALCs restored the phosphorylation of V5-tagged AMBN and OPN ( Figure 3A , B ) . Notably , the phosphorylation of V5-tagged OPN was totally eliminated when the Fam20c gene was disrupted in either WT or Fam20A-KO cells ( Figure 3C and Figure 3—figure supplement 1B–C ) . Furthermore , ectopic expression of Fam20A in the human osteosarcoma cell line U2OS , which produces Fam20C , but not Fam20A , significantly increased phosphorylation of V5-tagged OPN , ENAM and AMBN , without altering Fam20C expression levels ( Figure 3D–F ) . In contrast , no phosphorylation of these substrates was detected in Fam20C-KO U2OS cells , even when Flag-tagged Fam20A was overexpressed ( Figure 3D–F and Figure 3—figure supplement 2 ) . Collectively , these results indicate that Fam20A functions to increase Fam20C substrate phosphorylation when Fam20C is present . 10 . 7554/eLife . 06120 . 008Figure 3 . Fam20A enhances Fam20C-dependent phosphorylation . ( A and B ) Effect of Fam20A KO on the phosphorylation of AMBN ( A ) or OPN ( B ) . C-terminal V5-tagged AMBN or OPN ( AMBN-V5 or OPN-V5 ) was expressed in WT or Fam20A-KO ALCs and analyzed as in Figure 1C . Flag-tagged Fam20A ( 20A-Flag ) was co-expressed in Fam20A-KO ALCs as indicated . ( C ) Phosphorylation of OPN-V5 in WT , Fam20A-KO , Fam20C-KO , or Fam20A and Fam20C double-KO ( A&C-DKO ) ALCs . The experiment was performed as in ( B ) . Two different exposures of autoradiography are shown . ( D–F ) Effect of Fam20A overexpression on the phosphorylation of OPN ( D ) , ENAM ( 173–277 ) ( E ) or AMBN ( F ) in WT or Fam20C-KO U2OS cells . C-terminal V5-tagged OPN , ENAM ( 173–277 ) or AMBN ( OPN-V5 , ENAM-V5 or AMBN-V5 ) were expressed in WT or Fam20C-KO U2OS cells and analyzed as in Figure 1C . Fam20A-Flag was stably expressed as indicated . Fam20A-Flag in the cell lysates and endogenous Fam20C in the conditioned medium are also shown . Fam20C-KO U2OS cells were generated by means of transcription activator-like effector nuclease ( TALEN; Figure 3—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 00810 . 7554/eLife . 06120 . 009Figure 3—figure supplement 1 . CRISPR-mediated Fam20A KO and Fam20A/Fam20C double KO ( DKO ) in ALCs . ( A ) The schematic of the sgRNA-targeting site in the mouse Fam20a gene and the disrupted allele in Fam20A-KO ALCs ( clone A21 ) . The first exon of the mouse Fam20a gene is shown as a box . White , 5′ UTR; Orange , the coding sequence . Targeting sites and PAMs are indicated with blue and red bars , respectively . Inserted nucleotides are highlighted in red . ( B ) Disrupted mouse Fam20c allele in Fam20A/Fam20C-DKO ALC ( clone AC33 ) . Fam20A-KO ALC clone A21 ( Figure 3—figure supplement 1A ) was used as the parental cell line . Fam20c gene was targeted using sgRNA m20c #3 ( Figure 1—figure supplement 1A ) . ( C ) Endogenous Fam20C production by WT ALC or the KO clones used in this study . Cells were seeded at the same density and cultured overnight in serum-free media . Proteins in the conditioned media were precipitated using trichloroacetic acid ( TCA ) and separated by SDS-PAGE . Secreted Fam20C was detected by western blot ( upper ) . Equal protein loading of each sample was shown by Ponceau S staining ( lower ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 00910 . 7554/eLife . 06120 . 010Figure 3—figure supplement 2 . Transcription activator-like effector nuclease ( TALEN ) -mediated Fam20C KO in U2OS cells . ( A ) The schematic of the TALEN-targeting sites in the human FAM20C gene . ( B ) Disrupted human FAM20C allele in Fam20C-KO U2OS cells . Deleted region is indicated with red dashes . ( C ) Endogenous Fam20C production by WT or Fam20C-KO U2OS . The experiment was performed as in Figure 3—figure supplement 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 010 To decipher the mechanism by which Fam20A regulated Fam20C-dependent phosphorylation , we carried out in vitro kinase assays with the combination of recombinant Fam20A and Fam20C . When both proteins were incubated in the reactions , phosphorylation of ENAM , AMTN , or an N-terminal peptide of AMBN fused to a SUMO tag ( SUMO-AMBN-N ) , were all greatly increased ( Figure 4A ) . Recombinant Fam20B did not increase Fam20C-catalyzed ENAM phosphorylation ( Figure 4B ) , nor did Fam20A or Fam20C have any effect on Fam20B-catalyzed xylose phosphorylation of the proteoglycan decorin ( Figure 4C ) . These results suggest that the activating effect of Fam20A is specific to Fam20C substrate phosphorylation . 10 . 7554/eLife . 06120 . 011Figure 4 . Fam20A increases Fam20C kinase activity in vitro . ( A ) Time-dependent 32P incorporation into enamel matrix proteins by Fam20C , Fam20A , or the combination of Fam20A and Fam20C ( molar ratio of 1:1 ) . SUMO-AMBN-N , a N-terminal peptide of AMBN fused to a SUMO tag . ( B ) Effect of Fam20B on ENAM phosphorylation by Fam20C . ( C ) Effect of Fam20A or Fam20C on Fam20B-catalyzed phosphorylation of proteoglycan decorin . Fam20B specifically phosphorylated glycosylated decorin , which migrated as an elongated smear due to the heterogeneity of attached glycosaminoglycans ( Wen et al . , 2014 ) . ( D ) Effect of SxE motif mutations on ENAM phosphorylation by the combination of Fam20A and Fam20C . The experiment was performed as in Figure 1B , except that equal amounts of Fam20A and Fam20C were added to the reaction . DA , inactive Fam20C D478A . ( E ) Effect of pre-incubation of Fam20A and Fam20C on ENAM phosphorylation . Fam20A ( 20 μg/ml ) and Fam20C ( 0 . 5 μg/ml ) were pre-incubated with ATP or nonhydrolyzable AMP-PCP ( 50 μM ) , on ice or at 30°C for 1 hr as indicated . Recombinant ENAM ( 173–277 ) and excessive ATP were then added to the final concentration of 250 μg/ml and 1 mM , respectively . [γ-32P]-ATP was added on the second step to trace ENAM phosphorylation . ( F ) Comparison of the abilities of Fam20A WT and D430A to restore OPN phosphorylation in Fam20A-KO ALCs . The experiment was performed as in Figure 3B . For transfection , 0 . 6–6 ng of 20A-Flag ( as indicated ) and 2 μg of OPN-V5 DNA were used for each 35 mm dish . ( G ) Effect of Fam20A Q258E ( QE ) on ENAM phosphorylation by Fam20C in vitro . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 01110 . 7554/eLife . 06120 . 012Figure 4—figure supplement 1 . Mutagenesis scanning for the phosphorylation site of SUMO-AMBN-N . ( A ) The schematic of human AMBN and the SUMO-tagged AMBN N-terminal peptide ( SUMO-AMBN-N ) . Ser and Thr residues are highlighted and numbered . A red bar indicates the only SxE motif in SUMO-AMBN-N . SP , signal peptide . ( B ) SUMO-AMBN-N or its mutants were incubated with Fam20C alone or the combination of Fam20A and Fam20C ( molar ratio of 1:1 ) in the presence of [γ-32P]-ATP as indicated . S43 within an SxE motif appeared to be the primary phosphorylation site , either in the presence or absence of Fam20A . DA , inactive Fam20C D478A . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 01210 . 7554/eLife . 06120 . 013Figure 4—figure supplement 2 . Effect of Fam20A D430A on ENAM phosphorylation by Fam20C in vitro . Variable amount of Fam20A D430A ( DA ) was used . The molar ratio between Fam20A and Fam20C in each lane is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 01310 . 7554/eLife . 06120 . 014Figure 4—figure supplement 3 . Fam20A Q258E does not phosphorylate Fam20C . Inactive Fam20C D478A was used as the substrate to avoid Fam20C autophosphorylation . OPN was used as a positive control . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 014 To determine whether the phosphorylation sites changed in the presence of Fam20A , we used the ENAM ( 173–277 ) S191A and S216L mutants as substrates for in vitro kinase assays . When both Fam20A and Fam20C were included in the reactions , ENAM ( 173–277 ) was readily phosphorylated . However , the S191A/S216L double mutant was not , even though it contains 21 additional Ser/Thr/Tyr residues ( Figure 4D ) . These results demonstrate that the phosphorylation remains specific to SxE motifs . Further , we replaced each of the Ser and Thr residues with Ala in SUMO-AMBN-N . In vitro kinase assays confirmed that the phosphorylation was specific for SxE in the presence of Fam20A ( Figure 4—figure supplement 1 ) . As anticipated , the kinase activity of Fam20C was indispensable ( Figure 4D and Figure 4—figure supplement 1 ) . These results point to a model where Fam20A functions as an activator of Fam20C kinase activity . To verify this model , we sought to rule out the possibility that Fam20A could be phosphorylated and activated by Fam20C . We pre-incubated Fam20C and excessive Fam20A with ATP on ice or at 30°C prior to the addition of ENAM ( 173–277 ) and [γ-32P]-ATP , and monitored ENAM phosphorylation . Pre-incubation of Fam20A and Fam20C with ATP at 30°C would allow Fam20C to phosphorylate Fam20A; however , this did not increase subsequent ENAM phosphorylation when compared to conditions where pre-incubation was carried out on ice ( Figure 4E , group 2 and 3 ) . Furthermore , ENAM phosphorylation remained constant when Fam20A and Fam20C were pre-incubated with ATP or AMP-PCP , a non-hydrolyzable ATP analogue ( Figure 4E , group 3 and 4 ) . These findings suggest that the increase of substrate phosphorylation in the presence of Fam20A and Fam20C does not require Fam20C-dependent phosphorylation of Fam20A . To further confirm that the putative catalytic activity of Fam20A was not required for the increase in Fam20C-dependent substrate phosphorylation , we mutated the putative metal binding Asp of Fam20A ( D430A ) . Although mutation of the corresponding residue in Fam20C eliminated its kinase activity ( Figure 1A ) , expression of Flag-tagged Fam20A D430A in Fam20A-KO ALCs increased Fam20C-dependent OPN phosphorylation similar to that of WT Fam20A ( Figure 4F ) . Recombinant Fam20A D430A also enhanced ENAM phosphorylation by Fam20C in vitro , although slightly less effectively than WT Fam20A ( Figure 4—figure supplement 2 ) . Moreover , Fam20A Q258E , which acquires kinase activity , did not phosphorylate Fam20C and was less potent than WT Fam20A to potentiate Fam20C-catalyzed ENAM phosphorylation ( Figure 4G and Figure 4—figure supplement 3 ) . Collectively , our results indicate that Fam20A increases the kinase activity of Fam20C in a catalytically independent manner . Because Fam20A enhanced the activity of Fam20C by a mechanism that did not require Fam20A kinase activity , we tested whether Fam20A could directly interact with Fam20C . We incubated recombinant Fam20A with Fam20C and performed gel filtration analysis . Fam20A and Fam20C co-eluted in nearly equal amounts in the higher molecular weight fractions as compared to Fam20A or Fam20C alone , indicating that Fam20A and Fam20C formed a complex of molar ratio of 1:1 ( Figure 5A ) . This interaction was specific between Fam20A and Fam20C , because Fam20B and Fam20C did not co-elute during gel filtration ( Figure 5B ) . Furthermore , V5-tagged Fam20A , but not V5-tagged Fam20B , could be co-immunoprecipitated with Flag-tagged Fam20C when stably co-expressed in U2OS cells ( Figure 5C ) . Thus , Fam20A and Fam20C form a complex in vitro and in the secretory pathway of cells . 10 . 7554/eLife . 06120 . 015Figure 5 . Fam20A and Fam20C form a complex that is catalytically more active . ( A ) Gel filtration analysis of Fam20A/Fam20C complex . Purified Fam20A and Fam20C were pre-incubated on ice and passed through Superdex 200 . Proteins from different fractions were separated by SDS-PAGE and visualized by Coomassie staining . Fam20A or Fam20C alone were analyzed as controls . The calibration standard is shown on top . ( B ) Gel filtration analysis of Fam20B and Fam20C . ( C ) Co-immunoprecipitation of Fam20A and Fam20C from cells . Fam20C-Flag , Fam20A-V5 , or Fam20B-V5 was stably expressed in U2OS cells as indicated . ( D ) Co-immunoprecipitation of Fam20A-Flag and Fam20A-V5 from U2OS cells . ( E ) Analytical ultracentrifugation sedimentation velocity analysis of Fam20A/Fam20C complex . The plot represents the molecular mass distribution c ( M ) vs the apparent molecular mass ( Da ) . The main peak has a calculated weight of 245 kDa . ( F ) Fam20C kinase reaction initial velocities vs concentration of ENAM ( 173–277 ) S191A . Fam20C ( 2 μg/ml ) and Fam20A ( 40 μg/ml ) were used in this experiment . Data points are represented as mean ± SD and fitted by non-linear regression of the Michaelis-Menten equation . ( G ) Immunoprecipitation of Fam20A-Flag or Fam20C-Flag from either cell lysate ( C ) or the conditioned medium ( M ) . U2OS cells stably expressing Fam20A-Flag , Fam20C-Flag or both Fam20C-Flag and Fam20A-V5 were used . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 01510 . 7554/eLife . 06120 . 016Figure 5—figure supplement 1 . Saturation of Fam20C kinase activity by Fam20A in vitro . Fam20C kinase reactions were performed as in Figure 5F using ENAM ( 173–277 ) S191A as the substrate . Increasing amount of Fam20A was added to the reaction . The molar ratio between Fam20A and Fam20C in each lane is indicated . Saturation of ENAM phosphorylation was observed when 10 to 15-fold excess of Fam20A were used . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 016 Fam20C forms homodimers as judged by its elution profile during gel filtration ( Figure 5A ) . Fam20A largely eluted as a monomer; however , the elution profile revealed more heterogeneity , with significant amounts of protein in the fractions consistent with a dimer ( Figure 5A ) . Therefore , we ascertained if Fam20A was able to dimerize in cells . Indeed , Flag-tagged Fam20A co-immunoprecipitated with V5-tagged Fam20A , but not V5-tagged Fam20B , when stably co-expressed in U2OS cells , suggesting that Fam20A forms homodimers in cells ( Figure 5D ) . Analysis of the Fam20A/Fam20C complex by analytical ultracentrifugation revealed a molecular mass of 245 kDa , which is consistent with a heterotetrameric complex composed of two Fam20A and two Fam20C subunits ( Figure 5E ) . To further understand the effect of Fam20A on Fam20C-catalyzed phosphorylation , we set out to determine the steady-state kinetic parameters of Fam20C in the presence or absence of Fam20A , using ENAM ( 173–277 ) S191A as the substrate . When Fam20A was added in excess to the in vitro kinase reaction to maximize complex formation ( Figure 5—figure supplement 1 ) , the kcat value of Fam20C for ENAM increased by 19-fold ( from 0 . 04 s−1 to 0 . 76 s−1 ) , and the Km for ENAM decreased by ∼2 . 5-fold ( from 14 . 6 μM to 5 . 6 μM ) ( Figure 5F ) . This demonstrates that the Fam20A/Fam20C complex catalyzed substrate phosphorylation much more efficiently than Fam20C alone . Formation of the Fam20A/Fam20C complex raises a possibility that Fam20A associates with the Golgi more tightly than Fam20C and retains Fam20C in the Golgi to phosphorylate more substrates . However , when expressed in mammalian cells , both Fam20A and Fam20C are localized in the Golgi and are secreted into the conditioned medium at comparable levels ( [Tagliabracci et al . , 2012] and Figure 5G ) . We did not observe a redistribution of Flag-tagged Fam20C to the intracellular compartment when co-expressed with V5-tagged Fam20A ( Figure 5G ) . These results suggest that Fam20A does not serve as a retention mechanism for the Fam20A/Fam20C complex . To determine if Fam20A and Fam20C form a complex in vivo , we used lactating mouse mammary gland as a source of endogenous proteins . Fam20C is the bona fide kinase that phosphorylates casein , one of the most abundant and highly phosphorylated proteins in milk ( Tagliabracci et al . , 2012 ) . Fam20C and Fam20A mRNAs in the lactating mammary gland were dramatically elevated as compared to mammary gland from virgin animals , whereas mRNA levels of Fam20B , and other Fj family kinases remained relatively unaffected ( Figure 6A ) . Notably , the relative increase in Fam20A and Fam20C mRNAs in lactating mammary gland were virtually identical ( Figure 6A ) . In fact , the expression patterns of Fam20A and Fam20C were largely consistent throughout the process of pregnancy and lactation ( Figure 6B ) . The mRNA levels of both Fam20A and Fam20C in the mammary gland gradually increased from 13 to 18 days of pregnancy and decreased during lactation and involution , with Fam20A peaking just before and Fam20C peaking just after the birth of the new litter ( Figure 6B ) . Consistent with their mRNA levels , Fam20A and Fam20C proteins could only be detected in the mammary glands of lactating , but not virgin mice ( Figure 6C ) . Coordinated expression of Fam20A and Fam20C is consistent with the idea that they function together in the mammary gland during pregnancy and lactation . Indeed , Fam20A could be co-immunoprecipitated with Fam20C in the Golgi/ER-enriched fraction of the lactating mammary gland ( Figure 6D ) . These results demonstrate that Fam20A and Fam20C can form a functional complex in the mammary gland of pregnant/lactating animals . 10 . 7554/eLife . 06120 . 017Figure 6 . Upregulation and complex formation of Fam20A and Fam20C in the lactating mammary gland . ( A ) mRNA levels of Fam20C family members in the mouse mammary gland . The mammary glands from two non-lactating ( Virgin ) and two 10-day lactating ( Lact ) mice were isolated and the mRNA levels of Fam20 family members were determined by quantitative ( q ) RT-PCR . Data are represented as mean ± SD . ( B ) mRNA levels of Fam20A and Fam20C in the mammary glands of pregnant/lactating female mice at different stages . E−13 , 16 or 18: embryonic day 13 , 16 or 18; Lact-1 , 3 or 7: 1 , 3 or 7 days after lactation; Inv-1 or 7: 1 or 7 days after involution . The gestation period of mouse is 18–21 days . mRNA levels of Fam20A and Fam20C were determined by qRT-PCR . Data are represented as mean ± SD . ( C ) Protein levels of Fam20A and Fam20C in the mouse mammary glands . Endogenous Fam20A or Fam20C were detected from whole mammary gland lysate or the Golgi/ER-enriched fractions by western blot . ( D ) Co-immunoprecipitation of endogenous Fam20A and Fam20C from Golgi/ER-enriched fractions of the mouse lactating mammary gland . Polyclonal rabbit anti-Fam20C antibody was used for immunoprecipitating endogenous Fam20C . Normal rabbit IgG was used as control for immunoprecipitation . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 017 Several deletions , truncations and missense mutations in Fam20A cause AI in humans . Because the Fam20A deletions and truncations are likely inactivating , we tested whether the missense Fam20A mutants , L173R , G331D and D403N , could increase Fam20C activity . We co-expressed Flag-tagged Fam20A ( WT or AI mutants ) and V5-tagged OPN in Fam20A-KO ALCs , and analyzed radiolabeled phosphate incorporation into V5-immunoprecipitates . Expression of WT Fam20A , but not the AI mutants , markedly increased OPN phosphorylation ( Figure 7A ) . Moreover , these Fam20A mutants were poorly secreted when ectopically expressed in ameloblast cells ( Figure 7B ) , consistent with our previous structural modeling analyses that these mutations likely interfere with the proper folding of Fam20A ( Xiao et al . , 2013 ) . 10 . 7554/eLife . 06120 . 018Figure 7 . Fam20A mutants found in AI patients fail to enhance Fam20C activity , and WT Fam20A is able to increase the activities of Fam20C mutants . ( A ) Effect of Fam20A mutations on OPN phosphorylation . The experiment was performed as in Figure 3B . OPN-V5 was expressed in WT or Fam20A-KO ALC . Fam20A-Flag WT , L173R , G331D or D403N was co-expressed as indicated . ( B ) Effect of AI mutations on Fam20A secretion . Fam20A-Flag WT , L173R , G331D or D403N was expressed in ALCs . Fam20A-Flag was immunoprecipitated from conditioned media or cell lysate using anti-Flag antibody and detected by western blot . ( C ) Effect of Fam20A overexpression on Fam20C mutant-catalyzed OPN phosphorylation . OPN-V5 and Fam20C-Flag ( WT or mutants as indicated ) were expressed in U2OS cells with or without Fam20A-Flag . Secreted OPN-V5 was immunoprecipitated from conditioned media and detected by western blot . OPN phosphorylation was indicated by its mobility shift . D478A is the kinase-dead Fam20C mutation . Other Fam20C mutations were identified from Raine syndrome ( R . S . ) patients . ( D ) The working model of Fam20A/Fam20C complex . Upper: in the tissues and cells where both Fam20A and Fam20C are expressed , Fam20C forms a complex with Fam20A and exhibits high activity , resulting in high levels of substrate ( S ) phosphorylation; Lower: when Fam20A is disrupted , Fam20C activity becomes low and substrate phosphorylation is ineffective , leading to diseases like AI . DOI: http://dx . doi . org/10 . 7554/eLife . 06120 . 018 When expressed in U2OS cells , most Fam20C mutants found in Raine syndrome patients could not efficiently phosphorylate OPN . To test whether Fam20A could stimulate kinase activities of Fam20C mutants , we expressed Fam20A with Flag-tagged Fam20C mutants and monitored V5-tagged OPN phosphorylation by its mobility during SDS-PAGE . As expected , WT and non-lethal Fam20C mutants ( T268M and P328S ) , but none of the other mutants induced a mobility shift in OPN ( Figure 7C ) , consistent with previous reports ( Tagliabracci et al . , 2012 , 2014 ) . However , when Fam20A was co-expressed , we observed enhancement of Fam20C kinase activity in every mutant with the exception of T268M ( Figure 7C ) . Because several of the Fam20C mutations are predicted to prevent proper folding/stability , ectopic expression of Fam20A likely promoted not only Fam20C kinase activity , but also Fam20C stability . These results suggest that Fam20A could help stabilize Fam20C mutations and that recapitulating this effect by pharmacological means may be of therapeutic benefit to patients harboring these Fam20C mutations . In this study , we have shown that Fam20A and Fam20C form a functional complex to phosphorylate secreted proteins within SxE motifs . In the Fam20A/Fam20C complex , Fam20C is catalytically active , whereas Fam20A functions as an allosteric activator to increase Fam20C kinase activity . Because the residual time of secreted proteins within the secretory pathway is limited , the formation of Fam20A/Fam20C complex , whose turnover number ( kcat ) is dramatically increased , may reflect an important mechanism to achieve high stoichiometry of secreted proteins phosphorylation . Our model suggests that loss-of-function mutations in Fam20A result in inefficient phosphorylation of enamel matrix proteins by Fam20C , which prevents proper enamel formation and leads to AI . The degree of phosphorylation of enamel matrix proteins required for proper enamel formation is unknown . However , patients with hypomorphic mutations in Fam20C , including T268M and P328S , survive infancy but developed dental defects ( Fradin et al . , 2011; Rafaelsen et al . , 2013 ) , indicating that optimal Fam20C activity is critical for enamel formation . Our proposed mechanism for activation of Fam20C by Fam20A is supported by the fact that both Fam20A and Fam20C deficiency result in impaired biomineralization in humans and mice . In contrast to Fam20C , which is ubiquitously expressed , Fam20A expression has only been detected in dental tissues , lactating mammary gland , parathyroid gland and kidney ( Vogel et al . , 2012; Wang et al . , 2014 ) . Further , Fam20A deficiency does not cause complete loss of Fam20C-dependent substrate phosphorylation . These are consistent with the fact that patients harboring Fam20A mutations display mild and localized biomineralization phenotypes , as compared to the more severe forms caused by Fam20C mutations . It is increasingly apparent that in addition to phosphotranfer , kinase domain-containing proteins can play non-catalytic roles as scaffolds or allosteric regulators ( Zeqiraj and van Aalten , 2010; Shaw et al . , 2014 ) . Approximately 10% of the human kinome encodes pseudokinases , which are predicted to lack essential catalytic residues and primarily exert non-catalytic functions ( Manning et al . , 2002; Zeqiraj and van Aalten , 2010; Shaw et al . , 2014 ) . Several pseudokinase-kinase pairs have been revealed , where a pseudokinase binds and potentiates an active kinase . For example , the pseudokinase STRAD , together with MO25 , forms a complex with LKB1 and activates LKB1 kinase activity ( Zeqiraj et al . , 2009a , 2009b ) . Similarly , the pseudokinase domain of HER3 can dimerize with and allosterically activate catalytically active EGFRs ( Zhang et al . , 2006; Jura et al . , 2009 ) . Furthermore , the pesudokinase-like inactivated BRAF , generated through binding to ATP-competitive inhibitors or mutation of a residue in the catalytic spine ( C-spine ) , can still allosterically activate other RAF family kinases via dimerization ( Hatzivassiliou et al . , 2010; Heidorn et al . , 2010; Poulikakos et al . , 2010; Hu et al . , 2011 , 2013 ) . Fam20A lacks a negatively charged residue that is critical for kinase activity and therefore is a pseudokinase . Nevertheless , it binds and increases Fam20C activity in a phosphorylation-independent manner . Discovery of the Fam20A-Fam20C pair extends the concept that the activity of a kinase can be allosterically regulated by dimerization with an inactive paralog , which further expands the pseudokinase-kinase mode of regulation to protein phosphorylation within the secretory pathway . Our result suggests that Fam20A binds ATP , which may stabilize a conformational state that is required for the allosteric function ( Figure 2B ) . This mechanism has been reported for other pseudokinases . For instance , the crystal structure of the LKB1-STRAD-MO25 complex revealed that upon binding to ATP and MO25 , STRAD was maintained in an active conformation and engaged LKB1 as its ‘pseudo-substrate’ ( Zeqiraj et al . , 2009a , 2009b ) . In agreement with the potential role of ATP in stabilizing Fam20A conformation , Fam20A Q258E , which is able to hydrolyze ATP and therefore will not be locked in an ATP-binding conformation , is less potent than WT Fam20A to activate Fam20C ( Figures 2E , 4G ) . Although dimerization-induced allostery has emerged as a mechanism to regulate protein kinase activity , the interfaces used for mediating kinase domain association appear to be highly diverse ( Lavoie et al . , 2014 ) . The architecture of the Fam20A/Fam20C complex , as well as the detailed mechanism by which interactions between Fam20A and Fam20C increase the kinase activity of Fam20C , are currently unclear . However , the ability of Fam20A to increase the activity of Fam20C mutants identified in patients may lead to a method to allosterically boost Fam20C activity and this may have therapeutic potential . A crystal structure of the Fam20A/Fam20C complex will be of paramount importance . As mentioned above , in addition to its critical role in biomineralization , Fam20C also appears to be important for maintaining the secreted phosphoproteome . However , regulation of Fam20C-dependent secreted protein phosphorylation is poorly understood . Our data demonstrate that Fam20C activity can be allosterically potentiated by Fam20A , and this increased catalytic activity is critical for enamel formation . Because the expression of Fam20A is tissue-specific , it is unlikely that Fam20A functions as a global regulator of Fam20C , arguing the existence of additional regulatory mechanisms for Fam20C activity . Indeed , we show that both Fam20A and Fam20C mRNA levels are dramatically increased in the lactating mammary gland , suggesting transcriptional regulation , which may involve pregnancy-related hormones . In conclusion , our work suggests a model by which Fam20A regulates secretory pathway phosphorylation and tooth enamel formation via activation of Fam20C . We hypothesize that in cells where Fam20A and Fam20C are co-expressed , Fam20A forms a complex with Fam20C and enhances its kinase activity , leading to high levels of substrate phosphorylation ( Figure 7D ) . This would be especially important in the biological processes where large amounts of substrates need to be phosphorylated , such as teeth development or milk production . Conversely , Fam20A deficiency results in reduced Fam20C activity and basal levels of substrate phosphorylation ( Figure 7D ) . Phenotypes , such as enamel defects , will present when the basal phosphorylation is insufficient to support the corresponding biological process . Thus , our findings have uncovered a novel mechanism to regulate secretory pathway phosphorylation and biomineralization . Human cDNAs for FAM20A , FAM20B , FAM20C , ENAM and AMELX were from Open Biosystems . Human AMBN and AMTN cDNAs were from DNASU . For transient expression in mammalian cell , full length AMELX , AMBN or AMTN was cloned into pCDNA4 with a C-terminal V5 tag , and Fam20A , Fam20B or Fam20C was cloned into pCCF with a C-terminal Flag tag . ENAM ( 173–277 ) was cloned into pCDNA4 with a N-terminal signal peptide and a C-terminal V5 tag . For stable expression in mammalian cells , Fam20A , Fam20B or Fam20C was cloned into pQCXIP or pQCXIH ( Clontech , Mountain View , CA ) with a C-terminal Flag or V5 tag as specified . For recombinant expression in E . coli , AMBN ( 27–48 ) was cloned into pSUMO , and AMELX ( 17–191 ) , AMTN ( 17–219 ) or ENAM ( 173–277 ) was cloned into pET28 with a N-terminal 6× His tag . For insect cell expression , human Fam20A ( 63–529 ) or Fam20C ( 93–584 ) was cloned into a modified pI-secSUMOstar vector ( LifeSensors , Malvern , PA ) , in which the original SUMO tag was replaced by a MBP tag and a tobacco etch virus ( TEV ) protease site . Site-directed mutagenesis was performed using QuikChange ( Agilent Technologies , Santa Clara , CA ) . All the constructs were verified by DNA sequencing . MBP-tagged Fam20A and Fam20C proteins were expressed in Hi5 insect cells and purified from the conditioned medium as described previously ( Xiao et al . , 2013 ) . The MBP tag was removed using gel filtration chromatography following TEV protease digestion . 6× His-tagged AMELX , AMTN , ENAM and SUMO-AMBN-N were expressed in E . coli BL21 ( DE3 ) and purified by Ni-NTA-agarose chromatography ( Qiagen , Venlo , Netherlands ) as described previously ( Tagliabracci et al . , 2012 ) . Flag-tagged Fam20C , Fam20B and decorin were expressed in 293T cells and purified from the conditioned medium as described previously ( Wen et al . , 2014 ) . Flag ( M2 ) antibody was purchased from Sigma ( St . Louis , MO ) . V5 antibodies from Millipore ( Billerica , MA; AB3792 ) and Life Technology ( Carlsbad , CA; R960-25 ) were used for immunoprecipitation and immunoblotting , respectively . Rabbit polyclonal anti-Fam20C antibody was raised against and affinity purified with recombinant Flag-tagged Fam20C produced in 293T cells . Rabbit polyclonal anti-Fam20A antibody was raised against and affinity purified with recombinant Fam20A produced in insect cells . In vitro kinase assays using various Fam20A substrates were performed in a buffer containing 50 mM Hepes ( pH 7 . 0 ) and 10 mM MnCl2 at 30°C . 100 μM [γ-32P]-ATP ( specific activity , 500–2000 cpm/pmol ) was used in the reaction unless specified . Reactions were terminated by the addition of Laemmli buffer and 20 mM EDTA and boiling for 5 min . Proteins were separated by SDS-PAGE and visualized by Coomassie staining . 32P incorporation into the substrates was detected by autoradiography . For kinetic studies using ENAM ( 173–277 ) S191A as the substrate , reactions were performed in 50 mM Hepes ( pH 7 . 0 ) , 60 mM NaCl , 10 mM MnCl2 , 0 . 5 mg/ml BSA , 100 μM [γ-32P]-ATP ( specific activity , 5000 cpm/pmol ) and various amount of ENAM ( 173–277 ) S191A . Reactions were initiated by the addition of recombinant Fam20C ( 2 μg/ml ) or a combination of Fam20C ( 2 μg/ml ) and Fam20A ( 40 μg/ml ) , and incubated for 40 s at 30°C . Proteins were separated by SDS-PAGE and visualized by Coomassie staining . The gel was dried on a piece of filter paper and the ENAM band was cut out for scintillation counting to quantify the incorporated 32P radioactivity . Data points were fitted by non-linear regression of the Michaelis-Menten equation . ATPase assay was performed in a buffer containing 10 mM Hepes ( pH 7 . 4 ) , 50 mM NaCl , 10 mM MnCl2 , 0 . 5 mg/ml BSA , 0 . 5 mM ATP and 50 μg/ml protein at 30°C . Phosphate release was quantified by using a malachite green-based colorimetric assay for inorganic phosphate ( Maehama et al . , 2000 ) . For gel filtration analysis of the Fam20A/Fam20C complex , purified Fam20A and Fam20C proteins ( 1 mg/ml each ) were incubated on ice for 2 hr and then passed through a Superdex 200 column ( GE Healthcare , Piscataway , NJ ) with a buffer containing 25 mM Tris–HCl ( pH 7 . 4 ) and 150 mM NaCl . Fam20A or Fam20C alone , or the combination of Fam20B/Fam20C was analyzed the same way as controls . For the thermal stability shift assays , proteins were diluted to a final concentration of 2 μM in a buffer containing 10 mM Hepes ( pH 7 . 4 ) , 100 mM NaCl and 10 mM MnCl2 . ATP ( 1 mM ) was added as indicated . SYPRO Orange dye ( 5000× stock , Molecular Probes , Eugene , Oregon , S6650 ) was added to a final concentration of 5× to trace protein denaturation . Thermal scanning ( 25–85°C at 1°C/min ) was performed using a CFX96 Touch Real-Time PCR Detection System ( Bio-Rad , Hercules , CA ) with the detection channel of 610–650 nm , and the melting curves were normalized to a 0–100 range . Tm corresponding to the midpoint for the protein unfolding transition was calculated by fitting the increasing phase of the melting curve by nonlinear least-squares regression using sigmoidal equations ( GraphPad Prism ) . Sedimentation velocity experiment was performed using a Beckman ProteomeLab XL-I analytical ultracentrifuge . Purified Fam20A/Fam20C complex ( 1 mg/ml ) in 10 mM Hepes , pH 7 . 5 , 100 mM NaCl was spun at 30 , 000 rpm for 5 hr , and the 280 nm absorbance data were recorded . Data analysis was performed using the SEDFIT software ( Schuck , 2000 ) . The ameloblast-like cells ( ALCs ) were kindly provided by Dr John Bartlett at the Forsyth Institute with the permission of Dr Toshihiro Sugiyama at Akita University , Japan . ALC , U2OS and 293T cells were cultured in Dulbecco's modified Eagle's medium ( DMEM; Life Technology ) supplemented with 10% FBS ( Life Technology ) and 100 μg/ml penicillin/streptomycin ( Life Technology ) at 37°C in a 5% CO2 incubator . Transfection was carried out by using FuGENE-6 ( Promega , Madison , WI ) following the manufacturers' instructions . For metabolic radiolabeling experiments , ALCs or U2OS cells were seeded at 5 × 105 cells per well in 6-well plate; 20 hr later , cells were transfected with 4 μg of plasmid encoding C-terminal V5-tagged AMBN , ENAM ( 173–277 ) or OPN . 20 ng , or the amounts specified in Figure 4D , of pCCF-Fam20A-Flag was co-transfected , when the effect of Fam20A ectopic expression was ascertained . 2 days after transfection , metabolic labeling was started by replacing the medium with phosphate-free DMEM containing 10% dialyzed FBS and 1 mCi/ml 32P orthophosphate ( PerkinElmer , Waltham , MA ) . After labeling for 8 hr , the conditioned medium was collected and the cell debris was removed by centrifugation . V5-tagged proteins were immunoprecipitated from the supernatant and analyzed for protein and 32P incorporation by immunoblotting and autoradiography , respectively . Disruption of Fam20a or Fam20c gene in ALCs was carried out by means of CRISPR/Cas9 ( Ran et al . , 2013 ) . The guide sequences targeting exon 1 of mouse Fam20a or Fam20c gene were designed using the online tool at crispr . mit . edu and cloned into pX330 ( Addgene , Cambridge , MA ) . To generate Fam20A-KO or Fam20C-KO clones , ALCs were co-transfected with pX330 constructs containing the targeting sequences and pEGFP-C1 ( Clontech ) at molar ratio of 4:1 . 1 day after transfection , cells were trypsinized and single-cell sorted into 96-well plate using BD FACSJazz cell sorter . To screen cell clones with disrupted Fam20a or Fam20c gene , genomic DNA from expanded single clones was isolated using Quick-gDNA MiniPrep kit ( Zymo Research , Irvine , CA ) . The genomic region flanking the targeting sequence was amplified by PCR and subjected to DNA sequencing . To generate Fam20A/Fam20C double KO cell line , the Fam20c gene was disrupted in Fam20A-KO clone A21 using targeting sequence m20c#3 ( see below ) . The CRISPR targeting sequences used in this study are as follows: m20c #3: 5′-GTGGCGCGTCGGTCCAGCTT-3′ ( for clone C346 and AC33 ) ; m20c #4: 5′-GGGCTCCCCGGAGGATCGCG-3′ ( for clone C421 ) ; m20a #3: 5′-CAGGCGGCGGGGCGCTCGCC-3′ ( for clone A21 ) . The following primers were used to screen for the disrupted alleles: mouse Fam20a , 5′-GGTCCCCAAGTTCAGGGAAG-3′ ( forward ) and 5′-CTGTGACGGCAGAGTGAAGT-3′ ( reverse ) ; mouse Fam20c , 5′-CATGAAGATGATACTGGTGCG-3′ ( forward ) and 5′-GTCGCTGTTCACATTAAACAG-3′ ( reverse ) . FAM20C-KO U2OS cell line was generated by means of transcription activator-like effector nuclease ( TALEN ) . A TALEN binding pair was designed to target the exon 1 of human FAM20C gene . The genomic recognition sequences for the left and right TALEN arms are GCGCCGGTTCCGCGTGCT ( NN-HD-NN-HD-HD-NN-NN-NG-NG-HD-HD-NN-HD-NN-NG-NN-HD-NG ) and GGTGGCCTGCGCGCTGC ( NN-HD-NI-NN-HD-NN-HD-NN-HD-NI-NN-NN-HD-HD-NI-HD-HD ) , respectively . TALEN vectors were assembled using the Golden Gate method ( Cermak et al . , 2011 ) . The cutting activity of the designed TALEN pair was measured by the SURVEYOR assay ( Transgenomics , Omaha , NE ) . To isolate Fam20C-KO clones , U2OS cells were transfected with the TALEN plasmids by electroporation and subjected to limiting dilution . Cell clones were expanded and genomic DNA was isolated using Quick-gDNA MiniPrep kit ( Zymo Research ) . Disruption of FAM20C gene was assessed by PCR using primers 5′-GGACCCACACGCCCG-3′ ( forward ) and 5′-GCAGGATGCGGAGCG-3′ ( reverse ) and DNA sequencing . Loss of Fam20C expression was confirmed by immunoblotting . Mammary glands were harvested from virgin or lactating C57BL/6 female mice . For generating the whole tissue lysate , mammary glands were homogenized in a buffer containing 50 mM Tris–HCl ( pH 8 . 0 ) , 150 mM NaCl , 0 . 5% NP-40 , 10% glycerol , 0 . 5 mM EDTA and protease inhibitors cocktail . To enrich the ER/Golgi , mammary glands were homogenized in HME buffer containing 10 mM Tris–HCl ( pH 7 . 4 ) , 250 mM sucrose , 1 mM EDTA and protease inhibitors cocktail . The homogenate was centrifuged at 1000×g to remove nuclei and unbroken cells . The supernatant was centrifuged at 3000×g to pellet heavy mitochondria . The 3000×g supernatant was then centrifuged at 17 , 000×g to pellet the ER , Golgi and light mitochondria . This pellet was resuspended in HME buffer as the ER/Golgi-enriched fraction . For co-immunoprecipitation of Fam20A and Fam20C , NaCl and Triton X-100 were added to the ER/Golgi fraction to the final concentrations of 100 mM and 1% , respectively , prior to the addition of rabbit anti-Fam20C antibody . Normal rabbit IgG was used as control . Immunoprecipitates were subjected to a 6% polyacrylamide gel to separate Fam20A from the heavy chain . Fam20A and Fam20C were detected by immunoblotting . For qRT-PCR analysis , total RNA was isolated from lactating or non-lactating mammary glands using the NucleoSpin RNA kit ( MACHEREY-NAGEL , Bethlehem , PA ) . A panel of total RNA isolated from mammary glands at various pregnant/lactating stages was purchased from Zyagen ( San Diego , CA ) . cDNA was synthesized using the iScript kit ( Bio-Rad ) . qRT-PCR analysis was performed using the Power SYBR Green PCR Master Mix ( Applied Biosystems , Waltham , MA ) on Applied Biosystems 7500 Real-Time PCR System with primers as follows: Fam20A , 5′-GGCATCATTGACATGGCCGTCTTT-3′ ( forward ) and 5′-TTCATCCTGGGAATGTCGTCCGAA-3′ ( reverse ) ; Fam20B , 5′-TTCCAAATGGCATGCGATGGTCTG-3′ ( forward ) and 5′-TAACTGTGGTCCGTAGCTTGCACT-3′ ( reverse ) ; Fam20C , 5′-TGAAGATGATACTGGTGCGCAGGT-3′ ( forward ) and 5′-CAACAGCAATGTGCAAAGCGCAAG-3′ ( reverse ) ; Fam198A , 5′-TGACTTTCTGCTTCAGGTCCACGA-3′ ( forward ) and 5′-ATGCTGAAGGTTGCCAGCATTGTC-3′ ( reverse ) ; Fam198B , 5′-AAGCAATGATAGCCATTCCTCTG-3′ ( forward ) and 5′-CCACTCAGGTCTGGGAACC-3′ ( reverse ) ; Fjx1 , 5′-CATGCCAGGCTGTTTCCTTTCCAA-3′ ( forward ) and 5′-TCGGATCCAATCTCCACACAAGCA-3′ ( reverse ) . The expression level of target genes was normalized to that of GAPDH .	Some proteins must be modified in order to work effectively . One common modification is to add a phosphate group to the protein , which is performed by enzymes called protein kinases . Although most of the protein kinases work on proteins inside the cell , it was discovered recently that a small group of kinases work within the ‘secretory pathway’ and modify proteins that are released ( or secreted ) out of cells . One such secretory pathway kinase—called Fam20C—phosphorylates a wide range of secreted proteins and helps to ensure the proper development of bones and teeth . Specifically , Fam20C and a closely related protein called Fam20A are important for forming enamel , the hardest substance in human body , which makes up the outer surface of teeth . However , the exact role of Fam20A is unknown . Cui et al . now show that Fam20A binds to Fam20C , and this increases the ability of Fam20C to phosphorylate the proteins that form the ‘matrix’ that guides the deposition of the enamel minerals . Furthermore , mutations in Fam20A lead to the inefficient phosphorylation of enamel matrix proteins by Fam20C , and prevent proper enamel formation . The results raise the possibility that similar mechanisms of secretory kinase activation may also be important in other biological processes where many secreted proteins need to be phosphorylated rapidly .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2015	A secretory kinase complex regulates extracellular protein phosphorylation
Bone morphogenetic protein ( BMP ) signaling has emerged as an important regulator of sensory neuron development . Using a three-generation forward genetic screen in mice we have identified Megf8 as a novel modifier of BMP4 signaling in trigeminal ganglion ( TG ) neurons . Loss of Megf8 disrupts axon guidance in the peripheral nervous system and leads to defects in development of the limb , heart , and left-right patterning , defects that resemble those observed in Bmp4 loss-of-function mice . Bmp4 is expressed in a pattern that defines the permissive field for the peripheral projections of TG axons and mice lacking BMP signaling in sensory neurons exhibit TG axon defects that resemble those observed in Megf8−/− embryos . Furthermore , TG axon growth is robustly inhibited by BMP4 and this inhibition is dependent on Megf8 . Thus , our data suggest that Megf8 is involved in mediating BMP4 signaling and guidance of developing TG axons . During development , neurons of the peripheral nervous system ( PNS ) must project axons over long distances to form connections with their appropriate peripheral targets . To accomplish this task , axons of developing neurons rely on guidance provided by a variety of extracellular cues expressed in the surrounding environment and distal targets . Indeed , as developing axons extend into the periphery , they encounter a plethora of target-derived signals that they must interpret properly in order to accomplish proper axonal guidance , maturation , survival , target innervation , and synapse formation . Bone morphogenetic protein ( BMP ) signaling is an important regulator of sensory neuron development . BMPs are members of the TGFβ superfamily of secreted growth factors and are required for a wide range of developmental functions including gastrulation , mesoderm formation , neural patterning , left-right asymmetry , skeletal and limb development , kidney formation , and heart development ( Zhao , 2003 ) . In addition to patterning , BMPs also function as axon guidance molecules . BMP7 in the spinal cord roof plate repels axons of developing commissural neurons thus orienting their growth toward the floor plate ( Augsburger et al . , 1999; Butler and Dodd , 2003 ) . Moreover , BMP4 is a target-derived cue that augments development of somatosensory neurons in the trigeminal ganglia ( TG ) and dorsal root ganglia ( DRG ) . BMP4 is expressed in the developing craniofacial region and signals retrogradely to coordinate differential gene expression and patterning along the dorsoventral axis of the trigeminal ganglion ( Hodge et al . , 2007 ) . Furthermore , BMP4 appears to have a trophic effect on TG and DRG neurons as altering the level of BMP4 expression in the skin leads to changes in the peripheral innervation and number of sensory neurons found in both the TG and DRG ( Guha et al . , 2004 ) . Thus , BMP4 signaling contributes to sensory neuron maturation and target innervation; however , the mechanisms by which BMP4 acts on developing sensory neurons to promote axonal growth remain to be elucidated . We performed a forward genetic screen in the mouse to identify novel regulators of PNS development ( Merte et al . , 2010 ) . One of the lines that emerged from this screen has a mutation in the gene Megf8; loss of Megf8 function leads to defasciculation of the ophthalmic branch of the trigeminal nerve as well as defects in development of the heart , limb , skeleton , and left-right asymmetry . Given these phenotypes , we hypothesized that Megf8 interacts with the BMP4 signaling pathway . Indeed , loss of BMP signaling in trigeminal neurons leads to ophthalmic nerve defects that resemble those observed in Megf8 loss-of-function mouse lines . Furthermore , TG axon growth is robustly inhibited by BMP4 , and this inhibition is dependent on Megf8 expression . Taken together , these results show that Megf8 is a novel mediator of BMP signaling and is required for guidance of developing TG axons by target-derived BMP4 . We performed ENU mutagenesis and a recessive three-generation forward genetic screen in the mouse to identify novel growth and guidance cues for developing PNS axons ( Merte et al . , 2010 ) . One of the mutant mouse lines , Line 687 , exhibits severe defasciculation of the ophthalmic branch of the trigeminal nerve . In Line 687 mutants , the two major branches of the ophthalmic nerve initially form properly as tightly bound fascicles; however , as the axons extend into the target field , they prematurely defasciculate within the face ( Figure 1A ) . We mapped the genetic lesion responsible for the phenotype in Line 687 to a 2 . 6 Mb region of Chromosome 7 between rs3715453 and D7JHMI24 that contains 76 open reading frames ( Figure 1B ) . Based on bioinformatic analysis of genes within this region , we sequenced the 5′UTRs and coding exons of nine genes: Tmsb10 , Dmrtc2 , Zfp574 , Zfp526 , Gsk3a , Erf , Megf8 , BC024561 , and 2310004L02Rik . This analysis revealed a single base pair substitution , 5324T>C , in the gene multiple EGF-like-domains 8 ( Megf8 ) ( Figure 1C ) . Megf8 is a very large putative transmembrane protein , and it is predicted to have a signal peptide , a CUB domain , plexin-semaphorin-integrin ( PSI ) domains , Kelch domains , EGF/EGF-like domains , and a C-terminal transmembrane domain . The mutation in Line 687 ( Megf8L1775P ) is predicted to encode a single L>P amino acid substitution ( L1775P ) in the fourth Kelch domain of Megf8 ( Figure 1D ) . 10 . 7554/eLife . 01160 . 003Figure 1 . Disruption of Megf8 causes defasciculation of the TG ophthalmic nerve . ( A ) Whole-mount neurofilament staining of E11 . 5 control and Line 687 mutant littermates , showing the trigeminal ganglia ( TG ) and its three main projections: the ophthalmic ( OP ) , maxillary ( MX ) , and mandibular ( MD ) branches . ( B ) Schematic diagram of the region of Chromosome seven found to contain the Line 687 mutation , the markers used to diagnose linkage , and the frequency of recombination events observed at these markers . ORF , open reading frame . ( C ) Sequence data highlighting the mutation ( Megf8 5324T>C ) observed in Line 687 mutant DNA compared with C57BL/6 wild-type DNA . ( D ) Schematic diagram of Megf8 . The Line 687 point mutation induces a single amino acid substitution L1775P in a Kelch domain of Megf8 . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 00310 . 7554/eLife . 01160 . 004Figure 1—figure supplement 1 . Complementation analysis of Megf8Trap and Megf8L1775P alleles . ( A ) Schematic diagram of gene trap allele ( Megf8Trap ) . German Gene Trap Consortium cells G037A09 have an intron 2–3 insertion of a secretory trap vector such that the start ATG and signal peptide of Megf8 is captured and fused to a CD2-neomycin fusion protein instead of the remaining endogenous locus . SD = splice donor , SA = splice acceptor . ( B ) Whole-mount neurofilament staining of E11 . 5 embryos from an intercross of Megf8Trap/+ and Megf8L1775P/+ heterozygotes . Resulting Megf8L1775P/Trap embryos have the trigeminal defasciculation phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 004 In order to confirm that the Line 687 TG phenotype results from the Megf8L1775P mutation , we performed complementation analysis using a Megf8 secretory gene-trap line obtained from the German Gene Trap Consortium ( clone G037A09 , Megf8Gt ( CD2-neo ) GGTC or Megf8Trap ) . In this line , a CD2-neomycin secretory trap vector was virally inserted between exons two and three of Megf8; this approach results in a null , or hypomorphic , allele of Megf8 in which a CD2-neomycin fusion protein is brought to the cell surface instead of the endogenous protein ( Figure 1—figure supplement 1 ) . Megf8L1775P/Trap embryos phenocopy the Megf8L1775P/L1775P mutants and show defasciculation of the TG ophthalmic nerve ( Figure 1—figure supplement 1 ) . These results confirm that the Line 687 TG phenotype results from the Megf8L1775P mutation and demonstrate that Megf8 is required for proper growth and guidance of the TG ophthalmic nerve . We next examined the expression of Megf8 by in situ hybridization ( ISH ) using wild-type embryos . Whole-mount ISH of E8 . 5–E10 . 5 embryos showed that Megf8 is widely expressed , with strong expression in the somites , limb buds , primordial gut , developing eye , and in the pharyngeal arches ( Figure 2A–C ) . Megf8 is also highly expressed in the developing nervous system . Sensory neurons of the DRG and TG show strong expression of Megf8 throughout embryogenesis and into the postnatal period ( Figure 2D , E , Figure 2—figure supplement 1 ) . Megf8 is also expressed in the CNS including the developing neuroepithelium ( Figure 2F ) , postnatal hippocampus , layer 4/5 of the cortex , and the olfactory bulb ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 01160 . 005Figure 2 . Megf8 is expressed widely during development . ( A–C ) Whole-mount in situ hybridization ( ISH ) for Megf8 at E8 . 5 , E9 . 5 , and E10 . 5 . ( D ) ISH for Megf8 on E11 . 5 transverse cryosection shows expression in the DRG . ( E ) ISH for Megf8 on E11 . 5 transverse cryosection shows expression in the TG . ( F ) ISH for Megf8 on E10 . 5 paraffin sagittal section shows expression in the developing neuroepithelium . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 00510 . 7554/eLife . 01160 . 006Figure 2—figure supplement 1 . Megf8 is expressed throughout the developing nervous system . ( A ) In situ hybridization on coronal cryosections from E11 . 5-P0 shows strong Megf8 expression at all time points in the DRG and TG . ( B–D ) In situ hybridization on coronal cryosections at P0 shows strong Megf8 expression in the hippocampus ( B ) , layer 4/5 of the cortex ( C ) , and in the olfactory bulb ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 006 To gain insight into Megf8’s function during development we generated a conditional knockout mouse line using the Cre/LoxP system . We targeted the final exon of Megf8 , which encodes the transmembrane domain , intracellular C-terminus , and 3′UTR , to generate Megf8Flox conditional mutant mice ( Figure 3—figure supplement 1 ) . We then crossed Megf8Flox mice with the transgenic CMV-Cre line , in which cre recombinase is expressed in all cells , including germ cells ( Schwenk et al . , 1995 ) , to generate a Megf8 null allele ( Megf8− ) . Megf8−/− embryos phenocopy Megf8L1775P/L1775P embryos and exhibit dramatic defasciculation and undergrowth of the ophthalmic branch of the trigeminal nerve ( Figures 3A and 4H ) . In both the Megf8 null and Megf8L1775P/L1775P lines the maxillary and mandibular branches of the TG are largely unaffected . The maxillary branch shows no defasciculation phenotype although it is slightly undergrown , while the mandibular branch shows no abnormality ( Figure 4H , legend ) . 10 . 7554/eLife . 01160 . 007Figure 3 . Megf8 is required for development of the trigeminal ganglia , limb , skeleton , heart , and left-right asymmetry . ( A ) Whole-mount neurofilament staining of E11 . 5 Megf8+/+ and Megf8−/− littermates . The Megf8−/− null mutant phenocopies the point mutant Megf8L1775P/L1775P . ( B ) Whole-mount images of Megf8+/+ and Megf8−/− hindlimbs at E13 . 5 ( top ) and Megf8+/+ and Megf8L1775P/L1775P forelimbs at E16 . 5 ( bottom ) . ( C ) Alcian blue and alizarin red staining of E16 . 5 embryonic ribs/sternum . Megf8L1775P/L1775P mutants have a split sternum and delayed ossification of the rib cage . ( D ) Whole-mount images of the heart of freshly fixed E10 . 5 embryos with dissected pericardial cavity . Megf8L1775P/L1775P have complete left-right inversion of heart looping . Heart is outlined with dotted lines . ( E ) Whole-mount images of E11 . 5 Megf8+/+ and Megf8−/− littermates , showing reversal of embryonic turning and exencephaly in the Megf8−/− . ( F ) Whole-mount images of Megf8+/+ and Megf8−/− littermates at E13 . 5 , showing severe edema in the Megf8−/− . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 00710 . 7554/eLife . 01160 . 008Figure 3—figure supplement 1 . Generation of a conditional knock-out mouse line ( Megf8Flox ) . ( A ) Schematic diagram of Megf8 gene targeting strategy . This strategy targeted the final exon of Megf8 , which is 3 . 1 kb and comprises the transmembrane domain , intracellular C-terminus , and 3’′ UTR . A targeting vector was designed with an 8 . 5 kb long arm , a single loxP site located 500 bp upstream of the targeted exon , a frt-Neo-frt-loxP cassette placed 500 bp downstream of the exon , a 2 . 1 kb short arm , and a DTA selection cassette . Recombination with the endogenous locus resulted in integration of the 5′ loxP and 3′ frt-Neo-frt-loxP sites . Following germline transmission , male carriers were then mated with FlpE females to excise the neomycin cassette and generate the conditional allele . ( B ) Western blot of E12 . 5 TG and DRG lysates from Megf8+/+ , Megf8+/− , and Megf8−/− littermates . The blot was probed with rabbit anti-Megf8 and shows a loss of Megf8 protein in Megf8−/− neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 00810 . 7554/eLife . 01160 . 009Figure 3—figure supplement 2 . Megf8L1775P/L1775P mutants show defects in limb and skeletal development . ( A ) Alcian blue staining on E16 . 5 limbs shows that Megf8L1775P/L1775P mutants have digit duplication as well as duplication of bones in the hand . ( B ) Alizarin red staining on E16 . 5 limbs shows that Megf8L1775P/Trap mutants have duplication of the bones of the autopodium . ( C ) Alcian blue and alizarin red staining of whole E16 . 5 embryo skeletons . Megf8L1775P/L1775P mutants have a wider and shorter ribcage , delayed ossification of the ribcage , and a split sternum . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 00910 . 7554/eLife . 01160 . 010Figure 3—figure supplement 3 . Heart development in Megf8L1775P/L1775P mutant embryos . Sections from control ( A–C ) and Megf8L1775P/L1775P mutant ( D–F ) embryos . ( A and E ) Hematoxylin and eosin ( H and E ) staining of E15 . 5 control and mutant sections . The arrowhead in the control E15 . 5 embryo section points to the atrial septum; the arrowhead in the mutant embryo section points to the vestige of the endocardial cushion , above which the atrial septum is missing and below which the atrioventricular valves are poorly formed or atretic . ( B and E ) β-galactosidase and nuclear fast red staining of E12 . 5 embryos , which also carried the Tie2Cre/R26R alleles for labeling endothelium and endocardium and derived mesenchymal cells ( Kisanuki et al . , 2001; Soriano , 1999 ) . Note the normal endocardial cushion ( EC ) mesenchyme ( arrows ) . ( C and F ) H and E staining of E10 . 5 control and mutant sections . Note the apparently normal atrioventricular cushion mesenchyme at E10 . 5 ( arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 01010 . 7554/eLife . 01160 . 011Figure 4 . Megf8 is required for development of the PNS . ( A and D ) Whole-mount neurofilament staining of E11 . 5 Megf8+/+ and Megf8−/− littermates showing the DRG spinal nerves , which are undergrown in the Megf8−/− ( arrow ) . ( B and E ) Whole-mount peripherin staining of forelimbs from E13 . 5 Megf8+/+ and Megf8−/− littermates . The radial and ulnar nerves are undergrown in the Megf8−/−embryo . Limbs are outlined with dotted lines . Scale bar ( B and E ) represents 500 μm . ( C and F ) Whole-mount neurofilament staining of E11 . 5 Megf8+/+ and Megf8−/− littermates . The vagus/glossopharyngeal nerves are defasciculated in the Megf8−/− ( arrow ) . ( G ) Whole-mount neurofilament staining of E11 . 5 Megf8Flox/Flox and Wnt1-Cre; Megf8Flox/Flox littermates . ( H ) Quantification of ophthalmic branch phenotype for Megf8−/− ( KO ) , Megf8L1775P/L1775P ( L1775P ) , and Wnt1-Cre; Megf8Flox/Flox ( cKO ) compared to Megf8+/+ ( WT ) . Left: the number of branches at the nasociliary branch point was significantly greater for KO , L1775P , and cKO embryos . Right: the ophthalmic branch was undergrown in KO , L1775P , and cKO embryos . Four to seven embryos were analyzed per genotype . Error bars represent mean ± s . e . m . p<0 . 05 , one-way ANOVA . The relative outgrowth was also measured for the maxillary and mandibular branches of the TG for Megf8−/− embryos ( not shown ) . The maxillary branch was slightly undergrown compared to Megf8+/+ ( relative outgrowth 0 . 9 , p<0 . 05 ) while the mandibular branch was unaffected ( relative outgrowth 0 . 99 , p=0 . 7 ) . To assess defasciculation in the maxillary branch , the relative maxillary area was calculated and no difference was observed between Megf8+/+ and Megf8−/− ( relative area 0 . 94 , p=0 . 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 01110 . 7554/eLife . 01160 . 012Figure 4—figure supplement 1 . Conditional deletion of Megf8 from DRG neurons does not disrupt formation of the radial/ulnar nerves . Whole-mount peripherin staining of E13 . 5 forelimbs from Wnt1-Cre;Megf8Flox/+ and Wnt1-Cre;Megf8Flox/Flox littermates . Limbs are outlined with dotted lines . Scale bar represents 500 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 012 Consistent with its widespread expression during embryogenesis , we observed that loss of Megf8 leads to severe developmental abnormalities in several organ systems . Both Megf8−/− and Megf8L1775P/L1775P embryos display polydactyly in both the forelimbs and hindlimbs; this pre-axial polydactyly is completely penetrant , and embryos show six to seven digits per limb as well as duplication of bones in the hand ( Figure 3B , Figure 3—figure supplement 2 ) . They also exhibit skeletal abnormalities including delayed ossification of the rib cage , a wider and shorter rib cage , and a split sternum ( Figure 3C , Figure 3—figure supplement 2 ) . Furthermore , loss of Megf8 leads to a disruption of left-right patterning . Approximately one third of Megf8−/− embryos demonstrate a complete left-right inversion of heart looping ( Figure 3D ) or embryonic turning ( Figure 3E ) . Taken together with prior results , these findings indicate that Megf8 is required for development of left-right asymmetry . In addition to its role in development of the limb , skeleton , and left-right asymmetry , Megf8 is also required for normal heart development . In addition to the inversion of heart looping observed in some Megf8−/− and Megf8L1775P/L1775P embryos , loss of Megf8 function leads to a complete , or nearly complete , absence of the mitral and tricuspid valves , swollen atria , transposition of the outflow tract , pulmonary stenosis , and atrial and ventricular septal defects ( Figure 3—figure supplement 3 ) . These phenotypes were seen in all embryos at E14 . 5 or later , and they lead to severe peripheral edema ( Figure 3F ) and embryonic lethality by age E16 . 5 . Analysis at earlier stages revealed apparently normal endocardial cushion mesenchymal cells ( Figure 3—figure supplement 3 ) , which implies that the terminal pathologies result from improper remodeling events after E12 . 5 . Conditional mutation of the Megf8Flox allele using the Mesp1-Cre line ( Saga et al . , 1999 ) , which is active in all mesoderm of the heart , did not cause any heart phenotypes ( data not shown ) , implying that early heart and embryo laterality defects might be responsible for the constellation of later heart defects . These findings are consistent with a second mouse line harboring a Megf8 point mutation ( Megf8C193R ) that was identified in an independent forward genetic screen and which was shown to have polydactyly and defects in left-right patterning and heart development ( Zhang et al . , 2009 ) . Because Megf8 is expressed strongly throughout the nervous system , we next investigated the function of Megf8 in the developing PNS , in addition to its role in the TG . In Megf8−/− embryos , spinal nerves are shorter at E11 . 5 than in wild type ( Figure 4A , D ) , and by E13 . 5 the radial and ulnar nerves have not fully extended into the limbs , appearing shorter and less branched ( Figure 4B , E ) . At E13 . 5 the limbs appear immature in Megf8−/− embryos relative to wild type littermates , which may contribute to the radial and ulnar nerve phenotypes . Megf8−/− embryos also display defects in the developing vagus and glossopharyngeal nerves . At E11 . 5 these nerves are slightly defasciculated and several wayward axons branch away from the main fiber bundle ( Figure 4C , F ) . Overall , these findings demonstrate that Megf8 is required for proper axonal extension in several areas of the PNS , including the TG , DRG , and vagus/glossopharyngeal nerves ( Table 1 ) . Given the broad expression of Megf8 in a wide range of tissues during development , we next asked whether these defects , and in particular the TG defect , are due to a cell autonomous function of Megf8 in sensory neurons . To address this question , we crossed the conditional Megf8Flox line with the transgenic Wnt1-Cre line ( Danielian et al . , 1998 ) in which cre recombinase is expressed in neural crest-derived cells , including sensory neurons . Wnt1-Cre-mediated conditional deletion of Megf8 leads to defasciculation of the ophthalmic branch of the trigeminal nerve , phenocopying Megf8−/− and Megf8L1775P/L1775P mutant embryos ( Figure 4G ) . Although all Wnt1-Cre;Megf8Flox/Flox embryos showed defasciculation of the ophthalmic branch , the severity of the phenotype was more variable than in Megf8−/− and Megf8L1775P/L1775P embryos ( Figure 4H ) , and in a few embryos the ophthalmic branch defasciculated completely around the eye and failed to project into the periphery ( data not shown ) . These findings suggest that Megf8 mediates guidance of developing TG axons in a cell autonomous manner . Wnt1-Cre;Megf8Flox/Flox embryos do not exhibit a defect in radial/ulnar nerve development , however , suggesting that Megf8’s role in spinal nerve axon guidance is non-cell autonomous ( Figure 4—figure supplement 1 ) . Of note , Wnt1-Cre;Megf8Flox/Flox embryos do not exhibit the limb immaturity seen in Megf8−/− embryos . Thus , one possibility is that the radial/ulnar nerve phenotype observed in Megf8−/− embryos is secondary to an overall delay in limb development . An alternative explanation is that Megf8 may be required in spinal motor neurons , and that a loss of motor neuron axon guidance in Megf8−/− embryos causes a subsequent disruption in DRG sensory axon guidance; this indirect effect of Megf8 on DRG sensory neuron axon guidance would not be evident in Wnt1-Cre;Megf8Flox/Flox embryos . 10 . 7554/eLife . 01160 . 013Table 1 . Summary of Megf8−/− phenotypes and implicated BMPsDOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 013PhenotypeMegf8-/-BMP implicatedTrigeminal nerve ( V1 ) defasciculation100% ( 10/10 ) n . d . Trigeminal patterningn . d . BMP4 ( Hodge et al . , 2007 ) Polydactyly100% ( 9/9 ) BMP4 , BMP7 ( Dudley et al . , 1995; Dunn et al . , 1997 ) Reversed heart looping33% ( 7/21 ) BMP4 ( Fujiwara et al . , 2002 ) Reversed embryonic turning ( E11 . 5 ) 33% ( 5/15 ) BMP4 ( Fujiwara et al . , 2002 ) Edema ( E13 . 5+ ) 100% ( 6/6 ) DRG spinal nerves undergrown ( E11 . 5 ) 100% ( 10/10 ) BMP4 ( Guha et al . , 2004 ) Radial/ulnar nerves undergrown ( E13 . 5 ) 100% ( 2/2 ) BMP4 ( Guha et al . , 2004 ) Vagus defasciculation90% ( 9/10 ) n . d . Exencephaly36% ( 16/45 ) Disrupted BMP4 expression around V1100% ( 4/4 ) Phenotypes , stages of observation and penetrances of Megf8−/− mutants are listed . V1 refers to the ophthalmic branch of the trigeminal nerve . Bmp loss-of-function lines known to display similar phenotypes are noted . Our analysis demonstrates that Megf8 is required for development of the limb , skeleton , heart , left-right asymmetry , and PNS . The spectrum of phenotypes seen in Megf8−/− embryos is strikingly similar to that observed when BMP signaling is disrupted ( Table 1 ) . BMPs are members of the TGF-β superfamily of extracellular ligands and have been implicated in a wide range of developmental functions . We noted that loss of BMP4 , and to a lesser extent BMP2 and BMP7 , results in similar defects as those observed in the Megf8 mutants . Bmp4 and Bmp7 loss-of-function mouse lines exhibit pre-axial polydactyly ( Dudley et al . , 1995; Luo et al . , 1995; Dunn et al . , 1997; Selever et al . , 2004 ) . Bmp4 null embryos also display disrupted left-right patterning and loss of normal rightward heart looping ( Fujiwara et al . , 2002 ) . Loss of BMP4 , or its receptor BMPR2 , results in a wide range of heart defects , including atrial septal defects , ventricular septal defects , atrioventricular septal defects , and abnormal positioning or septation of the outflow tract ( Jiao et al . , 2003; Liu et al . , 2004; Beppu et al . , 2009 ) ; several of these cardiac phenotypes are exacerbated by a concomitant loss of BMP2 or BMP7 ( Rivera-Feliciano and Tabin , 2006; Goldman et al . , 2009 ) . Furthermore , BMP4 signaling is required for dorsoventral patterning of the TG ( Hodge et al . , 2007 ) , and modifying BMP4 expression in the skin leads to changes in the peripheral innervation and survival of sensory neurons ( Guha et al . , 2004 ) . Taken together , our findings demonstrate that loss of either BMP4 or Megf8 leads to a similar spectrum of defects in the limb , left-right asymmetry , and heart , and that both BMP4 and Megf8 are required for normal development of the TG . Based on these functional similarities , we hypothesized that Megf8 interacts with , or is a component of , the BMP4 signaling pathway , and that this interaction is required for axon guidance and peripheral target innervation by TG sensory neurons . To address the hypothesis that Megf8 and BMP4 cooperate to regulate growth and guidance of TG axons , we began by assessing the spatial relationship between Bmp4 expression and developing TG axons . To do so , we utilized a Bmp4lacZ mouse line in which a lacZ reporter cassette was inserted into the Bmp4 locus ( Lawson et al . , 1999 ) . Using whole-mount β-galactosidase and neurofilament co-staining , we were able to examine the expression pattern of Bmp4 relative to developing TG axons . At E10 . 5 , Bmp4 is expressed strongly in the eye and also in the dorsal portion of the face above the eye ( Figure 5A ) . These two areas of BMP4 expression flank a narrow corridor through which the developing ophthalmic nerve projects; this expression pattern suggests that BMP4 may act as a repulsive guidance cue for developing TG axons . One day later , at E11 . 5 , the TG axons have extended into their peripheral targets and have bifurcated into frontal and nasociliary nerves . At this stage , Bmp4 expression is striking; as the ophthalmic branch extends beyond the eye , the nasociliary nerve branches in a region devoid of Bmp4 expression and then immediately projects nasally as a tightly bundled fascicle through an area of strong Bmp4 expression while the axons of the frontal branch extend rostrally . Remarkably , the pattern of Bmp4 expression is disrupted in Megf8−/− embryos . In the absence of Megf8 , Bmp4 expression is lost at the point of defasciculation of the ophthalmic branch into the frontal and nasociliary branches ( Figure 5B; Table 1 ) suggesting that axonal innervation of the target region is necessary for proper expression of Bmp4 and that the absence of this BMP signaling leads to defasciculation of the nasociliary branch of the ophthalmic nerve . An alternative explanation is that Megf8 is required in cells of the craniofacial target tissues to regulate Bmp4 expression . ISH shows that Megf8 is expressed in craniofacial tissues as well as the TG , although Megf8 expression is far more robust in neurons of the TG ( Figure 2 ) . Interestingly , in Megf8−/− embryos Bmp4 expression is disrupted exclusively in the area around the ophthalmic branch but is intact throughout the rest of the embryo ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 01160 . 014Figure 5 . BMP signaling is required for proper extension of the TG ophthalmic nerve . ( A ) Whole-mount β-galactosidase and neurofilament ( NFM ) co-staining on Bmp4lacZ/+ embryos at E10 . 5 and E11 . 5 , showing the relationship between Bmp4 expression and the developing TG nerve . ( B ) Whole-mount β-galactosidase and neurofilament co-staining on E11 . 5 embryos: Bmp4lacZ/+;Megf8+/+ ( left ) and Bmp4lacZ/+;Megf8−/− ( center ) littermates , as well as Bmp4lacZ/+;Wnt1-Cre/Megf8Flox/Flox ( right ) . Bmp4lacZ expression is lost at the location where defasciculation of the TG ophthalmic nerve occurs in Megf8−/− and Wnt1-Cre;Megf8Flox/Flox embryos . The top row shows the whole head with all three TG branches . The bottom row is an enlargement of the top row showing the ophthalmic branch . The area of perturbed Bmp4 expression is outlined . The disruption of Bmp4lacZ expression was fully penetrant and observed in all Megf8−/− ( n = 4 ) and Wnt1-Cre;Megf8Flox/Flox ( n = 3 ) embryos assessed . ( C ) Whole-mount neurofilament staining of E11 . 5 Bmpr2Flox/+ and Wnt1-Cre;Bmpr2Flox/− littermates . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 01410 . 7554/eLife . 01160 . 015Figure 5—figure supplement 1 . Bmp4 expression in Megf8−/− embryo compared to wild-type littermate . Whole-mount β-galactosidase and neurofilament co-staining on E11 . 5 embryos: Bmp4lacZ/+;Megf8+/+ ( left ) and Bmp4lacZ/+;Megf8−/− ( right ) littermates . Bmp4lacZ expression is lost at the location where defasciculation of the TG ophthalmic nerve occurs in Megf8−/− embryos but is intact throughout the rest of the embryo . ( A ) Presentation of the whole embryo . ( B ) DRG and spinal nerves . ( C ) Forelimbs and hindlimbs . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 01510 . 7554/eLife . 01160 . 016Figure 5—figure supplement 2 . Bmpr2 and Megf8 are expressed throughout the developing TG . In situ hybridization on serial transverse cryosections at E12 . 5 shows strong expression of Bmpr2 ( A–C ) and Megf8 ( D–F ) throughout the developing TG . Representative images are shown for ophthalmic ( A and D ) , maxillary ( B and E ) , and mandibular ( C and F ) lobes of the TG . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 016 To directly test the idea that BMP4 is required for the growth and guidance of TG axons , we next sought to assess development of the trigeminal nerve in the absence of BMP4 signaling . Because Bmp4 null embryos die too early in development for our analysis , we used mice lacking the BMP receptor , BMPR2 , in cells of the neural crest lineage . We utilized mice harboring one mutant BMP receptor conditional allele ( Bmpr2Flox ) and one null allele ( Bmpr2− ) ( Beppu et al . , 2005 ) crossed to Wnt1-Cre mice to eliminate the receptor from TG sensory neurons . BMPR2 , one of several BMP receptors , is preferentially activated by BMP ligands and not by other TGF-β superfamily members ( Miyazono et al . , 2010 ) . Bmpr2 is robustly expressed in all three lobes of the developing TG , as is Megf8 ( Figure 5—figure supplement 2 ) . Interestingly , the ophthalmic branch of the trigeminal nerve in E11 . 5 Wnt1-Cre;Bmpr2Flox/− embryos is stunted and branches prematurely in a manner reminiscent of the phenotype seen in Megf8 loss-of-function lines ( Figure 5C ) . Furthermore , the maxillary and mandibular branches of the trigeminal nerve are unaffected in the Wnt1-Cre;Bmpr2Flox/− embryos , as in Megf8 null lines . These results demonstrate that BMP signaling is required specifically for development of the ophthalmic branch of the trigeminal nerve , and they suggest that both Megf8 and BMPR2 are necessary to mediate this signaling . We next asked whether BMP4 and Megf8 functionally interact to mediate growth and guidance of TG axons . To test this idea , we used an in vitro TG explant assay in which TG explants were cultured in a collagen gel and exposed to increasing amounts of bath applied BMP4 . Remarkably , BMP4 robustly inhibits the outgrowth of wild-type TG axons , and maximal doses inhibit axon outgrowth by more than 50% ( Figure 6A , C ) . We then asked whether loss of Megf8 would alter the response of TG axons to BMP4 by co-culturing explants from Megf8−/− embryos alongside littermate controls . Megf8−/− explants exhibit significantly greater axon outgrowth in the presence of BMP4 , but not under control culture conditions ( Figure 6B , D ) . These results demonstrate that Megf8−/− axons are less sensitive to the inhibitory effects of BMP4 and that the response of TG neurons to BMP4 is partially dependent on Megf8 expression . Together with our in vivo findings , these results support a model in which Megf8 functions within TG sensory axons , allowing them to respond to intermediate target-derived BMP4 and thus guide these axons to their appropriate peripheral target regions in the face . 10 . 7554/eLife . 01160 . 017Figure 6 . Megf8 mediates the inhibition of TG axon growth by BMP4 . ( A ) E12 . 5 wild-type TG explants cultured in 0–100 ng/ml BMP4 . ( B ) E12 . 5 TG explants from Megf8+/− and Megf8−/− littermates cultured side by side in 0–50 ng/ml BMP4 . ( C ) Quantification of ( A ) : axonal outgrowth of wild-type TG explants in the presence of 0–100 ng/ml BMP4 . Increasing doses of BMP4 caused robust inhibition of axon outgrowth ( p<0 . 05 , Student’s t test ) . Three to four explants were quantified for each concentration of BMP4 . Error bars represent mean ± SEM . ( D ) Quantification of ( B ) : relative axon outgrowth of TG explants from control ( Megf8+/+ or Megf8+/- ) and Megf8-/- littermates in the presence of 0–50 ng/ml BMP4 . BMP4 inhibition of axon outgrowth is partially lost in Megf8−/− explants ( p<0 . 05 , Student’s t test ) . The experiment was repeated three times , using three to four explants per BMP4 concentration in each experiment; each bar represents results from at least 10 explants . Error bars represent mean ± SEM . Scale bar represents 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01160 . 017 Analysis of Megf8L1775P/L1775P and Megf8−/− embryos demonstrates that Megf8 functions broadly during embryogenesis and is required for the development of several organ systems . Loss of Megf8 function leads to pre-axial polydactyly , skeletal defects , disruption of left-right patterning , and severe heart defects , all of which together result in embryonic lethality ( present study , and Zhang et al . , 2009 ) . Interestingly , mutations in Megf8 were recently identified in a small subset of children with Carpenter syndrome , which is an autosomal-recessive multiple-congenital-malformation disorder . Mutations were identified in several domains of Megf8 , including the Kelch domain adjacent to the Kelch domain harboring the L1775P point mutation , and resulted in thoracic skeletal defects , limb defects such as pre-axial polydactyly , congenital heart defects , and profound lateralization defects including transposition of the great arteries , dextrocardia , and complete situs inversus ( Twigg et al . , 2012 ) . These phenotypes mirror those seen in our Megf8 loss-of-function lines and reinforce the vital role for Megf8 throughout embryogenesis . Megf8 is also required for development of the peripheral nervous system , including the growth and guidance of the TG , spinal , and vagus/glossopharyngeal nerves . Megf8−/− embryos exhibit profound defasciculation of the ophthalmic branch of the TG , undergrown and underbranched spinal nerves , and defasciculation of the vagus/glossopharyngeal nerves . Megf8 is strongly expressed in sensory neurons throughout embryogenesis , and conditional deletion of Megf8 in neural crest-derived cells , including sensory neurons ( Wnt1-Cre;Megf8Flox/Flox ) , leads to a defect in the TG ophthalmic nerve that phenocopies the Megf8 null and Megf8L1775P/L1775P lines . These results strongly suggest that Megf8 functions cell autonomously in TG neurons to regulate axon guidance and innervation of peripheral targets . The phenotypes observed in Megf8L1775P and Megf8− mouse lines show a striking resemblance to those observed in mice lacking BMP4 signaling ( Table 1 ) . Like the Megf8−/− and Megf8L1775P/L1775P embryos , loss of BMP4 function leads to pre-axial polydactyly ( Dunn et al . , 1997 ) , loss of left-right patterning including reversed heart looping ( Fujiwara et al . , 2002 ) , and severe heart defects ( Jiao et al . , 2003; Liu et al . , 2004 ) . BMP4 signaling has recently been implicated in PNS development , where it is required by neurons in the TG for specifying positional identity ( Hodge et al . , 2007 ) and in both the TG and DRG for peripheral target innervation and survival ( Guha et al . , 2004 ) . Given these phenotypic similarities , we hypothesized that Megf8 is a component of the BMP4 signaling pathway and specifically regulates the response of TG neurons to target-derived BMP4 . We assessed Bmp4 expression during E10 . 5–E11 . 5 , which is the period when TG axons extend into their peripheral target fields and found that Bmp4 expression appears to define the permissible field for ophthalmic nerve target innervation suggesting that BMP4 may function as a repulsive guidance cue . Consistent with these findings , conditional deletion of the BMP receptor , Bmpr2 , in sensory neurons using Wnt1-Cre;Bmpr2Flox/− mice results in defasciculation of the ophthalmic nerve and phenocopies the Megf8 mutant lines . Thus , axons of TG ophthalmic sensory neurons require both Megf8 and BMP signaling to correctly navigate towards their target field . Remarkably , in each of the Megf8 and Bmpr2 loss-of-function lines , the TG maxillary and mandibular branches are largely unaffected despite widespread expression of Megf8 and BMPR2 within the TG ganglion , which demonstrates that Megf8 and BMPR2 function selectively in neurons of the TG ophthalmic branch to mediate axon guidance . We suggest that Megf8 is a mediator or a key component of the BMPR2 signaling pathways that control growth and guidance of ophthalmic nerve axons . The expression pattern of Bmp4 relative to the developing ophthalmic nerve suggests that BMP4 acts as an inhibitory cue to prevent developing TG axons from innervating inappropriate targets . This hypothesized role for BMP4 is supported by our in vitro findings , in which BMP4 treatment of TG explant cultures resulted in robust inhibition of axon outgrowth . This inhibition was partially lost in Megf8−/− explants , indicating that TG neurons require Megf8 expression to mediate their maximal response to the inhibitory effects of BMP4 . These findings suggest that Megf8 is necessary for BMP4 signaling in sensory neurons and that this signaling provides an important inhibitory cue to guide developing TG axons . In addition , our finding that Bmp4 expression is disrupted in Megf8−/− embryos at the site of defasciculation of the TG ophthalmic branch suggests a possible non-cell autonomous role for Megf8 in regulating Bmp4 expression , wherein BMP4 expression is dependent on signaling from incoming TG axons , which is disrupted by defasciculation in the Megf8−/− embryo . Alternatively , BMP4 may be directly regulated by Megf8 expressed in cells of the craniofacial region . Given that Bmp4 expression is preserved in all other areas of the Megf8−/− embryos as well as our in vitro findings that show Megf8 is required in TG neurons to mediate the inhibitory effects of exogenous BMP4 , we think it is unlikely that Megf8 predominantly regulates BMP4 signaling through modifying BMP4 expression . Future studies will address the Megf8 mechanism of action in developing neurons and , in particular , whether it is a mediator of expression or axonal localization of BMPR2 , BMP4 binding to BMPR2 , or BMP4–BMPR2 signaling in axons . Taken together , our findings support a model in which Megf8 is a mediator of BMP4 signaling in TG axons . Megf8 is expressed in TG sensory neurons where it acts cell autonomously to promote signaling in response to target-derived BMP4 . Megf8-mediated BMP4 signaling , which is inhibitory for TG axons , likely enables these axons to remain fasciculated en route to their appropriate targets in the face . Disruption of this signaling , either through loss of Megf8 or other components of the BMP4 pathway , leads to a corresponding loss of inhibition and premature defasciculation of the nerve . Thus , BMP4 signaling is required for TG axon guidance and Megf8 is a novel component of BMP4 signaling in TG sensory neurons . Given the numerous phenotypic similarities between Megf8 and Bmp loss-of-function mouse lines , including defects in the limb , heart , and left-right patterning , we propose that Megf8 mediates signaling by BMP4 and , perhaps , other TGFβ family members throughout the embryo during development . Isolated embryos were fixed overnight in 4% paraformaldehyde/PBS at 4°C . All subsequent washes and incubations were performed at room temperature on an orbital shaker . After fixation , embryos were washed in PBS ( three 10 min washes ) and dehydrated as follows: 50% methanol/PBS ( 1 hr ) , 80% methanol/PBS ( 2 hr ) , and 100% methanol ( overnight ) . Endogenous peroxidase activity was quenched by incubating the embryos overnight in a solution of 3% hydrogen peroxide , 70% methanol , and 20% DMSO . The embryos were next washed in five 45 min incubations in TNT ( 10 mM Tris-base , 154 mM NaCl , 0 . 1% Triton X-100 ) and then incubated for 3 days in primary antibody solution , which consisted of 2H3 anti-neurofilament ascites antibody ( 3 . 8 mg/ml stock , Developmental Studies Hybridoma Bank , University of Iowa , Iowa City , IA ) diluted 1:5000 in TNT with 0 . 02% sodium azide , 4% milk , 5% DMSO , and 2% sheep serum ( Millipore , Billerica , MA ) . Following primary antibody , embryos were washed in TNT ( five time , 45 min each ) , and then incubated for 2 days in secondary sheep anti-mouse IgG HRP-conjugated antibody ( Jackson ImmunoResearch Laboratories , West Grove , PA ) diluted 1:250 in TNT with 0 . 02% sodium azide , 4% BSA , 5% DMSO , and 2% sheep serum . The embryos were washed in TBS for four 45 min incubations followed by one overnight incubation . The following morning , the horseradish peroxidase reactions were developed with diaminobenzidine , washed in five incubations of TBS for 5 min each , and dehydrated through methanol as done previously ( 50% , 80% , 100% methanol ) . To visualize the staining , embryos were cleared in 1:2 benzyl alcohol/benzyl benzoate . In order to assess the defasciculation and outgrowth of the three trigeminal nerve branches , images of representative embryos stained with the whole-mount neurofilament assay were analyzed using ImageJ software ( Abramoff et al . , 2004 ) . The ophthalmic branch defasciculation was assessed by counting the number of branches that emerged from the branch point of the nasociliary nerve distal to the eye . The outgrowth of all three trigeminal nerve branches ( ophthalmic , maxillary , and mandibular ) was quantified by measuring the distance the branch travelled into the peripheral tissue after exiting the TG . This distance was then divided by the outgrowth measured in a littermate control in order to calculate the relative outgrowth of the nerve . In order to assess for possible defasciculation in the maxillary nerve , the area of the maxillary branch was measured beginning at the point of exit from the TG and extending to its innervation of peripheral tissues . This area was then divided by the area measured in a littermate control to calculate the relative area of the maxillary branch . Digoxigenin ( DIG ) -labeled cRNA probes were used for in situ hybridization . To generate probes directed against Megf8 , 1 . 1 kb and 2 . 5 kb ApaI fragments from MGC clone 5369271 ( Thermo Scientific , Pittsburgh , PA ) were cloned into pBK-CMV ( Agilent , Santa Clara , CA ) . A probe for Bmpr2 was amplified using gene-specific PCR primers ( Allen Institute for Brain Science , Seattle , WA ) and cDNA template prepared from P4 mouse brain . The resulting fragment was cloned into pCRII-TOPO ( Life Technologies , Carlsbad , CA ) . Embryos were fixed in 4% paraformaldehyde/PBS overnight at 4°C . For whole-mount preparations , embryos were then dehydrated into methanol and in situ hybridization was performed as described in Parr et al . ( 1993 ) with modifications ( Knecht et al . , 1995 ) . For sectioned tissue , after fixation embryos were washed in PBS , cryoprotected in 30% sucrose , embedded and frozen in Tissue-Tek OCT ( Sakura Finetek USA , Inc . , Torrance , CA ) , and serially sectioned at 20 μm . In situ hybridization was then performed on sectioned tissue as described previously ( Schaeren-Wiemers and Gerfin-Moser , 1993 ) . To assess the developing neuroepithelium , in situ hybridization was performed on formalin-fixed , paraffin-embedded tissue . Cartilage and bone were stained with alcian blue 8GX ( Sigma-Aldrich , St . Louis , MO ) and alizarin red S ( Sigma-Aldrich ) as previously described ( McLeod , 1980 ) . E13 . 5 embryos were eviscerated and fixed in 4% paraformaldehyde/PBS overnight at 4°C . The forelimbs and hindlimbs were then separated and immunostained with rabbit anti-peripherin ( Millipore , Billerica , MA , 1:2000 ) as described in Wickramasinghe et al . ( 2008 ) . Isolated embryos were fixed in 4% paraformaldehyde/PBS at 4°C for 30 min . After fixation , the embryos were washed three times with X-gal rinse solution ( 100 mM sodium phosphate pH7 . 3 , 2 mM MgCl2 , 0 . 01% sodium deoxycholate , and 0 . 02% NP40 ) and then stained overnight at room temperature in X-gal staining solution ( 5 mM potassium ferricyanide , 5 mM potassium ferrocyanide , and 1 mg/ml X-gal in X-gal rinse solution ) . They were then post-fixed in 4% paraformaldehyde/PBS at 4°C overnight to improve the stability of the β-galactosidase stain . Following β-galactosidase staining and post-fixation , the embryos were immunostained according to the whole-mount neurofilament protocol outlined above . Trigeminal ganglia ( TG ) were dissected from E12 . 5 embryos . The dorsal half of each TG was isolated in order to select for neurons that give rise to the ophthalmic branch . Explants were then cultured in collagen droplets in Neurobasal media ( Life Technologies , Carlsbad , CA ) supplemented with B27 , 20 ng/ml NGF , and 0–100 ng/ml BMP4 ( R&D Systems , Minneapolis , MN ) . After 40 hr in culture , the explants were fixed in 4% paraformaldehyde overnight at 4°C . They were then immunostained for neurofilament as follows . After fixation , explants were washed in PBS ( three 15 min washes ) , permeabilized in 0 . 25% Triton X-100/PBS for 1 hr at room temperature , incubated overnight at room temperature with 2H3 anti-neurofilament ascites antibody ( Developmental Studies Hybridoma Bank , University of Iowa , Iowa City , IA ) diluted 1:5000 in blocking buffer ( 5% sheep serum , 5% BSA , and 0 . 25% Triton X-100 in PBS ) , washed in 0 . 25% Triton X-100/PBS ( five 15 min washes ) , incubated overnight at room temperature with secondary sheep anti-mouse IgG HRP-conjugated antibody ( Jackson ImmunoResearch Laboratories , West Grove , PA ) diluted 1:250 in blocking buffer , washed in 0 . 25% Triton X-100/PBS ( five 15 min washes ) , and finally developed with diaminobenzidine to visualize staining . Images of the explants were quantified using ImageJ software ( Abramoff et al . , 2004 ) . In order to determine the axon outgrowth , we quantified the explant’s total area and then subtracted the small portion of this area that comprised the cell bodies; the remaining area represented the total axon outgrowth for a given explant . The relative axon growth was calculated by dividing an explant’s total axon outgrowth by the average outgrowth for control explants ( control genotype and 0 ng/ml BMP4 treatment ) . The Bmp4lacZ ( Lawson et al . , 1999 ) , BMP receptor ( Bmpr2 ) conditional knockout ( Beppu et al . , 2005 ) , Wnt1-Cre ( Danielian et al . , 1998 ) , and Mesp1-Cre ( Saga et al . , 1999 ) mouse lines were generated and genotyped as previously described . The Line 687 point mutant ( Megf8L1775P ) encodes a restriction enzyme-sensitive polymorphism and was genotyped by amplification of an approximately 600 bp fragment using the primers 5′-GGCAAGGGATGTGGTCATTAAG and 5′-CTTTTCTCCAAACCAGGCATGGAAA , followed by SacI digestion . The Megf8Trap line was genotyped by PCR using the primers 5′-GACCACCAGCCTGAGTGCAAAA and 5′-TCGAGCACAGCTGCGCAAGGAA . The Megf8Flox allele was genotyped using the primers 5′-AAGCCTGGATGCAAGGGCAGAT and 5′-CATAGGGCCACCATCAGCTCAT . The Megf8− allele was genotyped using the primers 5′-AAGCCTGGATGCAAGGGCAGAT and 5′-CATAGGGCCACCATCAGCTCAT .	The peripheral nervous system relays information between the brain and spinal cord ( the central nervous system ) and the rest of the body . During development , neurons of the peripheral nervous system must extend processes ( axons ) long distances to reach the cells that they will eventually form connections with . Signaling molecules tell neuronal processes which direction to move in , and also tell them when they have reached their intended destination . One group of molecules involved in the extension and guidance of neuronal processes are growth factors known as bone morphogenetic proteins ( BMPs ) . These proteins contribute to a range of developmental processes , including the formation of the limbs and the skeleton , as well as various organs . They also help to establish the correct left-right patterning of the embryo , and direct the migration of sensory neurons . Now , Engelhard et al . have used a genetic screen to identify additional signaling molecules involved in the development of the peripheral nervous system . They screened mice with a range of mutations , and found that animals with a mutant form of the gene that codes for a protein called MEGF8 closely resembled mice that lacked a member of the BMP family , BMP4 . These mutants showed abnormal development of the skeleton and heart , and had six or seven digits on each limb ( polydactyly ) . Given the similarities between mice that lacked the gene for BMP4 and those that lacked the gene for MEGF8 , Engelhard et al . explored these parallels further , and the results of a series of experiments were consistent with the two proteins being part of the same signaling cascade . In addition to identifying a novel signaling molecule that is involved in the formation of the peripheral nervous system , Engelhard et al . have provided new insights into the mechanisms by which one of the best known developmental signaling cascades is regulated .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2013	MEGF8 is a modifier of BMP signaling in trigeminal sensory neurons
Many organisms , including species from all kingdoms of life , can survive desiccation by entering a state with no detectable metabolism . To survive , C . elegans dauer larvae and stationary phase S . cerevisiae require elevated amounts of the disaccharide trehalose . We found that dauer larvae and stationary phase yeast switched into a gluconeogenic mode in which metabolism was reoriented toward production of sugars from non-carbohydrate sources . This mode depended on full activity of the glyoxylate shunt ( GS ) , which enables synthesis of trehalose from acetate . The GS was especially critical during preparation of worms for harsh desiccation ( preconditioning ) and during the entry of yeast into stationary phase . Loss of the GS dramatically decreased desiccation tolerance in both organisms . Our results reveal a novel physiological role for the GS and elucidate a conserved metabolic rewiring that confers desiccation tolerance on organisms as diverse as worm and yeast . Terrestrial organisms regularly encounter severe drought . For species with no means of preventing evaporative water loss , drought might result in desiccation , and eventually death . To cope with this environmental insult , many organisms enter an ametabolic state known as anhydrobiosis ( Keilin , 1959; Leprince and Buitink , 2015 ) . In this state , organisms can persist in the absence of water for a long period of time; when water becomes available , they exit the anhydrobiotic state and fully resume their normal activities . The nematode Caenorhabditis elegans and the budding yeast Saccharomyces cerevisiae are excellent anhydrobiotes . Studies of these two model organisms have revealed various strategies for desiccation tolerance , many of which appear to be broadly conserved among other anhydrobiotes ( Dupont et al . , 2014; Erkut and Kurzchalia , 2015 ) . One strategy for anhydrobiosis common to both worm and yeast is the biosynthesis and accumulation of trehalose ( Erkut et al . , 2011; Tapia and Koshland , 2014 ) , a disaccharide made of two alpha-linked glucose moieties ( Elbein , 2003 ) . In C . elegans , trehalose preserves the native packing of membranes in the dried state ( Erkut et al . , 2011; 2012 ) and stabilizes membranes against the adverse effects of fast rehydration ( Abusharkh et al . , 2014 ) . In yeast , trehalose also functions as a long-lived chaperone , preventing protein aggregation upon desiccation ( Tapia and Koshland , 2014 ) . These observations suggest that this disaccharide plays conserved roles in desiccation tolerance . However , the metabolic basis for synthesis of trehalose remains largely unknown . In this study , we sought to identify the source of trehalose carbons and the pathway ( s ) that promote trehalose biosynthesis and accumulation . Neither C . elegans nor S . cerevisiae invests in trehalose production during growth and development . By contrast , in their non-proliferative stages , i . e . , the dauer larva in C . elegans ( as shown in Penkov et al . , 2015 , and this study ) and stationary phase in yeast ( François and Parrou , 2001; Werner-Washburne et al . , 1993 ) , both species devote a substantial amount of their internal carbon reserve to trehalose biosynthesis . In the non-feeding dauer larva ( the only desiccation-tolerant stage of the C . elegans life cycle ) , trehalose levels rise dramatically upon exposure to mild desiccation stress ( preconditioning ) ( Erkut et al . , 2011 ) . Similarly , stationary phase yeast cells , which are tolerant to desiccation , also accumulate trehalose ( Calahan et al . , 2011 ) . Thus , in these specific developmental stages , these organisms must be able to divert available carbon sources to the production of sugars . In these non-growing stages , both worm and yeast must enter metabolic modes distinct from those that are active during growth . Reproductive stage larvae of C . elegans feed on bacteria , from which they ingest mostly lipids and proteins , and to some extent sugars; they assimilate these nutrients via glycolysis and/or the TCA cycle to produce energy ( Figure 1A , Figure 1—figure supplement 1A ) . On the other hand , when fed its preferred carbon source ( glucose ) , budding yeast grows exponentially and uses fermentative glycolysis for its energetic and biosynthetic needs ( Figure 1B ) . Under these conditions , the cells secrete ethanol as well as acetate . 10 . 7554/eLife . 13614 . 003Figure 1 . Metabolic modes of C . elegans and S . cerevisiae . ( A ) C . elegans reproductive larvae , which are feeding and growing , can utilize nutrients ( purple ) via TCA cycle and produce energy . Mitochondria are in a catabolic mode ( blue ) . ( B ) During fermentative growth , S . cerevisiae uses glucose to produce energy via glycolysis . ( C ) The non-feeding dauer larva utilizes internal TAG reserves via GS to drive gluconeogenesis and produce trehalose ( orange ) . Mitochondria are in an anabolic mode ( yellow ) . ( D ) In low glucose , high acetate , ethanol and glycerol regime , yeast switches to gluconeogenesis via GS . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 00310 . 7554/eLife . 13614 . 004Figure 1—figure supplement 1 . Metabolic pathways of glycolysis , gluconeogenesis , TCA cycle and glyoxylate shunt reactions during preconditioning . ( A ) Overview of reactions catalyzed by ( 1 ) citrate synthase , ( 2 ) aconitase , ( 3 ) isocitrate dehydrogenase , ( 4 ) α-ketoglutarate dehydrogenase , ( 5 ) succinyl CoA synthetase , ( 6 ) succinate dehydrogenase , ( 7 ) fumarate reductase , ( 8 ) fumarase , ( 9 ) malate dehydrogenase , ( 10 ) isocitrate lyase / malate synthase , ( 11 ) pyruvate dehydrogenase , ( 12 ) phosphoenolpyruvate carboxykinase , ( 13 ) pyruvate carboxylase , ( 14 ) fructose 1 , 6-bisphosphatase , ( 15 ) trehalase , ( 16 ) hexokinase , ( 17 ) ADP-dependent glucokinase , ( 18 ) phosphoglucose isomerase , ( 19 ) phosphofructokinase , ( 20 ) pyruvate kinase , ( 21 ) glutamate dehydrogenase , ( 22 ) glutamine synthetase , ( 23 ) phosphoglucomutase , ( 24 ) UDP-glucose pyrophosphorylase , ( 25a , 25b ) trehalose 6-phosphate synthase , ( 25b , 26 ) trehalose 6-phosphate phosphatase , ( 27 ) various lipases , ( 28 ) glycogen phosphorylase and ( 29 ) various aminotransferases . ( B ) Normalized expression levels of genes encoding the enzymes of these reactions . NP: Non-preconditioned , P: Preconditioned . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 004 Both species shift their metabolism during the transition to non-growing stages . The dauer larva relies on its internal carbon reserves , which are mainly triacylglycerols ( TAGs ) ( Hellerer et al . , 2007; Narbonne and Roy , 2008 ) , but must also retain the ability to produce sugars . In yeast , as glucose is consumed and glucose concentrations drop , cells undergo the diauxic shift , i . e . , a transition to respiratory metabolism ( Schweizer and Dickinson , 2004 ) . Following this shift , yeast produce acetyl-CoA from accumulated ethanol , acetate , and glycerol , and switch to oxidative phosphorylation via the TCA cycle ( Schweizer and Dickinson , 2004 ) , which provides energy as well as precursor metabolites for amino acid biosynthesis and gluconeogenesis . Finally , in the stationary phase , yeast cells accumulate trehalose and glycogen ( François and Parrou , 2001; Schweizer and Dickinson , 2004; Werner-Washburne et al . , 1993 ) . These observations imply that in order to synthesize trehalose , both organisms must undergo a transition to a gluconeogenic mode in which they synthesize glucose or glucose-6-phosphate from non-carbohydrate precursors . How is this transition implemented ? In theory , the TCA cycle could provide the intermediates required for gluconeogenesis , but this pathway generates substantial amounts of ATP , as well as NADH , which must be oxidized to NAD+ to maintain cellular redox balance ( Voet and Voet , 2010 ) . At first glance it seems counterintuitive that these two processes run in parallel , considering the low energetic demands of dauer larvae and stationary phase yeast cells . However , cells could be driven into gluconeogenesis via an alternate route , the glyoxylate shunt ( GS ) ( Figure 1C , D , depicted in red ) . The GS has been implicated in anhydrobiosis in the nematode Aphelencus avenae ( Madin et al . , 1985 ) . We hypothesized that , in C . elegans dauer larvae and stationary phase yeast cells , the GS serves a critical function in anabolic processes required for desiccation tolerance , in particular by enabling or promoting gluconeogenesis for trehalose biosynthesis . The GS , a shortcut in the TCA cycle ( Kornberg and Madsen , 1958 ) , is conserved in bacteria ( Kornberg , 1966 ) , fungi ( Lopez-Boado et al . , 1988; Lorenz and Fink , 2001 ) , protists ( Levy and Scherbaum , 1965; Nakazawa et al . , 2005 ) , nematodes ( Liu et al . , 1995; Madin et al . , 1985; Siddiqui et al . , 2000 ) , and plants ( Eastmond and Graham , 2001; Kornberg and Beevers , 1957 ) . It bypasses two CO2-releasing steps of the TCA cycle ( Figure 1C , Figure 1—figure supplement 1A , reactions 3 and 4 ) to produce succinate , and incorporates an additional molecule of acetyl-CoA to form L-malate from glyoxylate ( Figure 1C , D , Figure 1—figure supplement 1A , reaction 10 ) . Instead of remaining within the TCA cycle , excess malate can be converted into oxaloacetate and diverted into gluconeogenesis ( Figure 1C , Figure 1—figure supplement 1A , reaction 12 ) ( Voet and Voet , 2010 ) . Thus , the GS serves as a prototypical anaplerotic pathway , leading to the accumulation of critical TCA cycle intermediates , particularly oxaloacetate , which can be consumed for gluconeogenesis . Moreover , this pathway generates less ATP and NADH than the TCA cycle ( Kornberg , 1966 ) . To date , the biological importance of the GS has been largely ignored , and its physiological functions remain obscure . The GS has primarily been studied in the context of microbial sporulation and growth ( Kornberg and Krebs , 1957; Megraw and Beers , 1964 ) , fungal virulence ( Lorenz and Fink , 2001 ) , and plant seed germination ( Eastmond et al . , 2000 ) . However , the GS is not physiologically essential to any of these processes ( Voet and Voet , 2010 ) . On the other hand , it is astonishing that C . elegans , a nematode and thus a member of the animal kingdom , has the full set of enzymes required for the GS ( Liu et al . , 1995 ) . Although the GS has been proposed to be involved in sugar homeostasis in the worm ( Frazier and Roth , 2009 ) , its absence results in neither a detectable phenotype nor any effect on wild-type adult lifespan . However , it may be required for the extended longevity of some mitochondrial mutants ( Gallo et al . , 2011 ) . Thus , at present , no physiological role has been definitively assigned to this pathway in the worm . Here , we present evidence that the dauer larva is in a hypoaerobic , gluconeogenic state , which enables efficient production of trehalose using internal reserves ( TAGs and amino acids ) . Importantly , during preconditioning , the GS is the major pathway for conversion of TAGs into trehalose; in its absence , the dauer larva cannot produce sufficient trehalose to survive desiccation . Expanding our studies to the budding yeast , we discovered that S . cerevisiae utilizes a similar metabolic strategy , relying on the GS to drive trehalose synthesis and achieve desiccation tolerance . These results reveal , for the first time , a functionally conserved and central role for the GS in a process that is essential for survival under certain conditions . We characterized the energetic/metabolic states of the dauer larva and its parallel reproductive stage ( the L3 larva ) . Dauer larvae are metabolically less active than L3 larvae ( Burnell et al . , 2005; Kimura et al . , 1997; O'Riordan and Burnell , 1989; 1990; Vanfleteren and DeVreese , 1996 ) . As an indicator of metabolic activity , we compared the respiration rates of dauer and L3 larvae . To obtain large quantities of homogeneous L3 or dauer larva populations , we used the temperature-sensitive dauer-constitutive daf-2 ( e1370 ) strain . Oxygen consumption rates ( OCRs ) in these larvae were measured using an extracellular flux analyzer ( Figure 2—figure supplement 1 ) . To measure mitochondrial OCR , we specifically inhibited Complex IV with sodium azide ( Figure 2—figure supplement 1 ) and calculated the respiration rate as the difference between the overall OCRs of water- and azide-treated worms ( Figure 2A ) . In a given concentration of environmental oxygen , mitochondria of dauer larvae consumed ~5-fold less oxygen than those of L3 larvae ( Figure 2A ) , indicating that dauer larvae exist in a hypoaerobic mode . Moreover , dauer larvae contain much less ATP than L3 larvae ( Penkov et al . , 2015; Wadsworth and Riddle , 1989 ) , indicating that they are also hypometabolic . 10 . 7554/eLife . 13614 . 005Figure 2 . Energetic modes of C . elegans reproductive and dauer larvae . ( A ) Respiration rates in terms of OCR difference between water-treated and azide-treated worms ( n = 4 for each group ) . ANOVA shows that in both strains , L3 larvae consume significantly more oxygen than dauer larvae ( F1 , 12 = 1469 , p < 0 . 001 ) . There is also a minor effect of strain on oxygen consumption ( F1 , 12 = 6 . 864 , p = 0 . 022 ) , however there is no interaction between the larval stage and the strain ( F1 , 12 = 0 . 166 , p = 0 . 691 ) . Error bars show 95% confidence intervals . ( B ) Steady-state trehalose levels of daf-2 and daf-2;icl-1 , L3 and dauer larvae ( n = 3 for each group ) . L3 larvae produce less trehalose than dauer larvae ( F1 , 8 = 92 . 814 , p < 0 . 001 ) independent of the strain ( F1 , 8 = 0 . 083 , p = 0 . 781 ) . Error bars show standard error of the mean . p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 00510 . 7554/eLife . 13614 . 006Figure 2—figure supplement 1 . Details of oxygen consumption rate measurements . Oxygen consumption rates of L3 ( top panels ) and dauer ( bottom panels ) larvae of daf-2 ( left panels ) and daf-2;icl-1 ( right panels ) . Dashed and solid lines indicate OCR after water ( control ) and 20 mM sodium azide injection , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 006 Next , we compared trehalose levels in L3 vs . non-feeding dauer larvae . The latter accumulated substantially larger amounts of trehalose ( Figure 2B ) . This observation suggested that , in addition to being hypometabolic , dauer larvae rearrange their metabolism to favor intensive gluconeogenesis , leading to trehalose accumulation . To investigate this possibility , we adopted an approach that combined metabolic labeling with 2-dimensional high-performance thin-layer chromatography ( 2D-HPTLC ) . This relatively simple method enabled us to detect major small-molecules , including amino acids , sugars , and intermediates of the TCA cycle ( Figure 3—figure supplement 1A ) . First , we labeled C . elegans metabolites by feeding the worms 14C-acetate-supplemented bacteria until they formed L3 or dauer larvae . This labeling strategy allowed us to detect and identify metabolites derived from 14C-acetate that has entered the TCA cycle . Subsequently , we extracted the metabolites from worms , separated the extracts into organic and aqueous phases , and analyzed the latter with 2D-HPTLC . The aqueous phase of L3 extract contained many labeled compounds , including various amino acids ( Figure 3A , Figure 3—figure supplement 1A ) , but trehalose was not abundant ( Figure 3A , spot 1 ) . Thus , in this growth stage , the TCA cycle is mainly cataplerotic: in addition to reducing NAD+ and producing ATP , L3 larvae use intermediates to synthesize various building blocks such as amino acids , nucleotides , and sugars . By contrast , the aqueous fraction of dauer larvae contained only a limited number of metabolites , and fluorograms of this extract had one predominant spot , i . e . , trehalose ( Figure 3B , spot 1 ) . Other , barely detectable spots corresponded to glucose , glutamate , and glutamine ( Figure 3B , spots 2 , 3 , and 4 , respectively ) . These data suggest that metabolism in dauer larvae is almost entirely switched to a gluconeogenic mode in which sugars are produced by non-carbohydrate sources ( acetate/fatty acids ) . 10 . 7554/eLife . 13614 . 007Figure 3 . Metabolic modes of C . elegans reproductive and dauer larvae . ( A–C ) Radioactively labeled metabolites of daf-2 L3 , as well as non-preconditioned ( NP ) and preconditioned ( P ) dauer larvae . Enumerated spots indicate trehalose ( 1 ) , glucose ( 2 ) , glutamate ( 3 ) and glutamine ( 4 ) . ( D–F ) The same analysis for daf-2;icl-1 . Equivalent metabolome extracts were separated and exposed for 2 days for both strains and larval/experimental conditions . ( G ) Steady state trehalose levels before and after preconditioning in daf-2 and daf-2;icl-1 dauer larvae ( n = 3 for each group ) . Both strains elevate their trehalose levels upon preconditioning ( ANOVA for preconditioning reports F1 , 8 = 85 . 20 , p < 0 . 001 ) but to different extents ( ANOVA for strain reports F1 , 8 = 30 . 11 , p < 0 . 001; interaction between strain and preconditioning F1 , 8 = 11 . 26 , p = 0 . 010 ) . Error bars show standard error of the mean . p < 0 . 001 . ( H ) Induction of non-labeled and 14C-labeled trehalose upon preconditioning in daf-2 and daf-2;icl-1 dauer larvae expressed as fold changes ( n = 3 for each group ) . ANOVA shows that daf-2 larvae induce both non-labeled and labeled trehalose more than daf-2;icl-1 larvae ( F1 , 8 = 26 . 229 , p < 0 . 001 ) however induction in labeled trehalose does not differ from non-labeled ( F1 , 8 = 0 . 343 , p = 0 . 571 ) . Error bars show standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 00710 . 7554/eLife . 13614 . 008Figure 3—figure supplement 1 . Details for the detection of metabolites . ( A ) Map of metabolites in the 2D-HPTLC system . Amino acids , sugars and miscellaneous metabolites are indicated in red , green and blue , respectively . Metabolites are spread only on a part of the plate ( 5 . 1 × 5 . 1 cm ) . ( B ) Incorporation of 14C into TAGs after metabolic labeling . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 00810 . 7554/eLife . 13614 . 009Figure 3—figure supplement 2 . Trehalose 6-phosphate synthase ( TPS ) levels in worm and yeast . ( A ) TPS activity in worm lysates of daf-2 and daf-2;icl-1 background ( n = 3 for each group ) . Activity unit is defined as the amount of trehalose 6-phosphate ( nmol ) produced per min , normalized to total soluble protein amount in the lysate . Two-sample t-test shows no significant difference between TPS activities of different strains ( p = 0 . 783 ) . ( B ) Tps1 and Tps2 levels in wild type , ∆icl1 and ∆icl1/∆mls1 yeast . Both proteins are expressed at comparable levels in all strains . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 009 Previously , we showed that preconditioning of the dauer larva prior to harsh desiccation induces production of a massive amount of trehalose ( Erkut et al . , 2011 ) . For preconditioning , worms are treated with mild desiccation at 98% relative humidity ( RH ) for an extended period of time ( 4 days ) , after which they can survive in the almost complete absence of water ( Erkut et al . , 2011 ) . In this study , we preconditioned worms in this manner , and then analyzed the organic and aqueous fractions of radioactively labeled dauer larvae before and after preconditioning . In the organic phase , the amount of radioactivity incorporated into TAGs decreased substantially during preconditioning ( Figure 3—figure supplement 1B ) . At the same time , preconditioning dramatically increased the level of radioactively labeled trehalose ( Figure 3C , spot 1 ) , and the amounts of glutamate and glutamine also increased ( Figure 3C , spots 3 and 4 , respectively; this observation is discussed later ) . These results suggest that the dauer larva takes advantage of its gluconeogenic mode to boost trehalose synthesis upon desiccation stress . We next asked how the transition to this gluconeogenic mode is reflected in the transcriptome . Previously , we surveyed differential expression of C . elegans genes during preconditioning ( Erkut et al . , 2013 ) . In this study , we revisited our data to focus on genes involved in the TCA cycle and gluconeogenesis ( Figure 1—figure supplement 1A ) . Transcripts encoding enzymes required for gluconeogenesis , mdh-1 , mdh-2 , and pck-2 ( Figure 1—figure supplement 1B , enzymes 9 and 12 ) , were expressed at relatively high levels in dauer larvae even before preconditioning . Moreover , mdh-1 ( cytosolic malate dehydrogenase ) and pck-2 ( phosphoenolpyruvate carboxykinase ) , both of which are crucial for gluconeogenesis , were significantly upregulated during preconditioning , consistent with the increase in gluconeogenesis and sugar accumulation observed in the dauer larva . Collectively , our data demonstrate that , during dauer formation , worms enter a gluconeogenic mode associated with a large increase in the levels of sugars such as trehalose , and that this phenomenon is even more pronounced during preconditioning . We hypothesized that the GS in dauer larvae plays an important role in gluconeogenesis , and thus in trehalose biosynthesis . We tested this idea in worm lines having no functional GS . In plants , yeast , and bacteria , two enzymes are responsible for the GS: isocitrate lyase ( EC 4 . 1 . 3 . 1 ) , which breaks isocitrate down to glyoxylate and succinate , and malate synthase ( EC 2 . 3 . 3 . 9 ) , which condenses glyoxylate and acetyl-CoA to produce L-malate ( Figure 1C , Figure 1—figure supplement 1A ) ( Cozzone , 1998 ) . In C . elegans , these two enzymes are combined in one protein , ICL-1 ( formerly known as GEI-7 ) , which has both isocitrate lyase and malate synthase domains , and can thus carry out both reactions ( Liu et al . , 1995 ) . We produced a strain ( daf-2;icl-1 ) with a deletion mutation in icl-1 , and then exploited the daf-2 background to produce large populations of pure dauer larvae . The deletion in the icl-1 ( ok531 ) allele introduces a frame-shift and an early stop codon ( A373* ) , which should completely inactivate the GS strains harboring this mutation . Compared to daf-2 , the dauer and L3 larvae of daf-2;icl-1 exhibited no difference in respiration rate ( Figure 2A ) or basal trehalose levels ( Figure 2B ) , indicating that the GS has no influence on oxygen consumption or basal gluconeogenesis . By contrast , trehalose induction upon preconditioning differed dramatically between daf-2 and daf-2;icl-1 dauer larvae . We quantitated the total amount of trehalose , normalized to the amount of soluble protein , in both strains before and after preconditioning ( Figure 3G ) . Similar to our previous findings ( Erkut et al . , 2011 ) , the total trehalose level in daf-2 dauer larvae was ~100 µg trehalose/mg protein at baseline , and increased ~5-fold upon preconditioning ( Figure 3G , H ) . In daf-2;icl-1 , although the initial level of trehalose was the same as that of daf-2 , the increase was only 2-fold ( Figure 3G , H ) . This suggests that the major source of trehalose during preconditioning is the GS . To further investigate this possibility , we labeled daf-2;icl-1 larvae with 14C-Ac , as described above for daf-2 ( Figure 3D–F ) . Incorporation of radioactivity into trehalose was considerably reduced in daf-2;icl-1 relative to that in daf-2 ( compare Figure 3B and E ) . Nevertheless , as in daf-2 , the level of 14C-labeled trehalose increased in daf-2;icl-1 ( Figure 3E , F ) . Densitometry of fluorogram spots revealed that , during preconditioning , radioactively labeled trehalose increased ~6-fold in daf-2 , but only 2-fold in daf-2;icl-1 ( Figure 3H ) . The average increase in labeled trehalose in daf-2 was larger than the increase in total ( i . e . , unlabeled ) trehalose ( Figure 3H ) , suggesting preferential use of lipid sources for sugar production in this strain . By contrast , in daf-2;icl-1 , the levels of total and labeled trehalose increased to similar extents . These differences in trehalose induction levels between daf-2 and daf-2;icl-1 cannot be assigned to the trehalose biosynthetic pathway because the trehalose 6-phosphate synthase ( Figure 1—figure supplement 1A , reaction 25 ) activity of daf-2 is not higher than that of daf-2;icl-1 ( Figure 3—figure supplement 2A ) . Taken together , these results suggest that the utilization of acetate ( and thus fatty acids ) for gluconeogenesis and trehalose biosynthesis depends on the existence of a functional GS . Next , we asked whether the absence of the GS affects desiccation tolerance . For this purpose , we determined the survival rates of dauer larvae after mild ( 98% RH ) and harsh ( 60% RH ) desiccation . As described above and in our previous studies ( Erkut et al . , 2011 ) , preconditioning induced trehalose accumulation strongly in daf-2 and slightly in daf-2;icl-1 ( Figure 3G ) . As expected , a strain harboring a knockout of trehalose 6-phosphate synthase ( daf-2;∆∆tps ) was unable to synthesize trehalose ( Figure 4A ) . The desiccation survival assay revealed that all strains were equally tolerant to mild desiccation at 98% RH ( Figure 4B ) . However , daf-2;icl-1 was much more sensitive to harsh desiccation than daf-2 , and exhibited very poor survival under those conditions ( Figure 4B ) , although it was significantly more tolerant than daf-2;∆∆tps ( Figure 4B ) . These results suggest that a threshold level of trehalose must be reached during preconditioning in order for the worm to survive harsh desiccation . 10 . 7554/eLife . 13614 . 010Figure 4 . Effect of the glyoxylate shunt on desiccation tolerance . ( A ) Trehalose levels before ( NP ) and after ( P ) preconditioning in daf-2 , daf-2;∆∆tps and daf-2;icl-1 dauer larvae after separation with HPTLC and visualization via Molisch’s staining . Tre: Trehalose , Glc: Glucose . ( B ) Survival levels of the same strains at 98% and 60% RH after preconditioning ( dark and light boxes , respectively ) . Statistical comparison was done with beta regression followed by multiple hypothesis testing . Analysis of deviance results indicate that survival levels depend both on the strain ( χ22 = 124 . 64 , p < 0 . 001 ) and the RH ( χ12 = 141 . 46 , p < 0 . 001 ) . Error bars show 95% confidence intervals . p < 0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 01010 . 7554/eLife . 13614 . 011Figure 4—figure supplement 1 . GS is not involved in heat-shock stress in C . elegans and S . cerevisiae . ( A ) Survival rates of worms upon heat shock at 30°°C , 32°C , 34°C and 37°C for up to 16 hr . Red and blue lines show daf-2 and daf-2;icl-1 strains , respectively . Error bars show standard error of the mean calculated via beta regression . Analysis of deviance results indicate that survival does not depend on the strain ( χ12 = 0 . 221 , p < 0 . 638 ) but it depends on time of exposure ( χ12 = 56 . 071 , p < 0 . 001 ) and temperature ( χ12 = 86 . 041 , p < 0 . 001 ) . ( B ) Resistance of wild type and GS-deficient cells to heat stress . Cells at decreasing cell densities ( OD600 = 1 . 0–0 . 001 ) were subjected to a single heat shock ( 50°C for 30 min ) , and survival was estimated by spotting on YPD plates . ( C ) Cells at a constant cell density ( OD600 = 0 . 1 ) were subjected to increasing durations of heat shock ( 50°C for 45 , 60 and 90 min ) and survival was estimated by spotting onto YPD plates . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 011 To investigate the possibility that the GS might play a role in protection against environmental insults unrelated to desiccation stress , we also challenged dauer larvae with heat shock . Both daf-2 and daf-2;icl-1 dauer larvae survived at least 16 hr of heat shock at 30 or 32°C ( Figure 4—figure supplement 1A ) . At 34°C , survival dropped dramatically after 12 hr , with no dependence on strain ( Figure 4—figure supplement 1A ) . Finally , heat shock at 37°C could only be tolerated for 4 hr , again independent of the status of the GS shunt ( Figure 4—figure supplement 1A ) . These results suggest that the GS is specifically involved in tolerance of desiccation , and possibly related stresses , in C . elegans . In plants , the GS takes place in a specialized peroxisome called the glyoxysome ( Eastmond and Graham , 2001 ) , whereas in yeast , GS enzymes are distributed between the cytosol and peroxisomes ( Duntze et al . , 1969; Kunze et al . , 2006; McCammon et al . , 1990 ) . However , the localization of the GS pathway in nematodes has not been previously investigated . Using a bioinformatics tool ( Claros and Vincens , 1996 ) , we analyzed the C . elegans ICL-1 protein sequence . We identified a 20 amino acid N-terminal mitochondrial import sequence ( MSSAAKNFYQVVKSAPKGRF ) and calculated an 88% probability that the protein is imported into mitochondria . To determine the localization of ICL-1 ( and thus the site of GS activity ) in the worm , we generated a transgenic strain that expresses the ICL-1::GFP fusion protein under the control of the icl-1 promoter , which mimics endogenous expression . We first analyzed the localization of ICL-1::GFP in reproductive stage , actively feeding L3 larva . The protein was expressed at the highest levels in hypodermal cells ( Figure 5A ) , although strong expression was also detected in the pharynx ( Figure 5—figure supplement 1A ) and gut ( Figure 5—figure supplement 1B ) . In hypodermal syncytium , the protein was present in a tubular network interspersed with spherical structures ( Figure 5A ) , resembling mitochondrial staining of C . elegans ( Lee et al . , 2003 ) . To verify that ICL-1 is indeed localized to mitochondria , we fed worms the mitochondrial dye MitoTracker Red CMXRos ( Figure 5B ) . In hypodermal cells expressing ICL-1 , the MitoTracker and GFP signals fully overlapped , whereas seam cells did not express ICL-1 at all ( Figure 5C ) . It should be noted that TPS-1 , the key enzyme in trehalose biosynthesis , is almost exclusively localized to hypodermis and is not expressed in seam cells ( Penkov et al . , 2015 ) . 10 . 7554/eLife . 13614 . 012Figure 5 . ICL-1 is a mitochondrial protein . ( A ) Subcellular localization of ICL-1::GFP in L3 hypodermis ( B ) Mitochondrial staining of L3 hypodermis . ( C ) Colocalization of mitochondria and ICL-1::GFP in L3 hypodermis . Seam cells are circled with dashed curves . ( D ) Subcellular localization of ICL-1::GFP in dauer hypodermis . Seam cells are circled with dashed curves . ( E ) Subcellular localization of ICL-1::GFP in dauer gut . Gut lumen is shown as a dashed line . Scale bar corresponds to 10 µm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 01210 . 7554/eLife . 13614 . 013Figure 5—figure supplement 1 . Expression of ICL-1 in different tissues . ( A ) Pharynx . ( B ) Gut . Gut lumen is shown between dashed lines . Scale bar corresponds to 10 µl for both images . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 013 Because they are non-feeding and impermeable , dauer larvae cannot be stained with MitoTracker . Nevertheless , at the subcellular level , the distribution of the GFP signal in dauer larvae closely resembled the mitochondrial network ( Figure 5D ) . Once again , the hypodermis was the main tissue expressing ICL-1 , although the protein was also expressed in the gut ( Figure 5E ) . Collectively , these results indicate that C . elegans ICL-1 is mitochondrial , and suggest that in the worm , the GS occurs within or in association with mitochondria . Based on the striking conceptual similarity between C . elegans dauer larvae and S . cerevisiae cells entering stationary phase , we postulated the existence of a conserved mechanism for desiccation tolerance . As described earlier , dauer larvae accumulate large amounts of trehalose , despite the fact that worms in this stage of the life cycle do not feed or grow . A similar phenomenon occurs in budding yeast . In the presence of its preferred carbon source ( glucose ) , yeast uses fermentative glycolysis during rapid proliferation ( Figure 1B ) ; under these conditions , very little trehalose accumulates ( Tapia and Koshland , 2014 ) . However , once glucose concentration falls , the cells undergo a diauxic shift , thereafter using aerobic respiration for their energetic needs in order to continue proliferation ( Figures 1D , 6A ) ; eventually , as external energy sources are depleted , the cells enter stationary phase . Although both glycolytic and respiratory activities are low in stationary phase , the cells continue to accumulate trehalose and glycogen , which ultimately constitute >30% of total cell mass ( François and Parrou , 2001; Werner-Washburne et al . , 1993 ) . Thus , yeast might also rely on alternate carbon metabolism to generate trehalose . Therefore , we asked whether the GS in yeast can be used to drive gluconeogenesis for synthesis of trehalose . 10 . 7554/eLife . 13614 . 014Figure 6 . Growth of GS-deficient yeast cells in media with different carbon sources . ( A ) Growth of wild-type S . cerevisiae in YP + Glc medium in batch culture . The time of diauxic shift is shown with a dashed line . Error bars show standard deviation ( n = 3 ) . ( B–D ) Growth of wild-type ( WT ) or GS-deficient yeast in amino acid rich ( YP ) or minimal medium with glucose ( B ) , ethanol ( C ) and acetate ( D ) as the carbon source . Note that particularly with acetate as the primary carbon source , GS mutants ( ∆icl1 , ∆mls1 , ∆dal7 , ∆mls1/∆dal7 and ∆icl1/∆mls1/∆dal7 ) grow poorly regardless of amino acid availability . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 014 In stationary phase , yeast can consume ethanol , glycerol , and particularly acetate to generate acetyl-CoA , either directly or through gluconeogenesis . Acetyl-CoA can enter the TCA cycle as well as the GS ( Figure 1D ) . In yeast , the GS is carried out by isocitrate lyase , Icl1p ( Fernandez et al . , 1992 ) , and the malate synthases , primarily Mls1p but also Dal7p ( Hartig et al . , 1992 ) . We first compared the growth rates of WT yeast harboring mutations in GS components ( ∆icl1 , ∆mls1 , ∆dal7 , and combinations thereof ) under conditions in which we altered carbon sources as well as the availability of free amino acids ( Figure 6B–D ) . As expected , GS mutants exhibited no significant growth defect when grown in high glucose , irrespective of the presence or absence of amino acids ( Figure 6B ) . By contrast , when grown in high ethanol , GS mutants grew normally in YP medium , but their growth was impaired in minimal medium ( S min ) lacking free amino acids ( Figure 6C ) . This suggests that free amino acids can feed into carbon consumption in the TCA cycle , as C . elegans , and that in the absence of free amino acids , the GS plays an important role in carbon metabolism . In addition , we compared the growth rates of WT and GS-deficient cells growing on acetate as the sole carbon source ( Figure 6D ) . Under these conditions , yeast have high GS activity and exhibited elevated TCA-independent acetate metabolism ( Lee et al . , 2011; Schweizer and Dickinson , 2004 ) . Regardless of the availability of free amino acids , GS mutants grew very poorly under these conditions ( Figure 6D ) . We predicted that after the diauxic shift , GS-deficient yeast would exhibit reduced synthesis and accumulation of trehalose and glycogen . To test this idea , we quantitated these metabolites in WT and GS-deficient cells ( Figure 7A–D ) . In cells grown in glucose , following the diauxic shift ( 20 hr , black bars ) and in stationary phase ( 48 hr , grey bars ) , WT cells accumulated considerable amounts of trehalose ( Figure 7A ) and glycogen ( Figure 7C ) . By contrast , despite reaching high cell densities , GS-deficient cells contained low ( but detectable ) amounts of trehalose , but no detectable glycogen stores ( Figure 7A , C ) ( Figure 7—figure supplement 1A ) . We also measured trehalose and glycogen in WT and GS-deficient cells grown with acetate as a carbon source . Under these conditions , we observed an even greater accumulation of trehalose and glycogen in WT cells , whereas GS-deficient cells contained low levels of trehalose and no detectable glycogen ( Figure 7B , D ) . As controls , we also measured the amounts of trehalose and glycogen in mutant cells lacking trehalose synthase ( Tps1 ) or the heat-shock protein Hsp104 , both of which are important for yeast desiccation tolerance ( Tapia and Koshland , 2014 ) . As expected , ∆tps1 cells contained very low levels of trehalose , but high levels of glycogen , whereas ∆hsp104 cells had no defects in trehalose ( Figure 7—figure supplement 1B ) or glycogen storage ( Figure 7—figure supplement 1C ) . We also measured Tps1 and Tps2 amounts ( Figure 3—figure supplement 2B ) , which were unchanged in wild type and GS-deficient cells . This rules out the trivial explanation that GS-deficient cells have limitations in trehalose 6-phosphate synthase levels , and therefore have lower trehalose amounts . Collectively , these data show that , much like preconditioned C . elegans dauer larvae , S . cerevisiae cells rely on TCA cycle-independent acetyl-CoA consumption through the GS to synthesize trehalose and glycogen . 10 . 7554/eLife . 13614 . 015Figure 7 . Trehalose/glycogen synthesis and desiccation tolerance in GS-deficient yeast cells . ( A , B ) Steady-state trehalose levels of WT , ∆icl1 and ∆icl-1/∆mls1/∆dal7 strains in YP + Glc ( A ) and YP + Ac ( B ) media after 20 hr ( post-diauxic shift , dark bars ) and 48 hr ( stationary phase , light bars ) . ( C , D ) Steady-state glycogen levels under the same conditions . n . d . : Not detected/below assay sensitivity range . ( E , F ) Desiccation tolerance of the indicated WT and mutant yeast cells , measured after 24 hr ( E ) or 30 days ( F ) of desiccation . Error bars show 95% confidence intervals . p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 01510 . 7554/eLife . 13614 . 016Figure 7—figure supplement 1 . Cell density , trehalose and glycogen levels in WT and mutant yeast strains . ( A ) Maximum cell density attained by wild type and GS-deficient cells after 48 hr of growth in YP + Glc medium or YP + Ac medium . ( B ) Trehalose levels in WT , ∆hsp104 and ∆tps1 cells after 48 hr of growth in YP + Glc medium . ( C ) Glycogen levels in WT , ∆hsp104 and ∆tps1 cells after 48 hr of growth in YP + Glc medium . Error bars represent 95% confidence intervals . *p < 0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 016 Therefore , we asked whether this GS-dependent trehalose accumulation was required for desiccation tolerance in yeast . We grew saturated cultures of WT , GS-deficient , Tps1-deficient , or Hsp104 deficient cells , desiccated them for up to 30 days , and then rehydrated them . All strains survived very well after 1 day of desiccation ( Figure 7E ) . The viability of cells in WT cultures remained high after 30 days of desiccation ( Figure 7F ) . By contrast , ∆icl1 , ∆icl1/∆tps1 , and ∆icl1/∆hsp104 cells exhibited very poor desiccation tolerance after 30 days , with viability at least 10-fold lower than that of WT cells and comparable to that of ∆tps1 cells lacking trehalose altogether ( Figure 7F ) . Importantly , ∆icl1 cells supplemented with trehalose in the medium 24 hr before desiccation exhibited near-WT levels of desiccation tolerance ( Figure 7F ) . Finally , we expanded our study to investigate whether other environmental insults , such as heat-shock and freezing/thawing , were affected by GS deficiency . We first tested the ability of S . cerevisiae cells to tolerate elevated temperatures , subjecting wild type or GS-deficient mutants to heat shock at 50°C either at different cell densities , or for increasing amounts of time ( Figure 4—figure supplement 1B and C ) . Under both conditions , wild type as well as GS-deficient cells showed similar sensitivity to heat shock . Another environmental insult that yeast seasonally encounter is freezing and thawing . This stress could conceivably affect cell membranes and proteins similarly to desiccation . We subjected stationary phase cultures of WT or GS-deficient cells ( resuspended in water ) to multiple freeze-thaw cycles , and monitored viability by simple spotting and growth assays ( Figure 8 ) . Although a considerable proportion of WT cells survived even after six freeze-thaw cycles , GS-deficient cells underwent a dramatic loss in viability after just two cycles ( Figure 8 ) . 10 . 7554/eLife . 13614 . 017Figure 8 . Resistance of WT and GS-deficient cells to freezing and thawing . Equal numbers of cells were subjected to multiple freeze-thaw cycles , and survival estimated by spotting onto YPD plates . The plates were imaged after different times of recovery , to more carefully observe survival phenotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 13614 . 017 Together , our observations demonstrate that the GS plays an essential role in yeast desiccation tolerance by promoting gluconeogenesis for the synthesis of trehalose . Our data also suggest that GS-dependent trehalose synthesis is critical for survival under conditions of other water-related stresses , such as freezing . Here , we demonstrated that dauer larvae exist in a hypometabolic state in which metabolism is redirected largely towards gluconeogenesis . This state depends primarily on an active glyoxylate shunt ( GS ) , which serves as the main route for synthesis of trehalose from TAG reserves during preconditioning . We also showed that budding yeast undergoes a conceptually convergent process . In order to survive desiccation , stationary phase yeast cells must produce high levels of trehalose from acetate or glycerol . This conversion can be successfully accomplished only in the presence of a functional GS . In addition to the GS , gluconeogenesis/trehalose biosynthesis should use alternative carbon sources , such as amino acids , glycerol , and pyruvate , because daf-2;icl-1 accumulated some trehalose upon preconditioning . Amino acids , for example , can be converted into their corresponding α-keto acids , which then undergo a specific series of reactions to enter the TCA cycle , in which they are subsequently used for gluconeogenesis ( Figure 1—figure supplement 1A ) . However , even via utilization of amino acids or other metabolites , daf-2;icl-1 dauer larvae cannot tolerate desiccation as well as daf-2 . Our data indicate that an intermediate level of trehalose is insufficient for desiccation tolerance: Below some threshold level , trehalose cannot exert its protective effects and overcome the adverse consequences of desiccation . This hypothesis was previously explored in yeast ( Tapia et al . , 2015 ) . Our results strongly suggest that a conceptually similar trehalose threshold may exist in the worm as well . It is worth to note that gluconeogenesis is in general associated with the energetic needs of the organism . In C . elegans and the yeast , however , it is additionally used as a defense against environmental stress . Taking advantage of the GS , worms and yeast devote large amounts of resources to the production of trehalose , which in turn protects the organism against desiccation or other water-related stresses , such as freezing and thawing during winter . Although the GS was discovered and dissected at the molecular level almost 60 years ago , no clear physiological function has yet been assigned to it . For several decades , the germination of plant seeds was considered to be the most prominent process requiring this pathway , and it was assumed that production of sugars from seed oils was a prerequisite for germination . However , studies of Arabidopsis lines with no functional GS revealed that in the presence of light , this pathway is non-essential ( Eastmond et al . , 2000; Eastmond and Graham , 2001 ) . Other studies suggested a requirement for the GS in fungal virulence ( Lorenz and Fink , 2001 ) , although at present we have no mechanistic understanding of why this would be the case . To the best of our knowledge , our study provides the first clearly defined physiological role for the GS in C . elegans and S . cerevisiae . It remains unclear how the GS is regulated in animals . In the worm , regulation is directly connected to sensing of a desiccative environment ( hygrosensation ) . We previously showed that hygrosensation is at least partially mediated by head neurons ( Erkut et al . , 2013 ) . Although we still do not know the details of this sensation , one of its downstream effects is likely to be increased lipolysis . This process , followed by β-oxidation of fatty acids , yields acetyl-CoA , the fuel for GS . It is therefore reasonable to speculate that lipolysis determines the extent of the GS . Another finding that supports this view is the localization of ICL-1 within the organism . As shown above , ICL-1 is predominantly localized in the hypodermis . The major enzyme that synthesizes trehalose ( TPS ) is localized to the same tissue ( Penkov et al . , 2015 ) , but is not expressed in the gut . On the other hand , the major TAG deposit in the form of fat droplets resides in the gut , and to some extent in the hypodermis ( Mak , 2011 ) . Thus , synthesis of trehalose in the hypodermis must depend on the transport of fatty acids from the gut . Indeed , as we previously found , one of the fatty acid-binding proteins ( FAR-3 ) is strongly upregulated at both the transcriptional and translational levels upon preconditioning ( ~160 and four fold , respectively ) ( Erkut et al . , 2013 ) . FAR-3 is predicted to have a 20 amino acid signal sequence for secretion , and could thus be involved in the transport of fatty acids between cells . These observations strongly suggest that the regulation of the GS by substrate availability is a complex process that depends on interactions between different tissues . An interesting aspect of our study is that worm ICL-1 is localized to mitochondria , suggesting that the GS takes place in this organelle . By contrast , in plants and the yeast , this pathway is split between mitochondria and a specialized organelle ( glyoxysome or peroxisome ) or the cytoplasm . Furthermore , in contrast to yeast and many other organisms , C . elegans ICL-1 is a bifunctional enzyme with both glyoxylase and malate synthase activities . The physiological meaning of these differences remains elusive , but they suggest that transition into gluconeogenic mode is regulated differently in different organisms . In summary , we showed that dauer larvae and stationary phase yeast switch to a gluconeogenic mode , in which the GS plays an essential role . In both species , loss of the GS is deleterious during desiccation . Our results reveal a novel physiological role for the GS and a conserved mechanism by which diverse organisms can regulate their metabolism to achieve desiccation tolerance . C . elegans wild-type ( N2 ) , daf-2 ( e1370 ) III , icl-1 ( ok531 ) V , tps-1 ( ok373 ) X and tps-2 ( ok526 ) II strains were received from Caenorhabditis Genetics Center ( Minneapolis , MN ) , which is funded by the NIH Office of Research Infrastructure Programs ( P40 OD010440 ) . The glyoxylate shunt mutant icl-1 was outcrossed twice with N2 and subsequently crossed to daf-2 to generate daf-2 ( e1370 ) III;icl-1 ( ok531 ) V ( daf-2;icl-1 ) . The trehalose-deficient strains tps-2 ( ok526 ) II;tps-1 ( ok373 ) X ( ∆∆tps ) and tps-2 ( ok526 ) II;daf-2 ( e1370 ) III;tps-1 ( ok373 ) X ( daf-2;∆∆tps ) were previously generated in our group ( Penkov et al . , 2010 ) . The transgenic line was obtained by ballistic transformation of a fosmid construct encoding the C-terminal translational fusion protein ICL-1::eGFP , generated by our TransgeneOmics facility ( Sarov et al . , 2012 ) . The construct was isolated and purified using a FosmidMAX DNA purification kit ( Epicentre , Madison , WI ) and sequenced to confirm its identity . Microparticle bombardment was performed as explained elsewhere ( Sarov et al . , 2012 ) . Transgenic worms showing the GFP marker and rescue of Unc phenotype were screened for 2 generations to pick up an integrated line . This strain was then outcrossed twice with N2 and finally crossed to daf-2 to obtain daf-2 ( e1370 ) III;Is[icl-1::GFP+unc-119] ( daf-2;icl-1::GFP ) . Worms were maintained at 15°C on nematode growth medium ( NGM ) agar plates seeded with Escherichia coli NA22 ( Brenner , 1974 ) . Gravid adults on NGM agar plates were treated with alkaline hypochlorite solution ( i . e . , bleached ) to purify eggs . Dauer larvae of Daf-c strains were obtained by growing these eggs in complete S medium ( liquid culture ) ( Sulston and Brenner , 1974 ) at 25°C for 5 days unless stated otherwise . To obtain dauer larvae of other strains , we first let the eggs grow into gravid adults on sterol-depleted lophanol ( 4α-methyl-5α-cholestan-3β-ol ) -substituted agarose plates at 20°C for 4 days ( Matyash et al . , 2004 ) . Subsequently , these adults were bleached and their eggs were grown in cholesterol-free lophanol-substitued liquid culture at 25°C for 5 days . L3 larvae were obtained by growing eggs at 15°C in liquid culture for 3 days . To radioactively label lipids and sugars in worms , we let the eggs grow on NGM agar plates supplemented with bacteria mixed with 10 µCi 14C-labeled sodium acetate ( CH314COONa , Hartmann Analytic , Germany ) for 3 days at 15°C or 25°C until they became L3 or dauer larvae , respectively . The prototrophic Sacharomyces cerevisiae CEN . PK strain background was used in all experiments ( van Dijken et al . , 2000 ) . Strains that have been generated and used in this study are ∆icl1 ( MAT a Δicl1::NAT ) , ∆mls1 ( MAT a Δmls1::KanMX ) , ∆dal7 ( MAT a Δdal7::KanMX ) , ∆dal7/∆mls1 ( MAT a Δdal7::KanMX Δmls1::Hyg ) , ∆icl1/∆mls1/∆dal7 ( MAT a Δicl1::NAT Δmls1::KanMX Δdal7::Hyg ) , ∆tps1 ( MAT a ∆tps1::Hyg ) , ∆icl1/∆tps1 ( MAT a ∆icl1::NAT ∆tps1:Hyg ) , ∆hsp104 ( MAT a ∆hsp104::KanMX ) , ∆icl1/∆hsp104 ( MAT a ∆icl1::NAT ∆hsp104::KanMX ) , Tps1-FLAG ( MAT a Tps1-FLAG::NAT ) , Tps2-FLAG ( MAT a Tps2-FLAG::NAT ) , ∆icl1/Tps1-FLAG ( MAT a Δicl1::NAT Tps1-FLAG::Hyg ) , ∆icl1/Tps2-FLAG ( MAT a Δicl1::NAT Tps2-FLAG::Hyg ) , ∆mls1/Tps1-FLAG ( MAT a Δmls1::KanMX Tps1-FLAG::Hyg ) and ∆mls1/Tps2-FLAG ( MAT a Δmls1::KanMX Tps2-FLAG::Hyg ) . Gene deletions were performed using standard PCR-based strategies ( Longtine et al . , 1998 ) . Standard formulations for rich medium ( YP: yeast extract , peptone ) or synthetic minimal medium ( S: Yeast Nitrogen Base ( YNB ) and ammonium sulfate without amino acids ) with the specified carbon source were used . The carbon sources were 2% dextrose , 2% ethanol + 2% glycerol or 2% sodium acetate . Cell growth in a specified medium was measured using a serial dilution assay on plates . Briefly , cells were grown in YP with 2% glucose for 12 hr , after which they were harvested , washed twice in water , and serial diluted in water ( starting OD600 = 1 . 0 ) , following which , 5 µl drops were spotted onto agar plates containing YP or S medium with glucose , ethanol and glycerol , or acetate , and cell growth was measured by imaging the plates . Cell growth rates in YPD medium were measured by monitoring absorbance ( OD600 ) over time . Worms were harvested from liquid cultures or plates ( after radioactive labeling ) in distilled water and washed extensively to remove bacteria and debris . Preconditioning for subsequent biochemical analysis was done by first filtering dauer larvae on TETP membranes ( Merck-Millipore , Germany ) and then placing them in a controlled humidity chamber equilibrated at 98% RH ( Erkut et al . , 2011 ) . After 4 days of incubation at 25°C , these worms were collected in distilled water and frozen . L3 or non-preconditioned dauer larvae were frozen right after they were harvested . Desiccation survival assay was performed as described before ( Erkut et al . , 2013 ) . Briefly , in duplicate , 5 µl of worm slurry ( approximately 1000 worms ) in distilled water was dropped into the middle of a 35 mm plastic dish and placed into a controlled humidity chamber equilibrated at 98% RH . After 4 days of preconditioning at 25°C , one replicate was transferred to another controlled humidity chamber equilibrated at 60% RH and kept there for 1 day at 25°C . Meanwhile , the other replicate was left in the 98% RH chamber . Finally , worms were rehydrated with distilled water for 2–3 hr at room temperature and transferred to NGM agar plates seeded with E . coli . They were let recover at 15°C overnight . Next day , alive and dead worms were counted to calculate the survival rate . This experiment was carried out on 3 different days with 3 technical replicates on each day for each treatment . Desiccation tolerance assays were performed as described earlier ( Tapia and Koshland , 2014 ) , with slight modifications . Briefly , ~107 cells were collected from batch cultures ( grown for 96 hr in YPD ) , washed twice in dilute PBS , and brought to a final volume of 1 ml . Non-desiccated controls were plated on YPD agar for colony counting . Two hundred microliter aliquots were transferred to a 96-well tissue culture plate , centrifuged , and the excess water was removed . Cells were allowed to desiccate in a humid incubator at 27°C . Long-term desiccation experiments were kept for indicated time periods in a 96-well tissue culture plate at 27°C . Samples were resuspended in diluted PBS to a final volume of 200 µl , and plated for colony counting . The number of colony forming units per milliliter ( cfu/ml ) for each plate was measured , using an average from three independent controls . The relative viability of each experimental sample ( done in biological triplicate ) was determined by dividing the cfu/ml for that sample by the average cfu/ml of the control plates . Worms were collected from liquid cultures and incubated at elevated temperatures for 4 , 8 , 12 or 16 hr . After each time point , worms were allowed to cool down at room temperature and survival rate was calculated after counting the survivors . S . cerevisiae strains were grown to stationary phase ( 72 hr ) in YPD medium , after which cells were collected by centrifugation and washed twice with water . Subsequently , two different heat-stress survival assays were performed . In the first one , cells were resuspended at decreasing cell densities , starting at an OD600 of 1 . 0 and then serially diluted ( 1:10 ) up to an OD600 of 0 . 001 . These were subjected to severe heat shock at 50°C for 30 min . 5 µl from each of these samples were spotted onto YPD plates . Cells were allowed to recover for ~30 hr before imaging the plates , and estimating survival . In the second assay , cells were resuspended at a single cell density ( OD600 of 0 . 1 ) , and subjected to heat stress for 45 , 60 and 75 min . 5 µl of each suspension was spotted onto YPD plates and cells were allowed to recover for ~30 hr before imaging and estimating survival . WT or GS-deficient S . cerevisiae strains were grown to stationary phase ( 72 hr ) in YPD medium , after which cells were collected by centrifugation and washed twice with water . Subsequently cells were resuspended at at an OD600 of 0 . 1 . These were subjected to rapid freezing , followed by thawing at room temperature , for multiple cycles . 5 µl from each of these samples were spotted onto YPD plates . Cells were allowed to recover for the indicated times before imaging the plates , and estimating survival . Worms were collected in 1 ml distilled water and homogenized by freezing in liquid nitrogen and subsequent thawing in a sonication bath for 5 times . The debris was pelleted by centrifugation at 25 , 000 g for 1 min at 4°C . A micro BCA assay kit ( Thermo Fisher Scientific , Germany ) was used to determine total soluble protein amounts from the supernatant . Next , the pellet was resuspended and the homogenate was extracted according to Bligh and Dyer’s method ( Bligh and Dyer , 1959 ) . Briefly , homogenized sample in 1 ml water was mixed with 3 . 75 ml of chloroform–methanol ( 1:2 , v/v ) in glass tubes for at least 20 min . Then 1 . 25 ml of chloroform and 1 . 25 ml of water were added sequentially , with rigorous mixing after each addition . Phase separation was facilitated by centrifugation at 1 , 000 g for 15 min . Next , organic ( lower ) and aqueous ( upper ) phases were collected into fresh glass tubes using sterilized glass Pasteur pipettes . Organic fractions from radioactively labeled samples and all aqueous fractions were dried under vacuum with heating . Organic fractions from non-labeled samples were dried under nitrogen gas flow . All organic and aqueous fractions were dissolved in chloroform–methanol ( 2:1 , v/v ) and methanol–water ( 1:1 , v/v ) , respectively . Non-labeled samples were normalized according to total soluble protein amounts measured from homogenates . For each mg of protein , organic and aqueous fractions were dissolved in 166 µl and 332 µl of the corresponding solvent , respectively . Labeled organic samples were dissolved in 100 µl of the corresponding solvent and total radioactivity in each sample was measured by a scintillation counter . After sample homogenization and protein measurement prior to organic extraction , trehalose measurement was performed in some samples using a trehalose assay kit with a modified protocol ( Megazyme , Ireland ) . First , 40 µl of each homogenate supernatant was heated at 95°C for 5 min to inactivate endogenous enzymes . Next , reducing sugars in the homogenate were reduced to sugar alcohols by adding 40 µl of freshly prepared alkaline borohydride ( 10 mg/ml sodium borohydride in 50 mM sodium hydroxide ) into each tube and incubating at 40°C for 30 min with shaking at 300 rpm . Then , the mixture was neutralized by adding 100 µl of 200 mM acetic acid . Subsequently , the pH was adjusted by adding 40 µl of 2 M imidazole buffer ( pH 7 . 0 ) . 70 µl of the final mixture was transferred to a plastic cuvette and the reaction mixture was added ( 70 µl of 2 M imidazole buffer , 35 µl of NADP+/ATP mix , 7 µl of hexokinase/glucose 6-phosphate dehydrogenase mix and 700 µl of distilled water ) . The reaction was carried out at room temperature for 15 min . Then the basal absorbance at 340 nm was measured ( A1 ) . After that , 7 µl of trehalase was added and incubated for 15 min at room temperature before the final absorbance at 340 nm was measured ( A2 ) . Trehalose concentration was calculated from the difference of absorbance values and normalized to the protein amounts measured from the same samples . This experiment was carried out on 3 different days with 3 technical replicates on each day for each treatment . Median values of technical replicates were used for calculations . Trehalose and glycogen from yeast samples were quantified as described previously , with minor modifications ( Shi et al . , 2010 ) . Cell samples were collected and pelleted . Cell pellets were quickly washed with 1 ml of ice-cold water and then resuspended in 0 . 25 ml of 0 . 25 M sodium carbonate and stored at -80°C until processed . For batch cultures , 20 OD600 total cells were collected . After resuspension in water , 0 . 5 ml of cell suspension was transferred to two capped Eppendorf tubes ( one tube for glycogen assay and the other tube for trehalose assay ) . When sample collections were complete , cell samples ( in 0 . 25 M sodium carbonate ) were boiled at 95–98°C for 4 hr , and then 0 . 15 ml of 1 M acetic acid and 0 . 6 ml of 0 . 2 M sodium acetate were added into each sample . Each sample was incubated overnight with 1 U/ml amyloglucosidase ( Sigma-Aldrich , India ) rotating at 57°C for the glycogen assay , or 0 . 025 U/ml trehalase ( Sigma-Aldrich , India ) at 37°C for the trehalose assay . Samples were then assayed for glucose using a glucose assay kit ( Sigma-Aldrich , India ) . Glucose assays were done using a 96-well plate format . Samples were added into each well with appropriate dilution within the dynamic range of the assay ( 20–80 µg/ml glucose ) . The total volume of sample ( with or without dilution ) in each well was 40 µl . The plate was pre-incubated at 37°C for 5 min , and then 80 µl of the assay reagent from the kit was added into each well to start the colorimetric reaction . After 30 min of incubation at 37°C , 80 µl of 12 N sulfuric acid was added to stop the reaction . Absorbance at 540 nm was determined to assess the quantity of glucose liberated from either glycogen or trehalose . TPS activity assay was based on Dmitryjuk et al . ( 2014 ) . Worm homogenates were prepared in 1 ml of 0 . 9% NaCl ( w/v ) . Total soluble protein and the initial amount of trehalose were measured as described above . Reaction was carried out with 100 µl of the lysate in 40 mM acid-ammonia buffer ( pH 4 . 2 ) , 2 mM MgCl2 , 0 . 2 mM UDP-glucose , 0 . 2 mM glucose 6-phosphate and 80 µM trehalase inhibitor at 37°C for 30 min in a total volume of 0 . 5 ml . Subsequently , the resulting trehalose 6-phosphate was dephosphorylated with 1 U of alkaline phosphatase in 100 mM phosphoric buffer ( pH 8 ) at 37°C for 30 min . Reaction was stopped via heating the samples to 95°C for 5 min . Next , the final amount of trehalose was measured and the difference from initial level of trehalose was calculated . This difference was then normalized to the total soluble protein amount . The unit enzyme activity is defined as the normalized molar amount of trehalose 6-phosphate produced in 1 min . A 3X-FLAG epitope tag was added to the carboxy termini of Tps1 and Tps2 at the endogenous chromosomal locus in the indicated strains . Tps-FLAG containing wild type and GS-deficient strains were grown to stationary phase ( 72 hrs ) in YPD medium , cells were harvested by centrifugation , and proteins were extracted by first precipitating with 10% trichloroacetic acid ( TCA ) , followed by removal of TCA , and solubilization of the protein extracts in SDS-glycerol sample buffer , normalizing for total protein . Proteins were separated on an SDS-PAGE gel and detected with standard immunoblotting for the FLAG epitope , using a mouse anti-FLAG antibody , and HRP-conjugated rabbit anti-mouse IgG secondary antibody . High-performance thin-layer chromatography ( HPTLC ) was used to separate and visualize molecules of interest . The TLC system was developed using non-radioactive amino acid and sugar standards , based on Tweeddale et al . ( 1998 ) . Individual amino acid samples were first separated on 1 dimension for either mobile phases , visualized by ninhydrin staining , and their corresponding Rf values were calculated . Then they were mixed and separated on 2 dimensions . Individual Rf values calculated from the 1D TLC runs coincided largely with each molecule in question also on 2D . Furthermore , the positions of glutamate and glutamine were confirmed in another set of experiments , where glutamate- or glutamine-lacking mixtures of amino acids were separated on 2D . Localization of sugars on the 2D TLC system was done similarly , only using Molisch staining as the visualization method . Before any analysis , sample normalization following organic extraction was confirmed by loading 5 µl of each organic fraction on an HPTLC plate ( Merck , Germany ) , eluting with chloroform–methanol–water ( 45:18:3 , v/v/v ) and visualizing with copper acetate solution ( 3% copper acetate and 8% ortho-phosphoric acid in water ) after baking at 200°C . It was expected that the phospholipids should have comparable levels in every sample . Sugars were separated using chloroform–methanol–water ( 4:4:1 , v/v/v ) as the mobile phase and visualized with Molisch’s reagent ( 3% 1-naphtol in sulfuric acid ( 96% ) –water–ethanol ( 13:8:80 , v/v/v ) ) after baking at 180°C . For comparison of trehalose levels of dauer larvae before and after preconditioning , 8 µl of aqueous fraction was used . Triacylglycerol levels were compared by running 4 µl of the organic fraction with the solvent system petroleum ether ( b . p . 60–80°C ) –diethyl ether–glacial acetic acid ( 82:18:1 , v/v/v ) and visualizing with copper acetate . We used the 2-dimensional TLC approach to compare the metabolites in aqueous fractions of radioactively labeled worm extracts . An amount of aqueous fraction equivalent to 2700 worms ( L3 or dauer larvae ) was applied as a spot to an HPTLC plate and eluted on the first dimension with 1-propanol–methanol–ammonia ( 32% ) –water ( 28:8:7:7 , v/v/v/v ) . Then , the plate was dried for 15 min and eluted on the second dimension with 1-butanol–acetone–glacial acetic acid–water ( 35:35:7:23 , v/v/v/v ) . Finally , it was sprayed with EN3HANCE spray surface autoradiography enhancer ( Perkin Elmer , Waltham , MA ) and exposed on an X-ray film , which was developed by standard methods . Differences in the amounts of radioactively labeled trehalose were calculated via densitometry . For this purpose , TLCs were exposed to X-ray films for a shorter period ( 2 hr ) , which was optimized to prevent the saturation of spots . Subsequently , films were developed and scanned , after which trehalose spot intensities were calculated using Fiji ( fiji . sc ) software . daf-2:icl-1::GFP L3 and dauer larvae were grown in liquid culture for 3 days at 15 and 25°C , respectively . Subsequently , they were pelleted in a 15 ml tube and resuspended in 100 µl of the original culture medium still containing bacteria . MitoTracker Red CMXRos ( Thermo Fisher Scientific , Germany ) stock solution ( 1 mM in DMSO ) was diluted into 10 µM in the worm-bacteria suspension . Our previous experience shows that worms can tolerate up to 1% DMSO in the medium , therefore 1:100 dilution of the MitoTracker solution is the highest concentration that can be safely achieved . Worms were incubated in this solution for 1 . 5 hr at room temperature in dark . Next , the excess dye was washed off with M9 buffer and worms were resuspended again in 100 µl . They were incubated in M9 for 30 min so that they could defecate the excess dye in the gut . Finally , L3 and dauer larvae were anesthetized in 20 mM and 50 mM sodium azide , respectively . Meanwhile , agarose pads on microscope slides were prepared and 5 µl of worm suspension was transferred on them . They were immediately covered with a coverslip and sealed with nail polish . We used a Zeiss LSM 700 inverted laser scanning confocal microscope and a Zeiss LCI Plan-Neofluar 63×/1 . 3 Imm Corr DIC M27 objective to image mitochondria ( Zeiss , Germany ) . Simultaneously , GFP was excited at 488 nm and the emission below 550 nm was acquired by the first PMT while MitoTracker Red CMXRos was excited at 555 nm and the emission above 560 nm was acquired by the second PMT . Optical sections at 0 . 1×0 . 1×0 . 5 µm3 x-y-z resolution were collected in a 4D hyperstack . Final images were adjusted for intensity and merged in Fiji . No non-linear adjustments were done . Oxygen consumption of worms was measured with a Seahorse XFe96 system ( Seahorse Bioscience , North Billerica , MA ) . L3 and dauer larvae of daf-2 and daf-2;icl-1 strains were grown in liquid culture for 3 days . Then , they were collected and washed extensively to remove excess bacteria and debris . Approximately 100 worms were pipetted into each well of a 96-well Seahorse XFe assay plate , except for the 4 corner wells , which were used to estimate the background measurement . Initial oxygen consumption rate ( OCR ) was measured until the readings stabilized . Then , sodium azide with phenol red indicator was injected into half of the wells at a final concentration of 20 mM . The rest of the wells were injected only phenol red indicator in water . OCR was measured once again until it stabilized . Then 4 subsequent measurements were done in 6 . 5 min intervals . Finally , exact number of worms in each well were counted and used for normalization . A small number of abnormal readings were also filtered out at this stage . On average , 7–8 wells ( technical replicates ) were used for each condition . Normalized OCR values were averaged for the last 4 measurements for each strain , stage and injection . OCR after water injection was named as total OCR ( tOCR ) and OCR after azide injection was named as non-mitochondrial OCR ( nmOCR ) . The difference of tOCR and nmOCR was calculated for each time point ( measurement ) and named as mitochondrial OCR ( mOCR ) or respiration rate . Biochemical pathway analysis was done by querying the KEGG database for C . elegans proteins using NemaPath software ( www . nematode . net ) ( Wylie et al . , 2008 ) . Data acquired from these queries were cross referenced to our previously published microarray data ( Erkut et al . , 2013 ) . Sequences of worm proteins were obtained from WormBase ( www . wormbase . org ) . ICL-1 sequences were submitted to MitoProt ( ihg . gsf . de/ihg/mitoprot . html ) to predict the probability of mitochondrial import of ICL-1 ( Claros and Vincens , 1996 ) . Signal sequence for FAR-3 was predicted with SignalP 4 . 0 ( www . cbs . dtu . dk/services/SignalP-4 . 0 ) ( Petersen et al . , 2011 ) . All statistical analyses were done in R environment ( www . r-project . org ) . Trehalose and glycogen levels , as well as OCRs were compared with analysis of variance ( ANOVA ) followed by Tukey’s honestly significant differences ( HSD ) post-hoc test . Trehalose/glycogen amounts were log-transformed prior to model fit , normality was confirmed with QQ-plots and Shapiro-Wilk test , homoscedasticity with Levene’s test . Survival rates after desiccation and rehydration were compared with beta regression as described before ( Erkut et al . , 2013 ) , followed by Type II analysis of deviance for generalized linear models . Prior to beta regression , fit to beta distribution was confirmed with QQ-plots . Statistical power was calculated via power analysis when possible . The maximum Type I error rate was set as α = 0 . 05 for all tests . Data are presented as mean ± standard error for C . elegans trehalose levels and survival rates , and mean ± 95% confidence limit for other measurements unless stated otherwise .	Many organisms can survive losing all the water from their body in periods of severe drought by suspending their life . This ability is called anhydrobiosis ( from the Greek for ‘life without water’ ) . When the desiccated organisms encounter water again , they resume life as normal . Two organisms commonly used in research , a roundworm called Caenorhabditis elegans and a yeast called Saccharomyces cerevisiae , are anhydrobiotes . To survive without water , anhydrobiotes alter the chemical reactions that sustain their life , and so change their metabolic state . The organisms also produce molecules that preserve the structure of their cells . One such essential molecule is a sugar called trehalose . However , both worms and yeast can only enter anhydrobiosis during particular stages of life where they do not eat . So where does the trehalose come from ? Erkut et al . have now addressed this question by studying the metabolism of C . elegans and S . cerevisiae as these species entered anhydrobiosis . The experiments revealed that while preparing for desiccation , both species change their metabolism to favor creating sugars rather than releasing energy . In this process , the worms and yeast use a biochemical pathway called the glyoxylate shunt , which can convert fat or acetic acid into sugar . Genetic mutations that deactivate this pathway severely reduce the ability of both organisms to produce trehalose and tolerate desiccation . From these findings , Erkut et al . conclude that the source of trehalose in non-feeding worms is their fat deposits , while in yeast it is acetate: a molecule that is derived from ethanol , the end-product of the fermentation process . The glyoxylate shunt had been thought only to be a non-essential biochemical shortcut of another well-known metabolic pathway called the Krebs cycle . Now that Erkut et al . have shown that the glyoxylate shunt has its own specific biological role , further investigation is needed to understand how it is activated to act as a metabolic switch . The molecules that regulate similar metabolic transitions will also need to be identified in future studies . Ultimately , understanding these processes could present new ways of diagnosing and treating metabolic diseases such as diabetes and cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2016	The glyoxylate shunt is essential for desiccation tolerance in C. elegans and budding yeast
The Maternal Embryonic Leucine Zipper Kinase ( MELK ) has been reported to be a genetic dependency in several cancer types . MELK RNAi and small-molecule inhibitors of MELK block the proliferation of various cancer cell lines , and MELK knockdown has been described as particularly effective against the highly-aggressive basal/triple-negative subtype of breast cancer . Based on these preclinical results , the MELK inhibitor OTS167 is currently being tested as a novel chemotherapy agent in several clinical trials . Here , we report that mutagenizing MELK with CRISPR/Cas9 has no effect on the fitness of basal breast cancer cell lines or cell lines from six other cancer types . Cells that harbor null mutations in MELK exhibit wild-type doubling times , cytokinesis , and anchorage-independent growth . Furthermore , MELK-knockout lines remain sensitive to OTS167 , suggesting that this drug blocks cell division through an off-target mechanism . In total , our results undermine the rationale for a series of current clinical trials and provide an experimental approach for the use of CRISPR/Cas9 in preclinical target validation that can be broadly applied . Tumors of the breast can be divided into five distinct subtypes based on characteristic gene expression patterns . These breast cancer subtypes are referred to as Luminal A , Luminal B , Her2-enriched , normal-like , and basal ( Sørlie et al . , 2001 ) . Basal breast cancers ( BBCs ) comprise ~15% of all diagnosed breast cancers and express genes typically found in the basal/myoepithelial layer of the mammary gland ( Badve et al . , 2011; Rakha et al . , 2008 ) . BBCs are most frequently diagnosed in younger patients and present with advanced histologic grade , central necrosis , and high mitotic activity . Additionally , ~70% of basal breast cancers fail to express the estrogen receptor ( ER ) , the progesterone receptor ( PR ) , or the human epidermal growth factor receptor 2 ( HER2 ) ( Badve et al . , 2011 ) . Tumors that lack expression of ER , PR , or HER2 are referred to as ‘triple-negative’ breast cancers , and are irresponsive to hormonal or anti-HER2 therapies that have proven effective against receptor-positive cancers . Due to their resistance to targeted therapies as well as their rapid rate of cell division , basal breast cancers currently have the worst prognosis of any breast cancer subtype . Thus , there is an urgent need to develop new therapies that are effective against triple-negative or basal-type tumors . In recent years , significant progress has been made in the treatment of certain malignancies by targeting cancer cell ‘addictions’ , or genetic dependencies that encode proteins required for the growth of specific cancer types ( Luo et al . , 2009 ) . Drugs that block the function of a cancer dependency – like the antibody Herceptin in Her2+ breast cancer – can trigger apoptosis and durable tumor regression ( Weinstein , 2002 ) . Cancer cell addictions are often investigated through the use of different transgenic technologies to disrupt the expression of a specified gene . Two of the most popular methodologies are RNA interference , which destabilizes a targeted transcript , and CRISPR mutagenesis , which utilizes the nuclease Cas9 to induce frameshift mutations at a targeted locus . While CRISPR-mediated genetic engineering has been widely adopted since its discovery in 2013 , RNA interference remains popular due to its ability to deplete multiple isoforms of a protein , its reversibility , and its relative insensitivity to gene copy number ( Boettcher and McManus , 2015 ) . Moreover , the partial loss-of-function phenotype generated by RNAi may more accurately recapitulate the effects of drug treatment than the complete loss-of-function phenotype generated by a Cas9-induced frameshift mutation . Nonetheless , RNAi constructs exhibit limited specificity , and off-target knockdowns are an inherent and widespread problem in RNAi experiments ( Jackson et al . , 2003 , 2006; Singh et al . , 2011 ) . The Maternal Embryonic Leucine Zipper Kinase ( MELK ) has received substantial attention as a potential cancer cell addiction and promising target for drug development . MELK was first identified as an AMPK family member expressed in the mouse pre-implantation embryo ( Heyer et al . , 1997 ) , and has since been implicated in several cellular processes , including apoptosis ( Jung et al . , 2008 ) , splicing ( Vulsteke et al . , 2004 ) , and neurogenesis ( Nakano et al . , 2005 ) . MELK is also over-expressed in most types of solid tumors , including breast , colon , liver , lung , melanoma , and ovarian cancer ( Gray et al . , 2005 ) . Furthermore , many publications have reported that knocking down MELK using RNAi inhibited the proliferation of cell lines derived from these cancer types ( Gray et al . , 2005; Lin et al . , 2007; Kuner et al . , 2013; Du et al . , 2014; Kig et al . , 2013; Speers et al . , 2016; Alachkar et al . , 2014; Marie et al . , 2008; Nakano et al . , 2008; Hebbard et al . , 2010; Wang et al . , 2014; Choi et al . , 2011; Xia et al . , 2016; Gu et al . , 2013 ) . In particular , MELK has been identified as a key driver of basal-type breast cancer , suggesting a novel therapeutic approach to treat this disease ( Wang et al . , 2014 ) . In response to the widespread reports that MELK is a cancer dependency , several companies have developed small molecule inhibitors of MELK that block the activity of the kinase in vitro and that inhibit cancer cell proliferation at micromolar or nanomolar concentrations ( Beke et al . , 2015; Touré et al . , 2016; Johnson et al . , 2015a , 2015b; Chung et al . , 2012 ) . Additionally , four clinical trials have been launched to test the MELK inhibitor OTS167 in human cancers ( NCT01910545 , NCT02768519 , NCT02795520 , and NCT02926690 ) . As part of a project in our lab to characterize genes whose expression is associated with patient prognosis in cancer ( Sheltzer , 2013 ) , we identified MELK as highly-expressed in deadly tumors from multiple cancer types ( data not shown ) . We set out to use CRISPR/Cas9 to characterize the effects of MELK loss on tumorigenesis . Unexpectedly , we found that mutating MELK failed to affect the growth of every cancer cell line that we tested . Furthermore , the MELK inhibitor OTS167 remained effective against cells with null mutations in MELK , suggesting that its in vivo activity results from an off-target effect . We propose that CRISPR represents an essential modality to confirm putative cancer dependencies and drug specificity before preclinical findings are advanced to human trials . Over a dozen previous publications have reported that MELK is a cancer dependency , as blocking MELK with RNAi or small molecules inhibited the proliferation of cell lines derived from multiple tumor types ( Gray et al . , 2005; Lin et al . , 2007; Kuner et al . , 2013; Du et al . , 2014; Kig et al . , 2013; Speers et al . , 2016; Alachkar et al . , 2014; Marie et al . , 2008; Nakano et al . , 2008; Hebbard et al . , 2010; Wang et al . , 2014; Choi et al . , 2011; Xia et al . , 2016; Gu et al . , 2013 ) . However , several discrepancies exist in the literature on MELK . For instance , various publications disagree over the cell cycle stage affected by MELK inhibition ( Du et al . , 2014; Kig et al . , 2013; Alachkar et al . , 2014; Wang et al . , 2014; Beke et al . , 2015 ) , while other publications disagree over whether receptor-positive breast cancer cell lines are sensitive ( Lin et al . , 2007; Beke et al . , 2015; Chung et al . , 2012 ) or resistant ( Wang et al . , 2014 ) to MELK inhibition . To unambiguously determine the effects of MELK loss in cancer cell lines , we applied CRISPR/Cas9 to generate frameshift mutations in the MELK coding sequence . We designed seven guide RNAs ( gRNAs ) against MELK , five of which target the N-terminal kinase domain and two of which target the C-terminal kinase-associated domain ( Figure 1A and Supplementary file 1 ) . Then , we cloned each guide RNA into a GFP-expressing vector and transduced the guides into three Cas9-expressing cell lines: the triple-negative breast cancer cell lines Cal51 and MDA-MB-231 , reported to be addicted to MELK expression ( Wang et al . , 2014 ) , and the melanoma cell line A375 , from a cancer type that over-expresses MELK ( Gray et al . , 2005; Ryu et al . , 2007 ) . As negative controls in these assays , we also cloned and transduced three gRNA’s that target the non-essential and non-coding Rosa26 locus ( Supplementary file 1 ) . 10 . 7554/eLife . 24179 . 003Figure 1 . Mutation of the MELK kinase domain does not affect cancer cell proliferation or anchorage-independent growth . ( A ) Domain structure of MELK and locations of the sequences targeted by 7 MELK gRNAs . ( B ) Genomic DNA was purified from the indicated population of MDA-MB-231 cells and the targeted loci were amplified by PCR . Percent indel formation was estimated using TIDE analysis . The highlighted region indicates 20 nucleotides or 15 nucleotides of the sequence recognized by the guide RNA . Other sequence traces are presented in Figure 1—figure supplements 1–3 . ( C ) Western blot analysis of GFP+ MDA-MB-231 cells using the Abcam ab108529 MELK antibody . Alpha-tubulin levels were analyzed as a loading control . ( D–F ) Proliferation and doubling time analysis of A375 , Cal51 , and MDA-MB-231 cell lines transduced with 3 Rosa26 gRNAs or with 7 MELK gRNAs . ( G ) Images of colonies from the indicated Cal51 strains grown in soft agar . ( H–I ) Quantification of anchorage-independent growth in Cal51 or A375 cells transduced with the indicated gRNA . For each assay , colonies were counted in at least 15 fields under a 10x objective . Boxes represent the 25th , 50th , and 75th percentiles of colonies per field , while the whiskers represent the 10th and 90th percentiles . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 00310 . 7554/eLife . 24179 . 004Figure 1—figure supplement 1 . Mutation of MELK using seven different guide RNAs in the A375 melanoma cell line . Following sorting of GFP+ populations , genomic DNA was purified from each cell line and the targeted loci were amplified by PCR . The amplified fragments were sequenced using the forward and reverse PCR primers ( Supplementary file 2 ) , and indel formation was estimated using TIDE analysis . The highlighted region indicates 20 nucleotides or 15 nucleotides of the sequence recognized by the guide RNA . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 00410 . 7554/eLife . 24179 . 005Figure 1—figure supplement 2 . Mutation of MELK using seven different guide RNAs in the Cal51 triple-negative breast cancer cell line . Following sorting of GFP+ populations , genomic DNA was purified from each cell line and the targeted loci were amplified by PCR . The amplified fragments were sequenced using the forward and reverse PCR primers ( Supplementary file 2 ) , and indel formation was estimated using TIDE analysis . The highlighted region indicates 20 nucleotides or 15 nucleotides of the sequence recognized by the guide RNA . The heterogeneity in sequence reads at MELK g1 and MELK g3 in the Rosa26 gRNA line was observed in multiple samples , and likely represent polymorphisms present in the Cal51 population . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 00510 . 7554/eLife . 24179 . 006Figure 1—figure supplement 3 . Mutation of MELK using seven different guide RNAs in the MDA-MB-231 triple-negative breast cancer cell line . Following sorting of GFP+ populations , genomic DNA was purified from each cell line and the targeted loci were amplified by PCR . The amplified fragments were sequenced using the forward and reverse PCR primers ( Supplementary file 2 ) , and indel formation was estimated using TIDE analysis . The highlighted region indicates 20 nucleotides or 15 nucleotides of the sequence recognized by the guide RNA . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 00610 . 7554/eLife . 24179 . 007Figure 1—figure supplement 4 . Western blot analysis of MELK-disrupted cell populations . Following sorting of GFP+ populations , whole-cell lysate was collected and analyzed for MELK expression using the Abcam ab108529 antibody . Alpha-tubulin was analyzed as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 007 To assess the efficacy of our CRISPR system , we purified GFP+ populations from cell lines harboring each individual guide RNA , and then we analyzed the targeted loci by Sanger sequencing . TIDE analysis , which decomposes raw sequencing traces into linear combinations of indel mutations ( Brinkman et al . , 2014 ) , revealed high cutting efficiency at most targeted loci ( Figure 1B and Figure 1—figure supplements 1–3 ) . Across the 21 samples , the median level of indel formation was 80% . These values likely underestimate the true mutation frequency , as TIDE analysis is not able to detect missense mutations and large indels will not be efficiently amplified by PCR . Western blot analysis of MELK protein levels in A375 , Cal51 , and MDA-MB-231 sorted populations further confirmed that our CRISPR system effectively ablated MELK expression ( Figure 1C and Figure 1—figure supplement 4 ) . We next set out to determine whether MELK was required for cancer cell fitness . To test this , we measured cell proliferation over 15 days in culture in the 30 independent lines of A375 , Cal51 , and MDA-MB-231 that we had generated ( Figure 1D–F ) . Surprisingly , we failed to detect any difference in proliferative capacity between the cell lines with wild-type or mutant MELK . For instance , in the A375 cell line , we calculated a mean doubling time of 16 . 9 hr among cells transduced with Rosa26 guide RNAs , while the cell lines transduced with MELK gRNAs exhibited a mean doubling time of 16 . 8 hr . MELK gRNA-transduced cell lines also exhibited wild-type levels of growth under anchorage-independent conditions ( Figure 1G–I ) . These results call into question the notion that MELK is a genetic dependency either across cancer types or in triple-negative breast cancers . In order to assess whether a wider range of cancer cell lines were dependent on MELK for viability , we performed individual GFP dropout experiments in 13 Cas9-expressing cancer cell lines ( Shi et al . , 2015 ) . In these assays , cancer cells are transduced with GFP-expressing guide RNA vectors at low MOI to create mixed populations of GFP+ and GFP- cells . A guide RNA that induces mutations in a gene required for cancer cell fitness will drop out from the population , resulting in a decreasing ratio of GFP+ to GFP- cells over time . Additionally , we considered it possible that cells had adapted to MELK loss during the time required to sort and expand pure GFP+ cell populations to perform the experiments described in Figure 1 . For the following dropout assays , we monitored GFP levels directly following introduction of the gRNA virus , without selecting or expanding cell populations . Importantly , this strategy allows us to detect whether the mutation of MELK results in a transient or immediate loss of cell fitness . As negative controls in this experiment , we utilized three gRNA’s targeting Rosa26 , and as positive controls we designed six gRNA’s targeting the essential replication genes RPA3 and PCNA ( Supplementary file 1 ) . We first transduced these gRNA’s individually into seven triple-negative breast cancer cell lines ( Cal51 , HCC1143 , HCC1937 , HCC70 , MDA-MB-231 , MDA-MB-453 , and MDA-MB-468 ) . Over the course of five passages in culture , gRNA’s targeting Rosa26 typically depleted 1 . 2 to 2-fold ( Figure 2A ) . This low level of depletion may result from off-target mutagenesis or from cell cycle arrest caused by repeated DNA breaks ( Aguirre et al . , 2016 ) . Over the same period of time , gRNA’s targeting RPA3 and PCNA depleted 5-fold to 100-fold . These positive control guides exhibited varying degrees of dropout ( e . g . , compare RPA3 g1 and RPA3 g2 ) , which may result from variability in cutting efficiency or from functionally-important differences in the protein domains targeted by these guides . However , in every cell line tested , every single gRNA targeting RPA3 or PCNA dropped out to a greater degree than every single Rosa26 guide . In contrast to RPA3 and PCNA , the seven guides that targeted MELK typically depleted less than 2-fold . Across seven different gRNAs tested in seven different cell lines , we never observed a MELK guide deplete more than 2 . 5-fold . In six of the seven cell lines , a Rosa26 gRNA exhibited a higher level of depletion than every single MELK gRNA ( Figure 2B ) . We conclude that these seven triple-negative breast cancer cell lines are not dependent on MELK for cell fitness . 10 . 7554/eLife . 24179 . 008Figure 2 . Guide RNAs targeting MELK fail to drop out in triple-negative breast cancer cell line competition experiments . ( A ) The fold change in the percentage of GFP+ cells , relative to the percentage of GFP+ cells at passage 1 , is displayed for seven triple-negative breast cancer cell lines . ( B ) A table summarizing the results presented in ( A ) is displayed . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 00810 . 7554/eLife . 24179 . 009Figure 2—figure supplement 1 . Guide RNAs targeting MELK fail to drop out in several cancer cell lines . ( A ) The fold change in the percentage of GFP+ cells , relative to the percentage of GFP+ cells at passage 1 , is displayed for six additional cancer cell lines . ( B ) A table summarizing the results presented in ( A ) is displayed . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 00910 . 7554/eLife . 24179 . 010Figure 2—figure supplement 2 . Unbiased screens do not identify MELK as a cancer dependency . Gene essentiality data were examined from a kinome-wide siRNA screen ( Campbell et al . , 2016 ) , a genome-wide shRNA screen ( Marcotte et al . , 2012; Hart et al . , 2014 ) , and two genome-wide CRISPR screens ( Hart et al . , 2015; Tzelepis et al . , 2016 ) . In the kinome-wide screen , a gene with a Z score < −2 was considered essential . In the Hart et al . genome-wide CRISPR screen , a gene with a Bayes Factor >5 was considered essential . In the genome-wide shRNA screen and the Tzelepis et al . CRISPR screen , a threshold of p<0 . 01 was used to identify essential genes . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 010 To extend these observations , we repeated the GFP dropout experiments in six Cas9-expressing cell lines ( A375 , Cama1 , HCT116 , NCI-H1299 , T24 , and U118-MG ) from other cancer types previously suggested to require MELK expression . Consistent with our observations in the triple-negative breast cancer cells , guides targeting Rosa26 exhibited minimal dropout over five passages , while guides targeting RPA3 or PCNA were depleted up to 90-fold ( Figure 2—figure supplement 1 ) . However , 0 of the 7 MELK guides exhibited significant dropout in any of the cell lines tested ( maximum dropout: 1 . 8-fold ) . Again , guides targeting Rosa26 exhibited an equivalent or occasionally greater degree of depletion than guides targeting MELK ( Figure 2—figure supplement 1B ) . In total , this data suggests that MELK is not a common cancer dependency . Several laboratories have conducted genome-wide or kinase-focused screens to identify novel cancer addictions . If these unbiased screens indicated that cancer cell lines required MELK expression to proliferate , then that would bolster the contention that MELK could be a therapeutic target in cancer . We therefore examined data from four recent screens: a kinome-wide siRNA screen in 117 cancer cell lines ( Campbell et al . , 2016 ) , a genome-wide CRISPR screen in 6 cell lines ( Hart et al . , 2015 ) , a genome-wide shRNA screen in 72 cancer cell lines ( Marcotte et al . , 2012; Hart et al . , 2014 ) , and a genome-wide CRISPR screen in seven cell lines ( Tzelepis et al . , 2016 ) . Large-scale unbiased screens are prone to experimental artifacts , and variations in protocol , technology , or the method of analysis can cause different screens to yield different results ( Mohr et al . , 2014 ) . Nonetheless , each of these screens identified multiple mitotic kinases as essential in various cancer cell lines , including Aurora B , BubR1 , CDK1 , and Plk1 ( Figure 2—figure supplement 2 ) . In contrast , MELK was not identified as essential in a single experiment . These negative results include 13 triple-negative breast cancer cell lines tested by Campbell et al . and 16 triple-negative breast cancer cell lines tested by the Moffatt lab . Several other published pan-cancer or breast cancer-focused screens have failed to identify MELK as either a general cancer dependency or a triple-negative breast cancer dependency ( Silva et al . , 2008; Marcotte et al . , 2016; Cowley et al . , 2014 ) . Thus , while we do not consider results from large-scale screens to be dispositive , we believe that these findings , coupled with our own experimental evidence , suggest that MELK expression is not required for cancer cell proliferation . If MELK is not a cancer cell dependency , then drugs that inhibit MELK must either be ineffective at stopping cancer cell division or they must also act on other cellular targets . We therefore assessed the efficacy of the MELK inhibitor OTS167 ( alternately called OTSSP167 ) , a therapeutic agent being tested in several clinical trials . We treated a variety of cancer cell lines with 7-point serial dilutions of OTS167 , and we observed that OTS167 did in fact impede cell proliferation at nanomolar concentrations ( mean GI50 = 16 nM; see below ) . As OTS167 was able to inhibit growth despite the non-essentially of MELK , we considered the possibility that OTS167 acted through an off-target effect . To test this , we set out to determine whether MELK expression was actually required for OTS167 sensitivity . We calculated the GI50 value of OTS167 in A375 , Cal51 , and MDA-MB-231 cells that harbored gRNA’s targeting either Rosa26 or MELK , and we found that cell populations with wild-type or mutant MELK displayed equivalent sensitivity to the drug ( Figure 3 ) . For instance , in Cal51 cells transduced with Rosa26 gRNAs , the GI50 values ranged from 9 nM to 12 nM ( mean: 10 nM ) , while in Cal51 cells transduced with MELK gRNAs , the GI50 values ranged from 8 nM to 14 nM ( mean: 11 nM ) . As OTS167 exhibits nanomolar potency against cancer cell lines but is unaffected by mutations in MELK , this suggests that OTS167 blocks proliferation by inhibiting another target or targets . 10 . 7554/eLife . 24179 . 011Figure 3 . Mutating MELK does not affect OTS167 sensitivity . ( A ) Summary of GI50 values from OTS167 treatment of A375 cells harboring guide RNAs targeting Rosa26 or MELK . ( B ) 7 point dose-response curves of OTS167 in the indicated A375 cell lines . ( C ) Summary of GI50 values from OTS167 treatment of Cal51 cells harboring guide RNAs targeting Rosa26 or MELK . ( D ) 7 point dose-response curves of OTS167 in the indicated Cal51 cell lines . ( E ) Summary of GI50 values from OTS167 treatment of MDA-MB-231 cells harboring guide RNAs targeting Rosa26 or MELK . ( F ) 7 point dose-response curves of OTS167 in the indicated MDA-MB-231 cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 01110 . 7554/eLife . 24179 . 012Figure 3—figure supplement 1 . Receptor-positive breast cancer cell lines are sensitive to OTS167 . ( A ) Summary of GI50 values from OTS167 treatment of various breast cancer cell lines . ( B and C ) 7 point dose-response curves of OTS167 in the indicated cell lines . Note that the summary values in ( A ) represent averages from 2 or three replicate experiments per cell line , while individual replicate experiments are displayed in ( B ) and ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 01210 . 7554/eLife . 24179 . 013Figure 3—figure supplement 2 . OTS167 treatment , but not MELK mutation , causes the accumulation of multinucleate cells . ( A–C ) Cells were either untreated or treated with the indicated drug for 24 hr . Subsequently , cells were stained with Hoechst dye and at least 200 cells for each condition were examined . ( D ) Representative images of Cal51 cells stained with Hoechst . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 013 To further explore this observation , we tested the efficacy of OTS167 against a panel of triple-negative or receptor-positive breast cancer cell lines . MELK is significantly up-regulated in triple-negative tumors relative to receptor-positive tumors ( Wang et al . , 2014 ) , and one clinical trial ( NCT02926690 ) includes a dosage-escalation study of OTS167 in patients with triple-negative cancers . However , we observed no significant difference between the GI50 values of OTS167 according to receptor status ( Figure 3—figure supplement 1 ) . In triple-negative breast cancer cell lines , OTS167 inhibited growth by 50% at concentrations ranging from 10 nM to 42 nM ( mean: 19 nM ) , while in receptor-positive breast cancer cells GI50 values ranged from 9 nM to 21 nM ( mean: 14 nM ) . These results demonstrate that OTS167 is not specifically effective against triple-negative breast cancer cell lines , but instead remains remarkably potent against breast cancer cell lines that express hormone receptors . Lastly , to confirm that OTS167 treatment fails to phenocopy MELK mutations , we examined their effects on cell cycle progression . We found that treatment with OTS167 blocked cytokinesis in a dose-dependent manner , resulting in populations harboring 13% to 60% multinucleate cells . In contrast , cells transduced with either Rosa26 or MELK gRNAs progressed through the cell cycle without gross mitotic defects , and exhibited no significant difference in the frequency of multinucleate cells ( Figure 3—figure supplement 2 ) . These results demonstrate that OTS167 induces a cell cycle failure phenotype that is not recapitulated by mutagenizing MELK . We note that this observation is consistent with a recent publication that reported that , at certain concentrations , OTS167 was capable of inhibiting the mitotic kinases Aurora B , Haspin , and Bub1 ( Ji et al . , 2016 ) . We conclude that the anti-proliferative effects of OTS167 are not a result of its inhibition of MELK . To unambiguously demonstrate that MELK is dispensable for the proliferation of certain cancer cell lines , we used CRISPR to generate clonal cell lines that lack MELK protein . We transduced the MDA-MB-231 triple-negative breast cancer cell line with sets of 2 guide RNAs and then expanded clonal populations from single cells ( Figure 4A ) . Recombination between the chosen gRNA cut sites eliminates exon 3 , which encodes residues that are essential for ATP binding ( Cao et al . , 2013; Cho et al . , 2014 ) , as well as parts of exons 2 , 4 , and/or 5 . We used PCR to identify three independent clones that were homozygous for CRISPR-induced recombination , and then confirmed the loss of the intervening genetic material by sequencing across the gRNA cut sites ( Figure 4B–C ) . Additionally , we derived clones of Cal51 that have been transduced with single MELK gRNAs , and then identified three clones that harbored indels in the MELK kinase domain ( Figure 4—figure supplement 1 ) . Western blot analysis of the MDA-MB-231 and Cal51 clones with two antibodies that recognize distinct regions of MELK further verified the complete lack of MELK expression in all six derived cell lines ( Figure 4D–E and Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 24179 . 014Figure 4 . MELK-knockout cell lines proliferate at normal rates and remain sensitive to OTS167 . ( A ) Schematic of exons in the MDA-MB-231 MELK-KO g1/g6 knockout line . Half-arrows indicate positions of either cut-site or deletion-spanning primers used to screen these colonies . Primer sequences are presented in Supplementary file 2 . ( B ) PCR validation of 3 independent MELK-KO clones . Note that amplification of the MELK-KO g3/g5 DNA with deletion-spanning primers yielded deletion products of at least two distinct sizes . ( C ) Sanger sequence validation of 3 independent MELK-KO clones . While MELK-KO g1/g6 and g1/g5 harbor a single homozygous deletion , MELK-KO g3/g5 harbors at least two distinct deletions . ( D ) Western blot analysis of MELK-KO clones using an antibody that recognizes a region in the N-terminal kinase domain ( Abcam ab108529 ) . ( E ) Western blot analysis of MELK-KO clones using an antibody that recognizes a region in the C-terminal domain ( Cell Signal 2274S ) . ( F ) Proliferation analysis and doubling time measurements of MELK-KO cell lines . ( G ) Representative images of Rosa26 gRNA or MELK-KO clones either untreated or treated with 100 nM OTS167 and then stained with Hoechst dye . ( H ) The indicated cell lines were either left untreated or were treated with the cytokinesis inhibitor cytochalasin B or with OTS167 . Cells were then stained with Hoechst dye . For each experiment , at least 200 cells were counted . ( I ) Summary of GI50 values from OTS167 treatment of either MDA-MB-231 Rosa26 gRNA or MELK-KO clones . ( J ) 7 point dose-response curves of OTS167 in the indicated cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 01410 . 7554/eLife . 24179 . 015Figure 4—figure supplement 1 . Generation and analysis of Cal51 MELK-KO cell lines . ( A ) Sanger sequencing of 3 independent Cal51 MELK-KO clones transduced with single gRNAs targeting the MELK kinase domain . The highlighted regions indicate bases recognized by the gRNA . ( B ) Western blot analysis of MELK-KO clones using an antibody that recognizes a region in the N-terminal kinase domain ( Abcam ab108529 ) . ( C ) Western blot analysis of MELK-KO clones using an antibody that recognizes a region in the C-terminal domain ( Cell Signal 2274S ) . ( D ) Proliferation analysis and doubling time measurements of MELK-KO cell lines . ( E ) Summary of GI50 values from OTS167 treatment of either Cal51 Rosa26 gRNA or MELK-KO clones . ( F ) 7 point dose-response curves of OTS167 in the indicated cell lines . ( G ) Representative images of Rosa26 gRNA or MELK-KO clones either untreated or treated with 100 nM OTS167 and then stained with Hoechst dye . ( H ) The indicated cell lines were either left untreated or were treated with the cytokinesis inhibitor cytochalasin B or with OTS167 . Cells were then stained with Hoechst dye . For each experiment , at least 200 cells were counted . DOI: http://dx . doi . org/10 . 7554/eLife . 24179 . 015 The MDA-MB-231and Cal51 MELK-KO clones exhibited robust proliferation , demonstrating that MELK is fully dispensable for the growth of these cancer cell lines ( Figure 4F and Figure 4—figure supplement 1 ) . In fact , one MDA-MB-231 MELK-KO clone exhibited a significantly shorter doubling time than the Rosa26 gRNA-transduced cell lines , potentially due to the presence of additional mutations that were acquired during clonal expansion . The MELK-KO clones progressed through the cell cycle without gross abnormalities and accumulated few multinucleate cells ( Figure 4G and Figure 4—figure supplement 1 ) . OTS167 treatment of MELK-KO clones caused the formation of multinucleate cells , demonstrating that this drug blocks cytokinesis by inhibiting another cellular target ( Figure 4H and Figure 4—figure supplement 1 ) . Finally , serial dilution analysis revealed that the MDA-MB-231 and Cal51 MELK-KO clones exhibited equivalent OTS167 GI50 values compared to Rosa26 gRNA-transduced lines ( Figure 4I–J and Figure 4—figure supplement 1 ) . We conclude that MELK is not an absolute requirement for triple-negative breast cancer proliferation , and that OTS167 blocks growth in a MELK-independent manner . As a mitotic kinase highly expressed in many cancer types , MELK has been identified as a promising target for therapeutic intervention . However , through the use of CRISPR/Cas9-mediated mutagenesis , we have demonstrated that MELK is dispensable for growth in 13 out of 13 cancer cell lines tested , and that a MELK inhibitor currently in clinical trials blocks cell division by inhibiting another target . We believe that our results highlight the importance of using CRISPR/Cas9 technology to study and validate preclinical targets in cancer drug development . Previous research utilizing RNA interference to knock down MELK has indicated that MELK expression is required for cancer cell proliferation . However , a growing body of evidence has revealed that RNAi is prone to pervasive off-target effects . This problem is particularly challenging when RNAi is used to study putative cell cycle regulators , as multiple publications have reported that the cell cycle genes RAD51 and MAD2 are unusually sensitive to off-target RNAi inhibition ( Adamson et al . , 2012; Hübner et al . , 2010; Sigoillot et al . , 2012 ) . For instance , in a screen for genes whose depletion caused a bypass of the spindle assembly checkpoint , 34 of the top 34 candidate siRNA’s exhibited off-target down-regulation of Mad2 levels ( Sigoillot et al . , 2012 ) . Moreover , the expression of MELK is strongly cell-cycle regulated: MELK levels are typically low in G0/G1 , and peak in mitosis [ ( Badouel et al . , 2010 ) and our unpublished data] . A genetic or chemical treatment that induces a G1 arrest would therefore be predicted to down-regulate MELK , potentially confounding the analysis of knockdown efficiency . While Cas9 mutagenesis is also susceptible to off-target editing , to the best of our knowledge , the off-target loci affected by CRISPR are unlikely to substantially overlap with those that are affected by RNAi . Moreover , sequencing the locus targeted by Cas9 can provide an unbiased readout of mutagenesis efficiency that is not sensitive to cell state-dependent expression variability . Finally , unlike RNAi , CRISPR can be applied to generate clonal cell lines that harbor null mutations in a targeted gene . This technique bypasses the problems inherent in the analysis of mixed cell populations and partial loss-of-function phenotypes , and can provide significant insight into the genetic architecture of cancer . One limitation of CRISPR mutagenesis is that , over the time required to generate or select for a pure cell population , cells may engage compensatory mechanisms to buffer against the loss of a targeted protein . Thus , the analysis of knockout clones can be complemented with cell-cell competition assays , which allow less time for cells to adapt to gene loss and may reveal the presence of a transient or immediate fitness defect induced by CRISPR . We performed a total of 91 competition assays ( 7 MELK gRNAs in 13 different cell lines ) that failed to reveal an effect of MELK loss on cell fitness , further strengthening our conclusion that MELK is dispensable for cancer cell proliferation . CRISPR mutagenesis can also assist in the pharmacological study of potential drugs . Several lines of evidence indicate that OTS167 does indeed inhibit MELK: for instance , a crystal structure of OTS167 binding to the MELK kinase domain has been reported ( Cho et al . , 2014 ) . However , these structural and biochemical studies are unable to conclusively demonstrate that a phenotype in a living cell is due to an on-target effect . We believe that CRISPR represents a useful tool to gain genetic insight into this question: if a CRISPR-induced null mutation of a putative drug target fails to confer resistance to that drug , then that drug must act through alternate targets or mechanisms . While the MELK-KO cell lines that we generated remain exquisitely sensitive to OTS167 , at present , we do not know how OTS167 blocks cell division . One possibility , not ruled out by our studies , is that OTS167 exhibits polypharmacology ( Knight et al . , 2010 ) , and kills cancer cells by inhibiting multiple kinases , potentially including MELK . The analysis of drug-resistant alleles of other mitotic kinases that OTS167 has been shown to inhibit ( Ji et al . , 2016 ) may shed further light on the in vivo MOA of this compound . Our results leave open the question of what role , if any , MELK plays in mammalian biology and cell cycle progression . While MELK is up-regulated in diverse tumor types , it is also expressed in several normal cell lineages , including embryonic cells , hematopoietic cells , and neural progenitor cells ( Heyer et al . , 1997; Nakano et al . , 2005; Gil et al . , 1997 ) . MELK may be required at a certain developmental stage , or for a specific cell type or organismal process . Similarly , we cannot currently rule out the possibility that MELK plays a role in tumorigenesis in vivo that was not assessed in our current work . At a minimum , our results suggest that MELK is dispensable for mitotic progression in most cancers . MELK may function in an overlapping or redundant pathway with other mitotic kinases , several of which are up-regulated along with MELK in tumor cells ( Malumbres and Barbacid , 2007 ) . Synthetic lethal screens and further in vivo investigation will shed light on MELK’s function in development and cancer . Nonetheless , our data suggest that specific MELK inhibitors are unlikely to be useful monoagents in cancer therapy . The identity of every human cell line utilized in this paper was authenticated using STR profiling ( University of Arizona Genetics Core , Tucson , AZ ) . Cell lines were also confirmed to be negative for mycoplasma contamination using the MycoAlert Detection Kit ( Lonza , Switzerland; LT07-218 ) . Cell lines HCC70 , HCC1937 , MDA-MB-453 , MDA-MB-468 , and ZR-75–1 were grown in RPMI 1640 supplemented with 10% fetal bovine ( FBS ) , 2 mM glutamine , 1% Nonessential Amino Acids ( Life Technologies , Waltham , MA; BY00148 ) , and 100 U/ml penicillin and streptomycin . HCC1143 and NCI-H1299 were grown in RPMI 1640 supplemented with 10% FBS , 2 mM glutamine , and 100 U/ml penicillin and streptomycin . T24 was grown in McCoy’s 5A media supplemented with 10% FBS , 2 mM glutamine , and 100 U/ml penicillin and streptomycin . Cal51 , A375 , MDA-MB-231 , HCT116 , Cama1 , JIMT-1 , and U118-MG cells were grown in DMEM supplemented with 10% FBS , 2 mM glutamine , and 100 U/ml penicillin and streptomycin . T47D cells were grown in RPMI supplemented with 10% FBS , 6 . 94 μg/ml insulin ( Thermo Fisher , Waltham , MA; BN00226 ) , 2 mM glutamine , and 100 U/ml penicillin and streptomycin . MCF7 cells were grown in DMEM supplemented with 10% FBS , 0 . 01 mg/ml insulin , 2 mM glutamine , and 100 U/ml penicillin and streptomycin . Cell lines were kindly provided by the individuals thanked in the acknowledgments . All cell lines were maintained in a humidified environment at 37°C and 5% CO2 . Cell counting was performed using the Cellometer Auto T4 system ( Nexcelom , Lawrence , MA ) . Guide RNAs targeting protein domains in MELK , PCNA , and RPA3 were designed with assistance from Osama El Demerdash ( manuscript in preparation ) . Oligonucleotides were ordered from IDT and then cloned into the LRG 2 . 1 vector [a gift from Jun-Wei Shi ( University of Pennsylvania ) and Chris Vakoc ( Cold Spring Harbor Laboratory ) ] using a BsmBI digestion ( Shalem et al . , 2014 ) . Plasmids were amplified in Stbl3 E . coli ( Thermo Fisher; C737303 ) prepared using the Mix and Go transformation kit ( Zymo Research , Irvine , CA; T3001 ) . HEK293T cells were transfected using the calcium-phosphate method ( Smale , 2010 ) . Supernatant was harvested 48 to 72 hr post-transfection , filtered through a 0 . 45 μm syringe , and then frozen at −80° C for later use or applied directly to cells with 4 μg/mL polybrene . The culture media on target plates was changed 24 hr post-transduction . To measure cell proliferation , 100 , 000 cells of each strain were plated on a six well plate and then allowed to grow for 72 hr . Cells were then trypsinized , counted , and 100 , 000 cells were re-plated in fresh media . Cells were passaged five times , and cumulative population doublings and doubling times were calculated at each passage . Solutions of 1 . 0% and 0 . 7% Difco Agar Noble in sterile water were autoclaved and then allowed to equilibrate at 42°C or 37 . 5°C , respectively . The 1% agar solution was then mixed 1:1 with 2 ml of the appropriate media , supplemented with 2X the concentration of serum , glutamine , and penicillin/streptomycin . 1 mL of this mixture was then plated in one well of a six well plate and allowed to solidify at room temperature for 30 min , resulting in a 0 . 5% agar base layer . Cells of interest were then trypsinized and re-suspended in their appropriate media . 20 , 000 cells were diluted in 1 ml of 2X media and then mixed with 1 ml of the 0 . 7% agar solution . 500 μl of this mixture was then plated on the solidified base layer , resulting in 5000 cells in a 0 . 375% agar suspension . The agar was allowed to solidify at room temperature for one hour before being transferred to an incubator . Fresh 1X media was added to the surface of each well 24 hr after plating , and then were re-fed every 3 days . After 21 days of growth , colonies were scored under 10x magnification . All experiments were plated in triplicate and performed twice . Cells were lysed with RIPA buffer [25 mM Tris , pH 7 . 4 , 150 mM NaCl , 1% Triton X 100 , 0 . 5% sodium deoxycholate , 0 . 1% sodium dodecyl sulfate , protease inhibitor cocktail ( Roche , Indianapolis , Indiana ) , phosphatase inhibitor cocktail ( Roche ) ] . Lysates were quantified using the Pierce BCA Kit ( Thermo Scientific ) , and equal amounts of protein were denatured and loaded onto an 8% SDS-PAGE gel . The protein was transferred onto a polyvinylidene difluoride membrane using the Trans-Blot Turbo Transfer System ( Bio-Rad , Hercules , California ) . The membrane was blocked in 5% non-fat milk-TBST and then incubated with anti-MELK ( abcam , Cambridge , MA; ab108529 ) at a 1:3000 dilution , or blocked in 10% BSA-TBST and then incubated with anti-MELK ( Cell Signal , Danvers , MA; 2274S ) at a 1:1000 dilution . Anti-alpha-tubulin ( Sigma-Aldrich , St . Louis , MO; T6199 ) was used as a loading control at a 1:3000 dilution . All primary antibody incubations were performed overnight at 4°C . Following incubation , the membranes were washed and then incubated in either anti-rabbit secondary ( abcam; ab6721 ) at 1:50000 for MELK or anti-mouse secondary ( Bio-Rad; 1706516 ) at 1:10000 for Tubulin for 1 hr at room temperature . Genomic DNA was extracted from transduced cell lines using the QIAmp DNA Mini kit ( Qiagen , Germantown , MD; Cat . No . 51304 ) . Loci targeted by guide RNAs were amplified using the primers listed in Supplementary file 2 , and then sequenced using the forward and reverse primers at the Cold Spring Harbor Laboratory sequencing facility . Sequence traces were analyzed using TIDE ( Brinkman et al . , 2014 ) . For every cell line of interest , 10 , 000 cells were plated in 100 μl of media in an 8 × 4 matrix on a flat-bottomed 96-well plate . Cells were allowed to attach for 24 hr , at which point the media in every well was changed . 500 nM of OTSSP167 ( MedChem Express , Monmouth Junction , NJ; Cat . No . HY-15512A ) was added to one row of cells , and then 6 3-fold serial dilutions were performed . After 72 hr of growth in the presence of the drug , cells were trypsinized and counted using a MacsQuant Analyzer 10 ( Milltenyi Biotec , Germany ) . The fraction of cells recovered at every drug concentration , relative to a row of untreated cells , was determined . GI50 values were calculated using a four-parameter inhibition vs . concentration model in Prism 7 ( Graphpad , San Diego , California ) . Sensitivity experiments in Figure 4J and Figure 3—figure supplement 1 were performed 2–3 times each , while sensitivity experiments in Figure 3 and Figure 4—figure supplement 1 were performed once . Cells were transduced on day 0 with sgRNA lentiviral supernatant , which was then replaced with fresh media on day 1 . On day 3 , the baseline percentage of GFP+ cells was measured using a MacsQuant Analyzer 10 ( Milltenyi Biotec ) . Cells were then passaged every 3 or 4 days , according to their growth rate and confluence , and the percentage of GFP+ cells was measured at every split . Dropout values represent the fold decrease in GFP+ cells at each passage , relative to the GFP+ percentage on day 3 . In preliminary experiments with A375 and MDA-MB-231 , replicate dropout assays were highly reproducible across independent replicates . For that reason , GFP dropout experiments in the 13 tested cell lines were performed once . 10 , 000 ( A375 , Cal51 ) or 20 , 000 ( MDA-MB-231 ) cells of interest were plated in 250 μl of media in a flat-bottomed 24-well plate and allowed to attach for 24 hr . Then the media was replaced , and OTS167 or Cytochalasin B ( Cayman Chemical Company , Ann Arbor , MI; Cat . No . 11328 ) were added to control wells . Following an additional 24 hr period of growth , cells were stained with 2 . 5 μg/ml of Hoechst dye ( Thermo Fisher , Cat . No . H3569 ) for 30 min and imaged using appropriate filters . DNA staining experiments were performed twice .	Like a person who is dependent on coffee to be productive , cancer cells are dependent on the products of certain genes in order to dominate their environment and grow . Cancer cells will stop growing and die when the activity of these gene products is blocked . These genes are known as cancer dependencies or “addictions” . As a result , researchers are constantly looking for cancer dependencies and developing drugs to block their activity . It was previously believed that a gene called MELK was an addiction in certain types of breast cancer . In fact , pharmaceutical companies had developed a drug to block the activity of MELK , and this drug is currently being tested in human patients . However , Lin , Giuliano et al . have now taken a second look at the role of MELK in breast cancer , and have come to a different conclusion . Using a gene editing technology called CRISPR/Cas9 , Lin , Giuliano et al . removed MELK activity from several cancer cell lines . This did not stop cancer cells from multiplying , suggesting that MELK is not actually a cancer addiction . Additionally , when breast cancer cells that do not produce MELK were exposed to the drug that is supposed to block MELK activity , the drug still stopped cell growth . Since the drug works when MELK is not present in the cell , the drug must be binding to other proteins . This suggests that MELK is not the actual target of the drug . Lin , Giuliano et al . suggest that , in the future , CRISPR/Cas9 technology could be used to better identify cancer dependencies and drug targets before cancer drugs are given to human patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "cancer", "biology" ]	2017	CRISPR/Cas9 mutagenesis invalidates a putative cancer dependency targeted in on-going clinical trials
The eukaryotic cell membrane is connected to a dense actin rich cortex . We present FCS and STED experiments showing that dense membrane bound actin networks have severe influence on lipid phase separation . A minimal actin cortex was bound to a supported lipid bilayer via biotinylated lipid streptavidin complexes ( pinning sites ) . In general , actin binding to ternary membranes prevented macroscopic liquid-ordered and liquid-disordered domain formation , even at low temperature . Instead , depending on the type of pinning lipid , an actin correlated multi-domain pattern was observed . FCS measurements revealed hindered diffusion of lipids in the presence of an actin network . To explain our experimental findings , a new simulation model is proposed , in which the membrane composition , the membrane curvature , and the actin pinning sites are all coupled . Our results reveal a mechanism how cells may prevent macroscopic demixing of their membrane components , while at the same time regulate the local membrane composition . The lateral heterogeneity of lipids and proteins in the plasma membrane of eukaryotic cells is an important feature for regulating biological function . The most prominent concept for membrane organization , the lipid raft theory , relates lipid phase separation ( driven by interactions between cholesterol , sphingolipids , and saturated phospholipids ) to membrane protein partitioning and regulation ( Simons and Ikonen , 1997; Simons et al . , 2011 ) . Consequently , understanding lipid phase separation in membranes is a topic of extreme interest . A convenient starting point is to envision the membrane as a two-dimensional ( 2D ) fluid environment through which the various membrane components freely diffuse . This simple picture successfully captures ternary model membranes containing two phospholipid species and cholesterol . At low temperature , these systems macroscopically phase separate into liquid-ordered ( Lo ) and liquid-disordered ( Ld ) domains ( Veatch and Keller , 2003 ) and the nature of the transition is consistent with that of a 2D fluid ( Honerkamp-Smith et al . , 2008 , 2009 ) . Similar behavior was observed in plasma membrane-derived vesicles ( Baumgart et al . , 2007; Sezgin et al . , 2012 ) . Despite these successes for model membranes , there is growing consensus that this simple picture needs to be refined for the plasma membrane . For example , a remaining puzzle is that the Lo/Ld domains observed in model membranes grow macroscopic in size ( micrometers ) , whereas lipid domains in plasma membrane are postulated to be tiny ( nanometers ) ( Lenne and Nicolas , 2009 ) . Additionally , the temperature Tc below which Lo/Ld domains start to form in plasma membrane derived vesicles is distinctly below T = 37°C ( Sezgin et al . , 2012 ) , and so its relevance at physiological temperature requires further justification . In the case of a free-standing membrane ( i . e . , in the absence of an actin cortex ) , experiments have shown that lipid domains at temperatures above Tc can be induced by crosslinking low abundant membrane constituents ( Hammond et al . , 2005; Lingwood et al . , 2008 ) . Furthermore , there are numerous theoretical proposals of how a finite domain size above Tc might be accounted for: vicinity of a critical point ( Honerkamp-Smith et al . , 2009 ) , hybrid lipids ( Palmieri and Safran , 2013 ) , coupling between composition and membrane curvature ( Schick , 2012 ) , electrostatic forces ( Liu et al . , 2005 ) . In addition to this , the cortical cytoskeleton has also been identified as a key player affecting membrane domain formation ( Kusumi et al . , 2005 ) . The latter is a dense fiber network of actin and spectrin on the cytoplasmic side of the eukaryotic plasma membrane . This network is connected to the membrane via pinning sites , such as lipid-binding proteins , transmembrane proteins , or membrane-attached proteins ( Janmey , 1998; Mangeat et al . , 1999; Janmey and Lindberg , 2004; Saarikangas et al . , 2010 ) . This has led to the hypothesis of the membrane being laterally compartmentalized: the pinning sites structure the membrane into small compartments whose perimeters are defined by the underlying actin network ( the so-called ‘picket-fence’ model ) . This picket-fence network then acts as a barrier to diffusion , which elegantly accounts for confined diffusion of lipids and proteins observed in a single molecule tracking experiments ( Kusumi et al . , 2005 ) . In a recent series of simulations , it was subsequently shown that a picket-fence network also acts as a barrier to macroscopic phase separation of lipids ( Ehrig et al . , 2011; Fischer and Vink , 2011; Machta et al . , 2011 ) . Instead , a stable mosaic of Lo and Ld domains is predicted , with a domain structure that strongly correlates to the actin fibers . Moreover , this mosaic structure already appears at physiological temperatures . These simulation findings are promising in view of the lipid raft hypothesis , since rafts are postulated to be small , as opposed to macroscopic . In this paper , we present the first experimental confirmation of these simulation results . To this end , we use an in vitro model system consisting of a supported lipid bilayer bound to an actin network . Complementing previous studies ( Liu and Fletcher , 2006; Subramaniam et al . , 2012; Heinemann et al . , 2013; Vogel et al . , 2013 ) , our system enables direct observation of Lo/Ld domain formation in the presence of a lipid bound actin network using superresolution STED microscopy ( Hell and Wichmann , 1994; Hell , 2007 ) and fluorescence correlation spectroscopy ( FCS ) ( Magde et al . , 1972; Kim et al . , 2007 ) . Based on our results , we propose an extension of the picket-fence model by including a coupling of the local membrane curvature to the membrane composition . Computer simulations of this extended model show that the pinning effect of the actin network is dramatically enhanced by such a coupling . These results imply that even a low density of pinning sites can induce significant structuring of lipids and proteins in the plasma membrane . To determine the lateral movement of lipids in the membrane in relation to the actin fibers , we applied scanning FCS ( Digman et al . , 2005; Ries and Schwille , 2006; Digman and Gratton , 2009 ) . We used the same system as described above , but now with a lower density of actin such that single fibers could be resolved by confocal microscopy . The temperature was set to T = 32°C , that is slightly above Tc of the actin-free membrane . As a control , we started with actin fibers on a single component DOPC membrane including 1 mol% DOPE-biotin ( compare Figure 1—figure supplement 2B ) . A small circle ( diameter d = 500 nm ) was scanned over a single actin fiber , as depicted in Figure 2E . For each position on the circle , the diffusion of the Ld-marker was determined by standard FCS analysis ( Kim et al . , 2007 ) . The upper panel in Figure 2A shows the intensity of the Ld-marker ( red ) and of the actin marker ( green ) along the scan circle for the single component membrane ( A ) . The intensity maximum of the green channel indicates the position of the actin fiber ( corresponding to 0 and 180° ) , the intensity of the red channel shows the corresponding distribution of the Ld marker . In the single component membrane , the Ld marker was homogenously distributed . The lower panels of Figure 2A shows the decay of the autocorrelation function of the Ld marker measured along the scan circle , with the color representing the amplitude of the correlation . For the single component membrane , the autocorrelation analysis revealed no significant differences in mobility between areas with actin and without . Next , we used the same data to determine the diffusion of lipids across the scanning circle by calculating the pair-correlation function of ‘opposing’ pixels on the scanning circle , that is pixel pairs that are separated by a rotation of 180° ( Cardarelli et al . , 2012 ) . This correlation measures how long the probe on average needs to diffuse across the circle and is therefore more sensitive to detect diffusion barriers than the autocorrelation analysis of a single excitation spot . The resulting pair-correlation is shown in Figure 2B . For the single component membrane , the amplitude and correlation time revealed no clear directional dependence . The average correlation time across the circle was τ ≈ 30 ms , indicating that the pinning of the actin filament imposed no significant diffusion barrier for lipids . These results are in agreement with point FCS measurements on single component membranes in the presence of increasing actin densities , which are reported in Figure 1—figure supplement 2E . Also here , the lateral diffusion was remarkably insensitive to the presence of actin . 10 . 7554/eLife . 01671 . 008Figure 2 . Lipid diffusion was restricted by actin-organized domains . ( A ) Scanning-FCS ( mobility ) analysis of the Ld marker DPPE-KK114 in a single component DOPC membrane in the presence of a low density actin meshwork . The upper panel shows the intensity of the green channel ( actin ) and red channel ( Ld marker ) , indicating the position of the actin fiber on the scan orbit ( perpendicular lines ) . The lower panel depicts the auto-correlation decay for each pixel along the circular scan orbit with a diameter of d = 500 nm . The normalized auto-correlation amplitude is represented by the color . The mean transit time through the excitation spot ( mobility ) can be estimated by the transition from yellow to green . As expected the mobility along the scan orbit was homogenous . ( B ) Pair-correlation analysis of the same data as in A for opposing pixels on the scan orbit . The maxima in the pair-correlation represent the average time the probes need to move across the scan orbit . No significant directional dependence of the mobility was observed in case of simple one component membranes . ( C ) Same as in ( A ) but for ternary membranes ( same composition as in Figure 1 ) . ( D ) While the auto-correlation analysis seemed to be homogenous a distinct directional dependence of diffusion was revealed by the pair-correlation . In the direction along the actin fiber , a distinct correlation peak is visible with a maximum at τ ≈ 40 ms . In the perpendicular direction , the correlation amplitude is reduced , which indicates a diffusion barrier along this axis . ( E ) Representation of the scanning orbit over a single actin fiber bound to the membrane . The numbers represent the angles of the orbit ( F ) . Qualitative representation summarizing the results of the pair-correlation analysis with pronounced diffusion along ( drawn arrow ) and restricted diffusion perpendicular to the actin fiber ( dashed arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01671 . 008 In contrast to the single component membrane , a pronounced directional dependence of lipid diffusion was observed in the ternary system . The intensity distribution of the Ld-marker showed a slight increase at the position of the actin fiber ( upper panel Figure 2C ) , indicating a stabilized Ld phase along the actin fiber . While the autocorrelation analysis along the scan trajectory indicated no clear heterogeneities ( lower panel Figure 2C ) , the pair-correlation function revealed strong peaks at 0 and 180° ( i . e . , along the actin filament ) at delay times τ ≈ 30–50 ms ( Figure 2D ) . In the direction perpendicular to the actin fiber , a much weaker amplitude was observed , with the peak shifted to longer times τ > 50ms . This indicates that the actin stabilized Ld domains favored the diffusion of the Ld marker within the domain ( and thus along the actin fiber ) , while restricting diffusion across domain boundaries . These findings strikingly confirm the simulation prediction of Machta et al . ( 2011 ) , where lipid domains stabilized by actin pinning were also found to ‘compartmentalize’ lipid diffusion . We next consider how domain formation is affected when different pinning sites are used to bind the membrane to the actin network . To this end , we compared three biotinylated lipids with different partitioning values: DOPE-biotin ( i . e . , the same as used in the experiment of Figure 1 ) that partitions strongly into the Ld phase ( Lo% = 11 ± 2 ) , DSPE-PEG-biotin that partitions predominantly in the Lo phase ( Lo% = 78 ± 7 ) , and DPPE-biotin that partitions almost equally in both phases ( but with a small preference toward the Lo phase , Lo% = 59 ± 5 ) . The experimental procedure to determine the Lo% is outlined in Figure 3—figure supplement 1 . In Figure 3A , we show confocal images of the domain structure for the Ld-preferring pinning lipid DOPE-biotin . The figure confirms the previous experiment of Figure 1B: We again observe the formation of Ld domains along the actin fibers , as expressed by a positive Pearson coefficient PCC = 0 . 55 ± 0 . 02 . Pinning of the Lo-preferring lipid DSPE-PEG-biotin induced an ‘inverted’ domain structure , yielding a negative PCC = −0 . 37 ± 0 . 04 ( Figure 3C ) . The binding of actin to DPPE-biotin did not result in a strong localization of lipid domains along the fibers , as manifested by a small ‘but positive’ PCC = 0 . 07 ± 0 . 03 ( Figure 3B ) . The result of Figure 3B is surprising , since DPPE-biotin slightly prefers the Lo phase , and so a ‘negative’ PCC was expected . Notwithstanding that several mechanisms could be responsible for this ( e . g . , streptavidin binding to biotinylated lipids may induce lipid disorder by steric or electrostatic interactions with the membrane ) , it is interesting that a coupling between membrane composition and curvature also brings about this effect . To illustrate this point , we resort to computer simulations , where important parameters such as pinning density and phase partitioning of pinning sites can be systematically varied . 10 . 7554/eLife . 01671 . 009Figure 3 . The type of pinning site used to bind actin to the membrane strongly affects the domain structure . The Ld phase ( left column , magenta ) were stained with DSPE-KK114 , while actin ( middle column , green ) was stained with phalloidin-488 . The right column shows the overlay of both images . The membrane was imaged by confocal microscopy at T = 19°C , using the same lipid composition as in Figure 1 . ( A ) Binding of actin to the Ld preferring lipid DOPE-biotin resulted in Ld domains along the actin fibers ( as in Figure 1B , PCC = 0 . 55 ± 0 . 02 ) . ( B ) When actin was bound to DPPE-biotin , the correlation of domains with the actin fibers was significantly reduced , but remained detectable , with a slightly positive Pearson coefficient ( PCC = 0 . 07 ± 0 . 03 ) . ( C ) Binding of actin to the Lo preferring lipid DSPE-PEG-biotin resulted in correlated Lo domains along the actin fibers , that is the ‘inverse’ structure of ( A ) . In this case , the Pearson coefficient was negative ( PCC = −0 . 37 ± 0 . 04 ) . For the lipid phase partitioning values of the biotinylated lipids used in this experiment see Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01671 . 00910 . 7554/eLife . 01671 . 010Figure 3—figure supplement 1 . Lipid phase partitioning of biotinylated lipid–streptavidin complexes without actin . ( A ) Exemplary image of a phase separated membrane ( DOPC , DPPC , Cholesterol ) containing 1 mol% DSPE-PEG-biotin , and the green fluorescent Ld-marker ( DOPE-fluorescein; Avanti polar lipids Inc . ) . The biotinylated lipid was labeled by streptavidin stained with red fluorescent Atto655 ( Atto-Tec GmbH , Germany ) . ( B ) The partitioning of the biotinylated lipid streptavidin complex was calculated by determining the intensity of red fluorescence in the Lo phase ( int[Lo] ) and in the Ld phase ( int[Ld] ) across a domain boundary . The scan in B corresponds to the yellow box in the overlay image . The partitioning is defined as Lo%=100%×int ( Lo ) / ( int ( Lo ) +int ( Ld ) ) . ( C ) The resulting Lo% partitioning values for the biotinylated lipids used in this study with standard variation from three independent preparations . The partitioning values serve as input parameters for our simulation model . DOI: http://dx . doi . org/10 . 7554/eLife . 01671 . 010 We simulated a phase separating membrane coupled to a picket-fence network resembling actin using a model similar to Machta et al . ( 2011 ) but extended to include membrane curvature . The energy contains three terms: ℋsim=ℋHelfrich+ℋIsing+ℋx ( ‘Materials and methods’ ) . The first term describes the membrane elastic properties using the Helfrich form ( Helfrich , 1973 ) , with bending rigidity κ , and surface tension σ , the second term describes the phase separation using a conserved order parameter Ising model; the third term couples the phases to the local membrane curvature . The strength of the curvature coupling is proportional to the product of κ and the spontaneous curvature difference between Lo and Ld domains ( ‘Materials and methods’ ) . Since there is a substantial range in the experimentally reported values of κ and σ , a large uncertainty ( about one order of magnitude ) in the strength of the curvature coupling is implied ( Schick , 2012 ) . For this reason , we allow the curvature coupling strength to be scaled by a ( dimensionless ) factor g in our analysis . The latter is defined such that , for g > 0 , regions of positive curvature favor unsaturated lipids , that is Ld domains . For g = 0 our model reduces to the one of Machta et al . ( 2011 ) . In situations where curvature coupling is known to occur , one should restrict g to finite positive values . The influence of the actin network is incorporated via ( immobile ) pinning sites , which are distributed randomly along the actin fibers ( linear pinning density ρp ) . At the pinning sites , there is a preferred energetic attraction to one of the lipid species ( set by the Lo% ) . The pinning sites also locally fix the membrane height h , which we model by keeping h = 0 at these locations ( we assume the actin network to lie in a flat plane , providing the reference from which the membrane height is measured ) . Additionally , there is a steric repulsion between the membrane and the actin: directly underneath the actin fibers , the membrane height is restricted to negative values h < 0 . We first consider DPPE-biotin pinning sites that slightly prefer Lo domains ( Lo% = 59 ± 5 ) . Interestingly , the corresponding experiment ( Figure 3B ) revealed a positive PCC , implying a weak alignment of Ld domains along the actin fibers instead . This contradiction can be rationalized , however , when one considers the membrane curvature . In Figure 4G , we show how the simulated PCC varies with the curvature coupling parameter g , using pinning density ρp = 0 . 1/nm , corresponding to 20% of the total actin network being pinned . The key observation is that , at g ≈ 20 , the PCC changes sign and crosses the experimental value . We emphasize that the simulation data were not corrected for the optical point spread function ( PSF ) of the experiment; by artificially broadening the simulation images , lower values g ∼ 10 were obtained , precluding a precise determination . In addition , the value of g , where the PCC changes sign also depends on the pinning density: by increasing ρp , also g must increase to result in a positive PCC . In qualitative terms , the change of sign in the PCC reflects a competition between two effects: an energetic attraction between DPPE-biotin and saturated lipids , favoring alignment of Lo domains , vs a curvature-induced repulsion of these lipids away from the actin fibers . Due to the steric repulsion between the membrane and the actin fibers , the preferred curvature around the fibers is positive on average , thereby favoring Ld domains . At large g , the latter effect dominates , yielding a positive Pearson coefficient . In Figure 4E , we show a typical snapshot corresponding to g = 20 and ρp = 0 . 1/nm . In agreement with the experiment of Figure 3B , we observe a structure of small domains . In contrast , by using g = 0 , the domains grow to be much larger ( Figure 4B ) , contradicting the experimental observations . 10 . 7554/eLife . 01671 . 011Figure 4 . Simulation analysis of the influence of pinning and curvature coupling on lipid phase organization . All data refer to T = 19°C . The Ld lipids are shown as magenta , the Lo lipids as black , and the actin network is shown in green as an overlay . The pinning density ρp = 0 . 1/nm . ( A–C ) Simulation snapshots obtained without coupling to curvature ( g = 0 ) for the three species of pinning sites used in the experiments: Lo% = 11 ± 2 ( A ) , Lo% = 59 ± 5 ( B ) , and Lo% = 78 ± 7 ( C ) . No significant influence of the actin network is apparent . ( D–F ) Same as ( A–C ) but in the presence of curvature coupling ( g = 20 ) . For snapshots ( D ) and ( F ) , the lipid domains strongly correlate to the actin network , with Ld domains favoring actin in ( D ) , and the inverse pattern in ( F ) . The lower panels show height profiles of the images ( D–F ) scanned horizontally along the center of the image; the green dots indicate the positions of the actin fibers . ( G ) Pearson correlation coefficient PCC vs the curvature coupling g for the pinning species with Lo% = 59 ± 5 . For weak curvature coupling , the PCC is negative indicating alignment of Lo domains along actin . By increasing the curvature coupling , the PCC becomes positive and ‘meets’ the experimentally observed value ( conform Figure 4B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01671 . 011 Our model also predicts that , in the presence of curvature coupling , the pinning density required to induce domain alignment along the actin fibers can be much smaller . Previous simulations corresponding to g = 0 ( Ehrig et al . , 2011; Machta et al . , 2011 ) required rather large pinning densities , ρp ∼ 0 . 2 − 1 . 25/nm , in order to achieve this . In Figure 4A , we show a typical domain structure using DOPE-biotin pinning sites ( Lo% = 11 ± 2 ) at pinning density ρp = 0 . 1/nm in the absence of coupling to curvature ( g = 0 ) . Figure 4D shows the corresponding snapshot in the presence of curvature coupling , using g = 20 , which is the value of Figure 4G where the PCC changed sign . To reproduce the experimentally observed domain structure ( Figure 3A ) at low pinning densities , curvature coupling is thus essential . In Figure 4D , the average Ld domain size Rsim ≈ 40 nm , which is somewhat below Rexp ≈ 90 nm of the experiment ( Figure 1D ) . Note , however , that the experimental value likely presents an upper-bound , due to broadening by the PSF . The case of DSPE-PEG-biotin pinning sites ( Lo% = 78 ± 7 ) is considered in Figure 4C , F . Again , curvature coupling is required to reproduce the ( now inverted ) domain structure of the experiment ( Figure 3C ) . In the presence of curvature coupling , the length scale over which the effect of a pinning site propagates is thus enhanced dramatically . This is due to the elastic properties of the membrane: the bending rigidity and the surface tension define a length ξh = ( κ/σ ) 1/2 ( Schick , 2012 ) , which sets the scale over which the membrane height deformations propagate ( for our model parameters ξh ∼ 100 nm ) . When g > 0 , this scale couples to the composition , in which case a reduced pinning density already suffices to induce domain alignment . We emphasize that ξh is essentially independent of temperature , and so it is not necessary for the membrane to be close to a critical point . Also included in Figure 4 are typical membrane height profiles for the snapshots with curvature coupling ( D , E , F ) scanned along a horizontal line through the center of each image . These profiles qualitatively illustrate the curvature coupling effect: Ld domains ( magenta ) reveal positive curvature on average , while for Lo domains ( black ) the curvature is on average negative . In the absence of coupling to actin and g = 0 , the typical root-mean-square height fluctuation is h ≈ 3 . 6 nm . In the presence of actin and curvature coupling , these fluctuations are significantly reduced to h ≈ 2nm , which is very close to the value reported in Speck et al . ( 2010 ) . We have presented an experimental model system in which the response of membrane organization to a bound actin network can be accurately probed using superresolution STED microscopy and FCS . The application of our model system to a single component liquid-disordered membrane shows that actin has only a minor influence on the lateral distribution and dynamics of lipids ( Figure 1—figure supplement 2 , as well as Figure 2A , B ) . In contrast , under the same conditions using a ternary membrane , the effects are dramatic . In particular , by binding actin using pinning sites that attract the Ld phase , earlier simulation predictions ( Ehrig et al . , 2011; Fischer and Vink , 2011; Machta et al . , 2011 ) could finally be put to a stringent test . In agreement with these simulations , our experiments confirm the absence of macroscopic phase separation in the presence of actin . Additionally , our experiments revealed the alignment of Ld domains along the actin fibers , leading to a channel-like domain structure very similar to structures observed in simulations ( Ehrig et al . , 2011; Fischer and Vink , 2011; Machta et al . , 2011 ) . Finally , in agreement with the simulations of Machta et al . ( 2011 ) , the enhanced diffusion of unsaturated lipids along the Ld channels , and the hindered diffusion of these lipids in directions perpendicular , was confirmed . These findings demonstrate that relatively simple simulation models are capable to capture key essentials of lateral membrane organization and dynamics . However , by using pinning sites that weakly attract the Lo phase , our experiments also uncover phenomena that cannot be explained using these simulation models . The paradox is that , for the latter type of pinning site , one still observes alignment of Ld domains along the actin fibers , albeit weak . This indicates that there must be additional mechanisms at play—beyond the level of the pinning-lipid energetic interaction—playing a role in the lateral organization of the membrane . As possible candidate for such a mechanism , we considered the local membrane curvature , and a coupling of the latter to the lipid composition ( Schick , 2012 ) . A simulation model that incorporates these ingredients is able to reproduce the experimentally observed alignment of Ld domains , even when the pinning sites themselves energetically favor the Lo phase . The physical explanation is that the actin network locally induces regions with non-zero average curvature . The coupling of the curvature to the composition then causes these regions to prefer certain types of lipids . Provided the coupling is strong enough , it can overcome the pinning-lipid energetic interaction , which is how we interpret the experimental result . An additional finding is that , in the presence of curvature coupling , the effect of a pinning site extends over a much greater distance ( set by the elastic properties of the membrane ) . Hence , the pinning density can be much lower compared to models where such a coupling is absent . We emphasize once more that curvature coupling is not the only conceivable mechanism that could account for our experimental findings . However , a recent experimental study ( Kaizuka and Groves , 2010 ) of membrane phase separation using intermembrane junctions did uncover a very pronounced curvature coupling , making this a likely candidate . There are interesting implications of our results concerning the in vivo organization of the plasma membrane . Our experiments show that the type of lipid domain selected to be stabilized depends sensitively on the properties of the pinning species . A similar phenomenon , albeit on a larger scale , was observed by crosslinking GM1 with cholera toxin B ( Hammond et al . , 2005; Lingwood et al . , 2008 ) . Since the stabilized phase is determined by the properties of the pinning species , cells could locally sort their membrane components in this way . Moreover , this sorting mechanism persists to physiological temperatures , that is above the temperature of phase separation . At the same time , the pinning sites would naturally prevent the plasma membrane from phase separating at low temperatures . In the presence of curvature coupling , these effects are enhanced . We note that the proposed curvature effect in our experiments was likely limited by the Mica support . However , since the energy cost of lipid extraction far exceeds that of membrane de-adhesion from the support ( Helm et al . , 1991; Lipowsky and Seifert , 1991 ) , we still expect some effect . In free-standing membranes , or in cell membranes , the curvature-coupling is anticipated to be stronger . The recent findings of Kaizuka and Groves ( 2010 ) seem to support this view . Mica ( Muscovite , Pelco , Ted Pella , Inc . , Redding , CA ) was cleaved into thin layers ( ∼10 µm ) and glued ( optical UV adhesive No . 88 , Norland Products Inc . , Cranbury , NJ ) onto clean glass cover slides . Immediately before spin-coating the lipid solution , the MICA on top of the glass was cleaved again to yield a thin ( ∼1 µm ) and clean layer . Next , 30 µl of 2 g/l lipid solution in Methanol/Chloroform ( 1:1 ) were spin-coated ( 2000 rpm , for 30 s ) on top of the MICA . To remove residual solvent , the cover slide was put under vacuum for 20 min . The supported lipid bilayer was hydrated with warm ( 50°C ) buffer ( 150 mM NaCl Tris pH 7 . 5 ) for 10 min and then rinsed several times to remove excess membranes until a single clean bilayer remained on the surface . All lipids were purchased from Avanti Polar Lipids , Inc . , AL USA . The Ld phase was stained with far-red fluorescent DPPE-KK114 ( Kolmakov et al . , 2010 ) or green fluorescent DPPE-OregonGreen-488 ( Invitrogen , Darmstadt , Germany ) . The Lo phase was stained with DSPE-PEG ( 2000 ) -KK114 or DSPE-PEG ( 2000 ) -Cromeo-488 ( Honigmann et al . , 2013 ) . For imaging experiments , the concentration of fluorescent lipids was ∼0 . 1 mol%; for FCS experiments ∼0 . 01 mol% was used . Supported lipid bilayers were doped with biotinylated lipids ( 1 , 2-dioleoyl-sn-glycero-3-phosphoethanolamine-N- ( cap biotinyl ) ( DOPE-biotin ) , 1 , 2-dipalmitoyl-sn-glycero-3-phosphoethanolamine-N- ( cap biotinyl ) ( DPPE-biotin ) , 1 , 2-distearoyl-sn-glycero-3-phosphoethanolamine-N-[biotinyl ( polyethylene glycol ) -2000] ( DSPE-PEG-biotin ) , also purchased from Avanti ) that were used to bind actin fibers to the membrane . The following procedure was performed at 37°C to keep the membrane in the one phase region: The bilayer was incubated with 200 µl of 0 . 1 g/l streptavidin for 10 min and then rinsed several times to remove unbound streptavidin . Next , the membrane was incubated with 200 µl of 1 µM biotinylated phalloidin ( Sigma-Aldrich , Steinheim , Germany ) for 10 min and then rinsed several times to remove unbound phalloidin . Pre-polymerized actin fibers ( 500 µl with 7 µg/ml actin; Cytoskeleton Inc . , Denver , USA ) were then incubated with the membrane for 20 min and then rinsed several times to remove unbound actin . In case actin fibers were imaged , the actin was stained with green fluorescent phalloidin ( Cytoskeleton Inc . ) . The membrane bound actin network was stable for at least 24 hr . The density of the actin network was controlled by the amount of biotinylated lipids in the membrane ( Figure 1—figure supplement 2 ) . The local membrane height h ( x , y ) is a function of the lateral coordinates x and y , which are discretized on the sites of a L×L periodic lattice , L = 400a , with lattice constant a = 2 nm . The membrane elastic energy ℋHelfrich=∑a2 ( κ ( ∇2h ) 2+σ ( ∇h ) 2 ) /2 , with the sum over all lattice sites , and ∇ the gradient operator ( Helfrich , 1973 ) . The first term is the bending energy; the second term reflects the cost of area deformations . We use typical values , κ ∼ 2 . 7 × 10−19Nm and σ ∼ 2 × 10−5N/m , at the same time emphasizing that there is a considerable spread in the reported values of these quantities ( Schick , 2012 ) . This holds especially true for σ , whose value near a support may well be different . This , in turn , implies a large spread in the coupling to curvature strength ( Schick , 2012 ) . As stated before , we adopt the approach keeping κ and σ fixed , while allowing the curvature coupling strength to vary . To describe phase separation , we introduce the local composition s ( x , y ) , which reflects the lipid composition at site ( x , y ) . Experiments indicate that phase separation in membranes ( without actin ) is compatible with the universality class of the Ising model ( Magde et al . , 1972; Honerkamp-Smith et al . , 2009 ) . We therefore use a two-state description , s ( x , y ) = ±1 , where the positive ( negative ) sign indicates that the site is occupied by a saturated ( unsaturated ) lipid , leading to ℋIsing=−J∑s ( x , y ) s ( x' , y' ) , with the sum over all pairs of nearest-neighboring sites , and coupling constant J > 0 . To match the phase transition temperature of the Ising model to the experiment ( Machta et al . , 2011 ) , we choose J = 0 . 44kBTc , with Tc = ( 273 + 28 ) K the transition temperature of the membrane without actin , and kB the Boltzmann constant . It has also been shown experimentally that the membrane height and composition are coupled via the local curvature ( Baumgart et al . , 2003; Parthasarathy et al . , 2006; Yoon et al . , 2006; Parthasarathy and Groves , 2007; Kaizuka and Groves , 2010 ) . This motivates the term ℋx=gκδCa2∑s∇2h , with the sum over all lattice sites , δC ∼ 106m−1 the difference in the spontaneous curvature between Lo and Ld domains ( Leibler and Andelman , 1987; Liu et al . , 2005; Schick , 2012 ) , andg > 0 the dimensionless parameter introduced previously to reflect the fact that the model parameters are not known very precisely . To include actin , a network of line segments ( line thickness a ) was superimposed on the lattice , with a typical compartment size ∼100 nm , close to the experimental value ( Figure 1—figure supplement 2F ) . This network was the Voronoi tessellation of a random set of points ( Ehrig et al . , 2011; Machta et al . , 2011 ) . The network was fixed to the membrane via pinning sites , which were immobile , and distributed randomly along the fibers . The actin network and the pinning sites couple to both the composition and the membrane height . To realize the former , we replaced the composition variable at each pinning site by a fixed value s ( x , y ) = Lo%/50 − 1 , where Lo% is the partitioning fraction of the pinning lipid derived experimentally ( Figure 3—figure supplement 1 ) . To couple the pinning sites to the membrane height , we imposed h ( x , y ) = 0 at the pinning sites ( Speck and Vink , 2012 ) . Additionally , we included a steric repulsion between the membrane and the actin fibers: lattice sites underneath an actin fiber have their corresponding height variable restricted to negative values . The model ℋsim was simulated using the Monte Carlo method . To compute the gradient and Laplace operators , standard finite-difference expressions were used . The simulations were performed at conserved order parameter , using equal numbers of saturated and unsaturated lipids . Two types of Monte Carlo move were used . The first was a Kawasaki move ( Newman and Barkema , 1999 ) , whereby two sites of different composition were chosen randomly , and then ‘swapped’ . This move was accepted conform the Metropolis probability , Pacc=min[1 , e−Δℋsim/kBT] , with Δℋsim the energy difference , kB the Boltzmann constant , and T the temperature . The second move was a height move , whereby a new height was proposed for a randomly selected site; this height was optimally selected from a Gaussian distribution , as explained in Speck and Vink , 2012 . We emphasize that the moves were not applied to pinning sites . In addition , there is the steric repulsion constraint at sites that overlap with the actin network: for these sites , height moves proposing a positive value were rejected . Kawasaki and height moves were attempted with equal a priori probability , with production runs typically lasting 2·106 sweeps , prior of which the system was equilibrated for 4·105 sweeps ( one sweep is defined as L2 attempted moves ) . The temperature of the membrane and the surrounding buffer was controlled by a water cooled Peltier heat and cooling stage which was mounted on the microscope ( Warner Instruments , Hamden , CT , USA ) . The achievable temperature range with this configuration was between 7 and 45°C , with a precision of 0 . 3°C . The actual temperature directly over the membrane was measured by a small thermo-sensor ( P605 , Pt100 , Dostmann electronic GmbH , Wertheim-Reicholzheim , Germany ) . All experiments were performed on a confocal custom-built STED microscope whose main features are depicted in Figure 1—figure supplement 4 . The confocal unit of the STED nanoscope consisted of an excitation and detection beam path . Two fiber-coupled pulsed laser diode operating at λexc = 635 nm and λexc = 485 nm with a pulse length of 80 ps ( LDH-P-635; PicoQuant , Berlin , Germany ) were used for excitation of the green and far red fluorescence respectively . After leaving the fiber , the excitation beams were expanded and focused into the sample using an oil immersion objective ( HCXPLAPO 100x , NA = 1 . 4 , Leica Microsystems ) . The fluorescence emitted by the sample was collected by the same objective lens and separated from the excitation light by a custom-designed dichroic mirror ( AHF Analysentechnik , Tuebingen , Germany ) . In the following , the fluorescence was focused onto a multi-mode fiber splitter ( Fiber Optic Network Technology , Surrey , Canada ) . The aperture of the fiber acted as a confocal pinhole of 0 . 78 of the diameter of the back-projected Airy disk . In addition , the fiber 50:50 split the fluorescence signal , which was then detected by two single-photon counting modules ( APD , SPCM-AQR-13-FC , Perkin Elmer Optoelectronics , Fremont , CA ) . The detector signals were acquired by a single-photon-counting PC card ( SPC 830 , Becker&Hickl , Berlin , Germany ) . The confocal setup was extended by integrating a STED laser beam . A modelocked Titanium:Sapphire laser ( Ti:Sa , MaiTai , Spectra-Physics , Mountain View , USA ) acted as the STED laser emitting sub-picosecond pulses around λSTED = 780 nm with a repetition rate of 80 MHz . The pulses of the STED laser were stretched to 250–350 ps by dispersion in a SF6 glass rod of 50 cm length and a 120 m long polarization maintaining single-mode fiber ( PMS , OZ Optics , Ontario , CA ) . After the fiber , the STED beam passed through a polymeric phase plate ( RPC Photonics , Rochester , NY ) , which introduced a linear helical phase ramp 0 ≤ Φ ≤ 2π across the beam diameter . This wavefront modification gave rise to the doughnut-shaped focal intensity distribution featuring a central intensity zero . The temporal synchronization of the excitation and STED pulses was achieved by triggering the pulses of the excitation laser using the trigger signal from an internal photodiode inside the STED laser and a home-built electronic delay unit , which allowed a manual adjustment of the delay with a temporal resolution of 25 ps . The circular polarization of the STED and excitation laser light in the focal plane was maintained by a combination of a λ/2 and λ/4 retardation plates in both beam paths ( B Halle , Berlin , Germany ) . Integration of a fast scanning unit enabled rapid scanning of the excitation and STED beam across the sample plane . A digital galvanometric two mirror-scanning unit ( Yanus digital scan head; TILL Photonics , Gräfeling , Germany ) was used for this purpose . The combination of an achromatic scan lens and a tube lens in a 4f-configuration ( f = 50 mm and f = 240 mm , Leica , Wetzlar , Germany ) realized a stationary beam position in the back aperture of the objective , preventing peripheral darkening within the focal plane at large scan ranges , such as vignetting . The maximal frequency of the Yanus scanner depended on the scan amplitude and varied between 2 and 6 kHz for scan amplitudes up to 150 µm , respectively . The hardware and data acquisition was controlled by the software package ImSpector ( http://www . imspector . de/ ) . To analyze the temperature-driven phase separation of the membrane ( Figure 1C ) , we calculated the cumulant U1=[x2]/[\|x\|]2 , where the square brackets [⋅] denote an average over all pixel values x in the image . The latter were normalized beforehand such that the mean [x]=0 and the standard deviation [x2]−[x]2=1 . The average domain size R ( Figure 1D ) was extracted from the radial distribution function g ( r ) of the lipid intensity image , which represents the intensity correlations between two points of distance r . We observed that g ( r ) was largest at r = 0 , and decayed for r > 0 . As a measure of the domain size R , we used the criterion g ( R ) = 0 . 5 × g ( 0 ) . The Pearson correlation coefficient ( PCC ) between the actin and the lipid channels ( Figure 1E ) was calculated as the covariance of both channels divided by the product of the standard deviations of both channels . For each temperature up to 5 images from different parts of the sample were analyzed . Vesicles on top of the supported bilayer ( which were visible in the lipid channel as round bright structures ) were excluded from the analysis . For the analysis of Figure 2 , circular orbits 0 . 5–1 . 2 μm in diameter were scanned , at scanning frequency 4 kHz . The scanning orbit was subdivided into 64 pixels . For each pixel i , the fluorescence intensity Fi ( t ) was recorded as a function of time t for a duration of 30–60 s . The correlation between two pixels , i and j , was computed via the pair correlation function ( PCF ) Gij ( τ ) =Fi ( t ) Fj ( t+τ ) /Fi ( t ) Fj ( t ) −1 , where 〈⋅〉 denotes a time average . Figure 2A , C shows the autocorrelation ( i = j ) of each pixel along the orbit , whereas Figure 2B , D shows the correlation between pairs of pixels i and j separated by a rotation of 180° . The maximum of the PCF yields an estimate of how long the fluorescent probes on average need to diffuse from i to j ( Digman and Gratton , 2009 ) . To avoid crosstalk between the two excitation spots , the distance between pairs ( i . e . , the diameter of the scanning orbit ) was at least twice the size of the observation spot . In case of free Brownian diffusion , the PCF is homogenous around the scan orbit . In case the diffusion is hindered by obstacles , the maximum of the PCF is shifted to longer times , and its amplitude is decreased .	All cells are surrounded by a lipid membrane that protects the cell , controls the movement of molecules into and out of the cell , and passes messages about environmental conditions to the cell . This membrane is made of two layers of molecules called lipids , with various proteins embedded in it . There are many different types of lipid molecules that together help to keep the membrane flexible . Moreover , lipid molecules of particular types can also come together to form ‘rafts’ that help the membrane to carry out its various roles . Given the complexity of the cell membrane , cell biologists often use simpler model membranes and computer simulations to explore how the different types of lipid molecules are organized within the membrane . According to the ‘picket fence’ model the cell membrane is divided into small compartments as a result of its interaction with the dense network of actin fibers that acts as a skeleton inside the cell . Recent computer simulations have predicted that these interactions can influence the distribution of lipids and proteins within the membrane . In particular , they can prevent the drastic re-arrangement of lipids into regions of high and low viscosity at low temperature . This temperature dependent re-arrangement of the membrane is known as lipid phase separation . Honigmann et al . have now used computer simulations and two advanced techniques—super-resolution optical STED microscopy and fluorescence correlation spectroscopy—to explore the properties of a model membrane in the presence of a dense network of actin fibers in fine detail . The results show that , in agreement with the simulation predictions of the ‘picket fence’ model , the actin fibers bound to the membrane prevent lipid phase separation happening at low temperatures . Moreover , the actin fibers also help to organize the distribution of lipids and proteins within the membrane at physiological temperatures . Honigmann et al . also suggest that the actin fibers cause the membrane to curve in a way that can reinforce the influence of the ‘picket fence’ . The results show that the ‘raft’ and ‘picket fence’ models are connected , and that a cell can control the properties of its membrane by controlling the interactions between the membrane and the actin fibers that make up the skeleton of the cell .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2014	A lipid bound actin meshwork organizes liquid phase separation in model membranes
The cellular messenger cAMP regulates multiple cellular functions , including signaling in cilia and flagella . The cAMP dynamics in these subcellular compartments are ill-defined . We introduce a novel FRET-based cAMP biosensor with nanomolar sensitivity that is out of reach for other sensors . To measure cAMP dynamics in the sperm flagellum , we generated transgenic mice and reveal that the hitherto methods determining total cAMP levels do not reflect changes in free cAMP levels . Moreover , cAMP dynamics in the midpiece and principal piece of the flagellum are distinctively different . The sole cAMP source in the flagellum is the soluble adenylate cyclase ( SACY ) . Although bicarbonate-dependent SACY activity requires Ca2+ , basal SACY activity is suppressed by Ca2+ . Finally , we also applied the sensor to primary cilia . Our new cAMP biosensor features unique characteristics that allow gaining new insights into cAMP signaling and unravel the molecular mechanisms underlying ciliary function in vitro and in vivo . Cyclic adenosine 3’ , 5’-monophosphate ( cAMP ) controls various physiological functions , such as the heart beat ( Zaccolo , 2009 ) , learning and memory ( Lee , 2015; Morozov et al . , 2003 ) , olfaction ( Kaupp , 2010 ) , and fertilization ( Buffone et al . , 2014 ) . cAMP is synthesized by adenylate cyclases ( ACs ) and degraded by phosphodiesterases ( PDEs ) ( Francis et al . , 2011; Hanoune et al . , 1997; Steegborn , 2014 ) . Local cAMP signaling is achieved by targeting of signaling components to subcellular compartments and assembly of signaling complexes ( Willoughby and Cooper , 2007 ) . A prime example of a subcellular compartment is the cilium , where cAMP translates external stimuli into cellular responses ( Johnson and Leroux , 2010 ) . Cilia come in two different flavors: primary cilia and motile cilia . A prominent example of a motile cilium is the sperm flagellum that serves as a sensory antenna and propels sperm forward . cAMP-signaling pathways control different functions during the sperm’s journey to the egg . The principal cAMP source in sperm , the bicarbonate- and Ca2+-sensitive soluble adenylate cyclase SACY ( Esposito et al . , 2004; Hess et al . , 2005; Vacquier et al . , 2014; Xie et al . , 2006 ) , is activated by increasing bicarbonate concentrations ( Luconi et al . , 2005; Wennemuth et al . , 2003 ) after sperm are released from the epididymis . Sperm lacking SACY are immotile , which causes male infertility ( Esposito et al . , 2004; Hess et al . , 2005; Xie et al . , 2006 ) . The elevated cAMP levels activate protein kinase A ( PKA ) - the primary cAMP target in sperm ( Nolan et al . , 2004 ) . PKA activation in turn accelerates the flagellar beat ( Hess et al . , 2005; Wennemuth et al . , 2003; Xie et al . , 2006 ) . cAMP is also involved in a maturation process of sperm called capacitation that is essential for fertilizing the egg; stimulation of cAMP synthesis by bicarbonate seems to be required for protein tyrosine phosphorylation , which is a hallmark of capacitated sperm ( Visconti et al . , 1995 ) . However , the molecular mechanisms underlying these cAMP-signaling pathways in sperm are not well understood . Components of cAMP signaling have also been identified in immotile cilia , e . g . in the specialized cilium of mammalian olfactory neurons , where cAMP stimulates the electrical signal evoked by odorants ( Kaupp , 2010 ) . Primary cilia also host several cAMP-signaling components ( Johnson and Leroux , 2010 ) : the somatostatin 3 receptor ( SSTR3 ) , various adenylate cyclases ( AC3 , AC4 , AC6 , AC8 ) , PKA , and Epac2 ( exchange protein directly activated by cAMP ) ( Bishop et al . , 2007; Händel et al . , 1999; Kwon et al . , 2010; Masyuk et al . , 2006; Ou et al . , 2009 ) . An important signaling pathway in primary cilia is controlled by Sonic hedgehog ( Shh ) , which determines neuronal cell fate during development ( Chiang et al . , 1996; Ericson et al . , 1997; Huangfu et al . , 2003 ) . Shh signaling relies on PKA activation ( Mukhopadhyay et al . , 2013; Tuson et al . , 2011 ) . However , the physiological function of cAMP signaling in primary cilia and the underlying molecular mechanisms are ill-defined . Analyzing cAMP dynamics in vivo became feasible by the advent of genetically-encoded cAMP biosensors that rely on FRET ( Förster resonance energy transfer ) ( Castro et al . , 2014; Hong et al . , 2011; Sprenger and Nikolaev , 2013; Willoughby and Cooper , 2008 ) . The most advanced cAMP sensors are based on the cyclic nucleotide-binding domain ( CNBD ) of Epac1/2 ( Nikolaev et al . , 2004; Ponsioen et al . , 2004 ) or HCN channels ( Nikolaev et al . , 2006 ) . The CNBD is sandwiched between the fluorescent donor CFP ( cyan fluorescent protein ) and the fluorescent acceptor YFP ( yellow fluorescent protein ) . The KD values of the Epac1/2- or HCN2-based cAMP sensors range between 1–15 µM ( Hong et al . , 2011; Sprenger and Nikolaev , 2013; Willoughby and Cooper , 2008 ) , suitable to detect basal cAMP concentrations and changes in cAMP levels in the micromolar range ( Börner et al . , 2011 ) . In some cell systems , these sensors were able to report a cAMP increase , but their sensitivity was not sufficient to determine basal cAMP levels ( Börner et al . , 2011 ) . The analysis of cAMP dynamics in cilia and flagella is challenging for several reasons . First , the cAMP biosensor must be targeted to this cellular compartment; second , measuring low cAMP concentrations requires the affinity of the sensor to be high , in particular , in a subcellular compartment with only a few cAMP molecules present . Finally , quantitative characterization and calibration in cilia is technically demanding due to their small size compared to the cell soma . Ca2+dynamics have been measured in primary cilia using genetically-encoded biosensors . These studies revealed that the cilium represents a unique Ca2+-signaling compartment that is functionally distinct from the cytoplasm ( DeCaen et al . , 2013; Delling et al . , 2013 ) . Whether cAMP signaling in cilia and flagella is also functionally distinct from the cytoplasm , is ill-defined . Here , we describe a new FRET-based cAMP biosensor ( mlCNBD-FRET ) that is built from the CNBD of the bacterial MlotiK1 channel ( Nimigean et al . , 2004 ) and binds cAMP with nanomolar affinity ( Cukkemane et al . , 2007; Peuker et al . , 2013 ) . This biosensor features unique characteristics that enable its application in solution , in cell lines , and in vivo using kinetic , fluorometric , and live-cell imaging techniques . To target the sensor to cilia and flagella , we designed a cilia-specific targeting approach , and we generated transgenic mice expressing mlCNBD-FRET exclusively in the sperm flagellum . Our results reveal that the dynamics of total and free cAMP levels in sperm and the cAMP dynamics in the midpiece and principal piece of the flagellum are distinctively different . We investigated the regulation of cAMP dynamics in sperm and obtained new insights into the Ca2+-dependent control of cAMP synthesis . The CNBD from the bacterial MlotiK1 channel ( mlCNBD ) consists of a prototypical eight-stranded beta roll ( β1–8 ) flanked by three alpha helices ( αA-C , Figure 1A , B ) ( Clayton et al . , 2004; Cukkemane et al . , 2007; Nimigean et al . , 2004 ) . A phosphate-binding cassette ( PBC ) that interacts with the sugar and phosphate moiety of the cyclic nucleotide is the most conserved feature of CNBDs ( Figure 1B ) . Upon cAMP binding , mlCNBD undergoes a conformational change ( Schünke et al . , 2011 ) . We fused two variants of the green fluorescent protein - citrine ( Griesbeck et al . , 2001 ) and cerulean ( Rizzo et al . , 2004 ) - to the N- and C-terminus of mlCNBD to measure cAMP-induced conformational changes by FRET . Cerulean , the FRET donor , transfers energy to the FRET acceptor citrine . A histidine tag ( His10 ) was fused to the C-terminus of cerulean ( Figure 1C ) to purify the protein via cobalt immobilized-metal affinity chromatography ( Figure 1D ) . In size-exclusion chromatography , the mlCNBD-FRET protein eluted as a single peak with an apparent molecular mass of 67 kDa , close to the calculated molecular mass of 70 . 9 kDa ( Figure 1E ) . 10 . 7554/eLife . 14052 . 003Figure 1 . Generation and purification of mlCNBD-FRET . ( A ) Ribbon presentation of a single mlCNBD according to ( Schünke et al . , 2011 ) . The first ( Phe223 ) and the last amino acid ( Arg349 ) are indicated . ( B ) Structural features of mlCNBD . Alpha helices ( αA-C ) , beta rolls ( β1–8 ) , and the phosphate binding-cassette ( PBC ) are indicated . The arginine ( R ) crucial for cAMP-binding is boxed . ( C ) The mlCNBD-FRET biosensor . The sensor has been generated by fusing citrine and cerulean to the N- and C-terminus of mlCNBD , respectively . For purification , a His10 tag has been added to the C-terminus . ( D ) Purification of mlCNBD-FRET via cobalt immobilized-metal affinity chromatography . Representative elution profile for mlCNBD-FRET using a linear imidazole gradient . The absorption has been recorded at three different wavelengths ( 280 nm: protein , green; 433 nm: cerulean , black; 516 nm: citrine , red ) . The inset shows a representative Western blot for the purified mlCNBD-FRET and mlCNBD-FRET-R307Q protein , stained with an anti-His antibody . ( E ) Size-exclusion chromatography of the purified mlCNBD-FRET protein . Representative elution profile . The protein eluted in a main peak at 67 kDa ( peak maximum ) , close to the expected molecular mass of 70 . 9 kDa . A minor peak was observed that eluted earlier and represents the void volume . Fractions indicated by the grey line have been used for analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 003 Binding of cAMP to mlCNBD-FRET was measured in a spectrofluorometer . The two fluophores in the protein undergo FRET , indicating that they are in close proximity in the absence of cAMP ( Figure 2A ) . Addition of cAMP decreased FRET: the cerulean emission at 475 nm increased , whereas the citrine emission at 529 nm decreased ( Figure 2A ) . Thus , upon cAMP binding , the two fluorophores move further apart ( Figure 2B , C ) . In the following , changes of the cerulean/citrine emission ratio ( △F ) are used; an increase of this ratio reflects a cAMP increase and vice versa . We characterized cAMP binding to mlCNBD-FRET by measuring the change in △F caused by increasing cAMP concentrations ( Figure 2D ) . The dose-response relationship was fitted with a single binding-site model ( Cukkemane et al . , 2007 ) . The KD of mlCNBD-FRET for cAMP was 66 ± 15 nM ( n = 5 ) , similar to that for mlCNBD ( 68 ± 9 nM ) ( Cukkemane et al . , 2007 ) . This demonstrates that the two fluorophores do not interfere with cAMP binding , and that the sensor detects cAMP concentrations in the low to medium nanomolar range . Finally , a control mutant carrying an R307Q amino-acid exchange in the PBC that abolishes cAMP binding ( Figure 1B ) ( Bubis et al . , 1988; Harzheim et al . , 2008; McKay and Steitz , 1981; Zagotta et al . , 2003 ) , did not respond to cAMP concentrations up to 1 . 5 μM ( Figure 2D ) . 10 . 7554/eLife . 14052 . 004Figure 2 . Characterization of the purified mlCNBD-FRET . ( A ) Fluorescence spectra of mlCNBD-FRET at 430 nm excitation before ( black ) and after addition of 10 μM cAMP ( grey ) . ( B ) Schematic representation of the structural changes evoked by cAMP upon binding , FRET becomes smaller , indicating that cerulean and citrine move further apart . ( C ) Structural changes occurring after cAMP binding . The cAMP-free structure is shown in blue , the cAMP-bound structure is shown in green . Distances are presented in Ångstrom . ( D ) Binding of cAMP to mlCNBD-FRET ( black ) determined by fluorescence spectroscopy . Representative experiments showing an increase in the baseline-corrected cerulean/citrine emission ratio ( △F ) of mlCNBD-FRET ( 430 nm excitation ) after cAMP binding . Data have been fitted using a single binding-site model ( red line ) ( Cukkemane et al . , 2007 ) . As a control , mlCNBD-FRET-R307Q ( red dots ) has been used . Measurements have been performed using 1 μM protein . ( E ) Representative experiments showing an increase of △F of mlCNBD-FRET ( 430 nm excitation ) after cGMP binding . Data has been fitted ( see ( D ) , red line ) . Measurements have been performed using 1 μM protein . ( F ) Normalized fluorescence spectra of mlCNBD-FRET ( 430 nm excitation ) at different pH conditions . Spectra were normalized to the cerulean emission at 471 nm . ( G ) Normalized FRET ( 430 nm excitation , 529 nm emission ) at different pH values . Data have been taken from measurements shown in ( F ) and are presented as mean ± S . D . ; n = 3 . ( H ) Kinetics of cAMP binding to mlCNBD-FRET measured using the stopped-flow technique . Different cAMP concentrations ( in µM: 0 . 5 , 1 , 1 . 6 , 2 . 5 , 5 , and 10 ) were mixed with the purified mlCNBD-FRET protein ( 2 . 5 μM ) and the change in FRET was measured over time . Solid lines represent a global fit of a one-step model ( see materials and methods ) with the following parameters: kon = 2 . 6 107 M-1s-1 and koff = 12 . 8 s-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 004 We also determined the cGMP sensitivity of mlCNBD-FRET: cGMP enhanced △F at a tenfold higher concentration than cAMP ( Figure 2E ) . From the dose-response relationship , we determined for cGMP a KD of 504 ± 137 nM ( n = 6 ) , which is similar to the KD value of mlCNBD ( 499 ± 69 nM ) ( Cukkemane et al . , 2007 ) . The fluorescence of GFP derivatives , in particular YFP and citrine , is pH sensitive ( Griesbeck et al . , 2001 ) . Therefore , we measured the emission spectrum of mlCNBD-FRET ( while exciting cerulean ) at different pH values . The citrine emission peak at 529 nm increased at higher pH values and saturated at pH ≥ 7 . 5 , whereas the cerulean emission at 475 nm was largely pH-insensitive ( Figure 2F , G ) . The KD for cAMP binding was only modestly affected by changes in pH ( pH 6: 101 ± 5 nM , pH 7: 98 ± 21 nM; pH 7 . 5: 66 ± 15 nM , pH 8: 50 ± 13; n = 3 ) , indicating that the cAMP affinity might increase upon alkalization . Of note , it has been reported for other CNBDs that the cAMP affinity depends on the pH ( Gordon et al . , 1996; Kaupp and Seifert , 2002 ) . We also measured FRET using Fluorescence Lifetime Spectroscopy ( FLS ) . The fluorescence lifetime of the donor decreases during FRET . Because cAMP binding to mlCNBD-FRET reduces FRET , the cerulean lifetime should increase upon binding . The decay of cerulean fluorescence alone was fitted with a bi-exponential function ( weighted mean of the two lifetime constants τwm = 2 . 52 ± 0 . 09 ns , n = 12 ) . In mlCNBD-FRET , the cerulean lifetime decreased ( τwm = 2 . 38 ± 0 . 04 ns , n = 11 ) . Saturating cAMP ( 5 μM ) decreased FRET and , thereby , the cerulean lifetime increased ( τwm = 2 . 44 ± 0 . 03 ns , n = 5 ) . In summary , the mlCNBD-FRET sensor is suitable for both intensity- and lifetime-based approaches . The response time of the sensor to rapid changes in cAMP was determined using the stopped-flow technique . Different cAMP concentrations were mixed with purified mlCNBD-FRET protein . From the time course of the FRET response , we determined the time constant for the changes in fluorescence by numerically fitting a simple one-step model to the data ( Figure 2H , see materials & methods ) . According to this model , the overall on and off rates of ligand binding and changes in FRET are 2 . 5 ± 0 . 6 107 M-1 s-1 and 9 . 3 ± 6 . 7 s-1 ( n = 3 ) . The off rate is rate-limiting , resulting in a time constant of about 100 ms . Thus , mlCNBD-FRET allows measuring cAMP changes on a 100 millisecond time scale . We tested mlCNBD-FRET in a cellular environment using HEK293 cells . Western blotting confirmed the expression of mlCNBD-FRET ( Figure 3A ) ; the sensor was mainly localized to the cytosol , but a fraction also resided in the nucleus ( Figure 3B ) . To determine the cAMP-binding characteristics , we calibrated mlCNBD-FRET by bathing digitonin-permeabilized HEK293 cells in defined cAMP concentrations . The KD for cAMP binding was 73 ± 20 nM ( n = 11 , Figure 3C ) , which is similar to that of the purified mlCNBD-FRET ( 66 ± 15 nM , Figure 2D ) . Using the KD value and the cAMP null-point calibration , we determined a basal free cAMP concentration of 35 ± 1 nM ( n = 3 , Figure 3—figure supplement 1A ) . 10 . 7554/eLife . 14052 . 005Figure 3 . Characterization of mlCNBD-FRET in HEK293 cells . ( A ) . Representative Western blot using total protein lysates from mlCNBD-FRET and mlCNBD-FRET-R307Q-expressing cells , stained with an anti-His and an anti-GFP antibody . Calnexin ( Cal ) has been used as a loading control . ( B ) Immunocytochemistry . HEK293 cells expressing mlCNBD-FRET ( cerulean: blue , citrine: green ) have been labeled with an anti-GFP antibody and a fluorescent secondary antibody ( red ) . Scale bar 25 µm . ( C ) Ligand binding of cAMP to mlCNBD-FRET in HEK293 cells determined by fluorescence spectroscopy . Representative experiment showing an increase of the baseline-corrected cerulean/citrine emission ratio ΔF of mlCNBD-FRET ( 430 nm excitation ) at different cAMP concentrations . Cells have been permeabilized with 20 μM digitonin before addition of cAMP . Data have been fitted using a single binding-site model ( red line ) ( Cukkemane et al . , 2007 ) . ( D ) Changes in FRET in HEK293 cells expressing mlCNBD-FRET after stimulation with 40 μM NKH477/500 μM IBMX . FRET has been measured by fluorescence microscopy . Representative traces of raw data are shown above . HEK293 cells expressing mlCNBD-FRET-R307Q ( grey ) have been used as a control . Data are presented as mean ± S . D . ( mlCNBD-FRET: n = 31; mlCNBD-FRETR307Q: n = 3 ) . ( E ) Changes in FRET in HEK293 cells expressing mlCNBD-FRET after stimulation with 2 μM isoproterenol . Representative traces for the raw data are shown above . Data are presented as mean ± S . D . ; n = 9 . ( F ) Changes in FRET in HEK293 cells expressing mlCNBD-FRET after stimulation with 3 mM SNP . Representative traces for the raw data are shown above . Data are presented as mean ± S . D . ; n = 36 . ( G ) Changes in FRET in HEK293 cells expressing mlCNBD-FRET ( black ) or mlCNBD-FRET-R307Q ( grey ) after stimulation with 40 μM NKH477 . After reaching a steady-state , cells have been permeabilized using 1 μM digitonin . FRET has been measured using spectrofluorometer . Data are presented as mean ± S . D . ; n = 3 . ( H ) Changes in cerulean fluorescence lifetime using FLIM . HEK293 cells expressing mlCNBD-FRET were imaged under basal conditions , after addition of 20 μM digitonin , and the following addition of 5 μM cAMP . The cerulean fluorescence decay was recorded and fitted with a bi-exponential decay to calculate the lifetime . Data are presented as mean ± S . D . Representative images are shown using a look-up table ranging from 2 . 1 ns ( blue ) to 2 . 7 ns ( red ) . ( I ) Mean values of the two lifetimes averaged over different regions of interest in part ( H ) ; n = 8 for each condition . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 00510 . 7554/eLife . 14052 . 006Figure 3—figure supplement 1 . Characterisation of mlCNBD-FRET in HEK293 cells . ( A ) Calibration of mlCNBD-FRET in HEK293 cells . FRET was measured in a cuvette in a spectrofluorometer under basal conditions ( ES ) , after permeabilization with 20 μM digitonin , and the following addition of increasing cAMP concentrations ( in nM ) . According to the null-point calibration method , the difference in FRET ratio of the treated samples to the basal condition was determined and used to determine the basal cAMP concentration . ( B ) Changes in FRET in HEK293 cells expressing mlCNBD-FRET after stimulation with 40 μM NKH477/500 μM IBMX ( black ) . FRET has been measured using spectrofluorometer . Data are presented as mean ± S . D . DMSO ( 0 . 13% , red ) has been used as a control; n = 3 for each condition . ( C ) Similar to part ( B ) after stimulation with 2 μM isoproterenol ( black ) . ( D ) Similar to part ( B ) for stimulation with 3 mM SNP ( black ) . ( E ) Change in the cerulean fluorescence lifetime measured using Fluorescence Lifetime Spectroscopy ( FLS ) . Cells have been permeabilized with 20 μM digitonin followed by addition of cAMP . The cerulean fluorescence decay was recorded and fitted with a bi-exponential decay to calculate the lifetime . The mean of the two lifetimes was calculated and averaged over n = 3 experiments . ( F ) Changes in FRET in HEK293 cells expressing mlCNBD-FRET after stimulation with different concentrations of NKH477 . The FRET ratio has been calculated when reaching a maximum and normalized to the baseline ratio . Data is shown as mean ± S . D . ; n = 4 . ( G ) Changes in FRET in HEK293 cells expressing mlCNBD-FRET after alternatingly stimulating with 500 nM isoproterenol ( black ) followed by a wash-out with ES . As a control , cells were stimulated with buffer only ( red ) . Data is shown as mean ± S . D . ; n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 006 Next , we studied the cAMP dynamics of mlCNBD-FRET-expressing HEK293 cells by stimulation with a mixture of 40 µM NKH477 and 500 µM IBMX: NKH477 activates transmembrane adenylate cyclases that synthesize cAMP ( tmAC ) ( Hosono et al . , 1992 ) , whereas IBMX inhibits phosphodiesterases ( PDEs ) that hydrolyze cAMP ( Schmidt et al . , 2000 ) . NKH477/IBMX treatment increased △F , reflecting an increase of cAMP levels ( Figure 3D ) . Of note , the changes in citrine and cerulean emission are of opposite sign but similar time course , demonstrating that the changes in fluorescence were owing to FRET and not to fluorescence artefacts . The FRET change commenced within a few seconds after stimulation and reached a steady-state after 10 min ( ratio increase: 43 ± 4% , n = 31 , Figure 3D ) . HEK293 control cells expressing the cAMP-insensitive mlCNBD-FRET-R307Q mutant did not respond ( n = 3 , Figure 3D ) . We also studied whether the mlCNBD-FRET sensor reports a cAMP increase mediated by G protein-coupled receptors: stimulation with 20 µM isoproterenol , an agonist of beta-adrenergic receptors , rapidly changed △F by 47 ± 3% ( n = 9 , Figure 3E ) . mlCNBD-FRET also reports changes in cGMP; stimulation with 3 mM SNP , which releases nitric oxide ( NO ) that activates soluble guanylyl cyclases ( Denninger and Marletta , 1999 ) , changed △F by 25 ± 5% ( n = 36 , Figure 3F ) . Similar results were obtained using cell populations: NKH477/IBMX and isoproterenol treatment both evoked a larger change than SNP ( n = 3 , Figure 3—figure supplement 1B–D ) . The twofold difference in the maximal changes evoked by drugs that stimulate cAMP- or cGMP-synthesis factors seems small , considering that the respective KD values differ by 10fold . Probably , at rest , the sensor is partially occupied by cAMP and stimulation with NKH477/IBMX or isoproterenol saturates the response . To test the sensor with submaximal agonist concentrations , we stimulated cells with increasing concentrations of NKH477 and analyzed the dose-response relationship ( Figure 3—figure supplement 1F ) . The EC50 for NKH477 was 3 . 6 ± 0 . 6 μM ( n = 4 ) . Moreover , we analyzed whether mlCNBD-FRET also reliably reports a decrease of cAMP levels . HEK293 cell populations were pre-stimulated with NKH477 until steady-state before permeabilizing with 1 μM digitonin to release cAMP ( Figure 3G ) . In turn , the FRET ratio decreased ( Figure 3G ) , demonstrating that mlCNBD-FRET registers both , an increase and decrease of the intracellular cAMP concentration . To rule out any unspecific effects during permeabilization , we used mlCNBD-FRET-R307Q-expressing cells as a control . Here , the FRET ratio remained constant after NKH477 stimulation; only a small decrease occurred upon addition of digitonin ( Figure 3G ) . Similarly , we tested the reversibility of the sensor by alternatingly stimulating with isoproterenol followed by a wash-out step . Stimulation with 500 nM isoproterenol increased the FRET ratio ( Figure 3—figure supplement 1G ) . When reaching a maximum , the stimulus was removed and in turn , the FRET ratio decreased . Afterwards , a second stimulus of 500 nM isoproterenol was applied , which resulted in a similar increase as observed for the first stimulus ( Figure 3—figure supplement 1G ) . Finally , we also tested the performance of mlCNBD-FRET in a cellular environment by two different lifetime-based techniques . First , using FLS , we calibrated the sensor with defined cAMP concentrations in permeabilized , mlCNBD-FRET-expressing HEK293 cells . The cerulean lifetime increased after cAMP addition ( Figure 3—figure supplement 1E ) . The KD for cAMP derived from the dose-lifetime relationship was 99 . 0 ± 10 . 1 nM ( n = 3 ) , which is similar to the KD derived from fluorescence intensity-based FRET ( 73 ± 20 nM ) . Furthermore , the basal free cAMP concentration in HEK293 cells , calculated from lifetime and intensity-based measurements , was similar ( 48 . 7 ± 11 . 0 nM vs . 35 ± 1 nM , respectively , n = 3 ) . To evoke cAMP changes in intact cells , cells were treated with NKH477/IBMX , which also prolonged the cerulean lifetime compared to controls ( τwm = 1 . 88 ± 0 . 04 ns vs . τwm = 1 . 98 ± 0 . 03 ns , n = 9 ) . Second , we used FRET Fluorescence Lifetime Imaging ( FLIM ) . In mlCNBD-FRET-expressing HEK293 cells , the cerulean lifetime decreased as a result of FRET ( τwm =2 . 44 ± 0 . 02 ns vs . τwm =3 . 28 ± 0 . 07 ns for cerulean only , n = 9 ) . Upon addition of NKH477/IBMX , the cerulean lifetime increased when FRET decreased ( τwm =2 . 50 ± 0 . 02 ns , n = 9 ) . To explore the dynamic range , we first permeabilized cells to release intracellular cAMP , followed by addition of a saturating cAMP concentration ( 5 μM ) . The shift of lifetime upon changes in cAMP is illustrated in the color-coded FLIM images ( Figure 3H ) . The cerulean lifetime decreased after digitonin treatment from τwm = 2 . 44 ± 0 . 02 ns to τwm =2 . 38 ± 0 . 04 ns ( n = 8 ) , reflecting a cAMP decrease . Upon addition of cAMP , the lifetime increased to τwm =2 . 47 ± 0 . 05 ns ( n = 8 , Figure 3H , I ) . In summary , mlCNBD-FRET is an exquisitely sensitive biosensor for measuring cAMP dynamics in the nanomolar range , preferably using fluorescence intensity techniques , but also using lifetime-based techniques . To study cAMP dynamics in sperm flagella , we generated transgenic mice expressing mlCNBD-FRET under the control of the protamine 1 promoter ( Prm1 , Figure 4A ) . Transgenic mice were generated by pronuclear injection using standard procedures ( Ittner and Götz , 2007 ) . Genomic insertion of the transgene was confirmed by PCR ( Figure 4B ) . The mlCNBD-FRET protein was exclusively expressed in sperm and was mainly targeted to the flagellum ( Figure 4C–E ) . Prm1-mlCNBD-FRET males were fertile , demonstrating that mlCNBD-FRET does not impair sperm function ( Figure 4—source data 1 ) . 10 . 7554/eLife . 14052 . 007Figure 4 . Generation of a Prm1-mlCNBD-FRET transgenic mouse line . ( A ) Scheme of the Prm1-mlCNBD-FRET targeting vector . Expression of hemagglutinin ( HA ) -tagged mlCNBD-FRET is driven by the Protamine 1 promoter ( Prm1 ) ; arrows indicate the position of genotyping primers ( #1 , 2 ) . ( B ) Genotyping by PCR . In Prm1-mlCNBD-FRET mice , a 653 bp fragment is amplified ( 1 ) . The targeting vector served as a positive control ( + ) . ( C ) Western blot analysis of mlCNBD-FRET expression in lysates from different tissues from a transgenic male and a female . Lysates from HEK293 cells expressing mlCNBD-FRET served as positive control . Proteins have been labeled using an anti-GFP antibody . ( D ) Western blot analysis of mlCNBD-FRET expression in testis and sperm lysates from a wild-type and a transgenic male . Lysates from HEK cells expressing mlCNBD-FRET served as positive control . Proteins have been labeled using an anti-GFP antibody; tubulin has been used as a loading control . ( E ) Immunohistochemical analysis of mlCNBD-FRET expression in testis and sperm . Cryosections of mouse testis were probed with anti-HA antibody and fluorescent secondary antibody ( purple ) ; the fluorescence for cerulean ( blue ) or citrine ( green ) is also shown . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 00710 . 7554/eLife . 14052 . 008Figure 4—source data 1 . Fertility parameters of mlCNBD-FRET transgenic males . For matings , heterozygous males have been crossed with wild-type females . All data are represented as mean ± S . D . , n numbers are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 008 To calibrate mlCNBD-FRET in sperm , we followed a similar strategy as used for HEK293 cells . Titration of digitonin-treated sperm with cAMP decreased FRET , thereby increasing the FRET ratio ( Figure 5A ) . The mean KD value derived from the dose-response relation was 103 ± 31 nM ( n = 7 , Figure 5A ) . We also measured by FLS the cerulean lifetime of mlCNBD-FRET sperm . At rest , the cerulean lifetime was τwm = 2 . 0 ± 0 . 1 ns ( n = 4 ) ; addition of cAMP to digitonin-treated sperm prolonged the cerulean lifetime . At saturating cAMP concentrations ( 6 μM ) , the lifetime was τwm = 2 . 3 ± 0 . 03 ns ( n = 3 ) . The mlCNBD-FRET characteristics for cAMP binding in the different systems are summarized in a table ( Figure 5—source data 1 ) . For comparison , we have also included a summary of cAMP biosensors and their characteristics ( Figure 5—source data 2 ) . In summary , the sensor allows measuring changes in cAMP levels in sperm using fluorescence intensity and lifetime-based approaches . 10 . 7554/eLife . 14052 . 009Figure 5 . Characterization of cAMP dynamics in sperm . ( A ) Ligand binding of cAMP to mlCNBD-FRET in mouse sperm determined by fluorescence spectroscopy . Representative experiment showing an increase in the baseline-corrected cerulean/citrine emission ratio ( △F ) of mlCNBD-FRET ( 430 nm excitation ) after cAMP binding . Cells have been permeabilized with digitonin before addition of cAMP . Data have been fitted using a single binding-site model ( red line ) ( Cukkemane et al . , 2007 ) ; n = 7 . ( B ) Fluorescence spectra of mlCNBD-FRET at 430 nm excitation before ( black ) and after stimulation for 5 min with 25 mM bicarbonate ( red ) . ( C ) Changes in FRET after stimulation of a mlCNBD-FRET sperm with 25 mM bicarbonate . Sperm have been kept in 2 mM Ca2+ buffer . FRET has been measured using a spectrofluorometer . Data is shown as mean ± S . D . ; n = 3 . ( D ) Total cAMP content . Sperm have been stimulated with 25 mM bicarbonate for 1 or 10 min and the total cAMP content has been determined using an immunoassay . The p values according to Students t-test are indicated . Data are shown as mean ± S . D . ; n = 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 00910 . 7554/eLife . 14052 . 010Figure 5—source data 1 . Characteristics of mlCNBD-FRET . Binding affinities and the cerulean lifetime are shown as mean ± S . D . ; n numbers are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 01010 . 7554/eLife . 14052 . 011Figure 5—source data 2 . Characteristics of other cAMP biosensors . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 01110 . 7554/eLife . 14052 . 012Figure 5—figure supplement 1 . Characterisation of mlCNBD-FRET in mouse sperm . ( A ) Changes in pHi in wild-type sperm . Wild-type sperm have been loaded with BCECF and changes in pHi , after stimulation with 10 mM NH4Cl or 25 mM bicarbonate , have been measured using the stopped-flow technique . Data are shown as mean ± 95% CI ( confidence interval ) ; n = 4 . ( B ) Changes in FRET ratio of mlCNBD-FRET sperm after stimulation with 25 mM NH4Cl . Data is shown as mean ± S . D . ; n = 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 012 Changes in cAMP levels in sperm have previously been inferred from total cAMP concentrations measured with immunoassays . However , total cyclic nucleotide concentrations in cells are several orders of magnitude larger than free concentrations . Moreover , large volume ratios between the soma and small compartments like flagella might obscure the extent and time course of changes in cAMP . Therefore , we compared the dynamics of total and free cAMP in sperm upon stimulation with bicarbonate . Bicarbonate ( 25 mM ) increased the FRET ratio , reflecting an increase of the free cAMP concentration ( Figure 5B ) . cAMP commenced to rise within seconds , reached a steady-state after about 5 min , and persisted during bicarbonate stimulation ( Figure 5C ) . In contrast , total cAMP levels , after a rapid rise ( 1 min ) , declined again to basal levels within 10 min ( Figure 5D ) . Previous studies also reported a transient increase of cAMP during bicarbonate stimulation , although the extent and speed of recovery varied among species ( Battistone et al . , 2013; Brenker et al . , 2012 ) . In essence , the changes in total cAMP concentration unsatisfactorily reflect the changes in free cAMP levels in mouse sperm . Of note , we tested whether the bicarbonate-induced increase in FRET ratio reflects a change in cAMP levels and not a change in the intracellular pHi . Wild-type sperm were loaded with the fluorescent pH indicator BCECF and changes in pHi were measured after stimulation with 25 mM bicarbonate or 10 mM NH4Cl as a control . Stimulation with NH4Cl , evoked a pronounced alkalization and , in turn , decreased the FRET ratio ( Figure 5—figure supplement 1A , B ) . The kinetics of both changes were similar , indicating that the change in FRET is evoked by the change in pHi . In contrast , stimulation with NaHCO3 did not dramatically change pHi ( Figure 5—figure supplement 1A ) . These results support the notion that the changes in FRET evoked by NaHCO3 reflect changes in cAMP rather than pHi . Expression of mlCNBD-FRET also allows spatially resolving cAMP dynamics in sperm . The mlCNBD-FRET biosensor is predominantly expressed in the flagellum ( Figure 4E ) . Although the major if not only cAMP source is SACY ( Brenker et al . , 2012; Esposito et al . , 2004; Hess et al . , 2005 ) , a debate continues about the presence and function of transmembrane adenylate cyclases ( tmACs ) ( Buffone et al . , 2014; Vacquier et al . , 2014; Wertheimer et al . , 2013 ) , for a comprehensive discussion see ( Brenker et al . , 2012 ) . To gain more insight , a spatio-temporal analysis of tmAC- and SACY-dependent changes in cAMP is required . Stimulation by NKH477 and bicarbonate discriminates between tmACs and SACY activity , respectively . Thus , we measured cAMP dynamics in the freely beating flagellum of mlCNBD-FRET sperm upon perfusion with 40 μM NKH477 or 25 mM bicarbonate . mlCNBD-FRET is equally distributed along the flagellum , which allows distinguishing between the cAMP dynamics in the midpiece and principal piece ( Figure 6A; blue and green region , respectively ) . Invariably , no cAMP change in either the midpiece or principal piece was detected after perfusion with NKH477 ( 0 ± 3% , n = 3 , Figure 6B ) . Thus , our results rule out the presence of tmACs in the sperm flagellum . However , perfusion with bicarbonate increased cAMP levels in both , midpiece ( 26 ± 3% ) and principal piece ( 22 ± 10% ) ( n = 7 , Figure 6C ) . To compare the response kinetics in the two compartments , individual traces were fitted using a logistic regression ( Origin 9 . 0 ) and average data were plotted ( Figure 6D ) . The kinetics of the cAMP increase by bicarbonate was faster in the principal piece than in the midpiece , indicating that cAMP dynamics in these two compartments is differently regulated . 10 . 7554/eLife . 14052 . 013Figure 6 . Spatio-temporal cAMP dynamics in the sperm flagellum . ( A ) cAMP dynamics was analyzed in a region 20 µm in length in the midpiece ( blue ) and principal piece ( green ) of freely beating sperm . The cytoplasmic droplet is indicated with an arrow . ( B ) Changes in FRET after stimulation with 40 μM NKH477 . The perfusion with NKH477 is indicated with a grey box . Data for a representative cell are shown as mean ± S . D . in the midpiece ( blue ) and principal piece ( green ) . ( C ) Changes in FRET after stimulation with 25 mM bicarbonate . The perfusion with bicarbonate is indicated with a blue box . Individual traces have been fitted using logistic regression ( Origin 9 ) ( red line ) . ( D ) Average of the fitted data presented in ( C ) . The blue and green line represent the mean value for the midpiece and principal piece , respectively ( n = 7 ) . The blue and green areas represent the corresponding S . D . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 013 SACY activity is controlled by bicarbonate and Ca2+ ( Carlson et al . , 2007; Chen et al . , 2000; Jaiswal and Conti , 2003; Litvin et al . , 2003; Wennemuth et al . , 2003 ) . Ca2+ enhances the in vitro activity of native or heterologously expressed SACY in a dose-dependent manner ( Jaiswal and Conti , 2003 ) . In intact sperm , several read-outs including total cAMP content , flagellar beat frequency , stimulation of PKA activity , and increase of tyrosine phosphorylation have been used to assess SACY activity ( Carlson et al . , 2007; Navarrete et al . , 2015; Visconti et al . , 1995; Wennemuth et al . , 2003 ) . For example , stimulation of sperm with bicarbonate in so-called nominally Ca2+-free buffers , which contain an undefined free Ca2+ concentration [Ca2+]o in the low micromolar range ( Marín-Briggiler et al . , 2005 ) , does not increase the flagellar beat frequency ( Carlson et al . , 2007; Wennemuth et al . , 2003 ) . Recently , it has been proposed that Ca2+ regulates cAMP signaling in a biphasic manner ( Navarrete et al . , 2015 ) , i . e . cAMP signaling seems to be stimulated at extremely low [Ca2+]o . Although [Ca2+]i is not as well defined as [Ca2+] in cell-free assays ( Jaiswal and Conti , 2003 ) , it is reasonable to assume that [Ca2+]o affects [Ca2+]i accordingly . Given the limitations of indirect read-outs of free cAMP concentrations , we studied the Ca2+ regulation of SACY using mlCNBD-FRET . We followed the changes in cAMP upon stimulation with 25 mM bicarbonate in buffer with defined Ca2+ concentrations of 10 μM [Ca2+]o ( low ) and 2 mM [Ca2+]o ( normal ) under otherwise identical conditions ( Figure 7A ) . In agreement with indirect read-outs , bicarbonate did not change the cAMP concentration at 10 μM [Ca2+]o , whereas readjusting [Ca2+]o to 2 mM rapidly elevated cAMP levels ( Figure 7A ) . These results demonstrate that stimulation of SACY activity by bicarbonate requires Ca2+ and that the regulation by Ca2+ is fully and rapidly reversible . 10 . 7554/eLife . 14052 . 014Figure 7 . Ca2+ regulation of cAMP dynamics in sperm . ( A ) Changes in FRET after stimulation of mlCNBD-FRET sperm with 25 mM bicarbonate . Sperm have been either kept in 2 mM ( black ) or 10 μM ( grey ) Ca2+ buffer . FRET has been measured using a spectrofluorometer; n = 3 for each condition . ( B ) Changes in FRET after stimulation of a mlCNBD-FRET sperm kept 10 μM Ca2+ buffer with 25 mM bicarbonate , followed by addition of 2 mM Ca2+ ( final concentration ) ; n = 3 . ( C ) Changes in FRET of mlCNBD-FRET sperm kept in 2 mM Ca2+ buffer after the addition of 2 mM BAPTA ( final: 20 μM Ca2+ , black ) , kept in 10 μM Ca2+ buffer after the addition of 1 mM BAPTA ( final: 2 . 2 nM Ca2+ , red ) , or kept in 2 mM Ca2+ buffer after the addition of 3 mM BAPTA ( final: 443 nM Ca2+ , blue ) . Arrow indicates the addition of Ca2+or BAPTA , the dotted line indicates the baseline; n = 4 for each condition . ( D ) Changes in FRET of mlCNBD-FRET sperm kept in 10 μM Ca2+/1 mM BAPTA after addition of bicarbonate followed by 3 mM Ca2+ ( final: 2 mM Ca2+ ) . Data is shown as mean ± S . D; n = 4 . ( E ) Changes in pHi of wild-type sperm kept in 2 mM Ca2+ buffer after addition of 25 mM NH4Cl . Data is shown as mean ± S . D . ; n = 3 . ( F ) Changes in pHi of wild-type sperm kept in 2 mM Ca2+ buffer after addition of 3 mM BAPTA ( final: 443 nM Ca2+ , black ) or buffer ( red ) as a control . Data represents the mean of n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 014 Next , we studied basal SACY activity ( in the absence of bicarbonate ) when Ca2+ was stepped to various lower [Ca2+]o ( Figure 7C , D ) . Surprisingly , stepping [Ca2+]o from 2 mM to 20 μM moderately increased cAMP levels ( Figure 7C ) . A similar increase was observed when [Ca2+]o was decreased from 2 mM to 443 nM orfrom 10 μM to 2 . 2 nM ( Figure 7C ) . Stimulation with bicarbonate at 2 . 2 nM [Ca2+]o did not further enhance cAMP levels; however , when [Ca2+]o was re-adjusted to 2 mM , bicarbonate was able to stimulate SACY activity ( Figure 7D ) . As a control , we analyzed whether changing [Ca2+]o has an effect on the intracellular pHi . Thus , we measured changes in pH using BCECF in wild-type mouse sperm after changing [Ca2+]o . Reducing [Ca2+]o to 443 nM by addition of 3 mM BAPTA did not change pHi ( Figure 7F ) . We conclude that the △[Ca2+]o-evoked changes in FRET reflect changes in cAMP rather than in pHi . In summary , these results demonstrate that stimulation of SACY by bicarbonate requires Ca2+ and that Ca2+ alone , in the absence of bicarbonate , inhibits rather than activates SACY . Although components of cAMP signaling are also localized to primary cilia ( Berbari et al . , 2007; Bishop et al . , 2007; Händel et al . , 1999 ) , the physiological function of cAMP in primary cilia is ill defined . The analysis has been hampered by the lack of suitable tools to manipulate and analyze cAMP signaling in such a tiny cellular compartment . We set out to target mlCNBD-FRET to primary cilia . Protein import into the cilium is tightly regulated by the intraflagellar transport machinery ( Rosenbaum and Witman , 2002 ) . For targeting to cilia , we fused the mlCNBD-FRET sensor to the C-terminus of the ciliary somatostatin receptor 3 ( SSTR3 ) ( Händel et al . , 1999 ) ; as control , we fused a green-fluorescent protein ( eGFP ) to SSTR3 ( Figure 8A ) . Both constructs were expressed in IMCD3 cells , which carry primary cilia . In fact , both constructs were targeted to primary cilia ( Figure 8B , C ) . To test whether fusion of mlCNBD-FRET to SSTR3 affects the sensor properties , SSTR3-mlCNBD-FRET was expressed in HEK293 cells . Compared to mlCNBD-FRET , SSTR3-mlCNBD-FRET was localized to intracellular membranes rather than uniformly distributed throughout cells ( Figure 3B vs . 8D ) . Stimulation with isoproterenol ( 20 μM ) increased the FRET ratio by approximately 25% compared to 40% for the non-tagged sensor ( n = 9 , Figure 3E vs . 8E ) . Thus , the SSTR3-mlCNBD-FRET fusion protein allows measuring cAMP dynamics . The functionality of mlCNBD-FRET in primary cilia of IMCD-3 cells was tested by acceptor photobleaching experiments . Upon bleaching , the cerulean emission increased ( n = 11 , 43 ± 21% ) , demonstrating that the SSTR3-fusion protein undergoes FRET ( Figure 8F ) . In summary , mlCNBD-FRET is a versatile tool that can be used to measure cAMP dynamics with nanomolar sensitivity in solution , in different cell types , and small cellular compartments like the cilium or flagellum . 10 . 7554/eLife . 14052 . 015Figure 8 . Targeting mlCNBD-FRET to primary cilia . ( A ) Strategy to target a protein to cilia . The somatostatin receptor 3 ( SSTR3 ) has been fused to green fluorescent protein ( eGFP ) or mlCNBD-FRET . ( B ) Expression of eGFP in primary cilia of IMCD3 cells . An anti-acetylated tubulin antibody has been used as a marker for primary cilia . DNA has been labeled using DAPI . Scale bar is indicated . ( C ) Expression of mlCNBD-FRET in primary cilia of IMCD3 cells . Citrine fluorescence indicates the expression of mlCNBD-FRET . An anti-acetylated tubulin antibody has been used as a marker for primary cilia . DNA has been labeled using DAPI . Scale bar is indicated . ( D ) Representative image for HEK293 cells expressing SSTR3-mlCNBD-FRET . ( E ) Changes in FRET in HEK293 cells expressing SSTR3-mlCNBD-FRET ( see D ) after stimulation with 20 μM isoproterenol ( black ) or buffer only ( grey ) . FRET has been measured using fluorescence microscopy . Data are presented as mean ± S . D . ; n = 9 for each condition . ( F ) Acceptor photobleaching . The citrine ( acceptor ) fluorescence of mlCNBD-FRET in IMCD3 cells was bleached for 2 min with a mercury lamp using a 510/20 nm filter . A representative image is shown . The cerulean emission was recorded before and after acceptor photobleaching . Relative fluorescence intensities are color-coded from low ( blue ) to high ( red ) . Scale bars are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 14052 . 015 We developed a novel FRET-based biosensor to quantitatively measure cAMP dynamics . The sensor is quite versatile . First , unlike any other cAMP biosensor , the purified mlCNBD-FRET protein can be used in solution . Remarkably , the properties in solution and inside cells are by and large similar . Second , mlCNBD-FRET can be selectively targeted to subcellular compartments like primary cilia or sperm flagella to directly measure cAMP dynamics . Third , the sensor exhibits a significantly improved cAMP sensitivity compared to other single-protein biosensors , e . g . Epac-based FRET sensors . Thereby , it extends the range of accessible cAMP concentrations to the low nanomolar range . In fact , other cAMP sensors are unable to quantify resting cAMP levels in some cells ( Börner et al . , 2011 ) . Finally , the sensor is compatible with fluorescence intensity and lifetime-based approaches . These favorable characteristics prompted us to revisit a lingering debate about cAMP signaling in sperm . One issue concerns the presence , localization , and function of tmACs in sperm ( Brenker et al . , 2012; Buffone et al . , 2014; Vacquier et al . , 2014; Wertheimer et al . , 2013 ) ; for a detailed account of pros and cons see ( Brenker et al . , 2012 ) . SACY is the major source for cAMP in sperm and total cAMP levels in SACY-/- sperm are below the detection limit ( Xie et al . , 2006 ) . Members of the tmAC family and the corresponding G proteins reportedly localize to the sperm head , i . e . they are spatially segregated from SACY and PKA in the flagellum ( Wertheimer et al . , 2013 ) . Our result that bicarbonate , but not NKH477 , increases flagellar cAMP levels strengthens the notion that SACY represents the only cAMP source in the flagellum . Furthermore , considering that the head volume is much larger than that of the flagellum , our results also argue that the flagellum either is sealed up and that cAMP in the head does not propagate to the flagellum , or that NKH477-sensitive tmACs are absent altogether . A physical barrier that restricts diffusion of cAMP between midpiece and head is unlikely to exist , because Ca2+ entering the flagellum via CatSper channels propagates to the head ( Servin-Vences et al . , 2012; Xia et al . , 2007 ) . However , PDEs might serve as gatekeepers that prevent rapid cAMP exchange between head and flagellum . Another conundrum concerns the relationship between cAMP signaling and tyrosine phosphorylation during capacitation . An important requisite for capacitation in vitro is bicarbonate and the ensuing rise of cAMP followed by PKA activation . PKA activity must be maintained for at least 30 min to initiate tyrosine phosphorylation ( Morgan et al . , 2008 ) , which requires 90 min to complete . However , in mice , total cAMP levels return to basal values within 10 min ( Figure 5D ) . This begs the question how PKA activity is sustained when cAMP levels recover during bicarbonate stimulation . We resolve this conundrum in mouse sperm by showing that the free cAMP concentration stays up during measurements ( Figure 5D ) . Thus , total cAMP levels do not properly reflect free cAMP concentrations in live cells . This might be due to compartmentalization of cAMP signaling in sperm , similar to what has been shown for cardiomyocytes ( Zaccolo et al . , 2006 ) , or due to buffering by cAMP-binding proteins . A case in point is the rod photoreceptor ( Yau , 1994 ) . The free cGMP concentration in the dark is approximately 1 μM , whereas most of the total cGMP content ( 50 μM ) is bound to high-affinity sites of PDEs ( Cote et al . , 1984; 1986 ) . Light stimulation activates PDE , cGMP levels drop , and cyclic nucleotide-gated ion channels close , giving rise to a hyperpolarizing light response . Notably , a drop of the total cGMP concentration ( maximally 50% ) requires light intensities that are orders of magnitude larger than those that saturate the light response ( Cote et al . , 1984; 1986 ) . Likewise , the total cGMP concentration in sea urchin sperm is a few µM , whereas , at rest , the free cGMP must be ≤3 nM ( Bönigk et al . , 2009; Kaupp et al . , 2003 ) . The Ca2+ response in sea urchin sperm saturates at ≤20 pM of the chemoattractant , whereas approximately 5 nM chemoattractant are required to detect a noticeable rise of total cGMP concentration ( Kaupp et al . , 2003 ) . A third issue concerns the regulation of SACY by Ca2+ . SACY activity is stimulated by bicarbonate in buffer containing 2 mM Ca2+ , but not in low micromolar Ca2+ buffer ( Figure 7A ) ( Carlson et al . , 2007; Navarrete et al . , 2015; Wennemuth et al . , 2003 ) . Recent findings suggest that Ca2+ has a biphasic role in the regulation of cAMP signaling ( Navarrete et al . , 2015 ) : at micromolar Ca2+-conditions , bicarbonate does not stimulate PKA activity or increase tyrosine phosphorylation; however , chelating [Ca2+]o using EGTA in the presence of bicarbonate relieves the inhibition , stimulating PKA activity , and increasing tyrosine phosphorylation . Similarly , in CatSper1-/- sperm , [Ca2+]i is lower than in wild-type sperm ( Ren et al . , 2001 ) , yet tyrosine phosphorylation in the presence of bicarbonate is enhanced ( Chung et al . , 2014 ) . One interpretation of these seemingly puzzling results suggests that bicarbonate can stimulate SACY at normal and very low [Ca2+]i , but not at intermediate Ca2+ levels ( Navarrete et al . , 2015 ) . However , such a biphasic control by Ca2+ has not been observed in vitro . Instead , the activity of isolated SACY protein upon bicarbonate stimulation is steadily enhanced by Ca2+ in a dose-dependent manner ( Jaiswal and Conti , 2003 ) . Alternatively , Ca2+ might control cellular signaling downstream of cAMP . Here , we show that cAMP levels do not significantly change during bicarbonate stimulation at very low Ca2+ concentrations ( 2 nM ) ; thus , a bicarbonate-dependent cAMP increase is not underlying the stimulation of PKA activity and tyrosine phosphorylation under these conditions . However , under conditions that are expected to significantly lower [Ca2+]i , basal SACY activity is stimulated , indicating that SACY activity is controlled by a Ca2+-dependent negative feedback that operates in intact sperm , but is absent on the isolated or heterologously expressed protein . Regulation of SACY activity via another negative feedback has been described before ( Burton and McKnight , 2007; Nolan et al . , 2004 ) . In sperm lacking the catalytical PKA subunit Cα2 , both basal and bicarbonate-stimulated cAMP levels are higher than in wild-type sperm ( Burton and McKnight , 2007; Nolan et al . , 2004 ) , suggesting that PKA-dependent phosphorylation indirectly or directly down-regulates SACY activity . cAMP-signaling components are not uniformly distributed throughout sperm: PKA and SACY are exclusively localized to the flagellum ( Hess et al . , 2005; Nolan et al . , 2004; Wertheimer et al . , 2013 ) . However , even along the flagellum , cAMP signaling is compartmentalized; the cAMP dynamics stimulated by bicarbonate is different in the midpiece and the principal piece . We favor the idea that distribution and activity of PDEs is different in the midpiece and principal piece . PDE are targeted to distinct domains in sperm ( Bajpai et al . , 2006; Lefièvre et al . , 2002 ) . Moreover , at least two PDE isoforms and splice variants , PDE4 ( cAMP-specific ) and PDE1 ( Ca2+/calmodulin-dependent ) are involved in the control of sperm motility and tyrosine phosphorylation ( Fisch et al . , 1998; Leclerc et al . , 1996; Lefièvre et al . , 2002; Visconti et al . , 1995 ) . Using mlCNBD-FRET mice , it will now be possible to analyze cAMP dynamics in different subcellular compartments and unravel the mechanisms underlying the differential regulation of cAMP signaling . Finally , primary cilia constitute a unique Ca2+ compartment that is functionally distinct from the cytoplasm of the cell body ( Delling et al . , 2013; DeCaen et al . , 2013 ) . Ca2+ channels ( PKD-Like1/2 ) in the plasma membrane of primary cilia maintain a higher resting Ca2+ concentration compared to the cell body , despite steady Ca2+ efflux via the ciliary base ( Delling et al . , 2013; DeCaen et al . , 2013 ) . Considering that primary cilia also host cAMP-signaling components ( Johnson and Leroux , 2010 ) , we hypothesize that cAMP signaling in primary cilia might also be functionally distinct from the cell body . The combination of sensitive genetically-encoded cAMP biosensors and light-driven actuators ( Jansen et al . , 2015 ) provides the high spatial and temporal resolution that is needed to unravel cAMP-signaling pathways in cilia and flagella . The mlCNBD cDNA sequence ( cyclic nucleotide binding-domain ) of the cyclic nucleotide-gated K+ channel from Mesorhizobium loti ( MAFF303099 , mll3241 ) was amplified via PCR ( Cukkemane et al . , 2007 ) . For expression in mammalian cell lines , citrine and cerulean were amplified by PCR and fused to the N- and C-terminus of the mlCNBD via BamHI/HindIII or XhoI/ApaI , respectively . The C-terminus of cerulean contained a histidine ( His10 ) tag . The PCR product was cloned into a pcDNA3 . 1 ( + ) vector ( Invitrogen , Darmstadt , Germany ) using BamHI and ApaI ( pc3 . 1-mlCNBD-FRET ) . For expression in E . coli , citrine was amplified by PCR from pc3 . 1-mlCNBD-FRET and an NdeI site was added at the 5’end to allow in-frame cloning into the pET21a vector ( Novagen , Darmstadt , Germany ) . The mlCNBD fused to cerulean containing a His10 tag was amplified by PCR from pc3 . 1-mlCNBD-FRET and an EcoRI site was added at the 3’end . PCR products were cloned into pET21a using NdeI/EcoRI ( pET21a-mlCNBD-FRET ) . The mutation encoding R307Q was introduced using the QuikChange site-directed mutagenesis protocol ( Agilent Technologies , Santa Clara , CA ) . The mlCNBD-FRET protein was expressed in E . coli ( BL21 DE3 pLysE ) . Bacteria were grown at 37°C until density reached an OD600 = 0 . 6 . Then , expression of mlCNBD-FRET was induced using 0 . 6 mM IPTG . The cells were grown overnight at 20°C and harvested after 18 hr . The cells were lysed in lysis buffer ( 20 mM NaP , 500 mM NaCl , 10 mM DNAse , pH 7 . 4 ) using ultrasonification followed by centrifugation at 45 , 000 g , 20 min , and 4°C . The supernatant containing the mlCNBD-FRET protein was loaded onto a 1 ml HiTrap cobalt-IMAC column ( GE ) pre-equilibrated with the binding buffer ( 20 mM NaP , 500 mM NaCl , 10 mM imidazole , pH 7 . 4 ) in the Äkta HPLC system ( GE Healthcare , Solingen , Germany ) . After loading , the column was washed with 5 column volumes of binding buffer . Bound cAMP was removed from the protein by incubation for 10 min with 1 ml 8-CPT-cGMP ( 5 mM ) ( Biolog , Bremen , Germany ) . This step was performed three times followed by a washing step with 10 ml binding buffer . The mlCNBD-FRET was then eluted using a linear gradient ( 0–100% ) of elution buffer ( 20 mM Na3PO4 , 500 mM NaCl , 500 mM imidazole , pH 7 . 4 ) . To enrich properly folded protein , gel-filtration chromatography of mlCNBD-FRET was carried out using a Superdex 200 , Hi-Load 16/60 column ( Amersham Biosciences/GE Healthcare ) , which allows separating proteins with molecular weight between 30 and 200 kDa . The mlCNBD-FRET protein has an apparent molecular weight of 70 . 9 kDa . The column was washed and equilibrated with 2 column volumes of the gel-filtration buffer ( GF buffer , 10 mM K2HPO4 , 100 mM KCl , pH 7 . 4 ) . Subsequently , the sample was injected onto the column and protein elution was followed at 280 nm . The protein fractions corresponding to the monomer were collected . This sample was flash frozen in GF buffer containing 10% glycerol and stored at -80°C . The purified mlCNBD-FRET still contained cAMP ( Cukkemane et al . , 2007 ) . The cAMP was removed by incubation with the low-affinity analog 8-CPT-cGMP ( Cukkemane et al . , 2007 ) followed by extensive washing . To determine the cAMP content , protein samples were denatured and analysed as described before ( Cukkemane et al . , 2007 ) using a Zorbax SB-C18 column ( 5 μm , 250 mm x 0 . 46 mm , Agilent , Waldbronn , Germany ) . The column was calibrated and the detection limit for cAMP was determined by injecting increasing amounts of cAMP prepared in GF buffer . A linear gradient was produced from solution A ( 5 mM KH2PO4 , pH 5 . 0 ) and solution B ( 80% methanol ) at a flow rate of 0 . 3 ml/min . The absorbance for cAMP was detected at 256 nm with a retention time of 6 . 3 min . The final cAMP content of mlCNBD-FRET was below the detection limit . To determine the affinity of the purified mlCNBD-FRET , known protein concentrations ( ε = 53 , 290 M-1cm-1 , MW = 70 . 9 kDa , calculated by ProtParam tool ) were mixed with increasing concentrations of cAMP ( Sigma-Aldrich , Seelze , Germany ) . The fluorescence was observed in a spectrofluorometer ( Quantamaster 40 , PTI ) with excitation at 430 nm ( cerulean ) and emission at 471 nm ( cerulean ) and 529 nm ( citrine ) . Fluorescence spectra were recorded using 430 nm excitation . FRET was calculated from the ratio of the emission intensities for cerulean at 471 nm over citrine at 529 nm . The change in FRET △F was plotted versus the cAMP concentration and fitted with a substrate-depletion model ( Cukkemane et al . , 2007 ) to calculate the KD value for cAMP . To determine the effect of pH on the mlCNBD-FRET sensor , the protein was exposed to different pH solutions ( pH 6–8 ) . The buffer solutions consisted of 100 mM KCl/10 mM HEPES and the pH was adjusted using 1 M NaOH . The protein was kept in 100 mM KCl/10 mM HEPES pH 7 . 4 . For measurements , the protein solution ( 50 μM stock ) was mixed 1:1000 with the corresponding pH buffer and the final pH of the solution was controlled again using a pH meter . Fluorescence spectra were recorded in a spectrofluorometer ( Quantamaster 40 , PTI ) with excitation at 430 nm . The binding affinity was determined as described for the ligand-binding assay . For time-resolved recordings of FRET changes , purified mlCNBD-FRET protein was mixed 1:1 with cAMP in GF buffer in a rapid-mixing device ( SFM400 , Bio-Logic ) at a flow-rate of 9 . 5 ml/s . Cerulean was excited by the blue LED of a SpectraX Light Engine ( Lumencor , Beaverton , OR ) and the excitation light was passed through a 435/24 nm filter ( Semrock Brightline , Rochester , NY ) onto the μFC08 cuvette . The emission was collected through 485/15 nm ( cerulean ) and 535/15 nm filters ( citrine , Semrock Brightline ) . Fluorescence emission was recorded by photomultiplier modules ( 9656–20; Hamamatsu Photonics , Herrsching , Germany ) . Data acquisition was performed at 200 Hz with a data acquisition pad ( PCI-6221; National Instruments , Austin , TX ) and the Bio-Kine software v . 4 . 49 ( Bio-Logic ) . Numerical analysis was carried out with Dynafit ( v 3 . 28 . 070 , BioKin , Watertown , MA ) using a simple binding scheme . E + L ⇄KoffKon EL E , L , EL refer to the concentrations of the receptor , ligand , and receptor-ligand complex , respectively . kon describes the on rate for ligand binding along with the associated conformational change resulting in a FRET signal , koff describes the off rate of ligand dissociation from the receptor and the associated return to its resting-state conformation . HEK293 cells ( ATCC CRL-1573 , authentication method: STR profiling , mycoplasma test negative ) were electroporated with pc3 . 1-mlCNBD-FRET or pc3 . 1-mlCNBD-FRET-R307Q using the Neon 100 µl kit ( Invitrogen , Darmstadt , Germany ) and the MicroPorator ( Digital Bio , Invitrogen ) according to the manufacture’s protocol ( 3x 1245 mV pulses with 10 ms pulse width ) . Cells were transferred into complete medium composed of DMEM plus GlutaMax ( Life Technologies GmbH , Carlsbad , CA ) and 10% fetal bovine serum ( Biochrom , Berlin , Germany ) . For the selection of monoclonal HEK293 cells stably expressing mlCNBD-FRET , the antibiotic G418 ( 800 µg/ml , Invitrogen ) was added 24 hr after electroporation . Monoclonal cell lines were identified by Western blot and immunocytochemistry . Cells were cultured in Ibidi μ-slides ( Ibidi , Planegg , Germany ) attached to a home-built gravity-flow perfusion system . Measurements were performed using a CellR live-cell imaging system ( Olympus ) with 20x objective . Cells were perfused with extracellular solution ( ES: 120 mM NaCl , 5 mM KCl , 2 mM MgCl2 , 2 mM CaCl2 , 10 mM HEPES , 10 mM glucose , pH 7 . 4 ) to determine the basal FRET ratio and then the stimulus was added ( NKH477 and IBMX , Tocris , Wiesbaden , Germany; Isoproterenol and SNP , Sigma-Aldrich ) . FRET was recorded by exciting cerulean at 436 nm and measuring the emission of cerulean and citrine at 470 nm and 535 nm , respectively . The change in FRET ( emissioncerulean/emissioncitrine ) was calculated , corrected for bleed-through , normalized to the baseline , and plotted versus time . For population measurement using a spectrofluorometer ( Quantamaster 40 , PTI ) , HEK293 cells were trypsinized at 37°C for 5 min and washed twice with ES . Measurements were performed in PMMA cuvettes at 1x105 cells/ml under constant stirring ( Spinbar , Bel-art products , Wayne , NJ ) . Fluorescence spectra at steady-state were recorded with excitation at 430 nm and 0 . 5 s integration time . For time-resolved measurements , cells were excited at 430 nm and emission was recorded at 470 nm ( cerulean ) and 530 nm ( citrine ) at an acquisition frequency of 1 Hz . The change in FRET ( emissioncerulean/emissioncitrine ) was calculated , corrected for bleed-through , normalized to the baseline , and plotted versus time . For repetitive stimulation experiments , cells were stimulated with 500 nM isoproterenol and fluorescence was measured in a plate reader ( FLUOstar Omega; BMGLabtech ) at 29°C with excitation at 440 nm and emission at 485 and 520 nm . The wash-out was performed by removing the stimulus and addition of ES . The fluorescence emission ratio was normalized to the initial baseline fluorescence ratio and plotted as a function of time . A hemagglutinin ( HA ) tag was fused to the C-terminus , an EcoRI restriction site was added to the 5’ end , the internal BamHI site was deleted , and a BamHI and XbaI restriction site was added to the 3’ end by PCR . The PCR product was cloned into a pBluescript SK- vector ( Agilent Technologies , Santa Clara , USA ) using EcoRI and XbaI ( pB-mlCNBD-FRET ) . After sequencing , the mlCNBD-FRET-HA insert was excised and cloned into pPrCExV ( kind gift from Robert Braun , Jackson Laboratory ) using BamHI ( pPrCExV-mlCNBD-FRET ) to express mlCNBD-FRET under the control of the protamine-1 promotor that is exclusively active in post-meiotic spermatids ( Zambrowicz et al . , 1993 ) . Transgenic mice were generated via pronuclear injection following standard procedures ( Ittner and Götz , 2007 ) at the transgenic facility of the LIMES ( University of Bonn , Germany; licence number: 84–02 . 04 . 2012 . A192 ) . Mice were genotyped by PCR using mlCNBD-FRET-specific primers ( #1: 5’-GTACAAGGGTACCCAAGAAGTCCGTCGC-3’ , #2: 5’-CGAAGCACTGCACGCCCCAGGTC-3’ ) . Mice used in this study were 2–5 months of age . All animal experiments were in accordance with the relevant guidelines and regulations . Protein lysates were obtained by homogenizing cells or tissue in lysis buffer ( 10 mM Tris/HCl , pH 7 . 6 , 140 mM NaCl , 1 mM EDTA , 1% Triton X-100 , mPIC protease inhibitor cocktail 1:500 ) followed by trituration through a 18-gauge needle . Samples were incubated for 30 min on ice and centrifuged at 10 , 000 g for 5 min at 4°C . Prior to separation by SDS-PAGE , samples were mixed with 4x SDS loading-buffer ( 200 mM Tric/HCl , pH 6 . 8 , 8% SDS ( w/v ) , 4% β-mercaptoethanol ( v/v ) , 50% glycerol , 0 . 04% bromophenol blue ) and heated for 5 min at 95°C . Sperm samples used for SDS-PAGE were washed with 1 ml PBS and sedimented by centrifugation at 5000 g for 5 min . 1–2 x 106 cells were resuspended in 4 x SDS loading buffer and heated for 5 min at 95°C . For Western-blot analysis , proteins were transferred onto PVDF membranes ( Merck Millipore , Billerica , USA ) , probed with antibodies , and analysed using the Odyssey Imaging System ( LI-COR , Lincoln , NE ) detection-system . Primary antibodies: anti-HA 3F10 ( 1:5000; Roche , Basel , Switzerland ) , anti-α-tubulin ( 1:5000; Sigma-Aldrich ) anti-GFP antibody ( 1:5000; abcam , Cambridge , UK ) ; secondary antibodies: IRDye680 and IRDye800 antibodies ( 1:20 , 000 , LI-COR ) ; ICC: fluorescently-labeled antibodies ( 1:500; Dianova , Hamburg , Germany ) . Testes were fixed in 4% paraformaldehyde/PBS overnight , cryo-protected in 10 and 30% sucrose , and embedded in Tissue-Tek ( Sakura Finetek , Alphen aan den Rijn , Netherlands ) . To block unspecific binding sites , cryosections ( 16 µm ) were incubated for 1 hr with blocking buffer ( 0 . 5% Triton X-100 and 5% ChemiBLOCKER ( Merck Millipore ) in 0 . 1 M NaP , pH 7 . 4 ) . The primary anti-HA 3F10 antibody ( rat monoclonal; Roche , Basel , Switzerland ) was diluted 1:1000 in blocking buffer and incubated for 2 hr . Fluorescent secondary antibodies ( donkey anti-rat CY3; Dianova ) were diluted 1:500 in blocking buffer containing 0 . 5 mg/ml DAPI ( Life Technologies ) and pictures were taken on a confocal microscope ( FV1000; Olympus ) . Sperm were isolated by incision of the cauda followed by a swim-out in modified TYH medium ( 135 mM NaCl , 4 . 8 mM KCl , 2 mM CaCl2 , 1 . 2 mM KH2PO4 , 1 mM MgSO4 , 5 . 6 mM glucose , 0 . 5 mM sodium pyruvate , 10 mM L-lactate , 10 mM HEPES , pH 7 . 4 ) . For capacitation , the medium contained 3 mg/ml BSA and 25 mM of NaCl was substituted with 25 mM NaHCO3 . The pH was adjusted at 37°C . After 15–30 min swim-out at 37°C , sperm were collected and counted . After swim-out , sperm were kept in 500 μl tubes filled up with TYH buffer to avoid pH changes due to dissolution of ambient CO2 into the buffer . Measurements were performed in TYH buffer containing 1–5 x 104 sperm/ml . The fluorescence recording was performed at a spectrofluorometer ( Quantamaster 40 , PTI ) at 37°C as described for the mammalian cells lines . Nominally Ca2+-free TYH buffer was prepared by adding 10 μM Ca2+ to TYH buffer prepared without any Ca2+ . When adding 25 mM NaHCO3 ( 1:10 from 250 mM stock ) , the Na+ concentration in modified TYH was reduced to 110 mM NaCl . BAPTA ( 1 , 2-Bis ( 2-aminophenoxy ) ethane-N , N , N′ , N′-tetraacetic acid tetrapotassium salt , Sigma-Aldrich ) was dissolved in the modified TYH buffer ( pH 7 . 4 ) and added to sperm samples at a final concentration of 1 , 2 , or 3 mM . Cells were loaded with 10 μM ( HEK293 cells ) or 5 μM BCECF-AM ( sperm ) ( Invitrogen ) for 10 min at 37°C and 10% CO2 . Afterwards , HEK293 cells were washed three times in 1 ml ES before starting the measurement . For sperm , the dye was removed by single centrifugation ( 700 x g , 7 min , RT ) and resuspension in TYH buffer . For HEK293 cells , fluorescence was measured in a plate reader ( FLUOstar Omega; BMGLabtech ) at 29°C . For calibrating the intracellular pH ( pHi ) using the null-point method , cells were permeabilized with Triton‐X 100 ( final concentration of 0 . 1% ) followed by addition of different pH buffer solutions ( pH 5–8 ) . BCECF was excited at 440 nm and 485 nm and the emission was detected at 520 nm . The fluorescence emission ratio at both excitation channels was normalized to the initial fluorescence ratio and plotted as a function of pH . A linear regression analysis ( Origin; OriginLab ) was performed to calculate the pHi . Changes in pHi in sperm were either analysed in a rapid-mixing device ( SFM-400; Biologic ) in the stopped-flow mode or in a spectrofluorometer ( Quantamaster 40 , PTI ) at 37°C . In the stopped-flow device , the sperm suspension ( 5·106 sperm/ml ) was rapidly mixed 1:1 ( v/v ) at a flow rate of 0 . 5 ml/s with the respective stimulants . Fluorescence was excited by a SpectraX Light Engine ( Lumencor ) , whose intensity was modulated with a frequency of 10 kHz . The excitation light was passed through a BrightLine 452/45 nm filter ( Semrock ) onto the cuvette ( FC-15 , Biologic ) . Emission light was recorded in a dual-emission mode using BrightLine 494/20-nm and BrightLine 540/10-nm filters ( Semrock ) by photomultiplier modules ( H10723-20; Hamamatsu Photonics ) . The signal was amplified and filtered through a lock-in amplifier ( 7230 DualPhase , Ameteky ) . Data acquisition was performed with a data acquisition pad ( PCI-6221; National Instruments ) and Bio-Kine software v . 4 . 49 ( Bio-Logic , Illingen , Germany ) . The pH signal represent the ratio of F494/540 and is depicted as the percent of the relative change in ratio ( ΔR/R ) with respect to the mean of the first three data points at the onset of the signal . The control ( TYH ) signal was subtracted from the NH4Cl or bicarbonate traces . To analyze the effect of changing the extracellular Ca2+ on the intracellular pH , measurements were performed at the spectrofluorometer in respective TYH buffer containing 1 x 104 sperm/ml . BAPTA ( 1 , 2-Bis ( 2-aminophenoxy ) ethane-N , N , N′ , N′-tetraacetic acid tetrapotassium salt , Sigma-Aldrich ) was dissolved in the modified TYH buffer ( pH 7 . 4 ) and added to the sperm samples at a final concentrations of 3 mM . BCECF was excited at 452 nm and the emission was detected at 494 nm and 540 nm . The fluorescence emission ratio of 540/494 was calculated and plotted against time . The ratio was corrected for a drift in fluorescence . The drift was determined by calculating the slope of the response after addition of buffer experiment , which was multiplied to the time axis to calculate the drift over time . The drift was subtracted from the sample data . The drift corrected data was finally normalized to the initial emission ratio . After stimulation with 25 mM NaHCO3 , the reaction was quenched with HClO4 ( 1:3 ( v/v ) ; 0 . 5 M final concentration ) . Samples were frozen in liquid N2 , thawed , and neutralized by addition of K3PO4 ( 0 . 24 M final concentration ) . The salt precipitate and cell debris were sedimented by centrifugation ( 15 min , 15 , 000 g , 4°C ) . The cAMP content in the supernatant was determined by a competitive immunoassay ( Molecular Devices , Sunnyvale , CA ) including an acetylation step for higher sensitivity . Calibration curves were obtained by serial dilutions of cAMP standards . 96-well plates were analysed by using a microplate reader ( FLUOstar Omega; BMGLabtech ) . To determine the basal concentration of cAMP in HEK293 cells , the null-point method was used . HEK293 cells stably expressing mlCNBD-FRET were trypsinized and cells were suspended in 1 ml ES buffer ( 5 x 105 cells/ml ) . The fluorescence was recorded for 5 min in a spectrofluorometer ( Quantamaster 40 , PTI ) with excitation at 430 nm and emission at 471 nm ( cerulean ) and 529 nm ( citrine ) . Cells were then permeabilized with 20 μM digitonin to deplete intracellular cAMP and the concomitant FRET signal was measured . During permeabilization , the pHi might change if the extracellular and intracellular pH is different . Using the pH indicator BCECF , we determined a pHi of 7 . 2 ± 0 . 1 ( n = 4 ) in mlCNBD-FRET-expressing HEK293 cells . Accordingly , we set pHo during permeabilization to 7 . 2 to assure that no pH change occurs . Increasing concentrations of cAMP were then added to the cell suspension that eventually saturate the FRET sensor . The FRET ratio for each cAMP concentration was plotted against the cAMP concentrations and analysed by linear regression . The basal cAMP concentration was defined as the external cAMP concentration where no change in FRET was observed . The decay of the fluorescence lifetime was measured on a Trimscope II ( LaVision BioTec , Göttingen , Germany ) equipped with a FLIM x16 TCSPC detector . Cells were imaged using a 20x NA = 1 . 0 water-dipping objective and excited with an 80 MHz mode-locked laser tuned to 810 nm . Emitted photons were filtered through a 700 nm short-pass filter followed by a 495 nm dichroic mirror that reflected the fluorescence below 488 nm towards the FLIM detector . Images were acquired with 512x512 pixels , pixel size = 800 nm . The instrument response function ( IRF ) was obtained from second harmonic signals generated from a urea sample at the same excitation wavelength . Cells were imaged in ES in culture dishes at 23°C . To increase the intracellular cAMP concentration , cells were stimulated for 5 min using 40 µM NKH/500 μM IBMX . As a control , 0 . 05% DMSO was used . To permeabilize the cells and reduce the intracellular cAMP concentration , cells were treated with 20 μM digitonin for 10 min . To measure the maximal response , cells were incubated with 5 μM cAMP . Time constants of fluorescence decay were calculated using the FLIMfit software tool ( Warren et al . , 2013 ) . The fitting function consisted of a double exponential decay model that was convolved with the IRF . Pixels below 20 counts were excluded from the analysis . 15x15 pixels were then binned to increase photon count and improve the accuracy of the fit . The analysis generated two time constants ( τ1 and τ2 ) associated with the double exponential decay . Individual amplitudes ( a1 and a2 ) reflecting the magnitude of each decay constant were used to calculate the weighted average τaverage for every pixel using the following equation:τaverage=a1×τ1+a2×τ2a1+a2 Decay curves were recorded by Time-Correlated Single-Photon Counting ( TCSPC ) using the FluoTime 300 ( PicoQuant ) . Cells in solution ( 106 cells/ml in ES ) or purified protein ( 1 μM ) were measured in a cuvette and excited by a pulsed diode laser at 440 nm ( PicoQuant ) and fluorescence emission was detected at 480/20 nm . The IRF was obtained using Ludox suspension . The cerulean lifetime decay was measured for cells in ES before stimulation with 40 µM NKH/500 µM IBMX . Cells were suspended in 1 ml ES in a cuvette . To calibrate mlCNBD-FRET in cells using FLS , the decay of the cerulean lifetime was first recorded in ES followed by the addition of 20 µM digitonin to permeabilize the cells and then by addition of increasing cAMP concentrations . Fitting was performed using the PicoQuant FluoFit software . A double-exponential function was chosen for the analysis with iterative reconvolution using the IRF obtained from Ludox . The full-length mouse SSTR3 cloned into pEGFP-N1 was kindly provided by Greg Pazour ( UMass Medical School ) . For expression of SSTR3-tagged mlCNBD-FRET , the SSTR3 coding sequence was amplified by a recombinant PCR reaction to delete the internal NheI site . The fragment was subcloned into pc3 . 1-mlCNBD-FRET vie NheI/HindIII . During acceptor photobleaching , the fluorescence of the FRET donor is recorded while bleaching the FRET acceptor . IMCD-3 cells ( ATCC CRL-2123 , authentication method: STR profiling , mycoplasma test negative ) were transfected with pc3 . 1-SSTR3-mlCNBD-FRET using the Neon 100 µl kit ( Invitrogen ) and the MicroPorator ( Digital Bio ) according to the manufacture’s protocol . Afterwards , cells were plated on coverslips and the experiment was performed on an inverted microscope ( eclipse Ti , Nikon , Düsseldorf , Germany ) , equipped with EMCCD camera ( iXON Ultra 897 , Andor , Berlin , Germany ) , using 100x oil objective ( CFI Apo TIRF , NA 1 . 49 , Nikon ) . First , an image of the cilium was obtained in the donor channel using 405 nm laser excitation and 480/40 nm emission and an image in the acceptor channel using 488 nm laser excitation and 560/40 nm emission . Photobleaching was performed at 510/20 nm using a mercury lamp ( C-LHGFIE , Nikon ) for 2 min . This was followed by the acquisition of two images , one in the donor channel and another in the acceptor channel . The donor channel was compared to the images before and after acceptor photobleaching and the increase in intensity was quantified after background subtraction . The amount of bleaching was assessed by comparing citrine fluorescence before and after bleaching . For spatial-temporal-resolved measurements , sperm from transgenic mice were imaged with a dual-view microscopic setup . Cerulean was excited at 440/20 nm ( Lumencor Spectra X light engine ) combined with a 438/24 nm excitation filter ( Semrock ) . In addition , a dichroic mirror 455 nm DRLP and an additional filter with 460 nm ALP ( Edmund optics , Karlsruhe , Germany ) was used . Emission light was collected with an objective ( 40x , NA 0 . 95; Olympus , USA ) and detected with an EMCCD Camera ( iXon Ultra 897 , Andor ) . In the dual-view emission path , a beam splitter ( DV2 , 515 dxcr , Mag Biosystems , Exton , PA ) with a cerulean and citrine emission filter ( 475/28 nm and 535/50 nm , respectively; Semrock ) was used . Thereby , both channels were acquired simultaneously with an acquisition frequency of 1 Hz . The resulting video was analyzed with a self-developed image analysis plugin for ImageJ . First , the citrine and the cerulean channel was registered ( MultiStackReg , ImageJ ) . Second , the flagellum was tracked and the signal intensity of the citrine and the cerulean channel was determined along the flagellum . By using the pre-determined bleed-through value for the citrine channel ( 0 . 69 ± 0 . 01 ) , changes in FRET ( ( emissioncerulean – 0 . 69 x emissioncitrine ) /emissioncitrine ) was calculated , normalized to the baseline , and plotted versus time for each position along the arc length . The droplet separates the midpiece and the principal piece; thus , both regions were selected at a distance 5–10 µm apart from the droplet . The mean signal intensity and the standard deviation was determined in both regions for 20 µm along the arc length . The values were plotted over time and a logistic regression ( Origin 9 . 0 ) was used to determine the kinetics within the two regions . The parameters of the logistic regression for different experiments were used to determine the average kinetics within the two regions .	Cells can change the way they grow , move or develop in response to information from their environment . This information is first detected at the surface of the cell and then the information is relayed around the interior of the cell by signaling molecules known as “second messengers” . A molecule called cAMP is a well-known second messenger that is involved in many different signaling pathways . Therefore , the levels of cAMP in specific areas of the cell need to be precisely regulated to enable different signaling pathways to be activated at specific times and locations . Some cells have hair-like structures called cilia or flagella on their surface . Cilia and flagella are able to move the fluid that surrounds the cells or even move the cells themselves . The second messenger cAMP plays an essential role in making cilia move , but it is challenging to analyze the dynamics of cAMP – that this , how the levels of this molecule change over time – in these structures . The levels of cAMP in live cells can only be measured using fluorescent biosensors . Introducing these biosensors into specific cell structures is difficult and they are not sensitive enough to respond to low levels of cAMP . Furthermore , it is difficult to measure cAMP activity inside such tiny structures using these biosensors . Mukherjee , Jansen , Jikeli et al . now address some of these challenges by creating a new cAMP biosensor that has several unique features . Most importantly , it can respond to very low levels of cAMP , making it more sensitive than previous biosensors . Mukherjee et al . test this new biosensor in the flagella of sperm cells from mice , which reveals how the production of cAMP is regulated in the flagellum . The new biosensor also shows that different parts of the flagellum can have different cAMP dynamics . In the future , this new biosensor could be used to study cAMP in other structures and compartments within cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "cell", "biology", "tools", "and", "resources" ]	2016	A novel biosensor to study cAMP dynamics in cilia and flagella
Decision bias is traditionally conceptualized as an internal reference against which sensory evidence is compared . Instead , we show that individuals implement decision bias by shifting the rate of sensory evidence accumulation toward a decision bound . Participants performed a target detection task while we recorded EEG . We experimentally manipulated participants’ decision criterion for reporting targets using different stimulus-response reward contingencies , inducing either a liberal or a conservative bias . Drift diffusion modeling revealed that a liberal strategy biased sensory evidence accumulation toward target-present choices . Moreover , a liberal bias resulted in stronger midfrontal pre-stimulus 2—6 Hz ( theta ) power and suppression of pre-stimulus 8—12 Hz ( alpha ) power in posterior cortex . Alpha suppression in turn was linked to the output activity in visual cortex , as expressed through 59—100 Hz ( gamma ) power . These findings show that observers can intentionally control cortical excitability to strategically bias evidence accumulation toward the decision bound that maximizes reward . Perceptual decisions arise not only from the evaluation of sensory evidence , but are often biased toward a given choice alternative by environmental factors , perhaps as a result of task instructions and/or stimulus-response reward contingencies ( White and Poldrack , 2014 ) . The ability to willfully control decision bias could potentially enable the behavioral flexibility required to survive in an ever-changing and uncertain environment . But despite its important role in decision making , the neural mechanisms underlying decision bias are not fully understood . The traditional account of decision bias comes from signal detection theory ( SDT ) ( Green and Swets , 1966 ) . In SDT , decision bias is quantified by estimating the relative position of a decision point ( or ‘criterion’ ) in between sensory evidence distributions for noise and signal ( see Figure 1A ) . In this framework , a more liberal decision bias arises by moving the criterion closer toward the noise distribution ( see green arrow in Figure 1A ) . Although SDT has been very successful at quantifying decision bias , how exactly bias affects decision making and how it is reflected in neural activity remains unknown . One reason for this lack of insight may be that SDT does not have a temporal component to track how decisions are reached over time ( Fetsch et al . , 2014 ) . As an alternative to SDT , the drift diffusion model ( DDM ) conceptualizes perceptual decision making as the accumulation of noisy sensory evidence over time into an internal decision variable ( Bogacz et al . , 2006; Gold and Shadlen , 2007; Ratcliff and McKoon , 2008 ) . A decision in this model is made when the decision variable crosses one of two decision bounds corresponding to the choice alternatives . After one of the bounds is reached , the corresponding decision can subsequently either be actively reported , e . g . by means of a button press indicating a detected signal , or it could remain without behavioral report when no signal is detected ( Ratcliff et al . , 2018 ) . Within this framework , a strategic decision bias imposed by the environment can be modelled in two different ways: either by moving the starting point of evidence accumulation closer to one of the boundaries ( see green arrow in Figure 1B ) , or by biasing the rate of the evidence accumulation process itself toward one of the boundaries ( see green arrow in Figure 1C ) . In both the SDT and DDM frameworks , decision bias shifts have little effect on the sensitivity of the observer when distinguishing signal from noise; they predominantly affect the relative response ratios ( and in the case of DDM , the speed with which one or the other decision bound is reached ) . There has been some evidence to suggest that decision bias induced by shifting the criterion is best characterized by a drift bias in the DDM ( Urai et al . , 2018; White and Poldrack , 2014 ) . However , the drift bias parameter has as yet not been related to a well-described neural mechanism . Regarding the neural underpinnings of decision bias , there have been a number of reports about a correlational relationship between cortical population activity measured with EEG and decision bias . For example , spontaneous trial-to-trial variations in pre-stimulus oscillatory activity in the 8—12 Hz ( alpha ) band have been shown to correlate with decision bias and confidence ( Iemi and Busch , 2018; Limbach and Corballis , 2016 ) . Alpha oscillations , in turn , have been proposed to be involved in the gating of task-relevant sensory information ( Jensen and Mazaheri , 2010 ) , possibly encoded in high-frequency ( gamma ) oscillations in visual cortex ( Ni et al . , 2016; Popov et al . , 2017 ) . Although these reports suggest links between pre-stimulus alpha suppression , sensory information gating , and decision bias , they do not uncover whether pre-stimulus alpha plays an instrumental role in decision bias and how exactly this might be achieved . Specifically , it is unknown whether an experimentally induced shift in decision bias is implemented in the brain by willfully adjusting pre-stimulus alpha in sensory areas . Here , we explicitly investigate these potential mechanisms by employing a task paradigm in which shifts in decision bias were experimentally induced within participants through ( a ) instruction and ( b ) asymmetries in stimulus-response reward contingencies during a visual target detection task . By applying drift diffusion modeling to the participants’ choice behavior , we show that the effect of strategically adjusting decision bias is best captured by the drift bias parameter , which is thought to reflect a bias in the rate of sensory evidence accumulation toward one of the two decision bounds . To substantiate a neural mechanism for this effect , we demonstrate that this bias shift is accompanied by changes in pre-stimulus midfrontal 2–6 Hz ( theta ) power , as well as changes in sensory alpha suppression . Pre-stimulus alpha suppression in turn is linked to the post-stimulus output of visual cortex , as reflected in gamma power modulation . Critically , we show that gamma activity accurately predicted the strength of evidence accumulation bias within participants , providing a direct link between the proposed mechanism and decision bias . Together , these findings identify a neural mechanism by which intentional control of cortical excitability is applied to strategically bias perceptual decisions in order to maximize reward in a given ecological context . In three EEG recording sessions , human participants ( N = 16 ) viewed a continuous stream of horizontal , vertical and diagonal line textures alternating at a rate of 25 textures/second . The participants’ task was to detect an orientation-defined square presented in the center of the screen and report it via a button press ( Figure 2A ) . Trials consisted of a fixed-order sequence of textures embedded in the continuous stream ( total sequence duration 1 s ) . A square appeared in the fifth texture of a trial in 75% of the presentations ( target trials ) , while in 25% a homogenous diagonal texture appeared in the fifth position ( nontarget trials ) . Although the onset of a trial within the continuous stream of textures was not explicitly cued , the similar distribution of reaction times in target and nontarget trials suggests that participants used the temporal structure of the task even when no target appeared ( Figure 2—figure supplement 1A ) . Consistent and significant EEG power modulations after trial onset ( even for nontarget trials ) further confirm that subjects registered trial onsets in the absence of an explicit cue , plausibly using the onset of a fixed order texture sequence as an implicit cue ( Figure 2—figure supplement 1B ) . In alternating nine-minute blocks of trials , we actively biased participants’ perceptual decisions by instructing them either to report as many targets as possible ( ‘Detect as many targets as possible ! ”; liberal condition ) , or to only report high-certainty targets ( "Press only if you are really certain ! "; conservative condition ) . Participants were free to respond at any time during a block whenever they detected a target . A trial was considered a target present response when a button press occurred before the fixed-order sequence ended ( i . e . within 0 . 84 s after onset of the fifth texture containing the ( non ) target , see Figure 2A ) . We provided auditory feedback and applied monetary penalties following missed targets in the liberal condition and following false alarms in the conservative condition ( Figure 2A; see Materials and methods for details ) . The median number of trials for each SDT category across participants was 1206 hits , 65 false alarms , 186 misses and 355 correct rejections in the liberal condition , and 980 hits , 12 false alarms , 419 misses and 492 correct rejections in the conservative condition . Participants reliably adopted the intended decision bias shift across the two conditions , as shown by both the hit rate and the false alarm rate going down in tandem as a consequence of a more conservative bias ( Figure 2B ) . The difference between hit rate and false alarm rate was not significantly modulated by the experimental bias manipulations ( p=0 . 81 , two-sided permutation test , 10 , 000 permutations , see Figure 2B ) . However , target detection performance computed using standard SDT d’ ( perceptual sensitivity , reflecting the distance between the noise and signal distributions in Figure 1A ) ( Green and Swets , 1966 ) was slightly higher during conservative ( liberal: d’=2 . 0 ( s . d . 0 . 90 ) versus conservative: d’=2 . 31 ( s . d . 0 . 82 ) , p=0 . 0002 , see Figure 2C , left bars ) . We quantified decision bias using the standard SDT criterion measure c , in which positive and negative values reflect conservative and liberal biases , respectively ( see the blue and red vertical lines in Figure 1A ) . This uncovered a strong experimentally induced bias shift from the conservative to the liberal condition ( liberal: c = – 0 . 13 ( s . d . 0 . 4 ) , versus conservative: c = 0 . 73 ( s . d . 0 . 36 ) , p=0 . 0001 , see Figure 2C ) , as well as a conservative average bias across the two conditions ( c = 0 . 3 ( s . d . 0 . 31 ) , p=0 . 0013 ) . Because the SDT framework is static over time , we further investigated how bias affected various components of the dynamic decision process by fitting different variants of the drift diffusion model ( DDM ) to the behavioral data ( Figure 1B , C ) ( Ratcliff and McKoon , 2008 ) . The DDM postulates that perceptual decisions are reached by accumulating noisy sensory evidence toward one of two decision boundaries representing the choice alternatives . Crossing one of these boundaries can either trigger an explicit behavioral report to indicate the decision ( for target-present responses in our experiment ) , or remain implicit ( i . e . without active response , for target-absent decisions in our experiment ) . The DDM captures the dynamic decision process by estimating parameters reflecting the rate of evidence accumulation ( drift rate ) , the separation between the boundaries , as well as the time needed for stimulus encoding and response execution ( non-decision time ) ( Ratcliff and McKoon , 2008 ) . The DDM is able to estimate these parameters based on the shape of the RT distributions for actively reported ( target-present ) decisions along with the total number of trials in which no response occurred ( i . e . implicit target-absent decisions ) ( Ratcliff et al . , 2018 ) . We fitted two variants of the DDM to distinguish between two possible mechanisms that can bring about a change in choice bias: one in which the starting point of evidence accumulation moves closer to one of the decision boundaries ( ‘starting point model’ , Figure 1B ) ( Mulder et al . , 2012 ) , and one in which the drift rate itself is biased toward one of the boundaries ( de Gee et al . , 2017 ) ( ‘drift bias model’ , see Figure 1C , referred to as drift criterion by Ratcliff and McKoon ( 2008 ) ) . The drift bias parameter is determined by estimating the contribution of an evidence-independent constant added to the drift ( Figure 2D ) . In the two respective models , we freed either the drift bias parameter ( db , see Figure 2D ) for the two conditions while keeping starting point ( z ) fixed across conditions ( for the drift bias model ) , or vice versa ( for the starting point model ) . Permitting only one parameter at a time to vary freely between conditions allowed us to directly compare the models without having to penalize either model for the number of free parameters . These alternative models make different predictions about the shape of the RT distributions in combination with the response ratios: a shift in starting point results in more target-present choices particularly for short RTs , whereas a shift in drift bias grows over time , resulting in more target-present choices also for longer RTs ( de Gee et al . , 2017; Ratcliff and McKoon , 2008; Urai et al . , 2018 ) . The RT distributions above and below the evidence accumulation graphs in Figure 1B and C illustrate these different effects . In both models , all of the non-bias related parameters ( drift rate v , boundary separation a and non-decision time u + w , see Figure 2D ) were also allowed to vary by condition . We found that the starting point model provided a worse fit to the data than the drift bias model ( starting point model , Bayesian Information Criterion ( BIC ) = 7938; drift bias model , BIC = 7926 , Figure 2E , see Materials and methods for details ) . Specifically , for 15/16 participants , the drift bias model provided a better fit than the starting point model , for 12 of which delta BIC >6 , indicating strong evidence in favor of the drift bias model ( Kass and Raftery , 1995 ) . Despite the lower BIC for the drift bias model , however , we note that to the naked eye both models provide similarly reasonable fits to the single participant RT distributions ( Figure 2—figure supplement 2 ) . Finally , we compared these two models to a model in which both drift bias and starting point were fixed across the conditions , while still allowing the non-bias-related parameters to vary per condition . This model provided the lowest goodness of fit ( delta BIC >6 for both models for all participants ) . Given the superior performance of the drift bias model ( in terms of BIC ) , we further characterized decision making under the bias manipulation using parameter estimates from this model ( see below where we revisit the implausibility of the starting point model when inspecting the lack of pre-stimulus baseline effects in sensory or motor cortex ) . Drift rate , reflecting the participants’ ability to discriminate targets and nontargets , was somewhat higher in the conservative compared to the liberal condition ( liberal: v = 2 . 39 ( s . d . 1 . 07 ) , versus conservative: v = 3 . 06 ( s . d . 1 . 16 ) , p=0 . 0001 , permutation test , Figure 2F , left bars ) . Almost perfect correlations across participants in both conditions between DDM drift rate and SDT d’ provided strong evidence that the drift rate parameter captures perceptual sensitivity ( liberal , r = 0 . 98 , p=1e–10; conservative , r = 0 . 96 , p=5e–9 , see Figure 2—figure supplement 3A ) . Regarding the DDM bias parameters , the condition-fixed starting point parameter in the drift bias model was smaller than half the boundary separation ( i . e . closer to the target-absent boundary ( z = 0 . 24 ( s . d . 0 . 06 ) , p<0 . 0001 , tested against 0 . 5 ) ) , indicating an overall conservative starting point across conditions ( Figure 2—figure supplement 3D ) , in line with the overall positive SDT criterion ( see Figure 2C , right panel ) . Strikingly , however , whereas the drift bias parameter was on average not different from zero in the conservative condition ( db = –0 . 04 ( s . d . 1 . 17 ) , p=0 . 90 ) , drift bias was strongly positive in the liberal condition ( db = 2 . 08 ( s . d . 1 . 0 ) , p=0 . 0001; liberal vs conservative: p=0 . 0005; Figure 2F , right bars ) . The overall conservative starting point combined with a condition-specific neutral drift bias explained the conservative decision bias ( as quantified by SDT criterion ) in the conservative condition ( Figure 2C ) . Likewise , in the liberal condition , the overall conservative starting point combined with a condition-specific positive drift bias ( pushing the drift toward the target-present boundary ) explained the neutral bias observed with SDT criterion ( c around zero for liberal , see Figure 2C ) . Convergent with these modeling results , drift bias was strongly anti-correlated across participants with both SDT criterion ( r = –0 . 89 for both conditions , p=4e–6 ) and average reaction time ( liberal , r = –0 . 57 , p=0 . 02; conservative , r = –0 . 82 , p=1e–4 , see Figure 2—figure supplement 3B C ) . The strong correlations between drift rate and d’ on the one hand , and drift bias and c on the other , provide converging evidence that the SDT and DDM frameworks capture similar underlying mechanisms , while the DDM additionally captures the dynamic nature of perceptual decision making by linking the decision bias manipulation to the evidence accumulation process itself . As a control , we also correlated starting point with criterion , and found that the correlations were somewhat weaker in both conditions ( liberal , r = –0 . 75 . ; conservative , r = –0 . 77 ) , suggesting that the drift bias parameter better captured decision bias as instantiated by SDT . Finally , the bias manipulation also affected two other parameters in the drift bias model that were not directly related to sensory evidence accumulation: boundary separation was slightly but reliably higher during the liberal compared to the conservative condition ( p<0 . 0001 ) , and non-decision time ( comprising time needed for sensory encoding and motor response execution ) was shorter during liberal ( p<0 . 0001 ) ( Figure 2—figure supplement 3D ) . In conclusion , the drift bias variant of the drift diffusion model best explained how participants adjusted to the decision bias manipulations . In the next sections , we used spectral analysis of the concurrent EEG recordings to identify a plausible neural mechanism that reflects biased sensory evidence accumulation . Sensory evidence accumulation in a visual target detection task presumably relies on stimulus-related signals processed in visual cortex . Such stimulus-related signals are typically reflected in cortical population activity exhibiting a rhythmic temporal structure ( Buzsáki and Draguhn , 2004 ) . Specifically , bottom-up processing of visual information has previously been linked to increased high-frequency ( >40 Hz , i . e . gamma ) electrophysiological activity over visual cortex ( Bastos et al . , 2015; Michalareas et al . , 2016; Popov et al . , 2017; van Kerkoerle et al . , 2014 ) . Figure 3 shows significant electrode-by-time-by-frequency clusters of stimulus-locked EEG power , normalized with respect to the condition-specific pre-trial baseline period ( –0 . 4 to 0 s ) . We observed a total of four distinct stimulus-related modulations , which emerged after target onset and waned around the time of response: two in the high-frequency range ( >36 Hz , Figure 3A ( top ) and Figure 3B ) and two in the low-frequency range ( <36 Hz , Figure 3A ( bottom ) and Figure 3C ) . First , we found a spatially focal modulation in a narrow frequency range around 25 Hz reflecting the steady state visual evoked potential ( SSVEP ) arising from entrainment by the visual stimulation frequency of our experimental paradigm ( Figure 3A , bottom panel ) , as well as a second modulation from 42 to 58 Hz comprising the SSVEP’s harmonic ( Figure 3A , top panel ) . Both SSVEP frequency modulations have a similar topographic distribution ( see left panels of Figure 3A ) . Third , we observed a 59—100 Hz ( gamma ) power modulation ( Figure 3B ) , after carefully controlling for high-frequency EEG artifacts due to small fixational eye movements ( microsaccades ) by removing microsaccade-related activity from the data ( Hassler et al . , 2011; Hipp and Siegel , 2013; Yuval-Greenberg et al . , 2008 ) , and by suppressing non-neural EEG activity through scalp current density ( SCD ) transformation ( Melloni et al . , 2009; Perrin et al . , 1989 ) ( see Materials and methods for details ) . Importantly , the topography of the observed gamma modulation was confined to posterior electrodes , in line with a role of gamma in bottom-up processing in visual cortex ( Ni et al . , 2016 ) . Finally , we observed suppression of low-frequency beta ( 11—22 Hz ) activity in posterior cortex , which typically occurs in parallel with enhanced stimulus-induced gamma activity ( Donner and Siegel , 2011; Kloosterman et al . , 2015a; Meindertsma et al . , 2017; Werkle-Bergner et al . , 2014 ) ( Figure 3C ) . Response-locked , this cluster was most pronounced over left motor cortex ( electrode C4 ) , plausibly due to the right-hand button press that participants used to indicate target detection ( Donner et al . , 2009 ) . In the next sections , we characterize these signals separately for the two conditions , investigating stimulus-related signals within a pooling of 11 occipito-parietal electrodes based on the gamma enhancement in Figure 3B ( Oz , POz , Pz , PO3 , PO4 , and P1 to P6 ) , and motor-related signals in left-hemispheric beta ( LHB ) suppression in electrode C4 ( Figure 3C ) ( O'Connell et al . , 2012 ) . Our behavioral results suggest that participants biased sensory evidence accumulation in the liberal condition , rather than changing their starting point . We next sought to provide converging evidence for this conclusion by examining pre-stimulus activity , post-stimulus activity , and motor-related EEG activity . Following previous studies , we hypothesized that a starting point bias would be reflected in a difference in pre-motor baseline activity between conditions before onset of the decision process ( Afacan-Seref et al . , 2018; de Lange et al . , 2013 ) , and/or in a difference in pre-stimulus activity such as seen in bottom up stimulus-related SSVEP and gamma power signals ( Figure 4A shows the relevant clusters as derived from Figure 3 ) . Thus , we first investigated the timeline of raw power in the SSVEP , gamma and LHB range between conditions ( see Figure 4B ) . None of these markers showed a meaningful difference in pre-stimulus baseline activity . Statistically comparing the raw pre-stimulus activity between liberal and conservative in a baseline interval between –0 . 4 and 0 s prior to trial onset yielded p=0 . 52 , p=0 . 51 and p=0 . 91 , permutation tests , for the respective signals . This confirms a highly similar starting point of evidence accumulation in all these signals . Next , we predicted that a shift in drift bias would be reflected in a steeper slope of post-stimulus ramping activity ( leading up to the decision ) . We reasoned that the best way of ascertaining such an effect would be to baseline the activity to the interval prior to stimulus onset ( using the interval between –0 . 4 to 0 s ) , such that any post-stimulus effect we find cannot be explained by pre-stimulus differences ( if any ) . The time course of post-stimulus and response-locked activity after baselining can be found in Figure 4C . All three signals showed diverging signals between the liberal and conservative condition after trial onset , consistent with adjustments in the process of evidence accumulation . Specifically , we observed higher peak modulation levels for the liberal condition in all three stimulus-locked signals ( p=0 . 08 , p=0 . 002 and p=0 . 023 , permutation tests for SSVEP , gamma and LHB , respectively ) , and found a steeper slope toward the button press for LHB ( p=0 . 04 ) . Finally , the event related potential in motor cortex also showed a steeper slope toward report for liberal ( p=0 . 07 , Figure 4 , bottom row , baseline plot is not meaningful for time-domain signals due to mean removal during preprocessing ) . Taken together , these findings provide converging evidence that participants implemented a liberal decision bias by adjusting the rate of evidence accumulation toward the target-present choice boundary , but not its starting point . In the next sections , we sought to identify a neural mechanism that could underlie these biases in the rate of evidence accumulation . Given a lack of pre-stimulus ( starting-point ) differences in specific frequency ranges involved in stimulus processing or motor responses ( Figure 4B ) , we next focused on other pre-stimulus differences that might be the root cause of the post-stimulus differences we observed in Figure 4C . To identify such signals at high frequency resolution , we computed spectral power in a wide time window from –1 s until trial start . We then ran a cluster-based permutation test across all electrodes and frequencies in the low-frequency domain ( 1–35 Hz ) , looking for power modulations due to our experimental manipulations . Pre-stimulus spectral power indeed uncovered two distinct modulations in the liberal compared to the conservative condition: ( 1 ) theta modulation in midfrontal electrodes and ( 2 ) alpha modulation in posterior electrodes . Figure 5A depicts the difference between the liberal and conservative condition , confirming significant clusters ( p<0 . 05 , cluster-corrected for multiple comparisons ) of enhanced theta ( 2–6 Hz ) in frontal electrodes ( Fz , Cz , FC1 , and FC2 ) , as well as suppressed alpha ( 8—12 Hz ) in a group of posterior electrodes , including all 11 electrodes selected previously based on post-stimulus gamma modulation ( Figure 3 ) . The two modulations were uncorrelated across participants ( r = 0 . 06 , p=0 . 82 ) , suggesting they reflect different neural processes related to our experimental task manipulations . These findings are consistent with literature pointing to a role of midfrontal theta as a source of cognitive control signals originating from pre-frontal cortex ( Cohen and Frank , 2009; van Driel et al . , 2012 ) and alpha in posterior cortex reflecting spontaneous trial-to-trial fluctuations in decision bias ( Iemi et al . , 2017 ) . The fact that these pre-stimulus effects occur as a function of our experimental manipulation suggests that they are a hallmark of strategic bias adjustment , rather than a mere correlate of spontaneous shifts in decision bias . Importantly , this finding implies that humans are able to actively control pre-stimulus alpha power in visual cortex ( possibly through top-down signals from frontal cortex ) , plausibly acting to bias sensory evidence accumulation toward the response alternative that maximizes reward . Next , we asked how suppression of pre-stimulus alpha activity might bias the process of sensory evidence accumulation . One possibility is that alpha suppression influences evidence accumulation by modulating the susceptibility of visual cortex to sensory stimulation , a phenomenon termed ‘neural excitability’ ( Iemi et al . , 2017; Jensen and Mazaheri , 2010 ) . We explored this possibility using a theoretical response gain model formulated by Rajagovindan and Ding ( 2011 ) . This model postulates that the relationship between the total synaptic input that a neuronal ensemble receives and the total output activity it produces is characterized by a sigmoidal function ( Figure 6A ) – a notion that is biologically plausible ( Destexhe et al . , 2001; Freeman , 1979 ) . In this model , the total synaptic input into visual cortex consists of two components: ( 1 ) sensory input ( i . e . due to sensory stimulation ) and ( 2 ) ongoing fluctuations in endogenously generated ( i . e . not sensory-related ) neural activity . In our experiment , the sensory input into visual cortex can be assumed to be identical across trials , because the same sensory stimulus was presented in each trial ( see Figure 2A ) . The endogenous input , in contrast , is thought to vary from trial to trial reflecting fluctuations in top-down cognitive processes such as attention . These fluctuations are assumed to be reflected in the strength of alpha power suppression , such that weaker alpha is associated with increased attention ( Figure 6B ) . Given the combined constant sensory and variable endogenous input in each trial ( see horizontal axis in Figure 6A ) , the strength of the output responses of visual cortex are largely determined by the trial-to-trial variations in alpha power ( see vertical axis in Figure 6A ) . Furthermore , the sigmoidal shape of the input-output function results in an effective range in the center of the function’s input side which yields the strongest stimulus-induced output responses since the sigmoid curve there is steepest . Mathematically , the effect of endogenous input on stimulus-induced output responses ( see marked interval in Figure 6A ) can be expressed as the first order derivative or slope of the sigmoid in Figure 6A , which is referred to as the response gain by Rajagovindan and Ding ( 2011 ) . This derivative is plotted in Figure 6B ( blue and red solid lines ) across levels of pre-stimulus alpha power , predicting an inverted-U shaped relationship between alpha and response gain in visual cortex . Regarding our experimental conditions , the model not only predicts that the suppression of pre-stimulus alpha observed in the liberal condition reflects a shift in the operational range of alpha ( see Figure 5B ) , but also that it increases the maximum output of visual cortex ( a shift from the red to the blue line in Figure 6A ) . Therefore , the difference between stimulus conditions is not modeled using a single input-output function , but necessitates an additional mechanism that changes the input-output relationship itself . The exact nature of this mechanism is not known ( also see Discussion ) . Rajagovindan and Ding suggest that top-down mechanisms modulate ongoing prestimulus neural activity to increase the slope of the sigmoidal function , but despite the midfrontal theta activity we observed , evidence for this hypothesis is somewhat elusive . We have no means to establish directly whether this relationship exists , and can merely reflect on the fact that this change in the input-output function is necessary to capture condition-specific effects of the input-output relationship , both in the data of Rajagovindan and Ding ( 2011 ) and in our own data . Thus , as the operational range of alpha shifts leftwards from conservative to liberal , the upper asymptote in Figure 6A moves upwards such that the total maximum output activity increases . This in turn affects the inverted-U-shaped relationship between alpha and gain in visual cortex ( blue line in Figure 6B ) , leading to a steeper response curve in the liberal condition resembling a Gaussian ( bell-shaped ) function . To investigate sensory response gain across different alpha levels in our data , we used the post-stimulus gamma activity ( see Figure 3B ) as a proxy for alpha-linked output gain in visual cortex ( Bastos et al . , 2015; Michalareas et al . , 2016; Ni et al . , 2016; Popov et al . , 2017; van Kerkoerle et al . , 2014 ) . We exploited the large number of trials per participant per condition ( range 543 to 1391 trials ) by sorting each participant’s trials into ten equal-sized bins ranging from weak to strong alpha , separately for the two conditions . We then calculated the average gamma power modulation within each alpha bin and finally plotted the participant-averaged gamma across alpha bins for each condition in Figure 6C ( see Materials and methods for details ) . This indeed revealed an inverted-U shaped relationship between alpha and gamma in both conditions , with a steeper curve for the liberal condition . To assess the model’s ability to explain the data , we statistically tested three predictions derived from the model . First , the model predicts overall lower average pre-stimulus alpha power for liberal than for conservative due to the shift in the operational range of alpha . This was confirmed in Figure 6D ( p=0 . 01 , permutation test , see also Figure 5 ) . Second , the model predicts a stronger gamma response for liberal than for conservative around the peak of the gain curve ( the center of the effective alpha range , see Figure 6B ) , which we indeed observed ( p=0 . 024 , permutation test on the average of the middle two alpha bins ) ( Figure 6E ) . Finally , the model predicts that the difference between the gain curves ( when they are aligned over their effective ranges on the x-axis using alpha bin number , as shown in Figure 6—figure supplement 1A ) also resembles a Gaussian curve ( Figure 6—figure supplement 1B ) . Consistent with this prediction , we observed an interaction effect between condition ( liberal , conservative ) and bin number ( 1-10 ) using a standard Gaussian contrast in a two-way repeated measures ANOVA ( F ( 1 , 13 ) = 4 . 6 , p=0 . 051 , partial η2 = 0 . 26 ) . Figure 6F illustrates this finding by showing the difference between the two curves in Figure 6C as a function of alpha bin number ( see Figure 6—figure supplement 1C for the curves of both conditions as a function of alpha bin number ) . Subsequent separate tests for each condition indeed confirmed a significant U-shaped relationship between alpha and gamma in the liberal condition with a large effect size ( F ( 1 , 13 ) = 7 . 7 , p=0 . 016 , partial η2 = 0 . 37 ) , but no significant effect in the conservative condition with only a small effect size ( F ( 1 , 13 ) = 1 . 7 , p=0 . 22 , partial η2 = 0 . 12 ) , using one-way repeated measures ANOVA’s with alpha bin ( Gaussian contrast ) as the factor of interest . Taken together , these findings suggest that the alpha suppression observed in the liberal compared to the conservative condition boosted stimulus-induced activity , which in turn might have indiscriminately biased sensory evidence accumulation toward the target-present decision boundary . In the next section , we investigate a direct link between drift bias and stimulus-induced activity as measured through gamma . The findings presented so far suggest that behaviorally , a liberal decision bias shifts evidence accumulation toward target-present responses ( drift bias in the DDM ) , while neurally it suppresses pre-stimulus alpha and enhances poststimulus gamma responses . In a final analysis , we asked whether alpha-binned gamma modulation is directly related to a stronger drift bias . To this end , we again applied the drift bias DDM to the behavioral data of each participant , while freeing the drift bias parameter not only for the two conditions , but also for the 10 alpha bins for which we calculated gamma modulation ( see Figure 6C ) . We directly tested the correspondence between DDM drift bias and gamma modulation using repeated measures correlation ( Bakdash and Marusich , 2017 ) , which takes all repeated observations across participants into account while controlling for non-independence of observations collected within each participant ( see Materials and methods for details ) . Gamma modulation was indeed correlated with drift bias in both conditions ( liberal , r ( 125 ) = 0 . 49 , p=5e-09; conservative , r ( 125 ) = 0 . 38 , p=9e-06 ) ( Figure 7 ) . We tested the robustness of these correlations by excluding the data points that contributed most to the correlations ( as determined with Cook’s distance ) and obtained qualitatively similar results , indicating these correlations were not driven by outliers ( Figure 7 , see Materials and methods for details ) . To rule out that starting point could explain this correlation , we repeated this analysis while controlling for the starting point of evidence accumulation estimated per alpha bin within the starting point model . To this end , we regressed both bias parameters on gamma . Crucially , we found that in both conditions starting point bias did not uniquely predict gamma when controlling for drift bias ( liberal: F ( 1 , 124 ) = 5 . 8 , p=0 . 017 for drift bias , F ( 1 , 124 ) = 0 . 3 , p=0 . 61 for starting point; conservative: F ( 1 , 124 ) = 8 . 7 , p=0 . 004 for drift bias , F ( 1 , 124 ) = 0 . 4 , p=0 . 53 for starting point . This finding suggests that the drift bias model outperforms the starting point model when correlated to gamma power . As a final control , we also performed this analysis for the SSVEP ( 23–27 Hz ) power modulation ( see Figure 3B , bottom ) and found a similar inverted-U shaped relationship between alpha and the SSVEP for both conditions ( Figure 7—figure supplement 1A ) , but no correlation with drift bias ( liberal , r ( 125 ) = 0 . 11 , p=0 . 72 , conservative , r ( 125 ) = 0 . 22 , p=0 . 47 ) ( Figure 7—figure supplement 1B ) or with starting point ( liberal , r ( 125 ) = 0 . 08 , p=0 . 02 , conservative , r ( 125 ) = 0 . 22 , p=0 . 95 ) . This suggests that the SSVEP is similarly coupled to alpha as the stimulus-induced gamma , but is less affected by the experimental conditions and not predictive of decision bias shifts . Taken together , these results suggest that alpha-binned gamma modulation underlies biased sensory evidence accumulation . Finally , we asked to what extent the enhanced tonic midfrontal theta may have mediated the relationship between alpha-binned gamma and drift bias . To answer this question , we entered drift bias in a two-way repeated measures ANOVA with factors theta and gamma power ( all variables alpha-binned ) , but found no evidence for mediation of the gamma-drift bias relationship by midfrontal theta ( liberal , F ( 1 , 13 ) = 1 . 3 , p=0 . 25; conservative , F ( 1 , 13 ) = 0 . 003 , p=0 . 95 ) . At the same time , the gamma-drift bias relationship was qualitatively unchanged when controlling for theta ( liberal , F ( 1 , 13 ) = 48 . 4 , p<0 . 001; conservative , F ( 1 , 13 ) = 19 . 3 , p<0 . 001 ) . Thus , the enhanced midfrontal theta in the liberal condition plausibly reflects a top-down , attention-related signal indicating the need for cognitive control to avoid missing targets , but its amplitude seemed not directly linked to enhanced sensory evidence accumulation , as found for gamma . This latter finding suggests that the enhanced theta in the liberal condition served as an alarm signal indicating the need for a shift in response strategy , without specifying exactly how this shift was to be implemented ( Cavanagh and Frank , 2014 ) . The data analyzed in this study are publicly available on Figshare ( Kloosterman et al . , 2018 ) . Analysis scripts are publicly available on Github ( Kloosterman , 2018; copy archived at https://github . com/elifesciences-publications/critEEG ) . Sixteen participants ( eight females , mean age 24 . 1 years , ±1 . 64 ) took part in the experiment , either for financial compensation ( EUR 10 , - per hour ) or in partial fulfillment of first year psychology course requirements . Each participant completed three experimental sessions on different days , each session lasting ca . 2 hr , including preparation and breaks . One participant completed only two sessions , yielding a total number of sessions across subjects of 47 . Due to technical issues , for one session only data for the liberal condition was available . One participant was an author . All participants had normal or corrected-to-normal vision and were right handed . Participants provided written informed consent before the start of the experiment . All procedures were approved by the ethics committee of the University of Amsterdam . Regarding sample size , our experiment consisted of 16 biological replications ( participants ) and either three ( fifteen participants ) or two ( one participant ) technical replications ( i . e . experimental sessions ) . The sample size was determined based on two criteria: 1 ) obtaining large amounts of data per participant ( thousands of trials ) , which is necessary to perform robust drift diffusion modelling of choice behavior and obtain reliable EEG spectral power estimates for each of the ten bins of trials that were created within participants , and 2 ) obtaining data from a sufficient number of participants to leverage across-subject variability in correlational analyses . Thus , we emphasized obtaining many data points per participant relative to obtaining many participants , while still preserving the ability to perform correlations across participants . All participants were included in the signal-detection-theoretical and drift diffusion modeling analyses . One participant was excluded from the EEG analysis due to excessive noise ( EEG power spectrum opposite of 1/frequency ) . One further participant was excluded from the analyses that included condition-specific gamma because the liberal–conservative difference in gamma in this participant was >3 standard deviations away from the other participants . Stimuli consisted of a continuous semi-random rapid serial visual presentation ( rsvp ) of full screen texture patterns . The texture patterns consisted of line elements approx . 0 . 07° thick and 0 . 4° long in visual angle . Each texture in the rsvp was presented for 40 ms ( i . e . stimulation frequency 25 Hz ) , and was oriented in one of four possible directions: 0° , 45° , 90° or 135° . Participants were instructed to fixate on a red dot in the center of the screen . At random inter trial intervals ( ITI’s ) sampled from a uniform distribution ( ITI range 0 . 3–2 . 2 s ) , the rsvp contained a fixed sequence of 25 texture patterns , which in total lasted one second . This fixed sequence consisted of four stimuli preceding a ( non- ) target stimulus ( orientations of 45° , 90° , 0° , 90° respectively ) and twenty stimuli following the ( non ) -target ( orientations of 0° , 90° , 0° , 90° , 0° , 45° , 0° , 135° , 90° , 45° , 0° , 135° , 0° , 45° , 90° , 45° , 90° , 135° , 0° , 135° respectively ) ( see Figure 2A ) . The fifth texture pattern within the sequence ( occurring from 0 . 16 s after sequence onset ) was either a target or a nontarget stimulus . Nontargets consisted of either a 45° or a 135° homogenous texture , whereas targets contained a central orientation-defined square of 2 . 42° visual angle , thereby consisting of both a 45° and a 135° texture . 50% of all targets consisted of a 45° square and 50% of a 135° square . Of all trials , 75% contained a target and 25% a nontarget . Target and nontarget trials were presented in random order . To avoid specific influences on target stimulus visibility due to presentation of similarly or orthogonally oriented texture patterns temporally close in the cascade , no 45° and 135° oriented stimuli were presented directly before or after presentation of the target stimulus . All stimuli had an isoluminance of 72 . 2 cd/m2 . Stimuli were created using MATLAB ( The Mathworks , Inc , Natick , MA , USA; RRID:SCR_001622 ) and presented using Presentation version 9 . 9 ( Neurobehavioral systems , Inc , Albany , CA , USA; RRID:SCR_002521 ) . The participants’ task was to detect and actively report targets by pressing a button using their right hand . Targets occasionally went unreported , presumably due to constant forward and backward masking by the continuous cascade of stimuli and unpredictability of target timing ( Fahrenfort et al . , 2007 ) . The onset of the fixed order of texture patterns preceding and following ( non- ) target stimuli was neither signaled nor apparent . At the beginning of the experiment , participants were informed they could earn a total bonus of EUR 30 , - , on top of their regular pay of EUR 10 , - per hour or course credit . In two separate conditions within each session of testing , we encouraged participants to use either a conservative or a liberal bias for reporting targets using both aversive sounds as well as reducing their bonus after errors . In the conservative condition , participants were instructed to only press the button when they were relatively sure they had seen the target . The instruction on screen before block onset read as follows: ‘Try to detect as many targets as possible . Only press when you are relatively sure you just saw a target . ’ To maximize effectiveness of this instruction , participants were told the bonus would be diminished by 10 cents after a false alarm . During the experiment , a loud aversive sound was played after a false alarm to inform the participant about an error . During the liberal condition , participants were instructed to miss as few targets as possible . The instruction on screen before block onset read as follows: ‘Try to detect as many targets as possible . If you sometimes press when there was nothing this is not so bad’ . In this condition , the loud aversive sound was played twice in close succession whenever they failed to report a target , and three cents were subsequently deducted from their bonus . The difference in auditory feedback between both conditions was included to inform the participant about the type of error ( miss or false alarm ) , in order to facilitate the desired bias in both conditions . After every block , the participant’s score ( number of missed targets in the liberal condition and number of false alarms in the conservative condition ) was displayed on the screen , as well as the remainder of the bonus . After completing the last session of the experiment , every participant was paid the full bonus as required by the ethical committee . Participants performed six blocks per session lasting ca . nine minutes each . During a block , participants continuously monitored the screen and were free to respond by button press whenever they thought they saw a target . Each block contained 240 trials , of which 180 target and 60 nontarget trials . The task instruction was presented on the screen before the block started . The condition of the first block of a session was counterbalanced across participants . Prior to EEG recording in the first session , participants performed a 10-min practice run of both conditions , in which visual feedback directly after a miss ( liberal condition ) or false alarm ( conservative ) informed participants about their mistake , allowing them to adjust their decision bias accordingly . There were short breaks between blocks , in which participants indicated when they were ready to begin the next block . We calculated each participant’s criterion c ( Green and Swets , 1966 ) across the trials in each condition as follows:c =-12 [Z ( Hit‐rate ) + Z ( FA‐rate ) ]where hit-rate is the proportion target-present responses of all target-present trials , false alarm ( FA ) -rate is the proportion target-present responses of all target-absent trials , and Z ( . . . ) is the inverse standard normal distribution . Furthermore , we calculated objective sensitivity measure d’ using:d' =ZHit‐rate- Z ( FA‐rate ) as well as by subtracting hit and false alarm rates . Reaction times ( RTs ) were measured as the duration between target onset and button press . In order to be detected , the 40 ms-duration figure-ground targets used in our study undergo a process in visual cortex called figure-ground segregation . This process has been well characterized in man and monkey ( Fahrenfort et al . , 2008; Lamme , 1995; Lamme et al . , 2002; Supèr et al . , 2003 ) , and results from recurrent processing to extract the surface region in visual cortex . Figure-ground segregation is known to extend far beyond the mere presentation time of the stimulus , thus providing a plausible neural basis for the evidence accumulation process . Further , a central assumption of the drift diffusion model is that the process of evidence accumulation is gradual , independent of whether sensory input is momentary . Indeed , the DDM was initially developed to explain reaction time distributions during memory retrieval , in which evidence accumulation must occur through retrieval of a memory trace within the brain , in the complete absence of external stimulus at the time of the decision ( Ratcliff , 1978 ) . Our observed RT distributions show the typical features that occur across many different types of decision and memory tasks , which the DDM is well able to capture , including a sharp leading edge and a long tail of the distributions ( see Figure 2—figure supplement 2 ) . The success of the DDM in fitting these data is consistent with previous work ( e . g . Ratcliff , 2006 ) and might reflect the fact that observers modulate the underlying components of the decision process also when they do not control the stimulus duration ( Kiani et al . , 2008 ) . We fitted the drift diffusion model to our behavioral data for each subject individually , and separately for the liberal and conservative conditions . We fitted the model using a G square method based on quantile RT’s ( RT cutoff , 200 ms , for details , see Ratcliff et al . , 2018 ) , using custom code ( de Gee et al . , 2018 ) that was contributed to the HDDM 0 . 6 . 1 package ( Wiecki et al . , 2013 ) . The RT distributions for target-present responses were represented by the 0 . 1 , 0 . 3 , 0 . 5 , 0 . 7 and 0 . 9 quantiles , and , along with the associated response proportions , contributed to G square . In addition , a single bin containing the number of target-absent responses contributed to G square . Each model fit was run six times , after which the best fitting run was kept . Fitting the model to RT distributions for target-present and target-absent choices ( termed ‘stimulus coding’ in Wiecki et al . , 2013 ) , as opposed to the more common fits of correct and incorrect choice RT’s ( termed ‘accuracy coding’ in Wiecki et al . , 2013 ) , allowed us to estimate parameters that could have induced biases in subjects’ behavior . Parameter recovery simulations showed that letting both the starting point of the accumulation process and drift bias ( an evidence-independent constant added to the drift toward one or the other bound ) free to vary with experimental condition is problematic for data with no explicit target-absent responses ( data not shown ) . Thus , to test whether shifts in drift bias or starting point underlie bias we fitted three separate models . In the first model ( ‘fixed model’ ) , we allowed only the following parameters to vary between the liberal and conservative condition: ( i ) the mean drift rate across trials; ( ii ) the separation between both decision bounds ( i . e . , response caution ) ; and ( iii ) the non-decision time ( sum of the latencies for sensory encoding and motor execution of the choice ) . Additionally , the bias parameters starting point and drift bias were fixed for the experimental conditions . The second model ( ‘starting point model’ ) was the same as the fixed model , except that we let the starting point of the accumulation process vary with experimental condition , whereas the drift bias was kept fixed for both conditions . The third model ( ‘drift bias model’ ) was the same as the fixed model , except that we let the drift bias vary with experimental condition , while the starting point was kept fixed for both conditions . We used Bayesian Information Criterion ( BIC ) to select the model which provided the best fit to the data ( Neath and Cavanaugh , 2012 ) . The BIC compares models based on their maximized log-likelihood value , while penalizing for the number of parameters . In our task , only target-present responses were coupled to a behavioral response ( button-press ) , so we could measure reaction times only for these responses , whereas reaction times for target-absent responses remained implicit . Thus , in our fitting procedure , the RT distributions for target-present responses were represented by the 0 . 1 , 0 . 3 , 0 . 5 , 0 . 7 and 0 . 9 quantiles , and , along with the associated response proportions , contributed to G square . In addition , a single bin containing the number of target-absent responses contributed to G square . It has been shown that such a diffusion model with an implicit ( no response ) boundary can be fit to data with almost the same accuracy as fitting the two-choice model to two-choice data ( Ratcliff et al . , 2018 ) . In a diffusion model with an implicit ( no response ) boundary , both an increase in drift rate and drift criterion would predict faster target-present responses . However , the key distinction is that an increase in drift additionally predicts more correct responses ( for both target-present and target-absent responses ) , and an increase in drift criterion shifts the relative fraction of target-present and target-absent responses ( decision bias ) . Because a single bin containing the number of target-absent responses contributed to G square , our fitting procedure can distinguish between decision bias versus drift rate . Continuous EEG data were recorded at 256 Hz using a 48-channel BioSemi Active-Two system ( BioSemi , Amsterdam , the Netherlands ) , connected to a standard EEG cap according to the international 10–20 system . Electrooculography ( EOG ) was recorded using two electrodes at the outer canthi of the left and right eyes and two electrodes placed above and below the right eye . Horizontal and vertical EOG electrodes were referenced against each other , two for horizontal and two for vertical eye movements ( blinks ) . We used the Fieldtrip toolbox ( Oostenveld et al . , 2011 ) and custom software ( Kloosterman et al . , 2018 ) in MATLAB R2016b ( The Mathworks Inc , Natick , MA , USA; RRID:SCR_001622 ) to process the data ( see below ) . Data were re-referenced to the average voltage of two electrodes attached to the earlobes . We extracted trials of variable duration from 1 s before target sequence onset until 1 . 25 after button press for trials that included a button press ( hits and false alarms ) , and until 1 . 25 s after stimulus onset for trials without a button press ( misses and correct rejects ) . The following constraints were used to classify ( non- ) targets as detected ( hits and false alarms ) , while avoiding the occurrence of button presses in close succession to target reports and button presses occurring outside of trials: 1 ) A trial was marked as detected if a response occurred within 0 . 84 s after target onset; 2 ) when the onset of the next target stimulus sequence started before trial end , the trial was terminated at the next trial’s onset; 3 ) when a button press occurred in the 1 . 5 s before trial onset , the trial was extracted from 1 . 5 s after this button press; 4 ) when a button press occurred between 0 . 5 s before until 0 . 2 s after sequence onset , the trial was discarded . See Kloosterman et al . , 2015a and Meindertsma et al . ( 2017 ) for similar trial extraction procedures . After trial extraction , channel time courses were linearly detrended and the mean of every channel was removed per trial . Trials containing muscle artifacts were rejected from further analysis using a standard semi-automatic preprocessing method in Fieldtrip . This procedure consists of bandpass-filtering the trials of a condition block in the 110–125 Hz frequency range , which typically contains most of the muscle artifact activity , followed by a Z-transformation . Trials exceeding a threshold Z-score were removed completely from analysis . We used as the threshold the absolute value of the minimum Z-score within the block , +1 . To remove eye blink artifacts from the time courses , the EEG data from a complete session were transformed using independent component analysis ( ICA ) , and components due to blinks ( typically one or two ) were removed from the data . In addition , to remove microsaccade-related artifacts we included two virtual channels in the ICA based on channels Fp1 and Fp2 , which included transient spike potentials as identified using the saccadic artefact detection algorithm from Hassler et al . ( 2011 ) . This yielded a total number of channels submitted to ICA of 48 + 2 = 50 . The two components loading high on these virtual electrodes ( typically with a frontal topography ) were also removed . Blinks and eye movements were then semi-automatically detected from the horizontal and vertical EOG ( frequency range 1–15 Hz; z-value cut-off four for vertical; six for horizontal ) and trials containing eye artefacts within 0 . 1 s around target onset were discarded . This step was done to remove trials in which the target was not seen because the eyes were closed . Finally , trials exceeding a threshold voltage range of 200 μV were discarded . To attenuate volume conduction effects and suppress any remaining microsaccade-related activity , the scalp current density ( SCD ) was computed using the second-order derivative ( the surface Laplacian ) of the EEG potential distribution ( Perrin et al . , 1989 ) . We computed event-related potentials in electrode C4 by low-pass filtering the time-domain data up to 8 Hz followed by averaging all trials within participant per condition . We used a sliding window Fourier transform ( Mitra and Pesaran , 1999 ) ; step size , 50 ms; window size , 400 ms; frequency resolution , 2 . 5 Hz ) to calculate time-frequency representations ( spectrograms ) of the EEG power for each electrode and each trial . We used a single Hann taper for the frequency range of 3–35 Hz ( spectral smoothing , 4 . 5 Hz , bin size , 1 Hz ) and the multitaper technique for the 36–100 Hz frequency range ( spectral smoothing , 8 Hz; bin size , 2 Hz; five tapers ) . See Kloosterman et al . , 2015a and Meindertsma et al . ( 2017 ) for similar settings . Finally , to investigate spectral power also <3 Hz , we ran an additional time-frequency analysis with a window size of 1 s ( i . e . frequency resolution 1 Hz ) centered on the time point 0 . 5 s before trial onset ( frequency range 1–35 Hz , no spectral smoothing , bin size 0 . 5 Hz ) . Spectrograms were aligned to the onset of the stimulus sequence containing the ( non ) target , and ( in a separate analysis ) to the button press . Power modulations during the trials were quantified as the percentage of power change at a given time point and frequency bin , relative to a baseline power value for each frequency bin ( Figure 3 ) . We used as a baseline the mean EEG power in the interval 0 . 4 to 0 s before trial onset , computed separately for each condition . If this interval was not completely present in the trial due to preceding events ( see Trial extraction ) , this period was shortened accordingly . We normalized the data by subtracting the baseline from each time-frequency bin and dividing this difference by the baseline ( x 100% ) . For the analysis of raw pre-stimulus power modulations , no baseline correction was applied on the raw scalp current density values . We focused our analysis of EEG power modulations around target onsets on those electrodes that processed the visual stimulus . To this end , we averaged the power modulations or raw power across eleven occipito-parietal electrodes that showed stimulus-induced responses in the gamma-band range ( 59–100 Hz ) . See Kloosterman et al . , 2015a and Meindertsma et al . ( 2017 ) for a similar procedure . To determine clusters of significant modulation with respect to the pre-stimulus baseline without any a priori selection , we ran statistics across space-time-frequency bins using paired t-tests across subjects performed at each bin . Single bins were subsequently thresholded at p<0 . 05 and clusters of contiguous time-space-frequency bins were determined . Cluster significance was assessed using a cluster-based permutation procedure ( 1000 permutations ) . For visualization purposes , we integrated ( using the matlab trapz function ) power modulation in the time-frequency representations ( TFR’s , Figure 3 , left panels ) across the highlighted electrodes in the topographies ( Figure 3 , right panels ) . For the topographical scalp maps , modulation was integrated across the saturated time-frequency bins in the TFRs . To test at which frequencies raw prestimulus EEG power differed between the liberal and conservative conditions , we performed this analysis across electrodes and frequencies after taking the liberal – conservative difference at each frequency bin ( Figure 5A ) ( see Statistical comparisons ) . To test the predictions of the gain model , we first averaged activity in the 8–12 Hz range from 0 . 8 to 0 . 2 s before trial onset ( staying half our window size from trial onset , to avoid mixing pre- and poststimulus activity , also see Iemi et al . , 2017 ) , yielding a single scalar alpha power value per trial . If this interval was not completely present in the trial due to preceding events ( see Trial extraction ) , this period was shortened accordingly . Trials in which the scalar was >3 standard deviations away from the participant’s mean were excluded . We then sorted all single-trial alpha values for each participant and condition in ascending order and assigned them to ten bins of equal size , ranging from weakest to strongest alpha . Adjacent bin ranges overlapped for 50% to stabilize estimates . Then we averaged the corresponding gamma modulation of the trials belonging to each bin ( consisting of the average power modulation within 59–100 Hz 0 . 2 to 0 . 6 s after trial onset , see Figure 3 ) . Finally , we averaged across participants and plotted the median alpha value per bin averaged across participants against the mean gamma modulation . See Rajagovindan and Ding ( 2011 ) for a similar procedure . To statistically test for the existence of inverted U-shaped relationships between alpha and gamma , we performed a one-way repeated measures ANOVA on gamma modulation with factor alpha bin ( 10 bins ) to each condition separately and a two-way repeated measures ANOVA with factors bin and condition for testing the liberal–conservative difference ( Figure 6F ) . Given the model prediction of a Gaussian-shaped relationship between alpha and gamma , we constructed a Gaussian contrast using a Gaussian shape with unit standard deviation ( contrast values: −1000 , –991 , −825 , 295 , 2521 , 2521 , 295 , –825 , −991 , –1000 , values were chosen to sum to zero ) . For plotting purposes ( Figure 6C-F ) , we computed within-subject error bars by removing within each participant the mean across conditions from the estimates . To link DDM drift bias and gamma power modulation , we re-fitted the DDM drift bias model while freeing the drift bias parameter both for each condition as well as for the ten alpha bins , while freeing the other parameters ( drift rate , boundary separation , non-decision time ) for each condition and fixing starting point across conditions . We then used repeated measures correlation to test whether stronger gamma was associated with stronger drift bias . Repeated measures correlation determines the common within-individual association for paired measures assessed on two or more occasions for multiple individuals by controlling for the specific range in which individuals’ measurements operate , and correcting the correlation degrees of freedom for non-independence of repeated measurements obtained from each individual . Specifically , the correlation degrees of freedom were 14 participants × 10 observations – Number of participants – 1 = 140 – 14 – 1 = 125 . Repeated measures correlation tends to have greater statistical power than conventional correlation across individuals because neither averaging nor aggregation is necessary for an intra-individual research question . Please see Bakdash and Marusich ( 2017 ) for more information . We assessed the impact of single observations on the correlations by excluding observations exceeding five times the average Cook’s distance of all values within each condition ( five observations for liberal and four for conservative ) and recomputing the correlations . We used two-sided permutation tests ( 10 , 000 permutations ) ( Efron and Tibshirani , 1998 ) to test the significance of behavioral effects and the model fits . Permutation tests yield p=0 if the observed value falls outside the range of the null distribution . In these cases , p<0 . 0001 is reported in the manuscript . The standard deviation ( s . d . ) is reported as a measure of spread along with all participant-averaged results reported in the text . To quantify power modulations after ( non- ) target onset , we tested the overall power modulation for significant deviations from zero . For these tests , we used a cluster-based permutation procedure to correct for multiple comparisons ( Maris and Oostenveld , 2007 ) . For time-frequency representations along with spatial topographies of power modulation , this procedure was performed across all time-frequency bins and electrodes; for frequency spectra across all electrodes and frequencies; for power and ERP time courses , across all time bins . To test the existence of inverted-U shaped relationships between gamma and alpha bins , we conducted repeated measures ANOVA’s and Gaussian shaped contrasts ( see section Response gain model test for details ) using SPSS 23 ( IBM , Inc ) . We used multiple regression to assess whether starting point could account for the correlation between gamma and drift bias . We used Pearson correlation to test the link between parameter estimates of the DDM and SDT frameworks and repeated measures correlation to test the link between gamma power and drift bias ( see previous section ) .	How do you decide whether to buy a new car ? One factor to consider is how well the economy is doing . During an economic boom , you might happily commit to buying a new vehicle that goes on sale , but prefer to sit on your savings during a financial crisis , despite how good the offer may be . Adjusting how you make decisions in situations like this can help you optimize choices in an ever-changing world . It’s currently thought that when deciding , we accumulate evidence for each of the available options . When evidence for one of the options passes a threshold , we choose that option . External factors – such as a booming economy when considering buying a car – could bias this process in two different ways . The standard view is that they move the starting point of evidence accumulation towards one of the two choices , so that the threshold for choosing that option is more easily reached . Alternatively , they could bias the accumulation process itself , so that evidence builds up more quickly towards one of the choices . To distinguish between these possibilities , Kloosterman et al . asked volunteers to press a button whenever they detected a target hidden among a stream of visual patterns . To bias their decisions , volunteers were penalized differently in two experimental conditions: either when they failed to report a target ( a ‘miss’ ) , or when they ‘detected’ a target when in fact nothing was there ( a ‘false alarm’ ) . As expected , punishing participants for missing a target made them more liberal towards reporting targets , whereas penalizing false alarms made them more conservative . Computational modeling of behavior revealed that when participants used a liberal strategy , they did not move closer to the threshold for deciding target presence . Instead , they accumulated evidence for target presence at a faster rate , even when in fact no target was shown . Brain activity recorded during this task reveals how this bias in evidence accumulation might come about . When a volunteer adopted a liberal response strategy , visual brain areas showed a reduction in low-frequency ‘alpha’ waves , suggesting increased attention . This in turn triggered an increase in high-frequency ‘gamma’ waves , reflecting biased evidence accumulation for target presence ( irrespective of whether a target actually appeared or not ) . Overall , the findings reported by Kloosterman et al . suggest that we can strategically bias perceptual decision-making by varying how quickly we accumulate evidence in favor of different response options . This might explain how we are able to adapt our decisions to environments that differ in payoffs and punishments . The next challenge is to understand whether such biases also affect high-level decisions , for example , when purchasing a new car .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	Humans strategically shift decision bias by flexibly adjusting sensory evidence accumulation
Sedimentary rocks host a vast reservoir of organic carbon , such as 2-methylhopane biomarkers , whose evolutionary significance we poorly understand . Our ability to interpret this molecular fossil record is constrained by ignorance of the function of their molecular antecedents . To gain insight into the meaning of 2-methylhopanes , we quantified the dominant ( des ) methylated hopanoid species in the membranes of the model hopanoid-producing bacterium Rhodopseudomonas palustris TIE-1 . Fluorescence polarization studies of small unilamellar vesicles revealed that hopanoid 2-methylation specifically renders native bacterial membranes more rigid at concentrations that are relevant in vivo . That hopanoids differentially modify native membrane rigidity as a function of their methylation state indicates that methylation itself promotes fitness under stress . Moreover , knowing the in vivo ( 2Me ) -hopanoid concentration range in different cell membranes , and appreciating that ( 2Me ) -hopanoids' biophysical effects are tuned by the lipid environment , permits the design of more relevant in vitro experiments to study their physiological functions . Lipids play essential roles in compartmentalizing cells for specific functions and creating barriers that are selectively permeable to the environment . The composition of lipids in cell membranes varies significantly and the basic biophysical properties of membranes , such as rigidity and permeability , can be adjusted based on growth conditions and environmental stressors ( Lipowsky and Sackmann , 1995; Los and Murata , 2004; Neubauer et al . , 2014 ) . One well-studied example is cholesterol in eukaryotic membranes . This essential sterol plays diverse roles in maintaining membrane structural integrity , modifying membrane rigidity , serving as a biosynthetic precursor for steroid hormones , vitamin D , and bile acids , or acting as a protein modifier for signaling pathways ( Hanukoglu , 1992; Gallet , 2011; Song et al . , 2014 ) . In addition to their important biological functions , lipids are of interest because they are more geostable than other biomolecules . For example , hopanoid molecular fossils , ‘hopanes’ , date back over a billion years ( Brocks et al . , 2005 ) and are so abundant that the global stock of hopanoids that can be extracted from sedimentary rocks is estimated to be 1013 or 1014 tons , more than the estimated 1012 tons of organic carbon in all living organisms ( Ourisson et al . , 1984 ) . In contrast to steroids , hopanoids are a less well studied but evolutionarily significant and chemically diverse class of lipids that are thought to be sterol surrogates in bacteria ( Figure 1 ) ( Rohmer et al . , 1979; Ourisson et al . , 1987 ) . 10 . 7554/eLife . 05663 . 003Figure 1 . Structures of selected hopanoids , cholesterol , and squalene . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 003 The rich record of ancient lipids , including fossil hopanoids , has long been recognized to hold clues into the early history of life and past environments ( Ourisson et al . , 1984; Summons and Walter , 1990; Brocks and Pearson , 2005; Knoll et al . , 2007 ) . But being able to confidently interpret the meaning of any ancient molecular fossil poses considerable challenges . First , we must be able to identify potential sources for these compounds , demanding unambiguous chemical parity between modern and ancient structures . Once this is achieved , understanding whether particular environmental conditions regulate the production of specific hopanoid variants becomes important . But arguably , the most critical goal in advancing our understanding of ancient lipids is being able to identify specific biological functions for their counterparts in cells today . Myriad hopanoid structures are known to exist ( Ourisson et al . , 1984 ) , yet we only poorly understand the significance of this chemical diversity . For meaningful linkages to be made between modern compounds and ancient biomarkers , we must ( 1 ) study those hopanoids that leave a specific trace in sedimentary rocks ( e . g . , their chemical modifications are geostable ) , ( 2 ) identify their in vivo function ( s ) , and ( 3 ) evaluate whether the roles played by these lipids in modern organisms have been conserved over the course of evolution . Following the recognition in the early 1970s that hopanes are ubiquitous in sedimentary rocks , the occurrence of hopanoids in diverse organisms was documented , and insights were gained into their biosynthesis , biophysical properties , and cellular functions ( Ourisson et al . , 1987; Ourisson and Rohmer , 1992; Pearson , 2013 ) . For example , studies using hopanoid-deficient mutants have shown that hopanoids promote resistance to antibiotics , detergents , extreme pH , and high osmolarity ( Welander et al . , 2009; Sáenz , 2010; Schmerk et al . , 2011; Malott et al . , 2012; Kulkarni et al . , 2013 ) . Biophysical studies using mixtures of hopanoids and model lipids have demonstrated that , like cholesterol , bacteriohopanetetrol cyclitol ether can condense membranes at high temperatures but fluidize membranes at low temperatures ( Poralla et al . , 1980 ) . Similarly , bacteriohopanetetrol ( BHT ) and bacteriohopanemonol can condense model membranes , and diplopterol ( Dip ) can form liquid ordered microdomains ( Kannenberg et al . , 1983; Nagumo et al . , 1991; Ourisson and Rohmer , 1992; Sáenz et al . , 2012 ) . A recent molecular modeling study pointed out different behaviors between Dip and BHT in their specific location within lipid bilayers and their capacity to condense membranes , suggesting complex roles of hopanoids due to their structural diversity ( Poger and Mark , 2013 ) . These biophysical studies have provided insights into the physical capabilities of these hopanoids . However , whether hopanoids play the same roles in vivo has been unclear due to the differences in lipid composition and concentration between model and cellular membranes . Among various hopanoid modifications , methylation of C-2 on the A-ring has drawn attention from Earth scientists because this modification is preserved episodically in ancient sedimentary rocks dating back to 1 . 6 billion years ago; accordingly , it has been suggested that 2-methylated hopanes ( 2Me-hopanes ) , the molecular fossils of 2Me-hopanoids , could potentially serve as biomarkers to interpret events in the early history of life ( Brocks et al . , 2005; Rasmussen et al . , 2008 ) . For a time , it was thought that 2Me-hopanes were biomarkers of cyanobacteria , and hence the process of oxygenic photosynthesis ( Summons et al . , 1999 ) , but we now know this not to be the case ( Welander et al . , 2010; Ricci et al . , 2015 ) . Intriguingly , spikes in the C30 2Me-hopane index ( ratio of methylated short hopanes to total short hopanes ) through geologic time are correlated with episodes of oceanic anoxic events ( OAEs ) , which are thought to have imposed heightened stress on the biosphere ( Knoll et al . , 2007 ) . Towards the goal of finding a robust interpretation for 2Me-hopanes , we have elucidated the biosynthetic pathway of 2Me-hopanoids ( Welander et al . , 2010 , 2012 ) , linked 2Me-hopanoids to specific environments and producers through ( meta ) genomic studies ( Ricci et al . , 2014 ) , and identified the stress-responsive pathway regulating transcription of the 2-methylase ( hpnP ) in the model hopanoid-producing bacterium Rhodospeudomonas palustris TIE-1 ( Kulkarni et al . , 2013 ) . A major challenge in understanding the function of 2Me-hopanoids has been the lack of a clear phenotype in vivo , despite dedicated attempts to find one for the hpnP mutant ( Kulkarni et al . , 2013 ) . Recent data suggest this phenotypic silence results from changes to the lipidome that compensate for the loss of 2Me-hopanoids ( unpublished ) . To infer a specific in vivo function for 2Me-hopanoids , in vitro studies that mimic cellular composition are therefore necessary . We hypothesized that methylation at C-2 would change hopanoids' packing with other lipids and proteins and affect membrane biophysical properties such as rigidity . Furthermore , we reasoned that such an effect would depend on the specific lipid composition of the membrane . To test these hypotheses , we took advantage of the existence of specific hopanoid-mutant strains ( Welander et al . , 2010 , 2012 ) , and recently developed protocols allowing quantitative analysis and the purification of hopanoids and their 2-methylated species in large quantities ( Wu et al . , 2015 ) . This experimental foundation set the stage for what we now report: in vitro membrane studies to examine how 2Me-hopanoids affect membrane rigidity in the context of different lipid environments of relevance to the cell . Our finding that hopanoid methylation enhances membrane rigidity supports the interpretation that past intervals of heightened 2Me-hopane abundance record a history of stress . To test whether 2-methylation changes membrane rigidity , we measured the membrane rigidity of specific hopanoid biosynthetic mutants ( Table 1 ) ( Welander et al . , 2012 ) using fluorescence polarization . Figure 2 shows the whole cell membrane rigidity measured at 25°C and 40°C . As expected , at higher temperature , the cell membrane became less rigid across all strains . At 25°C , the Δshc mutant , which lacks all hopanoids , had the least rigid membrane . This could be caused by both the absence of hopanoids and the accumulation of the hopanoid biosynthetic precursor , squalene . The result is also consistent with the observation that in model lipid vesicles , hopanoids make the membrane more rigid ( Kannenberg et al . , 1985 ) . Interestingly , the production of only short-chain hopanoids ( ΔhpnH ) is sufficient to recover the rigidity level close to that of the WT . However , when an adenosine molecule is attached to short hopanoids and accumulates in the ΔhpnG mutant , rigidity decreases to Δshc levels . Furthermore , in ΔhpnN where hopanoids are unable to be transported to the outer membrane ( Doughty et al . , 2011 ) , membrane rigidity is similar to ΔhpnG . These results have two implications . First , the hydrophobic reporter dye we used for fluorescence polarization measurements , diphenyl hexatriene ( DPH ) , reflects mostly the rigidity of the outer membrane . Second , the types of hopanoids and their respective localization between the inner and outer membranes directly impact membrane rigidity . 10 . 7554/eLife . 05663 . 004Table 1 . Mutant strains used for the whole cell membrane rigidity measurementsDOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 004GeneFunctionDeletion effectshc ( hpnF ) Cyclization of squalene to form C30 hopanoids ( diploptene and diplopterol ) No hopanoid production and accumulation of squalenehpnHAddition of adenosine to diploptene to generate adenosylhopane , a precursor for extended hopanoid productionNo extended hopanoid production , accumulation of C30 hopanoidshpnGRemoval of adenine from adenosyl hopaneNo BHT and aminoBHT production , accumulation of adenosylhopanehpnNAn IM transporter that transports hopanoids to the outer membraneAbsence of hopanoids in the OM and accumulation of hopanoids in the IMhpnOProduction of aminoBHTNo aminoBHT productionhpnPMethyl transfer to A ring at C-2No hopanoid methylationThe function of the gene and the effect of its deletion are listed . 10 . 7554/eLife . 05663 . 005Figure 2 . Whole cell membrane fluidity . Error bars represent the standard deviation from three biological replicates ( total 21–26 technical replicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 005 Interestingly , no obvious impact on rigidity was found in the absence of either bacteriohopane aminotriol ( ΔhpnO ) or 2-methylhopanoid ( ΔhpnP ) or both ( ΔhpnOP ) compared to WT ( Figure 2 ) . This observation could have two possible explanations . One is that these specific hopanoids do not affect membrane rigidity in cells . The other explanation is that the cells might synthesize other lipids that compensate for their effects so that no phenotype is observed . We can distinguish between these two scenarios by measuring membrane rigidity using vesicles made of model lipids and purified hopanoids . However , to design experiments that are physiologically relevant , we first needed to quantify hopanoids distribution in both the inner ( IM ) and outer membrane ( OM ) of R . palustris TIE-1 . It is well appreciated that lipids can have different subcellular localization , even in bacteria ( Matsumoto et al . , 2006 ) . To understand the biological roles of hopanoids , especially for 2-methylhopanoids , we measured the amounts of different hopanoids in the IM and OM of R . palustris TIE-1 WT and ΔhpnP using a previously described protocol ( Figure 3 ) ( Morein et al . , 1994; Wu et al . , 2015 ) . Table 2 shows that the total yield of the fractionated membranes was about 10% weight of dried cells , which is comparable to 9 . 1% in Escherichia coli ( Neidhardt and Curtiss , 1996 ) . The yields of the total lipid extract ( TLE ) from the lyophilized fractionated membranes were ∼12–15% from the IM and ∼5% for the OM . This yield is reproducible and could be due to a larger proportion of membrane proteins in R . palustris TIE-1 , lower recovery after lyophilization , or loss of certain lipid classes that were not quantitatively extracted by procedures optimized for hopanoids . The TLE yield in the OM was at least 50% lower than that in the IM , which is expected because the outer leaflet of the OM consists of more hydrophilic lipid A and lipopolysaccharides that are not extractable by the hydrophobic Bligh-Dyer lipid extraction method we employed . 10 . 7554/eLife . 05663 . 006Figure 3 . Membrane fractionation and hopanoids analysis using GC–MS . ( A ) Three distinct bands were formed after ultracentrifugation in a Percoll gradient . ( B ) The bands were recovered and resuspended . ( C ) Samples were ultracentrifuged to pellet down the purified membranes , which sat on top of a transparent solid Percoll layer . ( D ) GC–MS of fractionated membranes of R . palustris TIE-1 WT and ΔhpnP . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 00610 . 7554/eLife . 05663 . 007Table 2 . Purification yields of membrane fractionation using Percoll gradientDOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 007Weight % membranes in dry cellsTotalWeight % TLE in membranesWTinner4 . 7 ± 0 . 512 . 4 ± 0 . 8mix3 . 7 ± 1 . 25 . 6 ± 0 . 1outer3 . 3 ± 0 . 411 . 6 ± 1 . 44 . 9 ± 0 . 6ΔhpnPinner4 . 2 ± 1 . 315 ± 2 . 3mix2 . 3 ± 0 . 88 . 3 ± 2 . 3outer2 . 9 ± 0 . 69 . 4 ± 2 . 25 . 3 ± 1 . 4The yields in wt% of membrane fractionation . Errors represent standard deviation from three biological replicates . To quantify hopanoids , TLE from IM and OM was analyzed by GC-MS using androsterone as an internal standard and the differences in ionization efficiencies between androsterone and hopanoids were calibrated by external standards using purified hopanoids ( Figure 3 ) . Such calibration was recently shown to be essential for accurate hopanoid quantification ( Wu et al . , 2015 ) . Using this approach , the exact wt% of each hopanoid in TLE was obtained . However , to put the numbers in context and compare the value in mol% , we assumed the average molecular weight of the total lipids is 786 g/mol , the same as dioleoyl phosphatidylcholine ( DOPC ) and E . coli polar lipid extract ( PLE ) . Because we could only confidently quantify ( 2Me ) -Dip and ( 2Me ) -BHT , we focused our analyses on these four hopanoids . Figure 4 shows the hopanoid quantification results . In both WT and ΔhpnP , each type of hopanoid is enriched in the OM compared to the IM . The total of these four hopanoids in the IM is ∼2 . 6 mol% of TLE , whereas in the OM , the value can reach 8–11 mol% . For individual hopanoids in WT , the mol% in IM and OM are 1% and 2% for Dip , 1% and 2 . 4% for 2Me-Dip , 0 . 4% and 3 . 4% for BHT , and 0 . 1% and 0 . 3% for 2Me-BHT , respectively . In ΔhpnP , the IM and OM values are 2% and 7 . 5% for Dip , and 0 . 5% and 4 . 3% for BHT , respectively ( Figure 4 ) . This quantitative measurement of hopanoid content within the IM and OM can be used to evaluate the impact of 2-methylation upon hopanoid subcellular distribution . 10 . 7554/eLife . 05663 . 008Figure 4 . Molar percentage of hopanoids in the inner membrane ( IM ) and outer membrane ( OM ) of WT and ΔhpnP determined by GC–MS . Error bars represent the standard deviation from three biological replicates . Total hopanoids = sum of ( 2Me ) -Dip and ( 2Me ) -BHT . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 008 The ratio of total hopanoids in the OM vs IM is 3 . 1 ± 0 . 4 and 4 . 4 ± 0 . 6 for WT and ΔhpnP , respectively , which is a significant difference ( p = 0 . 038 ) . Comparing the ratios between short ( ( 2Me ) -Dip ) and long ( ( 2Me ) -BHT ) chain hopanoids revealed an enrichment of long-chain hopanoids in the OM compared to the IM in both WT and ΔhpnP ( Figure 5 ) . Interestingly , about equal amounts of 2Me-Dip and Dip were found in both IM and OM , whereas 2Me-BHT is 16% and 8% of BHT in the IM and OM , respectively , suggesting hopanoid 2-methylation has neither strong nor consistent effects on the partitioning of short and long species between the IM and OM ( Figure 5 ) . However , our data do indicate that 2-methylation impacts that total amount of hopanoid enrichment in the OM . 10 . 7554/eLife . 05663 . 009Figure 5 . Partitioning of hopanoids in the inner membrane ( IM ) and outer mebraane ( OM ) of R . palustris TIE-1 . ( A ) Molar ratio between short ( Dip and 2Me-Dip ) and long ( BHT and 2Me-BHT ) hopanoids in WT and ΔhpnP . ( B ) Molar ratio between methylated and desmethylated hopanoid in WT . Error bars represent the standard deviation from three biological replicates . p = 0 . 015; p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 009 To put these numbers in context and gain a deeper understanding on how the amount , chain-length , and 2-methylation of hopanoids impact membrane biophysical properties , we performed membrane rigidity measurements . Small unilamellar vesicles ( SUVs ) are commonly used for such measurements because it is straightforward to control their lipid composition ( Hope et al . , 1985 ) . We used fluorescence polarization to measure the fluorescence of DPH , a reporter dye , in the presence of model lipids and purified hopanoids . Figure 6 shows how the membrane rigidity of model lipids , DOPC and E . coli PLE , responds to the presence of cholesterol , squalene , ( 2Me ) -Dip , and ( 2Me ) -BHT . Because the total hopanoids in cell membranes are ∼10 mol% ( Figure 4 ) , we varied the concentration of hopanoids between 5 and 20 mol% in SUVs , which our quantification results suggest are physiologically relevant concentrations . Starting with the simplest single lipid background , DOPC , the addition of cholesterol increases membrane rigidity and the magnitude of change is proportional to the amounts of cholesterol added ( black line , Figure 6A ) . This observation is consistent with the literature and validates our technical approach ( Vanblitterswijk et al . , 1987 ) . Interestingly , the addition of the hopanoid precursor , squalene , has the opposite effect as cholesterol and makes the membrane less rigid in a concentration-dependent fashion ( magenta line , Figure 6A ) . However , when Dip is added , the membrane rigidity is unaffected ( blue line , Figure 6A ) . Surprisingly , given the structural similarity between cholesterol and Dip , Dip does not seem to rigidify DOPC as extensively as does cholesterol . However , when Dip is further processed by the cell to produce BHT , it rigidifies DOPC vesicles in the same manner as cholesterol ( red line , Figure 6A ) . 10 . 7554/eLife . 05663 . 010Figure 6 . Membrane rigidity measurements at 25°C using model lipids . ( A ) Dioleoyl phosphatidylcholine ( DOPC ) and ( B ) E . coli polar lipid extract ( PLE ) were mixed with different mol% of cholesterol , squalene , and hopanoids . Error bars represent the standard deviation from three biological replicates ( total 21 technical replicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 01010 . 7554/eLife . 05663 . 011Figure 6—figure supplement 1 . Membrane rigidity measurements at 25°C and 40°C using model lipids . ( A ) DOPC and ( B ) E . coli polar lipid extract ( PLE ) were mixed with different mol% of cholesterol , squalene , and hopanoids . Error bars represent standard deviation from three biological replicates ( total 21 technical replicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 011 2-methylation of both Dip and BHT has striking effects on their ability to rigidify DOPC . Not only does 2-methylation increase DOPC membrane rigidity , but at 20 mol% , 2Me-Dip even outperforms cholesterol and BHT ( cyan line , Figure 6A ) . 2Me-BHT also increases membrane rigidity similar to cholesterol , even though the difference from BHT is much smaller than that between Dip and 2Me-Dip ( orange line , Figure 6A ) . This result demonstrates methylation itself changes the biophysical properties of hopanoids , which directly impacts membrane fluidity . This observation reinforces the interpretation that the lack of phenotype in the ΔhpnP strain may be due to cells synthesizing other lipids that functionally complement the absence of 2Me-hopanoids , rather than due to a lack of an impact at the molecular level per se . To determine whether the effects on membrane rigidity by these hopanoids hold in more physiologically relevant environments , we repeated the experiment using E . coli PLE as the main component to form SUVs ( Figure 6B ) . E . coli PLE from Avanti Polar Lipids consists of 67% phosphatidylethanolamine ( PE ) , 23 . 2% phosphatidylglycerol ( PG ) , and 9 . 8% cardiolipin ( CL ) ( in wt% ) and has an average molecular weight identical to DOPC ( 786 g/mol ) . Compared to DOPC alone , the SUVs from E . coli PLE are more rigid , probably because the difference in fatty acid chain saturation ( green square , Figure 6A , B ) . Even though E . coli PLE SUVs are more rigid than DOPC SUVs to start with , we observe the same trend , where adding cholesterol rigidifies the vesicles and squalene fluidizes them . However , the concentration dependence on the rigidity change is more dramatic for cholesterol compared to squalene ( black line , Figure 6B ) . Unlike in DOPC , Dip has a small rigidifying effect in the E . coli PLE background , yet similar to DOPC , no concentration dependence is observed ( blue line , Figure 6B ) . Addition of 2Me-Dip also rigidifies E . coli PLE , but in contrast to what was observed for DOPC , 2Me-Dip shows lower capacity to rigidify the membrane than cholesterol ( cyan line , Figure 6B ) . Interestingly , the extended hopanoid , BHT , also rigidifies the membrane , but seemed to saturate at 10 mol% ( red line , Figure 6B ) . Surprisingly , in sharp contrast to the DOPC background , 2Me-BHT strongly rigidifies the E . coli PLE membranes and has a concentration dependence ( orange line , Figure 6B ) . This result shows that the impact of 2-methylation and hopanoid extension on a membrane biophysical property depends on the specific lipid environment . As expected , when these experiments were repeated at 40°C , membranes were less rigid overall . However , similar trends compared to 25°C were observed in both DOPC and E . coli PLE background ( Figure 6—figure supplement 1 ) . Given that 2-methylation of hopanoids differentially impacts membrane rigidity based on the lipid context , to understand the physiological effects of 2-methylhopanoids in the IM and OM of R . palustris TIE-1 , we must characterize the composition of phospholipids in native membranes and perform biophysical experiments using these membranes . To determine the exact quantity of each phospholipid in R . palustris TIE-1 , we purified the IM and OM as described above and analyzed the lipid composition by LC-MS using electron spray ionization ( Malott et al . , 2014 ) . The elution profiles between strains and membranes look similar ( Figure 7 ) . A total of 33 major phospholipids were identified , including 10 PC , 9 PE , 7 PG , and 7 CL ( Table 3 ) . To determine the absolute quantity of each phospholipid , exogenous standards of PC ( 17:0/17:0 ) , PE ( 17:0/17:0 ) , and PG ( 17:0/17:0 ) that are absent in R . palustris TIE-1 were added as internal standards for LC-MS analyses . Although we can detect cardiolipins , we unfortunately are unable to quantify them due to the low solubility of the cardiolipin standard in the LC-MS solvent . 10 . 7554/eLife . 05663 . 012Figure 7 . LC-MS profiles of the inner membrane ( IM ) and outer membrane ( OM ) of R . palustris TIE-1 WT and ΔhpnP . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 01210 . 7554/eLife . 05663 . 013Table 3 . Annotation of phospholipids identified by LC-MS analyses ( see Figure 7 ) DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 013CompoundRT ( min ) [M+H]+[M−C3H7O2HPO4]+[M+NH4]+PG32:14 . 39549 . 4895PG34:24 . 89575 . 505PC34:25 . 94758 . 5694PG34:16 . 11577 . 5208PG36:26 . 2603 . 5364PE34:26 . 38716 . 5225PC_cyc35:16 . 76772 . 5851PG_cyc356 . 95591 . 5364PG_cyc37:17 . 05617 . 5521PC ( 35:2 ) 7 . 06772 . 5851PE_cyc35:17 . 28730 . 5381PC34:17 . 51760 . 5851PC36:27 . 65786 . 6007PG ( 17:0/17:0 ) 7 . 86579 . 5364PG36:17 . 99605 . 5521PE34:18 . 14718 . 5381PE36:28 . 27744 . 5538PC_cyc358 . 58774 . 6007PC_cyc37:18 . 72800 . 6164PE_cyc359 . 28732 . 5538PE_cyc37:19 . 42758 . 5694PC ( 17:0/17:0 ) 9 . 75762 . 6007PC36:19 . 91788 . 6164PE ( 17:0/17:0 ) 10 . 53720 . 5538PE36:110 . 71746 . 5694PC_cyc3711 . 23802 . 632PE_cyc3712 . 1760 . 5851PC36:012 . 79790 . 634PE3612 . 98748 . 5851PC36:413 . 39782 . 569CL70:415 . 091447 . 0373CL68:315 . 11421 . 0217CL72:415 . 311475 . 0686CL70:315 . 341449 . 053CL68:215 . 351423 . 0373CL72:315 . 581477 . 0843CL70:215 . 61451 . 0686The types of lipids ( PC: phosphatidylcholine , PE: phosphotidylethanolamine , PG: phosphatidylglycerol , CL: cardiolipin , cyc: cyclopropanation; the first number indicates the total number of carbon of the fatty acid chains and the second number indicates the number of double bonds in these chains ) and their retention time ( RT , min ) and m/z value of the base peak are shown . For PC and PE , the base peak is the proton adduct and for CL , the base peak is the ammonium adduct . For PG , the base peak indicates a loss of glycerophosphate ( −171 m/z ) . Table 4 shows the wt% of each identified phospholipid in the TLE . Among the identified phospholipids , the IM in WT and ΔhpnP has ∼50% , ∼36% , and ∼12–14% of PC , PE , and PG , respectively , whereas the OM in WT and ΔhpnP has ∼56–58% , ∼30–33% , and ∼11–12% of PC , PE , and PG , respectively . However , when we calculate the total identified phospholipid amount , it only accounts for ∼26–34 wt% of the original TLE samples ( Table 4 ) . This low value could be due to ( 1 ) low solubility of cardiolipins and other unidentifiable lipids in LC-MS , ( 2 ) differences in the ionization efficiency of the phospholipids with different chain length or saturation ( Myers et al . , 2011 ) , and ( 3 ) the ionization suppression effects occur from co-eluting lipids ( Brugger et al . , 1997; Furey et al . , 2013 ) . Nevertheless , the average molecular weight of the phospholipids without cardiolipin is 738 g/mol . Considering the higher molecular weight of cardiolipin , our calculation for the mol% of hopanoids in TLE using an average molecular weight of 786 g/mol may be very close to the real value in R . palustris TIE-1 . 10 . 7554/eLife . 05663 . 014Table 4 . Phospholipid compositions in the inner membrane ( IM ) and outer membrane ( OM ) of R . palustris TIE-1 WT and ΔhpnP analyzed by LC-MSDOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 014Weight % of TLECompoundRT ( min ) WT IMWT OMΔhpnP IMΔhpnP OMPC36:27 . 655 . 15 ± 0 . 475 . 34 ± 0 . 265 . 44 ± 0 . 575 . 94 ± 0 . 48PC_cyc37:18 . 724 . 33 ± 0 . 454 . 01 ± 0 . 484 . 07 ± 1 . 133 . 96 ± 1 . 24PC36:19 . 912 . 90 ± 0 . 31 . 84 ± 0 . 522 . 59 ± 0 . 932 . 15 ± 1 . 08PC34:17 . 512 . 66 ± 0 . 222 . 23 ± 0 . 332 . 53 ± 0 . 582 . 38 ± 0 . 57PC34:25 . 941 . 56 ± 0 . 121 . 24 ± 0 . 441 . 41 ± 0 . 511 . 44 ± 0 . 48PC_cyc358 . 580 . 23 ± 0 . 010 . 14 ± 0 . 050 . 20 ± 0 . 070 . 17 ± 0 . 1PC_cyc35:16 . 760 . 12 ± 0 . 010 . 07 ± 0 . 020 . 11 ± 0 . 040 . 09 ± 0 . 06PC ( 35:2 ) 7 . 060 . 09 ± 0 . 010 . 05 ± 0 . 030 . 08 ± 0 . 030 . 06 ± 0 . 05PC_cyc3711 . 230 . 07 ± 0 . 010 . 04 ± 0 . 010 . 06 ± 0 . 020 . 05 ± 0 . 03PC36:012 . 790 . 02 ± 00 . 01 ± 00 . 02 ± 0 . 010 . 01 ± 0 . 01Sum17 . 1214 . 9616 . 5016 . 25PE_cyc37:19 . 424 . 92 ± 0 . 363 . 36 ± 0 . 944 . 68 ± 1 . 54 . 18 ± 1 . 64PE34:18 . 142 . 49 ± 0 . 191 . 44 ± 0 . 492 . 28 ± 0 . 841 . 82 ± 0 . 92PE36:110 . 712 . 29 ± 0 . 281 . 08 ± 0 . 382 . 05 ± 0 . 851 . 43 ± 0 . 95PE36:28 . 271 . 88 ± 0 . 211 . 22 ± 0 . 461 . 74 ± 0 . 651 . 59 ± 0 . 72PE34:26 . 380 . 43 ± 0 . 020 . 23 ± 0 . 080 . 42 ± 0 . 160 . 33 ± 0 . 21PE_cyc359 . 280 . 20 ± 0 . 020 . 11 ± 0 . 040 . 19 ± 0 . 070 . 15 ± 0 . 09PE_cyc35:17 . 280 . 12 ± 0 . 010 . 06 ± 0 . 020 . 11 ± 0 . 050 . 09 ± 0 . 06PE_cyc3712 . 10 . 07 ± 0 . 010 . 04 ± 0 . 010 . 07 ± 0 . 020 . 05 ± 0 . 03PE3612 . 980 . 04 ± 00 . 03 ± 0 . 010 . 04 ± 0 . 010 . 03 ± 0 . 01Sum12 . 447 . 5811 . 589 . 67PG36:26 . 22 . 59 ± 0 . 241 . 68 ± 0 . 542 . 25 ± 0 . 671 . 87 ± 0 . 8PG36:17 . 991 . 13 ± 0 . 080 . 69 ± 0 . 190 . 98 ± 0 . 310 . 81 ± 0 . 35PG34:16 . 110 . 77 ± 0 . 170 . 5 ± 0 . 20 . 6 ± 0 . 240 . 51 ± 0 . 19PG_cyc37:17 . 050 . 11 ± 0 . 020 . 07 ± 0 . 050 . 08 ± 0 . 020 . 07 ± 0 . 04PG34:24 . 890 . 05 ± 00 . 03 ± 0 . 010 . 04 ± 0 . 010 . 03 ± 0 . 02PG_cyc356 . 950 . 02 ± 00 . 01 ± 00 . 02 ± 0 . 010 . 01 ± 0 . 01PG32:14 . 390 . 00 ± 00 . 01 ± 00 ± 00 . 01 ± 0Sum4 . 6733 . 973 . 31PC + PE + PGTotal % of TLE34 . 2325 . 5432 . 0629 . 23 Due to the limitations we encountered in quantifying native phospholipids , we elected to generate SUVs directly using the TLE from fractionated R . palustris TIE-1 IM and OM rather than reconstituting SUVs from commercially available sources . Because our main focus is on the effect of 2-methylhopanoids , we used the membranes from ΔhpnP and added 5 , 10 , and 20 mol% purified hopanoids to constitute the SUV lipid mixture . We also included WT IM and OM for comparison . Figure 8 shows membrane rigidity measurements at 25°C using R . palustris TIE-1 membranes . In both WT and ΔhpnP , the OM showed higher membrane rigidity than IM , which could be due to the higher hopanoid content in the OM ( Figure 4 ) and differences in phospholipid content . Compared to WT , the ΔhpnP IM and OM showed decreased rigidity . This contrasts with our whole cell membrane rigidity results in which no difference in rigidity was observed ( Figure 2 ) . This discrepancy could be due to the presence of lipid A and outer membrane proteins that may affect the overall membrane rigidity in whole cells . 10 . 7554/eLife . 05663 . 015Figure 8 . Membrane rigidity measurements at 25°C using total lipid extract from R . palustris TIE-1 inner membrane ( IM ) and outer membrane ( OM ) . The IM ( A ) or OM ( B ) from ΔhpnP was mixed with different mol% of hopanoids . Error bars represent the standard deviation from three biological replicates ( total 21 technical replicates ) . p < 0 . 001 ( relative to ΔhpnP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 01510 . 7554/eLife . 05663 . 016Figure 8—figure supplement 1 . Membrane rigidity measurements at 40°C using total lipid extract from R . palustris TIE-1 inner membrane ( IM ) and outer membrane ( OM ) . The IM ( A ) or OM ( B ) from ΔhpnP was mixed with different mol% of hopanoids . Small unilamellar vesicles from the lipid mixtures were prepared , and the membrane rigidity was measured by fluorescence polarization of a reporter dye diphenyl hexatriene . Error bars represent standard deviation from three biological replicates ( total 21 technical replicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05663 . 016 Similar to our observations in model lipids , the addition of Dip has no effect on membrane rigidity , both in the IM and OM . However , 2Me-Dip rigidifies both IM and OM in a concentration-dependent manner , as seen in the model lipid SUVs . Interestingly , unlike the effects BHT exerts in model lipids , it does not rigidify either the IM or OM . While it is possible that BHT has no effect on native membrane rigidity , given that ΔhpnP membranes contain BHT , it seems more likely that endogenous BHT is saturating its membrane rigidifying capacity , similar to what we observed for E . coli PLE ( red line , Figure 6B ) . When BHT is methylated , it rigidifies both the IM and OM . However , different trends can be seen between these membranes: in the IM , the more 2Me-BHT present , the higher the membrane rigidity , yet OM membrane rigidity appears to saturate by 5 mol% 2Me-BHT and remains constant between 5–20 mol% of 2Me-BHT ( Figure 8B ) . It is tempting to speculate that the reason less 2-methylation occurs for BHT than Dip in the OM ( Figures 4 , 5 ) is because less methylation of BHT is needed to significantly impact OM rigidity . We repeated these experiments at 40°C and observed similar trends as seen at 25°C ( Figure 8—figure supplement 1 ) . However , we had larger standard deviations than in our model lipid experiments , which could be due to higher heterogeneity in the samples extracted from IM and OM . Nevertheless , we find clear physiologically relevant distinctions in the rigidifying effects of both short and long 2Me-hopanoids on the IM and OM of R . palustris TIE-1 . Until now , a specific role for 2Me-hopanoids in living cells has evaded experimental detection , yet its identification has been of great interest for interpreting the extensive 2Me-hopane fossil record ( Welander et al . , 2010; Kulkarni et al . , 2013 ) . Our findings that 2Me-hopanoids rigidify membranes to different extents depending both on their specific structure ( short or long ) and lipid context not only provide a clear biological function for these compounds , but also help rationalize why previous efforts to identify such a function have been challenging . That methylation per se can contribute to rigidifying membranes may also help explain the association of methylated hopanoids in certain modern and ancient environments . Under what circumstances would adding a methyl group at the 2′ position of hopanoids , which seems a rather small modification , be beneficial ? Might there be a mechanistic explanation for the enrichment of 2Me-hopanes during stressful OAEs ( Knoll et al . , 2007 ) ? Several independent lines of evidence bridge our biophysical findings with the abundance patterns in the rock record , together suggesting that 2-methylhopanoids confer stress resistance: ( 1 ) 2Me-hopanoids are enriched in the outer membrane of akinetes , survival cell types of the cyanobacterium Nostoc punctiforme ( Doughty et al . , 2009 ) , ( 2 ) a stress-responsive pathway upregulates the HpnP methylase in R . palusris ( Kulkarni et al . , 2013 ) , ( 3 ) in modern environments , the capacity for 2Me-hopanoid production significantly correlates with organisms , metabolisms ( e . g . , nitrogen fixation ) , and environments that support plant–microbe interactions ( Ricci et al . , 2014 ) . This correlation , together with the observation that ( 2Me ) -hopanoids promote symbiosis ( Silipo et al . , 2014 ) , tempt us to speculate that 2Me-hopanoids may indirectly facilitate nitrogen fixation by enhancing bacterial survival under the stressful conditions that accompany the establishment of symbiosis ( Gibson et al . , 2008 ) . Going forward , it is worth critically examining this hypothesis; if correct , such an interpretation would indicate that spikes in the 2Me-hopane index may reflect episodes of particular environmental stresses favoring the growth of organisms capable of withstanding it using ( 2Me ) -hopanoids . How might the 2-methylation of hopanoids permit such an adventitious adaptation ? Molecular dynamic simulations of 2Me-hopanoid within a relevant lipid context are required to understand the interactions on an atomic scale . However , we may speculate about the mechanism of rigidification through lessons learned from studies of cholesterol , in which the number of methyl groups control its optimal molecular packing to geometrically complement phospholipid chains ( Bloch , 1979 ) . The addition or removal of methyl groups over the evolution of the cholesterol lipid family is thought to have optimally tuned cholesterol's ability to order or condense phospholipid membranes ( Miao et al . , 2002; Rog et al . , 2007 ) . We suggest that when a methyl group is added onto the 2′ position of A-ring in hopanoid , the stearic hindrance between the 2-methyl group and the methyl groups at the 4′ and 10′ position of the A-ring could transform the ring from a chair to a twisted conformation . The two additional 1 , 3-diaxial interactions elicited by hopanoid 2-methylation could mimic the smoothing and/or tilting effect known for cholesterol , thus rationalizing how 2-methylation may improve the ability of hopanoids to rigidify membranes ( Rog et al . , 2007 ) . The differences in ( 2Me ) -hopanoid distribution between the IM and OM pose many interesting questions about their role in maintaining membrane integrity and homeostasis . Compared to ( 2Me ) -Dip , the addition of a hydrophilic tail to form ( 2Me ) -BHT or even BHT-glucosamine ( Figure 1 ) may favor a stronger interaction with the outer leaflet of the OM , in which the lipid head group is heavily modified with hydrophilic molecules . The relative enrichment of ( 2Me ) -BHT in the OM is consistent with such a scenario . The hydrophilic tail of 2Me-BHT could also affect the vertical position of the 2-methyl group in the membrane compared to 2Me-Dip ( Rog et al . , 2007; Poger and Mark , 2013 ) , which may explain the difference between 2Me-BHT and 2Me-Dip in rigidifying membranes with different compositions . Future research will illuminate whether there are additional interactions between hopanoids and other membrane constituents ( e . g . , proteins or cell wall components ) that facilitate survival under stress . Finally , it is important to keep in mind that ( 2Me ) -hopanoids may act locally rather than globally with respect to influencing membrane biophysical properties . In our in vitro experiments , we did not observe a significant difference in membrane rigidity when less than 10 mol% of ( 2Me ) -hopanoids were used ( Figure 6 ) . This may be due to a critical concentration needed to trigger an effect , which is consistent with molecular dynamic simulations that demonstrate cholesterol starts to self-organize within membranes at concentration above 10 mol% ( Martinez-Seara et al . , 2010 ) . However , we hasten to point out that the mol% of ( 2Me ) -hopanoids in the native membrane experiment ( Figure 8 ) are not directly comparable to those using model lipids ( Figure 6 ) due to the presence of endogenous hopanoids in ΔhpnP membranes . The existence of BHT in ΔhpnP may explain why the membrane rigidifying effect of exogenously added BHT is saturated at 10 mol% in the E . coli PLE but has no impact on R . palustris TIE-1 IM and OM . In this context , it is noteworthy that cardiolipin significantly increases in the absence of all hopanoids ( unpublished ) . In E . coli , cardiolipin localizes to negatively curved regions of the cell ( Renner and Weibel , 2011 ) . Looking beyond the function of methylation , it is possible that certain hopanoid types could fulfill the geometry requirements of curved membranes and facilitate cell division or vesicle formation , consistent with both the microdomain features observed in prior subcellular hopanoid localization studies ( Doughty et al . , 2014 ) and the strong cell division defect displayed by a mutant lacking the ability to transport hopanoids to the OM ( Doughty et al . , 2014 ) . Going forward , consideration of other roles for structurally diverse hopanoids , including the possibility that some might influence membrane protein function ( Phillips et al . , 2009 ) , modify specific proteins or cell wall components through covalent linkages ( Jeong and McMahon , 2002; Silipo et al . , 2014 ) , or even play a role in signaling pathways in analogy to cholesterol and phosphatidylcholine ( Kuwabara and Labouesse , 2002; Aktas et al . , 2010 ) , will enhance our appreciation for this ancient lipid class . R . palustris TIE-1 wild type ( WT ) and mutant strains were grown as previously described ( Welander et al . , 2012 ) . Purified hopanoids ( ( 2Me ) -diplopterol , ( 2Me ) -bacteriohopanetetrol [BHT] ) were obtained by following the purification protocols ( Wu et al . , 2015 ) . E . coli polar lipid extract and 1 , 2-dioleoyl-sn-glycero-3-phosphocholine ( DOPC ) , 1 , 2-diheptadecanoyl-sn-glycero-3-phosphocholine ( PC[17:0/17:0] ) , 1 , 2-diheptadecanoyl-sn-glycero-3-phosphoethanolamine ( PE[17:0/17:0] ) , 1 , 2-diheptadecanoyl-sn-glycero-3-phospho- ( 1′-rac-glycerol ) ( PG[17:0/17:0] ) were from Avanti Polar Lipids ( Alabaster , AL ) . Squalene , cholesterol , pyridine , acetic anhydride , morpholinepropanesulfic acid ( MOPS ) , 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid ( HEPES ) , sodium succinate , 1 , 6-diphenyl-1 , 3 , 5-hexatriene ( DPH ) , Percoll , and tetrahydrofuran ( THF ) were from Sigma–Aldrich ( Milwaukee , WI ) . Yeast extract was from HIMEDIA ( Mumbai , India ) . Peptone was from BD Biosciences ( San Jose , CA ) . Methanol and dichloromethane ( DCM ) were HPLC grade from Alfa Aesar ( Ward Hill , MA ) . To prepare bacterial cells for measurements of membrane fluidity , single colonies of R . palustris TIE-1 WT and mutants were inoculated into 10 ml YPMS ( 0 . 3% yeast extract , 0 . 3% peptone , 50 mM MOPS , 5 mM succinate , pH 7 . 0 ) and grown at 30°C , 250 rpm for ∼72 hr to reach late stationary phase ( OD600 ∼1 . 0 ) . The cells ( 250 μl ) were spun down and the cell pellets were washed once with 50 mM HEPES , 50 mM NaCl , pH 7 . 0 ( buffer A ) . Pellets were resuspended in different amounts of the same buffer to adjust the final OD600 to ∼0 . 2 . To measure membrane fluidity , 4 . 3 μl of DPH ( 736 μM stock solution in ethanol; the concentration was determined by ε350 nm = 88 cm−1 mM−1 in methanol ) was added into 400 μl of the cell suspension . Samples were incubated in a 25°C or 40°C water bath without light for 30 min , followed by measurements of fluorescence polarization ( Fluorolog , HORIBA Instruments ( Edison , NJ ) . Instrument parameter: ex 358 nm , slit = 3 mm; em 428 nm , slit = 7 mm; integration time = 1 s ) ( Lin et al . , 2011 ) . Three biological replicates were measured , each containing 6–14 technical replicates . To reduce bias from the stability of the instrument and the samples , especially at 40°C , we randomized our data acquisition sequence . p value in this manuscript represents t-test using two-tailed equal variance . To prepare cell cultures for membrane fractionation , single colonies of R . palustris TIE-1 WT or ΔhpnP mutant were inoculated into 10 ml YPMS and grown for ∼4 days at 30°C , 250 rpm . The culture ( 0 . 5 ml ) was then inoculated into 1 l of YPMS in a 2-l flask and grown at 30°C , 250 rpm for 4 days before harvesting by centrifugation at 12 , 000×g for 20 min at RT . The typical yield was ∼1 . 8 g of wet cell paste per 1 l culture . To estimate the yield in dried cells , a small aliquot of the wet cell paste was lyophilized until there was no further change in weight . On average the weight of dried cells was one third of wet cells . The wet cell pastes were stored at −80°C before cell lysis . To lyse the cells , 19 ml of buffer A was added into ∼3 . 6 g of cells ( from 2 l culture ) and passed through a French Press twice at 14 , 000 psi , followed by sonication ( Sonic Dismembrator 550 , Fisher Scientific ( Waltham , MA ) , 1/8 inch tip , power output 3 . 5 , 1 s on , 4 s off , total on time 5 min at 4°C ) . The cell debris was spun down at 20 , 000×g , 20 min at 4°C . The supernatant containing cell membranes was transferred into 4-ml ultracentrifugation tubes ( ∼3 ml sample per tube ) and centrifuged at 80 , 000 rpm in a TLA-100 . 3 rotor for 1 hr at 4°C ( Optima MAX Ultracentrifuge , Beckman Coulter ( Brea , CA ) ) . The resulting membrane pellets in each tube were resuspended in 300 μl buffer A by pipetting while being sonicated in a bath sonicator ( VWR ( Radnor , PA ) B2500A-DTH , 42 kHz , RF Power 85 W ) . The suspension was combined into one single tube and sonicated again using the probe sonicator ( power level 2 . 5 , 1 s on , 4 s off , total on time 2 . 5 min , 4°C ) . To separate inner and outer membranes , ∼320 μl of membrane samples were laid on top of 3 ml 18% Percoll ( vol/vol in buffer A ) , followed by ultracentrifugation at 30 , 000 rpm in a TLA 100 . 3 rotor at 4°C for 15 min ( Morein et al . , 1994 ) . Three visually distinct bands were formed and a pipetman was used to take in sequence of the top band ( 1 ml ) , bottom band ( ∼200–250 μl ) , and the middle band ( 0 . 7–1 ml ) ( Figure 3 ) . The top and bottom bands constituted the IM and OM , determined by the presence and absence of NADH-oxidase activity , respectively . The band on top of the OM was less defined and exhibited some NADH-oxidase activity , which may be an artifact from sonication steps that mixed the IM and OM , and we therefore discarded it in our subsequent studies . To remove Percoll , samples from the same band were combined and centrifuged in a TLA 100 . 3 rotor at 50 , 000 rpm at 4°C for 1 . 5 hr . After centrifugation , a layer of the fractionated membrane was formed on top of a transparent Percoll layer ( Figure 3 ) . The membrane layers were collected by pipetting gently in water and/or scraped gently using a metal spatula . The membrane samples were then frozen at −20°C before being lyophilized . The total lipid extractions ( TLE ) from the lyophilized membranes were obtained by modified Bligh–Dyer extraction according to published protocols ( Kulkarni et al . , 2013 ) . GC-MS ( Thermo Scientific ( Waltham , MA ) Trace-GC/ISQ mass spectrometer with a Restek Rxi-XLB column [30 m × 0 . 25 mm × 0 . 10 μm] ) was used to quantify hopanoids from the TLE from the IM and OM . An internal standard , androsterone ( 750 ng ) was air dried with 100 μg of the TLE overnight at RT and derivatized with 50 μl acetic anhydride and 50 μl pyridine at 60°C for 30 min , followed by GC-MS analyses as described ( Welander et al . , 2009; Kulkarni et al . , 2013 ) . To account for the difference in ionization efficiencies between androsterone and hopanoids , calibration curves using androsterone and purified hopanoids were generated to quantify hopanoids ( Wu et al . , 2015 ) . LC-MS ( Waters ( Milford , MA ) Acquity UPLC/Xevo G2-S time-of-flight mass spectrometer with a CSH C18 column [2 . 1 × 100 mm × 1 . 7 μm] ) was used to quantify both phospholipids and hopanoids . Phospholipid internal standards ( PC[17:0/17:0] , PE[17:0/17:0] , PG[17:0/17:0] , 1 μg each ) were mixed with 100 μg of the TLE from fractionated membranes and air-dried overnight at RT . LC solvent ( 200 µl , Isopropanol:acetonitrile:water = 2:1:1 ) was then added into the samples , followed by sonication before analyses by LC-MS as described earlier ( Malott et al . , 2014; Wu et al . , 2015 ) . The column temperature was maintained at 55°C . A binary solvent system containing solvent A ( acetonitrile:water; 60:40 ) and solvent B ( isopropanol:acetonitrile; 90:10 ) , both with 10 mM ammonium formate and 0 . 1% formic acid was used . The flow rate was set at 400 μl/min and the elution program started at 40% B , increased linearly to 43% B in 2 min , then to 50% B in 0 . 1 min , followed by a linear increase to 54% B over 9 . 9 min , a jump to 70% B in 0 . 1 min , another linear increase to 99% B over 5 . 9 min , a subsequent decrease to 40% B in 0 . 1 min , and then maintained at the same level for 1 . 9 min . The eluents from the column were ionized by electrospray ionization ( ESI ) . MSE data from 100 to 1500 m/z were collected in either the positive or negative ion mode . MSE consisting of both low energy and high energy scans were obtained simultaneously . During data analysis product ions can be associated with parent ions if they are coincident in chromatographic time . Electrospray conditions were capillary voltage 2 . 0 kV , cone voltage 30 V , source offset 60 V , source temperature 120°C , desolvation temperature 550°C , cone gas 20 l/hr , and desolvation gas 900 l/hr . The TOF-MS was run in resolution mode , typically 32 , 000 m/Δm . The mass axis was calibrated with sodium formate clusters . Leucine enkephalin was used as a mass reference during acquisition . The data were collected in continuum mode , and then converted to centroid mode for quantitative analysis using the Quanlynx program ( Waters Corporation , Milford , MA ) ( Wu et al . , 2015 ) . Cholesterol , squalene , and hopanoid stock solutions were prepared at 1 mg/ml in THF and the E . coli PLE and DOPC were prepared at 10 mg/ml in DCM . To prepare lipid mixture , a total of 1 μmol of lipid was added into 0 . 5 ml of DCM and dried in a rotary evaporator . Because any residual solvents can cause high errors in fluidity measurements , the samples were placed under vacuum overnight to ensure complete removal of organic solvents . To prepare SUVs , 1 ml of buffer A was added into the glass vials containing dried lipid mixtures . Samples were suspended by sonication for 1 hr at RT in a bath sonicator ( VWR B2500A-DTH , 42 kHz , RF Power 85 W ) . The suspended lipids ( murky giant multilamellar vesicles ) were transferred into 1 . 5-ml eppendorf and flash frozen in liquid nitrogen for 3 min , followed by thawing in a 37°C water bath for 3 min . This freeze–thaw cycle , which breaks down the giant vesicles into smaller ones , was repeated two more times . SUVs were prepared by passing the samples through 0 . 1-μm polycarbonate membranes ( Whatman ) using Avanti mini-Extruder at RT ( Avanti Polar Lipids ) . The extrusion was performed a total of 11 times and the vesicle suspension became clear during the process . The sizes and stability of the SUVs was determined by dynamic light scattering ( Wyatt ( Santa Barbara , CA ) DynaPro NanoStar . Instrument parameters: acquisition time 5 s , number of acquisition 10 , laser wavelength 659 nm , laser power 10% , 25°C ) . The average size distribution of the SUVs was between 80 and 90 nm and remained stable for at least 4 hr at RT . SUVs after extrusion were diluted 1:1 in buffer A to reach a final concentration of 0 . 5 mM ( 400 μl total volume ) . DPH ( 1 . 8 μl of 44 . 5 μM stock solution in ethanol ) was added into the sample and vortexed immediately . The SUV-DPH samples were incubated in 25°C or 40°C water bath without light for at least 30 min before the fluorescence polarization was measured using the parameters described above . The concentration of the fluorescence reporter dye DPH and the instrument parameters were optimized to have a strong and linear signal output . Different amounts of purified hopanoids ( 5 , 10 , and 20 nmol ) were added to 100 nmol of the inner or outer membrane extracts of ΔhpnP ( assuming average molecular weight is 786 g/mol ) . The same procedures as described above were followed for the preparation of SUVs . Buffer A ( 600 μl ) was used to suspend the dried lipids so that the final lipid concentrations before the addition of DPH were between 0 . 088 and 0 . 1 mM ( 400 μl sample volume ) . To measure fluorescence polarization , DPH ( 1 . 8 μl of 7 . 4 μM stock solution in ethanol ) was used ( the final concentration of DPH was 0 . 03 μM , which was ∼0 . 03 mol% of the total lipids in the sample ) . Controls of membranes from WT or ΔhpnP only without addition of hopanoids were included .	The cell membrane that separates the inside of a cell from its outside environment is not a fixed structure . A cell can change the amount and type of different molecules in its membrane , which can alter the rigidity and permeability of the membrane and allow the cell to adapt to changing conditions . The cell membranes of many bacteria contain molecules called hopanoids . Hopanes are the fossilized forms of these molecules and many hopanes are found extensively in sedimentary rocks . For example , 2-methylated hopanes—the fossilized forms of hopanoids that have a methyl group added to a particular carbon atom—have been found in ancient rocks that formed up to 1 . 6 billion years ago . Many researchers have suggested that 2-methylated hopanes ( and other molecular fossils ) in sedimentary rocks could act as ‘biomarkers’ and be used to deduce what primitive life and ancient living conditions were like . Millions of years ago , several periods occurred where the Earth's oceans lost almost all of their oxygen; this likely placed all life on Earth under great stress . A greater proportion of the hopanes found in rocks formed during those periods are methylated than those seen in rocks from other time periods . However , it was difficult to interpret this observation about the fossil record , as the role of 2-methylated hopanoids in living bacterial cells was unknown . Wu et al . have now investigated the role of 2-methylated hopanoids by performing experiments on bacterial membranes and found that 2-methylated hopanoids help the other molecules that make up the membrane to pack more tightly together . This makes the membrane more rigid , and the extent of this stiffening depends on the length of the 2-methylated hopanoid and on the other molecules that are present in the membrane . A more rigid membrane would protect the bacteria more in times of stress; therefore , rock layers containing an increased amount of 2-methylhopane are likely to indicate times when the bacteria living at that time were under a great deal of stress .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	Methylation at the C-2 position of hopanoids increases rigidity in native bacterial membranes
The collapse of iconic , keystone populations of sockeye ( Oncorhynchus nerka ) and Chinook ( Oncorhynchus tshawytscha ) salmon in the Northeast Pacific is of great concern . It is thought that infectious disease may contribute to declines , but little is known about viruses endemic to Pacific salmon . Metatranscriptomic sequencing and surveillance of dead and moribund cultured Chinook salmon revealed a novel arenavirus , reovirus and nidovirus . Sequencing revealed two different arenavirus variants which each infect wild Chinook and sockeye salmon . In situ hybridisation localised arenavirus mostly to blood cells . Population surveys of >6000 wild juvenile Chinook and sockeye salmon showed divergent distributions of viruses , implying different epidemiological processes . The discovery in dead and dying farmed salmon of previously unrecognised viruses that are also widely distributed in wild salmon , emphasizes the potential role that viral disease may play in the population dynamics of wild fish stocks , and the threat that these viruses may pose to aquaculture . Pacific salmon ( Oncorhynchus spp . ) species have supported coastal ecosystems and Indigenous populations surrounding the North Pacific Ocean for tens of millennia . Today , through their anadromous life history , salmon continue to transport nutrients between aquatic and terrestrial environments ( Cederholm et al . , 1999 ) , supply the primary food sources for orca whales and sea lions ( Wasser et al . , 2017; Willson and Halupka , 1995; Chasco et al . , 2017; Thomas et al . , 2017 ) and provide economic livelihoods for local communities ( Noakes et al . , 2002 ) . In the Northeast Pacific , widespread declines of Chinook ( O . tshawytscha ) and sockeye ( O . nerka ) salmon have occurred in the last 30 years , leading some populations to the brink of extirpation ( Peterman and Dorner , 2012; Heard et al . , 2007; Miller et al . , 2011; Jeffries et al . , 2014 ) , and a cause of great concern to Indigenous groups , commercial and recreational fishers , and the general public . Although the exact number of salmon spawning in rivers is unknown , there are large declines in sockeye salmon over a large geographic area ( Peterman and Dorner , 2012 ) . Similarly , Chinook salmon stocks are at only a small percentage of their historic levels , and more than 50 stocks are extinct ( Heard et al . , 2007 ) . It is thought that infectious disease may contribute to salmon declines ( Miller et al . , 2011 ) , but little is known about infectious agents , especially viruses , endemic to Pacific salmon . Infectious disease has been identified as a potential factor in poor early marine survival in migratory salmon; an immune response to viruses has been associated with mortality in wild migratory smolts and adults ( Miller et al . , 2011; Jeffries et al . , 2014 ) , and in unspecified mortalities of salmon in marine net pens in British Columbia ( BC ) ( Miller et al . , 2017; Di Cicco et al . , 2018 ) . For instance , immune responses to viruses such as Infectious haematopoietic necrosis virus ( IHNV ) and potentially undiscovered viruses , have been associated with mortality in wild juvenile salmon ( Jeffries et al . , 2014 ) . This is an important observation as mortality of juvenile salmon can be as high as ~90% transitioning from fresh water to the marine environment ( Clark et al . , 2016 ) . Together , these suggest that there are undiscovered viruses which may contribute to decreased survival of Pacific salmon but a concerted effort to look for viruses that may contribute to mortality has been absent . Here , virus-discovery was implemented to screen for viruses associated with mortality . Together , sequencing of dead or moribund aquaculture salmon and live-sampled wild salmon , in-situ hybridization , and epidemiological surveys revealed that previously unknown viruses , some of which are associated with disease , infect wild salmon from different populations . Fish were screened against a viral disease detection biomarker panel ( VDD ) that elucidates a conserved transcriptional pattern indicative of an immune response to active RNA viral infection ( Miller et al . , 2017 ) . For instance , in a previous study , we showed that 31% of moribund Atlantic salmon were in a viral disease state , and half of these were not known to be positive for any known RNA viruses ( Di Cicco et al . , 2018 ) . Individuals that were strongly VDD-positive , but negative for any known salmon viruses ( e . g . Piscine orthoreovirus , Erythrocytic necrosis virus , Infectious pancreatic necrosis virus , Infectious hematopoietic necrosis virus , Infectious salmon anaemia virus and Pacific salmon paramyxovirus ) were subject to metatranscriptomic sequencing . The sequencing revealed viral transcripts belonging to members of the Arenaviridae , Nidovirales and Reoviridae , three evolutionarily divergent groups of RNA viruses ( Figure 1 ) that can be highly pathogenic ( Yun and Walker , 2012; Liang et al . , 2014; Weiss and Leibowitz , 2011 ) . One of the challenges of viral discovery in fish is that the proportion of viral transcripts in vertebrate metatranscriptomic libraries is small compared to the number of transcripts from the host and other contaminating sequences ( Geoghegan et al . , 2018; Zhang et al . , 2019 ) . However , we were able to achieve near-coding complete genomes for the three new viruses ( Figure 1—figure supplement 1A and B ) . The genomic organisation of the newly discovered viruses was consistent with related viruses in fish . For instance , SPAV has three genomic segments , as shown for other arenaviruses in fish ( Shi et al . , 2018 ) . High-throughput RT-PCR screening of >6000 wild juvenile Chinook and sockeye salmon showed dissimilar geographical distributions of infected fish , reflecting differences in epidemiological patterns of transmission and infection dynamics for each of the viruses ( Figure 2 ) . The distribution and abundance of the different viruses varied markedly . Arenaviruses were relatively common ( Figure 2—figure supplement 1 ) and geographically widespread in migratory juvenile Chinook and sockeye salmon in the marine environment ( Figure 2 , Figure 2—figure supplement 2 ) . Whereas , the nidovirus was spatially localised and predominantly observed at high prevalence over multiple years in Chinook salmon leaving freshwater hatcheries ( Figure 2 ) . Finally , the reovirus was detected only in farmed Chinook salmon ( Figure 2 and Figure 2—figure supplement 1 ) . With the exception of their relatively recent discovery in snakes ( Stenglein et al . , 2012 ) and frogfish ( Shi et al . , 2018 ) , arenaviruses were thought to solely infect mammals . The arenaviruses reported here share less than 15% amino-acid sequence similarity ( in the RdRp ) to those from mammals and snakes , and define a new monophyletic evolutionary group , the pescarenaviruses ( Figure 1A ) . The absence of clear sequence homology in the glycoprotein , the difference in genome segmentation ( Shi et al . , 2018 ) , as well as phylogenetic analysis of the replicase demonstrate that pescarenaviruses share a common but ancient ancestor with arenaviruses infecting snakes and mammals . We recommend these fish-infecting arenaviruses are assigned to the new genus Pescarenavirus , with those infecting Chinook and sockeye salmon being assigned to the species Salmon pescarenavirus ( SPAV ) , strains 1 and 2 , respectively . Farmed Chinook salmon positive for SPAV-1 displayed pathology and symptoms consistent with disease including inflammation of the spleen and liver , as well as tubule necrosis and hyperplasia in the kidney . Clinically , salmon presented with yellow fluid on the pyloric caeca and swim bladder , pale gills with haemorrhaging on the surface , and anaemia . Wild Chinook and sockeye that tested positive for arenavirus infection , but which were clinically healthy when sampled , showed few histological lesions . In-situ hybridization revealed that arenaviruses were concentrated mainly in macrophage-like cells , melanomacrophages , red-blood cells ( RBCs ) and endotheliocytes ( Figure 3 ) . These findings are consistent with localisation of arenaviruses in mammals and snakes , although in contrast to snakes and fish , mammalian red blood cells are not nucleated so the similarity likely only extends to nucleated cells . SPAV-1 and −2 shared similar cell tropism within Chinook and sockeye salmon , respectively ( Figure 3—figure supplement 1 ) . In one out of the eight Chinook samples examined , moderate chronic-active hepatitis was reported , and staining for SPAV-1 was identified in the area affected by inflammation ( Figure 3C and D ) , while in the other samples SPAV-1 was confined to reticuloendothelial cells in the liver tissue or in the sinusoids . More lesions were observed in dead farmed Chinook , where disease progression is more advanced . Our observations indicate that arenaviruses are replicating in red-blood cells , and occur in the macrophages and leukocytes that consume the infected cells . Moreover , the observed pathological changes in arenavirus-infected fish , including anaemia , and lesions in the gills , kidney and liver would be expected for viruses that infect red-blood cells . These results are the first empirical evidence for arenavirus infection in fish , and suggest that SPAV , like many other arenaviruses , has the potential to be a causative agent of disease . Sequencing of cultured Chinook salmon also revealed a previously undescribed nidovirus and reovirus . Phylogenetic analysis of the reovirus , named Chinook aquareovirus ( CAV ) , predicts that it is part of the genus , Aquareovirus ( Figure 1B ) . Rather than being most closely related to known reoviruses of salmon ( Winton et al . , 1981 ) , CAV groups with a growing number of aquareoviruses , some of which are known to cause haemorrhagic disease and have led to serious losses to aquaculture in China ( Nibert and Duncan , 2013; Wang et al . , 2012 ) . The observed clinical signs ( anemia , dark spleen , and blood-filled kidneys ) in dead farmed Chinook salmon with high loads of CAV are consistent with a haemorrhagic manifestation . The novel nidovirus , named Pacific salmon nidovirus ( PsNV ) , is most closely related to the recently described Microhyla alphaletovirus 1 , which forms a sister group to the coronaviruses ( Bukhari et al . , 2018 ) . This sequence , alongside PsNV are basal to all other Nidovirus families , and their long branch length suggests they each belong to a different genus ( Figure 1C ) . While not all coronaviruses cause serious disease , many do , such as SARS and MERS , which cause severe respiratory infections ( de Wit et al . , 2016 ) . Both SPAV-1 and SPAV-2 were relatively widespread along the coast of southwestern British Columbia , in ocean caught Chinook and sockeye salmon . Currently , it is unclear what is driving differences in SPAV-1 and SPAV-2 prevalence among regions , but the virus appears to be transmitted to juvenile salmon throughout southern BC soon after they enter the ocean , a period known to be critical to their survival ( Beamish et al . , 2012a ) . SPAV-1 was also relatively common in farmed Chinook populations . The distribution of SPAV-1 in wild Chinook populations was more localised to the west coast of Vancouver Island than SPAV-2 , which was most prevalent on the east coast of Vancouver Island , near the Discovery Islands and the Johnstone Strait , and was rarely detected in sockeye salmon in northern BC and Alaska ( Figure 2—figure supplement 2 ) . On the east coast of Vancouver Island , the Johnstone Strait and Discovery Islands have been identified as a potential choke point for the growth and survival of juvenile salmonids ( Healy et al . , 2017 ) . The availability of prey to juvenile sockeye in the northern Johnstone Strait is extremely low , resulting in food limitation and increased competition for prey ( Beamish et al . , 2012a; McKinnell et al . , 2014; Godwin et al . , 2015; Godwin et al . , 2018 ) . These regions of high SPAV-2 infection could represent a stressful part of juvenile sockeye outmigration , possibly resulting in higher susceptibility to infection . Moreover , SPAV-2 was detected at high loads in fish sampled from regions where finfish aquaculture facilities are abundant and accordingly , sea lice infestation is high ( Price et al . , 2011] . It remains an open question whether an alternative host could play a role in virus transmission between fish , and/or result in an increased susceptibility to infection ( Valdes-Donoso et al . , 2013 ) . The distribution of CAV was markedly different from SPAV . CAV was not detected in any juvenile wild or hatchery Chinook salmon , despite being detected in farmed fish on both the west and east coasts of Vancouver Island . Over 20% of moribund Chinook aquaculture fish tested positive for CAV , with most detections occurring in fish at least 1 . 5 years after ocean entry , well past the time when migratory salmon were sampled . Hence , infection by CAV may take a considerable time to develop , or be an infection that is only acquired by older fish . CAV was also detected in a small number of farmed Atlantic salmon ( seven positive detections of 2816 fish tested ) . The monophyletic grouping of CAV with other disease causing aquareoviruses and the consistency with haemorrhagic disease suggest that the virus is important to monitor in cultured fish , and potentially wild adults returning after several years at sea . PsNV distribution was strongly associated with a handful of salmon-enhancement hatcheries but was also detected in 18% of aquaculture Chinook and 3% of wild Chinook ( Figure 2—figure supplement 1 ) . In hatchery fish , infection by PsNV was primarily localised to gill tissue ( Figure 4A ) , reminiscent of the respiratory disease caused by the related mammalian coronaviruses such as MERS and SARS ( Figure 1C ) . PsNV is of particular concern as it proliferates while fish are undergoing smoltification , a process during which the gill tissue undergoes cellular reconfiguration to prepare for saltwater . Notably , branchial proliferation of no known cause was noted in some farmed salmon infected with PsNV . In one of the hatcheries , where pre- and post-release sampling took place , the virus increased in prevalence during smolt development in fresh water , was detected shortly post-release , and was barely detected in the month following ocean entry ( Figure 4B ) . This suggests that infected fish either cleared the infection , or did not survive after entry into the marine environment . The second interpretation is consistent with the lower rates of ocean survival in fish produced from hatcheries versus wild salmon ( Beamish et al . , 2012b ) . Viral disease is a potential threat to wild fish stocks; yet little is known about viruses circulating in wild , farmed , or hatchery salmon . Here , through metatranscriptomic surveys , we reveal several previously unknown viruses that were discovered in dead and dying aquaculture fish , and show them to also occur in wild and hatchery-reared fish . Depending on the viral and host species , the viruses range from being localised to widespread , from infecting <1% to >20% of fish , and being from within the limits of detection to very high loads . Our results are consistent with some of these viruses being causative agents of disease , making it critical to understand their possible roles in salmon mortality and the decline of wild salmon populations , and their potential interactions with net-pen fish farming and hatchery rearing . Viral discovery in moribund individuals followed by extensive surveillance and histopathological localisation are powerful tools towards the ultimate goals of identifying causative agents of disease and understanding the impact of infectious agents in wild populations . These insights are crucial as juvenile salmon that are in less than optimal health are expected to have lower rates of survival in the wild . Continued surveillance and knowledge of endemic and emerging virus infections in these iconic salmon species is beneficial for their conservation . Samples were provided by the Fisheries and Oceans , Canada Aquaculture Management Division and Salmon Enhancement Program . Additional samples were collected by the Hakai Institute Juvenile Salmon Program . Hatchery samples are identified by fin clipping , and in this study , wild fish could also encompass unmarked hatchery fish . DNA is extracted for detection of DNA viruses , bacteria and parasites from the same tissues from which we extract RNA to target RNA viruses . Nucleic acid extractions on the audit samples ( eight tissues-gill , atrium , ventricle , liver , pyloric caeca , spleen , head kidney and posterior kidney ) were as previously described ( Laurin et al . , 2019 ) . For the wild Chinook and sockeye samples , homogenization using Tri-reagent was performed in a Mixer Mill ( Qiagen , Maryland ) on each tissue independently ( five tissues- gill , liver , heart , head kidney and brain ) . Tri-reagent homogenates were organically separated using bromochloropropane , with the RNA-containing aqueous layer removed for RNA extraction and the lower DNA-containing organic layer separated from the organics using a TNES-Urea Buffer ( Asahida et al . , 1996 ) . For the DNA extractions , a pool of 250 μl ( 5 tissues contributing 50 μl each ) from each of the tissue TNES aqueous layers was processed for DNA using the BioSprint 96 DNA Blood kit ( Qiagen , Maryland ) and the BioSprint 96 instrument ( Qiagen , Maryland ) both based on manufacturer’s instructions . DNA was quantified using spectrophotometer readings performed on the Infinite M200Pro spectrophotometer ( Tecan Group Ltd . , Switzerland ) and normalised to 62 . 5 ng/μl using the Freedom Evo ( Tecan Group Ltd . , Switzerland ) liquid handling unit , based on manufacturer’s instructions . Similarly , a pool of 100 μl ( 5 tissues contributing 20 ul each ) of the aqueous layer was processed for RNA using the Magmax−96 for Microarrays RNA kit ( Ambion Inc , Austin , TX , USA ) with a Biomek NXP ( Beckman-Coulter , Mississauga , ON , Canada ) automated liquid-handling instrument , both based on manufacturer’s instructions . The quantity of RNA was analysed using spectrophotometer readings and normalised to 62 . 5 ng/μl with a Biomek NXP ( Beckman-Coulter , Mississauga , ON , Canada ) automated liquid-handling instrument , based on manufacturer’s instructions . Mixed tissue RNA ( 1 μg ) was reverse transcribed into cDNA using the superscript VILO master mix kit ( Invitrogen , Carlsbad , CA ) , following the manufacturer's instructions . We applied a panel of host biomarkers ( genes ) that when co-expressed are indicative of a viral disease state ( VDD ) ( Miller et al . , 2017 ) . Samples that displayed a positive viral disease state , but were not positive for viruses based on our 45 microbe panel screening , ( as described in Bass et al . , 2019 ) , were selected for high throughput sequencing of RNA ( dual RNA-seq ) to discover new viral agents . Total RNA from the mixed tissue samples was evaluated for quality using the Total RNA Pico chip on the Agilent 2100 Bioanalyzer ( Agilent , Santa Clara , CA ) and quantified using the Qubit RNA Br kit ( Invitrogen , Carlsbad , CA ) . A 1/100 dilution of the ERCC RNA Spike-In control mix 1 ( Ambion , Carlsbad , CA ) was added to each total RNA sample prior to ribosomal depletion and library preparation . The sequencing libraries and ribosomal removal were performed using the Epicentre ScriptSeq Complete Gold Kit ( Epidemiology ) ( Illumina , San Diego , CA ) according to manufacturer’s instructions and included a positive control ( Universal Human Reference RNA ) ( Agilent , Santa Clara , CA ) and negative control ( no total RNA ) . The rRNA depleted total RNA was purified using the Zymo RNA Clean and Concentrate-5 kit ( Zymo Research , Irvine , CA ) according to manufacturer’s instructions and quantified using the Qubit RNA HS kit ( Invitrogen , Carlsbad , CA ) . The ScriptSeq Index reverse primers were added to the cDNA during the final amplification step which involved 14 cycles . The 3’-terminal tagged cDNA and final amplified library were purified using the Agencourt AMPure XP system ( Beckman Coulter , Brea , CA ) . The final library size was determined using the HS DNA chip on the Agilent 2100 Bioanalyzer ( Agilent , Santa Clara , CA ) and the concentration was determined using the Qubit dsDNA HS kit ( Invitrogen , Carlsbad , CA ) . Sample libraries were normalised to 4 nM , pooled appropriately and denatured and diluted to obtain a final library of 17pM . Prior to loading into a v3 2 × 300 bp kit ( Illumina , San Diego , CA ) , 2% phiX was spiked in . Finally , a paired-end 251 bp sequencing run was performed on the Illumina MiSeq System ( Illumina , San Diego , CA ) , with four samples barcoded and pooled for each run . To sequence SPAV-2 , PsNV and CAV , the samples were prepared using the same method as above but sequenced by BC Cancer Agency using a HiSeq ( 2 × 125 ) protocol ( four different samples indexed over one lane ) . The quality of the raw reads was checked using FASTQC ( v0 . 11 . 7 ) ( https://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . Low quality reads or regions of adapter sequences were removed using Trimmomatic ( v0 . 36 ) ( Bankevich et al . , 2012 ) . Reads were aligned to the Atlantic Salmon genome using bwa mem ( v0 . 7 . 17-r1188 ) and unmapped reads were retained . The unmapped reads were then balanced using Trimmomatic and assembled into contigs using SPAdes ( v3 . 9 . 1 ) genome assembler ( Bankevich et al . , 2012 ) . Putative viral contigs were identified by aligning translated contigs using DIAMOND ( v0 . 9 . 16 . 117 ) ( Buchfink et al . , 2015 ) to the nr database . Reference alignments of all the reads to the viral contigs were used to ensure that no assembly artefacts occurred and the contigs were trimmed appropriately using Geneious ( V10 . 1 . 3 ) . Assembled sequences are available on Genbank ( BioProject: PRJNA547678 , Genbank accession numbers: MK611979 - MK611996 ) and raw sequencing reads have been uploaded to the Sequence Read Archive ( SAMN11974798 - SAMN11974801 ) . The phylogeny of each virus was resolved based on the predicted replicase ( CAV and SPAV ) and ORF1ab ( PsNV ) amino acid sequences , as nucleotide sequences were too dissimilar to reliably align . Alignments were generated with MAFTT ( v7 . 42 ) ( Katoh and Standley , 2013 ) employing the E-INS-i algorithm . This alignment algorithm is suited for evolutionarily distinct sequences with conserved motifs ( such as viral RNA polymerase ) that are embedded within long unalignable residues . The novel salmon viruses were aligned with other viral genomes with shared amino acid similarity as detected by DIAMOND ( Buchfink et al . , 2015 ) . In addition , viral genomes which are known to be evolutionarily related to these were included . The multiple protein alignments were then used as to build phylogenies using PhyML 3 . 0 ( Guindon et al . , 2010 ) plugin within Geneious with 100 bootstraps to generate branch support values . Trees are mid-point rooted for clarity only , and do not necessarily represent the ancestral relationship of the viruses . Assembled viral sequence contigs from the appropriate sample were imported into Primer Express v3 . 0 . 1 software ( Thermo Fisher Scientific , Waltham , MA ) where qPCR Taqman assays were designed using default parameters ( Supplementary file 1 ) . These assays were then tested using the Fluidigm BioMark microfluidics-based qPCR system following the same protocol as described below except with the new viral primer pairs included in the STA step and controls . From these initial screens , the most consistent assay was chosen and APC standards were constructed to include in future Fluidigm BioMark qPCR microbe panels . The assay-specific theoretical limit of detection was calculated as previously described ( Miller et al . , 2016 ) . The limit of detection was applied to categorise fish with amplifications above the 95% detection threshold that is the concentration of the analyte in the sample matrix that would be detected with high statistical certainty ( 95% of the time ) . Epidemiological maps were generated using these data with the limit of detection applied . The maps were created within R using ggplot2 ( Wickham , 2016 ) and ggmap ( Kahle and Wickham , 2013 ) . For all samples , after reverse transcription , resultant cDNA was combined with the normalised DNA in a ratio of 1:1 and used as the template for the specific target amplification ( STA ) step . The STA involves a pre-amplification of all primers to be run on a single dynamic array at low concentrations ( 0 . 2 μM of each of the primers ) , and upon completion , excess primers were removed by treating with Exo-SAP-IT ( Affymetrix , Santa Clara , CA ) according to manufacturer’s instructions and then diluted 1:5 in DNA re- suspension buffer ( Teknova , Hollister , CA ) . The 96 . 96 gene expression dynamic array ( Fluidigm Corporation , CA , US ) was run according to the procedure outlined previously ( Miller et al . , 2016 ) . Specifically , a 5 μl template mixture was prepared for each sample containing 1 × TaqMan Universal Master Mix ( No UNG ) , 1 × GE Sample Loading Reagent ( Fluidigm PN 85000746 ) and each of diluted STA’d sample mixtures . Five μl of Assay mix was prepared with 1 × each of the appropriate TaqMan qPCR assays ( agent probe in FAM-MGB and artificial positive construct ( APC ) probe in NED-MGB , 10 μM of primers and 3 μM of probes ) and 1 × Assay Loading Reagent ( Fluidigm PN 85000736 ) . Controls were added prior to running the dynamic array ( Miller et al . , 2016 ) . Note , APC clones to all assays were contained in a single serially diluted pool , loaded last , minimising the likelihood of contamination of any single APC clone . Once loading and mixing of the dynamic array was completed within the IFC HX controller , the array was transferred to the BioMark HD instrument and processed using the GE 96 × 96 Standard TaqMan program for qPCR which includes a hot start followed by 40 cycles at 95°C for 15 s and 60°C for 1 min ( Fluidigm Corporation , CA , USA ) . The data were analysed with Real-Time PCR Analysis Software ( Fluidigm Corporation , CA , USA ) . Chinook smolt samples positive for PsNV from 2014 were used for tissue localization ( Figure 4A ) . Gill , liver , heart , kidney , and brain were individually homogenized , processed for RNA extraction ( as described above ) , and 1 ug normalised RNA was used for reverse transcription . Resultant cDNA for each individual tissue was used as the template for PsNV relative quantification using an ABI 7900HT ( ABI ) in 384-well optical plates . The qPCR reaction volume was 12 μl , which comprised 6 μl of 2X TaqMan Gene Expression Master Mix ( ABI PN 4369016 ) , 4 . 3 μl of water , 0 . 22 μl of mixed forward and reverse primers ( 900 nM final concentration of each ) , 0 . 24 μl of each probe ( 200 nM final concentration; assay specific probe and APC control probe ) , and 1 μl of cDNA template . Temperature cycles included one 2 min hold ( 50°C ) , a 10 min denaturation ( 95°C ) , and 40 cycles of denaturation ( 95°C for 15 s ) , annealing and extension ( 60°C for 60 s ) . Amplification conditions on the ABI 7900 were not optimised for this platform , but rather closely reflected those used on the BioMark platform . Samples run on the ABI did not undergo STA enrichment . Standard curves were constructed using the same APC clone standards spiked in with CHSE DNA as on the BioMark . Serial dilutions were made to obtain concentrations of 24 , 1 . 2 × 102 , 6 × 102 , 3 × 103 , 1 . 5 × 104 , 1 . 5 × 105 copies of the clone per reaction . Clone standards , unknown samples , positive and negative controls were all run in duplicate . The ABI software calculates the relative copy number based upon the serial dilution of the standard curve . Before the discovery of these viruses clinical signs of disease and histopathological lesions were assessed for approximately 230 farmed Chinook salmon sampled in the Audit program . Consequently , gills , skeletal muscle , spleen , liver , heart , anterior and posterior kidney , pyloric caeca and brain from eleven samples of Chinook ( eight wild fish and three farmed fish ) and ten sockeye ( all wild fish ) positive for SPAV were histopathologically analysed to assess the presence of lesions . All tissues were fixed in 10% neutral buffered formalin , dehydrated through an ascending gradient of alcohol solutions , embedded in paraffin wax , cut at 3 . 5 μm thickness , and stained with routine hematoxylin and eosin ( H and E ) for morphological evaluation by light microscope . RNA-ISH was performed using RNAscope 2 . 5 HD Duplex assay ( Advanced Cell Diagnostics , Newark , California , USA , catalog# 322500 ) according to the manufacturer’s instructions . Briefly , consecutive sections of Chinook and sockeye salmon samples utilised for the histopathological analysis were dewaxed by incubating for 60 min at 60°C and endogenous peroxidases were quenched with hydrogen peroxide for 10 min at room temperature . Slides were then boiled for 30 min in RNAscope target retrieval reagents ( Advanced Cell Diagnostics , Newark , California , USA ) and incubated for 30 min in RNAscope Protease Plus reagent prior to hybridization . The slides underwent hybridization with RNAscope probes against a portion of SPAV-1 and SPAV-2 genome ( Advanced Cell Diagnostics , Newark , California , USA , catalog #513591-C2 and 538881-C2 , respectively ) . A RNAscope probe against Coil-p84 housekeeping gene in Chinook salmon ( Advanced Cell Diagnostics , Newark , California , USA , catalog #512391 ) was used as positive control probe to confirm the efficacy of the probes and the viability of the samples . Two samples which were negative for SPAV-1 and SPAV-2 were used as negative controls to confirm absence of background and ( or ) non-specific cross-reactivity of the assay . Signal amplification was performed according to the manufacturer’s instructions , followed by counterstaining with Gill’s hematoxylin and visualisation by bright field microscopy .	Keystone species are animals and plants that play a pivotal role in supporting the ecosystems they live in , making their conservation a high priority . Chinook and sockeye salmon are two such species . These fish play a central role in the coastal ecosystems of the Northeast Pacific , where they have supported Indigenous populations for thousands of years . The last three decades have seen large declines in populations of Chinook and sockeye salmon . One factor that may be involved in these declines is viral infection . In the last ten years , advances in DNA sequencing technologies have led to the discovery of many new viruses , and Mordecai et al . used these technologies to look for new viruses in Pacific salmon . First , Mordecai et al . looked for viruses in dead and dying salmon from farms and discovered three previously unknown viruses . Next , they screened for these viruses in farmed salmon , hatchery salmon and wild salmon to determine their distribution . Two of the viruses were present in fish from the three sources , while one of the viruses was only found in farmed fish . The fact that the three viruses are distributed differently raises questions about how the viruses are transmitted within and between farmed , hatchery and wild salmon populations . These findings will aid salmon-conservation efforts by informing the extent to which these viruses are present in wild salmon populations . Future work will focus on determining the risks these viruses pose to salmon health and investigating the potential for exchange between hatchery , farmed and wild salmon populations . While farmed Pacific salmon may pose some transmission risk to their wild counterparts , they also offer the opportunity to study disease processes that are not readily observable in wild salmon . In turn , such data can be used to develop policies to minimize the impact of these infectious agents and improve the survival of wild salmon populations .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "ecology", "short", "report", "microbiology", "and", "infectious", "disease" ]	2019	Endangered wild salmon infected by newly discovered viruses
The role of the cellular microenvironment in enabling metazoan tissue genesis remains obscure . Ctenophora has recently emerged as one of the earliest-branching extant animal phyla , providing a unique opportunity to explore the evolutionary role of the cellular microenvironment in tissue genesis . Here , we characterized the extracellular matrix ( ECM ) , with a focus on collagen IV and its variant , spongin short-chain collagens , of non-bilaterian animal phyla . We identified basement membrane ( BM ) and collagen IV in Ctenophora , and show that the structural and genomic features of collagen IV are homologous to those of non-bilaterian animal phyla and Bilateria . Yet , ctenophore features are more diverse and distinct , expressing up to twenty genes compared to six in vertebrates . Moreover , collagen IV is absent in unicellular sister-groups . Collectively , we conclude that collagen IV and its variant , spongin , are primordial components of the extracellular microenvironment , and as a component of BM , collagen IV enabled the assembly of a fundamental architectural unit for multicellular tissue genesis . A pivotal event in metazoan evolution was the transition from single-cell organisms to multicellular tissues ( Figure 1A ) . The cellular microenvironment is presumed to play an essential role in this transition , yet the mechanism remains obscure . The basement membrane ( BM ) , a specialized form of extracellular matrix ( ECM ) , is a hallmark morphological feature of the microenvironment of epithelial tissues , and its appearance within the non-bilaterian animal phyla suggests it was a prerequisite ( Sherwood , 2015; Hynes , 2012; Ozbek et al . , 2010 ) . The BM has numerous functions including maintaining tissue architecture and compartmentalization , organizing growth factor signaling gradients , guiding cell migration and adhesion , delineating apical-basal polarity modulating cell differentiation during development , orchestrating cell behavior in tissue repair after injury , and guiding organ regeneration ( Hynes , 2009; Yurchenco , 2011; Vracko , 1974; Pöschl et al . , 2004; Daley and Yamada , 2013; Wang et al . , 2008; Pastor-Pareja and Xu , 2011; Song and Ott , 2011 ) . 10 . 7554/eLife . 24176 . 003Figure 1 . Extracellular matrix of the non-bilaterian animal phyla . ( A ) The transition from single-cell organisms to complex multicellular animals was enabled by an extracellular matrix . ( B ) Electron microscopy ( EM ) and immunohistochemistry ( IHC ) of the Ctenophora species , Mnemiopsis ( IHC: 20X magnification ) , Pleurobrachia ( IHC: 20X magnification ) , and Beroe ( IHC: 40X magnification ) and ECM components of Ctenophora . ( C ) Electron microscopy ( EM ) and immunohistochemistry ( IHC ) of the non-bilaterian animal phyla , Cnidaria ( Nematostella; 20X magnification ) , Placozoa ( Trichoplax ) , and Porifera ( Homoscleromorpha and Demosponges ) and ECM components of Porifera , Placozoa , and Cnidaria . Demosponge EM reproduced from Figure 1E of Adams , et al . , Freshwater Sponges Have Functional , Sealing Epithelia with High Transepithelial Resistance and Negative Transepithelial Potential , PLoS ONE , 2010 , volume 5; Homoscleromorph EM reproduced from Figure 3B , Leys et al . , Epithelia and integration in sponges , Integrative and Comparative Biology , 2009 , volume 49 with permission from Oxford University Press; Homoscleromorph IHC reproduced from Boute et al . , Type IV collagen in sponges , the missing link in basement membrane ubiquity , Biology of the Cell , 1996 , volume 88 with permission from Wiley; Trichoplax EM reproduced from Ruthmann et al . , The ventral epithelium of Trichoplax adhaerens ( Placozoa ) : Cytoskeletal structures , cell contacts and endocytosis , Zoomorphology , 1986 , volume 106 with permission from Springer . ( D ) ECM components in choanoflagellates , the unicellular sister-group to metazoa . All scale bars 500 nm , unless otherwise noted . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 003 The basement membrane is a supramolecular scaffold , comprised of a toolkit of proteins including collagen IV , laminin , perlecan , and nidogen ( Hynes , 2012; Fahey and Degnan , 2010 ) . Among these proteins , recent studies reveal collagen IV is an ancient protein with up to six distinct genes ( COL4A1 , COL4A2 , COL4A3 , COL4A4 , COL4A5 , COL4A6 ) , essential for early development , that functions as a smart scaffold providing tensile strength to tissues , influencing cell behavior by tethering diverse macromolecules , including laminin , proteoglycans , growth factors , binding integrins ( Gupta et al . , 1997; Bhave et al . , 2012; McCall et al . , 2014; Fidler et al . , 2014; Pöschl et al . , 2004; Vanacore et al . , 2009; Cummings et al . , 2016; Wang et al . , 2008; Parkin et al . , 2011; Emsley et al . , 2000 ) . Disrupting collagen IV scaffolds causes BM destabilization and tissue dysfunction in mice , zebrafish , flies , and nematodes ( Pöschl et al . , 2004; Fidler et al . , 2014; Borchiellini et al . , 1996; Gupta et al . , 1997 ) . Collectively , these findings reveal that collagen IV , a component of the cellular microenvironment , is essential for tissue architecture and function; yet , the origin and molecular evolution of collagen IV remains obscure . Knowledge of collagen IV evolution may shed light on the fundamental features of the cellular microenvironment that enabled the transition from single-cell organisms to multicellular tissues . Together , the non-bilaterian animal phyla ( Ctenophora , Porifera , Placozoa , and Cnidaria ) represent this transition . Importantly , Ctenophora has recently emerged as one of the earliest-branching extant phyla ( Ryan et al . , 2013; Moroz et al . , 2014; Whelan et al . , 2015; Telford et al . , 2016 ) , along with the sponges ( Porifera ) ( Pisani et al . , 2015; Jékely et al . , 2015; Telford et al . , 2016 ) . Here , we sought to identify ECM components in Ctenophora along with the other non-bilaterian animal phyla , and compared the components to Bilateria and the metazoan sister-groups , Choanozoa , Filasterea , Amoebozoa , and Apusozoa . Our findings reveal that collagen IV and its truncated variant , spongin , are associated with the transition to multicellularity , and further that collagen IV , as a component of BM scaffolds , enabled the genesis of multicellular epithelial tissues . We characterized the extracellular matrix in Ctenophora ( comb jellies ) and the other non-bilaterian animal phyla through a combination of immunohistochemistry ( IHC ) , electron microscopy ( EM ) , RNA sequencing , and genomic and transcriptomic analyses . Three ctenophore species , Mnemiopsis leidyi , Beroe ovata , and Pleurobrachia pileus , were used for EM and IHC experiments . Systematic assessment by EM of a number of sections from similar areas in Mnemiopsis was conducted , and no organized basement membrane was encountered . Furthermore , tight junctions between cells and cellular polarization were not observed , both hallmarks of epithelial basement membrane tissue structure ( Figure 1B ) . IHC was congruent with this finding , indicating that collagen IV was dispersed throughout the tissue , surrounding and encompassing cells . In Beroe and Pleurobrachia , however , EM indicated an electron dense layer underlying cells along with cell polarization , and lateral tight junctions between cells , and IHC similarly showed a dense collagen IV layer underlying cells ( Figure 1B ) . Together , these features are congruent with basement membrane architecture and epithelial tissue . Basement membrane structures are prevalent throughout metazoa , from Cnidaria to vertebrates , and we sought to compare the basement membrane architecture in Ctenophora to that of other non-bilaterian animal phyla and Bilateria . Nematostella , along with other cnidarians , have a bilayer body structure composed of endoderm and ectoderm layer with an intervening mesoglea; however , the general BM structure is congruent with that of bilaterian organisms , including mammals . Nematostella demonstrated presence of basement membrane , characterized by polarized cells apical to an electron dense layer by EM , and a concentrated region of collagen IV underlying cell nuclei by IHC ( Figure 1C ) . We then characterized the ECM composition through analysis of transcriptomic and genomic data across the non-bilaterian animal phyla in comparison with Bilateria and unicellular sister-groups . Ctenophore genomic and transcriptomic data were publicly available from the Pleurobrachia Genome Browser on Neurobase ( http://neurobase . rc . ufl . edu/Pleurobrachia ) and the Mnemiopsis Genome Project Portal on the National Human Genome Research Institute site ( https://kona . nhgri . nih . gov/mnemiopsis/ ) . The ECM components in Nematostella are very similar to that of most bilaterian species BMs ( Hynes , 2012 ) , including human , mouse , zebrafish , Drosophila , and C . elegans , consisting of collagen IV , laminin , peroxidasin , collagen XV and XVIII , perlecan , nidogen , fibronectin , as well as spongin ( Figure 1C and Figure 2 ) . Ctenophora , however , revealed a simplified set of ECM proteins , with collagen IV and laminin as the only components identified across Beroe , Pleurobrachia , and Mnemiopsis ( Figure 1B and Figure 2 ) . Importantly , despite lacking the full gamut of bilaterian ECM proteins , ctenophore cells can still construct a prototypical basement membrane in Pleurobrachia and Beroe . Mnemiopsis , however , does not form a BM despite exhibiting the very same toolkit proteins and this may be a result of secondary loss event . 10 . 7554/eLife . 24176 . 004Figure 2 . Extracellular matrix gene content across bilaterian , non-bilaterian animal , and unicellular protist phyla . Protein BLAST searches using the human ortholog of each protein as bait was conducted for ECM gene content analysis . Where possible ( with exception of ctenophore species ) , we performed a search by protein name across each database . The databases used were Ensembl ( http://protists . ensembl . org ) , NeuroBase ( http://neurobase . rc . ufl . edu ) , AmoebaDB ( http://amoebadb . org ) and NCBI’s Blast ( https://blast . ncbi . nlm . nih . gov/Blast . cgi ) . Complete hits are denoted in green , while partial protein or domain sequences are denoted in orange . White boxes indicate absence of that protein/domain . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 004 Across Porifera and Placozoa , basement membranes are uncommon . Placozoa and the poriferan classes of calcareous and demosponges lack basement membranes ( Figure 1C ) ( Ozbek et al . , 2010; Leys et al . , 2009; Ruthmann et al . , 1986; Srivastava et al . , 2010 ) , suggesting the BM may have been secondarily lost in these lineages ( Cock , 2010 ) or that it is present only at specific stages during their life cycle ( Hynes , 2012 ) . Alternatively , basement membranes may have independently evolved in Ctenophora , Porifera , and Bilateria , a phenomenon that could have occurred because of shared inheritance of ECM proteins and domains from the last common ancestor of the non-bilaterian animal phyla . Investigation of the non-bilaterian animal phyla for components of the BM toolkit suggests that many of the components are present ( Srivastava et al . , 2010 , 2008 ) . Specifically , Placozoa contains the necessary components for a BM , including collagen IV , laminin , perlecan , and nidogen ( Srivastava et al . , 2008 ) and also exhibits spongin ( Figure 1C and Figure 2 ) ( vide infra ) . In Porifera , the ECM of homoscleromorph sponges contains basement membranes ( Boute et al . , 1996 ) , with collagen IV and laminin , but no spongin ( Figure 1C and Figure 2 ) . Demosponges , on the other hand , lack any detectable collagen IV and only contain laminin-related domains and spongin . In contrast , the Demosponge and Hexactinellidae classes of Porifera , which both lack collagen IV , do not have basement membranes ( Figure 1C ) ( Adams et al . , 2010 ) . Laminin architecture appears to be present in choanoflagellates ( Fahey and Degnan , 2012 ) , and additionally laminin-related genes appear in all other unicellular species including Capsaspora owczarzaki , Dictyostelium discoideum , and Thecamonas trahens ( Figure 2 ) . Although no other complete ECM components exist in unicellular choanoflagellates ( King , 2005 ) , several domains and fragments of ECM components have been identified , including collagenous repeats , laminin G ( globular ) domains and LN ( N-terminal ) domains , and fibronectin type II and III domains ( Figure 1D ) ( King et al . , 2008 ) . Furthermore , no other complete ECM components exist in other unicellular organisms , including Salpingoeca rosetta ( Choanozoa ) , Capsaspora owczarzaki ( Filasterea ) , Dictyostelium discoideum ( Amoebozoa ) , or Thecamonas trahens ( Apusozoa ) ; however , collagenous GXY repeats were identified in Monosiga , Salpingoeca , and Dictyostelium . Receptors for collagen binding , including integrin , dystrophin , and Discoidin domain receptors ( DDRs ) were analyzed across metazoa and unicellular protists ( Figure 2 ) . Integrins are highly conserved across metazoa and are found in the unicellular Capsaspora and Thecamonas . Dystrophin was identified in humans , fruit fly , Nematostella , Amphimedon , as well as in unicellular protists but is absent in Trichoplax , Oscarella , and ctenophores . DDR 1 and 2 was identified only in humans and fruit fly . Collectively , the ECM composition across the non-bilaterian animal phyla points to laminin and collagen IV as highly conserved components , and importantly , that they are associated with BM and epithelial tissue architecture . We characterized the gene and protein structure of collagen IV in Ctenophora and the other non-bilaterian animal phyla . In Mnemiopsis , 11 full-length collagen IV genes were discovered , which contrasts with the two genes typically present throughout invertebrates and the six genes typically found in vertebrates ( Khoshnoodi et al . , 2008 ) . The head-to-head orientation is a distinguishing characteristic of collagen IV genes among other collagens of Bilateria ( Kaytes et al . , 1988; Hudson et al . , 1993 ) . In Mnemiopsis , four genes occur on the same scaffold and exhibit a head-to-head orientation ( ML17501a and ML17504a , ML17502a and ML17503a ) ( designated as Group I ) ( Figure 3A ) . Furthermore , ML17502a and ML17503a genes are oriented in opposite directions , separated by ~2879 bases , and do not share the same promoter . The other seven Mnemiopsis genes: ML166441a , ML18175a , ML18197a , ML18198a , ML034334a , ML0343336a , and ML0343337a ( designated as Group II ) , are aligned individually on separate scaffolds or in a unidirectional tandem array . Transcriptomic analysis of adult Mnemiopsis reveals differential expression within each group of collagen IV genes . In Group I , the expression of ML17501a and ML17502a is significantly higher than other Group I genes . Similarly , ML034334a , ML0343336a , and ML0343337a show considerable increase in expression levels compared to the other Group II genes ( Figure 3B ) . 10 . 7554/eLife . 24176 . 005Figure 3 . Mnemiopsis reveals multiple duplications of collagen IV genes , and non-bilaterian animal phyla collagen IV organization is similar to Bilateria . ( A ) Collagen IV Gene Orientation . Mnemiopsis collagen IV genes were separated into two groups base on genomic orientation . Group I genes are found on the same scaffold ( colored in blue ) . Group II genes ( colored in orange ) are spread across four different scaffolds and do not have head-to-head orientation . ( B ) Transcriptome analysis of Mnemiopsis confirmed a total of 11 collagen IV genes and one NC1 proto-domain gene ( colored in grey ) . ( C ) Human COL4A1 spans 150 kb , contains 52 exons and has an intron composition of 95% . Mnemiopsis collagen IV genes are approximately one sixth the length of human collagen IV genes ranging in length from 1 to 23 kb with an intronic composition of 50–82% . ( D ) In Nematostella and Trichoplax , two collagen IV genes are located in head-to-head orientation on one genomic scaffold ( Nematostella , scaffold 14; Trichoplax , scaffold 235 ) , which indicates that they share one chromosome . The arrows at the top of each species indicate gene orientation: either minus or plus strands . The search for conserved domains revealed multiple collagens repeats ( PF01391 , light blue boxes ) and C4 domains ( PF01413 , dark blue ) further support that these genes belongs to collagen IV gene family . Pfam domains were identified using HMM against the genomic sequence . Mapping RNAseq reads to the genome strongly supports the proposed collagen IV genes model . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 00510 . 7554/eLife . 24176 . 006Figure 3—source data 1 . Mnemiopsis collagen IV gene expression by RNA-Seq ( RPKM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 00610 . 7554/eLife . 24176 . 007Figure 3—figure supplement 1 . In an effort to detect conservation of exon size between ML collagen IV genes a database composed of all the exons from each gene was compiled . Calculation of exon size frequency revealed a total of 75 different exon lengths ( ranging from 57 to 837 bp ) , which were repeated at least once . Only six of these exon sizes were repeated more than four times throughout the database ( upper graph ) . Conservation of exon length and position ( marked in red ) is present among some of the group II genes ( lower table ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 00710 . 7554/eLife . 24176 . 008Figure 3—figure supplement 1—source data 1 . Frequency of >exon lengths in Mnemiopsis collagen IV genes . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 00810 . 7554/eLife . 24176 . 009Figure 3—figure supplement 1—source data 2 . Conservation of exon length and position is present among some of the groups . Conserved exon lengths and positions are marked in red . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 00910 . 7554/eLife . 24176 . 010Figure 3—figure supplement 2 . The first NC1 domain ( red ) coding exon of collagen IV has several features conserved throughout the animal kingdom , including the collagenous domain ( yellow ) , which is a stretch of interrupted Gly-X-Y repeats at the 5’ end , and the presence of an HSQ coding region ( white text in red ) . In addition to these conserved features , Mnemiopsis Group II collagen IV genes encode for a cysteine loop region ( green ) on the collagenous domain/ NC1 domain junction exon . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 010 Mnemiopsis collagen IV genes are shorter in sequence length and typically have fewer exons and shorter intronic regions compared to their Bilaterian counterparts ( Figure 3C ) . Whereas bilaterian fibrillar collagen genes are composed of multiple exons with a length of 54 base pairs ( Yamada et al . , 1980 ) , exons of Mnemiopsis collagen IV genes range in size from 40 to 4867 bp . Among the Mnemiopsis genes , exons of 108 bp in length were found in several genes , including ML17503a , ML17504a , and ML16441a , but no increased frequency of 54 bp exons in length or any variation was detected ) ( Figure 3—figure supplement 1 ) . The presence of split glycine codons coding for the collagenous domain and codons on junctional exons ( e . g . collagenous domain/NC1 domain encoding exons ) are defining features of vertebrate collagen IV genes ( Quinones et al . , 1992 ) . Multiple genes from each group ( Group I: ML17501a , ML17502a , and ML17503a; Group II: ML034334a , ML034336a , and ML034337a ) possess split glycine codons on the collagenous domain/NC1 domain junction exon ( Figure 3—figure supplement 2 and Supplementary file 1 ) . We then determined the number of collagen IV genes in 10 other species from the Ctenophora phylum . Transcriptome analysis was conducted using in-house generated libraries for Mnemiopsis leidyi and Pleurobrachia pileus and publicly available libraries for Pleurobrachia bachei , Beroe ovata , Beroe abyssicola , Euplokamis sp . , Dryodora sp . , Vallicula sp . , Coeloplana sp . , and Bolinopsis sp . Across these ten species , a total of 118 unique collagen IV genes were detected . Each species contained a variable number of collagen IV genes , ranging between 4 and 20 genes each , as compared to the 2 to 6 genes in the other non-bilaterian animal phyla and Bilateria ( Figure 4A and B ) . All species , apart from the two Beroe species surveyed , contain a combination of both Group I and II chains , with the two Beroe species exhibiting only Group II chains . In addition to full-length collagen IV genes , two genes , with signal peptides , encoding only the NC1 domain were identified across ten species of Ctenophora . Expression of NC1 domains without collagenous tails is novel ( Figure 4C ) . Together , these findings show that the ECM of ctenophores contain both collagen IV and standalone NC1 genes , and that the number and diversity collagen IV genes exceeds that of any other metazoan group . 10 . 7554/eLife . 24176 . 011Figure 4 . Collagen IV in Ctenophora underwent numerous gene duplication events resulting in an unprecedented diversity . ( A ) Collagen IV chain distribution across non-bilaterian animal phyla and Bilateria . Two collagen IV chains are found across invertebrates , and six chains in chordate/vertebrate lineages . The poriferan class of Demosponges lacks collagen IV and BM . ( B ) Ctenophora collagen IV chains range from four to twenty distinct chains across species , indicating a variable number of gene duplication events . Ctenophora chains can be split into Group I , Group II , and NC1/C4 subgroupings . All ctenophore species contain Group I , II , and NC1 genes except for the two Beroe species , which lack Group I chains . ( C ) NC1 genes identified across Ctenophora were analyzed for signal peptide presence to determine whether sequences were truncated , or represented standalone NC1 proteins . Putative signal peptides were detected in at least three ctenophore NC1 genes , Mnemiopsis ( ml047918a ) , Pleurobrachia pileus ( pp_COL4_i ) , and Pleurobrachia bachei ( PBNC1_1 ) based on SignalP prediction ( http://www . cbs . dtu . dk/services/SignalP/ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 011 We also determined the number , structure , and orientation of collagen IV genes in Nematostella vectensis ( Cnidaria ) and Trichoplax adhaerens ( Placozoa ) . We found two genes in both species , and that they are homologous with Bilateria , with each demonstrating a ‘head-to-head’ orientation and homologous coding regions for both collagenous and non-collagenous domains ( Figure 3D ) . Complete genomic data was unavailable to determine the orientation of the two collagen IV genes in homoscleromorph sponges . In contrast , unicellular protists ( Choanozoa , Filasterea , Amoebozoa , Apusozoa ) do not contain collagen IV as determined by genomic analyses ( Figure 2 ) . Together , our findings show that head-to-head orientation of collagen IV is conserved across Bilateria , Cnidaria , Placozoa , and Ctenophora , whereas Ctenophora exhibits both head-to-head and tandem orientations ( Figure 3A ) . Several prominent structural domains characterize Bilaterian collagen IV chains ( Figure 5 ) . These include an N-terminal non-collagenous domain rich in cysteine and lysine residues ( NC3 ) ( Figure 5—figure supplement 1 ) , a large collagenous domain of Gly-Xaa-Yaa ( GXY ) repeats of ~1400 residues with interruptions in the GXY repeats ( Figure 6 ) , followed by a non-collagenous ( NC1 ) domain at the C-terminus of approximately ~230 residues ( Figure 5—figure supplement 2 ) . NC1 domains are comprised of two C4 domains , each containing a short , highly conserved HSQ residue motif proximal to the N-terminal side . We sought to determine whether these structural domains are characteristic of Ctenophora and the other non-bilaterian animal phyla . Indeed , these domains are conserved across Cnidaria , Placozoa , Porifera , and Ctenophora ( Figure 5 ) . The NC1 domain also contains a chloride-binding motif , which functions in binding extracellular chloride to signal the assembly of collagen IV networks and is conserved from vertebrates to Cnidaria and Placozoa ( Cummings et al . , 2016 ) . The chloride-binding motif was identified in both Ctenophora ( group II chains ) as well as in the two Homoscleromorph sponges analyzed , suggesting the chloride signaling function of NC1 domains is also conserved in Ctenophora and Porifera ( Figure 5—figure supplement 3 ) . Together , these findings reveal that the conserved structural features of bilaterian collagen IV extend across the non-bilaterian animal phyla , including Ctenophora . 10 . 7554/eLife . 24176 . 012Figure 5 . Collagen IV structural features are conserved across metazoa , and Ctenophora exhibits novel domains . Signature features of collagen IV are found in each of the identified in Mnemiopsis chains . The collagenous region ( yellow ) of each chain contains characteristic interruptions ( black lines ) of the Gly-X-Y motif repeats . Group II chains also possess a NC2 domain ( blue ) , which interrupts the collagenous region , and a cysteine loop ( green ) that is an extension of canonical NC1 domain ( red ) . The NC1 domain of each chain is composed of two C4 domains . Group II chains possess the chloride-binding motif ( purple ) within the NC1 domain . While conservation of most Mnemiopsis collagen IV features can be found throughout metazoan species the NC2 domain and cysteine loop are structural innovations restricted to Ctenophora . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 01210 . 7554/eLife . 24176 . 013Figure 5—figure supplement 1 . Multiple sequence alignment of the NC3 domain from various metazoan species reveals a high degree of conservation in this region among Mnemiopsis sequences , which is not seen in other metazoan sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 01310 . 7554/eLife . 24176 . 014Figure 5—figure supplement 2 . Multiple-sequence alignment of collagen IV NC1 domain sequences across human , mouse , zebrafish , fly , C . elegans , Nematostella , and Trichoplax , compared to the ctenophore representative , Mnemiopsis leidyi ( MLXXXXXX ) . Several hallmarks of collagen IV domains are present in Mnemiopsis: conservation of 12 cysteine residues throughout the NC1 ( highlighted in yellow ) , and conservation of an HSQ- motif at the N-terminal side of both C4 domains comprising the whole NC1 domain ( highlighted in red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 01410 . 7554/eLife . 24176 . 015Figure 5—figure supplement 3 . The chloride-motif has been identified previously in humans to Trichoplax . The chloride-motif is highly conserved in Mnemiopsis Group II chains but is absent in Group I chains , and the standalone NC1 gene . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 01510 . 7554/eLife . 24176 . 016Figure 5—figure supplement 4 . Sulfilimine bond crosslinking of collagen IV occurs between Methionine-93 and Lysine/Hydroxylysine-211 residues between adjoining NC1 domain interfaces . Sulfilimine crosslinking confers structural integrity to collagen IV networks . Sulfilimine bond crosslinking residues are conserved throughout Bilateria to Cnidaria . Ctenophora has no conservation of Met-93 and Lys/Hyl-211 residues , indicating lack of sulfilimine bond crosslinking within the phylum . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 01610 . 7554/eLife . 24176 . 017Figure 5—figure supplement 5 . Multiple sequence alignment of 32 partial ctenophore collagen IV sequences spanning 10 species reveals the NC2 domain ( highlighted in blue ) , spanning 38–44 amino acids . However , very little sequence homology was observed across species for this region . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 01710 . 7554/eLife . 24176 . 018Figure 5—figure supplement 6 . Multiple sequence alignment of partial sequences from 41 collagen IV genes exhibiting the cysteine-loop region of NC1 domains in Group II chains across Ctenophora . Cysteine-loop domains across species contain either three or four conserved cysteine residues . HSQ motif and preceding seven residues demarcate the classical start of the NC1 domain . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 01810 . 7554/eLife . 24176 . 019Figure 6 . Collagen IV in Ctenophora and the non-bilaterian animal phyla are structurally homologous to bilateria . Representative collagen IV sequences from human ( UniProt entries P02462 , P08572 , Q01955 , P53420 , P29400 , Q14031 ) , Drosophila ( UniProt entries P08120 , O18407 ) , C . elegans ( UniProt entries P17139 , P17140 ) , Nematostella , Trichoplax , and Mnemiopsis genomes . The following regions are depicted: ( based on prediction from http://www . cbs . dtu . dk/services/SignalP/; shown as orange arrow ) , NC3 domain ( black box ) , uninterrupted triple helical segments with at least three GXY repeats ( yellow boxes ) , NC2 domain ( blue box ) , cysteine loop ( green box ) , C4 domains ( based on conserved domain search at http://www . ncbi . nlm . nih . gov/Structure/cdd/wrpsb . cgi; shown as red boxes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 019 The NC1 domain is the molecular recognition module that directs the assembly of collagen IV protomers and networks ( Cummings et al . , 2016 ) . NC1 modules function in selecting collagen IV chains for trimerization , forming triple-helical protomers , and for oligomerization of protomers into networks ( Figure 5 ) ( Cummings et al . , 2016; Khoshnoodi et al . , 2008 ) . The NC1 modules are stabilized by sulfilimine cross-links , which connect methionine-93 ( Met-93 ) and lysine/hydroxylysine-211 ( Lys/Hyl-211 ) between adjoining protomers ( Fidler et al . , 2014; Vanacore et al . , 2009 ) . Sulfilimine cross-links are conserved throughout Bilateria , from Humans to C . elegans , and in Cnidaria , with the exception of Hydra ( Fidler et al . , 2014 ) . In contrast , this cross-link is absent in Ctenophora , owing to the absence of Met-93 and Lys/Hyl-211 residues ( Figure 5—figure supplement 4 ) . Thus , we sought to characterize the biochemical properties of ctenophore NC1 domains to ascertain whether they are stabilized by an alternative cross-linking mechanism . Uniquely , ctenophore collagen IV is distinguished from all other metazoans by the presence of two additional domains . One is a non-collagenous domain ( NC2 domain ) that is approximately 38–44 residues in length within the collagenous domain ( Figure 5—figure supplement 5 ) . The other domain is 11–13 residues in length and consists of 3–4 conserved cysteine residues , designated as the cysteine-loop , which is an extension of the canonical NC1 domain , and a candidate for an alternative cross-linking mechanism ( Figure 5—figure supplement 6 ) . As we previously established for bilaterian collagen IV , the presence of NC1 dimers after reduction of with mercaptoethanol , indicates cross-links ( Fidler et al . , 2014; Vanacore et al . , 2009 ) . Analysis in the three ctenophore species , Mnemiopsis , Beroe and Pleurobrachia , by SDS-PAGE and gel filtration chromatography , revealed the presence of NC1 hexamers , which upon reduction dissociated into dimers ( Figure 7A–D ) . Since the dimers lack Met-93 and Lys/Hyl-211 , the results indicate that ctenophore dimers are stabilized by the cysteine-loop ( Figure 7E ) . Hence , the cross-linking mechanism of ctenophores ( cysteine-loop ) is distinguished from that of Cnidaria and Bilateria ( sulfilimine cross-links ) . 10 . 7554/eLife . 24176 . 020Figure 7 . Ctenophora exhibits a novel collagen IV cross-linking mechanism . ( A ) Gel filtration chromatography elution profile of Mnemiopsis collagenase digest ( black ) and native , purified placental basement membrane NC1 hexamer ( dashed ) run successively . Three ctenophore species were digested with bacterial collagenase to solubilize NC1 hexamer for analysis of collagen IV crosslinking . ( B ) Western blot of gel filtration fractions encompassing elution of NC1 hexamer ( 12 mL to 16 . 4 mL ) from Mnemiopsis , Pleurobrachia , and Beroe , developed with NC1-specific monoclonal antibodies . HMW=high-molecular-weight complex . ( C ) Western blot of ctenophore NC1 hexamer separated by SDS-PAGE under reducing ( + ) and non-reducing ( - ) conditions ( 5% β-mercaptoethanol ) . ( D ) Reduction of the high-molecular-weight complex from Beroe ( first lane , >250 kDa ) following by alkylation results in formation of dimers at low DTT concentration , and complete reduction to monomers at high DTT concentration . ( E ) Structure of Ctenophora collagen IV group II chain , highlighting cysteine-loop region of the NC1 , and multiple-sequence alignment of cysteine-loop region of Group II chains of Mnemiopsis ( NC1 domain is partial sequence ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 020 NC1 domains are distinguishing domains of collagen IV that function as recognition modules in the assembly of collagen IV networks ( Cummings et al . , 2016 ) . Uniquely , all 10 ctenophore species possess a gene encoding only the NC1 domain , a feature not found in any other phyla . We performed phylogenetic analysis to compare the NC1 domains of non-bilaterian animal phyla with that of Bilateria ( Figure 8 ) . The results placed the NC1 domain of ctenophore collagen IV into two major groups , which are consistent with genomic orientation of Group I and Group II ( vida supra ) . We conducted additional phylogenetic analysis of the NC1 domain using RAxML to select the best tree from eleven models of evolution ( DAYHOFF , DCMUT , JTT , MTREV , WAG , RTREV , CPREV , VT , BLOSUM62 , MTMAM , and LG ) . Among these , the VT model yielded the tree with the best-fit ( Figure 8—figure supplement 1 ) . The distinction of the two groups coincides with the presence of the novel structural domains , NC2 and cysteine-loop , found exclusively in ctenophore Group II chains . Furthermore , these groups can be further subdivided into subgroups Group I ( A-E ) and Group II ( A-D ) based on phylogenetic affinity ( Figure 8—figure supplement 1 ) . RAxML phylogenetic analysis revealed a closer affinity between the ctenophore NC1 genes and collagen IV NC1 domain from the non-bilaterian animal phyla and Bilateria , as compared to ctenophore Group I and II chains ( Figure 8 ) . Furthermore , Group I and Group II ctenophore chains showed a much higher rate of divergence both within the phylum and as compared to the other non-bilaterian animal phyla and bilaterian collagen IV sequences . The unrooted tree topology illustrates the high phylogenetic affinity between ctenophore NC1 proteins and bilaterian NC1 domain that cluster closely . 10 . 7554/eLife . 24176 . 021Figure 8 . Evolutionary relationships of collagen IV and spongin NC1 domains across metazoa compared to Ctenophora . Metazoan collagen IV chains feature a sulfilimine bond cross-linked collagen IV network , with the exception of the cnidarian , Hydra , and the non-bilaterian animal phyla Porifera and Placozoa . However , the structural domains across bilaterians and the non-bilaterian animal phyla are homologous; however , Ctenophora also contains novel domains . Unrooted maximum likelihood tree of collagen IV NC1 domains in human , mouse , zebrafish , Trichoplax ( Placozoa ) , Pseudocorticium jarrei ( Porifera ) , and Oscarella sp . ( Porifera ) , in comparison with 10 ctenophore species . All analyses were based off amino acid sequence alignments of the NC1 domain , omitting the cysteine-loop region of Ctenophora NC1 domains . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 02110 . 7554/eLife . 24176 . 022Figure 8—figure supplement 1 . NC1 domain phylogeny across metazoa . Phylogenetic analysis was conducted using maximum likelihood ( using the RAxML software with the PROTGAMMAAUTO option ) on NC1 domains spanning human , mouse , zebrafish , Nematostella , Trichoplax , sponges , and 10 ctenophore species . Ctenophore collagen IV separates into two major groups apart from the other non-bilaterian animal phyla and Bilateria . These two groups are consistent with genomic orientation groupings of Group I and Group II and can be further subdivided into subgroups Group I ( IA-IE ) and Group II ( IIA-IID ) based on phylogenetic affinity . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 022 Spongins are a family of collagen IV-related proteins composed of a short collagenous domain attached to an NC1 domain . This protein family was first detected in the exoskeleton of demosponges and has been subsequently identified in cnidarians , across invertebrates ( with the exception of ecdysozoans , e . g . , C . elegans , Drosophila ) , and in basal chordates ( Aouacheria et al . , 2006; Exposito et al . , 1991 ) . Interestingly , spongins do not occur in vertebrates ( Aouacheria et al . , 2006 ) , and we did not detect them in Ctenophora . We examined the phylogenetic relationship of the NC1 domain of spongins to that of collagen IV NC1 domains ( Figure 8 and Figure 8—figure supplement 1 ) . Multiple sequence alignment showed conservation of seven cysteine residues between spongins and collagen IV , while the HSQ motif was absent in spongin sequences ( Figure 9 ) . Comparison of collagen IV sequences revealed four cysteine residues that are absent in spongin sequences . The spongin variants , however , do show conservation of three cysteines that are absent in collagen IV sequences . Collectively , the presence of collagenous domains and the conservation of collagen IV NC1 domain features within spongin NC1 domains , reveal that they are homologous to collagen IV protein domain structure , as previously noted ( Exposito et al . , 1991 ) . 10 . 7554/eLife . 24176 . 023Figure 9 . Spongins show conservation of key primary structure features within the NC1 region as compared to collagen IV . Multiple sequence alignment reveals the conservation of cysteine residues ( green arrows ) across all four families of collagen IV . Seven cysteines are common to all sequences , while spongins share three unique cysteine residues ( brown box ) . Likewise , the ctenophore NC1 protein and the NC1 domain of bilaterian collagen IV show conservation of four cysteine residues not found in spongin sequences ( red box ) . The ctenophore sequences also show conservation of the second HSQ motif ( purple arrows ) found within the bilaterian NC1 domain ( black box ) . No HSQ motifs were detected in the spongin sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 023 Within Ctenophora , collagen IV underwent numerous gene duplication events resulting in an unprecedented diversity in both gene sequence and organization in comparison to all other metazoans . Ctenophora contains between 4 and 20 collagen IV chains across species , and exhibits both head-to-head orientation and genes aligned individually on separate scaffolds or in a unidirectional tandem array . Interestingly , Ctenophora has both collagen IV genes and a standalone NC1 gene . In both Nematostella and Trichoplax , there are two collagen IV genes exhibiting head-to-head gene orientation , similar to that of Bilateria . In Porifera , the ECM of homoscleromorphs is composed of two collagen IV genes , while in demosponges , it is composed of spongin , a collagen IV variant . Spongin is absent in Ctenophora but is present in non-bilaterian animal phyla , invertebrates , and lower chordates along with collagen IV throughout invertebrates and lower chordates , with the exception of Drosophila and C . elegans . Structural and phylogenetic analysis of spongin shows it is homologous to collagen IV ( Figures 8 and 9 ) . Collectively , collagen IV genes are highly conserved across the non-bilaterian animal phyla and Bilateria; yet , in Ctenophora , these genes are more diverse and distinct including a novel cross-linking mechanism , with up to 20 distinct genes compared with six in vertebrates . Moreover , the collagen IV gene is absent in the unicellular sister-groups ( choanoflagellates , filastereans , amoebozoans , and apusozoans ) , suggesting it was an early metazoan innovation . To address the evolutionary origin of the collagen IV gene , we compared two scenarios , Ctenophora-first versus Porifera-first ( Ryan et al . , 2013; Moroz et al . , 2014; Whelan et al . , 2015; Telford et al . , 2016 ) . In both scenarios , collagen IV appeared in an early metazoan ancestor or a unicellular ancestor but was secondarily lost in demosponges and hexactinellid sponges ( Figure 10A and B ) . It is noteworthy that the NC1 gene is present only in Ctenophora ( Figure 10C and D ) , suggesting that this gene is a remnant from an early metazoan ancestor , and a forerunner to the NC1 domain of the ancestral collagen IV gene . This NC1 gene encodes a key recognition module that directs the assembly of collagen IV suprastructures ( Cummings et al . , 2016 ) . In comparison , spongin appeared after the divergence of the Ctenophora phylum in the Ctenophora-first hypothesis , yet in the Porifera-first hypothesis it would have appeared alongside collagen IV in the early metazoan ancestor or unicellular ancestor . With either hypothesis , the collagen IV gene coincided with the appearance of multicellular animals . Although laminin genes appear to have arisen prior to the metazoan lineage , with laminin-related genes appearing in unicellular choanoflagellates ( Fahey and Degnan , 2012 ) ( Figure 10A and B ) . 10 . 7554/eLife . 24176 . 024Figure 10 . Collagen IV and Laminin gene evolution under Ctenophora or Porifera-first hypotheses . ( A and B ) Comparison of collagen IV , spongin , and laminin gene evolution gain and loss evolutionary events in Ctenophora-first and Porifera-first hypotheses . ( C and D ) Comparison of NC1 gene evolution gain and loss evolutionary events in Ctenophora-first and Porifera first hypotheses . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 024 Collectively , we propose a model for collagen IV gene evolution that incorporates both Ctenophora-first and Porifera-first hypotheses ( Figure 11 ) . The presence of Gly-X-Y collagenous repeats , in the absence of a collagen IV gene , in choanoflagellates and amoebozoa , and the presence of a NC1 gene in the Ctenophora phylum suggests that Gly-X-Y repeats combined with an NC1 domain gene in an early metazoan ancestor , or possibly in a unicellular ancestor , forming an ancestral collagen IV gene . This combination of domains is analogous to the domain shuffling events that gave rise to the developmental protein , hedgehog ( Adamska et al . , 2007 ) . Within Ctenophora , collagen IV genes underwent unprecedented experimentation with several duplication events resulting in up to twenty distinct genes with both tandem and head-to-head organization . Within Porifera , the collagen IV gene was duplicated with a head-to-head orientation . This head-to-head feature was conserved in both sequence and gene structure throughout non-bilaterian animals and Bilateria ( Figure 11 ) , with the known exception of C . elegans ( Guo and Kramer , 1989 ) . Two additional rounds of genome duplication resulted in six collagen IV genes in the vertebrate subphylum . Spongin , a collagen IV variant , first appeared in Porifera and is conserved throughout invertebrates , with the exception of Ecdysozoa and is found in cephalochordates ( Branchiostoma floridae ) and tunicates ( Ciona intestinalis ) . The spongin gene arose either by domain shuffling of Gly-X-Y repeats from a unicellular ancestor and the NC1 gene , analogous to the assembly of the ancestral collagen IV gene , or diverged from an ancestral collagen IV gene ( Figure 11 ) . 10 . 7554/eLife . 24176 . 025Figure 11 . The non-bilaterian animal phyla reveal an evolutionary model for collagen IV . Based off our phylogenetic analysis: ( 1 ) the presence of the ancestral NC1 domain in Ctenophora may have resulted from tandem duplication of the ancestral C4 or conservation of the ancestral NC1 domain . The last common ancestor of ctenophore and the non-bilaterian animal phyla may have expressed both the ancestral C4 and the ancestral NC1 domain . ( 2 ) Intergenic duplication of the ancestral collagen IV resulted in the head-to-head orientation . Errors in gene duplication may have given rise to spongins as they lack the domain-swapping region of the NC1 domain ( a determined by predicative modeling ) and have truncated collagenous tails . ( 3 ) The presence of six collagen IV genes arranged in a head-to-head orientation in vertebrates likely resulted from the two rounds of genome duplication that occurred in the vertebrate lineage . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 025 Collagen IV protein , or its spongin variant , is a required ECM component for all extant multicellular animals , considering that all animals investigated contain either collagen IV or spongin , and that the essentiality of collagen IV during development has been established in several studies ( Gupta et al . , 1997; Pöschl et al . , 2004; Gotenstein et al . , 2010; Bhave et al . , 2012 ) . The collagen IV protein is associated with two distinct organizations of cells; one in which the ECM contains collagen IV broadly dispersed between communities of cells ( Mnemiopsis ) and the other in which collagen IV is a component of a well-defined BM underlying a layer of cells ( Beroe , Pleurobrachia , Homoscleromorph sponges , and Nematostella ) , a hallmark feature of epithelial bilaterian tissues ( Figure 12 ) . The absence of BMs in Trichoplax and Mnemiopsis suggests there is an unknown component that facilitates the assembly of collagen IV and laminin into a basement membrane . 10 . 7554/eLife . 24176 . 026Figure 12 . Collagen IV enabled the transition to multicellularity and the evolution of epithelial tissues in metazoa . Collagen IV was a primordial innovation in early metazoan evolution , providing the architectural foundation for ECM formation . Choanoflagellates exist as singular or in colonies , yet do not have an ECM . Spongins are similar in domain structure and phylogeny to NC1 domains across metazoan collagen IV , and are variants of collagen IV , arising during the divergence of demosponges . Collagen IV , as a member of the basement membrane toolkit , enabled the evolution of multicellularity . Basement membranes juxtaposed to plasma membrane underlying a layer of polarized cells are a fundamental architectural unit of epithelial tissues . A layer of apical/basal-polarized cells that are laterally connected by tight junctions between plasma membranes , and basally anchored via integrin receptors embedded in plasma membranes to a basement membrane suprastructure is a fundamental architectural unit . DOI: http://dx . doi . org/10 . 7554/eLife . 24176 . 026 Collectively , we conclude that collagen IV and its spongin variant are primordial components of the extracellular microenvironment , and collagen IV , as a component of BM , enabled the assembly of a fundamental architectural unit for the genesis and evolution of multicellular tissues ( Figure 12 ) . This unit is characterized by a layer of apical/basal-polarized cells that are laterally connected by tight junctions between plasma membranes , which are basally anchored via integrin receptors embedded in plasma membranes to a basement membrane supra-scaffold . In turn , this architectural unit served as the building block that enabled the formation and evolution of epithelial tissues , the ever-increasing complexity and size of organisms , and for the expansion and diversity of the animal kingdom . Transcriptomes used in this study were sequenced at the Vanderbilt Technologies for Advanced Genomics Core Facility ( VANTAGE , Nashville , TN ) . The Illumina TruSeq mRNA Sample Preparation Kit was used to convert the mRNA in 100 ng of total RNA into a library of template molecules suitable for subsequent cluster generation and sequencing on the Illumina HiSeq 2500 using the rapid run setting . The pipeline established in VANTAGE was followed and is briefly described below . The first step was a quality check of the input total RNA by running an aliquot on the Agilent Bioanalyzer to confirm RNA integrity . The Qubit RNA fluorometry assay was used to measure sample concentrations . The input-to-library prep was 100 ng of total RNA ( 2 ng/ul ) . The poly-A containing mRNA molecules were concentrated using poly-T oligo-attached magnetic beads . Following purification , the eluted poly ( A ) RNA was cleaved into small fragments of 120–210 base pair ( bp ) using divalent cations under elevated temperature . The cleaved RNA fragments were copied into first strand cDNA using SuperScript II reverse transcriptase and random primers . This step was followed by second strand cDNA synthesis using DNA Polymerase I and RNase H treatment . The cDNA fragments then went through an end repair process , the addition of a single ‘A’ base , and then ligation of the Illumina multiplexing adapters . The products were then purified and enriched with PCR to create the final cDNA sequencing library . The cDNA library then undergoes quality control by running on the Agilent Bioanalyzer HS DNA assay to confirm the final library size and on the Agilent Mx3005P qPCR machine using the KAPA Illumina library quantification kit to determine concentration . A 2 nM stock was created and samples pooled by molarity for multiplexing . From the pool , 12 pmoles were loaded into each well for the flow cell on the Illumina cBot for cluster generation . The flow cell was then loaded onto the Illumina HiSeq 2500 utilizing v3 chemistry and HTA 1 . 8 . The raw sequencing reads were processed through CASAVA-1 . 8 . 2 for FASTQ conversion and demultiplexing . The Illumina chastity filter was used and only the PF ( passfilter ) reads are retained for further analysis . Assembly of transcriptomes was performed using both Velvet/Oases and Trinity software packages with default settings ( see list of commands subsection below ) . List of commands used in sequence search: Velvet/Oases . velveth $outDir $hash_length -fastq -shortPaired $in_shuffled_ sequence_file velvetg $outDir -read_trkg yes oases $outDir -ins_length 150 Trinity Trinity . pl –output $outDir –seqType fq –JM 90G –left $file1 –right $file2 –CPU 16 Animals were initially fixed whole in cold 2 . 5% glutaraldehyde in 0 . 1M cacodylate buffer , pH7 . 4 overnight in the refrigerator . After this initial fixation , the samples were stable enough so that small portions of selected areas could be dissected out and fixed for a further 24 hr at 4°C in 2 . 5% glutaraldehyde in 0 . 1M cacodylate . Following fixation , the samples were washed in 0 . 1M cacodylate buffer , incubated 1 hr in 1% osmium tetroxide at RT then washed with 0 . 1M cacodylate buffer , dehydrated through a graded ethanol series and embedded in epoxy resin . Semi-thin sections ( 0 . 5 microns ) were cut , stained with toluidine blue and viewed by light microscopy to choose appropriate areas for study . Thin sections ( 70–80 nm ) were cut from these selected areas and contrasted using 2% uranyl acetate and Reynold’s lead citrate , and imaged on an FEI Tecnai T12 electron microscope . Whole ctenophore tissues were frozen in liquid nitrogen , pulverized in a mortar and pestle and then homogenized in 2 . 0 ml g−1 digestion buffer and 0 . 1 mg ml−1 Worthington Biochemical bacterial collagenase and allowed to digest at 37°C , with spinning for 24 hr . Liquid chromatography purification of solubilized NC1 varied by species based on protein yield . All ctenophore NC1s were purified by gel-exclusion chromatography ( GE Superdex 200 10/300 GL ) . For reduction and alkylation of collagen IV NC1 hexamers , fractions containing high-molecular-weight complex from size-exclusion chromatography were concentrated by ultrafiltration and reduced in TBS buffer with various concentrations of DTT . After incubation for 30 min at 37°C , samples were alkylated with twofold molar excess of iodoacetamide for 30 min at room temperature in the dark . After mixing with SDS loading buffer , samples were heated for 5 min in boiling water bath and analyzed by non-reducing SDS-PAGE . Collagenase-solubilized NC1 hexamers were analyzed by SDS-PAGE in 12% bis-acrylamide mini-cells with Tris-Glycine-SDS running buffer . Sample buffer was 62 . 5 mM Tris-HCl , pH 6 . 8 , 2% SDS ( w/v ) , 25% glycerol ( w/v ) , 0 . 01% bromophenol blue ( w/v ) . Western blotting of SDS-dissociated NC1 hexamer was developed with JK-2 , rat monoclonal antibody ( kindly provided by Dr . Yoshikazu Sado , Shigei Medical Research Institute , Okayama , Japan ) . All Western blotting in Figure 6 was done with Thermo-Scientific SuperSignal West Femto chemiluminescent substrate and digitally imaged with a Bio-Rad GelDoc . Whole ctenophore tissues were placed in 150 mL beaker and as much liquid was removed as possible . Each tube with tissue was filled with 100 mL ice-cold ctenophore fixation buffer 1 [80ul Glutaraldehyde ( 25% ) , 0 . 02% final concentration; 25 mL Paraformaldehyde ( 16% ) , 4 . 0% final concentration; 75 mL 0 . 2um-filtered seawater ( Red Sea Coral Pro Salt ) ] , inverted a few times gently , and left at 4 degrees Celsius for 5 min . Buffer 1 was then removed , and 100 mL of ctenophore fixation buffer 2 was added [25 mL Paraformaldehyde ( 16% ) , 4 . 0% final concentration; 75 mL 0 . 2um-filtered seawater ( Red Sea Coral Pro Salt ) ] . Buffer 2 was then removed and tissues were gently washed five times with cold 1X PBS . Fixation protocol adopted from Pang and Martindale , Ctenophore Whole-Mount Antibody Staining ( Pang and Martindale , 2008 ) . Tissues were then embedded in parafilm and sectioned onto individual slides for IHC staining . After deparaffinization and rehydration , tissues underwent heat-induced epitope retrieval with DAKO and microwaved for 15 min . Cold tap water was then run over tissues for 10 min , followed by two washed with 1X PBS and stored in 1X PBS . Immunostaining occurred at room temperature , with blocking by a 5% serum blocking buffer ( 1X PBS pH 7 . 4/5% normal goat serum/0 . 1% Triton X-100 ) for 60 min . All IHC for collagen IV was conducted with the rat monoclonal antibody ( mAb ) JK-2 , and antibody dilution was done in 5% serum blocking buffer accordingly . Alexa488 tagged anti-rat secondary was used for the fluorochrome-conjugated secondary antibody ( RRID:AB_10893331 ) , and dilution was also done in 5% serum blocking buffer . IHC images were taken on a Zeiss Axioplan microscope . Lenses used were a 20X lens ( Plan-APOCHROMAT 20X/0 , 75; ∞/0 , 17 ) and a 40X lens ( Plan-NEOFLUAR 40X/0 , 75; ∞/0 , 17 ) . Images were taken at room temperature , approximately 20 degrees Celsius . All images were done in an imaging medium of air . Fluorochromes used were Alexa488 ( green ) for collagen IV , and Hoescht stain ( blue ) was used for nuclei staining . The Camera for imaging was a Photometrics CoolSnap HQ , using Metamorph 7 . 7 . 0 . 0 software ( RRID:SCR_002368 ) . Slight gamma correction of ( < ± 0 . 2 ) after acquisition to adjust contrast . Images captured were merged with ImageJ64 , 1 . 48v ( RRID:SCR_003070 ) . The evolutionary relationship between collagen IV and spongins was analyzed using the NC1 domains of each of the 139 sequences in our dataset . NC1 domains were aligned using the Geneious alignment tool within Geneious software package ( RRID:SCR_010519 ) , version 8 . 1 . 9 with default settings ( Silvestro and Michalak , 2012 ) . The resulting sequence alignment , which was 881 amino acid sites in length , was used to reconstruct the phylogeny of NC1 domains under the maximum likelihood optimality criterion as implemented in the RAxML software ( RRID:SCR_006086 ) , version 8 . 2 . 3 ( Stamatakis , 2014 ) . The phylogenetic analysis was performed using the PROTGAMMAAUTO option , which selects the substitution model with the best fit to the alignment among a set of among a set of 11 models ( these were: DAYHOFF , DCMUT , JTT , MTREV , WAG , RTREV , CPREV , VT , BLOSUM62 , MTMAM , and LG ) . In the case of the NC1 domain phylogeny , the model with the best fit was the VT model ( Müller and Vingron , 2000 ) . Robustness in phylogeny inference was assessed with 100 bootstrap replicates .	The emergence of the diversity of multicellular animals involved cells joining together to form tissues and organs . The ‘glue’ that enabled the cells to work together is made of rope-like molecules called collagen , which assemble into scaffolds . These smart scaffolds tether proteins forming basement membranes that connect cells , provide strength to tissues , and transmit information that influences how the cells behave . How did collagen evolve over millions of years to enable the ever-increasing complexity , size and diversity of animals ? To investigate , Fidler , Darris , Chetyrkin et al . explored the tissues of the most ancient of currently living animals – the comb jellies and sponges . This revealed that among all the collagens that make up the human body , a type called collagen IV was a key innovation that enabled single celled organisms to evolve into multicellular animals . Collagen IV , as molecular glue , enabled the formation of a fundamental architectural unit of basement membrane and cells that allowed multicellular tissues and organs to evolve . The findings presented by Fidler , Darris , Chetyrkin et al . pose questions about how collagen IV glues cells together , and how information is stored in the rope-like scaffolds to influence cell behavior . Understanding these processes could ultimately lead to the development of new treatments for diseases in which the collagen smart scaffolds play a key role , such as in kidney diseases and cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2017	Collagen IV and basement membrane at the evolutionary dawn of metazoan tissues
Ubiquitylation ( ubi ) by the E3-ligases Mindbomb1 ( Mib1 ) and Neuralized ( Neur ) is required for activation of the DSL ligands Delta ( Dl ) and Serrate ( Ser ) to activate Notch signalling . These ligases transfer ubiquitin to lysines of the ligands' intracellular domains ( ICDs ) , which sends them into an Epsin-dependent endocytic pathway . Here , we have tested the requirement of ubi of Dl for signalling . We found that Dl requires ubi for its full function , but can also signal in two ubi-independent modes , one dependent and one independent of Neur . We identified two neural lateral specification processes where Dl signals in an ubi-independent manner . Neur , which is needed for these processes , was shown to be able to activate Dl in an ubi-independent manner . Our analysis suggests that one important role of DSL protein ubi by Mib1 is their release from cis-inhibitory interactions with Notch , enabling them to trans-activate Notch on adjacent cells . Signalling through the Notch signalling pathway is required in many developmental processes of probably all metazoans ( Bray , 2006; Fiúza and Arias , 2007 ) . Notch receptors are hetero-dimeric type 1 trans-membrane proteins , which are activated through ligands of the Delta/Serrate/Lag2 ( DSL ) protein family . The genome of Drosophila contains two DSL ligands , Delta ( Dl ) and Serrate ( Ser ) , which are structurally related in their extracellular domain ( ECD ) . Their binding elicits two consecutive proteolytic cleavages of Notch performed by Kuzbanian ( Kuz ) /ADAM10 and the γ-secretase complex , which results in the release of the intracellular domain ( NICD ) into the cytoplasm . The first , Kuz-mediated cleavage ( S2-cleavage ) occurs in the extracellular juxta-membrane region and removes the ecto-domain ( NECD ) . The shedding of NECD triggers the second cleavage within the transmembrane domain by the γ-secretase ( S3-cleavage ) , allowing the release of NICD . NICD enters the nucleus and associates with the CSL transcription factor Suppressor of Hairless ( Su ( H ) ) and co-factors to activate the transcription of target genes . In the absence of NICD , Su ( H ) acts as a repressor of transcription in association with Hairless ( H ) ( Barolo et al . , 2002; Brou et al . , 1994 ) . A number of studies have demonstrated the importance of endocytosis for Notch signalling and regulation of its activity ( Le Borgne et al . , 2005a ) . During signalling , the ligands must be endocytosed in the signal-sending cell to activate Notch in the signal-receiving one . Two membrane-associated E3-ligases , Neuralized ( Neur ) and Mindbomb1 ( Mib1 ) , play a crucial role in the activation of the ligands and their endocytosis ( Le Borgne et al . , 2005a ) . E3-ligases mediate the ubiquitylation ( ubi ) of specific substrates at lysines ( Ks ) . This label constitutes a common signal to initiate endocytosis and is thought to initiate endocytosis of the DSL ligands . In Drosophila the two ligases can functionally substitute for one another in at least some processes such as the wing primordium , suggesting that they perform similar functions , even though they share no obvious sequence similarity ( Weinmaster and Fischer , 2011 ) . Both ligases contain a Ring Finger domain ( RF ) that catalyses the ubi . neur is highly expressed in the mesoderm and then all over the neuroectoderm in the embryo and is restricted to neural precursor cells until the pupal stages . Mib1 is ubiquitously expressed , indicating that most DSL signalling is dependent on Mib1 . Recent analysis indicates that human Mib1 ( MIB1 ) binds to two peptide motifs in the ICD of the mammalian Ser ortholog Jagged1 ( Jag1 ) ligand termed the N- and C-Box ( McMillan et al . , 2015 ) . Two domains in the Mib1 N-terminus , MZM and REP , are responsible for this binding , with MZM binding the N- and REP the C-box ( McMillan et al . , 2015 ) . Dl has both of these motifs in its ICD; the N-box , also termed ICD2 , was shown to be essential for its function ( Daskalaki et al . , 2011 ) . Since MIB1 can substitute for Mib1 in Drosophila , it is likely that the interactions between Mib1 and the N-Box are conserved ( McMillan et al . , 2015 ) . Neur binds to a distinct site with the core consensus sequence NXXN , also termed ICD1 in Dl ( NEQN in Dl ) . Such a motif is present in the ICDs of both ligands of Drosophila ( Daskalaki et al . , 2011; Fontana and Posakony , 2009; Glittenberg et al . , 2006 ) . Both ligases are involved in endocytosis of the ligands and their ubi: we recently showed that Neur and Mib1 can ubiquitylate the ICD of Dl in an ICD1 and ICD2 dependent manner , respectively ( Daskalaki et al . , 2011 ) . Additionally , a K at position 742 ( K742 ) of the ICD was identified as the preferred target for Neur in Drosophila . Moreover , we found a good correlation between the ability of Dl to undergo ubiquitylation by Neur and Mib1 , its internalisation efficiency and its ability to signal . In mammals , a Dll1 variant in which all 17 Ks of its ICD are replaced by arginines ( Rs ) ( Dll1K17 ) is not ubiquitylated and inactive ( Heuss et al . , 2008 ) . Finally , the Drosophila endocytic adapter protein Liquid facets ( Lqf ) , which is the ortholog of Epsin , is absolutely required for the function of DSL ligands ( Overstreet et al . , 2004; Wang and Struhl , 2004 ) . The structural analysis of Lqf revealed that its ubiquitin binding motifs ( UIMs ) are necessary for function ( Overstreet et al . , 2004; Wang and Struhl , 2004; Ho et al . , 1989 ) . Two , not mutually exclusive , models have been suggested to explain why the activity of the ligands requires endocytosis . In the first model ( pulling force model ) the Epsin-mediated entry of ubiquitylated ligands in Clathrin-coated pits creates a pulling force that is essential for shedding of the NECD . In the second model ( recycling model ) , ligand signalling activity is thought to depend on events taking place after internalisation of the ligands into endosomes , such as enzymatic processing of the ligand into the active form or packaging into exosomes: accordingly , signalling activity would require recycling of the ligands to the cell surface to be able to activate Notch . Previous work has provided evidence for both models ( reviewed in [Weinmaster and Fischer , 2011] ) . Although Mib1 and Neur are clearly required for DSL signalling , their role is not entirely understood: ( 1 ) While endocytosis of Ser is virtually abolished in mib1 mutant cells , that of Dl is not obviously affected , although both Ser and Dl function is impaired ( Wang and Struhl , 2004 ) . ( 2 ) Recent work suggests that Neur might have a function that is separable from its ligase function and required for endocytosis ( Skwarek et al . , 2007 ) . ( 3 ) The phenotype of mib1 mutants is milder than that of mutants of other genes involved in Notch signalling . These results raise the possibility that one or both ligands might possess an activity that is independent of ubi . In addition to activating Notch in adjacent cells ( trans-activation ) , the ligands engage in inhibitory interaction with Notch molecules present in the same cell ( Klein et al . , 1997; Micchelli et al . , 1997 ) . This cis-inhibition occurs if DSL ligands are expressed above a threshold level . The underlying mechanism is not well understood , but cis-inhibition can be suppressed by co-expression of Notch , indicating that the ratio between Notch and ligand concentration in a cell is an important parameter ( Klein et al . , 1997 ) . Moreover , it has been shown that the ECD of the ligands is involved ( Glittenberg et al . , 2006 ) . Whether the ICD is involved , or cis-inhibition is influenced by Mib1 and Neur is not established , although there is evidence that enhanced/diminished ligand endocytosis relieves/aggravates ( respectively ) cis-inhibition of Ser and Dl ( Glittenberg et al . , 2006 ) Here , we tested the activity of a Dl variant ( termed DlK2R ) in which all 12 Ks of its ICD were replaced by structurally related arginines ( Rs ) . We show that DlK2R possesses signalling activity . However , its activity is reduced compared to Dl , indicating that ubi is required for the full function of Dl . The reduction in activity is in part caused by increased cis-inhibitory activity of DlK2R indicating that the ICD and its ubi contribute to the degree/strength of cis-inhibition of a ligand . We found that an important function of Mib1 during normal signalling is the release of the ligands from cis-inhibitory interaction with Notch . Our results revealed that Dl can signal in two ubi-independent modes , one dependent and one independent of Neur . The results also indicate that Neur and Mib1 have different mechanisms to activate Dl . In contrast to Mib1 , Neur can promote Dl signalling in an ubi-independent manner . We identified neuron sibling specification during larval neurogenesis and also specification of the sensory organ precursor of the bristle sensillum as naturally occurring instances of Dl/Notch signalling , where Dl requires Neur , but ubi seems to be dispensable . Finally , we found an instance of Dl induced Notch signalling that is independent of both Neur and ubi . During wing development of Drosophila interactions between dorsal and ventral cells mediated by the Notch pathway at the dorso-ventral ( D/V ) compartment boundary result in the establishment and maintenance of a stripe of expression of Wingless ( Wg ) that straddles the D/V boundary of the wing disc at the end of the third larval instar stage ( Figure 1A , arrow ) ( Klein , 2001 ) . During the initial phase , dorsal boundary cells signal to ventral boundary cells to activate Wg expression and to increase expression of Dl ( Micchelli et al . , 1997; de Celis and Bray , 1997; de Celis et al . , 1996; Troost and Klein , 2012 ) . The ventral boundary cells signal back through Dl to activate expression of Wg in dorsal cells ( Figure 1B , step 1 ) . After establishment , expression of Wg in boundary cells is maintained through a feedback loop ( DS-loop , Figure 1B steps 2 and 3 ) : it involves the induction of expression of Dl and Ser in cells adjacent to the boundary cells by Wg signalling ( Figure 1B step 2 ) . Wg expression at the D/V boundary is then maintained by back signalling from these cells to boundary cells through the Notch ligands ( Figure 1B step 3 ) . Loss of the activity of Notch results in the loss of expression of Wg along the D/V boundary and a reduction of the diameter of its ring-like domains of expression in the proximal wing . The reduction depends on the degree of loss of Notch pathway activity . Strong loss of activity , as observed in mutants of the γ-secretase component Psn , reduces the inner ring-like domain to a spot or causes its complete absence ( Figure 1C , arrowhead ) ( Klein and Martinez-Arias , 1998; Koelzer and Klein , 2006; Struhl and Greenwald , 1999a; Ye et al . , 1999 ) . Weaker loss of Notch activity , as observed in Ser mutants ( Klein and Martinez-Arias , 1998 ) , causes a reduction in diameter of the domains ( Figure 1D , arrowhead ) . The phenotype of mib1 null mutant wing discs resembles that of Ser mutants ( Figure 1E , arrowhead; compare with D ) . We previously showed that residual activity of Dl is responsible for the weaker phenotype of Ser mutants ( Klein and Martinez-Arias , 1999 ) . The resemblance prompted us to search for residual Notch pathway activity in mib1 null mutants . We found that the Notch pathway is indeed weakly active . The Notch activity reporter Gbe + Su ( H ) was initially expressed along the D/V boundary in early third larval instar stages of mib1 mutant wing discs and lost only later in development ( Figure 1F–I , arrow ) . Also the vestigial boundary enhancer ( vgBE ) ( Williams et al . , 1994 ) , a direct endogenous target of the Notch pathway , was expressed in early wing discs ( Figure 1J , K , arrow ) . In the notum Gbe + Su ( H ) is expressed in four stripes ( Figure 1L ) . We observed residual expression of stripe 3 in mib1 mutants ( S3; Figure 1L , M , arrow ) . S3 expression is dependent on Notch signalling , since depletion of the activity of Notch by expression of N-RNAi in the notum with ciGal4 resulted in the complete abolishment of its expression ( Figure 1N , O , arrow in O ) . Together , these results confirm that residual ligand-dependent Notch activity is present in mib1 mutants . Hairless ( H ) encodes a member of the transcriptional repressor complex assembled around Su ( H ) in the absence of Notch activity to repress expression of the target genes of the pathway ( Barolo et al . , 2002 ) . Loss of its function results in de-repression of a subset of target genes and enhancement of residual Notch signalling in Ser mutants , which is sufficient to re-establish the expression of Wg and Gbe + Su ( H ) along the D/V boundary of the wing disc ( Troost and Klein , 2012; Klein et al . , 2000 ) . We observed that , as in the case of Ser mutants , concomitant loss of H function partly restores the expression of Wg and maintains that of Gbe + Su ( H ) along the D/V boundary in mib1 mutants ( Figure 1—figure supplement 1A–C , arrow ) . The comparison with the expression of the dorsal selector Apterous ( Ap ) indicated that the expression of Gbe + Su ( H ) was restricted to dorsal boundary cells ( Figure 1—figure supplement 1D–F ) . Since Ser is unable to activate the Notch pathway in dorsal cells because of the activity of Fng ( Panin et al . , 1997 ) , this observation strongly suggests that Dl signalling is responsible for the re-appearance of the expression of the Notch target genes in mib1 H double mutants . The results indicate that a residual activity of the Notch pathway , probably induced by Dl , is present in the absence of mib1 function . Since Mib1 is the only known E3-ligase for Notch ligands known to be involved during wing development , the results raise the possibility that Dl has an activity that is independent of ubi . Although it has been shown that Mib1 is absent in the discs mutant for the mib1 null alleles used here ( Le Borgne et al . , 2005 ) , it might be possible that a long-lasting maternal component of mib1 exists that provides cells of the imaginal disc with residual Mib1 activity . It is also possible that another unidentified E3-ligase contributes to Dl signalling . These issues are addressed in the following . In the experiments described in the following , we expressed a set of UAS Dl constructs with ptcGal4 to test their activity in the wing imaginal disc . If not stated otherwise , the constructs are inserted in the attP landing site 51C ( Flybase: M{3xP3-RFP . attP'}ZH-51C ) to achieve similar expression levels and allow direct comparison ( Bischof et al . , 2007 ) . All constructs are HA-tagged at their C-terminus . ptcGal4 is expressed on the anterior side of the anterior-posterior boundary ( A/P boundary ) in a gradient that increases towards the posterior . Its posterior expression boundary coincides with the A/P boundary ( Figure 2A–C , arrow in B ) . Continuous expression of Dl-HA with ptcGal4 at 25°C induces ectopic expression of the Notch target gene Wg and the more sensitive Notch activity reporter Gbe + Su ( H ) in two stripes perpendicular to their normal expression along the D/Vboundary , as has been reported for untagged versions ( Doherty et al . , 1996 ) ( Figure 2D–F , white arrows in D , E ) . The anterior stripe ( a in Figure 2E ) is located in the anterior region of low expression , while the posterior stripe ( p in Figure 2E ) is adjacent to the ptc domain in posterior non-expressing boundary cells ( Figure 2F ) . Wg is not induced in regions of high Dl expression close to the A/P boundary ( Figure 2D , E , yellow arrow , F ) , because of cis-inhibition of Notch by high Dl concentrations ( Doherty et al . , 1996 ) . The cis-inhibition also interrupts expression of endogenous Wg along the D/V boundary ( Figure 2D , E , white and yellow arrows , F ) . If Dl is active in mib1 mutants , ectopic expression of Dl-HA should induce ectopic activation of Notch in this mutant ( Figure 2G–J ) . Indeed , expression of Dl-HA in mib1 mutants caused weak and diffuse ectopic expression of Wg in the mib1 mutant discs ( 100% penetrance , n = 7; Figure 2I , arrow ) . In addition , the diameter of the ring-like domains of Wg is increased ( Figure 2I , arrowhead , compare with H , arrowhead ) . Expression of the more sensitive Gbe + Su ( H ) was induced throughout the ptcGal4 expression domain , even in the notum where , in contrast to the wing , Notch activity does not induce expression of the ligands ( Figure 2J , arrow and arrowhead ) . Dl-HA was not able to induce ectopic expression of the activity reporters in Psn mutants , in which the pathway is interrupted on the signal-receiving side ( Figure 2—figure supplement 1A–C ) . To rule out that residual activity of Mib1 supports Dl signalling in the mutants , we generated a Dl variant where the core of the Mib1 binding box in its ICD ( ICD2 , see Figure 3A , [Daskalaki et al . , 2011] ) is mutated to alanine ( Dl-HAi2 ala ) . This variant induced activity of the Notch pathway in the mib1 mutant background in a manner comparable to Dl-HA throughout the ptcGal4 domain ( Figure 2K , L , arrow and arrowhead , compare with I , J ) . This finding indicates that Dl can activate the Notch pathway in a mib1 independent manner and strongly argues against a residual Mib1 activity in the mutants . The expression of a Ser-HA variant , also inserted into the 51C landing site , failed to induce ectopic expression of Wg and Gbe + Su ( H ) in mib1 mutants , indicating that it is completely dependent on Mib1 ( Figure 2M , N ) . Altogether , the results confirm on the one hand a strong dependency of Dl on the function of mib1 . On the other hand , they indicate that Dl , in contrast to Ser , can signal in the absence of the function of mib1 and raise the possibility that part of the activity of Dl is independent of ubi . Alternatively , it is possible that the residual activity of Dl is induced by the activity of an unidentified Dl specific E3-ligase . To discriminate between the possibilities , we generated a variant of Dl where all 12 Ks of its ICD are replaced by the structurally similar arginine ( R ) ( DlK2R-HA , Figure 3A ) and inserted it into the genomic 51C landing site . It has been shown that a corresponding variant of Dll1 is not ubiquitylated ( Heuss et al . , 2008 ) . In agreement , we here found that , in contrast to Dl-HA , DlK2R-HA is not ubiquitylated by Mib1 or Neur in S2 cells ( Figure 3—figure supplement 1 ) . Expression of DlK2R-HA with ptcGal4 resulted in the interruption of Wg expression along the D/V boundary suggesting a strong negative effect on Notch signalling at the D/V boundary ( Figure 3B , yellow arrow ) . The width of Wg interruption was much larger than that caused by Dl-HA suggesting that DlK2R-HA has a significantly stronger cis-inhibitory activity compared to Dl-HA ( Figure 3B compare with Figure 2E , yellow arrow ) . In an attempt to quantify the increase in cis-inhibition , we measured the width of the gap induced in the expression domain of Wg by expression of Dl-HA and DlK2R-HA ( Figure 3—figure supplement 2A–C ) . We found that the gap induced by Dl-HA was on average 3 , 3 ( n = 10 ) nuclei and that induced by DlK2R-HA 16 , 7 nuclei ( n = 9 ) . This dramatic increase in the gap size of Wg expression confirms the strong increase in the cis-inhibitory abilities of DlK2R-HA . However , DlK2R-HA was not completely inactive . It induced ectopic expression of Wg in posterior boundary cells in 44% of discs ( n = 9; Figure 3E , arrows ) , suggesting that it still possesses signalling activity . Indeed , we found that DlK2R-HA consistently induced ectopic expression of the more sensitive Gbe + Su ( H ) reporter in two stripes throughout the ptcGal4 domain ( Figure 3C , D , arrowhead ) . DlK2R-HA was not able to induce expression of Gbe + Su ( H ) in Psn mutant discs , indicating that its expression depends on the activity of the Notch pathway ( Figure 3F ) . Importantly , DlK2R-HA was able to induce strong expression of Gbe + Su ( H ) in mib1 null mutants , indicating that it can activate the Notch pathway in a mib1 independent manner ( Figure 3G–J ) . Also weak ectopic expression of Wg was induced in a fraction of mib1 discs ( 18% , n = 17; Figure 3I , arrow ) . The activity of DlK2R-HA was independent of the presence of the Neur binding site in its ICD , as a variant where the Neur binding site ( NEQNAV , see ( Fontana and Posakony , 2009 ) ; termed i1 in Figure 3A ) is mutated to alanines ( DlK2Ri1 ala-HA ) induces the expression of Gbe + Su ( H ) in mib1 mutants in a manner comparable to DlK2R-HA ( Figure 3K , L , arrowheads , compare with G ) . This finding excludes the possibility that so far undetected weak ubiquitous expression of Neur activates DlK2R-HA in mib1 mutants . In an earlier round of constructs UAS DlK2R-HA had been randomly inserted into the genome and a second- and third-chromosomal insertion had been selected for analysis . These constructs behaved in the same manner as the 51C insertion , but the phenotypes were more severe ( Figure 3—figure supplement 3 ) . Expression of each insertion by ptcGal4 resulted in a splitting of the wing primordium into two small halves ( Figure 3—figure supplement 3A–D , yellow arrow in A , B , D ) . A similar split was observed if a dominant-negative variant of Dl , Dlstu , or a Notch-RNAi construct is expressed ( Figure 3—figure supplement 3E–H , yellow arrow in E , G ) . This similarity further confirms that the loss of the Ks strongly increases the cis-inhibitory abilities of Dl . In contrast to Notch-RNAi and Dlstu , DlK2R-HA induced an ectopic stripe of Wg expression in the adjacent posterior boundary cells , indicating that it can activate Notch in trans in non-expressing neighbours ( Figure 3—figure supplement 3A–C , arrowheads in A , B , arrow in C ) . The phenotype of DlK2R-HA expression could be mimicked by co-expression of N-RNAi and Dl in wing imaginal discs ( Figure 3—figure supplement 3I , J , compare with A , B ) confirming the increase of cis-inhibitiory abilities without a loss of trans-signalling for the K free DlK2R-HA . We sought to suppress the strong cis-inhibitory effect of DlK2R-HA . In the case of Dl-HA , this had been achieved through co-expression with Notch ( Klein et al . , 1997; Doherty et al . , 1996 ) ( Figure 3M , yellow arrow , compare with Figure 2E ) . When co-expressed with Notch , DlK2R-HA ( in 51C ) produced the same phenotype . It induced a band of strong ectopic expression of Wg and Gbe + Su ( H ) and the endogenous expression of Wg along the D/V boundary was not interrupted ( Figure 3N–P , yellow arrow in N ) . This finding shows that the ability of DlK2R-HA to activate the Notch pathway is partly obscured by its increased cis-inhibitory abilities . The co-expression of Dl-HA and DlK2R-HA with Notch in mib1 mutants resulted in the induction of a broad stripe of expression of Gbe + Su ( H ) ( Figure 3Q–T ) . However , induction of Wg expression was severely reduced or absent . Although we did not examine the mechanistic basis of this defect in depth , a possible explanation for the reduction of activity of the Dl variants in mib1 mutants is the loss of the wing specific DS-loop . In line with this notion , we found that the ability of Dl-HA and DlK2R-HA to induce expression of endogenous Dl in mib1 mutant discs was severely reduced ( Figure 3—figure supplement 4A–G ) . Therefore , most of the Notch activity in mib1 mutants is generated by expression of the exogenous Dl constructs . It is probably too weak to consistently activate the expression of Wg . Using the MARCM technique , we co-expressed DlK2R-HA with Notch in Dl Ser double mutant cells and found that it induced expression of Wg in the mutant cells ( Figure 3U , V; arrows ) . Moreover , we expressed DlK2R-HA in mib1 mutant wing discs and found that the activity marker Gbe + Su ( H ) was induced also in cells of Dl Ser double mutant clones ( Figure 3W , X , arrow ) . These results indicate that the activation of the Notch pathway by DlK2R-HA is independent of the endogenous ligands . In summary , the results confirm that DlK2R-HA has a residual activity despite the loss of all Ks in its ICD . At least part of this activity is independent of mib1 , and probably ubi-independent . In order to test whether DlK2R-HA is endocytosed , we monitored its sub-cellular distribution in comparison to the endosomal marker Rab7-YFP , which marks maturing endosomes ( MEs ) ( Figure 4A–F; see also M and M ) . Dl-HA was found at the apical membrane ( Figure 4A , arrowhead ) and in Rab7-positive MEs ( Figure 4A–C , arrow ) , as previously described . We found that DlK2R-HA was distributed similarly to Dl-HA at the apical membrane ( Figure 4D , arrowhead ) and in Rab7 positive MEs ( Figure 4D–F , arrow ) and localised with similar frequency at Rab7 positive MEs as Dl-HA ( Figure 4G ) . Thus , DlK2R-HA is endocytosed into approximately the same number of endosomes . This is in agreement with previous findings , which show that endocytosis of Dl is not affected in a mib1 mutant background ( Le Borgne et al . , 2005 ) . Although the overall distribution in endosomes may not be affected by mib1 activity , we had earlier detected significant changes in endocytic rate of Dl upon compromising the intracellular motifs mediating Mib1 interaction ( Daskalaki et al . , 2011 ) . We therefore decided to assay possible stability differences between Dl-HA and DlK2R-HA . For this purpose , we ubiquitously expressed Dl-HA and DlK2R-HA for a limited time with tubGal4 tubGal80ts and evaluated their relative concentrations with Western blot analysis ( eight independent blots ) . This analysis revealed that DlK2R-HA is expressed at an approximately 2 . 5 x higher level than Dl-HA ( Figure 4I , J ) . Thus , the Ks in the ICD have a detectable negative effect on the stability of Dl ( Daskalaki et al . , 2011 ) . To pinpoint the phase in the endosomal pathway where the delay in degradation of DlK2R-HA is caused , we measured the fluorescence intensity of the HA staining of the Rab7 positive MEs shown in Figure 4A–F ( see M and M for details ) . We found that the summed fluorescence intensity in the MEs was significantly higher in discs that expressed DlK2R-HA compared to discs that expressed Dl-HA ( Figure 4H ) . This suggests that DlK2R-HA accumulates in MEs and is less efficiently transported to the lysosome than Dl-HA . Next , we expressed Dl-HA and DlK2R-HA in a pulse-chase experiment using a combination of ciGal4 and tubGal80ts , and measured the time required for their degradation . After 16 hr of expression all cells of the anterior compartment reliably expressed either Dl-HA or DlK2R-HA ( Figure 4—figure supplement 1A–E ) . After a chase of 24 hr , Dl-HA was completely degraded ( Figure 4—figure supplement 1C , D ) . At this time point , we still observed high DlK2R-HA levels , also at the plasma membrane ( Figure 4—figure supplement 1G , H , arrows in H ) . Low levels of DlK2R-HA were observed even after 36 hr of chase ( Figure 4—figure supplement 1I , J ) . These results further confirm that the degradation of DlK2R-HA is delayed and indicate that at least two effects cause this delay: ( 1 ) Inefficient endocytosis and ( 2 ) Less efficient transport to the lysosome after endocytosis . Altogether , the results are consistent with the documented requirement of ubi at several steps in the endosomal pathway . Dli1/2 is a randomly inserted variant that lacks the binding sites for Neur ( ICD1 ) and Mib1 ( ICD2 ) ( Figure 3A ) . As a consequence , it is not ubiquitylated by these E3-ligases ( Daskalaki et al . , 2011 ) . Based on the expression of Wg , the expression of Dli1/2 with ptcGal4 caused a phenotype that resembled that of DlK2R-HA , including the increased cis-inhibition judged from the large gap inflicted on the endogenous Wg stripe ( Figure 5A , compare with Figure 3B ) . The quantification of the gap in Wg expression confirmed the increase in cis-inhibition ( Figure 3—figure supplement 2 ) . This resemblance combined with our results raised the possibility that also Dli1/2 has residual activity , although a direct comparison of the two variants is not possible due to the unknown genomic location of Dli1/2 . We monitored the induction of the more sensitive Gbe + Su ( H ) upon expression of Dli1/2 . Indeed , we observed induction of ectopic expression of Gbe + Su ( H ) in wildtype , as well as in mib1 mutant wing discs ( Figure 5B–F , arrowhead in B , arrow and arrowhead in E ) . Moreover , its co-expression with Notch induced a broad stripe of ectopic expression of Wg and Gbe + Su ( H ) ( Figure 5G–I , arrow and arrowhead in G , H ) . Likewise , co-expression of Dli1/2 with Notch in mib1 mutant discs strongly induces expression of Gbe + Su ( H ) ( Figure 5K , L ) , but not of Wg ( Figure 5J , L ) . The activation of Wg expression achieved through co-expression of Notch and Dli1/2 in the wildtype is independent of the endogenous ligands , since it is also observed in Dl Ser double mutant MARCM clones ( Figure 5M , N , arrow ) . The observed phenotypes are similar to those of expression of DlK2R-HA ( compare with Figure 3 ) . If another E3-ligase could ubiquitylate Dli1/2 , we would expect Dli1/2 to signal more strongly than DlK2R-HA , since we know that ubi enhances the activity of Dl . The similarity of the phenotypes of Dli1/2 and DlK2R-HA suggests that no other E3-ligase is involved in Dl/Notch signalling in the wing disc . The results further support the notion that Dl has a residual activity that is independent of Mib1 and Dl-ICD ubi . This ubi-independent activity is likely to cause the weaker than expected mib1 mutant phenotype . The increase in cis-inhibition observed for DlK2R-HA and Dli1/2 also uncovers a so far little appreciated involvement of ubi of the ICD of Dl in this process . Thus , Mib1 might be required to overcome cis-inhibition through ubi of the Dl-ICD . The results so far suggest that the loss of Notch activity in mib1 mutants may result from excessive cis-inhibition caused by endogenous levels of the DSL ligands . Cis-inhibition is determined by the ratio between the concentrations of DSL ligands versus Notch ( Klein et al . , 1997 ) , implying a mechanism whereby DSL ligands and Notch expressed in the same cell engage in a non-productive stoichiometric complex ( see Figure 6A ) . To determine the extent to which Notch signalling could be augmented in mib1 discs by relieving cis-inhibition , we made several manipulations that imbalance the stoichiometry between Notch and its ligands . We induced Dl Ser double mutant clones in wildtype and mib1 mutant wing discs . It has been shown that the loss of function of Dl and Ser in cells of the wing pouch and the eye results in the activation of the Notch pathway by relieving cis-inhibition , when the mutant cells are located adjacent to ligand expressing wildtype cells ( Micchelli and Blair , 1999; Miller et al . , 2009 ) ( Figure 6A ) . Dl and Ser are widely expressed in the notum and in combination their patterns cover most of the notal area ( Troost et al . , 2015 ) ( Figure 6—figure supplement 1A–D ) . Analysis of Dl Ser mutant clones revealed that cis-inhibition occurs also throughout a large part of the notum of normal wing discs: we observed activation of the Notch pathway in mutant boundary cells , indicated by the ectopic activation of Gbe + Su ( H ) ( Figure 6B–D , arrowhead in C , D ) . Thus , cis-inhibition is not restricted to wing and eye development , but appears to be a frequently used mechanism to regulate Notch activity . We next tested whether relief of cis-inhibition can also be observed in mib1 mutant discs . Indeed , strong activation of Gbe + Su ( H ) in mutant cells at the boundary of Dl Ser clones was observed in the notum , indicating that the loss of the ligands can result in strong activation of the Notch pathway also in the absence of mib1 function ( Figure 6E–J , arrows ) . As expected , activation occurred only in areas where the ligands were expressed ( Figure 6G–I ) . Note , that this Notch activity in Dl Ser mutant cells is induced by the activity of ligands in adjacent ligand-expressing cells , although these are devoid of the function of mib1 function . Since Ser is not able to activate the Notch pathway in the absence of mib1 function , even when over-expressed ( see Figure 2M , N ) , the signalling must be mediated by Dl . Thus , Dl can signal in the absence of mib1 function without being over-expressed and ubiquitylated . The results also show that at least part of the loss of Notch activity in mib1 mutants is caused by cis-inhibition through the globally expressed ligands and supports the conclusion that Mib1 is required to overcome cis-inhibition during normal signalling . The experiments so far were performed with the artificial reporter construct Gbe + Su ( H ) . To confirm the results also with the promoter of an endogenous target gene , we monitored the expression of E ( spl ) mß-lacZ , which is expressed in a similar global manner as Gbe + Su ( H ) ( Cooper et al . , 2000 ) ( Figure 6—figure supplement 1E–G ) . E ( spl ) mß behaved like Gbe + Su ( H ) , as its expression was initiated in mutant boundary cells of Dl Ser double mutant clones in mib1 mutant discs ( Figure 6—figure supplement 1I–L , arrowhead ) . Cis-inhibition is mutual: the non-productive cis DSL-Notch complex inhibits the activity of both receptor and ligand ( Becam et al . , 2010 ) . We therefore next tested whether removing Notch expression in a group of cells would release the ligands from cis-inhibition and would result in trans-activation of the Notch pathway in adjacent cells . We created this situation by expression of a Notch-RNAi construct that efficiently depletes Notch expression if expressed with ciGal4 in the anterior compartment ( Figure 6M–O ) . In this situation , anterior cells that lack Notch were exposed to posterior Notch expressing cells at the A/P boundary . Consequently , the endogenous ligands of the anterior cells released from cis-inhibition should signal to posterior boundary cells . Indeed , this was observed in mib1 mutant wing imaginal discs: The expression of Gbe +Su ( H ) was induced in adjacent posterior boundary cells ( Figure 6O , arrows ) . This result confirms that endogenous Dl can signal in the absence of mib1 function . It also indicates that Dl , not engaged in cis-interaction , can to some extent out-compete cis-interacting ligands in adjacent cells for interaction with Notch . In mib1 mutant wing discs residual Gbe + Su ( H ) expression ( S3 ) is observed ( Figure 1L , arrow ) . S3 expression is lost in mib1 discs if the notum lacks expression of the ligands in this area , indicating that S3 is induced by Dl-dependent Notch signalling ( Figure 6K , L , arrow ) . However , expression of S3 in mib1 mutant discs is independent of Neur , as it is still detected in neur mutant clones ( Figure 6P , Q ) . We observed a S3-like residual expression domain of E ( spl ) mß-lacZ in mib1 mutants ( Figure 6—figure supplement 1H , J , arrow ) . The depletion of Notch with Notch-RNAi also abolished the expression of S3-like , indicating that it is dependent on Notch ( Figure 6—figure supplement 1M–O ) . Hence , we have identified a region of Notch signalling in the late larval notum that is induced by Dl independently of Mib1 and Neur and probably of ubi altogether . DSL signalling during wing development depends on Mib1 . The alternative E3-ligase for Dl , Neur , is only detected in sensory organ precursors , a small late arising minority of cells in the wing pouch . Hence , the experiments presented so far investigated the connection of Mib1 with the Ks in the ICD of Dl . In order to test the connection with Neur , we co-expressed it together with the Dl variants in a mib1 null mutant background . Expression of Neur alone had no effect on the expression of Wg in normal discs ( Figure 7A ) . Co-expression of Neur with Dl-HA in wildtype wing discs induces strong ectopic expression of Wg similar to that already seen with Dl-HA alone ( Figure 7B , arrows; compare with Figure 2E ) . However , we consistently observe a loss of the anterior stripe of Wg expression in the dorsal half of the pouch ( Figure 7B , arrowhead ) . As previously reported , expression of Neur in mib1 mutants re-established the expression of Wg along the D/V boundary in the area of the ptcGal4 domain ( Le Borgne et al . , 2005; Wang and Struhl , 2005 ) ( Figure 7C , arrow ) . The co-expression of Neur and Dl-HA in mib1 mutant wing discs resulted in induction of an ectopic expression domain of Wg that was comparable to that observed in wildtype discs ( Figure 7D , arrows; compare with B ) . To our surprise , co-expression with Neur dramatically enhanced also the weak activity of DlK2R-HA in normal and also mib1 mutant wing discs . Strong ectopic expression of Wg was observed in both genetic backgrounds ( Figure 7E , F , arrows ) . We confirmed the enhancement of DlK2R-HA by Neur with an independently generated second chromosome UAS neur insertion line ( Lai and Rubin , 2001 ) ( Figure 7G , arrows ) . Note , that the strength of Wg expression achieved by co-expression of Neur with DlK2R-HA or Dl-HA in mib1 mutants was comparable ( compare Figure 7D with F ) . This suggests that the Ks in the ICD of Dl are of minor importance for its activation by Neur . Moreover , cis-inhibition of DlK2R-HA is reduced to a level comparable to Dl-HA ( Figure 7E , compare with B ) . Using MARCM , we found that the activation of the Notch pathway by the co-expression of Neur and DlK2R-HA occurred in the absence of endogenous Dl and Ser ( Figure 7H , I , arrow ) . This excludes the possibility that the endogenous ligands are involved in the induction of the ectopic Notch activity . We co-expressed Mib1 together with DlK2R-HA and found that in contrast to Neur , co-expression of Mib1 failed to modulate the activity of DlK2R-HA ( Figure 7J , yellow arrow , compare with E ) . Thus , the enhancing effect on DlK2R-HA is a unique property of Neur . NeurΔNHR1 is a variant that is unable to bind to the ICD of Dl , because it lacks the necessary NHR1 binding domain ( Commisso and Boulianne , 2007 ) . In contrast to Neur , NeurΔNHR1 failed to rescue the expression of Wg along the D/V boundary in mib1 mutant discs and did not enhance the signalling activity of DlK2R-HA and Dl-HA in wildtype and mutant discs ( Figure 7K–N ) . Moreover , co-expression of Neur did not change the phenotype of Dli1/2 , which lacks the Neur binding site in its ICD ( Figure 7O , compare with Figure 5A ) . Hence , Neur must directly bind to the ICD of Dl-HA and DlK2R-HA to promote their signalling activity and reduce cis-inhibition of DlK2R-HA . The results reveal fundamental differences between Neur and Mib1 in the activation of Dl . We next asked whether Neur requires its E3-ligase activity for activation of Dl-HA and DlK2R-HA . To do so , we used a variant that lacks the RF , which catalyses the ubi reaction ( NeurΔRF ) ( Pavlopoulos et al . , 2001 ) . The subcellular localisation of this variant resembles that of Neur , indicating that the RF is not necessary for localisation of Neur ( Yeh et al . , 2001 ) . We have confirmed the correct localisation for the constructs used in our experiments ( Figure 7—figure supplement 1 ) . Similar to Neur , expression of NeurΔRF alone did not affect expression of Wg along the D/V boundary and did not significantly affect the activity of Dl-HA in wildtype wing discs ( Figure 7P , Q ) . However , unlike Neur , it does not abolish the dorsal anterior stripe of ectopic Wg expression observed if Dl-HA is expressed alone ( Figure 7Q , arrowhead , compare with B ) . It was also unable to rescue Wg expression in mib1 mutants ( Figure 7R ) . Nevertheless , NeurΔRF strongly enhanced the activity of Dl-HA in mib1 mutant discs ( Figure 7S ) . In comparison to Neur , a significant difference in the strength of the ectopic expression of Wg was observed between wildtype and mib1 mutant discs ( compare Figure 7Q , S and B–D ) , indicating a requirement of the RF for full activity . However , the result indicates that Neur can activate Dl in a manner independently of catalysing ubi of its ICD or other components of the endocytic machinery . NeurΔRF was also able to enhance the activity of DlK2R in wildtype discs , indicated by the induction of a stripe of ectopic expression of Wg in posterior boundary cells ( Figure 7T , arrows ) . A weak stripe was induced if both proteins were expressed even in mib1 mutant discs ( Figure 7U–W , arrows ) . Altogether , the results show that Neur , in contrast to Mib1 , can activate Dl in an ubi-independent manner . However , comparison revealed that the induction of Wg expression upon co-expression of Dl and NeurΔRF was stronger than that of DlK2R and NeurΔRF ( compare Figure 7S with V ) . In case of Neur the induction of Notch activity was comparable . This suggests that the Ks in the ICD of Dl are important for full activity , especially when Neur lacks its ability to ubiquitylate . The reason for this paradoxical requirement is not clear at the moment . A possible explanation could be a contribution of the Ks to the conformation of the ICD in a manner that cannot be completely replaced by the introduced Rs . Since Neur can stimulate Dl activity even in the absence of Ks in its ICD , we tested the activity of the various Dl variants in contexts , where Notch signalling is more dependent on endogenous Neur . We had shown earlier that two such developmental processes are ( a ) the asymmetric cell division of ganglion mother cells ( GMCs ) and ( b ) the selection of the sensory organ precursor ( SOP ) of macrochetae in the larval notum . During GMC asymmetric division the daughter cells rely on Dl/Notch signalling to turn on Hey expression in one of the two siblings ( Monastirioti et al . , 2010 ) ( Figure 8A–B' ) . neur loss of function severely compromises this signalling , whereas Mib1 only plays an accessory role ( Monastirioti et al . , 2010 ) : 80% of neur-clones lose Hey expression ( Figure 8C , C’ ) , which rises to 100% in a mib1 background . However , mib1 in the presence of Neur shows no loss of Hey . Using the MARCM system , we substituted endogenous Dl by DlK2R-HA or Dli1/2 in the developing late larval CNS . The clones were scored in the central brain and ventral nerve cord . As a control , we confirmed that removal of both endogenous ligands results in 96% of the lineages ( clones ) being Hey negative ( Figure 8D , D’ ) and this is fully rescued ( 0% Hey negative ) by UAS Dl-HA , but not Dli1/2 ( Figure 8E–F ) . When we expressed DlK2R-HA in Dl Ser MARCM clones , we obtained a full rescue of Hey expression; only 4% of lineages were Hey negative ( Figure 8G , G’ ) . Thus , unlike the wing pouch , where loss of Ks severely compromises Dl activity , in this context DlK2R-HA was able to sustain Notch signalling . Since Neur is needed but Ks are dispensable , we wondered whether the catalytic activity of Neur is necessary . We therefore generated neur mutant MARCM clones expressing NeurΔRF . These showed complete rescue of Hey ( 100% of clones are positive; Figure 8H , H’ ) . Taken together with the DlK2R-HA result , this supports a model where ubi is not required for Dl signalling in the context of GMC asymmetric cell division . Another Neur-dependent Notch signalling context is the selection of the sensory organ precursor ( SOP ) of the bristle sensillum in the notum . This occurs within proneural clusters , which are defined by the expression of proneural genes by Dl induced Notch signalling ( Modolell and Campuzano , 1998 ) ( Figure 9A ) . We recently introduced a MARCM based test system to test variants of Dl in this selection process ( Pitsouli and Delidakis , 2005 ) . MARCM clones double mutant for Dl and Ser cause the development of clusters of SOPs instead of single ones , termed neurogenic phenotype ( Figure 9A ) . This phenotype can be revealed by anti Hindsight ( Hnt ) staining , which labels mature SOPs ( Figure 9B , C , arrows ) . The neurogenic phenotype is completely suppressed by expression of Dl-HA ( Pitsouli and Delidakis , 2005 ) ( Figure 9D , E , arrows ) . Similarly , DlK2R-HA strongly suppressed the Dl Ser mutant neurogenic phenotype ( Figure 9F , G , arrows ) . This result suggests that the ubi-independent mechanism is sufficient for selection of most SOPs . To test whether DlK2R-HA requires the function of Neur during the selection process , we repeated the experiments with DlK2Ri1 ala-HA that lacks the Neur binding site . DlK2Ri1 ala-HA failed to suppress the neurogenic phenotype in Dl Ser double mutant MARCM clones , indicating that direct binding of Neur is required for the selection of the SOP by DlK2R-HA ( Figure 9H , I , arrows ) . Lqf binds ubiquitylated cargo by two UIMs . In order to further evaluate the role of ubi of the DSL ligands during SOP selection , we used a “reading “defective variant of Lqf in which UIM1 was mutated and UIM2 deleted ( Xie et al . , 2012 ) . This GFP-tagged variant , termed LqfUIM13E/3A-ΔUIM2-GFP is controlled by the endogenous lqf promoter and cannot recognise ubiquitylated cargo ( Xie et al . , 2012 ) . As previously shown , LqfUIM13E/3A-ΔUIM2-GFP partially rescues lqf null mutant flies , which normally die during embryogenesis , to the pharate adult stage ( Xie et al . , 2012 ) . The pharate adults displayed severe patterning defects that were similar to those described for mib1 mutants . The phenotype will be described in detail elsewhere . Importantly , the rescued flies displayed a nearly normal bristle pattern with only the occasional duplicated large bristle and a higher density of small bristles ( Figure 9J–L , arrowhead in K , L ) . A similar bristle phenotype was described for mib1 mutants ( Le Borgne et al . , 2005 ) . The analysis of the wing imaginal discs of the LqfUIM13E/3A-ΔUIM2-GFP rescued lqf flies revealed a nearly normal pattern of SOPs , just like in mib1 mutants ( Figure 9M , O , compare with A ) . Altogether , these data confirm that ubi-independent , but Neur dependent DSL signalling prevails during SOP selection . Note , that complete loss of lqf function caused a neurogenic phenotype in mib1 mutant discs ( Figure 9P , arrow ) , indicating that Lqf ( but not its UIMs ) is required for the ubi-independent selection of the SOP . Previous work established that the activity of the members of the DSL family depends on the function of the E3-ligases Neur and Mib1 that sends them into Lqf/Epsin dependent endocytosis . This activating endocytosis is different from bulk endocytosis , which is - in the case of Dl - not dependent on Mib1 or Lqf ( Le Borgne et al . , 2005; Wang and Struhl , 2005 ) . In order to account for these dependencies on E3-ligases , it has been suggested that ubi of the DSL protein ICDs by Neur and Mib1 initiates a special endocytosis event ( Weinmaster and Fischer , 2011 ) . This event either creates a pulling force that is essential for Notch S2 cleavage and subsequent ecto-domain shedding , or sends the ligands through a recycling pathway where they mature into the active form . However , it is not clear why in a given process only one of the two E3-ligases is required or why Neur strongly affects the endocytosis of Dl , but not of Ser , whereas Mib1 affects the endocytosis of the ligands in the opposite way ( Le Borgne et al . , 2005; Wang and Struhl , 2005; Le Borgne and Schweisguth , 2003 ) . Here , we further assessed the role of Ks and ubi during activation of Dl , using different tissues of Drosophila as test systems . One main finding of our work is that the activity of Dl has three components: One is dependent on Mib1/Neur/E3-ligase activity and probably ubi , the second is Neur-dependent , but ubi-independent and finally the third is independent of any of these factors . Our analysis suggests that the ubi-independent activities probably account for the residual Notch activity observed in mib1 null mutants . We identified two normal developmental processes , GMC division and SOP selection , which rely on the ubi-independent/Neur-dependent activity of Dl . The only example of a ubi-independent/Neur-independent case of Dl signalling is the S3 notum region expression of Gbe + Su ( H ) and the endogenous target gene E ( spl ) mß . Unfortunately , it is not known in which developmental process this expression domain is involved . The latter ubi-/Neur-independent pathway appears to be restricted to Dl as we did not observe ectopic activation of Gbe + Su ( H ) upon Ser expression in mib1 mutants . Recent evolutionary studies suggest that the ubi-independent signalling mode of Dl might be the evolutionary ancient one . Trichoplax adherens is the only known member of the basal metazoan phylum Placozoa . It possesses all crucial elements of the Notch pathway members except Mib1 and Neur ( Gazave , 2009 ) . The E3-ligases appear to be later additions . Moreover , the only ligand present in the Trichoplax genome belongs to the Dl family . In the light of our results , it is possible that Notch signalling in Trichoplax occurs solely by the probably more ancient ubi-independent mechanism . The Ser-like ligands were probably only introduced later in evolution , after the recruitment of the E3-ligases . Besides the discovery of the ubi-independent modes , we here confirm that the Ks in the ICD of Dl and Ser are important for their signalling abilities . In addition , our results indicate that the loss of the Ks in Dl also results in a defect in ligand degradation and delayed endocytosis , suggesting the Ks in the ICD are required for the correct degradation . This is in good agreement with our previous finding that the endocytosis of Dli1/2 that lacks the Mib1 and Neur binding sites is delayed ( Daskalaki et al . , 2011 ) . We did not find significant differences in the number of Rab7 positive endosomes that are also positive for Dl-HA or DlK2R-HA , however the total amount of DlK2R-HA on these endosomes was increased compared to Dl-HA . Together with the persistence of DlK2R-HA in the plasma membrane , we favour a model where ICD ubi enhances trafficking of Dl-HA at many stations along the endocytic route and ultimately increases its degradation . A novel finding is that Neur and Mib1 activate Dl by different mechanisms . In contrast to Mib1 , Neur can activate DlK2R-HA to a level that is very similar to that of Dl-HA upon co-expression , also in the absence of mib1 function . This activation requires direct binding of Neur to the ICD of Dl . This indicates that Neur can strongly activate Dl in a manner independently of ubi of its ICD . The findings are in agreement with the finding that Neur is involved in processes , which appear to depend on ubi-independent Dl signalling , e . g . the selection of neural fates in the brain and the PNS . Previous experiments suggested that Mib1 cannot rescue the embryonic defects of neur mutants ( Le Borgne et al . , 2005 ) . It is likely that Mib1’s lack of this second , ubi-independent function is the reason for this failure . One possibility for the activation of Dl by Neur might be that it acts as an adapter that connects Dl to the endocytic machinery in the identified developmental processes . In this way Dl could be incorporated into endocytic vesicles without ubi . This adaptor function of Neur is consistent with the observation that it frequently co-traffics with Dl and that expression of Dl can dramatically change the subcellular distribution of Neur ( Daskalaki et al . , 2011; Skwarek et al . , 2007 ) . Recent work indicates that this adapter function might be more widespread among E3-ligases than anticipated since it has been shown that also the E3-ligase Suppressor of deltex , which plays an important role in endosomal trafficking of Notch , induces the endocytosis of Notch in an ubi-independent manner ( Yamada et al . , 2011 ) . The selection of the SOP occurs within groups of cells , termed proneural clusters , which are defined by the expression of proneural genes ( Modolell and Campuzano , 1998 ) . It has previously been shown that this selection occurs within a subgroup of the proneural cluster ( Troost et al . , 2015 ) . It is termed the Neur group , since the cell that first expresses Neur will develop into the SOP and inhibits its neighbours by increased Dl signalling . The requirement for Neur to increase Dl signalling is puzzling since the cells of the subgroup also express Mib1 . The discovery of the ubi-independent activation of Dl by Neur provides an explanation why Neur might be required for correct SOP selection , since it could boost Dl signalling in a different manner than Mib1 . Interestingly , our results indicate that also Lqf contributes to SOP selection by an ubi-independent function , as the process is to a large extent rescued by LqfUIM13E/3A-ΔUIM2-GFP in lqf mutants . Hence , Lqf might be part of the Neur adapter complex . Another so far not well-appreciated finding is that the ICD contributes to cis-inhibition . It has previously been shown that the ECD of the ligands is involved in cis-inhibition ( Fleming et al . , 2013; Glittenberg et al . , 2006; Li and Baker , 2004 ) . Here , we found that the cis-inhibitory ability of Dl strongly increases if the Ks in its ICD are lost . We also showed that loss of mib1 results in a global increase of cis-inhibition by endogenously expressed DSL ligands . Thus , it appears that both parts of Dl are required for cis-inhibition: The ECD appears to execute the necessary physical interactions with Notch , while the ICD adjusts the strength of cis-inhibition via recruiting E3-ligases and promoting ubi , which suppresses cis-inhibition . How might Dl/Notch cis-inhibition be suppressed by ubi of the ligand ICD ? The ubiquitin moiety alone might already promote the separation of the Dl/Notch cis-pair on the cell surface . Alternatively , ubi enables binding of Epsin/Lqf via the UIMs , which separates the pair . The similarity between the phenotype of mib1 mutants and the LqfUIM13E/3A-ΔUIM2-GFP rescued lqf mutants favours the second possibility . Another possibility is that ubi/Lqf sends Dl into the recycling pathway as previously suggested ( Wang and Struhl , 2004 ) . Separation of the Dl/Notch ECDs is a likely event during this recycling thanks to the acidic pH in the endosome lumen . It is possible that the majority of newly synthesised Dl ( and probably Ser ) on its way to the cell surface is initially engaged in cis-interaction with Notch . Unless their separation is promoted by ubi , only a minor amount of Dl at the surface would be free to signal . This diminished signalling would justify the reduced ( but not eliminated ) Notch activity , observed in mib1 mutants ( e . g . S3 expression of Gbe + Su ( H ) ) . We found that co-expression of Neur also reduces the cis-inhibitory effect of DlK2R-HA and at the same time strongly enhances its signalling activity . To do so , Neur must directly bind to the ICD of DlK2R-HA . These results raise the possibility that the binding of Neur might directly separate the Dl/Notch cis-pair or , as a possible endocytic adapter , stimulates endocytic recycling upon which the cis-pair is separated . Future work is needed to clarify whether the cis-pair separation is the only mechanism through which ubi activates DSL signalling , or whether it also contributes to the pulling force which seems to be a prerequisite for Notch-activation in several contexts . mib11 , mib12 , mib1EY09780 ( Le Borgne et al . , 2005a ) , lqfL71 ( Xie et al . , 2012 ) and lqf1227 FRT2A ( Wang and Struhl , 2004 ) , PsnC2 FRT2A ( Struhl and Greenwald , 1999b ) , Dlrev10 SerRX82 FRT82B ( Micchelli et al . , 1997 ) , HE31 ( Bang and Posakony , 1992 ) . Ser- ( RRID:BDSC_59284 ) and Dl-MIMIC-GFP ( RRID:BDSC_59819 ) ( Nagarkar-Jaiswal et al . , 2015 ) . Neur1 ( Commisso and Boulianne , 2007 ) . All used alleles are null alleles . UAS constructs: UAS neur ( Weinmaster and Fischer , 2011; Lai et al . , 2001 ) , UAS neur-myc ( third chromosome ) and UAS neurΔRF-GFP ( Pavlopoulos et al . , 2001 ) , UAS neurΔNHR1-V5 ( Commisso and Boulianne , 2007 ) , UAS N-LV ( Loewer et al . , 2004 ) , UAS mib1 ( Lai et al . , 2005 ) , UAS N-RNAi ( RRID:BDSC_7078 ) , UAS Dli1 and UAS Dli1/2 ( Daskalaki et al . , 2011 ) , UAS Dlstu ( a gift of D . Horowicz and D . Henrique ) . Further stocks: ptcGal4 ( RRID:BDSC_2017 ) , tubRab7-YFP ( Marois et al . , 2006 ) , Gbe+Su ( H ) -lacZ ( Furriols and Bray , 2001 ) , tubGal4 tubGal80ts ( RRID:BDSC_5138 , RRID:BDSC_7018 ) , ciGal4 ( Croker et al . , 2006 ) and vgBE-GFP ( Zecca and Struhl , 2007 ) . Dl-HA was generated by introduction of a synthesized fragment from the NdeI restriction site of the ICD onwards . The DlK2R-HA was generated by replacing the ICD of Dl-HA by a synthesised ICD in which all lysines are replaced by arginines . Constructs were cloned in pUAST attB and inserted into the landing site at 51C . Gene synthesis was performed by GenScript . DlK2R-HA was used to generate DlK2Ri1 ala-HA ( NEQNAV to A , see [Fontana and Posakony , 2009] ) and Dli2 ala-HA ( IKNTWDK to A ) by site directed mutagenesis . All constructs were sequenced before injection and subsequently inserted into the attP landing site at 51C . The LqfUIM13E/3A-ΔUIM2-GFP attB construct was obtained from J . Fischer ( described in [Xie et al . , 2012] ) and inserted into the attP22A landing site . Transgenesis was partly performed by BestGene Inc . The necessary primers were purchased from Sigma-Aldrich . MARCM and conventional clones were induced during the first larval instar stage ( 24–48 hr after egg laying ) . Wing imaginal discs were dissected from third instar larvae and boiled in Laemmli buffer at 95°C for 10 min . Each lysate contained at least 10 discs . Proteins were separated by SDS-PAGE gel electrophoresis and blotted according to standard protocols . Antibodies used: HA ( anti-HA , High Affinity , Roche ( 3F10 ) RRID:AB_390919 ) , alpha-Tubulin ( Sigma-Aldrich T5168 , RRID:AB_477579 ) , HRP-conjugated secondary antibodies were purchased from Jackson Immuno Research . Antibody staining was performed according to standard protocols ( Klein , 2006 ) . Antibodies used: anti-Wg antibody ( DSHB Iowa RRID:AB_528512 ) , anti-ß-Gal ( Cappel/MP Biomedicals RRID:AB_2313831 ) , anti-Rab7 ( Tanaka and Nakamura , 2008 ) ( RRID:AB_2569808 ) . Fluorochrome-conjugated secondary antibodies were purchased from Invitrogen/Molecular Probes . Images were obtained with a Zeiss AxioImager Z1 Microscope equipped with a Zeiss Apotome or a Leica SP6 confocal microscope . For quantitative analysis of Dl-HA/DlK2R-HA in Rab7 positive vesicles , the image analysis software ‘Imaris’ ( RRID:SCR_007370 , Bitplane , Zurich , Switzerland ) was used . Z-Stacks of imaginal discs were acquired with a Zeiss AxioImager Z1 Microscope equipped with a Zeiss Apotome , applying the same microscopy hardware settings ( e . g . exposure time ) to ensure reproducibility between datasets . Individual Rab7 vesicle volumes were identified utilizing the ‘spot feature’ of Imaris with an initial size of 0 . 4 µm . Regrowing of volume , based on the local contrast diameter threshold was allowed later on . A quality threshold was placed on resulting spots to minimise background volumes . For further analysis only vesicles were included , which contain at least voxels with a Dl-HA/DlK2R-HA signal intensity threshold of 1 . Finally , the absolute cargo of Dl-HA/DlK2R-HA signal in Rab7 vesicles was defined as the median of the absolute signal sum per vesicle . The average of these values over several biological replicates is displayed in the graph . Statistics were applied as described above to calculate significance levels . The procedure of vesicle volume definition , thresholding and downstream analysis was the same for all datasets . In total 19373 ( Dl-HA ) and 15720 ( DlK2R-HA ) vesicles of 3 discs for each genotype were included in the analysis .	Cells use chemical signals to communicate , setting off chains of reactions known as signalling pathways . One key signalling pathway , thought to be required for the development of all animals , is called Notch . In fruit flies , signal proteins known as Delta and Serrate activate the Notch pathway by binding to receptors on the outside of the cell . To do so , the signal proteins first need to be activated themselves . Two enzymes known as Mindbomb1 and Neuralized activate Delta and Serrate . Both enzymes add a small unit called ubiquitin to specific locations on the signal proteins , but the effect that ubiquitin has on Notch signalling is not yet fully understood . Berndt , Seib , Kim et al . have now examined fruit flies that had a variety of genetic mutations . These included some flies that could produce mutant versions of the Serrate and Delta proteins that lacked the locations to which ubiquitin normally attaches . The results of the experiments reveal that Delta requires ubiquitin , Mindbomb1 and Neuralized to work at full capacity . However , Delta could still perform some of its roles without ubiquitin . Neuralized and Delta can partner up to send some signals independently of ubiquitin , and Delta can even send some signals on its own . Serrate , on the other hand , does not work at all without ubiquitin . The results presented by Berndt et al . help us to understand the role that ubiquitin plays in activating Notch signalling . Further work that builds on these findings could help to shed light on how uncontrolled Notch activation can contribute to a variety of diseases , including cancer , cardiovascular diseases and multiple sclerosis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2017	Ubiquitylation-independent activation of Notch signalling by Delta
A parental cancer diagnosis is psychologically straining for the whole family . We investigated whether a parental cancer diagnosis is associated with a higher-than-expected risk of injury among children by using a Swedish nationwide register-based cohort study . Compared to children without parental cancer , children with parental cancer had a higher rate of hospital contact for injury during the first year after parental cancer diagnosis ( hazard ratio [HR] = 1 . 27 , 95% confidence interval [CI] = 1 . 22-1 . 33 ) , especially when the parent had a comorbid psychiatric disorder after cancer diagnosis ( HR = 1 . 41 , 95% CI = 1 . 08-1 . 85 ) . The rate increment declined during the second and third year after parental cancer diagnosis ( HR = 1 . 10 , 95% CI = 1 . 07-1 . 14 ) and became null afterwards ( HR = 1 . 01 , 95% CI = 0 . 99-1 . 03 ) . Children with parental cancer also had a higher rate of repeated injuries than the other children ( HR = 1 . 13 , 95% CI = 1 . 12-1 . 15 ) . Given the high rate of injury among children in the general population , our findings may have important public health implications . Cancer is not only a leading cause of morbidity and mortality among the affected patients , it is also increasingly recognized as a contributor to ill-health in their significant others ( Sjovall et al . , 2009; Visser et al . , 2004; Kazak et al . , 2005; Krauel et al . , 2012 ) . In Sweden , the number of newly diagnosed cancer patients has more than doubled during the last forty years and a considerable proportion of these patients are parenting minor children ( National Board of Health and Welfare , 2014 ) . A diagnosis of cancer in parents has repeatedly been shown to exert psychological and social stress in their children ( Visser et al . , 2004 ) . Coping with cancer may affect the parenting of both the ill and well parents , further impacting the behavioral and social adaptability of the children ( Faulkner and Davey , 2002 ) . In contrast to the relatively rich literature on behavioral and mental well-being of children living with a parent with cancer , few studies have so far addressed somatic health outcomes among these children . In a recent study , we reported that children of parents with cancer had a higher risk of death , both due to cancer and other causes ( Chen et al . , 2015 ) . Injury is the most common cause of hospital care among children and accounts for almost one million child deaths annually worldwide ( Peden et al . , 2008 ) . Sociodemographic , behavioral and psychosocial factors of both children and their family are known determinants of injuries among children ( Horwitz et al . , 1988 ) . For example , childhood injury has been associated with male sex , risk-taking behavior , lack of parental supervision as well as poor mental health of the parents ( Matheny , 1986; Schwebel et al . , 2011; Morrongiello et al . , 2006b; Peden et al . , 2008; McKinlay et al . , 2010 ) . To our knowledge , no study has however specifically addressed the impact of parental cancer diagnosis on the risk of child injury . To this end , we leveraged the nationwide population and health registers in Sweden to explore the association between parental cancer diagnosis and the risk of hospital contact for injury among children . During follow-up , 15 , 377 exposed children ( incidence rate: 52 per 1000 person-years ) and 548 , 488 unexposed children ( incidence rate: 46 per 1000 person-years ) had a first hospital contact for injury . Adjusting for only attained age and sex , the exposed children had a 4% higher rate of hospital contact for injury ( hazard ratio [HR] , 1 . 04; 95% confidence interval [CI] , 1 . 02–1 . 05 ) than the other children . After adjustment for all covariates , the association became stronger ( HR , 1 . 07 [95% CI 1 . 05–1 . 09] ) ( Table 2 ) . Approximately 17% of hospital contacts among the exposed children occurred during the first year after cancer diagnosis , corresponding to an incidence rate of 60 per 1000 person-years and a HR of 1 . 27 ( 95% CI 1 . 22–1 . 33 ) . The rate increment decreased during the second and third years , and became null after three years ( Table 2 ) . 10 . 7554/eLife . 08500 . 004Table 2 . Hazard ratios for hospital contact for injury among children with parental cancer compared to children without parental cancer . DOI: http://dx . doi . org/10 . 7554/eLife . 08500 . 004Any Time After Parental Cancer DiagnosisFirst Year After Parental Cancer DiagnosisCharacteristicsNo . of Children With a Hospital Contact for InjuryPerson-yearsHR ( 95%CI ) p ( Wald Test ) No . of Children With a Hospital Contact for InjuryPerson-yearsHR ( 95%CI ) p ( Wald Test ) No parental cancer548 , 48811 , 879 , 0751548 , 48811 , 879 , 0751Parental cancer15 , 377298 , 3021 . 07 ( 1 . 05-1 . 09 ) 2 , 67444 , 6001 . 27 ( 1 . 22-1 . 33 ) Time since cancer diagnosis ≤ 1 year2 , 67444 , 6001 . 27 ( 1 . 22-1 . 33 ) ——— >1 and ≤3 years3 , 85074 , 0871 . 10 ( 1 . 07-1 . 14 ) <0 . 001——— > 3 years8 , 853179 , 6151 . 01 ( 0 . 99-1 . 03 ) ——— Sex of the cancer parent Male6 , 554126 , 2771 . 08 ( 1 . 05-1 . 11 ) 0 . 481 , 16618 , 9171 . 32 ( 1 . 24-1 . 40 ) 0 . 13 Female8 , 823172 , 0261 . 06 ( 1 . 04-1 . 09 ) 1 , 50825 , 6831 . 24 ( 1 . 18-1 . 31 ) Tobacco-related cancer † No12 , 008233 , 8481 . 07 ( 1 . 05-1 . 09 ) 0 . 722 , 14235 , 0801 . 29 ( 1 . 24-1 . 35 ) 0 . 13 Yes3 , 36964 , 4541 . 08 ( 1 . 04-1 . 12 ) 5329 , 5201 . 20 ( 1 . 10-1 . 31 ) Alcohol-related cancer ‡ No10 , 464201 , 3891 . 08 ( 1 . 05-1 . 10 ) 0 . 301 , 74528 , 5251 . 30 ( 1 . 24-1 . 37 ) 0 . 16 Yes4 , 91396 , 9131 . 06 ( 1 . 02-1 . 09 ) 92916 , 0761 . 23 ( 1 . 15-1 . 31 ) Predicted 5-year relative survival rate < 20% §93118 , 8451 . 02 ( 0 . 95-1 . 10 ) 1603 , 0411 . 15 ( 0 . 98-1 . 35 ) 20-80%7 , 112136 , 0801 . 08 ( 1 . 06-1 . 11 ) 0 . 211 , 24320 , 7361 . 27 ( 1 . 19-1 . 35 ) 0 . 38 ≥ 80% ¶ 7 , 334143 , 3771 . 06 ( 1 . 04-1 . 09 ) 1 , 27120 , 8241 . 30 ( 1 . 23-1 . 38 ) Parental psychiatric comorbidity after cancer diagnosis ‖ No14 , 630285 , 6211 . 06 ( 1 . 05-1 . 08 ) 0 . 0012 , 61143 , 6631 . 27 ( 1 . 22-1 . 32 ) 0 . 45 Yes74712 , 6811 . 21 ( 1 . 12-1 . 31 ) 639381 . 41 ( 1 . 08-1 . 85 ) HR , hazard ratio; CI , confidence intervalAdjusted for attained age , sex , number of siblings , gestational age , mode of delivery and birth weight of the child , paternal age at child's birth , maternal age at child's birth , maternal smoking during early pregnancy , and the highest educational level of the parents . †Tobacco-related cancers include cancers in lung , oesophagus , larynx , pharynx , mouth , lip , salivary glands , tongue , stomach , urinary bladder , kidney , uterine cervix , colon and pancreas . ‡Alcohol-related cancers include cancers in liver , oral cavity , pharynx , larynx , oesophagus , colorectum and breast . §Including cancers in esophagus , liver , gall bladder , biliary tract , pancreas , lung and stomach . ¶Including cancers in lip , breast , corpus uteri , testis , skin , thyroid and other endocrine glands , and Hodgkin’s lymphoma . ‖Including depression , anxiety disorders , stress reaction and adjustment disorder . The association was not modified by the sex of cancer parent or by the expected survival of cancer; the association did not differ between smoking/alcohol-related cancers and other cancers ( Table 2; all p>0 . 05 ) . However , children whose cancer parent had developed a comorbid psychiatric disorder after diagnosis had a higher rate of childhood injury ( HR , 1 . 21 [95% CI 1 . 12–1 . 31] , compared with children whose cancer parent had no such disease ( HR , 1 . 06 [95% CI 1 . 05–1 . 08] ) ( p = 0 . 001 ) . As in the overall analysis , the rate increment in these analyses was more prominent during the first year after diagnosis ( Table 2 ) . The overall association was significantly stronger for boys than for girls ( p for interaction , < 0 . 001 ) ( Table 3 ) . When focusing on the first year following parental cancer , no statistically significant difference was however detected between boys and girls ( p = 0 . 17 ) . Neither the overall association nor the association during the first year after parental cancer was modified otherwise by age at follow-up or number of siblings of the child ( Table 3 ) . 10 . 7554/eLife . 08500 . 005Table 3 . Hazard ratios for hospital contact for injury among children with parental cancer compared to children without parental cancer , according to sex , age and number of full and half siblings of the child . DOI: http://dx . doi . org/10 . 7554/eLife . 08500 . 005No Parental CancerAny Time After Parental Cancer DiagnosisFirst Year After Parental Cancer DiagnosisCharacteristics of the ChildNo . of Children With a Hospital Contact for InjuryPerson-yearsHR ( 95%CI ) No . of Children With a Hospital Contact for InjuryPerson-yearsHR ( 95%CI ) p for interactionNo . of Children With a Hospital Contact for InjuryPerson-yearsHR ( 95%CI ) p for interactionSex Male313 , 8065 , 966 , 45119 , 088150 , 0701 . 11 ( 1 . 08-1 . 13 ) < 0 . 0011 , 59422 , 7471 . 30 ( 1 . 24-1 . 37 ) 0 . 17 Female234 , 6825 , 912 , 62416 , 289148 , 2331 . 02 ( 0 . 99-1 . 05 ) 1 , 08021 , 8541 . 23 ( 1 . 16-1 . 31 ) Age ( years ) † < 335 , 157876 , 76111032 , 1061 . 21 ( 0 . 99-1 . 47 ) 571 , 1551 . 25 ( 0 . 96-1 . 63 ) 3-555 , 4521 , 508 , 188143911 , 7751 . 07 ( 0 . 97-1 . 18 ) 1313 , 1811 . 19 ( 1 . 00-1 . 42 ) § 6-11197 , 9844 , 625 , 78614 , 08588 , 1061 . 08 ( 1 . 05-1 . 12 ) 0 . 6075614 , 4351 . 24 ( 1 . 15-1 . 34 ) 0 . 72 12-15134 , 9952 , 500 , 21614 , 69883 , 1991 . 08 ( 1 . 04-1 . 11 ) 77711 , 4291 . 27 ( 1 . 18-1 . 37 ) ≥ 15124 , 9002 , 368 , 12416 , 052113 , 1161 . 06 ( 1 . 03-1 . 09 ) 95314 , 4001 . 32 ( 1 . 23-1 . 41 ) No . of full and half siblings‡ 056 , 1741 , 346 , 43511 , 14324 , 3011 . 05 ( 0 . 98-1 . 11 ) 2083 , 7591 . 30 ( 1 . 13-1 . 50 ) 1229 , 1454 , 982 , 31015 , 806113 , 7721 . 06 ( 1 . 04-1 . 09 ) 0 . 1398716 , 9071 . 24 ( 1 . 16-1 . 32 ) 0 . 74 2151 , 9853 , 242 , 39514 , 35685 , 3861 . 05 ( 1 . 02-1 . 09 ) 78212 , 7481 . 28 ( 1 . 19-1 . 38 ) ≥ 3111 , 1842 , 307 , 93514 , 07274 , 8431 . 11 ( 1 . 07-1 . 15 ) 69711 , 1871 . 31 ( 1 . 21-1 . 41 ) HR , hazard ratio; CI , confidence intervalAdjusted for attained age and sex of the child , interaction between sex of the child and cancer of the parents , number of siblings , gestational age , mode of delivery and birth weight of the child , paternal age at child's birth , maternal age at child's birth , maternal smoking during early pregnancy , and the highest educational level of the parents . †Adjusted for attained age of the child , interaction between attained age of the child and cancer of the parents , sex , number of siblings , gestational age , mode of delivery and birth weight of the child , paternal age at child's birth , maternal age at child's birth , maternal smoking during early pregnancy , and the highest educational level of the parents . ‡Adjusted for attained age , sex and number of siblings of the child , interaction between number of siblings of the child and cancer of the parents , gestational age , mode of delivery and birth weight of the child , paternal age at child's birth , maternal age at child's birth , maternal smoking during early pregnancy , and the highest educational level of the parents§p = 0 . 054 Among all hospital contacts , 96% were due to unintentional injuries ( HR , 1 . 07 [95% CI 1 . 05–1 . 09] ) . Parental cancer also tended to be associated with a higher rate of intentional self-harm ( HR , 1 . 09 [95% CI 0 . 95–1 . 25] ) and undetermined or other injuries ( HR , 1 . 11 [95% CI 0 . 98–1 . 26] ) , but not of assault ( HR , 0 . 99 [95% CI 0 . 87–1 . 13] ) . The associations did not appear to further differ by nature , body region , or mechanism of injury , or by place of injury occurrence , either during the entire follow-up or during the first year after cancer diagnosis ( Figure 1 ) . 10 . 7554/eLife . 08500 . 006Figure 1 . Hazard ratios for hospital contacts for injury among children with parental cancer compared to children without parental cancer , according to different characteristics of injury ( Hazard ratios were adjusted for attained age , sex , number of siblings , gestational age , mode of delivery and birth weight of the child , paternal age at child's birth , maternal age at child's birth , maternal smoking during early pregnancy , and the highest educational level of the parents ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08500 . 006 Among all events of injury , outpatient visit and hospitalization accounted for 83 . 5% and 16 . 5% respectively . Although the positive association was only statistically significant for outpatient visit during the entire follow-up ( outpatient visit HR , 1 . 08 [95% CI 1 . 06–1 . 10]; hospitalization HR , 1 . 03 [95% CI 0 . 99–1 . 08] ) , the association was statistically significant for both hospitalization ( HR , 1 . 18 [95% CI 1 . 07–1 . 31] ) and outpatient visit ( HR , 1 . 29 [95% CI 1 . 24–1 . 35] ) during the first year after cancer diagnosis . With 7-day washout periods , the mean number of hospital contacts for injury was 1 . 8 during the study period . Among children with one previous hospital contact , parental cancer was associated with a 1 . 24-fold rate of having a second hospital contact ( 95% CI 1 . 20–1 . 28 ) ( Table 4 ) . Similar patterns were observed for children with 2–4 previous injuries ( Table 4 ) . When all injuries were studied , children with parental cancer had a 13% higher rate of repeated injuries ( HR , 1 . 13 [95% CI 1 . 12–1 . 15] ) . Additional analyses with 14-day and 30-day washout periods showed similar results ( 14-day HR , 1 . 12 [95% CI 1 . 10–1 . 13]; 30-day HR , 1 . 10 [95% CI 1 . 08–1 . 12] ) . 10 . 7554/eLife . 08500 . 007Table 4 . Hazard ratios for hospital contact for injury among children with parental cancer compared to children without parental cancer , according to the number of previous hospital contact for injury of the childDOI: http://dx . doi . org/10 . 7554/eLife . 08500 . 007CharacteristicsNo . of Children With a Hospital Contact for InjuryPerson-yearsHR ( 95%CI ) No contact No parental cancer548 , 48811 , 879 , 0751 Parental cancer15 , 377298 , 3021 . 07 ( 1 . 05-1 . 09 ) One contact No parental cancer228 , 5601 , 542 , 9261 Parental cancer5 , 98131 , 3081 . 24 ( 1 . 20-1 . 28 ) Two contacts No parental cancer107 , 833454 , 0491 Parental cancer2 , 7969 , 1301 . 26 ( 1 . 20-1 . 32 ) Three contacts No parental cancer53 , 784176 , 2371 Parental cancer1 , 3773 , 6911 . 21 ( 1 . 13-1 . 31 ) Four contacts No parental cancer28 , 12674 , 4991 Parental cancer7221 , 6001 . 19 ( 1 . 07-1 . 33 ) HR , hazard ratio; CI , confidence interval*Adjusted for attained age , sex , number of siblings , gestational age , mode of delivery and birth weight of the child , paternal age at child's birth , maternal age at child's birth , maternal smoking during early pregnancy , and the highest educational level of the parents In this nationwide register-based study , we found that children having a parent with cancer had a higher rate of hospital contact for injury compared with other children . The rate increment was noted for children of all ages as well as for different kinds of injuries or places of injury occurrence , but was most pronounced immediately after the parent’s cancer diagnosis and among children with previous injuries . Comorbid psychiatric diagnoses after the cancer diagnosis rendered further higher rate increment of childhood injury . Although it has been suggested that adolescents are most prone to psychosocial problems at the time of stressful life experience , younger children are in greater need of supervision and parenting ( Phillips , 2014; Macpherson and Emeleus , 2007a; MacPherson and Emeleus , 2007b ) . The positive association between parental cancer and hospital contact for injury among children at all ages in the present study may therefore be jointly attributable to both the psychological distress among the children and the potential lack of parental supervision needed for injury prevention ( Morrongiello et al . , 2006a; Davis Kirsch et al . , 2003; Faulkner and Davey , 2002; Asbridge et al . , 2014; Bylund Grenklo et al . , 2013 ) . It has been debated whether boys and girls are differently affected by parental cancer ( Krattenmacher et al . , 2012; Visser et al . , 2005 ) . Our findings show that overall boys had a more pronounced rate increment than girls for injury . At the same age , boys are on average less mature in terms of social-emotional functioning compared to girls ( Visser et al . , 2005 ) . However , worth noting is that despite the overall difference , boys and girls had similarly increased rates of injury , during the first year after parental cancer diagnosis . Parental cancer was associated with a higher rate of injury , regardless of nature , mechanism , body region of the injury , or place of injury occurrence . Although a positive association was mainly noted for unintentional injuries , the lack of statistical significance for intentional injuries might be due to the relatively small number of intentional injuries observed . Interestingly and reassuringly , we found no increased rate of assault-related injuries after parental cancer . The fact that the higher rate of injury was noted not only at home , but also in transportation areas , in sports areas , etc . , suggests that efforts in preventing injuries in children living with a parent with cancer should include a larger circle of support . Children’s adjustment appears to vary at different stages of their parent’s cancer disease ( Nelson and While , 2002; Huizinga et al . , 2010 ) . Our results showed clearly that children had the highest injury rate increase during the first year after the parent’s cancer diagnosis . This finding corroborates earlier findings in indicating that a cancer diagnosis poses severe psychological distress immediately after the diagnosis , both among the cancer patients and among their children ( Fang et al . , 2012; Lu et al . , 2013; Fall et al . , 2009; Huizinga et al . , 2010 ) . Previous studies have demonstrated that the well-being of children living with a parent with cancer is largely dependent on the adjustment status of their parents to the cancer ( Krattenmacher et al . , 2012; Nelson and While , 2002; Thastum et al . , 2009; Huizinga et al . , 2011 ) . This was supported in our findings that children whose parent was also diagnosed with a psychiatric disorder after cancer diagnosis , appeared to have more pronounced rate increase of injury . Furthermore , among children with higher baseline risk of injury ( i . e . , children with previous hospital contact for injury ) , parental cancer was associated with an even more elevated risk for future injuries . These results highlight both a high-risk time window and high-risk groups for potential future interventions . Although the number of children living with a parent of cancer will undoubtedly increase due to the increasing cancer incidence and improving cancer survival , the postponement of childbearing , etc . , the proportion of such children is however still small , making dedicated intervention both feasible and viable . Conflicting results have been reported regarding whether maternal cancer has greater adverse impact on children than paternal cancer ( Visser et al . , 2005; Compas et al . , 1994 ) . In line with our previous finding on child mortality after parental cancer , the present study indicated no difference between maternal and paternal cancer in relation to the consequent risk of childhood injury ( Chen et al . , 2015 ) . In contrast to previous findings , we found no difference in child injury risk by the severity of parental cancer ( Krattenmacher et al . , 2012 ) . One potential explanation may be the fact that the severity of cancer does not always positively correlate with the adjustment status of the cancer patients . For example , in a recent large-scale study , it was reported that patients of cancers of relatively better survival ( e . g . , breast cancer ) had the highest prevalence of mental disorders , whereas patients of cancers with much severe prognosis ( e . g . , pancreatic cancer ) had the lowest prevalence ( Mehnert et al . , 2014 ) . This study is the first to use a population-based sample to examine the impact of parental cancer on the risk of childhood injury . The major strengths of our study include the nationwide cohort design , using the effective record linkage across the high quality Swedish population and health registers and the prospectively and independently collected information on exposure and outcome . These strengths enhance clearly the validity and generalizability of our findings . Some limitations of our study also deserve consideration . For instance , we had no information on the cohabitation or employment status of the parents . A cancer diagnosis may have considerable impact on the marital relationship and the family's economic status ( Wozniak and Izycki , 2014; de Boer et al . , 2009 ) which may in turn trigger additional psychological distress of the parents , leading to suboptimal parenting ( Tein et al . , 2000; Sallinen et al . , 2004 ) . Divorce or separation may further contribute to child injuries due to the departure of one parent from the household or simply a joint custody of the child between the parents . Therefore potential modifying effect of residence status within the family , as well as the cohabitation and employment status of the parents on the studied association deserves further investigation . Residual confounding due to unmeasured or unknown confounders is possible , however , with presumably small impact . In the multivariable models , only adjustment for the age of the child and the parental ages at child’s birth had noticeable impact on the increase in injury risk , whereas adjustment for other covariates including birth characteristics of the child and educational level of the parents , which have previously been suggested to be associated with both injury risk among the children and cancer risk among the parents , had rather negligible impact ( data not shown ) ( Innes and Byers , 2004; Sun et al . , 2010; Davey Smith et al . , 2007; Beiki et al . , 2014; Hemminki and Li , 2003 ) . The facts that the rate increment was mainly noted during the first three years after parental cancer diagnosis but not thereafter , and that the rate increment was independent of number of siblings or whether the cancer is smoking/alcohol-related or not , further alleviated concerns about residual confounding . Misclassification of injuries remains possible as only above 80% of outpatient visits were included in the Patient Register currently ( National Board of Health and Welfare , 2009 ) . However , such misclassification is largely administrative and arguably non-differential . Cancer parents with established contact with health care may be more likely to seek medical care for their children’s injury . Yet , the opposite can also be postulated that while coping with this major illness , parents are less likely to seek medical care for minor hassles of their children . Such misclassification , if it exists , should have little impact on inpatient care for injury – a proxy of more severe injury event – and could not explain the largely increased injury rate during the first three years after parental cancer diagnosis whereas not thereafter . In summary , children with a parent of cancer had a greater rate of hospital contact for injury , especially during the first year after cancer diagnosis . The association was also more pronounced for parental cancer with comorbid psychiatric disorders after the cancer diagnosis and among children with previous injuries . We conducted a historical cohort study from 2001 to 2010 , including all children born in Sweden during 1983–2002 ( n = 2 , 071 , 380 ) based on the Swedish Multi-Generation Register . The Swedish Multi-Generation Register contains information on all residents in Sweden who were born from 1932 onward and alive in 1961 , together with their parents ( Statistics Sweden , 2011 ) . To be included in the present study , a child must have both biological parents alive , free of cancer and identifiable from this register before the child’s birth ( n = 2 , 027 , 863 ) . All parents of these children were linked to the Swedish Cancer Register , which contains almost 100% complete information on all newly diagnosed cancer cases in Sweden since 1958 ( Barlow et al . , 2009 ) . Information on type of cancer and date of diagnosis was collected from this register . If both parents were diagnosed with a cancer , the first diagnosis was used . A hospital contact for injury was identified as either a hospitalization or an outpatient visit with injury according to the Swedish Patient Register . This register was initiated in 1964/1965 and has national coverage for hospital discharge records since 1987 ( Ludvigsson et al . , 2011 ) . Since 2001 , it also collects information on hospital-based outpatient specialist visits with over 80% coverage of the entire country ( Ludvigsson et al . , 2011 ) . Information collected includes dates of admission and discharge , primary as well as multiple secondary diagnoses , and additionally external causes of morbidity and mortality when applicable . All diagnoses and external causes are coded according to Swedish revisions of the International Classification of Diseases ( ICD ) . Since we were primarily interested in non-medical injuries , injuries due to complications of medical and surgical care were excluded from the definition of childhood injury in the present study . Thus , to be defined as a hospital contact for injury , the record had to have a main discharge diagnosis of injury ( ICD 10: S00-T98 except T80-T88 , T98 . 3 ) and an external cause ( ICD 10: V01-Y98 except Y40-Y84 , Y88 ) . In the primary analysis , we used the first hospital contact for injury during follow-up as the outcome and the date of admission or outpatient visit as the date of injury occurrence . To further examine whether the impact of parental cancer diagnosis differed between any hospital contact for injury ( i . e . , first hospital contact ) and repeated hospital contacts for injury , we analyzed children that had more than one hospital contact for injury during the study period . In this secondary analysis , all hospital visits within a 7-day time period ( wash-out period ) was counted as one contact ( i . e . , more likely referring to the same injury ) . In additional analyses , we also used 14-day and 30-day wash-out periods to assess the robustness of this definition . In the primary analysis , all children were followed from January 1 , 2001 or date of birth , whichever came later . Children without parental cancer contributed person-time to the unexposed period , whereas children with parental cancer contributed person-time first to the unexposed period and after date of parental cancer diagnosis to the exposed period . Children who had a parent diagnosed with cancer before January 1 , 2001 contributed all person–time to the exposed period . For both exposed and unexposed periods , the follow-up was censored on the date of first hospital contact for injury , emigration , death , 18th birthday , or December 31 , 2010 , whichever occurred first . As a result , 63 , 236 children who had died or emigrated or became 18 years old before/at the start of follow-up were excluded , leaving 1 , 964 , 627 children in the final analyses . In the secondary analysis , we specifically followed children ( both exposed and unexposed ) who already had one hospital contact for injury , from the end of wash-out period to the following injuries . For example , to examine the association of parental cancer diagnosis and a future injury among children that had already one hospital visit for injury , we followed all children with a first hospital contact for injury to the second one . Similar follow-ups were conducted when examining the risk of a third , fourth , etc . hospital contact for injury . Various characteristics in children and parents have been linked to both risks of child injury and parental cancer and therefore might either confound or modify the studied association ( Boutsikou and Malamitsi-Puchner , 2011; Morrongiello et al . , 2007; Bradbury et al . , 1999; Peden et al . , 2008; Innes and Byers , 2004; Hjern , 2012; Weitzman et al . , 1992; Weitoft et al . , 2003; Sun et al . , 2010; Hemminki and Li , 2003 ) . To address potential confounding and effect modification , we collected information on sex , gestational age , mode of delivery and birth weight of the child , maternal smoking during early pregnancy and maternal age at child’s birth from the Swedish Medical Birth Register , as well as number of full and half siblings of the child and paternal age at child’s birth through the Multi-Generation Register . The Medical Birth Register was established in 1973 and has covered over 99% of all births in Sweden since 1983 ( Centre for Epidemiology , National Board of Health and Welfare , 2003 ) . We further identified the highest educational level of the parents from the Swedish Register of Education ( Statistics Sweden . 2004 ) . The summary of data included in different registers used in the present study can be found on the homepages of the Swedish National Board of Health and Welfare ( http://www . socialstyrelsen . se/register ) as well as the Statistics Sweden ( http://www . scb . se/sv_/Vara-tjanster/bestalla-mikrodata/Vilka-mikrodata-finns/ ) . The authors confirm that , for approved reasons , some access restrictions apply to the data underlying the findings . The data used in this study are owned by the Swedish National Board of Health and Welfare and Statistics Sweden . According to Swedish law , the authors are not able to make the dataset publicly available . Any researchers ( including international researchers ) interested in obtaining the data can do so by the following steps: 1 ) apply for ethical approval from their local ethical review boards; 2 ) contact the Swedish National Board of Health and Welfare and/or Statistics Sweden with the ethical approval and make a formal application of use of register data . Contact emails for request of register data: Swedish National Board of Health and Welfare: registerservice@socialstyrelsen . se , Statistics Sweden: Mikrodata . individ@scb . se . Please visit http://www . socialstyrelsen . se/register/bestalladatastatistik/bestallaindividuppgifterforforskningsandamal ( the Swedish National Board of Health and Welfare ) and http://www . scb . se/sv_/Vara-tjanster/bestalla-mikrodata/ ( the Statistics Sweden ) for detailed information about how to apply for access to register data for research purposes . Pearson’s χ2 test was used to compare the distributions of different child’s and parental characteristics between the exposed and unexposed children . Cox proportional hazards regression was used to compare the rate of first hospital contact for injury between children with and without parental cancer . HR with 95% CI was estimated after adjustment for the covariates described above . To account for the correlation among children of the same parents , we used “clustered” ( sandwich ) standard errors in all models . Time since birth was used as the underlying time scale in the Cox models; no statistically significant violation of the proportional hazards assumption was detected from a test of the Schoenfeld residuals . Parental cancer diagnosis was treated as a time-varying exposure . To examine the specific impact of cancer diagnosis , independent of the later course of the disease , we calculated the HRs of first hospital contact for injury during the first year , >1 and ≤3 years , and >3 years after parental cancer diagnosis separately . Children with a parental cancer diagnosed before start of follow-up might not contribute to all three categories , depending on when the parental cancer was diagnosed and when follow-up was censored . Since we used time since birth as the underlying time scale , different HRs observed from these analyses did not conflict with the proportional hazards assumption tested . We sub-grouped parental cancer to explore whether maternal and paternal cancer had a different impact on child injury . To assess the impact of lifestyle factors as potential confounders for the studied association , we sub-grouped parental cancer as tobacco-related and other cancers , or alcohol-related and other cancers ( National Board of Health and Welfare , 2013; World Health Organization ) . To explore the potential modifying effect of cancer severity , we further categorized parental cancer as cancer with high , medium or low expected 5-year survival . The expected 5-year survival was indexed as the predicted 5-year relative survival rates of different cancer types based on the entire Cancer Register ( Talback et al . , 2004 ) . We further ascertained from the Patient Register hospital contacts for selected psychiatric comorbidity that were newly diagnosed after the cancer diagnosis among the parents . The psychiatric diagnoses considered were depression , anxiety disorders , stress reaction and adjustment disorder ( detailed diagnoses and corresponding ICD codes are listed in the Table 5 ) . We performed Wald tests to compare the HRs for different subgroups . 10 . 7554/eLife . 08500 . 008Table 5 . Swedish revisions of the international classification of diseases ( ICD ) for psychiatric comorbidity of the cancer parents . DOI: http://dx . doi . org/10 . 7554/eLife . 08500 . 008ICD 8 ( 1969-1986 ) ICD 9 ( 1987-1996 ) ICD-10 ( 1997-presesnt ) Depression296 . 2 , 298 . 0 , 300 . 4296B , 300E , 311F32-F39Anxiety disorders300 except 300 . 3 , 300 . 4300 except 300D , 300EF40 , F41 , F44 , F45 , F48Stress reaction and adjustment disorder307308 , 309F43 To assess whether the impact of parental cancer on child injury differed by sex , age or number of siblings of the child , we used formal tests of interaction of parental cancer with sex , age at follow-up ( <3 , 3–5 , 6–11 , 12–15 or ≥ 15 years ) , or number of full and half siblings ( 0 , 1 , 2 , ≥3 ) of the child . To examine whether the studied association differed for different types of injury , we further conducted separate analyses by manner or intent , nature , body region and mechanism of injury , as well as by place of injury occurrence . To assess whether the association varied by different severity of injury , we also examined separately the risk of hospitalization and outpatient visit for injury . Ordinary Cox proportional hazard regression was used to assess the association between parental cancer diagnosis and a new injury among children with at least one previous hospital contact for injury during follow-up . A conditional Cox model ( PWP-TT model ) was used to assess the overall association between parental cancer diagnosis and repeated injuries in children ( Amorim and Cai , 2015 ) . For all analyses , statistical significance was assessed using 2-tailed 0 . 05-level tests . Data preparation was performed using SAS version 9 . 4 . Statistical analyses were performed using Stata version 12 . 1 . The study was approved by the Central Ethical Review Board in Stockholm , Sweden . All individuals' information was anonymized and de-identified prior to analysis .	A diagnosis of cancer can be devastating for both a person and his or her family . Over the past 40 years , the number of individuals in Sweden diagnosed with cancer has more than doubled leaving growing numbers of families coping with the aftermath . Many individuals diagnosed with cancer have young children . Parents with cancer and their spouses often struggle to cope with disease and the demands of parenting simultaneously . In fact , previous research has shown children with a parent who has cancer have a greater risk of behavioral problems or distress than children with two healthy parents . Whether the stress of having a parent with cancer also affects the children’s physical wellbeing hasn’t been studied much . One concern in particular is whether these children may be at increased risk of injury . Injuries are the most common reason for a child to visit a hospital and in some cases lead to deaths . Children who are not well supervised or whose parents have poor mental health are at increased risk of injury . Coping with cancer and the mental anguish it causes may distract parents and possibly place their children at increased risk of injury . Based on data from nationwide population and health registers in Sweden , Chen , Regodón Wallin et al . now provide evidence that a child with a parent who has cancer is at a greater risk of injury than a child with two parents who are free of cancer . The analysis also revealed that the risk is particularly great if the parent with cancer also develops mental illness after the cancer diagnosis . The risk of injury is greatest in the first year after the parent’s diagnosis . Fortunately , the elevated risk of injury decreases overtime and is almost non-existing after the third year . The analyses suggest that providing extra support for parents with cancer might help to reduce the risk of injury in their children .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health" ]	2015	Childhood injury after a parental cancer diagnosis
Proteasome inhibition elicits an evolutionarily conserved response wherein proteasome subunit mRNAs are upregulated , resulting in recovery ( i . e . , ‘bounce-back’ ) of proteasome activity . We previously demonstrated that the transcription factor Nrf1/NFE2L1 mediates this homeostatic response in mammalian cells . We show here that Nrf1 is initially translocated into the lumen of the ER , but is rapidly and efficiently retrotranslocated to the cytosolic side of the membrane in a manner that depends on p97/VCP . Normally , retrotranslocated Nrf1 is degraded promptly by the proteasome and active species do not accumulate . However , in cells with compromised proteasomes , retrotranslocated Nrf1 escapes degradation and is cleaved N-terminal to Leu-104 to yield a fragment that is no longer tethered to the ER membrane . Importantly , this cleavage event is essential for Nrf1-dependent activation of proteasome gene expression upon proteasome inhibition . Our data uncover an unexpected role for p97 in activation of a transcription factor by relocalizing it from the ER lumen to the cytosol . Transcription factor Nuclear factor erythroid derived 2-related factor 1 ( Nrf1 ) belongs to the cap ‘n’ collar basic leucine zipper ( CNC-bZIP ) family of proteins that are known to be activated in response to cellular stress ( Sykiotis and Bohmann , 2010 ) . Other members of this family include p45 NF-E2 , Nrf2 , Nrf3 , Bach1 , and Bach2 ( Andrews et al . , 1993; Moi et al . , 1994; Oyake et al . , 1996; Kobayashi et al . , 1999 ) . The CNC-bZIP transcriptions factors heterodimerize with small Maf proteins ( Maf F , Maf G , and Maf K ) and preferentially bind to anti-oxidant response elements ( AREs ) present in the promoter region of their target genes ( Motohashi et al . , 2002 ) . The most studied member of the CNC-bZIP family is Nrf2 , which in response to oxidative stress directs a transcriptional program that helps maintain cellular redox homeostasis ( Kensler et al . , 2007 ) . Owing to their sequence similarity , Nrf1 was initially thought to have an overlapping function with Nrf2 in regulating antioxidant gene expression , but a growing body of emerging evidence contradicts this notion . Recently , we demonstrated that in mouse embryonic fibroblasts , Nrf1 but not Nrf2 is necessary for elevated expression of proteasome subunit mRNAs observed in cells treated with proteasome inhibitor , leading to a recovery or ‘bounce-back’ of proteasome activity ( Radhakrishnan et al . , 2010 ) . Consistent with our observation , TCF11 ( a longer isoform of Nrf1 found only in humans ) was subsequently reported to be a mediator of proteasome bounce-back response after proteasome inhibition in human cells ( Steffen et al . , 2010 ) . Thus it appears that Nrf1 functions to combat proteotoxic stress caused by proteasome inhibition in mammals akin to transcription factors RPN4 in yeast ( Xie and Varshavsky , 2001 ) and Cnc-C in Drosophila ( Grimberg et al . , 2011 ) . From the standpoint of cancer treatment , blockade of this bounce-back response may be a viable strategy to enhance the efficacy of proteasome inhibition therapy , given that Nrf1 depletion slows the rate of recovery of proteasome activity following transient application of a covalent proteasome inhibitor and results in enhanced proteasome inhibitor-mediated apoptosis in cancer cells ( Radhakrishnan et al . , 2010 ) . However , since the molecular requirements for Nrf1 activation have not been described , there remains no known mechanism to exploit this possibility . In addition to its role in the induced synthesis of proteasome subunit ( PSM ) genes , Nrf1 has also been found to regulate their basal expression in certain scenarios . For instance , in Nrf1−/− mouse neurons , PSM gene expression is diminished resulting in impaired proteasome function and neurodegeneration ( Lee et al . , 2011 ) . Also , a liver-specific knockout of Nrf1 in mice caused a similar attenuation of PSM expression in hepatocytes ( Lee et al . , 2013 ) . Despite the importance of Nrf1 in proteasome biology and its potential as an anti-cancer target , a thorough understanding of the molecular mechanism behind its activation is currently lacking . What is known is that Nrf1 exists in two forms , p120 and p110 , both of which are unstable and accumulate when the proteasome is inhibited ( Radhakrishnan et al . , 2010 ) . Whereas p120 is embedded in the ER membrane , p110 is soluble and can enter the nucleus ( Biswas and Chan , 2009 ) . However , the mechanism leading to the formation of these species is controversial , with one study pointing to proteolytic cleavage ( Wang and Chan , 2006 ) and another to differential glycosylation ( Zhang et al . , 2007 ) . We show here that , in striking contrast to other ER membrane-tethered transcription factors , the bulk of the Nrf1 polypeptide is inserted into the ER lumen . Activation of Nrf1 depends on p97/VCP-dependent transfer of the luminal segments of Nrf1 to the cytosolic side of the ER membrane , followed by a novel proteolytic processing step . To discriminate between the possibilities that p110 is derived by cleavage or deglycosylation of p120 , we first overexpressed 3×FlagNrf1HA in human HEK-293T and mouse NIH-3T3 cells . This construct contains three consecutive copies of the Flag tag at the N-terminus of Nrf1 and an HA tag at the C-terminus . In the presence of MG132 , regardless of the cell type used , we observed that although the anti-HA antibody was able to detect two different forms of Nrf1 ( ∼120 and ∼110 kDa ) , the anti-Flag antibody was able to detect only the ∼120 kDa form ( Figure 1A ) . The simplest explanation for this observation is that Nrf1 p120 was cleaved somewhere close to the N-terminus to yield p110 . To test this hypothesis and to identify the cleavage site , we overexpressed Nrf13×Flag ( Nrf1 with C-terminal triple Flag tag ) in HEK-293T cells and immunopurified the protein and subjected the p120 and p110 forms to Edman degradation-based N-terminal sequencing . Although the ∼120 kDa band confirmed the intact N-terminus of the full-length Nrf1 , sequence from the ∼110 kDa band was consistent with a new N-terminus starting with Leu-104 ( Figure 1B; putative cleavage site indicated with a scissor ) . 10 . 7554/eLife . 01856 . 003Figure 1 . Nrf1 p110 is derived from p120 by proteolytic processing . ( A ) Human HEK-293T or mouse NIH-3T3 cells were transduced with a retrovirus expressing 3xFlagNrf1HA and 72 hr later were left untreated or treated with 5 µM MG132 for 5 hr . The cell lysates were then used for immunoblotting with anti-Flag and anti-HA antibodies . β-actin protein levels were evaluated as a loading control . ( B ) Amino acids 99 to 108 from human Nrf1 are shown . The scissor indicates the cleavage position predicted by the Edman degradation sequencing . Also shown are mutants 1 through 6 with the mutated amino acids highlighted . ( C ) HEK-293T cells were transiently transfected with wild-type or various mutant constructs ( m1 through m6 ) of Nrf13×Flag and 48 hr later were left untreated or treated with 5 µM MG132 for 5 hr . The cell lysates were then processed for immunoblotting with anti-Flag antibody . β-actin was used as a loading control . ( D ) HEK-293 cells stably expressing either wild-type Nrf13×Flag or Nrf1 ( m1 ) 3×Flag were pulse-labeled for 1 hr with L-azidohomoalanine and then chased with MG132 , cycloheximide ( CHX ) , and excess methionine . The cells were harvested at the time points indicated ( from 0 min to 120 min ) and the lysates were subjected to immunoprecipitation with anti-Flag beads . Nrf1 species were visualized using the tetramethylrhodamine based click-chemistry method as described in ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 01856 . 003 To further confirm this observation , we introduced several mutations flanking the predicted cleavage site ( m1 through m6 in Figure 1B ) and compared these mutants to wild-type Nrf13×Flag for their ability to be processed into Nrf1 p110 . We found that all mutants were defective to varying degrees in their ability to be processed to the p110 form , although the defects of m4 and m6 were relatively modest ( Figure 1C ) . Interestingly , m4 and m6 are the only mutants that retain Trp-103 at the P1 position ( with respect to the cleavage site ) , implying a crucial role of this Trp residue for protease recognition . Although our data suggested that p110 was derived from p120 , there is no decisive evidence that proves that this is the case . To determine if there is a post-translational precursor–product relationship between Nrf1 p120 and p110 forms , we performed a pulse-chase experiment . We used L-azidohomoalanine ( analog of the amino acid L-methionine ) to pulse-label newly synthesized proteins and followed the resultant pool of Nrf1 during a chase period in which protein synthesis and proteasome-dependent degradation were both inhibited by the addition of cycloheximide and MG132 . We observed that over time , wild-type p120 was processed to p110 but the non-cleavable Nrf1 ( m1 ) 3×Flag was not processed ( Figure 1D ) . Taken together , our data are consistent with a model in which newly synthesized Nrf1 p120 undergoes proteolytic processing between Trp-103 and Leu-104 to generate the p110 form . Next , we asked if this proteolytic processing of Nrf1 is biologically relevant . To this end , we compared the ability of wild-type and non-cleavable mutant Nrf1 to reconstitute the bounce-back response in Nrf1−/− mouse embryonic fibroblasts ( Figure 2A ) . As expected , wild-type Nrf13×Flag was able to reinstate the induction of proteasome subunit ( PSM ) genes in response to MG132 treatment in these cells ( Figure 2B ) . In contrast , non-cleavable Nrf1 ( m1 ) 3×Flag was defective in inducing PSM genes under the same conditions , implicating a necessary role for proteolytic processing in Nrf1’s transcriptional program . 10 . 7554/eLife . 01856 . 004Figure 2 . Non-cleavable Nrf1 mutant is functionally defective . ( A ) Nrf1−/− MEFs were transfected with constructs expressing either Nrf13×Flag , Nrf1 ( m1 ) 3×Flag , or vector control and 48 hr later were subjected to Geneticin selection . These antibiotic resistant cells were treated with 1 µM MG132 for 10 hr as indicated . Lysates from these cells were immunoblotted with anti-Flag antibodies to detect exogenous Nrf1 . β-actin was used as a loading control . ( B ) RNA from Nrf1−/− MEFs transfected and selected as described above was used for quantitative RT-PCR to assess the mRNA levels of representative PSM genes . The values were normalized to GAPDH mRNA levels . Error bars denote SD ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01856 . 004 p97 is a homo-hexameric AAA ATPase that is implicated in numerous cellular processes ranging from cell-cycle regulation to membrane fusion and protein degradation ( Ye , 2006 ) . p97 has also been implicated in the turnover of the TCF11 isoform of human Nrf1 via the endoplasmic reticulum-associated degradation ( ERAD ) pathway ( Steffen et al . , 2010 ) . Similar to inhibition of the proteasome , depletion of p97 leads to the strong accumulation of ubiquitin-conjugated substrates ( Wojcik et al . , 2004 ) , underscoring its role in enabling degradation of a subset of the ubiquitinated proteome by the proteasome ( Kolawa et al . , 2013 ) . Given that cells depleted of p97 or treated with proteasome inhibitors accumulate both ubiquitinated proteins and Nrf1 , we asked if p97 depletion was also sufficient to trigger the Nrf1-mediated bounce-back response . To this end , we made use of NIH-3T3 mouse fibroblasts that have been engineered to stably express a doxycycline-inducible form of an shRNA targeting p97 . Unlike proteasome inhibition , depletion of p97 upon addition of doxycycline did not induce the bounce-back response ( Figure 3A; compare bars 1 and 3 for the PSM genes ) . By contrast , we observed that MG132-mediated bounce-back response was severely impaired when p97 was knocked-down ( Figure 3A; compare bars 3 and 4 for the PSM genes ) . 10 . 7554/eLife . 01856 . 005Figure 3 . p97 is required for processing and activation of Nrf1 . ( A ) NIH-3T3-p97sh cells stably expressing doxycycline ( Doxy ) -inducible shRNA targeting p97 were either mock-treated or induced with Doxy ( 1 µg/ml ) for 3 days after which the cells were further treated with DMSO or 1 µM MG132 as indicated for 10 hr . The RNA from these cells was used for quantitative RT-PCR to assess the mRNA levels of the representative PSM genes . The values were normalized to GAPDH mRNA levels . Error bars denote SD ( n = 3 ) . ( B ) NIH-3T3-p97sh cells were subjected to Doxy and MG132 treatments as above and the cell lysates were fractionated by SDS-PAGE and immunoblotted to detect p97 and Nrf1 . β-actin was used as a loading control . ( C ) NIH-3T3 ( left panel ) or HEK-293-Nrf13×Flag cells ( right panel ) were pulsed for an hour with 10 µM NMS-873 and then chased with MG132 plus cycloheximide ( CHX ) . The cells were harvested at the time points indicated ( from 0 min to 120 min ) and the lysates were fractionated by SDS-PAGE and immunoblotted to detect endogenous Nrf1 . β-actin was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 01856 . 005 To investigate the molecular basis for the lack of bounce-back response in p97-depleted cells , we examined the fate of Nrf1 . Consistent with Steffen et al . , 2010 , Nrf1 accumulated in cells depleted of p97 ( Figure 3B , compare lanes 1 and 3 ) . To address the key question of whether p97 contributes to processing of Nrf1 , which is essential for its activation , we evaluated the forms of Nrf1 that accumulated in control vs p97-depleted cells treated with MG132 . Whereas the control cells accumulated p110 ( indicative of Nrf1 activation ) , this form was markedly absent in p97-depleted cells ( Figure 3B , compare lanes 2 and 4 ) . This failure to process p120 is sufficient to account for the lack of bounce-back response in these cells . However , one concern was that due to the 3–4 days required for p97 knock-down , the p120 form we observed in p97-depleted cells could represent an aberrant dead-end species that only accumulated upon sustained deprivation of p97 activity , and was not a true intermediate in the processing pathway . To address this possibility , we made use of the recently described compound NMS-873 , a reversible inhibitor of p97 ( Magnaghi et al . , 2013; Polucci et al . , 2013 ) . Treatment of NIH-3T3 mouse fibroblasts or HEK-293-Nrf13×Flag cells with NMS-873 led to a robust accumulation of p120 Nrf1 ( Figure 3C; lane 1 ) , which was then efficiently converted to the p110 form with a half-life of ∼30 min upon washout of NMS-873 in the presence of cycloheximide and MG132 ( Figure 3C , lanes 2–6 ) . The overproduced protein behaved similarly in HEK-293 cells , although the conversion of p120 to p110 was not as efficient as for the endogenous protein ( Figure 3C , compare left and right panels ) . Thus , our results are consistent with a model in which p97 controls not only the degradation of Nrf1 , but also its processing to the active p110 form . It has been proposed that during synthesis , Nrf1 becomes anchored in the ER membrane via an N-terminal transmembrane region ( residues 7–24 ) ( Zhang and Hayes , 2010 ) . Given the role of p97 in Nrf1 processing and activation , we hypothesized that either p97 promotes processing of cytosolically-exposed Nrf1 and possibly its subsequent extraction from binding partners ( as is the case for Spt23 and Mga2 in yeast , Hoppe et al . , 2000; Rape et al . , 2001 ) , or p97 could be involved in re-positioning the TAD and DBD of Nrf1 from the luminal to the cytosolic side of the ER membrane to facilitate its processing . To distinguish between these possibilities , we performed a series of protease protection experiments to assess the topology of Nrf1 under various conditions ( Figure 4A–D ) . Regardless of p97 status ( normal in the absence of doxycycline , or depleted in the presence of doxycycline , Figure 4A ) , 3×FlagNrf1 was susceptible to Proteinase K ( PK ) digestion in the absence of detergent , implying that the N-terminus was oriented towards the cytosol at all times . Interestingly , however , the orientation of the C-terminus of Nrf13×Flag showed a strong dependence on p97 status ( Figure 4B ) . In mock-depleted cells , Nrf13×Flag was largely susceptible to digestion by PK ( Figure 4B , compare lanes 4 and 5 ) . In stark contrast , when p97 was depleted by the addition of doxycycline , Nrf13×Flag was largely resistant to PK ( Figure 4B , compare lanes 7 and 8 , or lanes 10 and 11 ) . The different behaviors of 3×FlagNrf1 and Nrf13×Flag and the failure to observe detectable trimming of Nrf13×Flag are consistent with a topology wherein only a very small amino-terminal segment of newly-synthesized Nrf1 is exposed to the cytosol ( amino acids 1–6 ) , whereas the bulk of Nrf1 ( amino acids 25–742 ) remains in the lumen of the ER when the retrotranslocation activity of p97 is blocked . Interestingly , when we overexposed our blots , a small , protease-resistant pool of p120 was detectable at steady-state ( Figure 4C , lanes 1 and 2 ) . This pool quickly disappeared upon chasing with cycloheximide plus MG132 ( lanes 4–11 ) . We propose that retrotranslocation of p120 constitutively occurs at such a fast rate that , at steady-state , only a tiny fraction of luminal p120 can be detected unless retrotranslocation is blocked by depletion of p97 as in Figure 4B . 10 . 7554/eLife . 01856 . 006Figure 4 . The C-terminus of Nrf1 is re-positioned from the lumen to the cytosol in a p97-dependent manner . ( A ) HEK-293 cells stably expressing doxycycline ( Doxy ) -inducible shRNA targeting p97 ( HEK-293-p97sh cells ) were transiently transfected with a 3×FlagNrf1 expression construct and 24 hr later were incubated in the presence or absence of Doxy for 4 days after which the cells were further treated or not with 5 µM MG132 for 5 hr . Microsomes prepared from these cells were subjected to protease protection assay with Proteinase K ( PK ) . SDS-PAGE followed by immunoblotting with anti-Flag antibody was used to detect Nrf1 . Formation of a discrete cleavage product of Calnexin in the absence of detergent ( Triton X-100 ) was used as an indicator for intact microsomes . ( B ) Same as ( A ) , except that cells were transfected with an Nrf13×Flag expression construct . ( C ) HEK-293 cells stably expressing the non-cleavable Nrf1 ( m1 ) 3×Flag were treated with MG132 plus cycloheximide ( CHX ) and cells were harvested at indicated time points . Microsomes from these cells were subjected to protease protection assay with PK . Samples were fractionated by SDS-PAGE followed by immunoblotting with anti-Flag antibody . ( D ) HEK-293 cells stably expressing either Nrf13×Flag or the non-cleavable Nrf1 ( m1 ) 3×Flag were pulsed for an hour with 10 µM NMS-873 and then chased with MG132 plus cycloheximide ( CHX ) . Microsomes prepared from cells harvested at the indicated time points ( 0 min and 120 min ) were used in a protease protection assay with PK . SDS-PAGE followed by immunoblotting with anti-Flag antibody was used to detect Nrf1 . ( E ) Total cell lysates from the pulse-chase experiment described above were subjected to deglycosylation by the enzyme Endoglycosidase H prior to SDS-PAGE and immunoblotting with anti-Flag . Species ‘a’ refers to full-length Nrf1 ( p120 ) that was fully glycosylated . Species ‘b’ refers to Nrf1 that was proteolytically processed to p110 and degylcosylated following p97-dependent retrotranslocation . Species ‘c’ refers to species ‘a’ that was deglycosylated with Endo H . Note that species ‘b’ migrates slightly more slowly than species ‘c’ even though both species were deglycosylated and species ‘b’ lacked the N-terminal 103 residues . The unexpectedly slow mobility of species ‘b’ presumably reflects acquisition of additional post-translational modifications upon retrotranslocation on Nrf1’s CTD to the cytosolic side of the ER membrane . ( TMD–transmembrane domain; Gly–N-linked glycan ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01856 . 00610 . 7554/eLife . 01856 . 007Figure 4—figure supplement 1 . Different forms of Nrf1 . HEK-293 cells stably expressing wild-type Nrf13×Flag were treated with MG132 and/or cotransin , an inhibitor of protein insertion into the Sec61 translocation channel ( Garrison et al . , 2005 ) as indicated ( lanes 1 to 4 ) and total cell lysates were prepared . Lanes 5 and 6 contain full-length Nrf13×Flag and Nrf1 ( 104-742 ) 3×Flag that were translated in vitro ( IVT ) in rabbit reticulocyte lysate in the absence of membranes . The HEK293 cell lysates and IVT reactions were examined by SDS-PAGE followed by immunoblotting with anti-Flag antibody . Different Nrf1 species are shown ( ungly–unglycosylated; unmod–unmodified ) . Nrf1 p120 ( species ‘a’ ) was converted to species ‘c’ , which comigrated with the primary translation product ( species ‘d’ ) upon expression in cells treated with cotransin , suggesting that the slow mobility of p120 arose from modifications ( e . g . , N-linked glycosylation ) that occurred within the endoplasmic reticulum . Retrotranslocation and processing of p120 ( species ‘a’ ) yielded p110 ( species ‘b’ ) . Species ‘b’ was not sensitive to endoglycosidase H ( Figure 4E ) , suggesting that it was deglycosylated upon retrotranslocation into the cytosol . Nevertheless , species ‘b’ migrated considerably more slowly than the primary translation product for Nrf1 ( 104-742 ) 3×Flag ( species ‘e’ ) , indicating that p110 must carry additional modifications that remain uncharacterized . Please note that deglycosylation by cytosolic enzymes converts the Asn at the site of glycosylation to Asp , which could influence migration on SDS-PAGE . DOI: http://dx . doi . org/10 . 7554/eLife . 01856 . 007 If the type II transmembrane orientation ( Ncytosol/Clumen ) for Nrf1 indicated by the protease protection experiments represents a true intermediate in its biogenesis , the accumulation of this species should be reversed upon restoration of p97 activity . To test if this is the case , we utilized the p97 inhibitor NMS-873 . Whereas Nrf13×Flag was largely resistant to digestion by PK in HEK293 cells treated with NMS-873 ( Figure 4D , lanes 1 and 2 ) , it became sensitive to PK upon washout of NMS-873 in the presence of MG132 plus cycloheximide ( Figure 4D , lanes 4 and 5 ) . In this case , the Nrf1 was detected as a p120 species , because the tagged protein was chased to p110 less efficiently than the endogenous protein ( Figure 3C ) , and the p110 that did form during the chase fractionated with soluble proteins upon preparation of the membrane fraction . Nrf1 ( m1 ) 3×Flag mirrored the behavior of the wild-type protein ( Figure 4D ) , indicating that the lack of processing observed for non-cleavable mutant was not due to defective retrotranslocation . To corroborate the results from the protease protection experiments , we also monitored the glycosylation status of Nrf1 . We reasoned that when the C-terminal domain ( CTD ) is oriented towards the lumen , Nrf1 should be glycosylated on a set of Asn-X-Ser/Thr sites C-terminal to the transmembrane domain ( Zhang et al . , 2007 ) , whereas when the CTD is retrotranslocated to the cytosol , it should become degylcosylated . To evaluate the glycosylation status of Nrf1 , we monitored its sensitivity to the degylcosylating enzyme Endoglycosidase H ( Endo H ) . Nrf1 p120 that accumulated in cells treated with NMS-873 was sensitive to Endo H ( Figure 4E , compare lanes 1 and 2 or 5 and 6 ) , whereas it became largely resistant to Endo H upon being chased in the presence of MG132 plus cycloheximide ( Figure 4E , compare lanes 3 and 4 or 7 and 8 ) . This behavior is consistent with the protease protection data that indicated that the CTD of p120 flips from a luminal to a cytosolic orientation upon restoration of p97 activity . Notably , the cytosolically-oriented Nrf1 that accumulated upon restoration of p97 activity migrated more slowly than expected for Nrf1 that has been deglycosylated and proteolytically processed . This was particularly evident when the migration of the primary translation product for amino acids 104–742 was compared to the species of Nrf1 that accumulated in cells subjected to various perturbations ( Figure 4—figure supplement 1 ) . Our data suggest that once it is retrotranslocated to the cytosol , Nrf1 is not only deglycosylated and proteolytically cleaved after Trp103 , but it also gains one or more additional modifications , the nature of which remains unknown . Mouse Nrf1 and its human ortholog TCF11 are transcription factors that enhance the accumulation of mRNAs that encode proteasome subunits upon inhibition of proteasome activity ( Radhakrishnan et al . , 2010; Steffen et al . , 2010 ) . Prior work established that Nrf1 is initially integrated into the ER membrane but is turned over rapidly ( Biswas and Chan , 2009; Radhakrishnan et al . , 2010 ) . Upon proteasome inhibition , it accumulates , migrates to the nucleus , and activates gene expression . In this work , we make several critical new observations about the biogenesis and processing of Nrf1 that allow us to formulate a model that explains how it becomes activated in response to proteasome inhibition . Specifically , we show that the domains of Nrf1 that are implicated in transcriptional activation are initially translocated into the lumen of the ER , but subsequently are retrotranslocated to the cytosolic side of the membrane in a manner that depends on the AAA ATPase p97/VCP . In unperturbed cells , the retrotranslocated Nrf1 is promptly degraded by the proteasome , resulting in a futile cycle of continuous synthesis , membrane insertion , retrotranslocation , and degradation ( Figure 5 , left-hand side ) . However , in cells with compromised proteasome activity , some of the retrotranslocated Nrf1 escapes degradation and is cleaved on the N-terminal side of Leu-104 to yield a p110 fragment that is no longer tethered to the ER membrane and can migrate to the nucleus ( Figure 5 , right-hand side ) . Importantly , this cleavage event is essential for Nrf1-dependent activation of proteasome gene expression upon inhibition of the proteasome . 10 . 7554/eLife . 01856 . 008Figure 5 . A model to explain Nrf1 activation . Newly synthesized Nrf1 p120 is initially inserted into the endoplasmic reticulum ( ER ) membrane in an Ncytosol/Clumen orientation . It is then rapidly and efficiently retrotranslocated to the cytosol , where it is immediately degraded by the proteasome . In cells deficient in proteasome activity , degradation of cytosolically-exposed Nrf1 is retarded , allowing sufficient time for proteolytic processing of Nrf1 p120 to yield the active p110 form which can then migrate to the nucleus , heterodimerize with small Maf proteins , and activate transcription of its targets ( e . g . , PSM genes ) by binding to antioxidant response elements ( ARE ) . Note that the exact mechanism of the retrotranslocation step is unknown; what is shown in the figure reflects one possibility among several . Also , although the putative protease is depicted to be in the cytosol , it remains possible that it is in the ER membrane or lumen . DOI: http://dx . doi . org/10 . 7554/eLife . 01856 . 008 This model raises two important questions . First , what is the identity of the protease that clips p120 to yield p110 ? This remains unclear , but the available evidence rules out the most likely candidates . Cleavage by the site-1 and site-2 proteases involved in mobilization of SREBP appears unlikely , because CHO cell lines deficient in these activities ( Rawson et al . , 1997 , 1998 ) were nonetheless able to accumulate p110 upon proteasome inhibition ( R Rawson , personal communication ) . In addition , the proteasome , which cleaves Spt23 and Mga2 from the ER membrane in yeast ( Hoppe et al . , 2000 ) , appears to be dispensable for cleavage because we observed robust formation of p110 in two different cell lines with three different proteasome inhibitors across a wide range of concentrations ( Figure 1 , and data not shown ) . Finally , an inhibitor of ER-localized rhomboid protease ( s ) ( Pierrat et al . , 2011 ) failed to impede the accumulation of p110 in cells treated with MG132 ( data not shown ) . The second question is why does p110 not normally accumulate in unperturbed cells ? Although it will be easier to address this question when the protease is in hand , a reasonable working hypothesis is that in unperturbed cells , p97-dependent retrotranslocation and proteasome-dependent degradation are so intimately coupled that there is no opportunity for the cleavage event to occur as Nrf1 is threaded from the membrane into the maw of the proteasome . However , when the proteasome is inhibited , Nrf1 accumulates on the cytosolic side of the membrane where it is susceptible to cleavage . Even if the protease occasionally intercedes before the proteasome can act and small amounts of p110 are generated , p110 is intrinsically unstable ( Radhakrishnan et al . , 2010 ) and thus does not accumulate to substantial levels . It should be immediately evident from the foregoing discussion how the activation of Nrf1 differs markedly from the activation of all other ER membrane-tethered transcription factors that have been studied to date , including SREBP , ATF6 , OASIS , CREB-H , Spt23 , and Mga2 ( Brown and Goldstein , 1997; Hoppe et al . , 2000; Ye et al . , 2000; Kondo et al . , 2005; Zhang et al . , 2006 ) . In all of those cases , the precursor is inserted into the ER membrane such that the functional domains involved in transcription are displayed on the cytosolic side of the membrane . Cleavages in the cytosolic or intramembrane domains can therefore liberate soluble fragments that translocate to the nucleus . By contrast , since Nrf1 begins its life with its functional domains in the ER lumen , it must be retrotranslocated to the cytosol before processing can yield a functional fragment that can relocalize to the nucleus . However , because the p97-dependent retrotranslocation pathway traversed by Nrf1 is normally employed to clear misfolded proteins from the ER ( Wolf and Stolz , 2012 ) , the inevitable consequence of Nrf1 piggybacking on this mechanism is that it is rapidly degraded by the proteasome as soon as it gains access to the cytosol . This precludes the accumulation of Nrf1 precursor on the cytosolic side of the ER membrane , except under circumstances where there is insufficient proteasome activity to keep pace with the appearance of substrates . Recently , Zhang and Hayes ( Zhang and Hayes , 2013 ) put forth a model for Nrf1 activation that shares some features with the model proposed here but is very different in its key aspects , leading to a fundamentally different view regarding the biogenesis and maturation of p120 into an active transcription factor . Specifically , Zhang and Hayes proposed that p120 spans the ER membrane at least three times . They argued that turnover of this species is promoted by calpain , not the proteasome . p120 that escapes degradation was envisioned to remain stably integrated as a multispanning transmembrane protein until it is mobilized by an unknown signal , which triggers retrotranslocation of the luminal domains followed by their deglycosylation to yield active Nrf1 . It was proposed that Nrf1 is later cleaved by the proteasome to yield lower molecular weight soluble species , but appearance of active Nrf1 precedes this cleavage event . However , the causal relationship between cleavage and activation could not be evaluated because the cleavage site ( s ) was not identified . Another key distinction is that Zhang and Hayes do not perform true chase experiments , and thus their data do not distinguish whether there is a precursor–product relationship between the various Nrf1 species , or they arise from newly-synthesized Nrf1 following different biogenetic pathways . We do not understand the basis for the multiple discrepancies between our data and those of Zhang and Hayes . Our observation that a functional transcription factor can begin its life in the lumen of the ER is unexpected but not without precedent . Ricin , a plant toxin , and some bacterial toxins including cholera toxin , and shiga toxin likewise are retrotranslocated from the ER lumen to the cytosol , where they exert their biological functions ( Inoue et al . , 2011 ) . Another example is the Hepatitis E Virus capsid protein ORF2 , which co-opts the ERAD machinery to gain access to the cytosol ( Surjit et al . , 2007 ) . To our knowledge , the only host protein that undergoes retrotranslocation from the ER but escapes degradation by the proteasome is calreticulin ( Afshar et al . , 2005; Gold et al . , 2010 ) . Upon inhibition of the proteasome , presumably the entire complement of ERAD substrates mimics Nrf1 in that these proteins are retrotranslocated from the luminal to the cytosolic side of the ER membrane but are not degraded . Our observations provoke the question of whether some of these proteins , like Nrf1 , exert novel functions upon their accumulation in the cytosol . An obvious question that emerges from our findings is why is Nrf1 regulated by such a baroque mechanism ? Targeting Nrf1 to the ER may ensure that the basal level of Nrf1 activity is essentially zero in the absence of proteasome stress , due to the normally tight coupling between retrotranslocation and degradation ( Wolf and Stolz , 2012 ) . In cells that experience insufficient proteasome activity—brought about either through environmental insults including proteasome or chaperone inhibitors produced by microbes , or through changes in gene expression—accumulation of active Nrf1 can restore an adequate level of proteasome function . Our finding that p97 is required for the Nrf1-mediated proteasome bounce-back response potentially has therapeutic implications . Previously , we demonstrated that in cancer cells , depletion of Nrf1 slows down recovery of proteasome activity upon transient inhibition of the proteasome and enhances apoptosis caused by a covalent proteasome inhibitor ( Radhakrishnan et al . , 2010 ) . Although inhibition of transcription factors with small molecules has proven difficult ( Berg , 2008 ) , p97 is a druggable target ( Chou et al . , 2011 , 2013; Magnaghi et al . , 2013 ) . Indeed , recent papers have reported enhanced killing of cancer cells when proteasome and p97 inhibitors are used in combination ( Auner et al . , 2013; Chou et al . , 2013 ) . It will be interesting to see whether blockade of the Nrf1-mediated bounce-back response through inhibition of p97 enhances the efficacy of proteasome inhibitor therapy in multiple myeloma or enables expansion of proteasome inhibitor therapy into new indications . 3×FlagNrf1 ( RDB-2411 ) and 3×FlagNrf1HA ( RDB-2412 ) constructs have been described previously ( Radhakrishnan et al . , 2010 ) . Human Nrf1 coding region along with a C-terminal 3×Flag sequence was cloned into pcDNA3 . 1+ ( Invitrogen , Carlsbad , CA ) to establish the wild-type construct Nrf13×Flag ( RDB-2867 ) . The following mutants were derived from Nrf13×Flag via site-directed mutagenesis using the indicated forward primers together with their corresponding reverse complement primers: Nrf1 ( m1 ) 3×Flag ( RDB-2868; forward primer: 5′-ACA GGT TCC AGG TGC CAA CCA CTG AGG TAG CTG CCG CGG CGG CTG CCC GAG ACC CAG AGG G-3′ ) , Nrf1 ( m2 ) 3×Flag ( RDB-2869; forward primer: 5′-GGT GCC AAC CAC TGA GGT AGC TGC CGC GCT GGT TCA CCG AGA-3′ ) , Nrf1 ( m3 ) 3×Flag ( RDB-2870; forward primer: 5′-CAC TGA GGT AAA TGC CGC GCT GGT TCA CCG AGA C-3′ ) , Nrf1 ( m4 ) 3×Flag ( RDB-2871; forward primer: 5′-GAG GTA AAT GCC TGG GCG GTT CAC CGA GAC CC-3′ ) , Nrf1 ( m5 ) 3×Flag ( RDB-2872; forward primer: 5′-CCA CTG AGG TAA ATG CCG CGG CGG TTC ACC GAG ACC CAG-3′ ) , and Nrf1 ( m6 ) 3×Flag ( RDB-2873; forward primer: 5′-ACT GAG GTA AAT GCC TGG GCG GCT GCC CGA GAC CCA GAG GGG TC-3′ ) . The doxycycline-inducible shRNA expression construct pTRIPZ-p97sh ( RDB-2874 ) targeting p97 was based on a 21-mer sequence ( AAC AGC CAT TCT CAA ACA GAA ) present in the coding region of both human and mouse genes and was cloned into the pTRIPZ vector ( Open Biosystems , Huntsville , AL ) . HEK-293T , HEK-293 , and NIH-3T3 cell lines were grown in Dulbecco’s modified Eagle’s medium ( DMEM ) supplemented with 10% fetal bovine serum ( Atlanta Biologicals , Norcross , GA ) , penicillin , and streptomycin ( Invitrogen ) at 37°C in 5% CO2 . Immortalized Nrf1−/− mouse embryonic fibroblasts ( MEFs ) ( Radhakrishnan et al . , 2010 ) were grown as above except that the medium was additionally supplemented with β-mercaptoethanol and non-essential amino acids ( Invitrogen ) . Transfection reagents Lipofectamine 2000 ( for HEK-293T , HEK-293 cells ) or Lipofectamine LTX ( for mouse embryonic fibroblasts ) were used for transient transfections as per manufacturer’s recommendations ( Invitrogen ) . For retroviral production , HEK-293T cells were transfected with the required retroviral construct along with helper plasmids . 48 hr after transfection , media supernatant containing the retrovirus was collected every 4–5 hr for 2 days . This retrovirus-containing medium , supplemented with polybrene ( 10 µg/ml ) , was used to transduce the target cells . NIH-3T3-p97sh cell line ( DTC-138 ) expressing doxycyline-inducible p97-specific shRNA was generated by transducing the mouse NIH-3T3 cells with the pTRIPZ-p97sh retroviral construct and selecting them in the presence of 7 . 5 µg/ml of puromycin . The HEK-293-p97sh cell line ( DTC-139 ) was generated similarly except that 2 µg/ml puromycin was used for selection . Stable cell lines HEK-293-Nrf13×Flag ( DTC-140 ) and HEK-293-Nrf1 ( m1 ) 3×Flag ( DTC-141 ) were established by transfecting HEK-293 cells with the constructs Nrf13×Flag and Nrf1 ( m1 ) 3×Flag respectively and subjecting the cells to 500 µg/ml Geneticin selection . Cells were lysed in RIPA buffer ( 50 mM Tris pH 7 . 4 , 150 mM NaCl , 1% NP40 , 1% Na . Deoxycholate , 0 . 1% SDS , 1 mM EDTA ) supplemented with a protease and phosphatase inhibitor cocktail ( Pierce , Rockford , IL ) . Immunoblots were performed with antibodies specific for Nrf1 ( 8052S; Cell Signaling , Danvers , MA ) , Flag tag ( A8592; Sigma–Aldrich , St . Louis , MO ) , HA tag ( 2013819; Roche Diagnostics , Indianapolis , IN ) , p97 ( sc-20799; SantaCruz Biotechnology , Dallas , TX ) , Calnexin ( sc-11397; SantaCruz Biotechnology ) , and β-actin ( A5441; Sigma–Aldrich ) . HEK-293T cell line was transiently transfected with the construct Nrf13×Flag and 48 hr later was treated with 5 µM MG132 for 5 hr . Cell lysate was prepared in lysis buffer ( 50 mM Tris pH 7 . 4 , 0 . 5 M NaCl , 1 mM EDTA , 1% Triton X-100 ) supplemented with protease and phosphatase inhibitor cocktail ( Pierce ) . The lysate was then subjected to immunoprecipitation with anti-Flag beads ( Sigma-Aldrich ) and eluted with Flag peptide ( Sigma-Aldrich ) . The eluate was resolved by SDS-PAGE and transferred on to a PVDF membrane in the presence of 10 mM CAPS ( 3-cyclohexylamino-1-propane sulfonic acid ) , pH 11 . 0 and 10% Methanol . The membrane was stained with Ponceau S solution , and the Nrf1-specific bands ( p120 and p110 ) were excised and sequenced using a protein micro-sequencer ( Procise cLC 492A sequencer; Applied Biosystems , Foster City , CA ) . Whereas the p120 sample confirmed the intact N-terminus , the p110 sample indicated a mix of amino acid sequences LVHRD and VHRD . The latter sequence is likely to be a frayed-end version of the former that was generated during sample processing ( F Rusnak , personal communication ) , thus revealing the N-terminus of p110 to be Leu-104 . RNA was isolated using the RNeasy kit ( Qiagen , Valencia , CA ) . cDNA was prepared using the Superscript III first strand synthesis kit ( Invitrogen ) according to the manufacturer’s recommendations . Quantitative PCR ( qPCR ) was performed using the SYBR GreenER supermix ( Invitrogen ) . The primers used to quantify murine proteasome subunits ( PSM ) and GAPDH mRNA levels have been described previously ( Radhakrishnan et al . , 2010 ) . The pulse-chase experiment to establish the precursor–product relationship between Nrf1 p120 and p110 was carried out using the Click-iT Metabolic labeling and detection kits ( Invitrogen ) as per the manufacturer’s recommendations . Briefly , HEK-293 cells stably expressing either wild-type Nrf13×Flag or Nrf1 ( m1 ) 3×Flag were starved for an hour in methionine-free medium and pulsed for an additional hour with 50 µM L-azidohomoalanine . Then the cells were washed with PBS and subjected to a chase with the addition of 5 µM MG132 , 50 µg/ml cycloheximide , and 2 mM L-methionine . The cells were harvested at different time points and the cell lysates were prepared in lysis buffer ( 50 mM Tris pH 7 . 4 , 0 . 5 M NaCl , 1 mM EDTA , 1% Triton X-100 ) supplemented with protease and phosphatase inhibitor cocktail ( Pierce ) . The lysates were then used for immunoprecipitation with anti-Flag beads ( Sigma-Aldrich ) . The immunoprecipitates were labeled with Tetramethylrhodamine ( TAMRA ) -alkyne and resolved by SDS-PAGE . The labeled bands were visualized in a Typhoon laser scan imaging system ( GE Healthcare , Pittsburgh , PA ) . The cells were washed once in PBS , resuspended in 10 mM Hepes-KOH pH 7 . 5 buffer and incubated on ice for 10 min . The swollen cells were then sedimented , resuspended in homogenization buffer ( 10 mM Hepes-KOH pH 7 . 5 , 10 mM KCl , 1 . 5 mM MgCl2 , 5 mM EGTA , and 250 mM sucrose ) and passed through a 27G syringe needle several times . The homogenate was then subjected to serial centrifugations at 600×g ( 10 min ) , 3000×g ( 10 min ) and 100 , 000×g ( 60 min ) . The microsomes collected at the end of the 100 , 000×g ultracentrifugation step were resuspended in membrane buffer ( 10 mM Hepes-KOH pH 7 . 5 , 50 mM KOAc , 2 mM Mg ( OAc ) 2 , 1 mM DTT , and 250 mM sucrose ) . Microsomes were mock-treated or subjected to Proteinase K ( 0 . 5 µg/µl ) treatment either in the absence or presence of 1% Triton X-100 for 60 min on ice . The samples were then precipitated by trichloroacetic acid ( TCA ) , resuspended in boiling sample buffer , resolved by SDS-PAGE and subsequently analyzed by immunoblotting with appropriate antibodies . Cells were lysed in IP buffer ( 50 mM HEPES/KOH pH 7 . 5 , 5 mM Mg ( OAc ) 2 , 70 mM KOAc , 0 . 2% Triton X-100; 10% glycerol , and 0 . 2 mM EDTA ) supplemented with a protease and phosphatase inhibitor cocktail ( Pierce ) . These cell lysates were used for deglycosylation reactions with the enzyme Endoglycosidase H for 1 hr at 37°C as per the manufacturer’s recommendations ( New England Biolabs , Ipswich , MA ) .	Cells exposed to high temperatures , infections and other forms of stress often produce oxygen ions and peroxide molecules that can cause damage to proteins and DNA . Cells therefore rely on molecular machines called proteasomes to eliminate damaged proteins , before they cause too much harm . Two related transcription factors—proteins that interact with DNA to ‘switch on’ the expression of genes—are involved in a cell’s responses to stress , but in different ways . Nrf2 switches on genes that limit the damage caused by oxygen ions and peroxide molecules , while Nrf1 switches on the genes that encode the components of the proteasome . As such , Nrf1 helps to restart proteasome activity if it has been shut off—a phenomenon known as ‘bounce-back’ . Within a cell , Nrf1 is known to start off embedded within the membranes of a structure called the endoplasmic reticulum . However , it is not clear how activated Nrf1 leaves this membrane and enters the nucleus to interact with the cell’s DNA . Now , Radhakrishnan et al . show that when Nrf1 is produced , most of its length is found inside the endoplasmic reticulum , with only a small piece being anchored in the surrounding membrane . This is unlike previously described transcription factors that associate with the endoplasmic reticulum , which are stuck to the outside of this structure . Radhakrishnan et al . also discovered that the activation of Nrf1 depends on an enzyme called p97 or VCP . This enzyme helps to flip Nrf1 from the inside of the endoplasmic reticulum to its outside surface . In most cells , the proteasome then breaks down this part of Nrf1 . However , if the proteasome is inhibited , an unknown enzyme cuts Nrf1 free from the endoplasmic reticulum , allowing it to migrate to the nucleus and promote the production of more proteasome components to counteract the inhibition . Interestingly , drugs that inhibit the proteasome are used to combat cancer because the build-up of damaged proteins is toxic to the cancer cells . By showing that p97 promotes the ‘bounce-back’ of the proteasome , the work of Radhakrishnan et al . suggests that combining existing proteasome inhibitors with drugs that inhibit p97 could eventually lead to new , more effective , therapies for cancer or other diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	p97-dependent retrotranslocation and proteolytic processing govern formation of active Nrf1 upon proteasome inhibition
The circuit mechanisms behind shared neural variability ( noise correlation ) and its dependence on neural state are poorly understood . Visual attention is well-suited to constrain cortical models of response variability because attention both increases firing rates and their stimulus sensitivity , as well as decreases noise correlations . We provide a novel analysis of population recordings in rhesus primate visual area V4 showing that a single biophysical mechanism may underlie these diverse neural correlates of attention . We explore model cortical networks where top-down mediated increases in excitability , distributed across excitatory and inhibitory targets , capture the key neuronal correlates of attention . Our models predict that top-down signals primarily affect inhibitory neurons , whereas excitatory neurons are more sensitive to stimulus specific bottom-up inputs . Accounting for trial variability in models of state dependent modulation of neuronal activity is a critical step in building a mechanistic theory of neuronal cognition . The behavioral state of the brain exerts a powerful influence on the cortical responses . For example , electrophysiological recordings from both rodents and primates show that the level of wakefulness ( Steriade et al . , 1993 ) , active sensory exploration ( Crochet et al . , 2011 ) , and attentional focus ( Treue , 2001; Reynolds and Chelazzi , 2004; Gilbert and Sigman , 2007; Moore and Zirnsak , 2017 ) all modulate synaptic and spiking activity . Despite the diversity of behavioral contexts , in all of these cases an overall elevation and desynchronization of cortical activity accompanies heightened states of processing ( Harris and Thiele , 2011 ) . Exploration of the neuronal mechanisms that underly such state changes has primarily centered around how various neuromodulators shift the cellular and synaptic properties of cortical circuits ( Hasselmo , 1995; Lee and Dan , 2012; Noudoost and Moore , 2011; Moore and Zirnsak , 2017 ) However , a coherent theory linking the modulation of cortical circuits to an active desynchronization of population activity is lacking . In this study we provide a circuit-based theory for the known attention-guided modulations of neuronal activity in the visual cortex of primates performing a stimulus change detection task . The investigation of the neuronal correlates of attention has a rich history . Attention increases the firing rates of neurons engaged in feature- and spatial-based processing tasks ( McAdams and Maunsell , 2000; Reynolds et al . , 1999 ) . Attentional modulation of the stimulus-response sensitivity ( gain ) of firing rates is more complicated , often depending on stimulus specifics such as the size and contrast of a visual image ( Williford and Maunsell , 2006; Reynolds and Heeger , 2009; Sanayei et al . , 2015 ) . In recent years there has been increased focus on how brain states affect trial-to-trial spiking variability ( Crochet et al . , 2011; Lin et al . , 2015; Doiron et al . , 2016; Stringer et al . , 2016 ) . In particular , attention decreases the shared variability ( noise correlations ) of the firing rates from pairs of neurons ( Cohen and Maunsell , 2009; Mitchell et al . , 2009; Cohen and Maunsell , 2011; Herrero et al . , 2013; Ruff and Cohen , 2014; Engel et al . , 2016 ) . The combination of a reduction in noise correlations and an increase in response gain has potentially important functional consequences through an improved population code ( Cohen and Maunsell , 2009; Rabinowitz et al . , 2015 ) . In total , there is an emerging picture of the impact of attention on the trial-averaged and trial-variable spiking dynamics of cortical populations . Phenomenological models of attentional modulation have been popular ( Reynolds and Heeger , 2009; Navalpakkam and Itti , 2005; Gilbert and Sigman , 2007; Ecker et al . , 2016 ) ; however , such analyses cannot provide insight into the circuit mechanics of attentional modulation . Biophysical models of attention circuits are difficult to constrain , due in large part to the diversity of mechanisms which control the firing rate and response gain of neurons ( Silver , 2010; Sutherland et al . , 2009 ) . Nonetheless , several circuit models for attentional modulation have been proposed ( Ardid et al . , 2007; Deco and Thiele , 2011; Buia and Tiesinga , 2008 ) , but analysis has been mostly confined to trial-averaged responses . Taking inspiration from these studies , mechanistic models of attentional modulation can be broadly grouped along two hypotheses . First , the circuit mechanisms that control trial-averaged responses may be distinct from those that modulate neuronal variability . This hypothesis has support from experiments in primate V1 showing that N-methyl-D-aspartate receptors have no impact on top-down attentional modulation of firing rates , yet have a strong influence of attentional control of noise correlations ( Herrero et al . , 2013 ) . A second hypothesis is that the modulations of firing rates and noise correlations are reflections of a single biophysical mechanism . Support for this comes from pairs of V4 neurons that each show strong attentional modulation of firing rates , also show a strong attention mediated reductions in noise correlation ( Cohen and Maunsell , 2011 ) . In this study we provide novel analysis of the covariability of V4 population activity engaged in an attention-guided detection task ( Cohen and Maunsell , 2009 ) that is consistent with the second hypothesis . Specifically , the modulation of spike count covariance between unattended and attended states has the same dimensionality as the firing rate modulation . We use the results from our dimensionality analysis to show that an excitatory-inhibitory recurrent circuit model subject to global fluctuations is sufficient to capture both the increase in firing rate and response gain as well as population-wide decrease of noise correlations . Our model makes two predictions regarding neuronal modulation: ( 1 ) that attentional modulation favors inhibitory neurons , and ( 2 ) that stimulus drive favors excitatory neurons . Finally , we show that our model predicts increased informational content in the excitatory population , which would result in improved readout by potential downstream targets . In total , our study provides a simple , parsimonious , and biologically motivated model of attentional modulation in cortical networks . Two rhesus monkeys ( Macaca mulatta ) with microelectrode arrays implanted bilaterally in V4 were trained in an orientation change detection task ( Figure 1a; see Materials and methods: Data preparation ) . A display with oriented Gabor gratings on the left and right flashed on and off . The monkey was cued to attend to either the left or right grating before each block of trials , while keeping fixation on a point between the two gratings . After a random number of presentations , one of the gratings changed orientation . The monkey then had to saccade to that side to obtain a reward . The behavioral task and data collection have been previously reported ( Cohen and Maunsell , 2009 ) . 10 . 7554/eLife . 23978 . 003Figure 1 . Attention increases firing rates and decreases trial-to-trial covariability of population responses . ( a ) Overview of orientation-change detection task; see ( Cohen and Maunsell , 2009 ) for a full description . ( b ) Firing rates of neurons in the unattended ( turquoise ) and attended ( orange ) states , averaged over 3170 units . The slight oscillation in the firing rate was due to the monitor refresh rate . ( c ) Attention significantly decreased the spike count correlation and covariance and slightly increased variance . Error bars provide the SEM . ( d ) Histograms of changes in covariance for each unit pair ( black ) and variance for each unit ( gray ) . In each case we consider the relative change [XA−XU]/max ( XA , XU ) , where X is either Cov ( ni , nj ) or Var ( ni ) . Data was collected from two monkeys over 21 and 16 recording sessions respectively . Signals were analyzed over a 200 ms interval , starting 60 ms after stimulus onset . DOI: http://dx . doi . org/10 . 7554/eLife . 23978 . 003 A neuron is considered to be in an 'attended state' when the attended stimulus is in the hemifield containing that neuron’s receptive field ( contralateral hemifield ) , and in an 'unattended state' when it is in the other ( ipsilateral ) hemifield . The trial-averaged firing rates from both attended and unattended neurons displayed a brief transient rise ( ∼100 ms after stimulus onset ) , and eventually settled to an elevated sustained rate before the trial concluded ( Figure 1b ) . During the sustained period the mean firing rate of attended neurons ( 22 . 0 sp/s ) was greater than that of unattended neurons ( 20 . 6 sp/s ) ( t test , P < 10−5 ) . A major finding of Cohen and Maunsell ( 2009 ) was that the pairwise trial-to-trial noise correlations of the neuronal responses decreased with attention ( Figure 1c , left , mean unattended 0 . 065 , mean attended 0 . 045 , t test , P < 10−5 ) . The noise correlation between neurons i and j is a normalized measure , ρij=Cov ( ni , nj ) /Var ( ni ) Var ( nj ) , where Cov and Var denote spike count covariance and variance respectively . Both spike count variance and covariance significantly change with attention ( ⟨VarU⟩trials=5 . 02⁢ spikes2 , ⟨VarA⟩trials=5 . 10⁢ spikes2 , t test , P < 10−3 , ⟨CovA⟩trials=0 . 252 spikes2 , t test , P < 10−5 ) , but the decrease in covariance ( 34 . 0% ) is much more pronounced than the increase in variance ( 1 . 61%; Figure 1c , middle and right ) . We therefore conclude that the attention mediated decrease in noise correlation is primarily due to decreased covariance . To further validate this observation , we consider the distributions of pairwise changes in covariance ( black ) and variance ( gray ) with attention over the entire data set ( Figure 1d ) . Covariance and variance are normalized by their respective maximal unattended or attended values ( see Methods: Comparing change in covariance to change in variance ) . The change in covariance with attention is concentrated below zero with a large spread , whereas the change in variance is centered on zero with a narrower spread . Taken together these results suggest that to understand the mechanism by which noise correlations decrease it is necessary and sufficient to understand how spike count covariance decreases with attention . A reasonable simplification of V4 neurons is that they receive a bottom-up stimulus alongside an attention-mediated top-down modulatory input . However , to properly model top-down attention we need to first understand the dimension of attentional modulation on the V4 circuit as a whole . Let Aϕ:ϕU ↦ ϕA denote the attentional modulation of measure ϕ from its value in the unattended state , ϕU , to its value in the attended state , ϕA . For example , the firing rate modulation Ar can be written as rA=Ar∘r𝐔 , where rA is an N×1 vector of neural firing rates in the attended state , rU denotes the firing rate vector in the unattended state , Ar is a vector the same size as r , and ∘ denotes elementwise multiplication . In this case , the entries ai of Ar are the ratios of the firing rates: ai=riA/riU ( Figure 2a ) . 10 . 7554/eLife . 23978 . 004Figure 2 . Rank one structure of attentional modulation of spike count covariance . ( a ) Attentional modulation of firing rate . Firing rates of neurons i and j ( black circles are modulated by bottom-up stimulus and top-down attention . ( b ) Two possible models of attentional modulation of covariance . Left: High-rank covariance modulation , in which attention modulates the shared variability of each pair of neurons . Right: Low-rank covariance modulation , in which attention modulates each neuron individually rather than in a pairwise manner . ( c–e ) The measured covariance values plotted against those predicted by the rank-1 model for data collected in one recording session , for c , the actual data ( ρ=0 . 77 ) , d , shuffled data ( ρshuf=0 . 22 , 100 shuffles ) , and ( e ) artificial upper-bound data ( ρub=0 . 90 , 10 realizations of the upper bound model ) . ( f ) Synthesis of c-e in a bar plot . The orange area represents the loss of model performance compared to the upper bound model , and the blue area represents the increase in model performance compared to model applied to shuffled data . ( g ) Rank-1 model performance reported for 21 recording sessions from one monkey . Each bar represents one recording session . Recordings from a mean of N=53 . 5 units in the right-hemisphere were analyzed , with maximum and minimum N of 80 and 35 , respectively . Error bars denote standard error of the mean . ( h ) Mean normalized performance ( relative to ρub ) for both hemispheres of two monkeys ( M1 and M2 ) . ( i ) , Analysis as in ( g ) , using leave-one-out cross-validation to test the predictive power of the model . ( j ) Mean normalized performance of the cross-validated data . DOI: http://dx . doi . org/10 . 7554/eLife . 23978 . 004 A less trivial aspect of attentional modulation is the modulation of covariance matrices: ( 1 ) 𝐂A=AC∘𝐂U . Here 𝐂𝐀 is the attended spike count covariance matrix , 𝐂𝐔 the unattended spike count covariance matrix , and AC is a matrix the same size as 𝐂𝐔 , consisting of entries gi⁢j , which we will call covariance gains . Unlike firing rates , the transformation matrix AC can be of varying rank . On the one hand AC could be constructed from the ratios of the individual elements: gi⁢j=ci⁢jA/ci⁢jU , with each pair of neurons ( i , j ) receiving an individualized attentional modulation gi⁢j of their shared variability ( Figure 2b , left ) . Under this modulation AC is a rank N matrix . A rank N AC will always perfectly ( and trivially ) capture the matrix mapping in Equation ( 1 ) . However , it is difficult to conceive of a top-down circuit mechanism that would allow attention to modulate each pair individually . On the other hand , gi⁢j could depend not on the specific pair ( i , j ) , but on the individual neurons of the pairing: gi⁢j=gi⁢gj ( Figure 2b , right ) . In this case , only N values are needed to characterize AC:AC=ggT , where 𝐠 is a N×1 column vector , meaning AC has rank of 1 . This is a more parsimonious and biophysically plausible scenario for attentional modulation , since in this case the covariance gain gi⁢j of neurons i and j is simply emergent from the attentional modulation of the individual neurons . To test whether AC is low rank we analyzed the V4 population recordings during the visual attention task ( Figure 1 ) , specifically measuring AC under the assumption that AC is rank 1: ( 2 ) 𝐂A=𝐠𝐠T∘𝐂U . Equation ( 2 ) is a system of N⁢ ( N-1 ) /2 equations of the form ci⁢jA=gi⁢gj⁢ci⁢jU in N unknowns 𝐠=[g1 , …⁢gN]T ( we only consider i≠j to exclude variance modulation from our analysis ) . For N>3 this is an overdetermined system , and we solve for 𝐠 using a nonlinear equation solver . Let 𝐠^ be the optimal solution obtained by the solver ( measured as a minimization of the L2-norm of the error; see Methods: objfxn ) . Then C^A:=𝐠^⁢𝐠^T∘CU provides an approximation to the attended covariance matrix . In an example data set from a single recording session with N=39 units , the correlation coefficient ρ of the actual attended covariance values from 𝐂𝐀 versus the approximated attended covariance values from C^A was 0 . 77 ( Figure 2c ) . A shuffled 𝐂𝐀 matrix provides a reasonable null model , and the example data set produces the lower bound correlation ρshuf=0 . 22 ( Figure 2d; see Materials and methods: Shuffled covariance matrices ) . Finally , a Poisson model that perfectly decomposes as Equation ( 2 ) , yet sampled with the same number of trials as in the experiment , gives an upper bound for the rank one structure , the example data yields ρub=0 . 90 ( Figure 2e; see Materials and methods: Upper bound covariance matrices ) . In total , the combination of ρ , ρshuf , and ρub ( Figure 2f ) suggests that the rank one model of attention modulation of covariance AC is well justified . We applied this analysis to 21 recording sessions from the right hemisphere of one monkey ( Figure 2g ) . For most of the recording sessions ρ is closer to ρub than ρshuf . The averaged performance of all sessions for both hemispheres of two monkeys generally agreed with this trend ( Figure 2h ) . We normalized ρ and ρshuf by ρub for each session to better compare different sessions that were subject to day-to-day variations outside of the experimenter’s control , such as the task performance or the internal state of the monkey . To further validate our model we show the distribution of gis computed from the entire data set ( Figure 3a ) . The majority of gi values are less than one , consistent with ⟨CovA⟩trials < ⟨CovU⟩trials ( Figure 1c ) . Further , there was little relation between the attentional modulation of firing rates , measured by riA/riU , and the attentional modulation of covariance through gi ( Figure 3b ) . This indicates that the circuit modulation of firing rates and covariance are not trivially related to one another ( Doiron et al . , 2016 ) . 10 . 7554/eLife . 23978 . 005Figure 3 . Covariance gain shows the attenuation of population-wide fluctuations with attention . ( a ) Distribution of covariance gains gi computed from the entire data set . ( b ) The relation between covariance gi and the attention mediated modulation of firing rates riA/riU . The correlation coefficients between the data sets were 0 . 036 and 0 . 051 for the right and left hemispheres , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 23978 . 005 We additionally tested the validity of our model in Equation ( 2 ) with a leave-one-out cross-validation analysis ( see Materials and methods: Leave-one-out cross-validation ) . We accurately predicted an omitted covariance Ci⁢jA ( Figure 2i and j ) , consistent with our original analysis ( Figure 2g and h ) . The individual session-by-session performance values for both the standard and leave-one-out setups are provided ( Appendix: Model performance for all monkeys and hemispheres ) . Finally , we investigated to what extent the actual value of the covariance gain gi of neuron i depends on the population of neurons in which it was computed . We solved the system of equations Ci⁢jA=gi⁢gj⁢Ci⁢jU using covariance matrices computed from recordings from distinct sets of neurons , overlapping only by neuron i . This gives two estimates of gi , that nevertheless agreed largely with one another ( Appendix: Low-dimensional modulation is intrinsic to neurons ) . This supported the hypothesis that covariance gain gi is an intrinsic property of neuron i . The standard and cross-validation tests verify that the low-rank model of attentional modulation defined in Equation ( 2 ) explains between 66 and 82% ( standard ) , or 56 and 77% ( cross-validation ) of the data . Taking this to be a positive result , we conclude that the covariance gain modulation depends largely on the modulation of individual neurons . Having described attentional modulation statistically our next goal is to develop a circuit model to understand the process mechanistically . Consider a network of N coupled neurons , and let the spike count from neuron i on a given trial be yi . The network output has the covariance matrix 𝐂 with elements ci⁢j=Cov⁢ ( yi , yj ) . In this section we identify the minimal circuit elements so that the attentional mapping AC:CU ↦ CA satisfies the following two conditions ( on average ) : C1: ci⁢jA=gi⁢gj⁢ci⁢jU ; attentional modulation of covariance is rank one ( Figure 2 ) . C2: gi < 1 ; spike count covariance decreases with attention ( Figure 1 ) . What follows is only a sketch of our derivation ( a complete treatment is given in Appendix: Network requirements for attentional modulation ) . If inputs are weak then yi can be described by a linear perturbation about a background state ( Ginzburg and Sompolinsky , 1994; Doiron et al . , 2004; Trousdale et al . , 2012 ) : ( 3 ) yi=yi⁢B+Li⁢ ( ∑k=1NJi⁢k⁢yk+ξi ) . Here yi⁢B is the background activity of neuron i , Ji⁢k is the coupling strength from neuron k to i , and Li is the input-to-output gain of neuron i . In addition to internal coupling we assume a source of external fluctuations ξi to neuron i . Here yi , yi⁢B , and ξi are random variables that vary across trials . The trial-averaged firing rate of neuron i is ri=⟨yi⟩/T ( where ⟨⋅⟩ denotes averaging over trials of length T ) . The background state has variability bi=Var ( yiB ) which we assume to be independent across neurons , meaning the background network covariance is B=diag ( bi ) . Finally , the external fluctuations have covariance matrix 𝐗 with element xij=Cov ( ξi , ξj ) . Motivated by our analysis of population recordings ( Figure 2 ) we study attentional modulations that target individual neurons . This amounts to considering only Ar:riU ↦ riA and AL:LiU ↦ LiA . Additionally , we assume that any model of attentional modulation must result in riA > riU ( Figure 1b ) . A widespread property of both cortical pyramidal cells and interneurons is that an increase of firing rate ri causes an increase of input-output gain L ( Cardin et al . , 2007 ) , thus we will also require LA > LU . Spiking covariability in recurrent networks can be due to internal interactions ( through Ji⁢k ) or external fluctuations ( through ξi ) , or both ( Ocker et al . , 2017 ) . Networks with unstructured connectivity have internally generated covariability that vanishes as N grows . This is true if the connectivity is sparse ( van Vreeswijk and Sompolinsky , 1998 ) , or dense having weak synapses where Ji⁢k∼1/N ( Trousdale et al . , 2012 ) or strong synapses where Ji⁢k∼1/N combined with a balance between excitation and inhibition ( Renart et al . , 2010; Rosenbaum et al . , 2017 ) . In these cases spiking covariability requires external fluctuations to be applied and subsequently filtered by the network . We follow this second scenario and choose 𝐗 so as to provide external covariability to our network . Recent analysis of cortical population recordings show that the shared spiking variability across the population can be well approximated by a rank one model of covariability ( Kelly et al . , 2010; Ecker et al . , 2014; Lin et al . , 2015; Ecker et al . , 2016; Rabinowitz et al . , 2015; Whiteway and Butts , 2017 ) ( we remark that Rabinowitz et al . , 2015 analyzed the same data set that we have in Figures 1 and 2 ) . Thus motivated we take the external fluctuations 𝐗 to be rank one with xi⁢j=xi⁢xj , reflecting a single source of global external variability ξ with unit variance ( neuron i receives ξi=xi⁢ξ ) . Combining this assumption with the linear ansatz in Equation ( 3 ) yields: ( 4 ) 𝐂≈ ( ( 𝐈-𝐊 ) -1⁢𝐋𝐱 ) ⁢ ( ( 𝐈-𝐊 ) -1⁢𝐋𝐱 ) T=𝐜𝐜T , where matrix 𝐊 has element Ki⁢j=Li⁢Ji⁢j and L=diag ( Li ) . We have also defined the vectors 𝐱=[x1 , … , xN]T and 𝐜=[c1 , … , cN]T with ci= ( ( 𝐈-𝐊 ) -1⁢𝐋𝐱 ) i . In total , the output covariability 𝐂 will simply inherit the rank of the input covariability 𝐗 . Attentional modulation affects ci through 𝐊 and 𝐋 and we easily satisfy condition 𝐂𝟏 with gi=ciA/ciU . What remains is to find constraints on 𝐉 and the attentional modulation of 𝐋 that satisfy condition 𝐂𝟐 . Let us consider the case where ciU , ciA > 0 so that condition 𝐂𝟐 is satisfied when ciA−ciU < 0 . For the sake of mathematical simplicity let us separate the population into q⁢N excitatory neurons and ( 1-q ) ⁢N inhibitory neurons ( 0 < q < 1 ) . Let all excitatory ( inhibitory ) neurons project with synaptic strength JE ( -JI ) , have gain LE ( LI ) , and receive the external inputs of strength xE ( xI ) . Finally , let the probability for all connections be p , and consider only weak connections ( J∝1/N and N large ) so that we can ignore the influence of polysynaptic paths in the network ( Pernice et al . , 2011; Trousdale et al . , 2012 ) . Then the attentional modulation of an excitatory neuron decomposes into: ( 5 ) cEA−cEU= ( LEA−LEU ) xE⏟direct external input+ ( LEA−LEU ) qpNJExE⏟external input filteredthrough the excitatory population− ( LIA−LIU ) ( 1−q ) pNJIxI⏟external input filteredthrough the inhibitory population . The first term is the direct transfer of the external fluctuations , and the second and third terms are indirect transfer of external fluctuations via the excitatory and inhibitory populations , respectively . Recall that LA−LU > 0 , meaning that for cEA−cEU < 0 to be satisfied we require the third term to outweigh the combination of the first and second terms . In other words , the inhibitory population must experience a sizable attentional modulation . A similar cancelation of correlations by recurrent inhibition has been recently studied in a variety of cortical models ( Renart et al . , 2010; Tetzlaff et al . , 2012; Ly et al . , 2012; Doiron et al . , 2016; Rosenbaum et al . , 2017 ) . In the above we considered weak synaptic connections where Ji⁢j∼1/N . Rather , if we scale Ji⁢j∼1/N , as would be the case for classical balanced networks ( van Vreeswijk and Sompolinsky , 1998 ) , then for very large N the solution no longer depends upon the gain L . Finite N or the inclusion of synaptic nonlinearities through short term plasticity ( Mongillo et al . , 2012 ) may be necessary to satisfy condition 𝐂𝟐 with large synapses . Furthermore , the large synaptic weights associated with Ji⁢j∼1/N do not allows us to neglect polysynaptic paths , as is needed for Equation ( 5 ) . Extending our analysis to networks with balanced scaling will be the focus of future work . In summary our analysis has identified two circuit features that allow recurrent networks to capture conditions 𝐂𝟏 and 𝐂𝟐 for attentional modulation . First , the network must be subject to a global source of external fluctuations that dominates network covariability ( 𝐂𝟏 ) . Second , the network must have recurrent inhibitory connections that are subject to a large attentional modulation ( 𝐂𝟐 ) . We next apply the intuition gained in the preceding section to propose a cortical model that captures key neural correlates of attentional modulation . We model V4 as a recurrently coupled network of excitatory and inhibitory leaky integrate-and-fire model neurons ( Tetzlaff et al . , 2012; Ledoux and Brunel , 2011; Trousdale et al . , 2012; Doiron et al . , 2004 ) ( Figure 4a ) . In addition to recurrent synaptic inputs , each neuron receives private and global sources of external fluctuating input ( Figure 4b ) . The global noise is an attention-independent source of input correlation that the network filters and transforms into network-wide output spiking correlations ( Figure 4c ) . 10 . 7554/eLife . 23978 . 006Figure 4 . Excitatory-inhibitory network model . ( a ) Recurrent excitatory-inhibitory network subject to private and shared fluctuations as well as top-down attentional modulation . ( b ) Example voltage trace from a LIF model neuron in the network . Top tick marks denote spike times . ( c ) Spike time raster plot of the spiking activity from the model network . ( d ) Population-averaged firing rate rE⁢ ( t ) of the excitatory population . Left: frequency distribution of population-averaged firing rate . ( e ) Transfer function fE between the effective input and the firing rate for a model excitatory neuron . The red segment represents the attentional shift in effective input and hence firing rate . ( f ) , Same as e , but for the inhibitory population . ( g ) Attention as a path through ( r¯E , r¯I ) space , and equivalently through ( IEeff , IIeff ) space . DOI: http://dx . doi . org/10 . 7554/eLife . 23978 . 006 While the linear response theory introduced in Equation ( 3 ) is well suited to study large networks of integrate-and-fire neurons driven by weakly correlated inputs ( Tetzlaff et al . , 2012; Ledoux and Brunel , 2011; Trousdale et al . , 2012; Doiron et al . , 2004 ) , the analysis offers little analytic insight . Instead , we consider the instantaneous activity across population α:ra ( t ) =1Nα∑iyiα ( t ) , where yi⁢α⁢ ( t ) is the spike train from neuron i of population α and Nα is the population size ( α=E or I ) . This approach reduces the model to just the two dynamic variables , the excitatory population rate rE⁢ ( t ) and the inhibitory population rate rI⁢ ( t ) ( rE⁢ ( t ) is shown in Figure 4d ) . Despite this severe reduction the model retains the key ingredients for attentional modulation identified in the previous section – recurrent excitation and inhibition combined with a source of global fluctuations . We take the population sizes to be large and consider a phenomenological dynamic mean field ( Tetzlaff et al . , 2012; Ledoux and Brunel , 2011 ) of the cortical network ( see Materials and methods: Mean field model ) : ( 6 ) τEdrEdt=−rE+fE ( μE+JEErE−JEIrI+σEξ ( t ) ) , τIdrIdt=−rI+fI ( μI+JIErE−JIIrI+σIξ ( t ) ) . The function fα is the input-output transfer of population α , taken to be the mean firing rate for a fixed input ( Figure 4e for the E population and Figure 4f for the I population ) . The parameter Jα⁢β is the coupling strength from population β to population α . Finally , μα and σα are the respective strengths of the mean input and the global fluctuation ξ⁢ ( t ) to population α ( throughout ξ⁢ ( t ) has a zero mean ) . To simplify our exposition we take symmetric coupling JE⁢E=JI⁢E≡JE and JE⁢I=JI⁢I≡JI and symmetric timescales τE=τI ( =1 ) . We set the recurrent coupling so that the model has a stationary mean firing rate ( r¯E , r¯I ) , about which ξ⁢ ( t ) induces fluctuations in rE⁢ ( t ) and rI⁢ ( t ) . Attention is modeled as a top-down influence on the static input: μα=μα⁢B+A⁢Δ⁢μα . Here μα⁢B is a background input , the parameter A models attention with A=0 denoting the unattended state and A=1 the fully attended state , and Δμα > 0 is the increase in μα due to attention . We note that the choice of representing the unattended state by A=0 and the attended state by A=1 is only due to convenience , and is not meant to make any statement about particular bounds on these states . In this model attention simply increases the excitability of all of the neurons in the network ( Figure 4a ) . This modulation is consistent with the rank one structure of attentional modulation in the data ( Figure 2 ) , since μα is a single neuron property . The attention-induced increase in ( μE , μI ) causes an increase in the mean firing rates ( r¯E , r¯I ) ( red paths in Figure 4e , f ) , consistent with recordings from putative excitatory ( McAdams and Maunsell , 2000; Reynolds et al . , 1999 ) and inhibitory neurons ( Mitchell et al . , 2007 ) in visual area V4 . Since fα is a simple rising function then there is a unique mapping of an attentional path in ( μE , μI ) space to a path in ( r¯E , r¯I ) space ( Figure 4g ) . In total , our population model has the core features required to satisfy Conditions C1 and C2 of the previous section . We next use our mean field model to investigate how attentional paths in ( r¯E , r¯I ) space affect population spiking variability . The global input ξ⁢ ( t ) causes fluctuations about the network stationary state: rα⁢ ( t ) =r¯α+δ⁢rα⁢ ( t ) . The fluctuations δ⁢rα⁢ ( t ) are directly related to coordinated spiking activity in population α . In particular , in the limit of large Nα we have that VE≡Var ( rE ) ∝⟨Cov ( yi , yj ) ⟩ , where the expectation is over ( i , j ) pairs in the spiking network . Thus , in our mean field network we require attentional modulation to decrease population variance VE . For sufficiently small σα the fluctuations δ⁢rE⁢ ( t ) and δ⁢rI⁢ ( t ) obey linearized mean field equations ( see Materials and methods: Mean field model , Equation ( 17 ) ) . The linear system is readily analyzed and we obtain the population variance VE computed over long time windows ( see Materials and methods: Computing VE ) : ( 7 ) VE=[LE⁢ ( JI⁢LI⁢ ( σE-σI ) +σE ) 1+JI⁢LI-JE⁢LE]2 . Here Lα≡fα′ is the response gain of neurons in population α . Equation ( 7 ) shows that VE depends directly on Lα , and we recall that Lα changes with attention ( the slope of fα in Figure 4e , f ) . Thus , while the derivation of VE requires linear fluctuations about a steady state , attentional modulation samples the nonlinearity in the transfer fα by changing the state about which we linearize . Any attention-mediated change in VE is not obvious since both LIA > LIU and LEA > LEU , meaning that both the numerator and denominator in Equation ( 7 ) will change with attention . We explore VE by sweeping over ( r¯E , r¯I ) space ( Figure 5a ) . When the network has high r¯E and low r¯I then VE is large , while VE is low for the opposite case of high r¯I and low r¯E . Along our attention path rE increases while VE decreases ( Figure 5b ) , satisfying our requirements for attentional modulation . The attention path that we highlight is just one potential path that reduces population variability , however all paths which reduce VE share a large attention-mediated recruitment of inhibition . If we start with the unattended state ( turquoise dot in Figure 5c ) we can label all ( ΔμE > 0 , ΔμI > 0 ) points that have a smaller population variance than the unattended point ( light green region in Figure 5c ) . These modulations all share that ΔμI > ΔμE ( Figure 5c , green region is below the Δ⁢μE=Δ⁢μI line ) . While the absolute comparison between Δ⁢μE and Δ⁢μI may depend on model parameters , a robust necessary feature of top-down attentional modulation is that it must significantly recruit the inhibitory population . This observation is a major circuit prediction of our model . 10 . 7554/eLife . 23978 . 007Figure 5 . Mean field model shows an attention mediated decrease in population variance . ( a ) An attentional path in excitatory-inhibitory firing rate space for which the population variance decreases . Colored contours define iso-lines of population variance in increments of 10 ( sp/s ) 2 . The attentional path links the unattended state ( A=0; turquoise point ) to the attended state ( A=1 , orange point ) . ( b ) Variance values as a function of the attentional path defined in a . ( c ) The modulation from an unattended state ( origin ) to an attended state over the input space ( Δ⁢μE , Δ⁢μI ) . Solid black line marks where VE remains unchanged , and the green region where ΔVE=VarA ( rE ) −VarU ( rE ) is less than zero . ( d ) The eigenvalue ( λ ) along the attentional path . With increased attention it becomes more negative , indicating that the state ( r¯E , r¯I ) is more stable . e , Autocovariance function of the excitatory population rate rE⁢ ( t ) in the attended and unattended state ( computed using Equation ( 19 ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23978 . 007 An intuitive way to understand inhibition’s role in the decrease in population variance is through the stability analysis of the mean field equations . The eigenvalues of the linearized system are λ1=−1−JILI+JELE < 0 and λ2=-1 ( see Materials and methods: Mean field model , Equation ( 18 ) ) . Note that the denominator of the population variance ( Equation 7 ) equals the square of the eigenvalue product λ1⁢λ2=1+JI⁢LI-JE⁢LE . The stability of the network activity is determined by λ1; the more negative λ1 , the more stable the point ( r¯E , r¯I ) , and the better the network dampens the perturbations about the point due to input fluctuations ξ⁢ ( t ) . The decrease of λ1 along the example attention path is clear ( Figure 5d ) , and overcomes the increase in the numerator of VE due to increases in LE and LI . The enhanced damping is why VE decreases , explicitly seen in the steeper decline of the excitatory population autocovariance function in the attended compared to the unattended state ( Figure 5e ) . This enhanced stability due to recurrent inhibition is a reflection of inhibition canceling population variability provided by external fluctuations and recurrent excitation ( Renart et al . , 2010; Tetzlaff et al . , 2012; Ozeki et al . , 2009 ) . Indeed , taking the coupling J to be weak allows the expansion ( 1+JI⁢LI-JE⁢LE ) -2≈1+2⁢JE⁢LE-2⁢JI⁢LI in Equation ( 7 ) , so that the attention mediated increase in LI reduces population variance through cancellation , as in Equation ( 5 ) . However , this expansion is not formally required to compute the eigenvalues λ1 and λ2 , and these measure the stability of the firing rate dynamics . We mention the expansion only to compare to the original motivation for inhibition . The expression for VE given above ( Equation 7 ) assumes a symmetry in the network coupling , namely that JE⁢E=JI⁢E≡JE and JE⁢I=JI⁢I≡JI . This allowed VE to be compactly written , facilitating the analysis of how attention affects both the numerator and denominator of Equation ( 7 ) . However , the linearization of the mean field equations and the subsequent analysis of population variability do not require this assumption ( see Materials and methods: Mean field model Equations ( 18–20 ) ) . To explore the robustness of our main result we let JI⁢E=α⁢JE and JI⁢I=β⁢JI , thereby breaking the coupling symmetry for α , β≠1 . The reduction in VE with attention is robust over a large region of ( α , β ) ( Figure 6a , green region ) . Focusing on selected ( α , β ) pairings within the region where VE decreases shows that the attentional path identified for the network with coupling symmetry produces qualitatively similar behavior in the more general network ( compare Figure 5c to Figure 6b–e ) . In total , the inhibitory mechanism for attention mediated reduction in population variability is robust to changes in the recurrent coupling with the network . 10 . 7554/eLife . 23978 . 008Figure 6 . The attention mediated reduction in population variance is robust to changes in strength of recurrent connectivity . ( a ) Sweep over α=JE⁢E/JI⁢E and β=JE⁢I/JI⁢I space ( with JE⁢E and JE⁢I fixed ) labeling the region where Δ⁢VE=VEU-VEA is positive ( grey ) and negative ( green ) . ( b–e ) Attentional path in excitatory-inhibitory firing rate space . The colored contours are as in Figure 5a . All calculations are done using Equations ( 18–20 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23978 . 008 While the reduced mean field equations are straightforward to analyze , a similar attenuation of pairwise covariance Cov ( yi , yj ) along the same attentional path occurs in the LIF model network ( Appendix: Spiking network ) . Using linear response analysis for the spiking network we can relate the effect of inhibition to previous work in spiking networks ( Renart et al . , 2010; Tetzlaff et al . , 2012; Ly et al . , 2012; Doiron et al . , 2016 ) . In particular , the attention-mediated decrease of Cov ( yi , yj ) occurs for a wide range of timescale , ranging as low as 20 ms . However , for short timescales that match the higher gamma frequency range ( approximately 60–70 Hz ) this attentional modulation increases Cov ( yi , yj ) ( Appendix 1—figure 6 ) . This finding is consistent with reports of attention-mediated increases of neuronal synchrony on gamma frequency timescales ( Fries et al . , 2001; Buia and Tiesinga , 2008 ) , particularly when inhibitory circuits are engaged ( Kim et al . , 2016 ) . An important neural correlate of attention is enhanced stimulus response gain ( McAdams and Maunsell , 2000 ) . The previous section outlines how the recruitment of recurrent inhibitory feedback by attention reduces response variability . However , inhibitory feedback is also a common gain control mechanism , and increased inhibition reduces response gain through the same mechanism that dampens population variability ( Sutherland et al . , 2009 ) . Thus it is possible that the decorrelating effect of attention in our model may also reduce stimulus response gain as well , which would make the model inconsistent with experimental data . To insert a bottom-up stimulus s in our model we let the attention-independent background input have a stimulus term: μα⁢B=kα⁢s+μ^α⁢B . Here kα is the feedforward stimulus gain to population α and μ^α⁢B is the background input that is both attention and stimulus independent . Our model captures a bulk firing rate rE rather than a population model with distributed tuning . Because of this the stimulus s should either be conceived as the contrast of an input , or the population conceived as a collection of identically-tuned neurons ( i . e a single cortical column ) . Straightforward analysis shows that the stimulus response gain of the excitatory population can be written as ( Materials and methods: Computing stimulus response gain ) : ( 8 ) GE≡d⁢r¯Ed⁢s=kE⁢VEσE+JI⁢LE⁢LI1+JI⁢LI-JE⁢LE⁢ ( kE-kI ) . If kE=kI then GE∝VE , and thus any attentional modulation that reduces population variability will necessarily reduce population stimulus sensitivity . However , for kE > kI the second term in Equation ( 8 ) can counteract this effect and decouple stimulus sensitivity and variability modulations . Consider the example attentional path ( Figure 4g ) with the extreme choice of kE=1 and kI=0 . In this case attention causes an increase in GE ( Figure 7a , b ) , while simultaneously causing a decrease in VE ( Figure 5a , b ) . This is a robust effect , as seen by the region in ( r¯E , r¯I ) space for which the change in VE from the unattended state is negative , and the change in GE is positive ( green region , Figure 7c ) . Further , for fixed kI the proportion of the gray rectangle that the green region occupies increases with kE > kI ( Figure 7d ) . Thus , the decoupling of attentional effects on population variability and stimulus sensitivity is robust to both attentional path ( Δ⁢μE , Δ⁢μI ) and feedforward gain ( kE , kI ) choices . The condition that kE > kI implies that feedforward stimuli must directly target excitatory neurons to a larger degree than inhibitory neurons ( or at least the inhibitory neurons subject to attentional modulation ) . This gives us a complementary prediction to the one from the previous section: while top-down attention favors inhibitory neurons , the bottom-up stimulus favors excitatory neurons . 10 . 7554/eLife . 23978 . 009Figure 7 . Attention model can capture increase in stimulus response gain GE despite decrease in population variance VE . ( a ) Attentional path through ( r¯E , r¯I ) space shows an increase in stimulus response gain . The shown path is the same path as in Figure 5 . ( b ) Values of GE along the path depicted in a . ( c ) The green region in ( r¯E , r¯I ) space denotes where ΔVE=VarA ( rE ) −VarU ( rE ) < 0 and ΔGE=GEA−GEU > 0 . Black lines are iso-lines of covariance and gain , along which those quantities do not change . ( d ) Percent area of the green region in c out of a constant rectangle , as the feedforward stimulus gain kE increases , with kI=0 . 2 held constant . DOI: http://dx . doi . org/10 . 7554/eLife . 23978 . 009 In total , our model of attentional modulation in recurrently coupled excitatory and inhibitory cortical networks subject to global fluctuations satisfies three main neural correlates of attention: ( 1 ) increase in excitatory firing rates and in ( 2 ) stimulus-response gain , with a ( 3 ) decrease in pairwise excitatory neuron co-variability . Attention serves to enhance cognitive performance , especially on discrimination tasks that are difficult ( Moore and Zirnsak , 2017 ) . Thus , it is expected that the attention-mediated reduction in population variability and increase in stimulus response gain subserve an enhanced stimulus estimation ( Cohen and Maunsell , 2009; Ruff and Cohen , 2014 ) . In this section we investigate how the attentional modulation outlined in the previous sections affects stimulus coding by the population . As mentioned above our simplified mean field model ( Equation 6 ) considers only a bulk response , where any individual neuron tuning is lost . As such a proper analysis of population coding is not possible . Nonetheless , our model has two basic features often associated with enhanced coding , decreased population variability ( Figure 5 ) and increased stimulus-response gain ( Figure 7 ) . Fisher information ( Averbeck et al . , 2006; Beck et al . , 2011 ) gives a lower bound on the variance of a stimulus estimate constructed from noisy population responses , and is an often used metric for population coding . The linear Fisher information ( Beck et al . , 2011 ) FIE⁢I computed from our two-dimensional recurrent network is: ( 9 ) FIEI=[GEGI][VECEICEIVI]−1[GEGI]=constant Here Vα=Var ( rα ) , Gα=d⁢r¯α/d⁢s , and CEI=Cov ( rE , rI ) . The important result is that FIE⁢I is invariant with attention , meaning that attention does not increase the network’s capacity to estimate the stimulus s . While the proof of Equation ( 9 ) is straightforward and applies to our recurrent excitatory-inhibitory population ( see Materials and methods: Fisher information ) , the invariance of the total information FE⁢I with attention is most easily understood by analogy with an uncoupled , one-dimensional excitatory population ( Figure 8a ) . Without coupling , the input to the population is simply kE⁢s+σE⁢ξ⁢ ( t ) , which is then passed through the firing rate nonlinearity fE . In this case the gain is GE=kE⁢LE , and assuming a linear transfer the population variance is VE=σE2⁢LE2 . In total the linear Fisher information from the uncoupled population is then: ( 10 ) F⁢IEuc=GE2VE= ( kE⁢LE ) 2σE2⁢LE2=kE2σE2 . 10 . 7554/eLife . 23978 . 010Figure 8 . Attention improves stimulus estimation by the excitatory population embedded within excitatory ( E ) -inhibitory ( I ) network . ( a ) Top: For a uncoupled excitatory population , the stimulus response gain GE increases with attention . Turquoise: unattended state; orange: attended state . Bottom: Population variance VE increases with attention . Stimulus-response curves same as above . Input variance is computed from all input to a population , including external noise and recurrent coupling . The Fisher information for the uncoupled E population is constant with attention because the squared gain GE2 and variance VE increase proportionally ( b ) Same as ( a ) but for the E population within the E-I network . Top: GE increases with attention . Bottom: VE decreases with attention , because the net input variance of the E population decreases with attention . ( c ) Total Fisher information for coupled E-I populations is constant with attention . By contrast , the Fisher information of the excitatory component FIE increases with attention . DOI: http://dx . doi . org/10 . 7554/eLife . 23978 . 010 The proportion LE2 by which attention increases the squared gain ( Figure 8a , top ) is exactly matched by the attention related increase in population variance ( Figure 8a , bottom ) , resulting in cancellation of any attention-dependent terms in FIE . The majority of projection neurons in the neocortex are excitatory , so we now consider the stimulus estimation from a readout of only the excitatory population . Combining our previous results we obtain: ( 11 ) FIE=GE2VE= ( JI⁢LI⁢ ( kE-kI ) +kE ) 2σE2-JI2⁢LI2⁢ ( σE⁢σI-σE2-σI2 ) -2⁢JI⁢LI⁢σE⁢ ( σI-σE ) . Restricting the readout to be from only the excitatory population drastically reduces the total information ( compare FIE⁢I to FIE in Figure 8c ) . As with the uncoupled population the response gain GE of the excitatory neurons in the coupled population increases with attention ( Figure 8b , top ) . Yet unlike the uncoupled population the net input variability to the E population is reduced by attention through a cancelation of the external variability ξ⁢ ( t ) via inhibition ( Figure 8b , bottom ) . These two components combine so that despite FIE < FIEI , we have that FIE does increase with attention ( Figure 8c ) . In sum , even though the total stimulus information in the network does not change with attention , the amount of information extractable from the excitatory population increases , which could lead to improved downstream stimulus estimation in the attended state . Our model does not consider a specific type of inhibitory neuron , and rather models a generic recurrent excitatory-inhibitory circuit . However , inhibitory circuits in cortex are complex , with at least three distinct interneuron types being prominent in many areas: parvalbumin- ( PV ) , somatostatin- ( SOM ) , and vasointestinal peptide-expressing ( VIP ) interneurons ( Rudy et al . , 2011; Pfeffer et al . , 2013; Kepecs and Fishell , 2014 ) . In mouse visual cortex , both SOM and PV cells form recurrent circuits with pyramidal cells , with PV cells having stronger inhibitory projections to pyramidal cells than those of SOM cells ( Pfeffer et al . , 2013 ) . Furthermore , PV and SOM neurons directly inhibit one another , with the SOM to PV connection being stronger than the PV to SOM connection ( Pfeffer et al . , 2013 ) . Finally , VIP cells project strongly to SOM cells ( Pfeffer et al . , 2013 ) and are activated from inputs outside of the circuit ( Lee et al . , 2013; Fu et al . , 2014 ) , making them an attractive target for modulation . Recent studies in visual , auditory , and somatosensory cortical circuits show that VIP cell activation provides an active disinhibition of pyramidal cells via a suppression of SOM cells ( Kepecs and Fishell , 2014 ) . Basal forebrain ( BF ) stimulation modulates both muscarinic and nicotinic ACh receptors ( mAChRs and nAChRs respectively ) in a fashion that mimics attentional modulation ( Alitto and Dan , 2012 ) . In particular , the recruitment of VIP cell activity in vivo through BF stimulation is strongly dependent on both the muscarinic and nicotinic cholinergic pathways ( Alitto and Dan , 2012; Kuchibhotla et al . , 2017; Fu et al . , 2014 ) , and it has thus been hypothesized VIP cells activation could be an important component of attentional modulation ( Alitto and Dan , 2012; Poorthuis et al . , 2014 ) . If we consider the inhibitory population in our model to be PV interneurons then the recruitment of VIP cell activity via top-down cholinergic pathways is consistent with our attentional model in two ways . First , activation of the VIP → SOM → pyramidal cell pathway provides a disinhibition to pyramidal cells , modeled simply as an overall depolarization to pyramidal cells in the attended state ( Figure 4 ) . Second , the activation of the VIP → SOM → PV cell pathway disinhibits PV cells , and the strong SOM → PV projection would suggest that the disinhibition is sizable as required by our model ( Figure 5c ) . Finally , a recent study in mouse medial prefrontal cortex reports that identified PV interneurons show an attention related increase in activity , and that optogenetic silencing of PV neurons impairs attentional processing ( Kim et al . , 2016 ) . However , our logic is perhaps overly simplistic and neglects the direct modulation of SOM cells via muscarinic and nicotinic cholinergic pathways ( Alitto and Dan , 2012; Kuchibhotla et al . , 2017 ) that could compromise the disinhibitory pathways . Further , there is evidence of a direct ACh modulation of PV cells ( Disney et al . , 2014 ) as opposed to through a disinhibitory pathway . Finally , there may be important differences across both species ( mouse vs . primate ) and visual area ( V1 vs . V4 ) that fundamentally change the pyramidal , PV , SOM , and VIP circuit that is understood from mouse V1 ( Pfeffer et al . , 2013 ) . Future studies in the inhibitory to excitatory circuitry of primate visual cortex , and its attentional modulation via neuromodulation , are required to navigate these issues . Finally , the simultaneous increase in response gain and decrease in noise correlations with attention requires excitatory neurons to be more sensitive to bottom-up visual stimulus than inhibitory neurons ( kE > kI , Figure 7 ) . In mouse visual cortex , GABAergic interneurons show overall less stimulus selectivity than pyramidal neurons ( Sohya et al . , 2007 ) , however this involves both direct feedforward and recurrent contributions to stimulus tuning . While our model simplified the feedforward stimulus gain kE and kI to be constant with attention , it is known that attention also modulates feedforward gain through presynaptic nACh receptors ( Disney et al . , 2007 ) . Notably , nAChRs are found at thalamocortical synapses onto layer 4 excitatory cells and not onto inhibitory neurons , suggesting that kE would increase with attention while kI would not . Thus , kE should also increase with attention while kI should not , further supporting that kE > kI . Our model considered the source of global fluctuations as external to the network . This choice was due in part to difficulties in producing global , long timescale fluctuations through strictly internal coupling ( Renart et al . , 2010; Rosenbaum et al . , 2017 ) . Our model assumed that the intensity of these external input fluctuation were independent of attention . Rather , attention shifted the operating point of the network such that the transfer of input variability to population-wide output activity was attenuated in the attended state . Recent analysis of population recordings show that generative models of spike trains that consider gain fluctuations in conjunction with standard spike emission variability capture much of the variability of cortical dynamics ( Rabinowitz et al . , 2015; Lin et al . , 2015 ) . Further , these gain fluctuations are well approximated by a one-dimensional , global stochastic process affecting all neurons in the population ( Ecker et al . , 2014; Rabinowitz et al . , 2015; Lin et al . , 2015; Ecker et al . , 2016; Engel et al . , 2016; Whiteway and Butts , 2017 ) . When these techniques are applied to population recordings subject to attentional modulation , the global gain fluctuations are considerably reduced in the attended state ( Rabinowitz et al . , 2015; Ecker et al . , 2016 ) . Our assumption that external input fluctuations to our network are attention-invariant is consistent with this statistical analysis since it is necessarily constructed from only output activity . Nevertheless , another potential model is that the reduction in population variability is simply inherited from an attention-mediated suppression of the global input fluctuations . Unfortunately , it is difficult to distinguish between these two mechanisms when restricted to only output spiking activity . However , a model where output variability reductions are simply inherited from external inputs suffers from two criticisms . First , it begs the question: what is the mechanism behind the shift in input variability ? Second , our model requires only an increase in the external depolarization to excitatory and inhibitory populations to account for all attentional correlates . An inheritance model would necessarily decouple the attentional mechanisms behind increases in network firing rate ( still requiring a depolarization ) and the decrease in global input variability . Thus , our model offers a parsimonious and biologically motivated explanation of these neural correlates of attention . Further work dissecting the various external and internal sources of variability to cortical networks , and their attentional modulation , is needed to properly validate or refute these different models . Our network model assumed attention-invariant external fluctuations and weak recurrent inputs , permitting a linear analysis of network activity . As a consequence the linear information transfer by the entire population was attention-invariant ( Figure 8 ) , because attention modulated the network’s transfer of signal and noise equivalently . However , this invariance was only apparent if the decoder had access to both the excitatory and inhibitory populations . However , most of the neurons in cortex that project between areas are excitatory . When the decoder was restricted to only the activity of the excitatory population then our analysis uncovered two main results . First , the excitatory population carried less information than the combined excitatory-inhibitory activity , suggesting an inherently suboptimal coding scheme used by the cortex . Second , the attention-mediated modulation of the inhibitory neurons increased the information carried by the excitatory population . This agrees with the wealth of studies that show that attention improves behavioral performance on stimulus discrimination tasks . Determining the impact of population-wide spiking variability on neural coding is complicated ( Averbeck et al . , 2006; Kohn et al . , 2016 ) . A recent theoretical study has shown that noise correlations that limit stimulus information must be parallel to the direction in which population activity encodes the stimulus ( Moreno-Bote et al . , 2014 ) . The fluctuations in our network satisfy this criteria , albeit trivially since all neurons share the same stimulus input . Indeed , in our network the external inputs appear to the network as s+x⁢ ( t ) , meaning that fluctuations from the noise source x⁢ ( t ) are indistinguishable from fluctuations in the stimulus s . This is an oversimplified view and assumes that the decoder treats the neurons as indistinguishable from one another , at odds with classic work in population coding ( Pouget et al . , 2000 ) . Extending our network to include distributed tuning and feature-based recurrent connectivity is a natural next step ( Ben-Yishai et al . , 1995; Rubin et al . , 2015 ) . To do this the spatial scales of feedforward tuning , recurrent projections , external fluctuations , as well as attention modulation must all be specified . It is not clear how noise correlations will depend on these choices yet work in spatially distributed balanced networks shows that solutions can be complex ( Rosenbaum et al . , 2017 ) . The role of inhibition in shaping cortical function is a longstanding topic of study ( Isaacson and Scanziani , 2011 ) , including recent work showing inhibition can actively decorrelate cortical responses ( Renart et al . , 2010; Tetzlaff et al . , 2012; Ly et al . , 2012 ) . Our work gives a concrete example of how this decorrelation can be gated and used to control the flow of information . Of interest are tasks that probe a distributed population where attention again decreases noise correlations between neurons with similar stimulus preference , yet increases noise correlations between cells with dissimilar stimulus preference ( Ruff and Cohen , 2014 ) . The circuit mechanisms underlying this neural correlate of attention are unclear . However , there is ample work in understanding how recurrent inhibition shapes cortical activity in distributed populations ( Isaacson and Scanziani , 2011 ) , including in models of attentional circuits ( Ardid et al . , 2007; Buia and Tiesinga , 2008 ) . Adapting our model to include distributed tuning is an important next step and will be a better framework to discuss the coding consequences of the attentional modulation circuits proposed in our study . Data was collected by from two rhesus monkeys with microelectrode arrays implanted bilaterally in V4 as they performed an orientation-change detection task ( Figure 1a ) ( Cohen and Maunsell , 2009 ) . All animal procedures were in accordance with the Institutional Animal Care and Use Committee of Harvard Medical School . Two oriented Gabor stimuli flashed on and off several times , until one of them changed orientation . The task of the monkey was to then saccade to the stimulus that changed . Each recording session consisted of at least four blocks of trials in which the monkey’s attention was cued to the left or right . We excluded from the analysis instruction trials which occurred at the start of each block to cue the monkey to one side to attend to , catch trials in which the monkey was rewarded just for fixating , and trials in which the monkey did not perform the task correctly . Moreover , the first and last stimulus presentations in each trial were not analyzed , to prevent transients due to stimulus appearance or change from affecting the results . The total number of trials included in the analysis from all the recording sessions was 42 , 496 . Each trial consisted of between 3 and 12 stimulus presentations , of which all but the first and last were analyzed . Recordings from the left and right hemispheres of each monkey were analyzed separately because the activities of the neurons in opposite hemispheres had near-zero correlations ( Cohen and Maunsell , 2009 ) . Neurons in the right hemisphere were considered to be in the attended state when the attentional cue was on the left , and vice-versa . We note that because our criteria for choosing which trials and units to analyze were based on different needs for data analysis compared to the original study ( Cohen and Maunsell , 2009 ) the specific firing rates and covariances differ quantitatively from those previously reported . In monkey 1 , an average of 51 . 1 ( min 35 , max 80 ) units were analyzed from the right hemisphere , and an average of 27 . 5 ( min 14 , max 56 ) units were analyzed from the left hemisphere . From monkey 2 , an average of 56 . 6 ( min 43 , max 71 ) units from the right hemisphere , and an average of 37 . 7 ( min 32 , max 46 ) units from the left hemisphere were analyzed . From each recording , spikes falling between 60 and 260 ms from stimulus onset were considered for the firing rate analysis , to account for the latency of neuronal responses in V4 . Let SU be the matrix containing spike counts of the neurons on trials in which they are in the unattended state , and SA the matrix containing spike counts of the neurons on trials in which they are in the attended state . Denote the unattended spike count covariance matrix by CU=Cov ( SU ) , and the attended one by CA=Cov ( SA ) . Attentional changes in covariance and variance were measured both on average ( Figure 1c ) and as distributions ( Figure 1d ) . The distributions of the normalized differences ( 12 ) CovA−CovUmax ( \|CovA\| , \|CovU\| ) and VarA−VarUmax ( \|VarA\| , \|VarU\| ) reveal a concentration of negative covariance changes , and a distribution of variance changes symmetric about zero . Here , CovA and CovU ( VarA and VarU ) are vectors containing covariance ( variance ) values of the entire data set . Note that the distributions are bounded between -2 and 2 by construction . When solving systems of the form of Equation ( 2 ) in order to quantify the fit of the model , a nonlinear equation solver ( fminunc ) in MATLAB was used . The solver found minima of an objective function which we defined as the Euclidean norm of the difference of the approximation of the attended covariance matrix and the original attended covariance matrix , in other words , the error of the approximation: ( 13 ) f ( g1 , . . . , gN ) =∑i < j ( giCU ( i , j ) gj−CA ( i , j ) ) 2 . For finite population sizes ( N < ∞ ) we expect our algorithm to extract some low-rank structure between arbitrary covariance matrices . Let CA be the principal square root of the attended covariance matrix , the unique positive-semidefinite square root of a positive-semidefinite matrix . Consider the symmetric matrix D=perm ( CA ) computed from the a random permutation of the upper-triangular entries of CA . Finally , let CshufA=real ( DD ) . The square root-permutation-squaring procedure guarantees a positive-semidefinite matrix , as the square of any matrix is positive-semidefinite . Shuffling removes any relation between 𝐂𝐔 and CshufA , and any remaining detected structure would be due to finite sampling . The shuffled covariance gain g^shuf provides the prediction C^shufA:=g^shufg^shufT∘CU , and ρshuf measures the relation between C^shufA and CshufA . Synthetic data shows that as population size N becomes large the coefficient ρshuf approaches 0 ( Appendix: Detected structure in random covariance matrices is a finite-size effect ) . The covariance matrices 𝐂𝐔 and 𝐂𝐀 are estimates obtained from a finite number of trials , and any estimation error will compromise the ability to detect rank one structure of AC . Here we outline an upper bound for the model performance based on a finite number of trials over which the covariance matrices were originally estimated . Let C^A:=𝐠^⁢𝐠^T∘CU with 𝐠^ minimizing the L2 norm of CA:=𝐠𝐠T∘CU . We remark that C^A perfectly decomposes according to the statistical model in Equation ( 2 ) . We used C^A to generate an artificial set of N correlated Poisson spike counts , using an algorithm based on a latent multivariate gaussian model ( Macke et al . , 2009 ) . We sampled these population spike counts with a fixed number of trials ( M ) with D be the resulting M×N matrix of Poisson samples for each process . Let CubA=Cov ( D ) be the 'upper bound' covariance matrix: a finite trial sampling approximation to the perfectly decomposable matrix C^A . Finally , we employ our algorithm to give C^ubA:=g^ubg^ubTCU , where the vector g^ub minimizes the L2 norm of the error . Since C^A is perfectly decomposable then for M→∞ we have C^ubA=CubA=C^A . Thus in the large M limit the coefficient ρub between elements of C^ubA and CubA converges to 1 ( Appendix: Performance limited by available number of trials ) . However , for finite M we have that ρub < 1 , solely due to inaccuracies in estimating C^A with CubA . To account for the possibility of particular strings of realizations D introducing random biases into CubA , we performed the following analysis on 10 independently generated upper-bound covariance matrices CubA . Instead of solving the system consisting of all Equations ( 2 ) , we remove one of them . Denote the complete set of equations by S , an individual equation as si⁢j:={Ci⁢jA=gigjCi⁢jU} and the set of equations with one of them removed as Sa⁢b:=S-sa⁢b . We then solve the system Sa⁢b . Denote the solution by 𝐠a⁢b . We can then compare Ca⁢bA and C^a⁢bA=𝐠a⁢b⁢ ( a ) ⁢𝐠a⁢b⁢ ( b ) ⁢Ca⁢bU . We do this for max ( 1000 , N ( N-1 ) /2 possible systems Sa⁢b . The ρ of the vector of resulting Ca⁢bA vs C^a⁢bA values is a measure of how well the system can predict one of its elements , or in other words , how well the structure holds together when one element is taken out . This leave-one-out cross-validation was performed for the shuffled and the upper-bound cases as well . The mean spiking activity over the population α ( =E or I ) is ( 14 ) rα⁢ ( t ) =⟨yi⁢α⁢ ( t ) ⟩i , where yi⁢α⁢ ( t ) =∑j=1ni⁢αδ⁢ ( t-ti⁢αj ) is the spike train of excitatory neuron i of population α , ni⁢α is the number of spikes from that neuron , and ti⁢αj is the time of spike j . We follow previous studies ( Tetzlaff et al . , 2012; Ozeki et al . , 2009; Ledoux and Brunel , 2011 ) and consider the firing rate dynamics of the E and I populations given by the system in Equations ( 6 ) :τE⁢d⁢rEd⁢t=-rE+fE⁢ ( μE⁢B+A⁢Δ⁢μE+JE⁢E⁢rE-JE⁢I⁢rI+σE⁢[1-χ⁢xE⁢ ( t ) +χ⁢x⁢ ( t ) ] ) , τI⁢d⁢rId⁢t=-rI+fI⁢ ( μI⁢B+A⁢Δ⁢μI+JI⁢E⁢rE-JI⁢I⁢rI+σI⁢[1-χ⁢xI⁢ ( t ) +χ⁢x⁢ ( t ) ] ) . Here μα⁢B is the attention independent drive to population α , A∈[0 , 1] is the attention variable , and Δ⁢μα is the maximal drive to population α due to attention . The parameter Jα⁢β is the coupling from population β to populations α . The stochastic processes xE⁢ ( t ) , xI⁢ ( t ) , and x⁢ ( t ) are the global fluctuations applied to the network . The excitatory and inhibitory populations have private fluctuations xα⁢ ( t ) and also common fluctuations x⁢ ( t ) given to both populations; the parameter χ scales the degree of private versus common fluctuations . We perform calculations for arbitrary χ and then take χ→1 to match the system given in Equations ( 6 ) . The total intensity of fluctuations to population α is set by σα . These simplified rate equations give an accurate picture of the long-timescale dynamics of networks of coupled spiking neuron models that are in the fluctuation driven regime ( Ledoux and Brunel , 2011 ) . The operative timescale reflects a combination of synaptic and membrane integration; since we are interested in spiking covariance over time windows that are much longer than these , we take them to be unity for simplicity . To give a quantitative match between the equilibrium statistics of the rate equations and the leaky integrate-and-fire ( LIF ) network simulations we take the transfer function f to be the inverse first passage time of an LIF neuron driven by white noise ( Ledoux and Brunel , 2011 ) : ( 15 ) fα ( I ) = ( ταπ∫ ( −VT+I ) /ηα ( −VR+I ) /ηαexp ( z2 ) erfc ( z ) dz ) −1 . The parameter ηα is the intensity of the external fluctuations given to the LIF neurons ( Appendix: Spiking model ) . The membrane timescale τ gives the dimensions of 1/s to the firing rate rα . The parameter VT denotes spike threshold while VR is the reset potential . Model parameters are given in Table 1 . 10 . 7554/eLife . 23978 . 011Table 1 . Model Parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 23978 . 011ParameterDescriptionValueτTime constants for membrane dynamics0 . 01 sVTSpike Threshold1VRSpike Reset0μEExcitatory baseline bias0 . 6089μIInhibitory baseline bias0 . 5388Δ⁢μEAttentional modulation of excitatory bias0 . 2624Δ⁢μIAttentional modulation of inhibitory bias0 . 3608JEExcitatory coupling constant1 . 5JIInhibitory coupling constant3σEAmplitude of external noise to E population0 . 3σIAmplitude of external noise to I population0 . 35cProportion of common noise to E and I populations1kESensitivity of E population to stimulus input1kISensitivity of I population to stimulus input0 If the input fluctuations , x⁢ ( t ) , xE⁢ ( t ) , and xI⁢ ( t ) are white noise processes then the nonlinearity in f makes the stochastic dynamics of rE⁢ ( t ) and rI⁢ ( t ) complicated ( non-diffusive ) . To simply the analysis we consider x⁢ ( t ) as the limiting process from:τx⁢d⁢xd⁢t=-x+τx⁢ξx⁢ ( t ) , for τx→0 , with ⟨ξx⁢ ( t ) ⟩=0 and ⟨ξx⁢ ( t ) ⁢ξx⁢ ( t′ ) ⟩=δ⁢ ( t-t′ ) . This makes x⁢ ( t ) sufficiently smooth in time ( the same is true for xE⁢ ( t ) and xI⁢ ( t ) ) . We restrict the coupling Jα⁢β such that for σα=0 the equilibrium point ( r¯E , r¯I ) is stable and given by:r¯E=fE ( μEB+AΔμE+JEEr¯E−JEIr¯I ) , ( 16 ) r¯I=fI ( μIB+AΔμI+JIEr¯E−JIIr¯I ) . For sufficiently small σα the fluctuations in population activity about the equilibrium firing rate , δ⁢rα⁢ ( t ) =rα⁢ ( t ) -r¯α , obey the linearized stochastic system:τEddtδrE= ( −1+LEJEE ) δrE−LEJEIδrI+LEσE ( 1−χxE ( t ) +χx ( t ) ) , ( 17 ) τIddtδrI=LIJIEδrE− ( 1+LIJII ) δrI+LIσI ( 1−χxI ( t ) +χx ( t ) ) . Here Lα=dfαdI\|I=Iαeff is the slope of the transfer function fα evaluated at the equilibrium point Iαeff=μα+AΔμα+JαEr¯E−JαIr¯I . Equation ( 17 ) is a two dimensional Ornstein-Uhlenbeck process ( Gardiner , 2004 ) that is readily amenable to analysis . Linear Fisher Information depends on the stimulus response gains and covariance matrix of the excitatory and inhibitory populations:FIEI=[GEGI][VECEICEIVI]−1[GEGI] ( 23 ) =GE2VI+GI2VE−2GEGICEIVEVI−CEI2 , When the input correlation 0≤χ < 1 we have: ( 24 ) VE= ( LE1+JI⁢LI-JE⁢LE ) 2⁢ ( JI2⁢LI2⁢ ( σE2+σI2-2⁢σE⁢σI⁢χ ) +2⁢JI⁢LI⁢σE⁢ ( σE-σI⁢χ ) +σE2 ) , ( 25 ) VI= ( LI1+JI⁢LI-JE⁢LE ) 2⁢ ( JE2⁢LE2⁢ ( σE2+σI2-2⁢σE⁢σI⁢χ ) +2⁢JE⁢LE⁢σI⁢ ( σI-σE⁢χ ) +σI2 ) , and ( 26 ) CEI=LELI ( 1+JILI−JELE ) 2 ( JEJILELI ( σE2+σI2−2σEσIc ) +JELEσE ( σE−σIχ ) −JILIσI ( σI−σEχ ) +σEσIχ ) . Inserting these expressions and those for GE and GI into Equation ( 23 ) and simplifying yields: ( 27 ) FIE⁢I=2⁢χ⁢kE⁢kI⁢σE⁢σI-kE2⁢σI2-kI2⁢σE2 ( χ2-1 ) ⁢σI2⁢σE2 . We remark that FIE⁢I is independent of LE and LI and thus independent of attentional modulation . Notice that we have re-introduced the correlation constant χ into the equations , rather than only considering the limit χ→1 . If χ=1 , the excitatory and inhibitory populations are receiving completely identical noise . If this is the case , the correlation cancellation would be perfect , leading to infinite informational content , as can be seen in Equation ( 27 ) .	The world around us is complex and our brains need to navigate this complexity . We must focus on relevant inputs from our senses – such as the bus we need to catch – while ignoring distractions – such as the eye-catching displays in the shop windows we pass on the same street . Selective attention is a tool that enables us to filter complex sensory scenes and focus on whatever is most important at the time . But how does selective attention work ? Our sense of vision results from the activity of cells in a region of the brain called visual cortex . Paying attention to an object affects the activity of visual cortex in two ways . First , it causes the average activity of the brain cells in the visual cortex that respond to that object to increase . Second , it reduces spontaneous moment-to-moment fluctuations in the activity of those brain cells , known as noise . Both of these effects make it easier for the brain to process the object in question . Kanashiro et al . set out to build a mathematical model of visual cortex that captures these two components of selective attention . The cortex contains two types of brain cells: excitatory neurons , which activate other cells , and inhibitory neurons , which suppress other cells . Experiments suggest that excitatory neurons contribute to the flow of activity within the cortex , whereas inhibitory neurons help cancel out noise . The new mathematical model predicts that paying attention affects inhibitory neurons far more than excitatory ones . According to the model , selective attention works mainly by reducing the noise that would otherwise distort the activity of visual cortex . The next step is to test this prediction directly . This will require measuring the activity of the inhibitory neurons in an animal performing a selective attention task . Such experiments , which should be achievable using existing technologies , will allow scientists to confirm or disprove the current model , and to dissect the mechanisms that underlie visual attention .	[ "Abstract", "Introduction", "Results", "Discussion", "Methods", "and", "materials" ]	[ "neuroscience" ]	2017	Attentional modulation of neuronal variability in circuit models of cortex
Fusion of mitochondrial outer membranes is crucial for proper organelle function and involves large GTPases called mitofusins . The discrete steps that allow mitochondria to attach to one another and merge their outer membranes are unknown . By combining an in vitro mitochondrial fusion assay with electron cryo-tomography ( cryo-ET ) , we visualize the junction between attached mitochondria isolated from Saccharomyces cerevisiae and observe complexes that mediate this attachment . We find that cycles of GTP hydrolysis induce progressive formation of a docking ring structure around extended areas of contact . Further GTP hydrolysis triggers local outer membrane fusion at the periphery of the contact region . These findings unravel key features of mitofusin-dependent fusion of outer membranes and constitute an important advance in our understanding of how mitochondria connect and merge . Membrane fusion underlies fundamental biological processes such as fertilization , virus entry into host cells or intracellular protein trafficking . Protein and lipid trafficking mainly involves SNAREs ( Soluble N-ethyl maleimide sensitive factor Attachment protein Receptors ) that are expressed on all intracellular compartments undergoing fusion except peroxisomes and mitochondria ( Cai et al . , 2007; Escobar-Henriques and Anton , 2013 ) . Mitochondria constitute a remarkably dynamic network with an organization and ultra-structure that is regulated by fusion and fission of mitochondrial outer and inner membranes ( Labbé et al . , 2014; Westermann , 2010 ) . Fusion and fission are crucial for all mitochondrial functions including oxidative phosphorylation , calcium signalling , apoptosis and lipid metabolism . Defects in mitochondrial fusion and fission are associated with numerous pathologies and severe neurodegenerative diseases ( Dorn , 2013; Liesa et al . , 2009 ) . Mitochondrial fusion and fission both depend on large GTPases of the Dynamin-Related Protein ( DRP ) family ( Labbé et al . , 2014; Low and Löwe , 2010 ) . To promote fission , soluble DRPs assemble into spirals around membrane compartments . GTP hydrolysis causes the spirals to constrict , reducing the diameter of the compartments , ultimately followed by their separation ( Bui and Shaw , 2013 ) . In contrast to fission , the role of DRPs in lipid bilayer fusion remains poorly understood . Among the three families of transmembrane DRPs implicated in fusion , Mitofusins and OPA1 mediate fusion of the mitochondrial outer and inner membranes , respectively , whereas atlastins promote homotypic fusion of ER tubules ( McNew et al . , 2013 ) . GTP binding and hydrolysis participate in trans auto-oligomerization of atlastins and OPA1 through their respective GTPase domains ( Rujiviphat et al . , 2012; Klemm et al . , 2011; Byrnes et al . , 2013; DeVay et al . , 2009; Meglei and McQuibban , 2009; Moss et al . , 2011; Saini et al . , 2014 ) . The resulting homotypic tethering of ER and mitochondrial inner membranes is accompanied by conformational rearrangements of the DRPs that are thought to trigger subsequent fusion of lipid bilayers . Based on structural insights gained from BDLP ( Bacterial Dynamin-Like Protein ) ( Low and Löwe , 2006; Low et al . , 2009 ) , a close relative of the yeast mitofusin Fzo1 , mitofusins may promote outer membrane tethering and fusion through similar processes of oligomerization and conformational rearrangement ( Cohen et al . , 2011; Escobar-Henriques and Anton , 2013 ) . As seen with spirals formed during membrane scission , DRPs are characterized by their ability to assemble into higher-order macromolecular structures ( Ingerman et al . , 2005; Low et al . , 2009; Mears and Hinshaw , 2008 ) . Whether DRPs participate in the formation of such structures during membrane attachment and fusion is unknown . In particular , the precise orchestration of events from the initial attachment of membranes to their ultimate fusion and the requirement for GTP binding and hydrolysis during these steps is elusive . By combining the power of an in vitro mitochondrial fusion assay with cryo-electron tomography ( cryo-ET ) , we undertook to visualize the junction between mitochondria , resulting in an unprecedented dissection of the outer membrane fusion process . The in vitro mitochondrial fusion assay ( Figure 1A ) allows us to distinguish between discrete steps of the mitochondrial fusion process ( i . e . attachment , fusion of outer membranes and fusion of inner membranes ) ( Hoppins et al . , 2009; Meeusen et al . , 2006 , 2004 ) . Purified mitochondria from wild-type yeast cells were brought into contact by centrifugation and were incubated at 4°C for 10 min to promote mitofusin-dependent attachment ( Meeusen et al . , 2004 ) . Subsequent incubation for 45 min at 25°C allows fusion of outer membranes but not inner membranes ( Figure 1A; top ) unless energy is regenerated ( Meeusen et al . , 2004 ) . Consistent with this , fusion reactions recurrently yielded 6 to 8% intermediates with fused outer membranes ( Figure 1B ) . Prolonged incubation of centrifuged mitochondria at 4°C ( Figure 1A; bottom ) decreases fusion of outer membranes but stabilizes attached intermediates ( Cohen et al . , 2011 ) with approximately 25% of mitochondria in close contact ( Figure 1C ) . We reasoned that with an imaging technique of sufficient resolution , it should be possible to visualize the contact sites and detect complexes that mediate mitochondrial attachment . 10 . 7554/eLife . 14618 . 003Figure 1 . In vitro outer membrane fusion and attachment assays . ( A ) Purified mitochondria are brought into contact by centrifugation . A 10 min incubation on ice promotes mitofusin-dependent attachment , which is essential for subsequent fusion of outer membranes at room temperature ( top ) . Prolonged incubation on ice prevents fusion of outer membranes but stabilizes attached intermediates ( bottom ) . Upon incubation at 25°C , fusion of inner membranes does not occur unless energy is regenerated . ( B ) Top: Fusion reactions were performed by mixing mitochondria isolated from cells expressing either the outer membrane protein OM45 tagged with GFP ( OM45-GFP ) or the mitochondrial matrix targeted mCherry ( Mito-mCherry ) . Bottom: Fluorescence microscopy of a fusion reaction . Co-localization of GFP and mCherry indicate intermediates with fused outer membranes ( white arrows ) , scale bars 1 μm . Top right: Fusion efficiency . Error bar represents the s . d . from three independent experiments . ( C ) Representative transmission electron micrograph of in vitro attachment reactions with mitochondria isolated from wild-type cells . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 003 Cryo-ET revealed two main populations of attached mitochondria , characterized by morphologically distinct contacts between outer membranes . 79% of the contact areas were organized as regions where opposing outer membranes approached one another to a distance ds of 1–3 nm ( Figure 2A , green bracket; Figure 2—figure supplement 1A ) , without observable density between the membranes . The contact areas displayed an average surface of 22 , 600 ± 3100 nm² with flattened outer membranes , prompting us to term this state ‘docked’ intermediates ( Figure 2A ) . At the edges of the contact areas , where the distance dl between outer membranes reached 6–9 nm ( Figure 2A , blue brackets; Figure 2—figure supplement 1B ) , we detected defined protein densities between the two outer membranes ( Figure 2A; yellow arrows ) . Tomographic reconstruction ( Figure 2—figure supplement 1C ) and 3D rendered volumes ( Figure 2B ) of mitochondria revealed that contact areas are delimited by a dense ring-like structure , termed the docking ring ( Video 1 ) . 10 . 7554/eLife . 14618 . 004Figure 2 . Cryo-ET of docked intermediates ( 79% of sampled wild-type mitochondria , see Table 1 ) . ( A ) Slices and zooms through tomographic volumes at different z-heights ( section planes indicated in B ) through the center ( left ) or edge ( right ) of a contact area defined by its major axis , la ( red bracket ) , and the perpendicular minor axis lb ( see B ) ; scale bars 100 nm . Dense protein complexes ( yellow arrows ) are visible at a distance of 6–9 nm from the outer membrane ( dl , blue bracket ) but not in the center of the contact area ( 1–3 nm , green bracket , ds ) . Sections through the edge of a contact area ( right ) reveal interstitial densities between the outer membranes . ( B ) 3D rendering of outer membranes ( red and orange ) of two closely apposed mitochondria and protein densities around the contacts area ( blue; the same color scheme is used in all figures ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 00410 . 7554/eLife . 14618 . 005Figure 2—figure supplement 1 . Cryo-ET of major mitochondrial population of wild-type attachment intermediates ( Docked , 79% , see Table 1 ) . ( A ) Histogram of distances between outer membranes of attached mitochondria . Density between outer membranes were observed at distances > 3 . 5 nm . 79% of all contacts were closer than 3 . 5 nm . ( B ) Histogram of outer membrane distance dl at the periphery of contact areas . ( C ) Slices through tomographic volumes of the contact area; scale bars 100 nm . One out of 20 slices in the stack is shown . Red arrows , interstitial density between outer membranes . Note that this density is observed only around the edge of the contact area . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 00510 . 7554/eLife . 14618 . 006Video 1 . 3D rendering of docked mitochondria as shown in Figure 2 . Outer membranes in red and orange , distinct density in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 006 Contact areas of the remaining 21% of attached mitochondria were smaller ( 6900 ± 1700 nm² ) with an average distance of 6 . 7 ± 0 . 5 nm between outer membranes ( Figure 3 ) . In contrast to docked intermediates , there were clear densities between the apposed outer membranes over the complete contact area ( Figure 3A; Figure 3—figure supplement 1 ) . Frequently , this density appeared to consist of more or less regular repeats of globular protein units ( Figure 3B; red arrows ) . These units , with an average spacing of 4 . 6 ± 0 . 7 nm ( n = 13 ) , occasionally extended two protrusions toward each membrane ( Figure 3B; inserted zoom ) . We refer to these attached mitochondria as tethered intermediates , as they appeared to be tethered by proteins . 10 . 7554/eLife . 14618 . 007Figure 3 . Cryo-ET analysis of tethered intermediates ( 21% of sampled wild-type mitochondria , see Table 1 ) . ( A ) 3D rendering of two closely apposed mitochondria shown in B , top row and Figure 3—figure supplement 1 . ( B ) Slices and zooms ( indicated by black and red dashed boxes ) through tomographic volumes; scale bars 100 nm . Blue bracket: outer membrane distance; red arrows , interstitial density . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 00710 . 7554/eLife . 14618 . 008Figure 3—figure supplement 1 . Cryo-ET analysis of the minor mitochondrial population of wild-type attachment intermediates ( Tethered , 21% , see Table 1 ) . Stack of tomographic slices of mitochondria shown in Figure 3A with details of contact area . One out of 20 slices is shown; scale bars 100 nm; red arrows , interstitial density . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 008 The ring of densities ( docked intermediates , Figure 2B ) and the regular repeat of globular protein densities ( tethered intermediates , Figure 3 ) indicate that protein complexes may be responsible for promoting mitochondrial attachment . As the homotypic attachment of outer membranes and their subsequent fusion depends on the GTPase activity of mitofusins ( Cohen et al . , 2011; Koshiba et al . , 2004 ) , we assessed the impact of GTP hydrolysis on the formation of the tethered and docked intermediates . Cryo-ET of in vitro attachment reactions treated with a non-hydrolysable GTP analog ( GMP-PNP ) revealed two distinct populations of attached mitochondria with different distances between their outer membranes ( Figure 4 ) . 10 . 7554/eLife . 14618 . 009Figure 4 . Cryo-ET of mitochondrial attachment intermediates upon GMP-PNP treatment ( Tethered , 86% , see Table 1 ) . ( A-B ) Example slices and zooms ( black and red dashed boxes ) through tomographic volumes; scale bars 100 nm; blue bracket: distance between outer membranes; red arrows: regularly spaced interstitial protein densities . Note the repeat distance of 4 . 3 nm . ( C ) 3D rendering of two closely apposed mitochondria shown in A . The zoomed densities correspond to the regularly spaced interstitial protein densities zoomed in A . ( D ) Histogram of distances between outer membranes for all major populations of attached intermediates . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 00910 . 7554/eLife . 14618 . 010Figure 4—figure supplement 1 . Cryo-ET of a tethered mitochondrial intermediate upon GMP-PNP treatment ( Tethered , 86% , see Table 1 ) . ( A ) Consecutive slices through tomographic volume ( compare Figure 4A ) ; scale bars 100 nm . One out of 20 slices is shown . Red arrows , densities between outer membranes . ( B–D ) Statistics of populations of attached intermediates . ( B ) Contact area of mitochondria at conditions indicated . ( C ) Contact area ratio normalized to mitochondrial dimensions . The area is compared to the diameter of the smaller of the two appressed mitochondria . ( D ) Distance distribution between outer membranes in mitochondrial contacts . The major population of attached intermediates seen in wild-type conditions ( docked ) differs significantly from those observed upon GMP-PNP treatment . Error bars , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 010 In contrast to non-treated samples , only 14% of mitochondria formed contacts reminiscent of docked intermediates and with significantly smaller contact areas ( Table 1 ) . In the remaining 86% of attached mitochondria , the outer membranes were 6 . 3 ± 0 . 2 nm apart ( Figure 4A , blue bracket ) . Densities were observed between apposed membranes with frequent repeats of 3–4 nm globular structures with , occasionally , two protrusions extending towards each membrane ( Figure 4A , inserted zoom and Figure 4B , red arrows ) . These densities and their regular spacing ( 4 . 3 ± 0 . 6 nm; n = 24 ) were strikingly similar to those seen in wild-type tethered intermediates ( Figure 3B ) . Moreover , among the attached intermediates we identified ( Table 1 ) , this sub-population observed upon treatment with GMP-PNP displayed features that were essentially identical to the tethered intermediates seen in wild-type conditions both in terms of morphology ( Figure 4C; Figure 4—figure supplement 1A; Video 2 ) and statistics ( Figure 4D; Figure 4—figure supplement 1B–D ) . 10 . 7554/eLife . 14618 . 011Table 1 . Characteristics of mitochondrial attachment intermediates identified in this study . Abbreviations: o . e . overexpression , i . a . inter alia . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 011Condition ( n ) Contact type ( n ) %Densities organizationDistance ± SE / nmContact area ± SE / nm²Wild-Type ( 52 ) Docked ( 41 ) 79Docking ring2 . 1 ± 0 . 122600 ± 3100Tethered ( 11 ) 21Interstitial/Few6 . 7 ± 0 . 56900 ± 1700Wild-Type + GMP-PNP ( 22 ) Tethered ( 19 ) 86Interstitial/Few6 . 3 ± 0 . 210100 ± 1300Other ( 3 ) 14i . a . Docking ring2 . 6 ± 0 . 311600 ± 3900Fzo1 O . E . ( 19 ) Abortive ( 14 ) 74Interstitial/Numerous8 . 8 ± 0 . 45000 ± 900Other ( 5 ) 26Docking ring2 . 5 ± 0 . 44200 ± 120010 . 7554/eLife . 14618 . 012Video 2 . 3D rendering of tethered mitochondria upon addition of GMP-PNP as shown in Figure 4 . Outer membranes in red and orange , distinct density in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 012 These observations suggest that inhibition of GTP hydrolysis induces the accumulation of tethered mitochondria at the expense of docked intermediates ( 21% tethered intermediates in wild-type conditions against 86% upon GMP-PNP treatment ) . To find out if inhibition of GTP hydrolysis blocks formation of the docking ring at an early stage of the mitochondrial attachment process , we introduced GMP-PNP at distinct time points of the in vitro fusion reaction and analyzed the ratios of tethered and docked intermediates by cryo-ET ( Figure 5A ) . Docked intermediates were predominantly found when GMP-PNP was added 10 , 20 or 40 min after centrifugation ( Figure 5B , blue squares ) , indicating that the ring had been already formed and that it was not dissolved by GMP-PNP . In contrast , addition of GMP-PNP prior to or immediately after centrifugation resulted in strong accumulation of tethered intermediates ( Figure 5B , red circles ) , while docked intermediates were suppressed . Strikingly , addition of GMP-PNP two minutes after centrifugation resulted in equal proportions of tethered and docked intermediates ( Figure 5B ) . 10 . 7554/eLife . 14618 . 013Figure 5 . Cryo-ET time course experiment of attached mitochondria treated with GMP-PNP . ( A ) GMP-PNP was added at the time points indicated ( dashed black lines ) . ( B ) Proportion of docked and tethered intermediates observed before and after GMP-PNP addition . Time t = 0 indicates the start of the incubation period on ice . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 01310 . 7554/eLife . 14618 . 014Figure 5—figure supplement 1 . Hybrid intermediate captured upon GMP-PNP addition two minutes after centrifugation . Left: Slices through tomographic volume of two attached mitochondria; scale bars 100 nm . The blue bar indicates the typical distance between outer membranes of GMP-PNP-treated mitochondria ( about 6 nm ) . The green bar indicates a small area of closer contact . Right: 3D rendering of two mitochondria . Note the absence of protein densities between outer membranes in close contact . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 014 These observations demonstrate not only that tethering precedes docking , but also that transition from one state to the other is controlled by GTP hydrolysis . Consequently , the tethered intermediate with its repeating globular protein densities represents an early stage of the mitochondrial outer membrane fusion process . GTP hydrolysis then allows this stage to evolve towards the docked state with the docking ring . Consistent with this , a discrete category of intermediates was detected among the mixed population of tethered and docked intermediates when GMP-PNP was added two minutes after centrifugation . This category was characterized by apposed membrane regions sandwiching protein densities identical to those in tethered intermediates and regions reminiscent of docked contact areas where outer membranes approached to within less than 3 nm ( Figure 5—figure supplement 1 and Video 3 ) . This hybrid category of attached mitochondria probably represents a transition state from mitochondrial tethering towards mitochondrial docking . 10 . 7554/eLife . 14618 . 015Video 3 . 3D Rendering of tethered intermediate upon addition of GMP-PNP after two minutes , as shown in Figure 5—figure supplement 1 . Outer membranes in red and orange , distinct density in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 015 Of all attached mitochondria , docked intermediates would be the most suitable for outer membrane fusion . The large contact surface and close approach of outer membranes would poise them for fusion . The proportion of docked intermediates increases when outer membrane fusion can proceed but decreases when fusion is inhibited , as we showed by inhibiting GTP hydrolysis . The question of how docking promotes the formation of intermediates with fused outer membranes ( Figure 6—figure supplement 1 ) was nonetheless puzzling . Fusion may occur anywhere within the contact area devoid of visible densities because this area corresponds to regions where membrane contact is closest . Alternatively , the docking ring might trigger fusion over the whole rim of the contact area , similar to the 'vertex ring' that assembles at the edge of docked vacuoles ( Wang et al . , 2002; Wickner , 2010 ) . This ring , composed of SNAREs , small GTPases , tethering factors and lipid microdomains , promotes membrane fusion at the periphery of contact regions between vacuoles , generating a lumenal vesicle that degrades in the fused organelle . In the context of mitochondrial fusion , such an intralumenal outer membrane vesicle would block the subsequent attachment of inner membranes . Indeed , we did not find fused outer membrane intermediates of this kind . However , close inspection of large populations of attached intermediates ( from 87 high magnification micrographs ) by Transmission Electron Microscopy ( TEM ) of stained plastic sections revealed a sub-category of mitochondria in close contact , presumably in the docked configuration , in which the outer membranes were fused near the rim of the contact region ( Figure 6—figure supplement 2A and B ) . While the switch from docking to fusion is bound to be a rapid , transient process and cryo-ET can only sample specimen volumes of the order of 10–7 nanoliters , we succeeded in capturing such a docked intermediate . The tomographic volume shows that the outer membranes of two mitochondria were locally fused to form a toroid , 40 nm pore on one side of the contact area ( Figure 6A ) . The toroid pore formed in the path of the docking ring ( Figure 6B and Video 4 ) . Densities in the vicinity of the fusion pore were sparse , suggesting that in this region the docking ring structure was in the process of disassembly ( Figure 6B ) . These observations provide a proof of principle that the docked intermediates are competent for effective outer membrane fusion , raising the question of which molecular events trigger the transition from docking to outer membrane fusion . 10 . 7554/eLife . 14618 . 016Figure 6 . Stages of outer membrane fusion . ( A and B ) Cryo-ET of docked mitochondria with partially fused outer membrane . ( A ) Slices through tomographic volume; scale bars 100 nm . The red circle highlights the region of outer membrane fusion . ( B ) 3D rendering . Two mitochondria are joined by one continuous outer membrane ( red ) . The inter-membrane spaces are connected by a toroid pore of 40 nm diameter . ( C and D ) Fluorescence microscopy of in vitro outer membrane fusion . GMP-PNP was added at the beginning ( tethering , T ) or at the end ( docking , D ) of the 10 min incubation period . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 01610 . 7554/eLife . 14618 . 017Figure 6—figure supplement 1 . Intermediate with fully fused outer membrane . Top: TEM micrograph of an intermediate with completely fused outer membrane , obtained by in vitro outer membrane fusion of mitochondria isolated from wild-type cells . Lower panel: Zoom of the contact region shows the apposition of inner membranes ( yellow dashed lines ) and continuous fused outer membrane ( blue dashed lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 01710 . 7554/eLife . 14618 . 018Figure 6—figure supplement 2 . Intermediates with partially fused outer membranes . ( A and B ) Left: Electron micrograph of two pairs of attached mitochondria with partially fused outer membranes in stained thin plastic sections of wild-type mitochondria obtained as for Figure 6—figure supplement 1 . Right panels: Zoom of the contact region shows the continuous inner membranes ( yellow dashed lines ) and the cross section through the toroid pore opening at the right ( A ) or the left ( B ) edge of the contact region of outer membranes ( blue dashed lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 01810 . 7554/eLife . 14618 . 019Video 4 . 3D rendering of partially fused mitochondria as shown in Figure 6 . Outer membranes in red , distinct density in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 019 To this end , we compared the effect of GMP-PNP on in vitro fusion efficiency either before tethering ( i . e . after centrifugation; see Figure 5B ) or after docking ( i . e . after 10 min incubation on ice; see Figure 5B ) ( Figure 6C; see Figure 1B ) . Strikingly , the extent of outer membrane fusion impairment was similar ( 40 to 50% ) , irrespective of whether GMP-PNP was added at the beginning or at the end of the 10 min incubation period ( Figure 6D ) . This indicates that it is not important whether the mitochondria were tethered or docked . Hence , GTP hydrolysis is not only required for the transition from tethering to docking but also for the transition from docking to fusion . To further evaluate the functional relationship between mitofusins , mitochondrial docking and mitochondrial fusion , we took advantage of the fact that both the absence or the accumulation of mitofusins inhibits mitochondrial fusion in vivo ( Cohen et al . , 2011; Escobar-Henriques et al . , 2006; Koshiba et al . , 2004 ) . Consistent with this , absence or 50-fold overexpression of Fzo1 ( Figure 7A; fzo1△ and FZO1 o . e . ) totally abolished respiratory growth at 30°C ( Figure 7B ) . 10 . 7554/eLife . 14618 . 020Figure 7 . Overexpression of Fzo1 . ( A ) Total protein extracts of fzo1Δ cells transformed with an empty vector ( fzo1Δ ) , pRS314-FZO1 ( WT ) or pRS414-TEF-FZO1 ( FZO1 o . e . ) were analyzed by anti-Fzo1 and anti-Pgk1 immunoblotting . Fzo1 is overexpressed about 50 fold in FZO1 o . e . as compared to WT conditions . ( B ) Serial dilutions of cells from A grown in the presence of glucose or glycerol as the sole carbon source at 30°C . Lack or overexpression of Fzo1 both abolishes respiration and , therefore , growth on glycerol , consistent with inhibition of mitochondrial fusion . ( C ) Mitochondrial morphology in WT and FZO1 o . e . cells . Left: Representative morphologies . Right: Percentage of WT and FZO1 o . e . cells with indicated mitochondrial morphologies . Error bars represent the s . d . from three independent experiments . ( D ) Left: TEM analysis of in vitro outer membrane fusion reactions performed with mitochondria isolated from wild-type cells or cells overexpressing Fzo1 . Note that mitochondria from Fzo1 o . e . cells are smaller than from wild-type cells . Right: Effect of Fzo1 overexpression on outer membrane fusion and attachment in vitro . ( E ) Slices through tomographic volume of mitochondrial attached intermediates upon Fzo1 overexpression ( abortive , 74% , see Table 1 ) ; outer membrane distance ( blue bracket ) and densities between outer membranes ( red arrows ) are indicated . ( F ) 3D rendering of two closely apposed mitochondria shown in E . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 02010 . 7554/eLife . 14618 . 021Figure 7—figure supplement 1 . Absence or accumulation of Fzo1 . ( A ) Tomographic slice of in vitro attachment reactions with mitochondria isolated from fzo1△ cells; scale bar 100 nm . ( B ) Levels of Ugo1 and Mgm1 in whole-cell extracts prepared from wildtype or Fzo1-overexpressing cells . Ugo1 levels did not vary , which is consistent with an imbalance with mitofusins upon Fzo1 overexpression . Note that the ratio between long and short forms of Mgm1 was slightly shifted toward the short form in cells overexpressing Fzo1 . This may contribute to the changes in cristae morphology upon Fzo1 overexpression as seen in the electron micrographs shown in ( D ) . ( C ) Mitochondrial morphology in representative WT and FZO1 o . e . cells . ( D ) Electron micrographs from Figure 7D at higher resolution . Fusion intermediates ( fused outer membranes , separated inner membranes ) and attached intermediates ( attached outer membranes ) are indicated by red and green arrowheads , respectively . ( E-F ) in vitro mitochondrial attachment upon Fzo1 overexpression . ( E ) Method summary: Mitochondria isolated from cells expressing either mito-GFP or mito-mCherry were mixed in equal amounts and processed for in vitro attachment reactions before analysis by fluorescence microscopy . Attached mitochondria ( red-red; green-green; green-red ) were counted in reactions stopped before centrifugation ( t −10 ) , after centrifugation ( t 0 ) or after 10 min incubation on ice ( t +10 ) . ( F ) Ratios of attached mitochondria from wild-type ( WT , blue ) or Fzo1 overexpressing ( FZO1 o . e . , red ) cells at t −10 , t 0 and t +10 . Ratios were normalized to the WT attachment at t 0 . Mitochondria from Fzo1-overexpressing mitochondria were attached twice as frequently than mitochondria from wild-type cells at all time points , both after and before centrifugation ( t −10 ) . With Fzo1 overexpression , centrifugation stimulated in vitro attachment by a factor of two , compared to wild-type ( t 0 ) . In contrast , incubation on ice ( t +10 ) had a weak effect on attachment in both wild-type and Fzo1 overexpressing conditions ( compare t +10 with t 0 ) , consistent with the requirement of this step for the transition from tethering to docking rather than for de novo attachment in vitro . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 02110 . 7554/eLife . 14618 . 022Figure 7—figure supplement 2 . Cryo-ET of mitochondria from Fzo1-overexpressing cells . ( A–C ) Minor population of attached intermediates ( Other , 26% , see Table 1 ) . ( A ) Slice through tomographic volume . ( B ) Stack representation of slices through tomographic volume with details of mitochondrial contacts . One out of 20 slices is shown . Red arrows , interstitial densities . All scale bars 100 nm . ( C ) 3D rendering of two closely apposed mitochondria . Rings of densities around the edge of contact areas are less extensive than in wild-type . ( D–E ) Cryo-ET of major population of attached intermediates ( Abortive , 74% , see Table 1 ) ( D ) Histogram of closest outer membrane distances in attached mitochondria . ( E ) Slices through tomographic volume of the mitochondria shown in Figure 7F . One out of 20 slices is shown . Red arrows , interstitial density between outer membranes . All scale bars 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 022 In the absence of Fzo1 , cryo-ET analysis revealed small mitochondria well separated from each other but mitochondrial attachment was not detected ( Figure 7—figure supplement 1A ) , which is consistent with the essential function of mitofusins in mitochondrial anchoring ( Cohen et al . , 2011; Koshiba et al . , 2004 ) . On the other hand , the fusion defect caused by the overexpression of Fzo1 may result from the accumulation of Fzo1 molecules on outer membranes and an imbalance with other proteins implicated in mitochondrial fusion , such as Ugo1 and Mgm1 ( Figure 7—figure supplement 1B ) . Notably , Fzo1 overexpression also induces a specific phenotype of mitochondrial aggregation ( Figure 7C ) in which mitochondrial puncta aggregate in one region of the cell cortex ( Figure 7—figure supplement 1C ) . This phenotype suggests that , as well as inhibiting fusion , overexpression of Fzo1 may also promote the attachment of mitochondria to each other . To verify this , in vitro fusion assays were performed using mitochondria isolated from cells overexpressing Fzo1 . As expected , in vitro outer membrane fusion was strongly inhibited but mitochondrial attachment was significantly increased , with 50% of mitochondria attached to others ( Figure 7D and Figure 7—figure supplement 1D ) . Consistent with their compromised fusing ability , mitochondria from Fzo1-overexpressing cells were significantly smaller than controls . Further analysis indicated that Fzo1 overexpression induced a two-fold increase in mitochondrial attachment both before and after centrifugation ( Figure 7—figure supplement 1E–F ) . Thus , mitochondria purified from Fzo1-overexpressing cells retained an increased attachment capacity , indicating that the in vivo aggregation phenotype ( Figure 7C ) is caused , at least in part , by perturbations that are intrinsic to outer membranes . To better characterize these perturbations , attached mitochondria were analyzed by cryo-ET . While intermediates with a docking ring were formed ( Figure 7—figure supplement 2A–C ) , they were observed less frequently ( 26% ) and the average contact surface was significantly smaller compared to docking intermediates observed under wild-type conditions ( Table 1 ) . In the remaining attached intermediates ( 74% ) , the surface of apposition was also reduced ( 4970 ± 930 nm² ) , with outer membranes on average 8 . 8 ± 0 . 4 nm apart ( Figure 7E; blue bracket; Figure 7—figure supplement 2D–E ) . Importantly , numerous densities accumulated in regions of closest contact between mitochondria ( Figure 7F ) , but the densities were disorganized and did not form globular protein repeats ( Figure 7E–F ) . These attached mitochondria , which were clearly distinct from tethered intermediates observed under wild-type conditions , likely correspond to artefactual and abortive fusion intermediates . These results indicate that while normal levels of Fzo1 are required for the formation of bona fide docking rings , overexpression of the mitofusin induces the formation of protein aggregates that perturb the regulated sequence of events required to reach productive mitochondrial docking . Densities detected at the junctions of tethered and docked mitochondria correspond to protein complexes responsible for promoting productive attachment of outer membranes prior to their fusion . Several protein factors either within or extrinsic to the mitochondrial outer membrane may assemble into such complexes ( Coonrod et al . , 2007; Hoppins et al . , 2009; Sesaki and Jensen , 2001; 2004 ) . Notably , the units with a defined central density and protrusions extending towards each lipid bilayer bridged the apposed outer membranes ( Figure 8A ) . These units were stabilized in the presence of GMP-PNP ( Figure 4 ) and their arrangement may reflect the self-association in trans of a factor extruding from outer membranes ( Figure 8B ) . These features thus point to a transmembrane GTPase specialized in connecting outer membranes , raising the possibility that the observed units are indeed trans-oligomers of Fzo1 molecules . 10 . 7554/eLife . 14618 . 023Figure 8 . Fzo1 enrichment at mitochondrial junctions . ( A ) Cryo-ET of tethered mitochondria . Slices through tomographic volumes and zooms on contact regions ( red and blue boxes ) ; Outer membranes on zooms are delimited by red bars; scale bars 10 nm . ( B ) Scheme representing the central density ( yellow circle ) and extensions ( blue lines ) to outer membranes ( red bars ) for densities detected at the junction of tethered intermediates . ( C–E ) Q-Dot labelling of FZO1 and FZO1-AVI mitochondria . ( C ) Control with non-labelled Fzo1 . Tomographic slice and zoom with Q-Dots on the outer membrane . ( D ) Q-Dots per contact area ( left ) or on mitochondrial surface excluding the contact area ( right ) for FZO1 ( white ) and FZO1-Avi ( black ) . ( E ) Slices through tomographic volumes of Fzo1-Avi mitochondria labelled with Q-Dots ( yellow arrows ) . Zooms are indicated by dashed red boxes . Scale bars 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 02310 . 7554/eLife . 14618 . 024Figure 8—figure supplement 1 . Biotinylation and Q-Dot labelling of Fzo1-Avi . In vitro biotinylation ( orange dots ) and Q-Dot labelling ( green hexagons ) for outer membrane attachment assays with mitochondria containing mito-mCherry ( red matrix ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14618 . 024 The presence and active involvement of mitofusins in mitochondrial tethering and docking is not only consistent with the established function of these DRPs in attachment and fusion of outer membranes ( Cohen et al . , 2011; Hermann et al . , 1998; Ishihara et al . , 2004; Koshiba et al . , 2004; Legros et al . , 2002; Shutt et al . , 2012 ) but also with their documented accumulation at mitochondrial junctions ( Hoppins et al . , 2009 ) . To validate Fzo1 as a potential component of the densities found at mitochondrial contact sites , we thus devised an in situ protein labeling strategy for cryo-ET . Recent experiments to label mitochondria for cryo-ET with a purified , biotin-labelled protein import substrate conjugated with streptavidin-coated Quantum Dots have been successful ( Gold et al . , 2014 ) . To biotinylate mitofusin molecules in situ , Fzo1 was C-terminally fused to the Avi tag , a 15 amino-acid peptide that is recognized and can undergo specific biotinylation by the E . coli biotin ligase BirA ( Beckett et al . , 1999; van Werven and Timmers , 2006 ) . Mitochondria isolated from wild-type ( FZO1 ) or FZO1-Avi cells were biotinylated in vitro with recombinant BirA before processing for outer membrane attachment assays , followed by incubation with streptavidin-coupled Q-Dots to label the Fzo1-Avi construct ( Figure 8—figure supplement 1 ) . While Q-Dots were rarely found on the surface of tagged or untagged mitochondria ( Figure 8C and D , right graph ) , they were enriched at the junction of attached FZO1-Avi mitochondria as compared to attached FZO1 mitochondria ( Figure 8D , left graph ) . Moreover , Q-Dots , and therefore the FZO1-Avi constructs , were located either at the periphery of docked contact sites or between tethered contact sites ( Figure 8E ) . These data confirm the accumulation of mitofusins at mitochondrial contact sites and corroborate the likely contribution of Fzo1 to densities found at junctions of tethered and docked intermediates . Initially outer membranes of two attached mitochondria are tethered by globular protein repeats and membranes approach to within 6 nm ( Figure 9 , step 1 ) . These tethered intermediates are seen in wild-type conditions ( Figure 3 ) and accumulate upon addition of GMP-PNP ( Figure 4 ) . Subsequently , GTP hydrolysis allows the fusion process to evolve progressively towards mitochondrial docking ( Figure 5 ) , as evidenced by the hybrid intermediates observed upon addition of GMP-PNP two minutes after initiation of tethering ( Figure 5—figure supplement 1 ) . Consistent with this , GTP hydrolysis by trans-complexes of atlastins precedes vesicle fusion and is essential to promote tethering of proteoliposomes ( Liu et al . , 2015; Saini et al . , 2014 ) . The docked intermediates are characterized by a docking ring of protein density that surrounds extended areas with outer membranes separated by less than 3 nm and devoid of visible densities ( Figure 9 , step 2 ) . The capture of docked intermediates with partly fused outer membranes , the location of the fusion pore in the path of the docking ring undergoing disassembly and the inhibition of fusion upon treatment of docked mitochondria with GMP-PNP , demonstrates that docking is the stage that precedes merging of outer membranes ( Figure 6 ) . We do not exclude that small protein complexes , which are not visible by cryo-ET , might reside between outer membranes and participate in the fusion process . However , our observations do suggest that the docking ring of protein densities is the driving force for subsequent bilayer merging . We thus propose that the fusion of bilayers is initiated by further GTP hydrolysis in the path of the docking ring where the outer membrane curvature is most pronounced ( Figure 9 , step 3 ) . This step may trigger the disassembly of the docking ring . Tethering ( Figure 9 , step 4 ) and fusion of the inner membrane ( Figure 9 , step 5 ) mediated by OPA1/Mgm1 then completes the mitochondrial fusion process . Taking into account the essential role of mitofusins in mitochondrial attachment and fusion ( Hermann et al . , 1998; Koshiba et al . , 2004 ) , we obtained several lines of evidence that Fzo1 contributes to the formation of the macromolecular assemblies we discovered at mitochondrial junctions . Overexpression of Fzo1 inhibited mitochondrial docking and fusion but stimulated the formation of artefactual tethering intermediates that are characterized by an accumulation of protein aggregates at mitochondrial junctions ( Figure 7 ) . This result demonstrates once more that absence of docking correlates with deficient mitochondrial fusion and implies that normal levels of mitofusins are required for the formation of the docking rings . It is also important to realize that increased levels of mitofusins correlate with the accumulation of protein densities at mitochondrial junctions . However , whether these aggregates correspond to abortive mitofusin oligomers remains speculative . GMP-PNP that may bind to mitofusins and inhibit their GTPase activity ( Amiott et al . , 2009; Ishihara et al . , 2004 ) , induced the accumulation of tethered intermediates and prevented progression towards the docked stage ( Figure 4 ) . Moreover , Fzo1 enrichment at mitochondrial contact sites was confirmed by Q-dot labeling ( Figure 8 ) . To our knowledge , no other GTPase has been shown to be involved in outer membrane fusion . In this context , our observations converge to propose that Fzo1 is at least a component of globular protein repeats and docking rings , possibly together with other , as yet unknown factors . In fact , the units composed of a central density with protrusions extending toward each outer membrane suggest that they may be Fzo1 trans-oligomers ( Figure 8A ) . So far , the only structural insight into mitofusins derives from the HR2 domain of Mfn1 that has been proposed to tether outer membranes at a distance of 16 nm ( Koshiba et al . , 2004 ) . This would exceed the 6 nm membrane spacing we measured in tethered intermediates . The overall morphology of the HR2 dimer can hardly account for the distinctive shape of the protein units we observed . However , the structure of the bacterial Fzo1 homolog BDLP in the open conformation ( Low and Löwe , 2006; Low et al . , 2009 ) and the established trans-interaction of atlastins through their GTPase domain ( Klemm et al . , 2011; Byrnes et al . , 2013; Liu et al . , 2015; Saini et al . , 2014 ) allow us to propose an alternative model . Similar to BDLP , mitofusins bound to GMP-PNP or GTP may adopt an open conformation that , in analogy to atlastins , would promote trans-oligomerization of Fzo1 molecules through their GTPase domain . Conformational changes of the mitofusin oligomers upon GTP hydrolysis would pull the outer membranes closer together . This model is not only consistent with the shape of the protein units we visualized but also with the essential requirement of GTP hydrolysis for transition from mitochondrial tethering to mitochondrial docking we unraveled in this study . Hence , the structure of mitofusins with or without bound GTP will be instrumental to evaluate this model . An absolute prerequisite for mitochondrial fusion in vivo is that the tips of mitochondrial tubules come close enough to promote attachment between outer membranes . In yeast cells , this requires the participation of actin filaments ( Simon et al . , 1995; Smith et al . , 1995 ) . In the in vitro fusion system , the essential role of the cytoskeleton is replaced by centrifugation to bring mitochondria into close enough contact for fusion to proceed ( Meeusen et al . , 2004 ) . However , centrifugation is not sufficient , and an incubation of at least 10 min on ice , previously suggested to promote Fzo1 association in trans , has been shown to be essential for in vitro fusion ( Meeusen et al . , 2004 ) . Our results reveal that incubation on ice promotes the transition from tethering to docking , which is an active process that requires several rounds of GTP hydrolysis to progressively bring opposing outer membranes closer over an extended surface area but is not sufficient for their effective fusion . The observation that fusion starts at the edge of the docking ring and also depends on GTP hydrolysis suggests that the transition from tethering to docking brings membranes closer over an extended area . This would induce a locally increased curvature of the lipid bilayer , which may be critical . An ultimate cycle of GTP hydrolysis in this region of local membrane curvature would therefore lead to fusion instead of bringing membranes closer together . Notably , micrographs taken from gastric mucosa of a mole , featured in Chapter 7 of Don W . Fawcett’s 'The Cell' , present a series of three pairs of mitochondria proposed to represent successive stages of mitochondrial fission ( Fawcett , 1981 ) . The middle stage intermediate from this series and the fusion events identified in our study actually appear strikingly similar . This not only suggests that the sequence of events shown in this chapter might in fact correspond to successive stages of fusion but also that our model of outer membrane fusion may apply in vivo with mitochondria from mammalian cells . Whereas in vitro local membrane deformation would be promoted exclusively by successive cycles of GTP hydrolysis after mitochondria become tethered through centrifugation , in vivo the transition from tethering to docking would also involve the action of the cytoskeleton . In the cell , the process of mitochondrial fusion as indicated by our in vitro studies would thus be regulated by cytoskeletal factors . Regulation of mitochondrial fusion in vivo also involves post-translational modification of mitofusins ( Anton et al . , 2011; Cohen et al . , 2008; Shutt et al . , 2012 ) . In yeast , the efficient fusion of outer membranes requires Fzo1 ubiquitylation by the Mdm30 ubiquitin ligase and subsequent degradation by the proteasome ( Cohen et al . , 2008 ) . While its precise function is yet to be fully characterized , this regulation was previously shown to take place at the stage of mitochondrial attachment ( Cohen et al . , 2011 ) . It is therefore conceivable that the UPS-dependent regulation of Fzo1 participates in regulating proper assembly of docking rings , which is consistent with the observation that high levels of Fzo1 inhibit mitochondrial docking ( Figure 7 ) . At this stage , we cannot exclude that ubiquitylation and degradation of Fzo1 regulates the transition from docking to effective fusion of outer membranes . Further investigations will be required to answer the fascinating question of how mitochondrial fusion is regulated . Our work provides a detailed dissection of the outer mitochondrial membrane fusion process in vitro and highlights the crucial involvement of Fzo1 in this system . Similar mechanisms involving atlastins or OPA1/Mgm1 are likely to be active in the fusion of ER and mitochondrial inner membranes , respectively . Future challenges include deciphering the precise composition of the complexes mediating outer membrane fusion , and dissecting the processes regulating their formation and function . The S . cerevisiae strains and plasmids are listed in Supplementary file 1 . Standard methods were used for growth , transformation and genetic manipulation of S . cerevisiae . Complete media and minimal synthetic media [Difco yeast nitrogen base ( Voigt Global Distribution Inc; Lawrence , KS ) , and drop-out solution] supplemented with 2% dextrose ( YPD and SD ) or 2% glycerol ( YPG and SG ) were prepared as described ( Sherman et al . , 1986 ) . Mitochondrial fractions for in vitro attachment and fusion were prepared as previously described ( Ingerman et al . , 2007 ) . Cells were cultured to stationary phase in dextrose medium and then shifted to glycerol medium ( or fresh dextrose medium for cells affected in respiration ) to grow to a final OD 0 . 8–1 . 0 . Cell walls were disrupted by incubation at 30°C with 100 mM Tris-HCl pH 9 . 4 and 50 mM β-mercaptoethanol for 20 min and subsequently with 1 . 2 M sorbitol plus zymolyase ( Zymo Research; Irvine , CA ) for 30 min . The resulting spheroplasts were lysed in cold NMIB ( 0 . 6 M sorbitol , 5 mM MgCl2 , 50 mM KCl , 100 mM KOAc , 20 mM Hepes pH 7 . 4 ) by douncing . After centrifugation at 4°C of the lysate at 3000 x g for 5 min , the supernatant was further centrifuged at 10170 x g , 4°C for 10 min to yield a pellet enriched in mitochondria . Protein concentration in mitochondria-enriched fractions was determined by Bradford assay ( Bio-Rad Protein Assay; Bio-Rad Laboratories GmbH , Germany ) . Homotypic and heterotypic attachment/fusion reactions were respectively carried out with 0 . 5 mg of type 1 mitochondria or by mixing 0 . 25 mg of type 1 mitochondria with 0 . 25 mg of type 2 mitochondria . Mitochondria were then brought in vicinity by centrifugation at 10170 x g , 4°C for 10 min . Pellets were left on ice for 10 min before the supernatant was replaced by Stage 1 buffer ( 20 mM Pipes–KOH pH 6 . 8 , 150 mM KOAc , 5 mM Mg ( OAc ) 2 , 0 . 6 M sorbitol ) . Mitochondria were then left on ice for 30 min in attachment assays or incubated at 25°C for 45 min in outer membrane fusion assays . Resulting attachment and fusion reactions were subsequently processed for TEM , cryo-ET or fluorescence microscopy analysis . Depending on experiments , discrete variations in attachment/fusion reactions described above were used . To assess the effect of GMP-PNP on attachment of wild-type mitochondria ( Figure 4 ) , GMP-PNP ( 1 . 5 mM , Sigma-Aldrich; St Louis , MO ) was added during mixing of mitochondria ( prior centrifugation ) and was kept at constant concentration during the whole attachment reactions ( including in Stage 1 buffer ) . In contrast , 6 mM GMP-PNP were added where indicated in time course experiments ( Figure 5 ) . Similarly , 12 mM GMP-PNP were added at tethering or docking stages in outer membrane fusion reactions shown in Figure 6C . To evaluate effects of Fzo1 overexpression on attachment and fusion , in vitro assays were performed with mitochondria-enriched fractions prepared from wild-type ( MCY553 ) or Fzo1-overexpressing cells ( MCY1222 ) grown in dextrose medium . For analysis of attachment and outer membrane fusion intermediates by TEM , the fusion reactions were mixed with 20 volumes of fixative solution ( 3% paraformaldehyde , 1 . 5% glutaraldehyde , 2 . 5% sucrose , 100 mM sodium cacodylate , pH 7 . 4 ) for 20 hr and subsequently washed twice with 100 mM sodium cacodylate , pH 7 . 4 . To improve contrast , samples were post-fixed with 1% osmium tetroxide , 100 mM sodium cacodylate , pH 7 . 4 , for 1 hr , after which samples were washed in distilled water . Then samples were embedded in 4% agar and washed with 50 mM acetate buffer pH 5 . 2 before they were stained with 1% uranyl acetate overnight ( 12 h ) at 4°C . For plastic embedding , the samples were dehydrated in an ethanol gradient series ( 1 × 20 min 30% , 2 × 20 min 50% , 2×30 min 70% , 2 × 30 min 90% , 1 × 60 min 100% ethanol ) followed by a switch to 1 , 2 propylenoxid ( 3 x 20 min 100% ) and subsequently infiltrated using the Low Viscosity Premix Kit-Medium ( Agar Scientific , England; 2 × 20 min 30% , 2 × 30 min 50% , 2 × 30 min 75% , overnight 100% , 2 × 2 h 100% ) . Polymerization was carried out at 65°C for 16 h . Thin sections ( 60–70 nm ) were prepared with an Ultracut S microtome ( Reichert , Germany ) , collected on 100 mesh copper grids coated with Pioloform FN 65 ( Wacker Polymer Systems GmbH , Germany ) and were double stained with 2% uranyl acetate for 2 min and lead citrate for 1 min . Sections were inspected with a transmission electron microscope ( EM 208S; FEI , Germany ) at 80 kV equipped with 2 x 2 k CCD camera ( Gatan , Inc; Pleasanton , CA ) . Ratios of attached and fused mitochondria were obtained by dividing the number of mitochondria in physical contact or those with fused outer membranes over the total number of mitochondria ( n > 300 ) . The fused intermediates shown in Figure 6—figure supplement 1 and 2 were obtained from the analysis of 87 high magnification micrographs containing one pair of attached or fused intermediates each . For cryo-ET , mitochondria were washed twice with 320 mM trehalose , 20 mM Tris pH 7 . 3 , 1 mM EGTA . Samples were mixed 1:1 with fiducial gold markers ( 6 or 10 nm gold particles conjugated to protein A , Aurion , Netherlands ) and immediately plunge-frozen in liquid ethane on Quantifoil holey carbon grids ( Quantifoil Micro Tools , Germany ) . Single tilt series ( typically ± 60° , step size 1 . 5–2 . 0° ) were collected at 300 kV with an FEI Polara or FEI Titan Krios electron microscope equipped with a post-column Quantum energy filter and a K2 Summit direct electron detector ( Gatan ) at 6 µm underfocus . Magnifications were chosen to give an object pixel size of 3 . 5 Å or 3 . 3 Å , respectively . The total electron dose per tilt series was 90–120 e-/Å² . If dose fractionation was used , frames were aligned using the IMOD software package ( Kremer et al . , 1996 ) . Tilt series were aligned to gold fiducial markers , binned 2-fold and tomograms were reconstructed by back-projection with IMOD . A final filtering step applying non-linear anisotropic diffusion ( Frangakis and Hegerl , 2001 ) was performed to increase contrast . Tomograms were manually segmented with the program Amira ( FEI ) . Distances in tomographic data were analyzed using IMOD . The following parameters were measured for each contact: the radii r of mitochondria perpendicular and parallel to the contact area , the major axis la and the minor axis lb of the contact area assuming elliptical geometry and the distance between the outer membranes d . The contact area A was calculated assuming it to be elliptical , A = ( πlalb ) /4 . The normalized contact area ratio R was calculated , taking the radius r of the smaller of the two involved mitochondria into account , R = la/ ( 2r ) . For fluorescence microscopy analysis , the fusion reactions were fixed with two volumes of 8% formaldehyde in phosphate-buffered saline ( PBS ) . Aliquots were immobilized on microscope slides by mixing 1:1 with 2% low melting point agarose ( Sigma-Aldrich ) in NMIB . The ratios of fused mitochondria were obtained by dividing the number of GFP and mCherry signals co-localizing with each other over the total number of OM45-GFP ( obtained from strain #779 ) and mito-mCherry ( obtained from strain #980 ) mitochondria ( n > 1000 ) . The levels of fused mitochondria were then determined by subtracting the ratios obtained for fusion reactions stopped at the mixing step ( at the beginning of the reaction ) from those obtained from reactions stopped at t = 45 min ( at the end of the reaction ) . Cells grown in SD medium were collected during the exponential growth phase ( OD600 = 0 . 5–1 ) . Total protein extracts were prepared by the NaOH/trichloroacetic acid ( TCA ) lysis technique ( Volland et al . , 1994 ) . Proteins were separated by SDS-PAGE 8% and transferred to nitrocellulose membranes ( AmershamTM HybondTM-ECL; GE Healthcare , UK ) . Primary antibodies for immunoblotting were mouse anti-Pgk1 ( clone 22C5D8; Abcam , UK ) , rabbit anti-Fzo1 ( Covalab , France ) , anti-Ugo1 ( Covalab ) , anti-Mgm1 ( kind gift from Andreas Reichert ) and mouse anti-Por1 ( clone 16G9E6BC4; Abcam ) . Primary antibodies were detected by secondary anti-mouse or anti-rabbit antibodies conjugated to horseradish peroxidase ( HRP , Sigma-Aldrich ) , followed by incubation with the Clarity Western ECL kit ( Bio-Rad ) . Immunoblotting images were acquired with a Gel DocTM XR + ( Bio-Rad ) and processed with the Image Lab 3 . 0 . 1 software ( Bio-Rad ) . Cultures grown overnight in SD medium were pelleted , resuspended at OD600 = 1 , and serially diluted ( 1:10 ) five times in water . Three microliters of the dilutions were spotted on SD and SG plates and grown for 3 days ( dextrose ) or 5 days ( glycerol ) at 30°C . Mitochondrial morphology was analyzed in cells expressing mito-GFP from the p426-TEF-mitoGFP plasmid ( MC244 ) . Strains were grown in dextrose medium to mid-log phase and fixed with 3 . 7% formaldehyde . Morphology phenotypes were assessed in at least 100 cells . Data reported are the mean and standard deviation ( SD , error bars ) of three independent experiments . Fluorescence microscopy was carried out with a Zeiss Axio Observer Z1 microscope ( Carl Zeiss Microscopy GmbH , Germany ) . For in vitro fusion assays a 100X oil immersion objective and the following filter sets were used: 10 Alexa Fluor 489 ( Excitation BP 450–490 , Beam Splitter FT 510 , Emission BP 515–565 ) for GFP , mCherry ( Excitation BP 542–582 , Beam Splitter FT 593 , Emission BP 604–679 ) for mCherry . Images were acquired with an SCMOS ORCA FLASH 4 . 0 charge-coupled device camera ( Hamamatsu Photonics K . K . , Japan ) and the Zen 2012 Package Acquisition/Analyse software before processing with ImageJ . Mitochondrial morphology was analyzed with a 63X oil immersion objective and an FITC filter ( Filter set 44 , Excitation BP 475/40 , Beam Splitter FT 500 , Emission BP 530/50 ) for GFP . Cell contours were visualized with Nomarski optics . Images were acquired with an ORCA-R2 CCD camera ( Hamamatsu ) and processed with ImageJ . Specific labelling of mitofusins on attached mitochondria was achieved using Quantum Dots coupled to streptavidin ( QDot 525ITK streptavidin; Life Technologies; Carlsbad , CA ) , which required specific biotinylation of Fzo1 . For this purpose , the FZO1 ORF placed under control of its own promoter , was fused at its 3’-end in tandem with a sequence encoding for the Avi tag , a 15 amino-acid peptide that can be recognized and undergo specific biotinylation by the E . coli biotin ligase BirA ( Beckett et al . , 1999; van Werven and Timmers , 2006 ) . The resulting pFZO1-AVI plasmid ( MC369 ) was introduced in fzo1△ cells using a plasmid shuffling strategy to yield the FZO1-AVI yeast strain ( MCY1155 ) that expresses Fzo1-Avi as the sole source of mitofusin in the cell . After verifying that Fzo1-Avi is competent for mitochondrial fusion in vivo , mitochondria were purified to promote biotinylation of Fzo1-Avi in vitro . Briefly , 0 . 5 mg of total mitochondrial enriched fractions prepared from FZO1 ( MCY1154 ) or FZO1-Avi ( MCY1155 ) cells were incubated at 25°C for 60 min with 10 µg BirA and 1X biotin in biomix A + B buffer ( Biotinylation kit purchased from GeneCopoeia Inc; Rockville , MD ) before processing for attachment assays . Following 30 min incubation on ice in Stage 1 buffer , attachment reactions were incubated for 60 min at 4°C with 50 nM streptavidin-coupled QDs before processing for cryo-ET analysis .	Yeast and other eukaryotic cells contain distinct compartments that have specific roles . For example , compartments called mitochondria – which are surrounded by two layers of membrane – provide the energy needed for many cell processes . The organization of the network of mitochondria in a cell has a large effect on their capacity to provide energy . Mitochondria can fuse together to make larger compartments or divide to make smaller ones . Defects in fusion or division of mitochondria can reduce the amount of energy that is provided , which , in humans and animals can lead to diseases that affect various organs , especially those in the nervous system . When two mitochondria fuse they must first attach to each other and then merge their outer membranes . Proteins called mitofusins are known to be involved in these processes , but the molecular details of how they take place were not clear . Brandt , Cavellini et al . investigated how mitochondria isolated from budding yeast cells attach to each other . The experiments found that two mitochondria first become loosely attached by mitofusins . These proteins then promote a tighter attachment in which the outer membranes of the two mitochondria come into contact over a larger area . This contact area is determined by a linear arrangement of proteins referred to as the docking ring . Brandt , Cavellini et al . further observed that local fusion between the outer membranes takes place at the edge of the contact area in the path of the docking ring . Future research will need to address how mitochondria attach to each other in living cells and how the process is regulated .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	A mitofusin-dependent docking ring complex triggers mitochondrial fusion in vitro
The budding yeast centromere contains Cse4 , a specialized histone H3 variant . Fluorescence pulse-chase analysis of an internally tagged Cse4 reveals that it is replaced with newly synthesized molecules in S phase , remaining stably associated with centromeres thereafter . In contrast , C-terminally-tagged Cse4 is functionally impaired , showing slow cell growth , cell lethality at elevated temperatures , and extra-centromeric nuclear accumulation . Recent studies using such strains gave conflicting findings regarding the centromeric abundance and cell cycle dynamics of Cse4 . Our findings indicate that internally tagged Cse4 is a better reporter of the biology of this histone variant . Furthermore , the size of centromeric Cse4 clusters was precisely mapped with a new 3D-PALM method , revealing substantial compaction during anaphase . Cse4-specific chaperone Scm3 displays steady-state , stoichiometric co-localization with Cse4 at centromeres throughout the cell cycle , while undergoing exchange with a nuclear pool . These findings suggest that a stable Cse4 nucleosome is maintained by dynamic chaperone-in-residence Scm3 . In all eukaryotes , accurate segregation of genetic material constitutes the basis of cell division and inheritance . Chromosome segregation is controlled by a complex signalling network targeting the kinetochore—a protein superstructure of some 100 polypeptides , anchoring chromosomes to the mitotic spindle through interaction with a specialized region of the chromosome , the centromere . The chromatin structure of centromeres is distinguished from other chromosome regions by nucleosomes containing a distinct variant of histone H3 , called CENP-A or CenH3 ( Biggins , 2013; Westhorpe and Straight , 2013 ) . Unlike other organisms , in which centromeres encompass extended regions with tens or thousands of CENP-A nucleosomes , the centromere of the budding yeast Saccharomyces cerevisiae is fully specified by a short DNA segment ( CEN , ∼125 bp ) ( Gaudet and Fitzgerald–Hayes , 1987; Murphy et al . , 1991 ) . This so-called ‘point’ centromere consists of a single nucleosome-like chromatin particle containing Cse4 , the yeast ortholog of CENP-A ( Stoler et al . , 1995; Meluh et al . , 1998 ) . Classic genetic , molecular , and biochemical studies have defined three contiguous centromeric DNA elements CDEI , CDE II , and CDE III that direct assembly of a Cse4 nucleosome by sequence-specific DNA binding factors CBF1 and CBF3 ( Cai and Davis , 1990; Lechner and Carbon , 1991 ) and Scm3 , a Cse4-specific chaperone ( Camahort et al . , 2007; Mizuguchi et al . , 2007; Stoler et al . , 2007; Xiao et al . , 2011; Cho and Harrison , 2011b ) . The singular nature of centromeric nucleosomes of budding yeast thus offers a simplified biological system for detailed study of the biogenesis , maintenance , and dynamics of centromere–kinetochore interactions . Despite this simplicity , the architecture of Cse4 nucleosomes has become the subject of much debate . Cse4 nucleosomes have been reported to differ from the canonical nucleosome not only by the replacement of both molecules of histone H3 by the Cse4 variant , but also by the presence of chaperone Scm3 and dislocation of histones H2A-H2B ( Mizuguchi et al . , 2007; Xiao et al . , 2011 ) , or by existence of a hemisome particle bearing half the histone content ( Dalal et al . , 2007; Furuyama et al . , 2013 ) . Moreover , live cell microscopy of GFP-tagged Cse4 have variously indicated that the number of Cse4 molecules associated with centromeres may be either several-fold greater than the two Cse4 molecules within a nucleosome ( Coffman et al . , 2011; Lawrimore et al . , 2011 ) , or oscillate during mitosis from one to two molecules per centromeric nucleosome ( Shivaraju et al . , 2012 ) . Thus , the fundamental composition and stability of the Cse4 nucleosome has been obfuscated by the recent microscopic studies . To assess those claims , we have taken a direct approach to monitor the fate of Cse4 molecules throughout the cell cycle in live yeast . We utilize the photoconvertible fluorescent protein tdEos in fluorescence pulse-chase experiments to mark pre-existing Cse4 and document its complete replacement at centromeres with newly synthesized molecules early in S phase . We find that after this transient replacement , Cse4 remains stably associated with centromeres for the rest of the cell cycle , without additional Cse4 deposition in anaphase . Importantly , we show that recent discrepant claims can be attributed to reliance on GFP fusion to the C-terminus of Cse4 , which causes impaired cell growth , temperature-dependent lethality , and extra-centromeric nuclear accumulation . By contrast , an insertion of GFP or tdEOS within the unstructured N-terminal tail of Cse4 avoids such deleterious phenotypes . Hence , many of the conflicting properties of C-terminally tagged Cse4 reflect the behavior of functionally impaired protein rather than native Cse4 . To analyze the localization of Cse4 in live cells , we introduced a fluorescent protein tag at an internal Xba I site ( corresponding to Leu81 within the long N-terminal tail of Cse4 ) based on the original studies of Stoler et al . ( 1995 ) and Chen et al . ( 2000 ) ( Figure 1A ) . These workers showed that insertions or deletions within the N-terminal tail do not impose any deleterious growth phenotype at all tested temperatures , as long as a 33-residue essential END domain , that interacts with the Ctf19-Mcm21-Okp1 kinetochore sub-complex , is preserved . Thus , as schematically depicted in Figure 1B , the flexible N-terminus of Cse4 is well suited to accommodate internal protein tags . In contrast , the extreme C-terminal residues of Cse4 ( QFI , aa 227-229 , located close to the structured part of the nucleosome [Tachiwana et al . , 2011] ) mediate recognition by CENP-C ( Kato et al . , 2013 ) and an adjoining tag is likely to impair this interaction . Moreover , functionality of Drosophila CENP-A/CenH3 is also preserved by an internal insertion of GFP but not by a C-terminal fusion ( Schuh et al . , 2007 ) . 10 . 7554/eLife . 02203 . 003Figure 1 . Internal tagging of Cse4 confers exclusive centromeric localization and preserves wild type phenotype . ( A ) Alternative tag locations at Leu81 ( internal XbaI site ) or at the C-terminus of Cse4 are indicated by green triangles . Unstructured N-terminal tail ( aa1-135 ) is depicted in grey while region corresponding to the known 3D structure of mammalian CENP-A ( aa134-226 ) is shown as solid black and red ( loops and α-helices of histone-fold domain ) . Functionally important END region ( aa28-60 ) and C-terminal CENP–C interaction region QFI ( aa227-228 ) are highlighted in blue and yellow . ( B ) Schematic position of fluorescent protein tags in relation to the overall nucleosome structure . Monomeric GFP tag is shown in green while Cse4 histone-fold domains are highlighted in red inside nucleosome core . Unstructured N-terminal tails of Cse4 are depicted as dashed lines for illustrative purposes . ( C–E ) Distribution of tagged Cse4 in live cells containing Cse4 tagged internally with GFP ( C ) or tdEos ( D ) , or the C-terminal GFP fusion ( E ) . Cell cycle stages are indicated in DIC panels . In addition to G1 , S , and G2 , individual stages of mitosis are identified as: M—metaphase , A—anaphase , T—telophase . Fluorescence images are shown as negatives to reveal residual intracellular autofluorescence and the diffuse nuclear presence of C-terminally tagged Cse4 . ( F ) Viability test of strains containing wild-type or tagged Cse4 . 10 μl of 10-fold serial dilutions of equivalent log-phase cultures were spotted on YPD plates and incubated overnight at 38°C or for 36 hr at 24°C . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 00310 . 7554/eLife . 02203 . 004Figure 1—figure supplement 1 . Fluorescence of centromeric clusters containing C-terminally tagged Cse4 is slightly elevated but does not double in anaphase . ( A ) Representative examples of metaphase and telophase cells ( outlined ) shown at the same brightness scale . Clusters in surrounding cells may be out of focus . ( B ) Intensity of individual centromeric clusters in metaphase ( M ) and telophase ( T ) was measured by photometry ( 5 s exposure ) . Minimum/1st quartile/median/3rd quartile , and maximum values are displayed for each group of 20 measurements . Prior to measurement , clusters were separated from lower frequency components of the image ( diffuse fluorescence present in nuclei and intracellular autofluorescence ) by processing the image with wavelet function and adding together scales 1+2+3 ( 1 , 2 and 4 pixels across ) ( see Figure 7—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 00410 . 7554/eLife . 02203 . 005Figure 1—figure supplement 2 . Internally tagged Cse4 accumulates at levels comparable to wild-type protein . ( A ) Whole-cell lysates of cells containing wild-type or tagged Cse4 were probed with affinity-purified anti-Cse4 antibody . Arrows indicate bands corresponding to wild-type Cse4 ( 27 kDa ) and Cse4 fused to GFP ( 57 kDa ) or tdEos ( 87 kDa ) . A control blot probed for histone H4 is shown below . ( B ) Anti-Cse4 antibody was used to estimate the level of endogenous wild-type Cse4 ( present in WT lysate ) by comparison to indicated amounts of purified recombinant H6-Cse4 introduced into a lysate of the strain carrying Cse4-GFPinternal . The * denotes a common background band used to normalize different lysates; see below . The presence of the Histidine-tag visibly retards the electrophoretic mobility of recombinant Cse4 and Cse4-GFP . ( C and D ) A similar analysis of Cse4 internally tagged with GFP ( C ) or tdEos ( D ) . In these cases , purified proteins were introduced into WT lysate . ( E ) Plot of band intensity ( panel B ) for recombinant Cse4 ( black dots ) to estimate the endogenous Cse4 level ( red dot ) . ( F and G ) Estimates of endogenous levels of internally tagged Cse4 ( panels C and D ) . Insets show the intensity of the common background band used to compare Cse4-GFPinternal and Cse4-tdEosinternal lysates to WT lysate . The resulting normalization factor is given in red . After normalization ( protein fmoles/normalization factor ) , we obtained a 1:1 ratio for both Cse4:Cse4-GFPinternal and Cse4:Cse4-tdEosinternal . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 005 Accordingly , we investigated the behavior of Cse4 internally tagged with GFP ( Cormack et al . , 1997 ) or the photoconvertible fluorophore tdEos ( tandem dimer Eos; Nienhaus et al . , 2006 ) . For both constructs , we replaced the wild-type CSE4 gene in a haploid yeast strain , yielding tagged strains displaying normal bud morphology ( Figure 1C , D ) , and viability at normal and elevated growth temperatures indistinguishable from the wild-type strain ( Figure 1F ) . Live cell imaging of internally tagged Cse4-GFP reveals fluorescence exclusively in a single dot or a pair of dots ( Figure 1C ) , corresponding to the clusters of yeast centromeres ( Jin et al . , 1998; Meluh et al . , 1998; Chen et al . , 2000; Jin et al . , 2000 ) . Identical results are obtained for internal Cse4-tdEos fusion , despite the larger tag size ( Figure 1D ) . Additionally , the red emission of photoconverted tdEos avoids intracellular autofluorescence and improves contrast–nonetheless , no nuclear fluorescence is detectable outside of centromeric clusters . Curiously , S phase centromeres display weaker tdEos fluorescence ( Figure 1D ) , a phenomenon further explored below . Live cell imaging studies relying on a GFP fusion to the C-terminus of Cse4 reported unusual properties of Cse4 ( Coffman et al . , 2011; Lawrimore et al . , 2011; Shivaraju et al . , 2012 ) . Therefore , for comparison with our internal GFP fusions , we examined a representative C-terminally tagged Cse4-GFP strain ( MSY173 , obtained from Jennifer Gerton's laboratory; Shivaraju et al . , 2012 ) . For this strain , we confirm the presence of fluorescent centromeric clusters , though centromere intensity appears slightly elevated ( Figure 1—figure supplement 1 ) . However , in contrast to internally tagged Cse4 , we clearly detect extra-centromeric fluorescence throughout nuclei at every stage of the cell cycle ( Figure 1E , Figure 1—figure supplement 1 ) . This difference is confirmed by Western blot analysis of whole cell extracts , showing ∼twofold excess of C-terminal over internal Cse4-GFP fusion of comparable size and blotting efficiency ( Figure 1—figure supplement 2A ) , while internally tagged Cse4 is present at levels close to wild-type Cse4 ( Figure 1—figure supplement 2B–G ) . Most importantly , the strain carrying the C-terminal tag shows substantially reduced viability . The C-terminal Cse4-GFP strain exhibits slow growth in rich medium even at 24°C , and is not viable at 38°C , while none of the internal fusions have growth defects at either temperature ( Figure 1F ) . Taken together , our results demonstrate that fusion of a fluorescent protein tag to the C-terminus impairs Cse4 function . Accordingly , we only used internal tags to further explore the physiological dynamics of Cse4 . After synthesis and protein folding , tdEos fluorophores undergo relatively slow maturation to a green fluorescent state ( Nienhaus et al . , 2006 ) ( Figure 2A , Figure 6—figure supplement 1 ) . However , upon exposure to violet light , such mature fluorophores undergo almost instantaneous , irreversible photoconversion to a red-emitting state . To follow the fate of Cse4-tdEos in living cells by fluorescence pulse-chase analysis ( Figure 2B ) , we photoconverted Cse4-tdEos in asynchronously growing yeast to mark its initial distribution at different cell cycle stages ( Figure 2C ) . This reveals centromeric clusters in all cells , including the aforementioned weak signal in early S phase . Cells are then allowed to advance into the cell cycle , and re-imaged 40 min later . Figure 2D shows that centromeric clusters typically retain pre-existing Cse4 , with the striking exception of cells crossing the G1/S boundary ( magenta outlines ) , which lose centromeric fluorescence . This indicates that pre-existing Cse4 is not maintained or recycled in S phase . An additional round of photoconversion at the end of the experiment confirms loading of new Cse4 molecules at centromeric clusters ( Figure 2E ) , in accordance with previous studies showing Cse4-GFP deposition in S phase ( Pearson et al . , 2004; JW , personal communication ) . 10 . 7554/eLife . 02203 . 006Figure 2 . Pre-existing Cse4 is removed and exchanged for new Cse4 molecules at G1/S transition . ( A ) Relevant fluorescence states of a tdEos-tagged protein molecule are depicted schematically after its synthesis and folding , fluorophore maturation and irreversible photoconversion . Excitation and emission peak wavelengths are indicated ( see Figure 6—figure supplement 1 for additional details ) . ( B ) Pulse-chase experimental scheme . After initial photoconversion ( pulse at t0 ) , red-fluorescent Cse4 is followed into later stages of the cell cycle ( chase until tx ) . ( C–E ) Cells containing Cse4-tdEos were imaged immediately after pulse ( C ) and following 40 min chase ( D ) . At the end , additional photoconversion ( 2ndPC ) was used to confirm sufficient Z-stack range ( E ) . Three cells that crossed G1/S boundary are outlined in magenta while all other cells are outlined in grey , based on DIC images . ( F ) An example of a telophase cell followed until mother cell entered S phase , while the bud-derived daughter remained in G1 . ( G ) An example of S phase cell followed into telophase . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 006 Figure 2F shows a specific example of S phase replacement of Cse4 . A cell in telophase displays equivalent Cse4-tdEos fluorescence on both centromere clusters . Thereafter , the mother cell , which enters S phase sooner than the daughter , loses pre-existing centromeric signal , whereas centromeres of the daughter cell , still in G1 , are still occupied by pre-existing Cse4 . A second round of photoconversion confirms that mother cell centromeres contain newly deposited Cse4 . By contrast , photoconversion of a cell in S phase reveals weak fluorescence of the centromeric cluster ( Figure 2G ) . Upon advancement to telophase , the original cluster separates into two , each still showing weak fluorescence . A second round of photoconversion reveals a substantial increase of signal at these telophase clusters , a phenomenon attributable to the maturation of the tdEos fluorophore , as shown below . Taken together , our results document that early in S phase , Cse4 molecules are eliminated and replaced by newly synthesized molecules , thereafter remaining stably associated with centromeres through the rest of the cell cycle . The results also indicate the absence of a persistent pool of free nuclear or cytoplasmic Cse4 , as we fail to observe carry-through of pre-existing Cse4 into S phase . The fluorescence of Cse4 clusters increases gradually from S phase through mitosis for both tdEos and GFP insertions ( Figure 3A ) . Such a pattern could be caused by either a continuous deposition of newly synthesized fluorescent molecules ( precluded by results above ) , or an ongoing maturation of fluorophores already deposited in S phase . We tested the second scenario by measuring the fluorescence of newly formed centromere clusters after protein synthesis was blocked with cycloheximide ( Figure 3B ) . We find that their brightness increases with time until it reaches a plateau at ∼70 min , remaining stable for at least 120 min thereafter , notwithstanding the cycloheximide-induced block in cell cycle progression ( Figure 3C , D ) . This intensity profile indicates that the half-time of maturation at 25°C is approximately 40 min—similar to the time observed for half-maximal increase of Cse4-tdEos fluorescence in a population of asynchronously growing cells ( Figure 3A ) . Hence , fluorophore maturation is sufficient to explain the gradual rise of Cse4-tdEos fluorescence through the cell cycle after early S phase deposition . 10 . 7554/eLife . 02203 . 007Figure 3 . Cell cycle-dependent increase in centromere cluster intensity is a result of fluorophore maturation . ( A ) Relative intensity of centromeric clusters in asynchronously growing cells as the function of cell cycle stage and approximate time since entry into S phase . Values were corrected per 16 centromeres , to account for the presence of replicated ( 32 ) centromeres in a single ‘dot’ in S and G2 . ∼7500 and ∼11 , 000 photons were detected during 5 s exposure for G1 clusters containing Cse4 with internal GFP or tdEos , respectively—other results were normalized against those values . Standard deviation of each sample is indicated . ( B ) Schematic of the experiment to measure maturation rate of fluorophores present on Cse4-tdEos . Following α-factor synchronization , cycloheximide ( CHX , 0 . 2 mg/ml ) was added 10 min after entry into S phase . ( C ) Images of S phase cells at different sampling points ( tx ) after bud emergence . Cells which entered S phase prior to addition of cycloheximide are outlined . ( D ) Relative fluorescence of Cse4-tdEos centromere clusters in S phase in the absence of protein synthesis . Plateau value ( average of points representing tx >100 min , ∼1750 photons detected during 1 s exposure ) was used for normalization and standard deviations are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 007 To gain insight into the mechanism of Cse4 replacement , we investigated the role of DNA replication by analysis of synchronized cells in which replication is blocked with hydroxyurea ( Figure 4A ) . We photoconverted Cse4-tdEos in late G1 , ∼30 min after release from α-factor arrest ( corresponding to ∼15 min prior to bud emergence ) , and followed the fate of the pre-existing Cse4 thereafter . Figure 4B shows that , as expected , control untreated cells lose strong centromeric fluorescence upon entry into S phase . However , cells treated with hydroxyurea uniformly retain pre-existing Cse4 on centromere clusters , regardless of bud emergence . This suggests that the removal of old Cse4 from centromeres is associated with DNA replication . 10 . 7554/eLife . 02203 . 008Figure 4 . Removal of pre-existing Cse4 is associated with DNA replication . ( A ) Experimental scheme to assess role of DNA replication on the removal of pre-existing Cse4 . α-factor synchronized cells were released into control medium or one with 0 . 2 M hydroxyurea ( HU ) . Cse4-tdEos was photoconverted prior to bud emergence and then followed after sizable buds became evident . ( B ) Examples of cells released from α-factor block directly into control or hydroxyurea ( +HU ) containing medium . Time of photoconversion and observation after chase is indicated . Only cells on which buds appeared during the observation period are outlined . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 008 The gradual increase of Cse4 fluorescence through the cell cycle conflicts with the discrete twofold increase reported for C-terminally tagged Cse4-GFP at anaphase ( Shivaraju et al . , 2012 ) . To further examine this issue , we used a targeted FRAP procedure to detect any deposition of internally tagged Cse4-tdEos at anaphase ( Figure 5A ) . We photoconverted Cse4-tdEos in a metaphase cell to reveal the pair of centromere clusters ( Figure 5B1 ) . One of those clusters was then photobleached with a pulsed dye laser beam focused to a diffraction-limited spot , without affecting fluorescence of the other cluster ( Figure 5B2 ) . Upon progression through anaphase , we find that only one red-fluorescent cluster is visible at telophase as well ( Figure 5B3 ) . This indicates that additional Cse4 deposition did not occur on the bleached cluster , nor did Cse4 exchange between the two clusters . A second photoconversion conducted at the end of the experiment ( uncovering additional fluorophores that completed maturation in the meantime ) confirms that the targeted cluster remains functional and segregates to the opposite pole ( Figure 5B4 ) . Hence , our results indicate a compositional stasis for Cse4 after S phase deposition , and do not support a second wave of Cse4 deposition in anaphase . 10 . 7554/eLife . 02203 . 009Figure 5 . There is no additional Cse4 deposition in anaphase . ( A ) Scheme of the experimental test for Cse4 deposition in anaphase . All operations were carried on a selected metaphase cell in the specified order . 3D diffraction-limited spot , generated with a galvano-controlled MicroPoint 551 nm dye laser system , was used for targeted photobleaching . ( B ) An example of the metaphase cell subjected to a targeted photobleaching of photoconverted Cse4-tdEos centromeric cluster . Images were acquired at stages indicated in panel A . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 009 Recent advances in fluorescence microscopy enable localization of molecules in live and fixed cells with sub-diffraction accuracy ( Sengupta et al . , 2012 ) . The newly developed multifocal microscope ( MFM ) allows 3D imaging of the entire yeast cell volume in a single exposure ( Abrahamsson et al . , 2013 ) and , when combined with PALM ( Betzig et al . , 2006 ) , permits super-resolution localization of single fluorescent molecules with lateral accuracy of ∼20 nm and axial accuracy of ∼50 nm within a depth of ∼4 μm ( Hajj et al . , unpublished data ) . We applied this combined approach to analyze the volumetric distribution of Cse4-tdEos molecules within centromeric clusters in paraformaldehyde-fixed cells . As illustrated in Figure 6A , individual tdEos fluorophores are detectable simultaneously at different depths within a fixed cell , and a low photoconversion rate ensures observation of well-separated single-molecule fluorescence events ( Figure 6B ) . A resulting plot of the 3D distribution of all independent detections inside an anaphase cell ( assembled with ViSP software; El Beheiry and Dahan , 2013 ) reveals both centromere clusters as compact groups of 20 and 22 tdEos fluorophores ( Figure 6C; Video 1 ) . These should not be construed to reflect the total number of Cse4 molecules present at the clusters , because incomplete maturation and the initial photobleaching prior to PALM ( necessitated by paraformaldehyde-induced conversion—JW , personal communication ) leave only a fraction of total fluorophores detectable as single-molecule events . Moreover , the existence of the reversible dark state of red tdEos may cause multiple detections of some fluorophores ( Annibale et al . , 2010; Lee et al . , 2012; Figure 6—figure supplement 1 ) . Despite this , MFM-PALM localization of individual fluorophores allows estimation of the overall dimensions of centromere clusters . We find that Cse4 clusters in G1 are typically ∼450 nm across ( Figure 6D , F; Video 2 ) , clearly indicating that their wide-field image ( the sum of all individual Airy disks ) would significantly exceed the diffraction limit ( in this case an Airy disk with FWHM ∼225 nm ) . Strikingly , anaphase clusters are more compact and asymmetric , on average approximating an ellipsoid of 350 nm × 200 nm ( still above the diffraction limit—Figure 6E , G; Video 3 ) . This change corresponds to ∼threefold reduction in the volume of the cluster and thus higher spatial density of centromeres . Frequently , the shortened polar axis coincides with the direction of the mitotic spindle extending between anaphase clusters ( Video 1 ) . Such substantial dimensions of Cse4 clusters and their compaction in anaphase have important implications for photometric measurements of fluorescence intensity ( see below ) . 10 . 7554/eLife . 02203 . 010Figure 6 . Centromeric clusters become more compact during anaphase . ( A ) An example of MFM-PALM image with nine simultaneously acquired Z-planes . Gold Nanorods are indicated ( blue circles ) and distance above the glass surface is listed for other tiles . Anaphase cell outline ( red ) is based on a separate bright-field MFM image . Two single-molecule events of Cse4-tdEos are visible inside the cell . ( B ) A time-trace representation of the number of tdEos fluorophore detections per frame during single MFM-PALM acquisition series . Events lasting >1 frame were considered to represent the same fluorophore . ( C ) A projection of all 73 independent single-fluorophore detections in the above image series . Image volume is indicated and Z position of individual localizations is color-coded . Each event is depicted as a dot 50 nm across ( increased from average 20 nm lateral precision to facilitate visualization at this image scale ) . Number of events concentrated at each of the two centromeric clusters is listed . See Video 1 . ViSP software ( El Beheiry and Dahan , 2013 ) was used for projections in C–E . ( D ) A representative example of a 3D distribution of Cse4-tdEos molecules on the G1 centromere cluster . Each event ( total of 21 independent detections ) is depicted as a 20 nm dot , corresponding to the average lateral localization precision . Total volume of 1 μm3 is shown , with color-coded Z distance . See Video 2 . ( E ) A representative late anaphase centromere cluster depicted as above . Total of 24 independent detections are plotted . See Video 3 . ( F ) Compilation of detections from 10 G1 clusters , center-aligned and projected onto XY plane . Grey circle depicts cross-section of a sphere ( ∼450 nm across ) sufficient to contain majority of detected Cse4-tdEos molecules . ( G ) Compilation of total detections from 10 late anaphase and telophase clusters projected onto XY plane ( center-aligned , long axis rotated horizontally ) . Grey ellipse depicts cross-section of an ellipsoid ( ∼350 nm equatorial diameter and ∼200 nm polar distance ) sufficient to contain majority of detected Cse4-tdEos molecules . In both cases , distribution in Z is comparable to that along X-axis ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 01010 . 7554/eLife . 02203 . 011Figure 6—figure supplement 1 . tdEos fluorophore undergoes transitions between multiple fluorescent and dark states . Different states of tdEos fluorophore ( and mEos variants , including mIRIS [Wiedenmann et al . , 2011] ) are shown schematically . Newly synthesized non-fluorescent tdEos undergoes fast folding followed by slow maturation into green emitting state ( λexc/λem = 506 nm/516 nm , respectively ) ( Nienhaus et al . , 2006 ) . Green tdEos can undergo irreversible photobleaching as well as conversion to a long-term reversible dark state , from which it is ‘rescued’ by 405 nm illumination ( JW personal communication ) . 405 nm light also induces irreversible photoconversion from green- to red-emitting form ( Nienhaus et al . , 2006 ) . Interestingly , PFA fixation also results in the conversion of some fluorophores into the red-emitting state ( JW , personal communication ) . Red tdEos may be bleached irreversibly or enter a reversible dark state as well—again subject to rescue by 405 nm illumination ( Annibale et al . , 2010 ) . In addition , both fluorescent forms undergo short-term ‘blinking’ ( not depicted in the figure ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 01110 . 7554/eLife . 02203 . 012Video 1 . 3D representation of tdEos fluorophore distribution in anaphase cell from Figure 6C . Each event is depicted as a dot 50 nm across instead of the actual average localization precision ( 20 nm lateral/50 nm axial ) . The original color coding of axial distance from Figure 6C is maintained . The video was assembled in ViSP software ( El Beheiry and Dahan , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 01210 . 7554/eLife . 02203 . 013Video 2 . 3D representation of Cse4-tdEos distribution in G1 centromere cluster from Figure 6D . Each event is depicted as a dot 20 nm across instead of the actual average localization precision ( 20 nm lateral/50 nm axial ) . The box encloses 1 μm3 volume and two artificial sizing marks ( red and blue ) are present at the corners . The original color coding of axial distance ( Figure 6D ) is maintained . The video was assembled in ViSP software ( El Beheiry and Dahan , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 01310 . 7554/eLife . 02203 . 014Video 3 . 3D representation of Cse4-tdEos distribution in late anaphase centromere cluster from Figure 6E . Each event is depicted as a dot 20 nm across instead of the actual average localization precision ( 20 nm lateral/50 nm axial ) . The box encloses 1 μm3 volume and two artificial sizing marks ( red and blue ) are present at the corners . The original color coding of axial distance ( Figure 6E ) is maintained . The movie was assembled in ViSP software ( El Beheiry and Dahan , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 014 In addition to centromere clusters , we also observe individual fluorescent events scattered throughout the cytoplasm . Due to the absence of a persistent free Cse4 pool ( as demonstrated by pulse-chase experiments in Figure 2 ) , those are unlikely to represent free Cse4 molecules . Because GFP is known to be resistant to proteolytic degradation ( Chiang et al . , 2001 ) , we speculate that these cytoplasmic events correspond to fluorophore moieties persisting after proteolytic degradation of unincorporated Cse4 ( Collins et al . , 2004 ) . Such residual fluorophores would not be distinctly detectable in live cells due to their mobility and dispersal in the cytoplasmic volume ( ∼40-fold larger than the nucleus ) . To estimate the number of Cse4-GFP molecules present at a centromere cluster , we compared its fluorescence intensity to that of TetR-GFP bound to a defined number of tet operator sites ( tetO ) ( Michaelis et al . , 1997 ) . To minimize the background caused by free TetR-GFP molecules , we expressed TetR-GFP from a weakened , non-induced URA3 promoter ( Roy et al . , 1990 ) . Figure 7A shows that a fluorescent dot is detectable against a diffuse nuclear background even in cells containing 7x tetO and becomes clearly apparent in the case of 14x tetO . When compared at an identical brightness scale , it is evident that the intensity of Cse4-GFP cluster lies between that of 28 and 42 GFPs , the maximum number that can be present on 14x and 21x tetO , respectively , as tetracycline repressor is a homodimer . Furthermore , we performed photometric measurements of tetO arrays and centromeric clusters after precise background subtraction . We utilized wavelet filtering ( Berry and Burnell , 2011 ) to separate small scale features ( e . g . , clusters ) from larger patterns ( e . g . , nuclear and cytoplasmic fluorescence ) ( Figure 7—figure supplement 1 ) . Figure 7B shows that median intensity of wavelet filtered Cse4-GFP clusters corresponds to ∼36 GFP molecules . Given the scatter of measured values , this is consistent with two Cse4 molecules for each of the 16 centromeres clustered together in telophase . 10 . 7554/eLife . 02203 . 015Figure 7 . Two Cse4 molecules are present on each centromere . ( A ) Comparison of Cse4-GFP centromere clusters with TetR-GFP bound to arrays of 7 , 14 or 21 tetO , displayed within the same brightness range . Representative telophase cells are outlined . Clusters in surrounding cells may be out of focus . ( B ) Fluorescence intensity of tetO arrays and centromeric clusters was measured in telophase cells ( 2 s exposure ) . Minimum/1st quartile/median/3rd quartile and maximum values are displayed for each group of 50 measurements . Prior to measurement , clusters were separated from lower frequency components of the image ( diffuse fluorescence in nuclei and intracellular autofluorescence ) by processing the image with wavelet function and adding together scales 1 , 2 and 3 ( 1 , 2 and 4 pixels FWHM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 01510 . 7554/eLife . 02203 . 016Figure 7—figure supplement 1 . Wavelet filtering allows precise separation of cluster signal from nuclear and cellular background . ( A ) Images of the representative telophase cells ( outlined ) are shown for Cse4-GFPinternal strain or strains containing integrated tetO arrays ( 7 , 14 , 21 or 112 repeats ) and TetR-GFP under constitutive low level expression from mutated URA3 promoter . Image containing extracted small-scale structures ( sum of wavelet scales 1 , 2 , and 3 ) is presented together with the larger structures , representing background ( including cells and nuclei; sum of remaining wavelet scales and residual ) . Simple arithmetic addition of all scales ( 1 through 8 ) and residual returns the image indistinguishable from the original ( within the precision of the 32-bit floating-point format ) . ( B ) An example of intensity profile ( in ADUs ) across cell , nucleus and cluster present in a telophase cell from the strain containing 21x tetO and TetR-GFP . Scales 1+2+3 accurately separate cluster signal from background ( represented by higher scales ) . ( C ) Similar example of intensity profile and effects of wavelet filtering for a telophase cell containing Cse4-GFPinternal . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 016 The high stability of centromeric Cse4 nucleosomes after S phase suggests a need for special maintenance mechanism ( s ) . Scm3 is the Cse4-specific chaperone required for Cse4 deposition and maintenance at centromeres ( Camahort et al . , 2007; Mizuguchi et al . , 2007; Stoler et al . , 2007 ) . Scm3 is recruited to centromeres by sequence-specific factor CBF3 and itself possesses AT-rich DNA binding activity ( Xiao et al . , 2011; Cho and Harrison , 2011a ) . We demonstrated previously that Scm3-GFP localizes to centromeres at every stage of the cell cycle , including anaphase of mitosis , and is also distributed diffusely throughout the nucleus ( Xiao et al . , 2011 ) ; Scm3-tdEos has an identical distribution ( Figure 8—figure supplement 1A ) . Comparison of total nuclear and centromeric fluorescence reveals ∼fourfold excess of free Scm3-tdEos throughout the nucleus compared to the centromeric-bound protein ( Figure 8—figure supplement 1B ) . To assess the stability of centromeric Scm3 , we applied targeted laser photobleaching of Scm3-tdEos at one of two centromere clusters ( Figure 8A ) . In contrast to the stability of Cse4-tdEos , Scm3 fluorescence reappears on the cluster within several minutes ( between 4 and 11 min in this example ) , demonstrating that , unlike Cse4 , Scm3 undergoes exchange between centromeres and the free nuclear pool . When measured throughout the cell cycle , the average recovery time ( at which fluorescence is detected again ) is ∼5 min ( Table 1 ) . Such dynamic exchange of Scm3 with persistent , steady-state occupancy may ensure continuing integrity of the singular centromeric nucleosome after Cse4 deposition in S phase . 10 . 7554/eLife . 02203 . 017Figure 8 . Scm3 dynamically interacts with centromeres at levels equivalent to Cse4 . ( A ) Scm3-tdEos fluorescence recovery after targeted photobleaching . The experiment was performed essentially as shown in Figure 5A , except that recovery was monitored by repetitive imaging , without additional photoconversion . In this example , images were acquired only when indicated . Arrowhead indicates targeted centromere cluster . ( B ) Pulse-chase demonstrates overall stability and cell-cycle persistence of Scm3-tdEos . Photoconverted Scm3 molecules were followed after approximately one and two cell cycles ( 2 . 5 and 4 hr , respectively ) . Fluorescent images are displayed with the same intensity range . ( C ) Fluorescence intensity of centromeric clusters containing Scm3-tdEos ( black circles ) or Cse4-tdEosinternal ( open circles ) in late anaphase/telophase . Average and standard deviation are shown as a function of excitation time to illustrate photostability of photoconverted tdEos . ∼10 , 000 photons were initially detected in 5 s exposure in both cases . Representative images of individual cells containing Cse4-tdEosinternal or Scm3-tdEos are shown at the same brightness scale . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 01710 . 7554/eLife . 02203 . 018Figure 8—figure supplement 1 . Scm3-tdEos is present on centromeres and in the nucleus at every stage of the cell cycle . ( A ) Distribution of Scm3-tdEos in live cells . Cell cycle stages are indicated in DIC panels and red fluorescence , after photoconversion , is shown as in Figure 1 . ( B ) Majority of Scm3 is dispersed throughout nucleus . Average ratio of fluorescence present at centromeres ( ±standard deviation ) through the cell cycle is indicated . ( C ) Scm3-tdEos is not displaced from centromeres in S phase but its signal diminishes , probably due to exchange with new , less mature molecules . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 01810 . 7554/eLife . 02203 . 019Figure 8—figure supplement 2 . Similar amounts of Cse4 and Scm3 molecules reside at centromeres . ( A ) Average intensity ( photons/s , ±standard deviation ) of centromeric clusters containing Cse4-GFPinternal or Scm3-GFP ( n = 8 ) , measured during continuous excitation . ( B ) Sample images of cells containing Cse4-GFP or Scm3-GFP are shown at the same brightness scale . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 01910 . 7554/eLife . 02203 . 020Table 1 . Recovery time of centromeric Scm3-tdEos after targeted photobleachingDOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 020Cell cycle stageMean recovery time ( min ) Standard deviationSample sizeG15 . 22 . 59S4 . 51 . 64metaphase4 . 92 . 37anaphase4 . 72 . 013telophase5 . 12 . 79Note: After targeted photobleaching of photoconverted Scm3-tdEos centromere clusters with 551 nm dye-laser , cells were imaged with stepwise focus changes within -1 μm to +1 μm Z range ( 7 steps , 333nm apart , 5 sec . exposure per step ) . G2 clusters were excluded due to their extended size . A pulse-chase experiment shows that total pre-existing Scm3 persists through multiple cell cycles , with gradual dilution during consecutive cell divisions , indicating its low rate of turn-over ( Figure 8B ) . Moreover , steady-state fluorescence of centromeric Scm3 shows a mild decrease in S phase , suggesting synthesis of new Scm3 molecules with immature fluorophores during that stage ( Figure 8—figure supplement 1C ) . Comparison of centromeric Scm3 and Cse4 fluorescence in telophase , when the majority of tdEos fluorophores are mature , shows that their intensities closely overlap and follow similar photobleaching curves ( Figure 8C ) . An identical result is also obtained with the GFP tag ( Figure 8—figure supplement 2 ) . Taking into account that Cse4 remains stable after deposition and Scm3 interacts dynamically and persistently with centromeres , this indicates that near equimolar levels of both proteins coexist on centromere clusters throughout the cell cycle . In a side-by-side comparison , we document that a C-terminal GFP tag impairs Cse4 functionality , causing severe growth defects and substantial extra-centromeric accumulation . We observed similar growth defects with a FLAG epitope tag as well ( GM , unpublished data ) . The extreme C-terminal residues of Cse4 specify recognition by Mif2 , the yeast CENP-C inner kinetochore protein ( Kato et al . , 2013 ) . Accordingly , a C-terminal fusion is likely to affect such interaction , perturbing kinetochore functionality . Thus , the molecular phenotypes of C-terminally tagged Cse4-GFP reflect properties of functionally impaired Cse4 , rather than the native protein . Similarly , partial loss of function was also observed for C-terminally tagged CENP-A/CenH3 in mouse ( Kalitsis et al . , 2003 ) and Drosophila ( Schuh et al . , 2007 ) . Recent claims of altered cell cycle dynamics and/or substantially increased centromere localization were based on such compromised Cse4 fusions , despite their temperature-sensitive phenotype and evident extra-centromeric distribution ( Coffman et al . , 2011; Lawrimore et al . , 2011; Shivaraju et al . , 2012 ) . On the other hand , the very first epitope tag in Cse4 consisted of an insertion within the N-terminal tail , at codon 81 ( Stoler et al . , 1995 ) . At this location , a GFP tag does not affect Cse4 functionality and cell growth ( Chen et al . , 2000 ) . We find that even the insertion of tdEos tag ( twice the size of GFP ) at the same location is well tolerated , causing no detectable growth phenotypes , and similar findings were obtained for up to four GFP copies ( R Baker , personal communication ) . Thus , internally tagged Cse4 fusions should be used in imaging studies as a preferred reporter of the composition and dynamics of centromeric nucleosomes . The internal photoconvertible tdEos tag allows a direct analysis of Cse4 dynamics in live cells , minimizes autofluorescence , improves signal to noise , and enables excitation at low energies to limit phototoxicity and cell cycle perturbation . This reveals replacement of Cse4 exclusively in early S phase , linked to DNA synthesis—consistent with the timing of centromere replication ( McCarroll and Fangman , 1988; Pohl et al . , 2012 ) . Our data elaborate on the S phase deposition of Cse4 reported by Pearson et al . ( 2004 ) , by showing this process as a removal of pre-existing Cse4 followed by the deposition of newly synthesized molecules , without recycling of old Cse4 . Subsequently , Cse4 remains stably bound to centromeres for the remainder of the cell cycle until the next S phase . Furthermore , targeted photobleaching experiments show no second wave of Cse4 deposition in anaphase . Taken together , our findings provide compelling evidence that Cse4 is replaced in S phase and remains static on centromeres for the rest of the cell cycle . In this context , budding yeast Cse4 has no epigenetic role in kinetochore inheritance , in contrast to the inheritance of CENP-A on regional centromeres of other organisms ( De Rop et al . , 2012 ) . The gradual increase in fluorescence intensity observed for Cse4-GFP and Cse4-tdEos after S phase deposition is a manifestation of fluorophore maturation . Accordingly , interpretation of fluorescence intensities for proteins undergoing synthesis and exchange at a highly specific moment of the cell cycle requires caution . Furthermore , in conventional microscopy , centromere clusters frequently appear more point-like in anaphase and telophase than in G1 , which may give the impression of a rise in fluorescence when viewed against the increased nuclear background caused by C-terminal Cse4-GFP fusions ( Joglekar et al . , 2006; Aravamudhan et al . , 2013 ) . A new super-resolution 3D-PALM approach allowed mapping of the actual spatial distribution of individual Cse4 molecules in the centromere cluster , indicating that it should not be treated as a point source for photometric analysis , and providing resolution superior to previous results based on bulk analysis ( Haase et al . , 2013 ) . Moreover , 3D-PALM directly reveals that centromere clusters contract in anaphase . This may be a consequence of the hydrodynamic drag of segregating chromosomes , and is consistent with EM tomography showing congregation of the plus ends of spindle microtubules during anaphase ( O'Toole et al . , 1999 ) . Such compaction of centromere clusters leads to ∼threefold higher spatial density of centromeres , increasing the likelihood that individual Cse4 molecules on separate centromeres come into proximity sufficient for FRET . This may explain the higher FRET efficiency reported in anaphase ( Shivaraju et al . , 2012 ) as interactions between centromeres , without the need to invoke structural oscillation of the centromeric nucleosome between hemisome and octasome . Previous biochemical and molecular genetic evidence led to a model for a single centromeric nucleosome per yeast chromosome , each containing two Cse4 molecules , located at the ∼125 bp CEN sequence common to all 16 yeast chromosomes ( Chen et al . , 2000; Smith , 2002; Furuyama and Biggins , 2007 ) . In contrast , the use of C-terminally tagged Cse4 yielded estimates ranging from 1 to 8 Cse4 molecules per centromere ( Joglekar et al . , 2006; Coffman et al . , 2011; Lawrimore et al . , 2011; Shivaraju et al . , 2012; Aravamudhan et al . , 2013 ) . These discrepant results can be attributed to inaccuracies in estimating spot intensity in the presence of substantial nuclear background , failure to account for the full extent of the centromere cluster ( which clearly exceeds the diffraction disk , especially in interphase ) in the measurement aperture , or treatment of the cluster as a point source with Gaussian intensity distribution . Interestingly , bimolecular fluorescence complementation ( BiFC ) experiments demonstrated that the C-terminal Cse4-GFP fusion is deposited on centromeres as a pair during S phase , and the fluorescence intensity of a ‘lagging’ centromere in a dicentric chromosome at anaphase is consistent with the presence of two Cse4-GFP molecules ( Aravamudhan et al . , 2013 ) . Our photometry measurements of internally tagged Cse4-GFP—taking into account the dimensions of centromere clusters—also support the presence of two molecules of Cse4 in the singular centromeric nucleosome . The Scm3 chaperone persists at centromeres in every stage of the cell cycle ( Xiao et al . , 2011 ) . This steady-state centromeric occupancy is the result of continuous dynamic exchange , on a timescale of several minutes , with a large nuclear pool of free Scm3 molecules . Such exchange was also observed by Luconi et al . ( 2011 ) in anaphase , although authors did not reliably observe Scm3 in other stages of the cell cycle . Scm3 may dissociate stochastically , and re-associate onto centromeres through interactions with Ndc10 and AT-rich CEN DNA ( Xiao et al . , 2011; Cho and Harrison , 2011a ) . This dynamic property explains the lack of Scm3 in biochemical purifications of kinetochores ( Westermann et al . , 2003; Akiyoshi et al . , 2009 ) , its absence as a stable component of reconstituted Cse4 octasome ( Dechassa et al . , 2011 ) and fluctuations in measurements of Scm3 occupancy by ChIP ( Luconi et al . , 2011; Mishra et al . , 2011; Shivaraju et al . , 2011; Xiao et al . , 2011 ) . As a Cse4-specific histone chaperone , Scm3 needs not , in principle , be retained at centromeres once assembly of the centromeric nucleosome has been accomplished in S phase . Indeed , biochemical experiments document classic chaperone properties for the conserved Cse4-binding domain of Scm3 ( Dechassa et al . , 2011; Shivaraju et al . , 2011; Xiao et al . , 2011 ) , and NMR and crystal structures of this domain show that DNA binding by Cse4-H4 in the nucleosome is physically incompatible with continued Scm3 interaction ( Zhou et al . , 2011; Cho and Harrison , 2011b ) . However , full-length Scm3 ( containing the Ndc10 and DNA binding domains ) is enriched at centromeres through all of the cell cycle stages ( Xiao et al . , 2011 ) , consistent with live cell imaging , and genetic studies suggesting the importance of Scm3 after Cse4 deposition ( Camahort et al . , 2007 ) . Steady-state occupancy with dynamic exchange has been described for other chromatin proteins , notably the heterochromatin protein HP1 , which functions as a platform for assembling gene silencing complexes ( Cheutin et al . , 2003 ) . Thus , it is highly likely that Scm3 remains after deposition of Cse4 to safeguard the integrity of the singular centromeric nucleosome on each budding yeast chromosome . A model showing the overall fate of Cse4 in the cell cycle is depicted in Figure 9 . In G1 , a stable Cse4 nucleosome is maintained by steady-state occupancy of Scm3 , which would capture and redeposit Cse4-H4 if any stochastic dissociation occurs . Early in S phase , centromeric nucleosomes are disrupted , leading to removal and degradation of old Cse4 , with kinetochore detachment ( Kitamura et al . , 2007 ) . The centromeric nucleosome is then re-established in a step-wise process , most likely starting with binding of CBF1 to CDEI , and CBF3 to CDEIII with assistance of Scm3 ( Camahort et al . , 2007; Mizuguchi et al . , 2007 ) . Subsequently , two Scm3-Cse4-H4 heterotrimers are recruited by Ndc10 , the dimeric component of CBF3 ( Cho and Harrison , 2011a ) . Scm3 then deposits each Cse4-H4 on CEN DNA through a dimer intermediate to form a ( Cse4-H4 ) 2 tetrasome ( Dechassa et al . , 2014 ) . During this step , CDEII DNA out-competes Cse4-H4 contacts with Scm3 , which nonetheless remains in close proximity through interactions with Ndc10 and AT-rich CEN DNA . Assembly of two H2A-H2B dimers is likely to follow , although their topography may be altered , as indicated by lack of formaldehyde cross-linking ( Mizuguchi et al . , 2007; Xiao et al . , 2011; Krassovsky et al . , 2012 ) . Thus , the stable state of Cse4-nucleosomes is octameric , although transient , sub-octameric intermediates may occur during assembly or disassembly . By remaining in close proximity , Scm3 serves not as a structural replacement for H2A-H2B ( contrary to our initial model in Mizuguchi et al . ( 2007 ) ) , but rather as a persistent chaperone-in-residence to insure against catastrophic loss of the singular Cse4 nucleosome . Given that fungal Scm3 orthologs possess a diversity of DNA binding motifs ( Aravind et al . , 2007 ) , the centromeric persistence of this chaperone through the majority of the Schizosaccharomyces pombe cell cycle ( Pidoux et . al , 2009; Williams et al . , 2009 ) or the entirety of the Saccharomyces cerevisiae cell cycle ( this study ) may be a common theme of CENP-A/CenH3 chaperone function . We hope that our findings and clarification of the fates of Cse4 and Scm3 will enable constructive dissection of the mechanisms underlying kinetochore establishment and maintenance to ensure accurate segregation of daughter chromosomes . 10 . 7554/eLife . 02203 . 021Figure 9 . Model of Cse4 replacement and re-establishment of point centromere during cell cycle ( see text for details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 021 All strains were derived from Saccharomyces cerevisiae W1588-4C ( Table 2 ) except strain MS173 containing C-terminal Cse4-GFP ( MATa his3-1 leu2-0 ura3-0 Cse4-GFP::SpHIS5 ) , which was obtained from Jennifer Gerton , Stowers Institute . Constitutive activity of a mutant URA3 promoter ( −80 to −109 deletion; Roy et al . , 1990 ) was used for low level expression of TetR-GFP . 10 . 7554/eLife . 02203 . 022Table 2 . Saccharomyces cerevisiae strainsDOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 022StrainGenotypeMBY507MATa ade2 CSE4-GFP-CSE4 can1-100 his3-11 , 15 leu2-3 , 112 trp1-1 ura3-1 RAD5JBY119†MATa ADE2 dynLC::hphMX4 cse4::natMX4 can1-100 his3-11 , 15 leu2-3 , 112::LEU2-CSE4-tdEOS-CSE4 trp1-1 ura3-1 RAD5JBY111‡MATa ADE2 dynLC::hphMX4 SCM3-tdEOS-kanMX4 can1-100 his3-11 , 15 leu2-3 , 112 trp1-1 ura3-1 RAD5JBY251§MATa ADE2 can1-100 his3-11 , 15 leu2-3 , 112::LEU2-Δ80ura3p-TetR-GFP-TAP-ADHt trp1-1 ura3-1::pRS406-7xtetO RAD5JBY252§MATa ADE2 can1-100 his3-11 , 15 leu2-3 , 112::LEU2-Δ80ura3p-TetR-GFP-TAP-ADHt trp1-1 ura3-1::pRS406-14xtetO RAD5JBY253§MATa ADE2 can1-100 his3-11 , 15 leu2-3 , 112::LEU2-Δ80ura3p-TetR-GFP-TAP-ADHt trp1-1 ura3-1::pRS406-21xtetO RAD5JBY254§MATa ADE2 can1-100 his3-11 , 15 leu2-3 , 112::LEU2-Δ80ura3p-TetR-GFP-TAP-ADHt trp1-1 ura3-1::pRS306-112xtetO RAD5MSY173#MATa his3-1 leu2-0 ura3-0 Cse4-GFP::SpHIS5Notes: Cse4-GFPinternal , Xiao et al . , 2011 . †Cse4-tdEosinternal , this paper . ‡Scm3-tdEosC-terminal , this paper . §7 , 14 , 21 or 112 tetO repeats , respectively , and TetR-GFP , this paper . #Cse4-GFPC-terminal , Shivaraju et al . , 2012 . Histidine-tagged versions of Cse4 , Cse4-GFPinternal , and Cse4-tdEosinternal were expressed in Rosetta ( DE3 ) Escherichia coli strain ( Novagen , San Diego , CA ) under control of T7 promoter ( pET15b vector ) . Bacterial cells were lysed in 6 M guanidine-hydrochloride , 50 mM Tris–HCl pH8 , 0 . 5 mM DTT , 20 mM imidazole buffer , and sonicated . Lysates were absorbed with HisTrap resin ( GE Healthcare , Uppsala , Sweden ) , washed with 8 M urea , 50 mM Tris–HCl pH8 , 0 . 5 mM DTT , 20 mM imidazole buffer and bound protein eluted with the same buffer containing 500 mM imidazole . Concentration of full-length recombinant proteins was assayed by densitometric analysis of Coomassie-stained ( Simply Blue Safe Stain , Invitrogen , Carlsbad , CA ) SDS-PAGE gel containing known amounts of BSA ( fraction V , Sigma-Aldrich , St . Louis , MO ) . Total cellular extracts were prepared by boiling pelleted yeast samples in SDS loading buffer . After SDS-PAGE , Western blots were probed with affinity-purified rabbit anti-Cse4 ( Mizuguchi et al . , 2007 ) or anti-H4 antibodies ( Upstate , Lake Placid , NY ) , followed by anti-rabbit IgG-HRP ( Life Technologies , Grand Island , NY ) . Chemiluminescence was detected with ImageQuant LAS3000 ( FujiFilm , Tokyo , Japan ) . Serial dilutions of recombinant proteins ( added to indicated lysates; see Figure 1—figure supplement 2 ) were used to estimate the amounts of endogenous Cse4 , Cse4-GFP , and Cse4-tdEos present in yeast lysates after Western blotting with anti-Cse4 antibody . The bud size and the number/position of centromere clusters were used to assign stages of the cell cycle . For synchronization , low density cultures ( OD600 <0 . 3 ) in CSM medium ( MP Biomedicals , Santa Ana , CA ) supplemented with 400 μg/ml adenine ( Sigma-Aldrich , St . Louis , MO ) , were exposed to 5 μg/ml of α-factor ( Sigma-Aldrich ) for 90 min , collected by filtration , washed with sterile water and released into CSM+adenine medium . Entry into S phase ( bud emergence , ∼45 min after release ) was monitored by DIC . Release medium containing 0 . 2 M hydroxyurea ( Sigma-Aldrich ) was used to inhibit DNA replication . To block protein synthesis , release medium was supplemented with 200 μg/ml of cycloheximide ( Sigma-Aldrich ) 10 min after first detection of cells entering S phase ( ∼55 min after release ) . Hamamatsu C9100-13 camera ( −94°C , 0 . 63 MHz , 16-bit ADC; Hamamatsu , Bridgewater , NJ ) was used typically with EM gain of 50 ( conversion factor 0 . 044 e−/ADU , readout noise 0 . 470 e−RMS , thermal current 0 . 014 e−/s , established experimentally—see Berry and Burnell , 2011 ) . IR was blocked with FF01-750/SP filter ( Semrock , Rochester , NY ) . Zeiss AxioObserver Z1 microscope ( Carl Zeiss Microscopy , Thornwood , NY ) was equipped with Zeiss Plan-Apochromat 150x NA1 . 35 glycerine-immersion objective , P-737 piezoelectric stage ( Physik Instrumente , Auburn , MA ) , Zeiss Colibri and Lumencor Spectra-6 ( Lumencor , Beaverton , OR ) illuminators , and custom fluorescence cubes ( Table 3 ) . Yeast were grown in complete darkness in the CSM+adenine medium ( at 25°C , 250 rpm , final OD600 ≤0 . 3 ) , manipulated only under dim red light ( 660 nm ) and imaged in CellAsic Y04C microfluidic chambers ( CellASIC , Hayward , CA ) . 671 nm narrowband illumination ( #65-233; Edmund Optics , Barrington , NJ ) was used for DIC . To minimize phototoxicity , low level excitation ( ∼7 W/cm2 , 1–5 s exposure ) was used for fluorescence imaging and 405 nm light ( ∼0 . 7 W/cm2 , 7–10 s ) for tdEos photoconversion . Typically , Z-stacks consisted of 13 steps , 333 nm apart . 10 . 7554/eLife . 02203 . 023Table 3 . Light sources and filters used for wide field fluorescence imagingDOI: http://dx . doi . org/10 . 7554/eLife . 02203 . 023FluorophoreLight source & channelFilter cubeExcitationBeamsplitterEmissionGFPColibri/LED470* Spectra-6/C§FF01-475/28† LL01-488/1†T495LP‡FF01-525/50†tdEos ( red emission ) Colibri/LED555* Spectra-6/GY§BP550/25* FF01-543/3†FT570* FF568-Di01†BP605/70* FF01-593/46†tdEos ( photoconversion ) Colibri/LED405* Spectra-6/V§FF01-405/10†59004BS‡59004M‡Source: *Zeiss . †Semrock . ‡Chroma . §Lumencor . Raw 16 bit images were converted into FITS format ( Supplementary file 1 contains batch FITS converter macro for ImageJ ) and calibrated in 32-bit floating-point space using bias , thermal and flat-field frames ( AIP4WIN , Berry and Burnell , 2011 ) . Z-stacks were reduced to the composite image only for the presentation purposes , by projecting individual layers , with centromeres in focus , onto a common plane and the identical brightness range was kept for all comparable panels of any given Figure . All intensity measurements were carried on calibrated , unreduced Z-stacks with aperture photometry in AIP4WIN software , using typical FWHM of centromere cluster ( 4 pixels = 428 nm ) as a radius of measurement aperture and an outer background annulus ( 5 pixels = 535 nm wide , area 4 times larger than measurement aperture – see Berry and Burnell , 2011 for discussion of photometry techniques ) . Background-corrected signal was converted into photoelectrons ( equivalent of detected photons ) using experimentally established camera parameters ( see Berry and Burnell , 2011 for details ) . For strains with considerable nuclear background ( tetO/TetR-GFP and C-terminal Cse4-GFP strains ) , the signal corresponding to tetO arrays or centromeric clusters was separated from the diffuse nuclear and cellular background by à trous wavelet transform of 32-bit floating point images ( see above ) using AIP4WIN software ( Berry and Burnell , 2011 ) . Wavelet scales 1 , 2 and 3 were added together to include all objects up to 4 pixels FWHM across and the intensity of spots was measured by aperture photometry as above . Galvano-controlled MicroPoint system ( Photonic Instruments , Saint Charles , IL ) was used for targeted photobleaching . 551 nm pulsed dye laser was focused to a diffraction-limited spot ( FWHM ∼210 nm ) and centromeres were targeted in real-time during initial Z-stack acquisition , after photoconversion . Following recovery , additional Z-stacks were acquired from time to time . Multifocus microscope ( Abrahamsson et al . , 2013 ) was used for 3D-PALM . The system contained MFM grating ( designed to yield 380 nm spacing between consecutive planes in the multifocal image ) , matching corrective grating/prism , Nikon 100x NA1 . 4 oil-immersion objective and Andor DU897+ camera ( −70°C , EM gain = 250 , 70 ms exposure; Andor Technology USA , South Windsor , CT ) , yielding final image voxel of 120 × 120 × 380 nm . Paraformaldehyde-fixed yeast cells were attached to concanavalin A-coated cover slips containing immobilized 550 nm bare Gold Nanorods ( 25 nm diameter , 550 nm emission; NanoPartz , Loveland , CO; Shtengel et al . , 2009 ) . For limited photoconversion , a 405 nm laser ( Coherent , Santa Clara , CA ) was used at 0 . 2W/cm2 , and red tdEos fluorophores were detected under 561 nm laser illumination ( 2 kW/cm2; Cobolt , San Jose , CA ) . Images were corrected for distortion and transmission , converted into 3D stacks , then individual events were identified and their 3D coordinates determined with FISHQuant ( Mueller et al . , 2013 ) . Finally , residual 3D drift of the sample was corrected in MatLab ( MathWorks , Natick , MA ) based on Gold Nanorod fiducials . ViSP software ( El Beheiry and Dahan , 2013 ) was used for visualization and presentation of results . Full details for MFM-PALM are available upon request ( Hajj et al . , unpublished data ) .	When cells multiply , it is essential for each new cell to get a copy of the organism's genetic blueprint . If an error occurs during cell division , and one of the daughter cells ends up with too many or too few copies of a chromosome , the cell can die or malfunction . Errors during cell division can , for example , cause cancer . Before a cell divides , it must create an exact copy of each of its chromosomes . The two copies of the chromosome are linked together at a region called the centromere . To separate them , structures called microtubules attach to each side of the centromere via a structure called the kinetochore . The kinetochore then sends out signals orchestrating how the microtubules should move in order to pull the chromosomes apart . In yeast , it is known that a protein called Cse4 must be present at the centromere for cell division to be successful . However , researchers have come to conflicting conclusions about how many copies of this protein are needed and how they function as the chromosome copies are separated . Wisniewski et al . now reveal that a 'tag' scientists use to make Cse4 more visible under a microscope may have skewed the results of some studies . Attaching a tag to the end of the protein interferes with its function , slowing down cell growth , and even killing cells at high temperatures . This could explain the disagreements about how Cse4 works . Placing a tag inside Cse4 , on the other hand , allows the protein to behave normally . Using such an internal tag , Wisniewski et al . found that , as the cell copies its chromosomes , old Cse4 is removed and replaced by new molecules . Those proteins then remain attached to the centromere throughout cell division . A second protein called Scm3 helps to hold the Cse4 in place . By clarifying the number and behavior of various crucial components of the kinetochore , this work opens avenues to better understand the process of chromosome separation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "cell", "biology" ]	2014	Imaging the fate of histone Cse4 reveals de novo replacement in S phase and subsequent stable residence at centromeres
The tuberculosis ( TB ) epidemic is fueled by a parallel Human Immunodeficiency Virus ( HIV ) epidemic , but it remains unclear to what extent the HIV epidemic has been a driver for drug resistance in Mycobacterium tuberculosis ( Mtb ) . Here we assess the impact of HIV co-infection on the emergence of resistance and transmission of Mtb in the largest outbreak of multidrug-resistant TB in South America to date . By combining Bayesian evolutionary analyses and the reconstruction of transmission networks utilizing a new model optimized for TB , we find that HIV co-infection does not significantly affect the transmissibility or the mutation rate of Mtb within patients and was not associated with increased emergence of resistance within patients . Our results indicate that the HIV epidemic serves as an amplifier of TB outbreaks by providing a reservoir of susceptible hosts , but that HIV co-infection is not a direct driver for the emergence and transmission of resistant strains . Among the estimated 1 . 5 million people who died from TB in 2013 , 360 , 000 were HIV co-infected and 200 , 000 cases were caused by multidrug-resistant TB ( MDR-TB ) ( World Health Organization , 2015 ) . Until the late 1980s , reports of MDR-TB were rare , and transmission of such strains was even less frequent ( Reves et al . , 1981; Small et al . , 1993; Wells et al . , 2007 ) . The MDR-TB burden surged concurrently with the human immunodeficiency virus ( HIV ) pandemic and most reported early MDR-TB outbreaks mainly affected HIV co-infected individuals in hospitals and prisons ( Small et al . , 1993; Wells et al . , 2007; Ritacco et al . , 1997 ) . There are good epidemiological reasons to suspect that the HIV and MDR-TB pandemics are fueling each other . Not only does HIV infection render people more susceptible to develop active TB by weakening their immune system , but anti-TB drugs can also directly interfere with antiretroviral treatment . Rifampicin ( RIF ) , one of the cornerstones in anti-TB therapy , has been shown to significantly lower serum concentrations of HIV protease and reverse transcriptase inhibitors ( Burman et al . , 1999; Centers for Disease Control and Prevention , 1998 ) . To make matters worse , HIV co-infection is also associated with malabsorption of anti-TB drugs . This pattern is particularly pronounced for RIF , but seems to hold true for most anti-TB drugs ( Patel et al . , 1995; Peloquin et al . , 1993 ) . HIV co-infection might also directly contribute to the accumulation of resistance in Mtb . First , as resistance mutations generally entail a fitness cost to the bacterium ( at least initially ) , some resistant strains might be more successful in HIV+ hosts with weakened immunity leading to a reduced selective pressure on the bacillus . Second , some antiretroviral drugs used to treat HIV might have a mutagenic effect on mycobacterial genomes , but this has yet to be investigated ( McGrath et al . , 2014 ) . HIV co-infection and very low CD4 lymphocyte counts ( <100 cells/mm3 ) , a hallmark of advanced HIV infection , have been shown to be risk factors for developing resistance to RIF and to a lesser degree isoniazid ( INH ) ( Bradford et al . , 1996; Burman et al . , 2006; Li et al . , 2005; Porco et al . , 2013 ) . However , a systematic review of 32 studies assessing MDR-TB prevalence by HIV status did not demonstrate an overall association between acquired MDR-TB and HIV , but suggested that HIV co-infection is a risk factor for contracting primary MDR-TB ( Suchindran et al . , 2009 ) . In summary , the association between HIV co-infection and Mtb drug resistance remains unclear , with a number of studies yielding conflicting results ( Small et al . , 1993; Chum et al . , 1996; Lukoye et al . , 2011; Meyssonnier et al . , 2012; Robert et al . , 2003 ) . Attempts have also been made to model the impact of HIV on TB incidence and resistance ( Sergeev et al . , 2012 ) , but in lieu of empirical data , such studies relied on a number of assumptions on both host and pathogen biology as well as the interactions between them . It is beyond doubt that HIV has been a driver of increased TB incidence globally , but a recent review of the subject actually found HIV co-infection to be associated with decreased rates of TB transmission within households and between close contacts ( Kwan and Ernst , 2011 ) . This observation is possibly explained by differing manifestation of TB in HIV positives , namely less frequent cavitation and lower pulmonary bacillary load ( Kwan and Ernst , 2011 ) . External factors such as social isolation or HIV infected patients being followed up more closely than HIV negatives may also contribute to this pattern ( Kwan and Ernst , 2011 ) . Indeed , in a low-incidence setting of close follow-up , HIV co-infection was associated with reduced TB transmission ( inferred by clustering ) and TB among HIV co-infected was at least partly due to transmission from HIV-negative patients ( Fenner et al . , 2012 ) . In the current work we aimed to directly investigate the impact of HIV co-infection on the evolution of antibiotic resistance emergence and on transmission dynamics . We analyzed the genomes of 252 isolates belonging to the largest reported outbreak of MDR-TB in South America , caused by the M strain ( Ritacco et al . , 1997; Eldholm et al . , 2015 ) . The isolates were collected from patients with known HIV status from the mid-90s until 2009 , providing important temporal information . To assess the impact of HIV co-infection on Mtb evolutionary rates , we estimated mutation rates in the terminal branches of a time-labelled phylogenetic tree , roughly corresponding to the evolutionary history of individual clinical Mtb isolates within sampled patients . We also inferred transmission networks by implementing a novel epidemiological model accounting for the long latency of TB . Finally , we estimated the length of the latent period by combining the results of the phylogenetic reconstruction and inferred transmission networks . We found that HIV status of the host does not affect the mutation rate of Mtb , and that drug resistance is not more likely to evolve in HIV positive than HIV negative patients . Together these findings suggest that HIV co-infection is not a direct driver of Mtb drug resistance , which fits well with the distribution of the global burden of TB , MDR-TB and HIV . Reconstructed transmission networks did not reveal a significant impact of HIV co-infection on the ability of patients to transmit TB . However , our estimates of TB latency confirm that HIV co-infection accelerates progression to active TB . After filtering out positions with low mapping quality and removal of single nucleotide polymorphisms ( SNPs ) in problematic regions , a total of 509 SNPs separating the 252 isolates were used to construct a Bayesian phylogeny ( Figure 1 ) ( Eldholm et al . , 2015 ) . The majority of the isolates in the study shared the same six mutations yielding resistance to INH , RIF , streptomycin , kanamycin , pyrazinamide and ethambutol ( Eldholm et al . , 2015 ) . The bulk of resistance mutations evolving within the outbreak were thus made up of ethionamide ( ETH ) and fluoroquinolone ( FLQ ) resistance mutations . The HIV status was known for all patients in the study , of which 60 . 7% were HIV positive . 10 . 7554/eLife . 16644 . 003Figure 1 . Whole-genome Bayesian evolutionary phylogeny of the M outbreak . The peripheral color strips indicate the HIV status of patients from which the clinical isolates were collected and the resistance burden of the isolate . The scale bar is given in years since the most recent common ancestor of the outbreak . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 003 Based on the available data we considered that the sequenced outbreak isolates represented about one third of the total number of individuals belonging to the outbreak . Based on available RFLP data and estimates of the proportion of MDR-cases in Argentina belonging to the M strain , the outbreak is believed to have caused about 550 cases between 1992 and 2002 , of which 109 genomes were available for study ( 20% ) . A large fraction of isolates from before 2001 , which includes the peak of the outbreak , were lost in a freezer accident . From 2003 to 2009 , the M strain caused 228 cases in Argentina , of which 143 genomes were available ( 63% ) , 116 isolates were sequenced from HIV positive patients and 85 from negatives . Lost isolates amounted to 40 positives and 25 negatives . For these years there are hence no reason to suspect any bias in the HIV status of the sampled patients ( χ2; p = 0 . 53 ) . Lost samples can potentially inflate the length of the terminal branches as they can result in missing internal nodes , but any inflation in branch length is thus expected to apply equally to branches leading to isolates sampled from HIV positive and negative patients . To investigate the impact of HIV-TB coinfection on the accumulation of mutations in Mtb genomes we directly counted the mutations occurring on terminal branches by performing an ancestral reconstruction analysis in PAML ( Table 1 , Figure 2—figure supplement 1 ) ( Yang , 2007 ) . We observed no significant differences in the rate at which substitutions accumulate in the genomes of strains evolving in HIV positive and negative patients ( Figure 2a ) . However , terminal branches were significantly longer and contained significantly higher numbers of mutations in HIV negative patients than in positives ( Table 1 and Figure 2—figure supplement 1 ) , possibly reflecting a slower progression of TB in HIV negatives relative to positives . 10 . 7554/eLife . 16644 . 004Table 1 . Number of SNPs accumulated in clinical isolates . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 004Host HIV status n Mutations total Mean number per isolate χ2 p-valueNegative 99 262 2 . 646 < 0 . 001 Positive 153 277 1 . 810 10 . 7554/eLife . 16644 . 005Figure 2 . Impact of HIV co-infection on Mtb evolution . Left: Rate of evolution ( substitutions/site/year ) on terminal branches ( p = 0 . 1920 ) . Right: resistance load ( number of antimicrobials to which resistance-conferring mutations were found in clinical Mtb isolates , stratified by HIV status of the host . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 00510 . 7554/eLife . 16644 . 006Figure 2—figure supplement 1 . Evolution of Mtb within patients as a function of HIV status . From top to bottom: Rate of evolution ( substitutions/site/year ) ( p=0 . 1920 ) . Terminal branch lengths ( p=0 . 0006 ) . Number of SNPs on terminal branches ( p=0 . 0009 ) . *** denotes p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 006 Hypothesizing that HIV-coinfection could either be a direct driver of resistance emergence or increase the susceptibility to contract additionally resistant isolates , we tested whether patient HIV status was associated with resistance load , by counting the number of resistance determinants present in each Mtbisolate ( Supplementary file 1: Sample information ) and stratifying the data by HIV status . We found that the resistance load was near identical between Mtb isolates from HIV positive and negative patients ( mean = 5 . 99 and 6 . 04 respectively ) ( Figures 1 and 2b ) . These results are in line with those from a very recent study of drug resistant TB in Kwazulu-Natal ( Cohen et al . , 2015 ) . The analysis above does not distinguish between mutations that emerged in the patients included in our sample ( acquired resistance ) and those acquired earlier in unsampled patients and subsequently transmitted to the patients in our sample ( primary resistance ) . To investigate the impact of HIV co-infection on evolution of new resistance , we collected available treatment history for patients from whom isolates with terminal branch resistance mutations had been sampled ( Supplementary file 2: Treatment histories ) . We excluded secondary mutations in resistance genes , namely katG and rpoB mutations , in isolates already harboring high-level resistance mutations in these genes . These mutations could either be random events or be involved in fitness compensation , but not resistance per se . We then excluded isolates collected from patients who had not been treated with drugs relevant for the terminal branch resistance mutation , as these most likely represent mutations that evolved in unsampled cases and subsequently transmitted to a sampled secondary case . This left 13 resistance mutations that evolved with high probability during therapy in 11 patients ( Table 2 ) . Nine events of acquired resistance occurred in seven HIV negatives and four in HIV positives . Based on the frequency of HIV co-infection among the sampled patients , HIV negative patients were overrepresented among cases of acquired resistance , but the difference was not significant ( p= 0 . 24 , Fisher’s exa ct test ) . While the sample size is arguably small , this finding does also not implicate HIV as a driver of Mtb drug resistance within the outbreak . 10 . 7554/eLife . 16644 . 007Table 2 . Identified events of within-patient acquired resistance . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 007Isolate ID HIV Treatment history Mutation Acquired resistance 107 - follow-up ( ETH* treated ) ethA L225fs ETH 108 - follow-up ( ETH and FLQ treated ) ethA S208P ETH 516 - follow-up ( unknown treatment ) pncA D129G PZA 1757 - follow-up ( ETH and FLQ treated ) ethA H22P ETH 2098 - follow-up ( ETH and FLQ treated ) ethA F302S + gyrB D461V ETH + FLQ 2485 - follow-up ( unknown treatment ) ethA G437fs ETH POGU - follow-up ( ETH and FLQ treated ) ethA R259fs + gyrB R292G ETH + FLQ 110 + follow-up ( ETH and FLQ treated ) gyrB R446S FLQ 257 + follow-up ( ETH and FLQ treated ) inhA -15 C>T ETH 1298 + follow-up ( ETH and FLQ treated ) gyrA D94N FLQ 2569 + follow-up ( ETH treated ) ethA S251fs ETH Patient received the ETH analogue prothionamide To investigate the impact of HIV co-infection on transmission of Mtb , we implemented a new method to infer transmission events based on the timed phylogenetic tree ( Figure 1 ) . This was needed because a phylogeny is not directly informative about transmission events as a result of within-host diversity and evolution ( Didelot et al . , 2016; Pybus and Rambaut , 2009 ) . Our methodology is briefly outlined below and explained in more details in the materials and methods section . A coalescent within-host model ( Didelot et al . , 2014 ) was combined with a Susceptible-Exposed-Infectious-Removed ( SEIR ) epidemiological model ( Lekone and Finkenstädt , 2006 ) . The likelihood of transmission from one host to another can be computed under this combined model , and this calculation was performed for all pairs of individuals with one acting as potential infector and the other as potential infectee . The likelihood calculation relies solely on the dates at which the two individuals were sampled , their relative position on the phylogeny , and whether the putative infector was smear positive or negative . It does not incorporate other information such as HIV status , so that these effects can be tested separately . The SEIR model was set up with parameters for latency ( mean of 5 years with 95% CI 46 days – 18 . 5 years ) and infectious period ( mean of 120 days with 95% CI 3–443 days ) . The infectious period includes time from symptom onset to infection clearance . A standard method for diagnosing TB is direct microscopy of sputum smears . If bacteria are visible under the microscope , the case is denoted smear positive . If no bacteria are observed , but Mtb can be cultured from the sputum , the case is culture positive but smear negative . Smear positive cases transmit TB far more efficiently on average than smear negative cases . We thus applied a so-called smear-correction , penalizing transmission event likelihoods involving a smear negative transmitter by multiplying likelihood values with 0 . 05 . The resulting matrix contains likelihoods of all possible transmission events ( Figure 3—source data 1 ) . For each of the 252 sampled cases in the outbreak , we extracted the most likely transmitter , resulting in 251 identified transmission pairs . Examples of transmission graphs and transmission events mapped on the phylogeny are shown in Figure 3 whereas full transmission graphs are presented as figure supplements ( Figure 3—figure supplements 1 and 2 ) . Figure 3—source data 2 provides the links between transmission graph nodes and sample IDs . We performed a simulation analysis to test the accuracy of our transmission analysis method , and sensitivity analyses to ensure that our results were robust to parameter choice ( see Materials and methods ) . 10 . 7554/eLife . 16644 . 008Figure 3 . Reconstruction of transmission events . ( A ) Graphs representing two selected high-likelihood transmission chains . The colors of the edges indicate the probabilities of each transmission event from high ( red ) to lower ( orange ) . Patient HIV-status is indicated by grey ( negative ) and blue ( positive ) . ( B ) The corresponding transmission chains annotated in the timed phylogenetic tree . Red color highlights isolates linked by transmission events from a single source . Branches in magenta indicate subsequent transmission from a secondary case to additional cases ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 00810 . 7554/eLife . 16644 . 009Figure 3—source data 1 . Likelihood matrix of all possible pairwise transmission events . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 00910 . 7554/eLife . 16644 . 010Figure 3—source data 2 . Conversion table linking transmission graph nodes and sample IDs . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 01010 . 7554/eLife . 16644 . 011Figure 3—figure supplement 1 . Inferred transmission graph including all 251 transmission events ( grey boxes HIV negative; blue HIV positive ) . Graph edges colored by likelihood from high ( red ) to low ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 01110 . 7554/eLife . 16644 . 012Figure 3—figure supplement 2 . Inferred transmission graph including only the most likely transmissions after applying various cut-offs ( grey boxes HIV negative; blue HIV positive ) . Graph edges colored by likelihood from high ( red ) to low ( yellow ) . ( A ) Top 45% likely transmissions . ( B ) Top 35% likely transmissions . ( C ) Top 25% likely transmissions . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 01210 . 7554/eLife . 16644 . 013Figure 3—figure supplement 3 . Top 25% likely transmission events mapped on the timed phylogeny . Red coloring is used to highlight isolates linked by transmission events from a single source . Branches in magenta indicate isolates transmitted further from one of the secondary cases to other cases ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 013 Next , we analyzed transmission events as a function of the HIV status of the transmitter of transmitter-receiver pairs . We found no significant effect of HIV status on the ability of patients to cause secondary TB cases ( Table 3 ) . Due to incomplete sampling , a proportion of identified transmission pairs are expected to be spurious , as unsampled intermediary hosts go undetected . To account for this , the analyses were repeated including only the most likely transmission events using three thresholds of increasing stringency ( top 45% , 35% or 25% most likely transmissions ) . These subsets are expected to be increasingly enriched for true transmission pairs , but subsampling did not affect the original findings ( Table 3 ) . We also explicitly investigated the distribution of the number of transmissions per transmitter to test whether this could be affected by HIV status , but detected no significant differences between HIV-status of transmitters ( Table 4 ) . The 25% most likely infection events were mapped onto the time-labelled phylogeny for a visual integration of the modelled transmission links ( Figure 3—figure supplement 3 ) . 10 . 7554/eLife . 16644 . 014Table 3 . Number of reconstructed transmission events . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 014Transmission event cut-off Donor HIV status Observed Expected Obs/Exp χ2 p valueAll transmissions Negative 80 98 . 61 0 . 81 0 . 3185 Positive 171 152 . 39 1 . 12 Top 25% events Negative 20 24 . 75 0 . 81 0 . 2205 Positive 43 38 . 25 1 . 12 Top 35% events Negative 30 34 . 57 0 . 87 0 . 3185 Positive 58 53 . 43 1 . 09 Top 45% events Negative 36 44 . 39 0 . 81 0 . 1060 Positive 77 68 . 61 1 . 12 10 . 7554/eLife . 16644 . 015Table 4 . Distribution of transmissions as a function of HIV status of transmitter . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 015All transmission events Transmissions per transmitter: Kruskal-Wallis p value HIV status none 1 2 3 4 5 6 7 8 9 10 11 neg 50 37 4 2 1 5 0 0 0 0 0 0 0 . 075 pos 63 58 11 10 5 2 2 1 0 0 0 1 Top 25% likely transmission events Transmissions per transmitter: HIV status none 1 2 3 4 5 neg 83 15 0 0 0 1 0 . 304 pos 121 25 3 4 0 0 Top 35% likely transmission events Transmissions per transmitter: HIV status none 1 2 3 4 5 neg 75 21 2 0 0 1 0 . 505 pos 111 33 4 4 0 1 Top 45% likely transmission events Transmissions per transmitter: HIV status none 1 2 3 4 5 neg 69 27 2 0 0 1 0 . 324 pos 100 39 7 5 1 1 To further assess performance of the epidemiological modelling , we investigated whether six pairs of isolates with known epidemiological links ( epi-pairs ) had been identified by the transmission analysis . Four pairs of household contacts were identified as likely transmission pairs by the genomic analysis . All four were among the 35% most likely transmission events , and two among the top 25% . The SNP differences between these epi-pairs ranged from one to three SNPs ( Supplementary file 3: SNP distances between epi-pairs ) . The remaining two epi-pairs were not identified as likely transmission events . These included one pair of household contacts and one pair of isolates from the same patient taken 4 . 5 years apart . The genomic differences were nine and five SNPs respectively , which explains why the model did not identify these as likely transmission pairs . Interestingly , drug resistance had evolved in one of the epidemiologically linked isolates in both of these pairs , but in none of the four other pairs . We previously showed that a large number of mutations can hitchhike in the genetic background of resistance mutations sweeping to fixation and hypothesized that such selective sweeps could potentially confuse the reconstruction of transmission events ( Eldholm et al . , 2014 ) . These two cases might well exemplify such a situation . However , it cannot be ruled out that the epi-links actually represent independent sources of infection ( re-infection in the serially sampled patient ) . We then set out to estimate the effect of HIV co-infection on the length of TB latency . For pairs of samples connected by a transmission event , transmission of Mtb from host A to B must happen after the date of the node connecting the two isolates in the Bayesian phylogeny ( Figure 1 ) ( Didelot et al . , 2012 , 2013 ) . The date of transition from silent infection to active TB is unknown , but must happen before sampling time , when the active status is known . An upward biased estimate for the length of latency period of individual j is therefore given by the difference between the date of the MRCA of the transmitter i and the receiver j ( when j was not yet infected ) and the date of sampling of j ( by which time j had developed active TB ) . Although this estimator clearly overestimates the latency period , there is no a priori reason to suspect that the bias should be different between HIV negatives and positives . Any significant difference is therefore likely to reflect a difference in length of the actual latency period . Accordingly , we extracted the length ( in years ) of the branches separating the MRCA and recipient of the transmission pairs and stratified the data by HIV status of the recipient . As we did not have an exhaustive sampling of all isolates in the outbreak , not all individuals would have donors present in the phylogenetic tree . To account for this , we analyzed branch lengths of the receiver for all 251 inferred transmissions , and separately for the 45% , 35% and 25% best supported transmission events , respectively . Again , we expected the proportion of genuine transmissions to increase in frequency as we restricted the analysis to a smaller subset of the best-supported transmissions . The length of the branches leading to HIV negative hosts was significantly longer than for HIV positive hosts when including all 251 estimated transmission events ( p<0 . 001 ) , and this difference remained significant for all three subsampling regimes ( Figure 4 ) . 10 . 7554/eLife . 16644 . 016Figure 4 . Estimating latency time as a function of HIV status . ( A ) For pairs of samples connected by a transmission event from i to j , transmission of Mtb is expected to have occurred on the terminal branch above j . Even though we do not know exactly when j went from latent TB to active TB , the latent period is included in the length of the terminal branch leading to j ( see main text ) . We therefore use this branch length as an upwardly biased estimate for latency time . ( B ) For transmission pairs in the calculated transmission networks , the length ( in years ) of terminal branches leading to the recipient of the pairs ( overestimated latency period ) was extracted and stratified by HIV status of the recipient . To account for incomplete sampling , the analyses were performed on all 251 calculated transmission events as well as subsets including only the most likely transmission pairs ( top 45 , 35 and 25% ) . *denotes p<0 . 001 , denotes p<0 . 05 as determined by unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 016 When including only the top 25% of transmission events , the average branch lengths were 5 . 56 versus 4 . 65 years for HIV negative and positive receivers , respectively . A comprehensive review of 52 studies found that the average time from TB symptom onset to diagnosis ( diagnostic delay ) is approximately two months , with no significant difference between high and low income countries ( Sreeramareddy et al . , 2009 ) . Another meta-study found that HIV positive status was associated with both increased and decreased diagnostic delay , depending on study setting ( Storla and Yimer , 2008 ) . The study most relevant to the current setting was conducted in 2005 in Buenos Aires and other Argentinean provinces and found a delay of about three months , with no significant effect of HIV status ( Zerbini et al . , 2008 ) . As the difference in TB activation time we infer between HIV- and HIV+ is several times higher than the diagnostic delay reported in any setting , we feel confident that it reflects faster progression to active TB in HIV+ patients . The fact that HIV co-infection significantly increases the rate of reactivation of latent TB is well documented . A comprehensive study from the United States found the rate of reactivation to be 25-fold higher in HIV co-infected individuals relative to their HIV-free peers ( 1 . 82 vs 0 . 072 per 100 person-year ) ( Shea et al . , 2014 ) . However , our outbreak analysis is necessarily restricted to people who develop active TB , and in this subset of cases , HIV co-infection seems to be associated with a relatively modest acceleration of TB progression , speeding up the process by about 11 months . We do not know when individual patients contracted TB and HIV respectively . Hypothetically , the accelerating effect of HIV co-infection on TB progression is likely to be underestimated in patients who were infected with HIV significantly later than TB . Conversely , patients who were infected with TB late in the study period might be enriched for HIV co-infection as these patients were more likely to develop TB in time for inclusion in the study . However , as the study period was relatively long , we do not believe this potential bias to have significantly affected our results . The single most important impact of HIV infection in this large multi-decade outbreak of MDR-TB seems to be an increase in the proportion of patients who develop active TB . The HIV prevalence in Argentina is approximately 0 . 4% ( in 2001 and 2014 ) ( World Health Organization , 2013 ) , whereas the proportion of HIV co-infected individuals is 60 . 7% within the M outbreak . These numbers demonstrate that HIV infection is a massive risk factor for developing TB with the MDR M strain . We found that HIV co-infection is associated with a moderately faster , yet statistically significant , progression to active TB . As this subtle effect of HIV status on time to active TB cannot explain the far higher incidence of the M strain in HIV positives , this suggests that the main effect of HIV co-infection is to increase the absolute risk of developing active TB . In other words , we surmise that a large proportion of HIV negatives infected by the M strain will not progress to active TB but for those that do , the latency period is only slightly longer than for HIV positives . This study encompasses an outbreak within which resistance to six common anti-TB drugs evolved early on , and our results are thus mainly restricted to the evolution of resistance to second-line drugs such as ETH and FLQ in individual isolates . Extrapolation of these findings to evolution of resistance to first-line drugs thus requires caution . However , the physiological and societal impact of HIV on TB patients as well as the fitness constraints associated with new Mtb resistance mutations should be fundamentally similar regardless of drug class . It should be noted that free access to highly active antiretroviral therapy in Argentina from 1997 is likely to have mitigated the accelerating effect of HIV on TB progression ( Waisman , 2001; Gupta et al . , 2015 ) . Clinical data on HIV progression was not available and we were thus unable to quantify the effect of anti-retroviral treatment by stratifying our analyses by CD4 counts . We predict CD4 counts would correlate negatively with TB progression . However , we do not expect that antiviral therapy ( and increased CD4 counts ) would nullify the effect of HIV on TB progression . Indeed , it has been shown that TB incidence during highly active anti-retroviral treatment is significantly higher than background levels even though a number of possible confounders makes the exact quantification of the effect of antiretroviral therapy challenging ( Gupta et al . , 2015; Girardi et al . , 2005; Lawn et al . , 2005; Lawn and Wood , 2005 ) . Accurate reconstruction of transmission chains of bacteria with extended periods of within-host evolution remains challenging ( see Materials and methods ) . But even though some artefactual transmissions are bound to be included among the reconstructed high-confidence events , we are confident that the overall pattern of transmission is shaped by actual events and is hence robust . Restricting the transmission network analyses to the most likely transmission events did not affect our finding that HIV status does not significantly impact the transmissibility of Mtb ( Table 3 and Table 4 ) . We also found that HIV co-infection does not affect the rate of Mtb evolution within patients . In fact , Mtb was found to accumulate more mutations in HIV negatives . This likely reflects the slower progression to active disease in this group , with these patients harboring Mtb for a more extended period relative to HIV positives . This pattern holds true also for antimicrobial resistance mutations , which were found to evolve significantly more often in HIV negatives than in HIV positives . We previously showed that the largest clade in the M outbreak had evolved resistance to six antimicrobials by 1979 , well before the HIV epidemic reached Argentina ( Eldholm et al . , 2015 ) , a finding which has been replicated for another highly resistant Mtb lineage in South Africa ( Cohen et al . , 2015 ) . To put our results in a global context , we retrieved data on the burden of TB , MDR-TB and HIV globally from the World Health Organization ( WHO ) Global Health Observatory Data Repository ( http://apps . who . int/gho/data/node . main ) . We observed a strong correlation between TB and MDR-TB prevalence ( Figure 5a ) as well as a correlation between HIV and TB burden between countries ( Figure 5b ) . We also recovered a highly significant correlation between HIV and MDR-TB ( Figure 5c ) . However , when correcting the MDR-TB burden for total TB burden , the correlation vanished ( Figure 5d ) . This is in line with our results on the M outbreak that HIV is a driver of TB in general , but does not disproportionately contribute to the rise of MDR-TB lineages . 10 . 7554/eLife . 16644 . 017Figure 5 . Correlations between global patterns of HIV , TB and MDR-TB prevalence . Clockwise: Per country prevalence of MDR-TB as a function of TB prevalence ( p=2 . 2 × 10−16 ) ; TB prevalence as a function of HIV prevalence ( p=5 . 9 × 10−6 ) ; MDR-TB prevalence as a function of HIV prevalence ( p=1 . 6 × 10−4 ) ; Proportion of MDR-TB cases among TB patients as a function of HIV prevalence ( p=0 . 8 ) . All values are log-transformed . The depth of shading of individual dots reflect the TB prevalence in individual countries . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 01710 . 7554/eLife . 16644 . 018Figure 5—source data 1 . Global per-country health , economy and disease metrics . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 01810 . 7554/eLife . 16644 . 019Figure 5—figure supplement 1 . Correlations between global patterns of HIV , TB and MDR-TB prevalence restricted to the top 50% countries in terms of GDP per capita . Clockwise: Per country prevalence of MDR-TB as a function of TB prevalence ( p=1 . 5 × 10−15 ) ; TB prevalence as a function of HIV prevalence ( p=5 . 3 × 10−5 ) ; MDR-TB prevalence as a function of HIV prevalence ( p=5 . 1 × 10−5 ) ; Proportion of MDR-TB cases among TB patients as a function of HIV prevalence ( p=0 . 11 ) . All values are log-transformed . The depth of shading of individual dots reflect the TB prevalence in individual countries . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 019 By combining Bayesian evolutionary analyses and the reconstruction of transmission networks based on a new epidemiological model , we were able to directly assess the impact of HIV on the evolution and transmission of the single most widespread MDR-TB strain reported to date in South America . The main pre-extensively resistant ( pre-XDR ) clade within the outbreak evolved before the HIV epidemic in Argentina , but HIV patients at a major hospital in Buenos Aires played a central role in fueling the epidemic in the early 1990s ( Ritacco et al . , 1997; Eldholm et al . , 2015 ) , by providing the strain with a large and spatially restricted reservoir of individuals susceptible to develop active TB . Once the outbreak erupted , we find that HIV co-infection did not play a role in accelerating Mtb mutation rates; neither did HIV co-infected patients cause secondary TB cases at significantly higher rates than their HIV negative peers did . Our findings confirm that HIV co-infected patients have increased susceptibility to contract TB , but strongly suggest that they do not drive the evolution of Mtb resistance within an outbreak , nor do they act as super-spreaders of MDR-TB . All available isolates belonging to the M outbreak as assessed by IS6110 RFLP were included in the study ( see ( Eldholm et al . , 2015 ) for additional information on samples ) . The exact number of lost isolates is not known . No IS6110 RFLP data are available for isolates from before 1992; a freezer accident also contributed significantly to sample loss . The protocols used for DNA isolation , preparation of sequencing libraries and SNP calling are described in ( Eldholm et al . , 2015 ) , as are the methods for phylogenetic evolutionary inferences , testing of tip-based calibrations and molecular dating . Sequence reads from the study can be found under European Nucleotide Archive accession PRJEB7669 . Briefly , BEAST 1 . 7 . 4 ( Drummond and Rambaut , 2007 ) was used to infer a phylogeny , branch lengths and evolutionary rates using a general time reversible substitution model with variation among sites simulated using a discrete gamma distribution with four rate categories . We assumed a lognormal relaxed clock to allow variation in rates among branches in the trees . Trees were calibrated using tip dates only with sample time span ranging from October 1996 to December 2009 . Following appropriate testing , we applied an exponential demographic model . Posterior distributions of parameters , including branch lengths and substitution rates were estimated by Markov chain Monte Carlo ( MCMC ) sampling . In this study , we aimed to test for evolutionary differences between strains evolving in HIV positive and negative patients . Because we can only be confident about the HIV status from which the samples were collected from , we restricted these analyses to terminal branches in the tree . We estimated the rates of evolution on terminal branches and compared those leading to HIV- and HIV+ hosts using two sample unpaired t-tests . We used the baseml model implemented in PAML program to perform the empirical Bayesian reconstruction of ancestral sequences . High-likelihood resistance mutations in the genes embB , ethA , gidB , gyrA , gyrB , katG , ndh , mshA , pncA , rpoB , rpsL and rrs were identified as described previously ( Eldholm et al . , 2015 ) . The code used to reconstruct transmission events is available at https://github . com/xavierdidelot/TransPairs . We wanted to reconstruct likely transmission events between sampled individuals , with the added difficulty that we knew a significant proportion of infected individuals were not sampled , so that some of the sampled individuals would have been infected by unsampled individuals . To avoid this difficulty , we developed the following inferential framework in which the likelihood of direct transmission from any sampled host to any other can be calculated . We consider a Susceptible-Exposed-Infectious-Removed ( SEIR ) model where individuals move from E to I at rate γ1 and from I to R at rate γ2 . We also assume that within-host coalescence happens at a constant rate α as in previous work ( Didelot et al . , 2014 ) . We want to calculate the likelihood Li→j of transmission from host i to host j ( Figure 4 ) . Let ti and tjdenote the known times at which the two hosts are sampled . Let ti , j denote the time at which the samples from i and j last shared a common ancestor , which is known from the timed phylogeny ( Figure 1 ) . Let s denote the unknown time at which i transmitted to j , assuming that this is indeed what happened . s is unknown but is greater than ti , j and smaller than both ti and tj . With these notations:Li⟶j=p ( ti , tj , ti , j∣i⟶j ) =∫p ( ti , tj , ti , j∣s ) p ( s ) ds∝∫s=ti , jmin ( ti , tj ) p ( ti∣s ) p ( tj∣s ) p ( ti , j ) ∣s ) ds The first term in the integral is the probability of host i being removed at time ti given that he was infectious at time s and is exponentially distributed with rate γ2:p ( ti∣s ) =γ2e−γ2 ( ti−s ) The second term in the integral is the probability of host j being removed at time tj given that he was exposed at time s and so is a convolution of the exponentials with rates γ1 and γ2:p ( tj∣s ) =γ1γ2 ( e−γ2 ( tj−s ) −e−γ1 ( tj−s ) ) γ1−γ2 The third term in the integral is the probability that coalescence of the two lines present in host i at time s happens at time ti , j and also that either i was infectious at time ti , j and stayed so until s or that host i was latent at time ti , j and became infectious ( but not removed ) by time s , leading to:p ( ti , j∣s ) =αe−α ( s−ti , j ) 2γ1e−γ2 ( s−ti , j ) −γ2e−γ2 ( s−ti , j ) −γ1e−γ1 ( s−ti , j ) γ1−γ2 By injecting the last three equations into the first we get the likelihood of transmission from i to j . These calculations were made for all putative infector-infectee pairs using γ1 = 0 . 2 per year and γ2 = 3 per year and the previously estimated within-host coalescent rate α = 0 . 83 per year ( Didelot et al . , 2014 ) . The likelihoods of transmission from smear negative individuals was multiplied by 0 . 05 to reflect the lower infectiousness of these individuals . The SEIR epidemiological model assumed in the calculations above implies that there is random mixing between the individuals , with every infectious individual being a priori equally likely to infect any susceptible individual . Although the assumption of random mixing is appealing in theory , in practice human population are well known to behave differently , with for example a strong effect of the household structure in the transmission patterns of many pathogens ( Cauchemez et al . , 2004; Whittles and Didelot , 2016 ) . Here we did not have information on the structure of the population and so could not integrate it in our model . Application of our methodology in a setting where such information is available could be performed simply by multiplying the likelihood values with the a priori probabilities of transmission caused by the host population structure . From the full matrix of transmission likelihoods between all pair of strains , we aimed to reconstruct disease transmission as accurately as possible . For each pair [i , j] of the transmission matrix , we started by removing the lowest likelihood value ( i infecting j or j infecting i ) . From the remaining transmission events , we used Edmonds algorithm implemented in the RBGL R package ( Carey et al . , 2016 ) to find the spanning arborescence of minimum weight ( sometimes called an optimum branching ) . An optimum branching is a graph defined as a set of directed edges that contain no cycles and such that no two edges are directed towards the same node . In our reconstruction , such a graph contains n nodes , n being the number of isolates and n-1 directed edges representing the transmission events . As our sampling of the outbreak was not exhaustive , we know that a proportion of direct transmission events did not happen . To deal with that situation , we used various thresholds of inferred transmission events with the highest associated likelihoods to plot the transmission graphs and analyze the distribution of transmission events . In order to test the accuracy of the above method of reconstruction of transmission chains , we simulated an SEIR model for a population of N = 3000 individuals , with a transmission rate of β = 0 . 001 per year , a rate of becoming infectious when exposed γ1 = 0 . 2 per year , and a rate of being removed when infectious γ2 = 3 per year . These values of N and β were selected to produce simulated outbreaks of roughly the same size as in the real data , and these values are not used for inference . The transmission tree generated by this simulation was recorded . A timed phylogeny was then constructed from the transmission tree , using a coalescent within-host evolutionary model with coalescent rate α = 0 . 83 per year , and leaves were randomly removed from this tree to simulate incomplete sampling of cases , keeping two thirds of the leaves in the second half of the outbreak to emulate the sampling frame in our study . The resulting phylogeny ( Figure 6 ) was then analyzed in exactly the same way as the real data: the likelihood of transmission was computed for every pair of leaves , a transmission tree was deduced using Edmonds algorithm , and only the 25% , 35% or 45% most likely transmission links were retained to account for incomplete sampling . When applying these three thresholds to the simulated data , we found that the proportion of correctly inferred links were 74% , 69% and 63% , respectively . These results conform with our expectation given that there is significant uncertainty about who infected whom based on genomic data alone when accounting for extended periods of within-host evolution ( Didelot et al . , 2016; Biek et al . , 2015; Didelot et al . , 2014; Worby et al . , 2014 ) . 10 . 7554/eLife . 16644 . 020Figure 6 . Timed phylogeny used in simulation of SEIR model . DOI: http://dx . doi . org/10 . 7554/eLife . 16644 . 020 The results of our transmission analysis are based on three parameter values , namely a mean latent period of 5 years , a mean infectious period of 120 days and a smear correction of 0 . 05 by which the likelihood of transmission from smear negative individuals is multiplied . These parameters were selected based on the literature and clinical experience . The latent period can vary extensively between people . Approximately two months of diagnostic delay ( Sreeramareddy et al . , 2009 ) plus two months from treatment onset to clearance of the MDR infection ( Brust et al . , 2013 ) suggests that 120 days is a reasonable estimate of infectious period . Finally , A 20-fold decreased transmissibility of smear-negative cases was chosen as a reasonable parameter ( Ma et al . , 2015 ) . We performed a sensitivity analysis to test how reliable our results would be if any of these parameters were inaccurate . For each of the three parameters , we ran the analysis again considering double and half of their specified values above , and compared the reconstructed transmission links with those of the main analysis . In each case , the proportion of links identical with the main analysis was between 91% and 99% . We also performed an analysis in which no smear correction was applied and recovered 90 . 5% of the links in the main analysis . HIV prevalence expressed as% population between the age of of 15 and 49 was downloaded form the World Bank Data website ( http://data . worldbank . org/indicator/SH . DYN . AIDS . ZS ) . TB and MDR-TB prevalence data was obtained from the World Health Organization ( http://www . who . int/tb/country/data/download/en/ ) . For TB prevalence , data was available for all countries for the year 2013 and point estimates of prevalence by 100 k individuals were retrieved ( e_prev_100 k ) . For MDR-TB prevalence , the data was collected less systematically , and relies on a mix of surveillance , surveys and models . We used the estimated number of MDR-TB cases among all notified pulmonary TB cases ( e_mdr_num ) , expressed as prevalence per 100 k individuals by dividing by country population size estimates from the same source . We calculated the proportion of MDR-TB cases by dividing the prevalence of MDR-TB by the prevalence of TB . All four variables ( HIV- , TB- , MDR-TB- prevalence and the ratio of MDR-TB/TB prevalence were transformed as log ( x+1 ) prior to analyses . Pearson correlation coefficients were used to test for significant associations between the prevalence of TB and MDR-TB , HIV and TB , HIV and MDR-TB and finally HIV and of the MDR-TB/TB ratio . The robustness of the prevalence estimates likely vary between countries due to difference in methodology and surveillance effort , which may lead to some biases in the correlations reported in Figure 5 . We reasoned that more robust estimates should be obtained in countries with more developed economies and public health institutions . Thus , we additionally retrieved estimates for 2013 GDP per capita ( http://data . worldbank . org/indicator/NY . GDP . PCAP . CD ) and health expenditure ( % ) http://data . worldbank . org/indicator/SH . XPD . TOTL . ZS . For all countries . Health expenditure was transformed into absolute health expenditure per capita , by multiplying by GDP and dividing by population size of the countries . The source data used in these analyses is provided in Figure 5—source data 1 . We then recomputed the correlations reported in Figure 5 on different fractions ( 25% , 50% and 75% ) of the countries with highest GDP or health expenditure per capita . Prevalence estimates from countries with lower GDP are indeed likely to be less robust as the coefficients between the significant correlations in Figure 5 ( panels A , B and C ) were substantially higher for the countries with high GDP . However , importantly , we never recovered a significant correlation between the prevalence of HIV and the proportion of TB that were MDR-TB . In Figure 5—figure supplement 1 , we report the correlations between the same variables than in Figure 5 for the 50% countries with highest GDP .	Tuberculosis is an infectious disease caused by a bacterium called Mycobacterium tuberculosis that causes more deaths worldwide than any other infection . Individuals who are infected with the Human Immunodeficiency Virus ( HIV ) , which weakens the immune system , are particularly vulnerable to tuberculosis . However , treating individuals who are infected with both HIV and tuberculosis is complicated because the drugs currently used to treat one infection can interfere with the effectiveness of the drugs used to treat the other . Tuberculosis is generally treated with antibiotics . However , some strains of M . tuberculosis are difficult to treat as they have evolved to resist the effects of multiple types of antibiotics . These “multidrug-resistant” bacteria appear to be particularly common in areas where HIV infections are also common . However , it was not known whether HIV directly influences whether M . tuberculosis bacteriaevolve into drug-resistant forms . Eldholm , Rieux et al . have now analyzed the genomes , or total genetic content , of 252 samples of M . tuberculosis taken from the largest outbreak to date of multidrug-resistant tuberculosis in South America . This made it possible to identify the genetic mutations that enable the bacteria to resist antibiotic treatment . Using mathematical models to reconstruct the spread of multidrug resistant M . tuberculosis bacteria during the outbreak also made it possible to assess who transmitted tuberculosis to whom . The results suggest that M . tuberculosis does not evolve drug resistance any faster in patients with HIV than otherwise . Furthermore , patients infected with both HIV and tuberculosis did not transmit tuberculosis to others more often than patients who did not have HIV . However , being infected with HIV did increase the likelihood that an individual would contract tuberculosis . HIV also increased the rate at which the symptoms of tuberculosis progressed in an individual . To clarify the effect of HIV on the spread of tuberculosis , similar studies are needed that collect more complete patient data , including their anti-HIV treatment history and their degree of immune weakening .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2016	Impact of HIV co-infection on the evolution and transmission of multidrug-resistant tuberculosis
There is growing recognition that co-morbidity and co-occurrence of disease traits are often determined by shared genetic and molecular mechanisms . In most cases , however , the specific mechanisms that lead to such trait–trait relationships are yet unknown . Here we present an analysis of a broad spectrum of behavioral and physiological traits together with gene-expression measurements across genetically diverse mouse strains . We develop an unbiased methodology that constructs potentially overlapping groups of traits and resolves their underlying combination of genetic loci and molecular mechanisms . For example , our method predicts that genetic variation in the Klf7 gene may influence gene transcripts in bone marrow-derived myeloid cells , which in turn affect 17 behavioral traits following morphine injection; this predicted effect of Klf7 is consistent with an in vitro perturbation of Klf7 in bone marrow cells . Our analysis demonstrates the utility of studying hidden causative mechanisms that lead to relationships between complex traits . Epidemiological and clinical research has identified a profusion of correlated physiological traits , as well as a remarkably high incidence of co-occurrence and comorbidity among disorders . Various studies have shown that such connections among diseases are typically attributable to a common underlying genetic or molecular mechanism ( Rzhetsky et al . , 2007; Oti et al . , 2008; Barabási et al . , 2011; Cotsapas et al . , 2011; Lee et al . , 2012; Cross-Disorder Group of the Psychiatric Genomics Consortium , 2013 ) . Disclosure of unexpected relationships among disease phenotypes and understanding of their common mechanisms opens the way to improved disease classification and treatment . In particular , it may allow a drug approved for one disease to be used for the treatment of another disease , and provide us with the means to anticipate undesired off-target effects ( e . g . , Ashburn and Thor , 2004; Dudley et al . , 2011 ) . Advanced computational methods have made it possible to study the mechanisms underlying trait connections in an unbiased manner . One approach is to derive trait connectivity based on trait–trait comorbidity , co-occurrence , and correlations ( Hidalgo et al . , 2009; Shi et al . , 2010; Blednov et al . , 2012 ) . As an example , Figure 1—figure supplement 1A illustrates a sample network and Figure 1—figure supplement 1B depicts a group of correlated traits in this network . Relying entirely on trait information , however , makes it difficult to identify the shared mechanisms and to distinguish shared molecular mechanisms from shared environmental influences . Alternatively , a common way to improve predictions is by integrating relationships between genes and traits , using gene–trait correlations , associations , or causal mutations ( Rzhetsky et al . , 2007; Cotsapas et al . , 2011; Baker et al . , 2012; Hwang et al . , 2012; Gat-Viks et al . , 2013 ) . Such pairwise gene–trait connections were used to construct two-layer clusters ( ‘biclusters’ ) consisting of groups of traits linked to the same group of genes . For example , Figure 1—figure supplement 1C depicts a bicluster for the sample network in Figure 1—figure supplement 1A . Notably , although such ‘gene-based’ approaches provide a list of putative non-environmental mechanisms , their utilization has two major drawbacks . First , these approaches assign a single mechanistic layer whereas in fact what is affected by genetic variation is a number of molecular components ( such as transcripts ) , which affect the physiological traits secondarily; thus , the model should include a series of mechanistic layers and the propagation of information between them . Secondly , gene-based approaches do not distinguish between reactive , independent , and causative relationships , whereas molecular components should be related causatively to the group of traits . For example , although transcript g2 is reactive to the p4 trait but does not cause it ( Figure 1—figure supplement 1A ) , it is still part of a two-layer model ( Figure 1—figure supplement 1C ) . Thus , a valid and reliable methodology for understanding similarities among distinct traits should identify a series of layers and causative relationships . We have developed GEMOT , a methodology for constructing a causative model of trait–trait connections . GEMOT addresses the above challenges by constructing three-layer modules in which each module consists of a group of molecular mechanisms ( here , gene transcripts within a ‘transcripts layer’ ) translating between a genetic layer and a layer of traits . In particular , a GEMOT module is focused specifically on causative , non-environmental relationships rather than on relationships of any kind , and accordingly it includes only transcripts that are part of causative relationships ( referred to as ‘driver transcripts’ , see Figure 1A , Figure 1—figure supplement 1D ) . Naively , systematic identification of GEMOT modules could be obtained by a detailed reconstruction of all relationships among variants , traits and transcripts ( e . g . , Neto et al . , 2010; Hageman et al . , 2011; Wang and van Eeuwijk , 2014 ) . However , such a detailed reconstruction is a major computational and statistical burden , especially considering the large number of components . This problem is avoided in GEMOT by the use of a stepwise heuristic approach that drastically reduces the search space and allows scalability to large networks . 10 . 7554/eLife . 04346 . 003Figure 1 . Schematic illustration of the GEMOT algorithm . ( A ) An overview of GEMOT for the scenario depicted in Figure 1—figure supplement 1A . GEMOT incorporates 3 stages ( I , II and III ) that are schematically described in B–D , E–F and G–I , respectively . Stage I: ( B ) High and low link-potential scores for pairs of variants and traits . A typical calculation for variant–trait pairs ( left: v1–p1 , right: v1–p2 ) . Shown are variant–transcript associations ( y-axis ) and transcript–trait correlations ( x-axis ) for each transcript ( black dots ) . GEMOT uses a threshold ( horizontal dashed line ) to compare the distribution of transcript–trait correlation in high and low transcript–variant associations scores . Link potential is high ( left ) when the distributions of correlation differ between the high and low association range more than expected by chance; link potential is low ( right ) where no difference is observed . Notably , a high link potential reflects the potential that some transcripts may translate between a variant and a trait , although such transcripts are not yet specified . ( C ) Bipartite graph construction . GEMOT constructs a graph whose two parts are variants ( squares ) and traits ( diamonds ) ; edge weights are the link-potential scores ( solid lines , high; dashed lines , low ) . ( D ) Bipartite module identification . GEMOT identifies ‘heavy’ biclusters in the bipartite graph ( in this example , 1 module ) . Stage II: ( E ) Transcript link score . The input is provided by all calculated correlation and association scores ( such as the three plots on the left ) . On the right: given a transcript , GEMOT aggregates and ranks all its scores in a horizontal track ( red , correlations; blue , associations ) and uses the distribution of ranks to score the transcript for significance . ( F ) Tripartite module construction . GEMOT adds high-scoring transcripts from E , referred to as ‘candidate transcripts’ ( circles ) , to the module . Stage III: ( G ) Causality p value scores . GEMOT uses a statistical score to assess causative relationships ( blue , significant; white , non-significant ) for each transcript ( row ) and trait ( column ) in a module . Non-causative relationships attain non-significant scores ( cartoon examples on the left ) . ( H ) Module refinement . Starting with the causality p value scores for the tripartite module ( from G ) , GEMOT eliminates non-causative transcripts ( left ) and non-affected traits ( right ) in an iterative manner . ( I ) The resulting GEMOT module ( arrows , causative relationships ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 00310 . 7554/eLife . 04346 . 004Figure 1—figure supplement 1 . An example of GEMOT methodology compared to alternative methods . ( A ) A sample group of related traits and their related mechanisms . The illustration exemplifies a scenario of two causative gene transcripts ( g1 and g3 ) translating between a certain genetic variant ( v1 ) and 2 traits ( p1 , p4 ) . An additional transcript g2 is reactive to trait p4 and an additional trait p3 is affected by v1 independently of any driver transcript . Traits p4 and p6 are influenced by the same environmental factor . ( B ) Trait-based grouping methods , applied on the data from A . All four traits are grouped according to either shared environmental factors or shared molecular factors , but the underlying mechanisms are not specified . ( C ) Gene-based methods , applied on the data from A . The model consists of a group of traits ( p1 , p3 , p4 ) that share enriched relationships to a group of transcripts ( v1 , g1–g3 ) . Although environment-based relationships are purposely omitted here , the model still includes various types of relationships among transcripts and traits . ( D ) Our 3-layer GEMOT methodology , applied on the sample shown in A . The GEMOT module consists of causative transcripts called ‘driver transcripts’ ( g1 and g3 , middle ) translating between a certain genomic interval ( variant v1 ) and a group of traits affected by the drivers ( p1 , p4 ) . The model avoids components that are related only through reactive or independent relationships ( p3 , g2 ) or by environmental factors ( p6 ) . Diamonds , traits; squares , genetic variants; circles , transcripts; empty circles , driver transcripts . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 00410 . 7554/eLife . 04346 . 005Figure 1—figure supplement 2 . Evaluation of methods for identifying ( broad sense ) causative relationships . ( A ) Causal models M1–M5 . ( B ) Shown is an error matrix , where each cell i , j , represents whether prediction in row i is TP , TN , FP or FN , when the data were generated using the model in column j . ( C and D ) AUC ( top ) and FDR ( bottom ) scores ( y-axis ) when using synthetic models of types M1–M3 ( C ) or when using synthetic models M1–M5 ( D ) . Scores are presented across varying levels of σ ( x-axis , using fixed λ = 4; left panels ) or across varying values of λ ( x-axis , using fixed σ = 0 . 6; right panels ) . Presented are four methods ( color coded ) : an AIC-based method ( Lee et al . , 2009 ) , parametric and non-parametric QTLHot ( Neto et al . , 2013 ) , and its modification in GEMOT ( ‘Materials and methods’ ) . The results indicate that GEMOT performs similarly or better than the existing methods in identifying broad-sense causality . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 005 We applied GEMOT to a large dataset of 1738 traits spanning a broad spectrum of physiological , biochemical , clinical , and behavioral traits that were measured across the genotyped BXD mouse recombinant inbred strains ( a cross between the C57BL/6J and DBA/2J strains [Wang et al . , 2003] ) . We used measurements of transcript levels in bone marrow-derived myeloid cells ( Gerrits et al . , 2009 ) . The modules were used to uncover shared driver transcripts underlying collections of closely related or distinct traits . Notably , many of the findings were supported by independent knowledge or data . We also demonstrated the tissue specificity of modules , based on a post-processing analysis of gene expression in additional tissues . Our study highlights the power of causative reconstruction combined with grouping of complex traits to reveal a comprehensive picture of phenome connections . Global identification of traits that share common causal transcripts and genetic mechanisms requires a reliable reconstruction of a global causal network among variants , transcripts and traits—a notoriously difficult problem , particularly in the case of a large collection of traits and high throughput data . We designed GEMOT , a scalable algorithm for the systematic construction of three-layer modules , each consisting of a group of traits , their shared causal driver transcripts and a genetic variant . GEMOT is based on the notion of ‘linked relationships’ between a variant , a transcript , and a trait . Such relationships incorporate a transcript that is both associated with a variant and correlated with a trait , without a direct evaluation of the causative relationship between the three components . In particular , it relies on the observation that causative relationships are expected also to be linked relationships ( but not vice versa ) . It is therefore possible to start by constructing candidate modules based on the potential of variants and traits to be linked through transcripts . The internal structure is then constructed within each of these modules . Notably , the linked relationships are exploited to avoid global reconstruction of the particular types of relationships , which are then confirmed only at the validation stage . GEMOT's input is a collection of traits , genotyping , and molecular data across a population of individuals . GEMOT consists of three main stages ( see ‘Materials and methods’ , Figure 1A ) . In stage I , GEMOT identifies ‘bipartite modules’ consisting of a single genomic interval and multiple traits that are connected by linked relationships through certain transcripts ( e . g . , traits p1 , p3 , p4 and variant v1 in Figure 1A , left ) . In stage II , candidate transcripts are added to the modules solely on the basis of their linked relationships ( e . g . , candidate transcripts g1–g3 , Figure 1A , middle ) ; the resulting ‘tripartite modules’ , however , are not limited to causative relationships . Finally , in stage III , GEMOT refines the tripartite modules by investigating the causal relationships within them and eliminating non-causative transcripts . The output GEMOT module , therefore , finally consists of the validated driver transcripts , the trait ( s ) they affect , and a single causal genomic interval ( e . g . , driver transcripts g1 , g3; traits p1 , p4; variant v1 in Figure 1A , right ) . Overall , each of the first two stages is aimed at filtering relevant objects for the next stage: candidate modules are selected on the basis of their potential to include candidate transcripts ( stage I ) , and candidate transcripts are selected on the basis of an efficient score that is expected to be high in true driver transcripts ( stage II ) . The final stage ( stage III ) is aimed at validating the causative relationships in the candidate modules from the previous stage . In the following we provide additional details about the three GEMOT stages . I . We start by calculating associations between each transcript and each variant ( a ‘variant–transcript association’ ) and correlations between each transcript and each trait ( a ‘transcript–trait correlation’ ) . We combine these two measures in a statistical test to identify variant–trait pairs that have high potential to be linked through many transcripts . We call this scoring scheme a ‘link potential’ ( Figure 1B ) . From these link-potential scores we construct a bipartite graph whose two parts are variants and traits , where edge weights are the link-potential scores ( Figure 1C ) . We use a biclustering approach ( the REL algorithm; [Gat-Viks et al . , 2010] ) to identify within this graph heavy ‘bipartite modules’ , each consisting of a single genomic interval ( harbouring a consecutive list of variants ) and a collection of traits ( Figure 1D ) . Importantly , such bipartite modules do not represent pleiotropy in general , but only pleiotropy that is likely mediated through transcripts . II . We next apply a statistical test to identify transcripts whose linked relationships in the module are higher than expected by chance . This is done by evaluating the correlations and associations of transcripts with the module's traits and genomic interval respectively , computing ‘transcript link scores’ , and using it to filter promising ‘candidate transcripts’ ( Figure 1E ) . We then add the candidate transcripts to the bipartite modules , thus forming ‘tripartite modules’ ( Figure 1F ) . III . In the third stage the aim is to investigate the internal relationships within a module , thus allowing identification of driver transcripts and their affected traits . Recent methods allow structural reconstruction of a causality network ( e . g . , Neto et al . , 2010; Hageman et al . , 2011; Wang and van Eeuwijk , 2014 ) , and can therefore be applied on each module to reveal its internal structure . However , since such methods may fail owing to a lack of scalability to large networks , we use an alternative approach that aims only to identify the relationships among layers , without the need to infer the causative relationships within each of the layers . To that end , we devise the following 2-step procedure: We first assess the causality among all candidate transcripts and traits in a module , assuming a single representative variant for the module's genomic interval ( Figure 1G ) . We use a ‘causality p value’ score to assess the quality of such causative relationships . Next , the modules are refined by the iterative removal of transcripts and traits whose causality p values are non-significant ( Figure 1H ) . The output is a list of ‘GEMOT modules’ , each consisting of a single genomic interval , a group of validated ‘driver transcripts’ , and their affected traits ( Figure 1I ) . Notably , the GEMOT algorithm is a unified pipeline that integrates several independent procedures , including biclustering , causality testing and network reconstruction . In this study we use the ReL biclustering algorithm ( Gat-Viks et al . , 2010 ) ; the causality testing proposed by Neto et al . ( 2013 ) ; and a tailored network reconstruction scheme . However , the GEMOT pipeline is general and can be applied using alternative procedures ( e . g . , biclustering [Tanay et al . , 2002]; network reconstruction [Neto et al . , 2010; Hageman et al . , 2011; Wang and van Eeuwijk , 2014]; causality testing [Lee et al . , 2009; Neto et al . , 2013] ) . We applied GEMOT to study phenome connections using gene expression in myeloid Gr-1+ cells ( Gerrits et al . , 2009 ) and 1738 traits across recombinant inbred BXD mice ( see ‘Materials and methods’ ) . Using GEMOT , we found 40 bipartite modules , 11 tripartite modules , and 8 mature GEMOT modules with non-overlapping sets of drivers ( permutation-based FDR < 10−3 , ‘Materials and methods’; see Figure 2A , Figure 2—figure supplements 1 , 2 and Supplementary file 1A , B ) . For comparison , no GEMOT modules were found in 100 permutation tests , and the average number of bipartite and tripartite modules in 100 permutation tests was 32 and 0 . 06 , respectively ( Figure 2—figure supplement 2 ) . As expected , the observed link potential scores in mice were much higher than would be expected by chance ( p < 10−10 , permutation test; Figure 2—figure supplement 3 and ‘Materials and methods’ ) . It is highly unlikely , therefore , that our GEMOT modules were generated at random ( p < 0 . 01 ) . Notably modules nos . 1–3 , 6–8 show strong correlations between traits , whereas the remaining modules show moderate or weak trait–trait correlations ( Figure 2—figure supplement 4 ) . 10 . 7554/eLife . 04346 . 006Figure 2 . GEMOT modules in BXD mouse strains . ( A ) Shown is a module identifier ( column 1 ) , its genomic interval ( column 2 ) , the numbers of driver transcripts and traits in the module ( columns 3 and 4 , respectively ) , and the main characteristic of its traits ( column 5 , see description in Supplementary file 1A ) . ( B ) Global visualization of the GEMOT modules . The graph presents the genomic intervals ( squares , bottom ) , the transcripts ( circles , middle ) and traits ( diamonds , top ) of all eight resulting GEMOT modules . The transcripts , which are unique to each module , are connected to their module's variants and traits . Sets of traits known to be related to the same process are enclosed in a box and labeled ( top ) . The module's genetic and transcripts layers are color coded as in A; the traits are color coded based on their gender: female ( white ) , male ( gray ) or both ( black ) . Notably , some modules have overlapping collections of traits , or their traits relate to the same process . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 00610 . 7554/eLife . 04346 . 007Figure 2—figure supplement 1 . Linkage disequilibrium . Scatter plots of linkage disequilibrium ( correlation coefficient , y-axis ) between all variants in the genome ( x-axis ) and the single representative genetic variant of a single module ( marked by arrows ) . The analysis was applied only on those BXD mouse strains that were profiled in the myeloid gene-expression dataset ( Gerrits et al . , 2009 ) and were therefore used by the GEMOT algorithm in this study . Notably , for each module , none of the genomic positions outside its genomic interval was in absolute linkage disequilibrium that is larger than 0 . 75 . Within the module's genomic interval , linkage disequilibrium reached 1 , as expected . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 00710 . 7554/eLife . 04346 . 008Figure 2—figure supplement 2 . Application of GEMOT on real and on permuted data . ( A ) Overview of the GEMOT algorithm applied on real data ( in myeloid tissue ) compared to permuted data ( see ‘Materials and methods’ ) . ( B ) A larger number of modules for real data than for permuted data . The histogram shows numbers of modules in the permuted dataset ( x-axis , 100 repeats ) . Numbers of modules in the real data are indicated by gray arrows . The three plots record the numbers of modules that were constructed during stages I to III of the GEMOT algorithm ( left to right , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 00810 . 7554/eLife . 04346 . 009Figure 2—figure supplement 3 . Higher link potential scores for real data than for permuted data . Shown is a histogram representing the frequency ( y-axis ) of different link-potential ranges ( x-axis ) generated on the basis of real data ( myeloid tissue , gray ) compared to permuted data ( black; see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 00910 . 7554/eLife . 04346 . 010Figure 2—figure supplement 4 . Trait- trait correlations in GEMOT modules . ( A ) Shown is the average trait–trait absolute correlation coefficient ( y-axis ) in each of the GEMOT modules ( x-axis ) . ( B ) . Matrices of trait–trait Pearson correlation coefficient ( blue/red scale indicates negative/positive correlations ) for three exemplified modules: modules nos . 2 ( left ) , 3 ( middle ) and 4 ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 01010 . 7554/eLife . 04346 . 011Figure 2—figure supplement 5 . Characterization of GEMOT modules in BXD mouse strains . ( A and B ) Evaluation of pairs of traits predicted to be in the same group based on the GEMOT methodology ( blue ) , the trait-based hierarchical clustering approach ( ‘HC’ , green ) , the gene-based INVAMOD method ( red ) , and random trait pairs ( black ) . Results are presented for GEMOT's 40 bipartite modules compared to the top 40 predictions from each of the existing methods . ( A ) Absolute correlation coefficients between the two traits in a pair ( x-axis ) vs their best average absolute-correlation with a gene transcript ( y-axis ) . ( B ) Percentages of matched biological annotations ( terms ) within the description of 2 traits in a pair ( y-axis ) vs their trait–trait absolute correlation ( x-axis ) . The higher the percentage of matched annotation terms , the higher the agreement between trait-pairing predictions and prior knowledge . The plots were generated using the moving average of a window of 200 trait pairs . In both cases , GEMOT outperformed the other methods along a wide range of trait–trait correlation values . ( C and D ) Evaluation of pairs of traits predicted to be in the same group , presented as in plots A and B but using the 8 GEMOT modules and eight top-scoring groups from each of the existing methods . As with the 40 bipartite modules ( in plots A and B ) , GEMOT outperformed the other methods along a wide range of trait–trait correlation values . ( E ) Correlations between trait pairs . Shown are the absolute correlation coefficients between trait pairs from modules of different grouping methods ( left to right: random , GEMOT , HC , and INVAMOD; x-axis ) . Results are presented for GEMOT's 40 bipartite modules ( left ) and 8 GEMOT modules ( right ) ; in each plot we use the same number of top-scoring groups from each of the compared methods . In each box the central mark is the median; edges are the 25th and 75th percentiles; whiskers extend to the 10th and 90th percentiles . The plot highlights the wide range of trait–trait correlations generated by GEMOT and the gene-based methods . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 011 Given that our findings were significant , we next aimed to obtain a global perspective on the resulting modules . To that end we generated a graph of the modules , where the transcripts layer of each module is connected to the module's traits and genomic interval ( Figure 2B ) . On this graph we marked groups of similar traits , such as different behavioral and physiological responses after ethanol stimulation , or the results of different thermal nociception tests . Notably , the trait collections of the different modules are partly overlapping . For example , five of the modules relate to morphine responses ( nos . 2 , 5–8 ) and two modules relate to thermal nociception ( nos . 3 and 5 ) . Nevertheless , although some of the modules have overlapping traits , there is no overlap between the drivers and variants of the different modules ( Figure 2A and Supplementary file 1B ) , demonstrating GEMOT's ability to predict several distinct underlying mechanisms for the same collection of traits . Furthermore , a focused examination revealed that the traits of each of the modules shared a unique characteristic . For example , module nos . 6 and 2 relate to place preference following morphine injection , but at distinct time intervals ( Figure 2 ) ; similarly , module nos . 1 and 3 relate to an anxiety assay , but with and without ethanol stimulation , respectively . Taken together , our results indicated high-level organization of overlapping collections of traits , while each module reflects a unique molecular and genetic signature that underlies a different trait characteristic . Notably , some of the resulting modules consisted of multiple traits that are related to the same process , whereas others consisted of a collection of distinct traits ( Figure 2 and Supplementary file 1B ) . For example , a module related to morphine response ( module no . 2 ) consists of 17 different traits that were measured following treatment of mice with morphine at different time points and in various behavioral assays . Similarly , a module related to an anxiety assay ( module no . 1 ) consists of two different traits that were measured following treatment with ethanol at different time points . In contrast , and consistently with our goal of identifying novel relationships among traits , module nos . 3 , 4 and 5 suggest previously unknown connections between traits . We next characterized pairs of traits within each group of traits ( ‘trait pairs’ ) to show that the quality of these pairs is not lower than in existing methods . We focused on three main properties of trait pairs: the correlation among traits in a pair; the correlation between a trait pair and the transcripts; and the knowledge-based relationships among traits . As a reference we demonstrate these properties for modules that were generated using three alternative methods: ( i ) the trait-based hierarchical clustering approach ( denoted ‘HC’; [Hastie et al . , 2009] , as in Figure 1—figure supplement 1B ) ; ( ii ) the gene-based INVAMOD algorithm , which identify pleiotropic genetic variants and their associated groups of traits in an agglomerative manner ( Gat-Viks et al . , 2013; as illustrated in Figure 1—figure supplement 1C ) ; and ( iii ) a set of randomly sampled trait pairs ( 5% of all possible trait pairs ) . In particular , the bipartite modules were compared to the top 40 groups from each method ( Figure 2—figure supplement 5A , B ) . Similar results were obtained when the eight GEMOT modules were compared to the top eight groups from each of the alternative methods ( Figure 2—figure supplement 5C , D ) . For each property , we first explain the metric of evaluation and then present the results with GEMOT and with the alternative methods . The morphine module ( module no . 2 , see Figure 3 ) exemplifies the ability of GEMOT to suggest an underlying mechanism for an entire group of traits known to be tightly related . The module consisted of a collection of 17 behavioral assays in the recombinant mouse strains , all carried out to measure their responses to injection of morphine ( 50 mg/kg , over different periods of time ) . Measured parameters included distance traveled , place preference , and vertical activity ( [Philip et al . , 2010]; Supplementary file 1A ) . All module traits showed a strong positive correlation with each other ( \|r\| values ranged from 0 . 6 to 0 . 99 , Figure 2—figure supplement 4 ) and shared similar peaks within the genomic interval of module no . 2 ( Figure 3A , top ) . 10 . 7554/eLife . 04346 . 012Figure 3 . Characteristics of module no . 2 . ( A ) Genetic associations . Shown are the association scores ( y-axis ) across the genomic positions of chromosome 1 ( x-axis ) for four module traits ( top ) and for seven selected drivers ( bottom ) in myeloid cells . The position of the Klf7 gene is marked below the x-axis . ( B–E ) Characterization of driver groups I and II . ( B ) A matrix of traits ( rows ) vs drivers ( columns ) , where the blue/red scale indicates negative/positive Pearson correlation coefficients among them . ( C ) The transcript causality p value scores ( y-axis ) are shown for each module transcript ( log scale ) , assuming a representative variant in the module's genomic interval ( rs13475891 in chr1:62 Mbp ) . ( D ) The histogram represents the Pearson correlation coefficient ( y-axis ) of Klf7 with the remaining drivers ( x-axis ) . ( E ) Genetic effect size of variant rs13475891 on the driver transcripts . The histogram represents the average expression levels of DBA2-carrying individuals minus the B6-carrying individuals ( y-axis ) for each module driver ( x-axis ) . ( F ) Distribution of the causality −log p value scores of Idh1 ( blue ) and of Klf7 ( black ) on each of the remaining drivers in the module . Causality p values were calculated by positioning Idh1 or Klf7 between the module variant and a driver from this module . ( G ) Validation of the effect size of Klf7 perturbation of knockdown ( left ) or overexpression ( right ) on other drivers in bone-marrow hematopoietic stem cells ( y-axis ) . The ‘effect size’ of perturbation ( either knockdown or overexpression ) on a certain transcript g is defined as the difference between the ( log-scaled ) expression of g in normal cells to that in the perturbed cells . In each panel , the first and second columns refer to groups I and II , respectively . **a significant t-test p value for determining whether the mean effect size is different from zero ( FDR < 0 . 06 ) . ( H ) Scatter plot , where for each driver in group I ( a black square ) the y-axis shows the transcript causality score ( for morphine-response traits in this module; p values ) , and the x-axis shows the significance of Klf7 knockdown effect on its transcription level ( t-test p value ) . The plot indicates that in group I , drivers with highly significant causality on behavioral responses to morphine were also significantly influenced by Klf7 knockdown . ( I ) Overall model illustration of module no . 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 01210 . 7554/eLife . 04346 . 013Figure 3—figure supplement 1 . Characterization of module no . 2 drivers . ( A ) Driver–driver correlations in module no . 2 . A matrix of drivers ( columns ) vs drivers ( rows ) , where the blue/red scale indicates negative/positive Pearson correlation coefficients . ( B ) Causality p value score profiles . A matrix of traits ( columns ) vs driver transcripts ( rows ) , where the blue/white scale indicates the significance of causative relationships among the trait , the driver , and a representative variant in the module's genomic interval ( rs13475891 in chr1:62 Mbp ) . ( C ) Relationship between module no . 2 drivers and human glioblastoma survival . The figure depicts the significance of differences in Kaplan–Meier survival plots ( y-axis; −log p value ) between groups of patients with glioblastoma , where groups are based on their levels of expression of a driver in module no . 2 ( x-axis ) . Black/white indicates grouping based on up-regulation/down-regulation vs intermediate regulation of a driver . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 01310 . 7554/eLife . 04346 . 014Figure 3—figure supplement 2 . Relationships among components in module no . 2 . ( A ) Left: Klf7 expression levels ( y-axis ) of all individuals carrying the C57BL/6J allele ( left column ) and DBA/2J allele ( right column ) for the representative variant rs13475891 of module no . 2 . Right: Scatter plot , where for each individual the y-axis shows the measured level of a morphine-related trait ( distance travelled in 0–180 min in males and females ) and the x-axis shows the Klf7 expression . ( B ) Plots are shown as in A , but for the Idh1 transcript . ( C ) Level of a morphine-related trait ( distance travelled in 0–180 min in males and females; y-axis ) of all individuals carrying the C57BL/6J allele ( left column ) and the DBA/2J allele ( right column ) . The plot shows higher levels of the morphine-related trait in C57BL/6J-carrying individuals . ( D ) Effect size of variant rs13475891 on the traits in module no . 2 . The histogram represents average expression levels of DBA/2J-carrying individuals minus the B6-carrying individuals ( y-axis ) for each module trait ( x-axis ) . The plot indicates a higher level of the morphine-related trait in C57BL/6J-carrying individuals , as exemplified in C . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 01410 . 7554/eLife . 04346 . 015Figure 3—figure supplement 3 . Causality of cis-associated transcripts . ( A ) Causality p value scores of Klf7 and Idh1 on several drivers in module no . 2 . The histogram represents the causality p-value scores ( y-axis ) of Idh1 ( blue ) and Klf7 ( black ) on the module's transcripts ( x-axis ) . The plot indicates a stronger influence of Klf7 than of Idh1 . Causality p-value scores were calculated as −log p values in a causality test ( see ‘Materials and methods’ ) , where the expression levels of Klf7 ( or Idh1 ) were positioned between the genetic variant of module no . 2 and the gene expression of another driver transcript in this module . ( B ) Causality p-value scores ( y-axis ) of Klf7 ( left column ) and Idh1 ( right column ) on drivers in groups I ( blue ) and II ( red ) , indicating a stronger influence of Klf7 than of Idh1 in both driver groups . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 01510 . 7554/eLife . 04346 . 016Figure 3—figure supplement 4 . Causative relationships of module no . 2 across multiple tissues . Significance of causative relationships ( blue , high; white , low ) among traits ( rows ) vs the expression of driver transcripts ( columns ) in nine different tissues or cell types ( color coded ) . In all cases a representative variant in the genomic interval of the module was used for analysis of causality ( see Supplementary file 1C ) . Myeloid cells are shown at the top , indicating that causality relationships in this module are observed in myeloid cells , but hardly if at all in other tissues . Source data: 1Gerrits et al . , 2009; 2Geisert et al . , 2009; 3Mozhui et al . , 2012; 4Gatti et al . , 2007; 5Alberts et al . , 2011 . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 01610 . 7554/eLife . 04346 . 017Figure 3—figure supplement 5 . Causative relationships of module no . 2 across time points . A plot of causality −log p values of Klf7 ( y-axis ) , assuming a fixed genetic variant in the Klf7 gene and various behavioral assays ( blue , place preference; green , distance travelled; orange , vertical activity ) that were performed at various time points ( y-axis ) after morphine injection ( data from Philip et al . , 2010 ) . The plot indicates that the causative influences of Klf7 on the three different behavioral assays act mainly between 30 and 120 min after morphine injection . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 017 Module no . 2 consists of a group of 32 driver transcripts ( e . g . , Klf7 , p35 , Lrrk2 ) , all associated with the module's genotype and strongly correlated with each other and with the module traits ( Figure 3A , bottom; Figure 3B; Figure 3—figure supplement 1A ) . For all driver transcripts , the causative relationships were much preferable to the alternative relationships ( p value ≤ 0 . 005 , permutation-based FDR < 6 × 10−5 , ‘Materials and methods’; Figure 3C and Figure 3—figure supplement 1B ) . We found two main groups of drivers ( Figure 3D , E , Figure 3—figure supplement 1A ) . The first ( denoted ‘group I’ ) consists of 21 genes whose transcript levels are negatively correlated with the morphine traits . In this group , individuals carrying the DBA/2J allele in the module's variant have higher gene expression values than those of individuals carrying the C57BL/6J allele ( e . g . , Klf7 in Figure 3—figure supplement 2A ) . The second driver group ( denoted ‘group II’ ) has the opposite correlation with the morphine traits and the opposite genetic effect ( e . g . , Idh1 , in Figure 3—figure supplement 2B ) . These observations coincide with the fact that for all module traits , individuals carrying the C57BL/6J-allele have higher trait values ( Figure 3—figure supplement 2C , D ) . Notably , causality p values in group I are more significant than in group II ( p < 0 . 05 , t-test; Figure 3C ) ; one possible reason is that the two groups might relate to distinct mechanisms that differ in their causality strength . The role of the Klf7 gene in morphine module no . 2 is particularly interesting . Klf7 and Idh1 are the only two cis-associated module drivers and their causative role in mediating morphine traits is highly significant . We hypothesized that the cis-associated variation in gene-expression levels may lead to variation in the trans-associated module drivers . To test this hypothesis we used the causality p value score ( ‘Materials and methods’ ) , but utilized a cis-associated gene ( Klf7 or Idh1 ) as the transcript positioned between the module's genomic interval and another module transcript . Using these scores we found that the causative p values of Klf7 on the remaining module transcripts were substantially more significant than the causative p values of Idh1 on those transcripts ( Figure 3F , Figure 3—figure supplement 3A , p < 10−7 , K-S test ) . This finding holds for each of the driver groups I and II ( Figure 3—figure supplement 3B ) , suggesting that Klf7 , but not Idh1 , likely affects the other module drivers , which in turn affect behavioral activity in response to morphine . The finding of positive and negative correlations of Klf7 with the drivers in groups I and II is particularly intriguing because it suggests that Klf7 is an activator of group I and a repressor of group II ( Figure 3D ) . To validate the suggested central role of Klf7 in mediating variation in other drivers , we analyzed the influences of knockdown and overexpression of Klf7 on gene expression in bone-marrow hematopoietic stem cells from the C57BL/6J mouse strain ( termed Klf7KD and Klf7OE , using three and four biological repeats , respectively; data were taken from [Schuettpelz et al . , 2012] ) . Indeed , whereas knockdown of Klf7 led to a down-regulation of group I transcripts ( p < 1 . 5 × 10−5 , FDR < 5 × 10−5 , t-test ) , it had no influence in group II ( p > 0 . 32; Figure 3G , left ) , in agreement with the predicted role of Klf7 as an activator and repressor of groups I and II , respectively . Furthermore , we found that the more significant the influence of Klf7 knockdown on a driver in group I , the stronger the causative role of the driver on behavioral morphine traits ( Figure 3H ) . Overexpression of Klf7 had the opposite effect , with significant down-regulation of transcripts in group II ( p < 0 . 05 , FDR < 0 . 06 , t-test ) and even a small increase of transcripts in group I ( p < 0 . 002 , FDR < 0 . 005; Figure 3G , right ) , consistently with our model . Taken together , the two lines of evidence—both natural and experimental perturbations—indicated that Klf7 is a key driver mediating the effects of additional drivers in groups I and II , which in turn affect morphine response diversity ( Figure 3I ) . Module no . 2 may affect morphine responses through a variety of mechanisms . For example , the p35/Cdk5 driver directly phosphorylates the opioid receptor ( Xie et al . , 2009; Pareek et al . , 2012 ) , and the morphine adduct MO-GSH is controlled by the Idh1 and Ggt1 drivers ( Correia et al . , 1984; Kumagai et al . , 1990; Muller and Do , 2012 ) . Furthermore , morphine treatment may exert its action through cell migration and cell invasion processes ( Gach et al . , 2011 ) : the p35 driver , as part of the p35/Cdk5 complex , affects the Rac/Cdc42 complex through PAK inhibition , resulting in altered cell migration ( Nikolic et al . , 1998 ) , while the Mmp13 driver alters cell migration in response to morphine treatment because of its ability to degrade collagen ( Gach et al . , 2011; Wang et al . , 2013 ) . Both the Klf7 and the Cdk5/p35 drivers activate p27 by expression or phosphorylation ( Laub et al . , 2001; Smaldone et al . , 2004 ) , and p27 in turn affects the Rho GTPases Rac and RhoA , which then alter cell invasiveness and infiltration ( Kawauchi et al . , 2006 ) . In glioblastoma , for example , the p27/Rho pathway affects infiltration of tumor cells ( Ruiz-Ontañon et al . , 2013 ) . In agreement with this prediction , using the Rembrandt database ( Madhavan et al . , 2009 ) we found that all four top-ranked Klf7-mediated drivers attain significant effects on survival of patients with glioblastoma ( Kaplan–Meier p < 1 × 10−11 , 1 . 4 × 10−4 , 1 × 10−4 and 0 . 01 for the four drivers Ntng2 , p35 , Rexo2 and Lrrk2 , respectively; in all cases , FDR < 0 . 05 , Figure 3—figure supplement 1C ) , supporting the role of module no . 2 in cell invasiveness . Further experimental studies are required in order to test these suggested pathways and search for additional mechanisms . Our findings in module no . 2 agree well with previous studies showing that several driver genes participate in the morphine response . For example , the Klf7 transcript is up-regulated in response to morphine ( being one of the top ten up-regulated genes [Suzuki et al . , 2003] ) . Both the p35 driver and its activated protein Cdk5 were up-regulated in response to acute morphine but down-regulated on exposure to chronic morphine ( Ferrer-Alcón et al . , 2003 ) . The coregulated Idh1 and Ggt1 drivers ( Muller and Do , 2012 ) are responsible for the synthesis and degradation , respectively , of the reduced form of glutathione ( GSH ) . A conjugated form of morphine and GSH ( MO-GSH ) attains higher morphine reactivity ( Correia et al . , 1984; Kumagai et al . , 1990 ) that may alter the morphine responses . However , whereas key roles of Klf7 have been reported primarily in neurons ( Bieker , 2001; Laub et al . , 2005 ) , here we found a causative effect of Klf7 on behavioral responses to morphine that was specific to myeloid tissue ( Figure 3—figure supplement 4 ) . We cannot as yet explain this observation; however , p35/Cdk5-mediated neutrophil secretion ( Rosales et al . , 2004 ) and the cytokine-mediated regulation of Klf7 by morphine in lymphocytes ( Suzuki et al . , 2003 ) potentially provide an explanation for this tissue specificity . Furthermore , morphine injection leads to a reduction in neutrophil infiltration 30–120 min after treatment ( Clark et al . , 2007 ) , in agreement with the timing of causative relationships between Klf7 and behavioral assays at 30–120 min after morphine injection ( Figure 3—figure supplement 5 ) . Taken together , our results suggest that in vivo behavioral responses to morphine are affected not only by neuronal activity , but also through certain components of the immune system . In the following we demonstrate GEMOT's ability to identify a model for previously uncharacterized connections , with either strong ( module no . 3 ) or weak ( module no . 4 ) correlations among traits . Module no . 3 ( Figure 4A ) shows the ability of GEMOT to group a variety of distinct traits . This module consists of a genomic interval in chr4:133–142 Mbp and five correlated traits , namely pain response ( thermal nociception [Philip et al . , 2010] ) , lens weight ( Zhou and Williams , 1999 ) , eye weight ( with or without correction for brain weight [Zhou and Williams , 1999] ) , and ethanol response ( place preference [Cunningham , 1995] ) . All traits were found to be strongly intercorrelated , with \|r\| values ranging from 0 . 67 to 0 . 9 ( Figure 2—figure supplement 4 ) . Eya3 and Cd52 were proposed as cis-associated drivers , a suggestion further supported by the known role of Eya3 in eye development ( Tadjuidje and Hegde , 2013 ) and the involvement of Cd52 in pain signaling ( Poh et al . , 2012 ) . Figure 4—figure supplement 1 shows that the significant causative role of these drivers can be found along the entire myeloid pathway ( including stem cells [Lin−Sca-1+c-Kit+] , common progenitors of the myeloid and erythroid lineages [Lin−Sca-1−c-Kit+] , erythroid [TER-119+] and myeloid [Gr-1+] lineages ) , but not when using data from the lymphoid , eye , or brain tissues . This suggests that Eya3 and Cd52 play a role in pain processes and eye conditions mainly through their functionality in myeloid cells . 10 . 7554/eLife . 04346 . 018Figure 4 . Characteristics of module nos . 3 ( A ) and 4 ( B ) . Left: Genetic associations . Association scores ( y-axis ) across the genomic positions ( x-axis ) for the module's driver transcripts , based on gene-expression data in myeloid cells . Middle: Matrix of correlations among traits ( columns ) vs driver transcripts ( rows ) ; the blue/red scale indicates negative/positive correlation coefficients . Right: Matrix of traits ( columns ) vs driver transcripts ( rows ) , where the blue/white scale indicates the significance of their causative relationships based on gene-expression data from the myeloid tissue . A histogram depicting transcript causality p value scores is shown as in Figure 3C . A representative variant in the module's genomic interval is assumed ( see Supplementary file 1C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 01810 . 7554/eLife . 04346 . 019Figure 4—figure supplement 1 . Causative relationships of module no . 3 in multiple tissues . Significance of causative relationships ( blue , significant; white , non-significant ) among traits ( columns ) vs the expression of driver transcripts ( rows ) in nine different tissues or cell types ( left to right matrices ) . A representative variant in the module's genomic interval is assumed ( see Supplementary file 1C ) . Notably , causative relationships in this module are clearly observed along the entire myeloid lineage ( stem cells [Lin−Sca-1+c-Kit+] , common progenitors of the myeloid and erythroid lineages [Lin−Sca-1−c-Kit+] , erythroid [TER-119+] , and myeloid [Gr-1+] lineages ) , but not in lymphoid tissues or other tissues under study . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 01910 . 7554/eLife . 04346 . 020Figure 4—figure supplement 2 . Causative relationships of module no . 4 in multiple tissues . Profiles of causality p values across module no . 4 , shown as in Figure 4—figure supplement 1 . A representative variant in the module's genomic interval is assumed ( see Supplementary file 1C ) . Causative relationships in this module are clearly observed in myeloid tissue , but not in the other tissues under study . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 020 Module no . 4 ( Figure 4B ) demonstrates GEMOT's ability to identify a group of traits that show weak correlations among themselves , but share the same driver transcripts . The module consists of a genomic interval in chr11:8–19 Mbp and five traits: the measures of two hippocampal structures ( volume and age-related lesions ) from two distinct publications ( Jucker et al . , 2000; Martin et al . , 2006 ) , locomotor response to cocaine ( Philip et al . , 2010 ) , number of liver tumors ( Lee et al . , 1995 ) , and number of haematopoietic stem cells ( Liang et al . , 2007 ) . The module's traits show weak and moderate correlations , with \|r\| values ranging from 0 . 01 to 0 . 58 ( Figure 2—figure supplement 4 ) . Of six suggested drivers , Ythdf2 and Aldh6a1 were predicted as the top drivers ( Figure 4B , right panel ) . The validity of this prediction is supported by the known involvement of Aldh6a1 in brain structure ( Marcadier et al . , 2013 ) , and is further supported by a recent report ( Meyer and Jaffrey , 2014 ) that N6-methyladenosine ( m6A ) modification of RNA , whose readers are Ythdf1–3 , causes an altered locomotor response to cocaine . In this module , a significant causative role was found in myeloid cells , but not in brain , eye , liver , or lymphoid tissue ( Figure 4—figure supplement 2 ) , suggesting a novel function of myeloid cells in regulating neurobiological and behavioral traits—an intriguing possibility that warrants future investigation . We next investigated the ability of GEMOT to identify the subsets of traits that are caused by the same transcripts . We first focused on studying GEMOT's utility for small sub-networks of co-mapped components , where each sub-network consists of a subset of traits caused by the same transcripts , as well as additional transcripts and traits that are independently or reactively related ( Figure 5A ) . Such sub-networks mimic the tripartite modules that serve as input at the third stage of the GEMOT algorithm . A single synthetic data collection consisted of genotyping , phenotyping , and gene expression for 100 such sub-networks with two characteristic parameters: number of traits and noise level; in all cases we used 100 individuals ( ‘Materials and methods’ ) . Using these synthetic data , GEMOT performance was compared to that of three alternative network reconstruction methods: Hageman et al . , 2011; Wang and van Eeuwijk , 2014 ( ‘QPSO’ ) ; and Neto et al . , 2010 ( ‘QTLNet’ ) ( see parameter selection in Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 04346 . 021Figure 5 . Comparative performance analysis on simulated models . ( A ) Illustration of a sub-network model , in which all components are mapped to the same genetic variant v1 but not necessarily through the same relationships . In particular , the model includes k + 2 traits , with k traits p1 , . . , pk that share the same underlying transcripts g1 , g2 , g3 , and two additional traits pk + 1 and pk + 2 that are affected through other mechanisms ( see detailed in Figure 5—figure supplement 1A ) . ( B and C ) Accuracy assessments of the synthetic sub-network depicted in A . Accuracy ( y-axes ) is compared across methods and different data parameters . Results are shown in models of different noise levels ( x-axis , log-scaled; B ) with either 5 ( left ) or 7 ( right ) traits , or over different numbers of traits ( x-axis; C ) with either a low noise level = 0 . 5 ( left ) or a high noise level = 2 ( right ) . The accuracy metric evaluates whether p1 , . . , pk , but not pk + 1 , . . , pk + 2 , share the same mechanisms , as detailed in Figure 5—figure supplement 1B . Plots depict alternative network construction methods ( color coded , see Figure 5—figure supplement 2 ) , indicating that GEMOT has an advantage over existing methods with noise levels ranging between 0 . 25 and 1 , which is the relevant range for the mouse data in this study ( see Figure 5—figure supplement 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 02110 . 7554/eLife . 04346 . 022Figure 5—figure supplement 1 . Accuracy assessment of predicted shared mechanisms in synthetic ( co-mapped ) sub-networks . ( A ) Detailed illustration of the synthetic network ‘Net-1’ ( a simplified illustration of the same network is shown in Figure 5A ) . Among the k + 2 traits , k traits PC={p1C , . . . , pkC} are jointly affected by the same transcripts . ( B ) Accuracy assessment is summarized in an error matrix , where each cell i , j shows the prediction in row i for the mechanism of trait pt is TP , TN , FP or FN , while the actual class of trait pt is in column j . The two actual classes are pt ∈ PC and pt ∉ PC , as exemplified in A . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 02210 . 7554/eLife . 04346 . 023Figure 5—figure supplement 2 . Evaluation of the compared network reconstruction methods . ( A ) Accuracy metrics ( y-axes , according to the definition in Figure 5—figure supplement 1B ) for three sub-network models ( Figure 5—figure supplements 1A , 4A; color coded ) with noise level 0 . 5 and 7 traits , shown across different method parameters ( x-axes ) . The plots indicate that for Wang and van Eeuwijk , 2014 ( top ) , Hageman et al . , 2011 ( middle ) , and Neto et al . , 2010 ( bottom ) , the best performances are attained with τ = 0 . 1 , α = 0 . 05 , and α = 0 . 25 , respectively . These method parameters were used throughout the study . ( B ) Shown is the running time ( y-axis , log scaled ) across compared methods ( color coded ) and different numbers of traits ( x-axis ) . The running time refers to construction of 100 sub-networks of type ‘Net-1’ ( as demonstrated in Figure 5A ) with noise level 0 . 5 . The reported running time are from a Linux machine with 2 . 6 GHz AMD Opteron 6238 processors . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 02310 . 7554/eLife . 04346 . 024Figure 5—figure supplement 3 . Relevant range of synthetic data parameters . Shown are the absolute correlation coefficients between transcripts and traits ( A ) , between transcripts and variants ( B ) , and between traits and variants ( C ) for different datasets ( x-axis ) . Datasets ( from left to right ) : all traits , transcripts and variants in mouse myeloid data ( Gerrits et al . , 2009 ) ( first column ) ; the components of module no . 2 ( second column ) ; synthetic sub-networks with noise level = 0 . 5 ( third column ) ; and synthetic sub-networks with noise level = 2 ( fourth column ) . In each box the central mark is the median; edges are the 25th and 75th percentiles; whiskers extend to the most extreme data points not considered outliers; and outliers are plotted individually . Notably , synthetic sub-networks with high noise level = 2 resemble the background distribution in mouse data . In contrast , synthetic networks with lower noise levels ( noise level = 0 . 5 ) show a signal that is higher than the background distribution , similarly to the inter-relations within the desired GEMOT modules . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 02410 . 7554/eLife . 04346 . 025Figure 5—figure supplement 4 . Accuracy assessment in additional sub-networks . ( A ) Detailed illustration of two synthetic networks , ‘Net-2’ and ‘Net-3’ . ( B ) Accuracy metrics ( y-axes , according to the error matrix in Figure 5—figure supplement 1B ) are shown over different numbers of traits ( x-axes ) . Results are shown for Net-2 with noise level = 0 . 5 ( left ) , for Net-3 with noise level = 0 . 5 ( middle ) , and for Net-2 with noise level = 2 ( right ) . Plots depict alternative methods of network construction ( color coded ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 02510 . 7554/eLife . 04346 . 026Figure 5—figure supplement 5 . Accuracy assessment in a large biological network . Performance is evaluated for the ability to group traits based on their co-mapping to the same variant ( denoted naive grouping , A ) and for the ability to identify groups of traits sharing the same causal transcripts ( grouping-by-causality , B ) . ( A ) Naive grouping evaluation . Left: An error matrix , where each cell i , j shows the predicted class in row i for a trait pt is TP , TN , FP or FN while the actual class of trait pt is in column j . The actual class for a trait pt is either pt ∉ P ( s ) or pt ∉ P ( s ) , where P ( s ) is the collection of all traits in a sub-network s . The predicted classes are defined similarly as pt ∉ Q ( s ) and pt ∈ Q ( s ) , where Q ( s ) is the predicted group of traits that attain the largest intersection with the traits in a sub-network s . Right: The error matrix ( from left panel ) is used to calculate the accuracy metric ( y-axis ) across different numbers of traits ( x-axis ) . Plots depict alternative grouping methods ( color coded ) . The lower performance of GEMOT is expected , since it is designed for grouping by causality , rather than a naive grouping of traits . ( B ) grouping-by-causality evaluation . An error matrix ( left ) and an accuracy plot that utilizes this error matrix ( middle ) , are presented as in A but for the case of grouping-by-causality . PC ( s ) is the subset of traits that share the same causal transcripts within a sub-network s , as exemplified in Figure 5—figure supplement 1A . Right: The error matrix ( from left panel ) is used to calculate the FDR metric ( y-axis ) across different numbers of traits ( color coded ) and alternative grouping methods . The plots clearly show that GEMOT's ability to group by causal relationships has an advantage over existing methods . Metrics: Accuracy = ( TPR + TNR ) /2 , FDR = FPa/ ( FPa + TP ) ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04346 . 026 Performance was evaluated using an accuracy metric , which reflects the ability of a method to discern the correct subset of traits sharing the same transcripts ( e . g . , traits p1 , . . . , pk but not pk + 1 and pk + 2 in Figure 5A and Figure 5—figure supplement 1B for details ) . Figure 5B presents the accuracy for synthetic data collections of varying levels of noise ( using the sub-networks from Figure 5A with 5 or 7 traits ) . GEMOT displayed the best accuracy for noise levels ranging between 0 . 25 and 1 , with lower accuracy for higher noise levels . Analysis of various network properties in both mouse and synthetic data shows that sub-networks with a noise level that do not exceed 1 are more likely to represent real biological modules ( Figure 5—figure supplement 3 ) . Unlike the compared methods , GEMOT's accuracy remained high with an increasing number of traits ( Figure 5C ) ; similar results were obtained for alternative network structures ( Figure 5—figure supplement 4 ) . Overall , in our simulation , GEMOT outperformed the compared algorithms in handling a growing number of traits and in identifying the correct groups of traits when using biologically-relevant parameters . These results do not rule out the possibility that for other tissues , conditions or organisms , utilizing the alternative methods as part of the third stage of the GEMOT algorithm may enhance its performance . We next aimed to characterize GEMOT's utility for a large biological network that included groups of traits that share the same causal transcripts . Accordingly , each synthetic network included 100 traits , 200 transcripts and 100 variants , featuring five co-mapped sub-networks . A singe data collection consists of 100 networks , each containing five co-mapped sub-networks that carry the same number of traits ( ‘Materials and methods’ ) . We compared GEMOT to two alternative trait-grouping methods: the trait-based iterative clique enumeration ( ICE ) approach ( Shi et al . , 2010 ) and the gene-based INVAMOD approach ( Gat-Viks et al . , 2013 ) ( ‘Materials and methods’ ) . Notably , network construction methods ( e . g . , Neto et al . , 2010; Hageman et al . , 2011; Wang and van Eeuwijk , 2014 ) could not be compared owing to an unrealistic running time in the case of large networks . The analysis suggested that although all compared methods successfully discern all traits in a sub-network ( Figure 5—figure supplement 5A ) , GEMOT attains higher accuracy in discerning those traits that share the same driver transcripts ( Figure 5—figure supplement 5B ) . Notably , the GEMOT algorithm is tailored for identification of causative relationships , unlike the compared methods , explaining why GEMOT succeeded in discriminating the correct subsets of co-regulated traits . We set out to identify the molecular and genetic mechanisms underlying connections among groups of traits . To that end , we combined module identification ( Gat-Viks et al . , 2010 ) with causality testing ( Neto et al . , 2013 ) in a unified pipeline that relies on the definition of linked relationships so that candidate modules can be filtered out prior to the validation stage . Our results in mice highlighted three types of high-order organization of traits . ( i ) Groups of tightly related traits that share the same transcripts mechanisms ( modules 1 , 2 , 6 , 7 , 8 , e . g . , Figure 3 ) . ( ii ) Groups of distinct traits that share the same transcripts mechanism , but not necessarily high correlations among them ( modules 3 , 4 , 5 , e . g . , Figure 4 ) . ( iii ) Different groups commonly have overlapping traits , but typically differ in their underlying mechanisms ( Figure 2B ) . Our study emphasizes the need for methodologies for constructing causative models that underlie connections among traits . Whereas previous trait-grouping methods have used genetic or molecular data separately , and thus did not validate causative transcripts ( e . g . , Shi et al . , 2010; Gat-Viks et al . , 2013; as illustrated in Figure 1—figure supplement 1B , C ) , the GEMOT method aims at directly filtering and validating such relationships . Our simulations showed that GEMOT is superior to these methods in identifying trait groups that share the same underlying transcripts ( Figure 5—figure supplement 5B ) . Another strategy is to use network reconstruction methods to construct groups of related traits ( e . g . , Neto et al . , 2010; Hageman et al . , 2011; Wang and van Eeuwijk , 2014 ) . These methods can be applied in the case of either the complete biological network or the sub-networks within the tripartite modules . Whereas these methods are limited in their scalability and may be particularly inefficient when applied on a large number of components , our approach can be scaled to larger networks , but can construct the network only partially . For example , in comparing both running time and accuracy under increased network sizes we found GEMOT to be more scalable than the alternative network construction methods ( Figure 5C and Figure 5—figure supplements 2B , 4B ) . Our methodology opens the door to a variety of future research directions . One possibility is that GEMOT will be applied on a compendium of molecular data from multiple tissues . In such cases , GEMOT predictions can be used to simultaneously identify both the biological mechanism and its relevant tissue . Second , GEMOT is applicable to a variety of molecular data types in addition to gene-expression data . For example , its application in blood cytokines or plasma lipids ( Teslovich et al . , 2010 ) is expected to make it possible to identify molecular factors acting at the cell–cell communication level . Similarly , future extensions of GEMOT may provide the means to include environmental factors as part of the module . Third , monitoring of additional strains would allow discriminating between several alternative genomic intervals for the same module , which may arise due to linkage disequilibrium between chromosomally distinct loci . Fourth , characterization of GEMOT modules that share a similar collection of traits but have different genomic intervals may reflect gene–gene interactions that lead to connections among traits . Finally , GEMOT can potentially be further improved by the construction of internal causative relationships within the transcripts and the traits layers . For example , some drivers may control other drivers , which in turn affect a collection of traits ( as exemplified in the case of Klf7 in Figure 3 ) . It should be noted , however , that GEMOT cannot distinguish cases of multiple drivers that are part of the same regulatory circuit from cases of multiple drivers that act through several distinct circuits . Rather , its predictions provide biological or clinical hypotheses for additional experimental investigations . Overall , our approach paves the way to the simultaneous study of several mechanistic layers underlying connections among traits , providing a multilayered view of phenotypic connections . Because the GEMOT methodology is a general one and can be applied to the study of other taxa , this approach may facilitate our understanding of the molecular mechanisms underlying human disease . GEMOT is designed to identify three-layer modules in which driver components translate between a single genomic interval and a collection of traits . As input , GEMOT takes three types of objects: ( 1 ) a collection of traits across a population of individuals; ( 2 ) genotyped genetic variants for these individuals; and ( 3 ) high-throughput gene expression data across the same population . Our algorithm incorporates three stages ( Figure 1A ) . The first stage constructs candidate bipartite modules consisting of a group of traits and a genomic interval . In the second stage , candidate transcripts are added to each module from the previous stage , thus forming tripartite modules . The final stage validates the actual drivers and refines the modules accordingly . The GEMOT code is available at http://csgi . tau . ac . il/gemot/ . In the following we first define the construction of a bipartite graph and then explain the identification of bipartite clusters within this graph as previously described ( Gat-Viks et al . , 2010 ) . We define a bipartite graph whose two parts correspond to genetic variants and traits , and in which the edges reflect the potential of a variant and a trait to have significant linked relationships ( Figure 1C ) . Edge weights are calculated as follows ( Figure 1B ) : First , for each pair of a genetic variant and a transcript , we evaluate the genetic association between the expression of the transcript and the candidate genetic variant . This yields a ‘variant–transcript association score’ . In this study , for the case of homozygous recombinant inbred strains the association score is a −log t-test p value for the different gene-expression values between the strains carrying the two possible variant alleles . For other cases , such as an outbred population , other standard association scores can be applied ( Falconer and Mackay , 1996 ) . Secondly , for each pair of a transcript and a trait , we calculate the absolute Pearson correlation coefficient across genetic backgrounds . We term this score the ‘transcript–trait correlation score’ . Finally , for each genetic variant and each trait we compare the distribution of transcript–trait correlations in high and low transcript–variant association scores ( a statistical t-test ) . We assign such a t-test p value to five different transcript–variant association cutoffs ( the five cutoffs partition the association range into 6 equally sized groups ) and record the top −log t-test p value across these five cutoffs . We refer to the recorded −log p values as ‘link potentials’ and use them as edge weights in the bipartite graph . Within this graph we use the ReL software package ( Gat-Viks et al . , 2010 ) to identify significant biclusters ( Figure 1D ) . Briefly , the ReL algorithm starts with a set of seed clusters consisting of one trait and one variant whose link potential exceeds a certain initialization cutoff , cs . A trait or a variant can be included in a cluster if and only if its average link potential exceeds an improvement cutoff ci ( here , cs = 180 , ci = 90 ) . Each bipartite cluster is subject to iterative improvements by addition or removal of traits and variants based on this cutoff . We refer to the bipartite clusters as ‘bipartite modules’ and further improve them in the following stages . Notably , ReL provides the same results when applied with or without Boneferroni correction for the gene–variant association −log p values and link potentials scores , since the construction of the biclusters is robust to an additive rescaling of these scores . In this stage a rough list of candidate transcripts is constructed for each module ( Figure 1E ) . To this end , for each transcript in a given module we rank its correlations and associations with all traits and variants in the input dataset and record its ranks of associations and correlations within the module . We next compare the distribution of recorded ranks in this module with the distribution of all ranks . The two distributions are compared using a Kolmogorov–Smirnov ( K-S ) test , a nonparametric test that may be used to compared two samples . We refer to the K-S p value for a transcript in a module as a ‘transcript link score’ . Only transcripts with significant link scores are added to their module . Such transcripts are called ‘candidate transcripts’ and the resulting extended modules are referred to as ‘tripartite modules’ ( Figure 1F ) . In the following we first define the causality test and then describe the procedures for identifying driver transcripts and for module refinement . All mice data was taken from a previously produced body of work . We applied our analysis to data obtained from homozygous BXD recombinant inbred mouse strains ( Peirce et al . , 2004 ) generated by crossing C57BL/6J and DBA/2J inbred strains for many generations . Microarray expression data in myeloid cells across 24 BXD strains have been measured ( Gerrits et al . , 2009 ) . To identify high-quality candidates we selected 5786 genes whose variation in expression across BXD strains , based on average intensities of the genes , was higher than expected . Expected variance values were calculated using a sliding window along the genes' average intensities . Gene-expression values were log10-transformed and normalized by Z-score normalization . All 2885 traits and 3796 genetic variants across BXD strains were downloaded from the WebQTL dataset ( Wang et al . , 2003 ) . Trait values were normalized by Z-score normalization . Given the strains in the gene expression and trait datasets , we restricted our analysis to 1738 traits that had at least 15 strains in common . Other compared cell types or tissues ( Gatti et al . , 2007; Geisert et al . , 2009; Gerrits et al . , 2009; Alberts et al . , 2011; Mozhui et al . , 2012 ) were similarly preprocessed . Supplementary file 1C records the particular representative variants that were used for causality testing in each predicted GEMOT module . To assess the corresponding false discovery rates , we generated negative controls based on a permutation test in which the transcript levels of each transcript were randomly shuffled and the GEMOT modules were recomputed ( a process that was repeated 100 times ) . In each repeat , a variety of statistics ( such as the number of identified modules ) were recorded . The permutation-based ‘false discovery rate’ ( FDR ) is the ratio of the averaged number of statistics that were declared significant using the permuted data to the number of statistics that were declared significant using the original ( non-permuted ) data . In this study , GEMOT was applied using transcript link score cutoff = 10−95 for identifying candidate transcripts ( stage II , permutation-based FDR < 0 . 09 ) and transcript causality p value cutoff = 0 . 005 ( stage III , permutation-based FDR < 6 × 10−5 ) . To investigate the performance of the causality score we simulated triplets of objects , each consisting of a variant v , a transcript g and a trait p . In all such triplets we assume 100 homozygous individuals . The genotyping of each variant was generated by sampling a vector of values 0 and 1 from a binomial distribution ( with p = 0 . 5 ) . Based on these genotyping values , the values of g and p were generated according to the following five different models ( denoted M1–M5 ) , as depicted in Figure 1—figure supplement 2A: M1 , a causative model v → g → p; M2 , an independent model v → g , v → p; M3 , a reactive model v → p → g; M4 , v → g → p and v → p; and M5 , v → p → g and v → g . Each arrow in these five models was simulated as a linear expression with a normally distributed error term . For example , based on model M1 , data were generated as g = α⋅v + ε , p = λ⋅g + ε , ε ∼N ( 0 , σ2 ) ; similarly , model M4 was generated as g = α⋅v + ε , p = α⋅v + λ⋅g + ε , ε ∼N ( 0 , σ2 ) . A single ‘synthetic collection’ consisted of 250 relationships from each of the five models , a total of 1250 samples . Results are displayed for many different collections , each generated using different combination of λ and σ values with α = 0 . 5 . GEMOT's causality p values were compared to those obtained by two alternative methods: QTLHot ( Neto et al . , 2013 ) and an AIC-based method ( Lee et al . , 2009 ) . For a given significance threshold we evaluated the ability to identify causal relationships using true positive ( TP ) , true negative ( TN ) , false positive ( FP ) and false negative ( FN ) counts , which were defined according to the broad-sense definition of causality ( Figure 1—figure supplement 2B ) . The area under the receiver operating characteristic ( ROC ) curve ( the AUC ) was calculated accordingly , where the higher the AUC the better the method . In addition , a balanced false discovery rate ( FDR ) can be used as a criterion for comparisons of methods , computed as FDR = FPa/ ( FPa + TP ) where FPa accounts for imbalanced data by dividing FP by the ratio between the negative and positive synthetic datasets , calculated as FPa = FP/π0 , π0 = ( FP + TN ) / ( TP + FN ) . The method with lowest FDR is regarded as the best method ( when all methods use the same p value cutoff ) . Notably , GEMOT attains similar performance to that of the compared methods for models M1–M3 ( Figure 1—figure supplement 2C ) , but outperforms the existing methods when adding synthetic M4 and M5 samples ( Figure 1—figure supplement 2D ) . The synthetic data analysis is focused on two simulations with increasing complexity of the input network: ( i ) sub-network analysis , in which the input is a sub-network where all components are co-mapped to the same variant; and ( ii ) network analysis , in which the input is a large network comprising several co-mapped sub-networks .	Many individuals who have diabetes also have other diseases that affect the heart and blood vessels . It is not uncommon for human diseases to occur together like this; and understanding the relationships between diseases and other traits can make it easier to diagnose conditions . Furthermore , it can also help researchers develop treatments that are more precisely targeted to each condition and cause fewer side effects . Two conditions or traits tend to occur together if they are caused by mutations in the same gene or genes; or if they involve processes within cells that share the same proteins and other molecules . However , in most cases the genes and molecular mechanisms involved are not yet known so it is more difficult to work out how the traits are connected . Computing techniques make it possible to assess the relationships between hundreds or thousands of traits at the same time . These high volume analyses can also allow scientists to identify less obvious relationships that might be missed in more traditional types of study . Here , Oren et al . created a new computer algorithm to identify related traits , their shared genetic basis , and the molecular mechanisms behind them . The algorithm is called GEMOT and uses a three-step approach to sift through a large amount of data . Oren et al . tested GEMOT using a database of 1738 documented traits—including diseases and behaviors—in laboratory mice . Oren et al . identified many clusters of traits in the mice and the underlying genetic and molecular mechanisms that link them . For example , they found that a mutation in a gene called Klf7 affected the expression of other genes that are involved in making new cells in the bone marrow . In turn , these changes influenced 17 different behaviors in the mice after they were injected with the painkiller morphine . In humans , the same genes that underlie behaviors related to morphine treatment have been linked to the survival rate of patients with a form of brain cancer . This suggests that—alongside providing pain-relief—morphine may influence how the tumor grows . The algorithm developed by Oren et al . can now be used to further explore the impact of the environment on the relationships between traits .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology" ]	2015	Linking traits based on their shared molecular mechanisms
Hebbian plasticity is thought to require glutamate signalling . We show this is not the case for hippocampal presynaptic long-term potentiation ( LTPpre ) , which is expressed as an increase in transmitter release probability ( Pr ) . We find that LTPpre can be induced by pairing pre- and postsynaptic spiking in the absence of glutamate signalling . LTPpre induction involves a non-canonical mechanism of retrograde nitric oxide signalling , which is triggered by Ca2+ influx from L-type voltage-gated Ca2+ channels , not postsynaptic NMDA receptors ( NMDARs ) , and does not require glutamate release . When glutamate release occurs , it decreases Pr by activating presynaptic NMDARs , and promotes presynaptic long-term depression . Net changes in Pr , therefore , depend on two opposing factors: ( 1 ) Hebbian activity , which increases Pr , and ( 2 ) glutamate release , which decreases Pr . Accordingly , release failures during Hebbian activity promote LTPpre induction . Our findings reveal a novel framework of presynaptic plasticity that radically differs from traditional models of postsynaptic plasticity . Learning and memory are thought to require synaptic plasticity , which refers to the capacity for synaptic connections in the brain to change with experience . The most frequently studied forms of synaptic plasticity are long-term potentiation ( LTP ) and long-term depression ( LTD ) , which respectively involve long-lasting increases and decreases in synaptic transmission . LTP and LTD can be expressed either postsynaptically ( LTPpost or LTDpost ) , as changes in AMPA receptor ( AMPAR ) number , or presynaptically ( LTPpre or LTDpre ) , as changes in glutamate release probability ( Pr ) ( Padamsey and Emptage , 2011; Bliss and Collingridge , 2013; Larkman and Jack , 1995; Lisman , 2003; Lisman and Raghavachari , 2006; Padamsey and Emptage , 2014 ) . Traditionally , postsynaptic NMDA receptor ( NMDAR ) activation is believed to be important for both pre- and postsynaptic forms of plasticity ( Bliss and Collingridge , 2013; Lüscher and Malenka , 2012 ) . Postsynaptic changes , in particular , have been causally and convincingly linked to NMDAR-dependent Ca2+ influx , which , via the activation of postsynaptic Ca2+-sensitive kinases and phosphatases , triggers changes in the number of synaptic AMPARs ( Lüscher and Malenka , 2012 ) . The link between postsynaptic NMDAR activation and presynaptic plasticity , however , is less clear . In the case of LTPpre induction , it is traditionally thought that Ca2+ influx through postsynaptic NMDARs triggers the synthesis and release of a retrograde signal , most likely nitric oxide ( NO ) , which in turn triggers increases in Pr ( Padamsey and Emptage , 2014; Garthwaite and Boulton , 1995 ) ( though other forms of presynaptic plasticity exist [Yang and Calakos , 2013; Castillo , 2012] ) . Consistent with this , several studies have demonstrated that LTPpre induction is impaired by the blockade of NMDAR or NO signalling ( Ryan et al . , 1996; Ratnayaka et al . , 2012; Emptage et al . , 2003; Bliss and Collingridge , 2013; Enoki et al . , 2009; Nikonenko et al . , 2003; Stanton et al . , 2005; Padamsey and Emptage , 2014; Johnstone and Raymond , 2011 ) . However , some groups have found that presynaptic enhancement can be induced in the presence of NMDAR antagonists ( Blundon and Zakharenko , 2008; Zakharenko et al . , 2003; Bayazitov et al . , 2007; Zakharenko et al . , 2001; Stricker et al . , 1999 ) ( but see [Grover and Yan , 1999a; Grover , 1998] ) . Under these conditions , presynaptic plasticity relies on L-type voltage-gated Ca2+ channel ( L-VGCC ) activation ( Blundon and Zakharenko , 2008; Zakharenko et al . , 2003; Bayazitov et al . , 2007; Zakharenko et al . , 2001 ) , but may still depend on NO signalling ( Padamsey and Emptage , 2014; Pigott and Garthwaite , 2016 ) . These findings suggest that LTPpre may require neither the activation of postsynaptic NMDARs nor NMDAR-dependent NO synthesis; nonetheless , results across studies are largely inconsistent ( Padamsey and Emptage , 2014 ) , and the exact mechanism and retrograde signal underlying LTPpre induction remains unclear . The role of glutamate signalling in presynaptic plasticity is also unclear . Glutamate release is of course necessary to drive the postsynaptic depolarization required for the induction of both LTPpre and LTPpost . However , this does not explain why any given synapse needs to release glutamate in order to be potentiated , since depolarization triggered by one synapse can affect another , either directly via electrotonic spread , or indirectly via the actions of dendritic or somatic spikes . The necessity for site-specific glutamate release in LTP induction , at least in the case of LTPpost , is instead imposed by the strict requirement of postsynaptic NMDAR-mediated Ca2+ influx for potentiation ( Bliss and Collingridge , 2013; Lüscher and Malenka , 2012 ) . However , that NMDARs may not to be necessary for the induction of LTPpre ( Blundon and Zakharenko , 2008; Padamsey and Emptage , 2014 ) suggests that the role of synapse-specific glutamate release in presynaptic plasticity may be different . Indeed , a common finding across a number of studies is that high Pr synapses are more likely to show LTDpre , whereas low Pr synapses are more likely to show LTPpre ( Ryan et al . , 1996; Slutsky et al . , 2004; Larkman et al . , 1992; Hardingham et al . , 2007; Sáez and Friedlander , 2009 ) . Moreover , glutamate release can induce LTDpre by acting on presynaptic NMDARs ( McGuinness et al . , 2010; Rodríguez-Moreno et al . , 2013 ) , or metabotropic glutamate receptors ( mGluRs ) in the case of younger tissue ( Zakharenko et al . , 2002 ) . Thus , enhanced glutamate release at a presynaptic terminal , unlike at a dendritic spine ( Lüscher and Malenka , 2012; Harvey and Svoboda , 2007; Matsuzaki et al . , 2004; Makino and Malinow , 2009 ) , may not necessarily result in enhanced potentiation , but instead promote depression . Several studies have also demonstrated that presynaptic terminals initially releasing little or no glutamate are reliably potentiated following tetanic stimulation ( Ryan et al . , 1996; Emptage et al . , 2003; Slutsky et al . , 2004; Larkman et al . , 1992; Hardingham et al . , 2007; Sáez and Friedlander , 2009; McGuinness et al . , 2010; Enoki et al . , 2009 ) . How low Pr synapses , including those that are putatively silent , can undergo activity-dependent potentiation raises questions as to the necessity of synapse-specific glutamate release in presynaptic plasticity . Here we re-examined the mechanisms underlying activity-dependent presynaptic changes at CA3-CA1 hippocampal synapses , with a particular focus on understanding the role of glutamate in presynaptic plasticity . We find that , contrary to current thinking , Hebbian activity , via L-VGCC-triggered NO signalling , is sufficient to induce LTPpre without the need for synapse-specific signalling by glutamate . When glutamate release occurs , it inhibits LTPpre and instead promotes LTDpre by activating presynaptic NMDARs . Thus , for presynaptic potentiation to occur , a presynaptic neuron must not only fire together with its postsynaptic partner , but it must also fail to release glutamate . Our findings reveal a novel set of rules and mechanisms governing presynaptic plasticity that are distinct from those associated with traditional , postsynaptic models of plasticity . We started by examining how manipulating glutamatergic signalling at synapses would affect activity-driven changes in presynaptic function . We recorded excitatory postsynaptic potentials ( EPSPs ) in CA1 neurons in cultured hippocampal slices . Cells were recorded using patch electrodes ( 4–8 MΩ ) and EPSPs were evoked by Schaffer-collateral stimulation . Baseline EPSP recordings were kept short ( 5 min ) to minimize dialysis as we found that longer baseline recordings prevented LTPpre induction ( see Materials and methods ) . For LTP induction we used a pairing protocol , in which individual presynaptic stimuli were causally paired with postsynaptic spiking , 60 times at 5 Hz . Because LTPpre is preferentially induced under conditions of strong postsynaptic depolarization ( Padamsey and Emptage , 2014; Blundon and Zakharenko , 2008; Zakharenko et al . , 2003; Bayazitov et al . , 2007; Zakharenko et al . , 2001; Stricker et al . , 1999 ) , we paired presynaptic stimuli with a current injection of sufficient amplitude to generate 3–6 postsynaptic spikes over a 50–60 ms time course . Spikes tended to broaden over the time course of injection , and the resulting waveform resembled a complex spike ( Figure 1A ) , which is known to efficiently drive LTP in vitro ( Thomas et al . , 1998; Remy and Spruston , 2007; Golding et al . , 2002; Hardie and Spruston , 2009 ) , and has been recorded in the hippocampus in vivo ( Ranck , 1973; Grienberger et al . , 2014 ) . We found that this pairing protocol produced robust and reliable LTP ( fold ΔEPSPslope: 1 . 88 ± 0 . 24; n = 6 cells; vs 1 . 0: p<0 . 05; Figure 1B , C ) , which had a presynaptic component of expression , as assessed by a decrease in the paired pulse ratio ( PPR ) ( ΔPPR: −0 . 39 ± 0 . 15; n = 6 cells; vs 0: p<0 . 05; Figure 1D ) . Changes in PPR were evident across a range of intervals; however , we chose to measure PPR at an interval of 70 ms , at which plasticity-induced changes tended to be maximal ( Figure 1—figure supplement 1A ) . We then examined the effects of elevating glutamate release during LTPpre induction . One physiological means of transiently increasing glutamate release probability ( Pr ) is to elevate the frequency of presynaptic activity ( Dobrunz and Stevens , 1997; Dobrunz and Stevens , 1997 ) . We therefore repeated our LTP experiments , but during induction , in the place of single presynaptic pulses , we used short , high frequency bursts of presynaptic stimuli to increase Pr ( Figure 1—figure supplement 2 ) . The burst consisted of two pulses , delivered 5 ms apart , and resembled high-frequency bursting activity recorded in CA3 neurons in vivo ( Kowalski et al . , 2016 ) . Remarkably , we found that pairing burst stimulation with postsynaptic depolarization produced significantly less LTP compared to single pulse pairings ( fold ΔEPSPslope: 1 . 36 ± 0 . 13; n = 6 cells; vs . single pulse pairings: p<0 . 05 ) , and was accompanied by no significant changes in PPR ( ΔPPR: 0 . 00 ± 0 . 04; n = 6 cells; vs 0: p=0 . 84; vs . single pulse pairings: p<0 . 01; Figure 1D ) . These findings suggest that high frequency presynaptic activity inhibits the induction of LTPpre . To examine the effects of presynaptic stimulation alone , we repeated our experiments , but during LTP induction we omitted postsynaptic depolarization ( unpaired stimulation ) . Under these conditions , when single presynaptic stimuli were delivered , we observed no significant change in the EPSP ( fold ΔEPSPslope: 0 . 93 ± 0 . 10; n = 5 cells; vs 1 . 0: p=0 . 62; Figure 1E , F ) or PPR ( fold ΔPPR: 0 . 01 ± 0 . 09; n = 5 cells; vs 0: p=0 . 81; Figure 1F ) . However , when high frequency bursts were delivered during induction , we observed a robust decrease in the EPSP ( fold ΔEPSPslope: 0 . 42 ± 0 . 08; n = 8 cells; vs . single pairing: p<0 . 01; Figure 1E , F ) and an increase in PPR ( fold ΔPPR: 0 . 36 ± 0 . 04; n = 8 cells; vs . single pulse pairing: p<0 . 01; Figure 1G; Figure 1—figure supplement 1B ) , suggesting that we had induced LTD with a presynaptic component of expression . Collectively , these findings demonstrate that high frequency presynaptic stimulation not only inhibits the induction of LTPpre , but also promotes the induction of LTDpre . We next tested whether the effects of high frequency presynaptic stimulation on LTPpre and LTDpre were in fact due to synapses releasing glutamate more reliably , as opposed to other effects , such as an increase in Ca2+ influx at the presynaptic terminal . To do so , we used glutamate uncaging instead of high-frequency presynaptic stimulation to artificially elevate Pr at synapses during LTP induction . Glutamate uncaging was restricted to single synapses , and activity-dependent changes in presynaptic function ( i . e . Pr ) were assessed by imaging postsynaptic Ca2+ transients ( Emptage et al . , 2003; Emptage et al . , 1999 ) . This technique relies on the fact that at most CA3-CA1 synapses single quanta of glutamate , through AMPAR-mediated depolarization , generate sufficient Ca2+ influx from NMDAR and voltage-gated Ca2+ channels ( VGCCs ) to be detected by Ca2+-sensitive dyes ( Padamsey and Emptage , 2011; Grunditz et al . , 2008; Emptage et al . , 1999 ) . Consequently , the proportion of trials in which single presynaptic stimuli generate postsynaptic Ca2+ transients can be used to calculate Pr at single synapses ( Emptage et al . , 1999 ) . Notably , estimates of Pr measured at resting membrane potential are resilient to large perturbations of postsynaptic Ca2+ influx ( Figure 2—figure supplement 1; also see Figure 6—figure supplement 1A–C ) . CA1 pyramidal neurons were filled with the Ca2+-sensitive dye Oregon Green BAPTA-1 and a fluorescently-coated glass electrode was used to stimulate Schaffer-collaterals in the vicinity of an imaged dendrite ( Figure 2A ) . Dendritic spines were sequentially scanned in order to identify those that were responsive to stimulation . To increase the likelihood of visually identifying responsive synapses , especially those with low basal release probabilities , we delivered two presynaptic stimuli , 70 ms apart , to transiently increase Pr ( Figure 2B ) . When a synapse was found that responded to stimulation , it always responded in an all-or-none manner , with Ca2+ transients largely restricted to the spine head . As expected , Ca2+ transients were more likely to be elicited by the second of the two presynaptic stimuli because of the effects of short-term facilitation . Pr was calculated as the proportion of trials in which the first of the two presynaptic stimuli generated a fluorescent increase in the spine head; the second of the two presynaptic stimuli was ignored . Because of the additional time required to measure Pr , most of our imaging experiments were done in the absence of electrophysiological recordings in cells bolus-loaded with Ca2+ sensitive dye; cells were only transiently patched for plasticity induction ( see Materials and methods ) . Consistent with our electrophysiological results , pairing single presynaptic stimuli with postsynaptic complex spikes ( 60 pairing at 5 Hz ) evoked an increase in Pr ( ΔPr: 0 . 19 ± 0 . 03; n = 14 spines; vs . 0: p<0 . 01; Figure 2B , D , F ) . We then repeated the experiment but this time , during LTP induction , each presynaptic stimulus was coupled with photolysis of caged glutamate at the synapse , regardless of whether the synapse released glutamate or not , in order to artificially elevate Pr during stimulation . We adjusted the laser power to ensure that photolysis mimicked the fluorescent changes elicited by uniquantal glutamate release evoked by single presynaptic stimuli ( ΔF/F; photolysis vs . stimulation: 0 . 46 ± 0 . 07 vs . 0 . 55 ± 0 . 09; n = 15 spines; p=0 . 23; Figure 2A ) . Remarkably , under these conditions , increases in Pr at the target synapse were effectively abolished ( ΔPr: −0 . 02 ± 0 . 02; n = 15 spines; photolysis vs . control: p<0 . 01; Figure 2B , D , F ) . This demonstrates that , consistent with our hypothesis , transiently elevating glutamate signalling at synapses inhibited the induction of LTPpre . In eight of the synapses imaged under these conditions , LTP induction was repeated for a second time , but in the absence of caged glutamate during photolysis; in these experiments , the expected increase in Pr was observed ( ΔPr: 0 . 23 ± 0 . 02; n = 8 spines; vs . control: p=0 . 48; Figure 2Biii , post-photolysis control in Figure 2D ) . Increases in Pr were also observed in a subset of control experiments , in which LTP induction was conducted in the presence of caged glutamate , but in the absence of photolytic laser exposure ( ΔPr: 0 . 25 ± 0 . 03; n = 8 spines; vs . control: p=0 . 15 ) . These results suggest that the inhibitory effect of photolysis on Pr was due to glutamate release , as opposed to non-specific effects of uncaging . We also examined the effects of glutamate photolysis delivered in the absence of postsynaptic depolarization ( unpaired stimulation ) . Delivery of 60 presynaptic stimuli at 5 Hz , consistent with our electrophysiological recordings , produced no changes in Pr at the majority of synapses imaged ( Figure 2C , E , F ) . We did , however , notice that synapses with initially high release probabilities ( Pr >0 . 5 ) , showed a modest decrease in Pr following unpaired stimulation ( Figure 2E ) ; this decrease was not likely to be detected by electrophysiological recordings because high Pr synapses comprise an estimated <10% of synapses in our preparation ( Ward et al . , 2006 ) . Remarkably , when we coupled each presynaptic stimulus with glutamate photolysis , we now observed decreases in Pr at all imaged synapses , regardless of their initial Pr ( ΔPr photolysis vs . control: −0 . 33 ± 0 . 08 vs . −0 . 12 ± 0 . 06; n = 9 , 10 spines p<0 . 05; Figure 2C , E , F ) . These findings suggest that elevated glutamate release decreases Pr , and does so regardless of the level of postsynaptic depolarization that accompanies presynaptic activity; Pr changes induced by paired or unpaired stimulation were always more negative compared to controls when glutamate signalling was augmented . Given that transiently elevating glutamate release probability , either by presynaptic bursts or glutamate photolysis , inhibited the induction of LTPpre , we asked if glutamate signalling was required at all for driving increases in Pr during paired stimulation , as traditionally believed . Physiologically , glutamate is clearly necessary for driving the postsynaptic spiking required for LTP , and all major classes of glutamate receptors including: AMPARs , Kainate receptors ( KARs ) , NMDARs , and mGluRs can contribute to membrane depolarization ( Grienberger et al . , 2014; Schiller and Schiller , 2001; Grover and Yan , 1999b; Chemin et al . , 2003 ) . If , however , membrane depolarization is the only function of glutamate in LTPpre induction , then a presynaptic terminal could in principle be potentiated even if it failed to release glutamate , provided that its activity coincided with strong postsynaptic depolarization , as driven by glutamate release at other co-active synapses . If this is the case , we reasoned that we should be able to experimentally trigger LTPpre in a full glutamate receptor blockade provided that , during presynaptic stimulation , we supplemented the depolarizing effects of glutamate with somatic current injection . If , however , glutamate is additionally required for some form of synapse-specific signalling , as in the case of LTPpost induction , then the induction of LTPpre should not be possible in full glutamate receptor blockade no matter how much we depolarize the neuron during presynaptic stimulation . To test this possibility we attempted to induce LTPpre at CA3-CA1 synapses with all known glutamate receptors ( AMPARs , KARs , NMDARs , and mGluRs ) pharmacologically inhibited ( 10 µM NBQX , 100 µM D-AP5 , 0 . 5 mM R , S-MCPG , 100 µM LY341495 ) ; we used AP5 instead of MK-801 in order to block both ionotropic and metabotropic effects associated with NMDAR activation ( Nabavi et al . , 2013 ) . Given the additional time required for these experiments , we recorded from CA1 neurons using high-resistance patch electrodes ( 18–25 MΩ ) to limit the effects of postsynaptic dialysis . Following pharmacological abolishment of the EPSP , we delivered paired stimulation as before , during which strong postsynaptic depolarization again took the form of a complex spike induced by somatic current injection ( Figure 3A ) . The antagonist cocktail was then washed out in order to recover the EPSP . As expected with the use of high concentrations of NBQX ( Holbro et al . , 2010 ) , EPSP recovery was never complete and varied across experiments ( Figure 3B , C ) , and so it was necessary to compare the EPSP recorded from the pathway receiving paired stimulation to a second , independent control pathway recorded simultaneously ( Figure 3A , B ) . We found that paired stimulation induced a robust enhancement of the EPSP in the stimulated pathway relative to the control pathway ( fold ΔEPSPslope; paired vs . control: 1 . 12 ± 0 . 13 vs . 0 . 71 ± 0 . 12; n = 7 cells; p<0 . 05; Figure 3B , D ) ; this enhancement was not seen when pairings were anti-causal , with presynaptic stimuli following postsynaptic spiking ( Figure 3—figure supplement 1 ) . Causal pairings resulted in a 1 . 72 ± 0 . 21 fold potentiation , which we estimated by normalizing the fold change in the EPSP of the paired pathway to that of the control pathway . Notably , EPSP recovery of the control pathway was not significantly different from experiments in which drugs were applied in the absence of paired stimulation ( control vs . drugs-only: 0 . 71 ± 0 . 12 vs . 0 . 54 ± 0 . 11; n = 7 , 5 cells; p=0 . 59; Figure 3C , D ) , suggesting that LTP was restricted to only synapses that were active during the pairing . LTP was also associated with a significant decrease in PPR ( paired vs . control ΔPPR: −0 . 28 ± 0 . 06 vs . 0 . 03 ± 0 . 03; n = 6 cells; p<0 . 05; Figure 3E ) , that again , was only found in the paired pathway , suggesting that LTP induction was both presynaptic and site-specific . Similar site-specific enhancements in presynaptic function could be induced under full glutamate receptor blockade in acute hippocampal slices ( Figure 3—figure supplement 2 ) . In contrast to these findings , several studies have demonstrated that NMDAR blockade alone impairs LTP induction , even presynaptically ( Ryan et al . , 1996; Ratnayaka , 2012; Emptage et al . , 2003; Bliss and Collingridge , 2013; Enoki et al . , 2009; Nikonenko et al . , 2003; Stanton et al . , 2005; Padamsey and Emptage , 2014 ) . However , it is important to recognize that NMDARs , like all glutamate receptors , can contribute to postsynaptic depolarization . The NMDARs are particularly potent sources of depolarization , especially given their role in dendritic ( Schiller and Schiller , 2001; Losonczy and Magee , 2006 ) and somatic spiking ( Grienberger et al . , 2014 ) . Thus , it is possible that NMDAR blockade inhibits LTPpre expression by inhibiting postsynaptic depolarization . This is less likely to be an issue when strong postsynaptic depolarization is driven via somatic current injection , as in our experiment ( Figure 3 ) , than when depolarization is driven by presynaptic stimulation alone ( e . g . tetanic stimulation ) . To test this reasoning , we induced LTP using standard 100 Hz tetanic stimulation to drive postsynaptic spiking . This protocol produced robust potentiation of the recorded EPSP ( Figure 3—figure supplement 3A , D ) , and an increase in presynaptic efficacy ( Figure 3—figure supplement 3E ) . As in previous studies , NMDAR inhibition with AP5 abolished LTP induction , including its presynaptic component of expression ( Figure 3—figure supplement 3B , D ) . However , we found that if we augmented the levels of postsynaptic depolarization by current injection during tetanic stimulation , then LTP induction in AP5 was rescued , at least presynaptically ( Figure 3—figure supplement 3C , E ) . These findings suggest that the importance of postsynaptic NMDAR signalling in LTPpre induction is to provide a source of depolarization rather than any necessary source of synapse-specific signalling . These findings also underscore the importance of taking the level of postsynaptic depolarization into consideration when LTP is induced following the blockade of one or more glutamate receptor class . Collectively , our findings suggest that the role of glutamate signalling ( including postsynaptic NMDAR signalling ) in LTPpre induction is to drive postsynaptic depolarization . Physiologically , this means that a presynaptic terminal could in principle be potentiated if it fails to release glutamate , provided that its activity coincides with postsynaptic spiking , which could be triggered by glutamate release at other co-active synapses . We then returned to Ca2+ imaging to determine whether we could directly observe increases in Pr at single synapses associated with the induction of LTPpre in full glutamate receptor blockade ( Figure 4 ) . Because spine Ca2+ transients , in contrast to EPSPs , are resilient to partial AMPAR blockade ( Emptage et al . , 2003 ) , we found that they recovered well following drug washout , despite the difficulties associated with washing out high concentrations of NBQX ( Holbro et al . , 2010 ) . Consistent with electrophysiological findings , causal pairing of pre- and postsynaptic spiking in full glutamate receptor blockade produced robust and reliable increases in Pr ( ΔPr: 0 . 38 ± 0 . 07; n = 8 spines; vs 0: p<0 . 01; Figure 4A–C ) . No such changes were elicited by drug application in the absence of pairing ( ΔPr: 0 . 01 ± 0 . 02; n = 9 spines; vs . causal pairing: p<0 . 01 ) , or by either presynaptic stimulation alone ( ΔPr: −0 . 03 ± 0 . 03; n = 8 spines; vs . causal pairing: p<0 . 001 ) , or postsynaptic stimulation alone ( ΔPr: −0 . 00 ± 0 . 03; n = 8 spines; vs . causal pairing: p<0 . 001 ) , or when postsynaptic spiking preceded , rather than followed , presynaptic stimulation during pairing ( ΔPr: −0 . 02 ± 0 . 05; n = 8 spines; vs . causal pairing: p<0 . 001 ) ( Figure 4B , C ) . The induction of LTPpre in the absence of glutamatergic signalling was therefore Hebbian , requiring presynaptic activity to be causally paired with postsynaptic spiking . We next investigated the mechanism by which paired stimulation could trigger increases in Pr in the absence of glutamatergic signalling . The requirement for postsynaptic depolarization in the induction of LTPpre suggests a need for a diffusible retrograde messenger . One promising , albeit still controversial , retrograde signal implicated in LTPpre induction is nitric oxide ( NO ) ( for review see [Padamsey and Emptage , 2014] ) . Although NO synthesis has classically been associated with the activation of postsynaptic NMDARs ( Garthwaite and Boulton , 1995 ) , there is some suggestion that Ca2+ influx from L-type voltage-gated Ca2+ channels ( L-VGCCs ) , which have previously been implicated in LTPpre ( Bayazitov et al . , 2007; Zakharenko et al . , 2001 ) , could trigger NO production ( Pigott and Garthwaite , 2016; Sattler et al . , 1999; Stanika et al . , 2012 ) ; though definitive proof of a causal link between L-VGCC activation and NO synthesis at Schaffer-collateral synapses is lacking . We reasoned that if NO synthesis in CA1 neuronal dendrites can be triggered by L-VGCC activation , then NO production could occur in a manner dependent on postsynaptic depolarization , but independent of synapse-specific glutamatergic signalling . To test this , we first asked whether LTPpre , induced in glutamate receptor blockade , was dependent on L-VGCC activation and NO signalling . In keeping with our hypothesis , we found that pairing-induced increases in Pr ( ΔPr: 0 . 34 ± 0 . 04; n = 10 spines; p<0 . 01 ) were reliably abolished by bath application of the L-VGCC antagonist nitrendipine ( 20 µM ) ( ΔPr: −0 . 03 ± 0 . 04; n = 8 spines; vs . blockade: p<0 . 001 ) and by the NO scavenger carboxy-PTIO ( cPTIO ) , either bath applied ( 50–100 µM ) ( ΔPr: −0 . 01 ± 0 . 04; n = 8 spines; vs . blockade: p<0 . 01 ) or injected into the postsynaptic neuron ( ΔPr: −0 . 04 ± 0 . 06; n = 8 spines; vs . blockade: p<0 . 001 ) ( Figure 5A ) . We confirmed our findings in acute slices , and found that nitrendipine and cPTIO blocked presynaptic enhancements induced under glutamate receptor blockade ( Figure 5—figure supplement 1 ) , suggesting that , as in cultured slices , presynaptic efficacy in acute slices was similarly regulated by L-VGCC and NO signalling . We then examined whether NO production could be driven by postsynaptic depolarization in a L-VGCC-dependent manner . We transiently patched onto CA1 neurons in order to load them with the conventionally used NO-sensitive dye , DAF-FM ( 250 µM bolus-loaded ) , and then measured fluorescent changes in the apical dendrites prior to and following postsynaptic depolarization in glutamate receptor blockade . Given the poor signal-to-noise ratio associated with DAF-FM imaging , we drove strong postsynaptic depolarization by elevating extracellular K+ to 45 mM , as previously described ( Sattler et al . , 1999; Stanika et al . , 2012 ) . Under these conditions , we observed increases in fluorescence in neuronal dendrites ( Figure 5B , C ) . These increases were dependent on NO synthesis as they could be prevented by postsynaptic injection of cPTIO ( ΔF/F; control vs . cPTIO: 0 . 38 ± 0 . 04 vs . −0 . 03 ± 0 . 05; n = 5 cells/condition; p<0 . 05 ) or bath application of the NO synthase ( NOS ) inhibitor L-NAME ( ΔF/F: 0 . 00 ± 0 . 05; n = 5 cells; vs . control: p<0 . 05 ) . Importantly , fluorescent increases were reliably abolished with nitrendipine ( ΔF/F: −0 . 02 ± 0 . 06; n = 5 cells; vs . control: p<0 . 05 ) ( Figure 5B , C ) , suggesting that NO synthesis required L-VGCC activation . We then attempted to image NO release in response to more physiologically-relevant forms of postsynaptic stimulation , such as the complex spikes we were using to induce LTP . To do so , we pre-loaded slices with the NO-sensitive dye 1 , 2-Diaminoanthraquinone ( DAQ; 100 µg/mL ) , as previously described ( Chen et al . , 2001 ) . We then patched onto a single cell and imaged the DAQ-associated changes in the cell after stimulating the cell with 600 complex spikes , delivered at 5 Hz ( Figure 5D ) . Stimulation was performed in full glutamate receptor blockade . This protocol took advantage of the fact that DAQ forms an insoluble fluorescent precipitate upon reacting with NO , meaning that fluorescence would accumulate with stimulation and not readily wash away ( von Bohlen und Halbach et al . , 2002 ) . We found that with 600 complex spikes , the accumulated NO signal in the dendritic arbour was sufficiently large to detect by our setup ( Figure 5D , E ) . Notably , no signal was detected in the absence of any stimulation ( ΔF/F; stimulated vs . unstimulated: 2 . 97 ± 0 . 48; vs 0 . 22 ± 0 . 73; n = 9 , 7 cells; p<0 . 05 ) , or when stimulation was delivered in the presence of nitrendipine ( ΔF/F: −0 . 14 ± 0 . 65; n = 7 cells; vs . control: p<0 . 05 ) or L-NAME ( ΔF/F: 0 . 24 ± 0 . 43; n = 8 cells; vs . control: p<0 . 05 ) . These findings suggest that postsynaptic depolarization alone can drive NO release from neuronal dendrites in a L-VGCC dependent manner . Once NO is released , is it alone sufficient to induce LTPpre at active presynaptic terminals ? To address this , we examined whether increases in Pr could be elicited when presynaptic stimulation was paired with rapid photolytic release of NO ( 0 . 5–1 mM RuNOCl3 ) , in the absence of postsynaptic depolarization . We used Ca2+ imaging to determine basal Pr at a single synapse . We then paired 30–60 presynaptic stimuli , delivered at 5 Hz in full glutamate receptor blockade , with brief photolysis of NO , which was targeted to the spine head in order to emulate postsynaptic NO release . As with our standard LTP induction protocol , pairing was causal , with each NO photolysis event timed to occur 7–10 ms after each presynaptic stimulus . Under these conditions , we found significant increases in Pr when assessed 30 min post-pairing ( ΔPr: 0 . 29 ± 0 . 07; n = 10 spines; p<0 . 01; Figure 5F–H ) . No such changes were produced when pairing occurred in the presence of bath-applied cPTIO ( ΔPr: 0 . 03 ± 0 . 05; n = 8 spines; vs . causal pairing: p<0 . 05; Figure 5G , H ) , suggesting that LTPpre did not result from non-specific effects associated with photolysis . Remarkably , when pairing was reversed such that presynaptic stimuli followed NO photolysis , no significant change in Pr was observed ( ΔPr: −0 . 01 ± 0 . 04; n = 8 spines; vs . causal pairing: p<0 . 01; Figure 5F–H ) , suggesting that NO-mediated potentiation was Hebbian , requiring presynaptic activity to precede , rather than follow , NO release . Previously , the effects of NO on synaptic efficacy have primarily been examined by recording EPSPs in acute slices ( Padamsey and Emptage , 2014 ) . We therefore sought to confirm our findings using NO photolysis in the same preparation ( Figure 5—figure supplement 2 ) . We loaded CA1 pyramidal neurons in acute slices with caged NO ( 100 µM RuNOCl3 ) while recording EPSPs in the presence of AP5 . Wide-field photolysis was triggered using a 1 ms flash from a UV lamp . Causal pairings of presynaptic activity with photolysis resulted in an enhancement of the EPSP and a decrease in PPR . These increases were absent when pairings were anti-causal , or when pairings occurred in the presence of cPTIO , which instead resulted in a modest depression of the EPSP . These findings confirm that NO can trigger LTPpre provided that its release precedes rather than follows presynaptic activity . Our findings suggest that a presynaptic terminal need not release glutamate in order to become potentiated , provided that its activity precedes strong postsynaptic depolarization . In fact , in our initial experiments we found that glutamate release , if anything , inhibited LTPpre and promoted LTDpre ( Figures 1 and 2 ) . These findings , however , were based on elevating glutamatergic release probability at the synapse either by using high-frequency presynaptic bursts or glutamate photolysis . We therefore sought to examine whether endogenous glutamate release also had a similar effect of inhibiting LTPpre and promoting LTDpre . To investigate , we conducted single-spine Ca2+ imaging experiments in control conditions and under glutamate receptor blockade to examine how changes in Pr were affected by glutamate signalling . Remarkably , we found that increases in Pr produced in glutamate receptor blockade were significantly larger than under control conditions ( ΔPr; blockade vs . control: 0 . 34 ± 0 . 04 vs . 0 . 18 ± 0 . 02; n = 10 spines; p<0 . 05; Figure 6A–C ) , suggesting that even endogenously released glutamate reduced elevations in Pr induced by paired stimulation . We also examined the effects of glutamate receptor blockade on unpaired stimulation , during which single presynaptic stimuli ( 60 pulses at 5 Hz ) were delivered in the absence of postsynaptic depolarization . As before ( Figure 2E ) , this protocol reliably induced decreases in Pr at synapses with high release probabilities ( Pr >0 . 5 ) under control conditions ( Figure 6D ) . However , no such decreases were observed in glutamate receptor blockade ( ΔPr blockade vs . control: 0 . 00 ± 0 . 03 vs . −0 . 21 ± 0 . 05; n = 10 , 9 spines; p<0 . 05; Figure 6D , E ) . These findings suggest that endogenous glutamate release depresses Pr regardless of the level of postsynaptic depolarization . Across conditions , Pr changes were always more positive compared to controls when glutamate signalling was inhibited . How might glutamate release drive decreases in Pr ? We have previously reported functional and immunohistological evidence for the existence of presynaptic NMDARs at CA3-CA1 synapses ( McGuinness et al . , 2010 ) ; notably , these receptors act as reliable detectors for uniquantal glutamate release ( McGuinness et al . , 2010 ) , and have been implicated in LTDpre ( Rodríguez-Moreno et al . , 2013; Rodríguez-Moreno and Paulsen , 2008; Min and Nevian , 2012; Nevian and Sakmann , 2006; Andrade-Talavera et al . , 2016; Sjöström et al . , 2003 ) . We therefore examined whether glutamate was acting on these receptors to drive decreases in presynaptic efficacy . Given the difficulties associated with selectively blocking pre- , as opposed to post- , synaptic NMDARs , several groups have investigated the role of presynaptic NMDARs in plasticity by comparing the effects of bath application of AP5 or MK-801 , which blocks both pre- and postsynaptic NMDARs , with that of intracellular MK-801 application , which selectively blocks postsynaptic NMDARs ( Nevian and Sakmann , 2006; Corlew et al . , 2008; Corlew et al . , 2007; Cormier and Kelly , 1996 ) . We sought to use a similar approach . However , because MK-801 does not readily washout , and since postsynaptic NMDARs greatly contribute to spine Ca2+ influx ( Grunditz et al . , 2008; Emptage et al . , 1999; Holbro et al . , 2010 ) , we first examined whether the permanent loss of postsynaptic NMDAR signalling affected our ability to measure Pr using postsynaptic Ca2+ imaging . We found that at about 50% of synapses , NMDAR blockade reduced , but did not entirely abolish synaptically-evoked Ca2+ transients ( Figure 6—figure supplement 1A-C ) . The residual Ca2+ transients were mediated by activation of voltage-gated Ca2+ channels ( VGCCs ) in response to AMPAR-mediated depolarization , and could be used to accurately measure Pr ( Figure 6—figure supplement 1D , E ) . Importantly , the average Pr of these synapses did not significantly differ from that of synapses lacking a residual Ca2+ transient in NMDAR blockade ( ΔPr; AP5-sensitive vs . AP5-insensitive: 0 . 42 ± 0 . 07 vs . 0 . 47 ± 0 . 11; n = 8 spines/condition; p=0 . 67; Figure 6—figure supplement 1B , C ) . These findings suggest that , in NMDAR receptor blockade , VGCC-dependent spine-Ca2+ influx can be used as a means of calculating Pr at a sizeable and representative proportion of presynaptic terminals; nonetheless the use of VGCC-dependent Ca2+ transients presents an inevitable selection bias in our study . Using VGCC-dependent spine Ca2+ transients , we found that when both pre- and postsynaptic NMDARs were blocked by bath application of either AP5 or MK-801 , paired stimulation triggered increases in Pr ( ΔPr: 0 . 34 ± 0 . 03; n = 18 spines ) that were not significantly different from those produced in full glutamate receptor blockade ( p>0 . 99 ) , but that were greater than increases in Pr produced under control conditions ( p<0 . 01 ) ( Figure 6A–C ) . Bath blockade of NMDARs , like glutamate receptor blockade , also blocked decreases in Pr produced by unpaired stimulation ( ΔPr: −0 . 02 ± 0 . 02; n = 17 spines; vs . control; p<0 . 05; vs . blockade: p>0 . 99; Figure 6D , E ) . Notably , bath application of MK-801 produced similar effects as AP5 , suggesting that the effects of NMDARs on Pr are associated with its ionotropic , rather than metabotropic effects ( Nabavi et al . , 2013 ) ( ΔPr; bath MK-801 vs . AP5 paired stimulation: 0 . 33 ± 0 . 04 vs . −0 . 32 ± 0 . 03; n = 8 , 9 spines; p=0 . 47; bath MK-801 vs . AP5 unpaired stimulation: 0 . 00 ± 0 . 03 vs . −0 . 04 ± 0 . 03; n = 8 , 9 spines; p=0 . 27; Figure 6B , C ) To then specifically block postsynaptic NMDARs , we bolus-loaded cells intracellularly with MK-801 ( see Materials and methods ) . Since MK-801 can have off-target effects on voltage-gated channels ( Jaffe et al . , 1989; Kim et al . , 2015 ) , we ensured our loading protocol reliably abolished NMDAR-mediated EPSPs and NMDAR-mediated spine Ca2+ transients without impacting L-type voltage gated Ca2+ currents ( Figure 6—figure supplement 2 ) . In contrast to bath application of AP5 or MK-801 , we found that with postsynaptic application of MK-801 , increases in Pr produced by paired stimulation ( ΔPr: 0 . 16 ± 0 . 04; n = 9 spines; vs . control: p>0 . 99; Figure 6A–C ) and decreases in Pr produced by unpaired stimulation ( ΔPr: −0 . 25 ± 0 . 07; n = 9 spines; vs . control: p>0 . 99; Figure 6D , E ) did not significantly differ from control conditions ( p>0 . 99 ) , and were significantly different from changes in Pr induced in glutamate receptor blockade ( p<0 . 05 ) and extracellular NMDAR blockade ( p<0 . 05 ) . Collectively , these results suggest that glutamate release acts on pre- , but not post- , synaptic NMDARs to drive long-lasting decreases in Pr observed during both paired and unpaired stimulation . Notably , these decreases were independent of endocannabinoid signalling ( Figure 6—figure supplement 3 ) . We confirmed the effects of presynaptic NMDARs on presynaptic plasticity using electrophysiological recordings , both in organotypic ( Figure 6—figure supplement 4 ) and acute slices ( Figure 6—figure supplement 5 ) . Using PPR changes to monitor presynaptic plasticity , we found that bath , but not intracellular blockade of NMDARs augmented LTPpre and abolished LTDpre , consistent with our Ca2+ imaging results . Our findings that presynaptic NMDAR activation depresses Pr would explain why glutamate photolysis in our earlier experiments inhibited LTPpre and promoted LTDpre ( Figure 2 ) . To confirm this , we repeated our photolysis experiments in the presence of MK-801 , either intracellularly or extracellularly applied , again to differentially block pre- and postsynaptic NMDAR signalling ( Figure 6—figure supplement 4 ) . Consistent with our hypothesis , we found that bath , but not intracellular , application of MK-801 blocked the inhibitory effects of photolysis on LTPpre during paired stimulation , and prevented photolysis from inducing LTDpre during unpaired stimulation . To directly assess the involvement of pre- and postsynaptic NMDARs in presynaptic plasticity , we differentially targeted these receptors for genetic deletion . We cultured hippocampal slices from a mouse line in which the gene encoding GluN1 , the obligatory NMDAR subunit , was floxed ( Grin1fx/fx ) . We then virally injected Cre recombinase either into the CA3 or CA1 region to knockout pre- or postsynaptic NMDARs at Schaffer-collateral synapses . NMDAR currents were selectively abolished in targeted regions by 15 days post-injection ( Figure 7—figure supplement 1 ) . We first examined plasticity in Cre-injected and control Grin1fx/fx slices using electrophysiology ( Figure 7A ) . In these experiments PPR was measured throughout the experiment . Paired stimulation resulted in LTP ( fold ΔEPSPslope; control: 1 . 63 ± 0 . 08; CA3 KO: 2 . 09 ± 0 . 17; CA1 KO: 1 . 28 ± 0 . 06; n = 14 , 12 , 12 cells; p<0 . 001/condition; Figure 7B ) with a presynaptic component of expression ( ΔPPR; control: −0 . 23 ± 0 . 04; CA3 KO: −0 . 47 ± 0 . 06; CA1 KO: −0 . 22 ± 0 . 02; n = 13 , 12 , 12 cells; p<0 . 001/condition; Figure 7C ) that was evident across conditions , regardless of whether pre- or postsynaptic NMDARs were knocked out . PPR decreased by 5 min after plasticity induction ( p<0 . 05; Figure 7A ) , and continued to decrease over the duration of the recording , in line with previous findings that the expression of LTPpre evolves over time ( Bayazitov et al . , 2007 ) . Notably , slices lacking presynaptic NMDARs showed the greatest magnitude of LTP and the largest decrease in PPR , suggesting that LTPpre expression was strongest in this condition ( fold ΔEPSPslope; CA3 KO vs . control: p<0 . 05; CA3 KO vs CA1 KO: p<0 . 01; Figure 7B ) ( ΔPPR; CA3 KO vs . control: p<0 . 01; CA3 KO vs CA1 KO: p<0 . 01; Figure 7C ) . LTP magnitude of control slices exceeded that of slices lacking postsynaptic NMDARs ( p<0 . 05 ) , although PPR changes were of comparable magnitude in both conditions ( p>0 . 99 ) , suggesting that loss of postsynaptic NMDARs likely only impaired post- , but not pre- , synaptic plasticity . Collectively , these findings confirm that LTPpre induction is impaired by presynaptic NMDAR activation , and does not strictly require postsynaptic NMDAR activation . We also examined LTD in Grin1fx/fx slices . To induce LTD , we used our unpaired , high-frequency burst stimulation protocol ( 2 pulses at 200 Hz repeated 60 times at 5 Hz; as in Figure 1E ) . This protocol produced robust depression of recorded EPSPs across conditions ( fold ΔEPSPslope; control: 0 . 55 ± 0 . 06; CA3 KO: 0 . 65 ± 0 . 05; CA1 KO: 0 . 47 ± 0 . 06; n = 9 , 9 , 8 cells; p<0 . 01/condition; Figure 7D ) . However , increases in PPR were only seen in control ( ΔPPR = 0 . 34 ± 0 . 05; n = 8 ) and postsynaptic NMDAR knockout slices ( ΔPPR = 0 . 33 ± 0 . 05; n = 8; vs . control: p>0 . 99 ) , and were absent in presynaptic NMDAR knockout slices ( ΔPPR = 0 . 00 ± 0 . 05; n = 8; vs . control: p<0 . 01; vs . CA1 KO: p<0 . 01; Figure 7D , F ) . These findings confirm that pre- , but not post- , synaptic NMDARs are essential for LTDpre induction . Notably , loss of presynaptic NMDARs did not abolish LTD of the EPSP , suggesting that a postsynaptic component of LTD was likely still present in this condition . Changes in PPR , when present , were observed by 5 min following LTD induction ( p<0 . 05; Figure 7D ) and increased across the duration of the experiment , suggesting that the expression of LTDpre , like that of LTPpre , evolved over time . Lastly , we used Ca2+ imaging to directly examine changes in Pr at single synapses in Grin1fx/fx slices ( Figure 8 ) . In these experiments we assessed Pr at multiple time points following plasticity induction . Consistent with electrophysiological results , we found that genetic knockout of presynaptic NMDARs led to greater increases in Pr following paired stimulation ( ΔPr: CA3 KO vs . control: 0 . 37 ± 0 . 03 vs 0 . 20 ± 0 . 02; n = 12 spines/condition; p<0 . 001; Figure 8A–C ) , and abolished decreases in Pr triggered by unpaired stimulation ( ΔPr: CA3 KO vs . control: 0 . 00 ± 0 . 04 vs −0 . 45 ± 0 . 05; n = 12 , 11 spines; p<0 . 0001; Figure 8E–G ) . These effects were evident within 5–15 min after plasticity induction ( p<0 . 05 ) , and were maintained throughout the 45 min post-induction imaging period ( p<0 . 01; Figure D , H ) . We aligned changes in Pr with time course measurements of PPR obtained in electrophysiological experiments ( Figure 7A , D ) and found good agreement between both measures following paired ( Figure 8D ) and unpaired ( Figure 8H ) stimulation . Collectively , these findings confirm that glutamate release drives decreases in Pr via activation of presynaptic NMDARs . Such decreases occur independent of the level of postsynaptic depolarization; across conditions , changes in Pr following paired and unpaired stimulation were always more positive when presynaptic NMDAR signalling was absent . Our study is robust because we examine presynaptic plasticity under a diverse range of experimental conditions . We used both Ca2+ imaging and PPR to assess presynaptic plasticity in cultured and acute hippocampal slices using a number of pharmacological and genetic manipulations . With such diverse experimental techniques , preparations , and manipulations , we found consistent support for the proposed model of presynaptic plasticity . PPR and Ca2+ imaging are markedly different techniques to assess presynaptic efficacy , each with its own assumptions , advantages , and disadvantages . Ca2+ imaging provides a powerful means to monitor Pr at single synapses in brain slices ( Padamsey and Emptage , 2011 ) . The excellent signal-to-noise with which this technique can be used to detect uniquantal glutamate release makes Pr estimates robust to large changes in postsynaptic Ca2+ . Indeed , either removal of extracellular Mg2+ ( Figure 2—figure supplement 1 ) or bath application of AP5 ( Figure 2—figure supplement 1 ) , which more than doubles or halves the Ca2+ transient respectively , has no effect on Pr calculations made at resting membrane potential . Nonetheless , Ca2+ imaging may bias synapse selection , particularly in favour of synapses producing large Ca2+ transients ( Padamsey and Emptage , 2011 ) . Such selection bias , conversely , is absent in PPR calculations , which reflect aggregate changes in Pr over a larger number of stimulated synapses . However , PPR is an indirect measures of Pr , and may be confounded by factors such as postsynaptic receptor desensitization ( Yang and Calakos , 2013 ) ; though , not at the interpulse intervals used in this study ( Arai and Lynch , 1998 ) . Despite these caveats , in our study , results from PPR measurements and Ca2+ imaging were consistent with one another across a range of experimental conditions , making it unlikely that our assessment of presynaptic plasticity was confounded . Both of these techniques , therefore , present valid means of measuring presynaptic efficacy , at least in the context of this study . The proposed model provides a mechanism by which presynaptic terminals releasing little or no glutamate can become potentiated provided that their activity is accompanied by strong postsynaptic depolarization . Notably , most central synapses have low glutamate release probabilities , with some synapses appearing to release no glutamate in response to presynaptic stimulation ( Voronin and Cherubini , 2004; Stevens , 2003 ) . This is true for synapses recorded in both in vitro and ex vivo preparations from young and adult rodents . In fact , under electron microscopy , a significant portion of synapses ( up to 35–50% ) in the adult rodent hippocampus have presynaptic zones lacking synaptic vesicles in their near proximity ( <170 nm ) ; these so-called ‘nascent zones’ have been hypothesized to be functionally silent ( Bell et al . , 2014 ) . Although the existence of bona fide presynaptically silent synapses remains controversial ( Voronin and Cherubini , 2004 ) , the low release probabilities ( average Pr of approximately 0 . 2 [Murthy et al . , 1997] ) of central synapses suggests that it is possible that activity at a presynaptic terminal may not elicit glutamate release at the synapse , but may still coincide with strong postsynaptic depolarization , driven by glutamate release at other co-active synapses . Under such conditions , the mechanisms proposed in this study could enable presynaptic induction of Hebbian potentiation at these synapses . Our finding that presynaptic enhancements can occur without glutamatergic signalling at the synapse raises the question as to why many studies show that LTP induction can be abolished or impaired by blockade of one or more glutamate receptor subtypes ( Holbro et al . , 2010; Collingridge et al . , 1983; Bashir et al . , 1993 ) . To address this question , it is first necessary to recognize that not all LTP induction protocols are associated with presynaptic enhancements ( Padamsey and Emptage , 2014 ) . This is because LTPpre induction requires higher levels of postsynaptic depolarization than LTPpost induction ( Padamsey and Emptage , 2014; Zakharenko et al . , 2003; Bayazitov et al . , 2007 ) . Whether presynaptic enhancements are obtained will therefore depend on the levels of postsynaptic depolarization achieved during LTP induction , which in turn will be influenced by a variety of experimental factors , including the frequency and intensity of stimulation ( Padamsey and Emptage , 2014 ) . Nonetheless , even studies reporting LTPpre also find that inhibition of glutamate receptors , in particular NMDARs , abolish or reduce presynaptic enhancements ( Ryan et al . , 1996; Ratnayaka et al . , 2012; Emptage et al . , 2003; Bliss and Collingridge , 2013; Enoki et al . , 2009; Nikonenko et al . , 2003; Stanton et al . , 2005; Padamsey and Emptage , 2014; Zakharenko et al . , 2003; Bayazitov et al . , 2007; Zakharenko et al . , 2001 ) . In such cases it is important to recognize that AMPARs , KARs , NMDARs , and mGluRs can all contribute to postsynaptic depolarization ( Grienberger et al . , 2014; Schiller and Schiller , 2001; Grover and Yan , 1999b; Chemin et al . , 2003 ) . Given that presynaptic changes rely on the voltage-dependent release of NO , it is possible that blockade of any of these glutamate receptor classes would abolish or reduce LTPpre in an indirect way , by reducing postsynaptic depolarization and the activation of L-VGCCs . This may explain , in part , why experimental manipulations that augment the levels of postsynaptic depolarization reliably rescue LTP in AMPAR ( Holbro et al . , 2010; Fuenzalida et al . , 2010 ) , NMDAR ( Padamsey and Emptage , 2014; Grover and Teyler , 1992; Zakharenko et al . , 2003; Bayazitov et al . , 2007; Zakharenko et al . , 2001; Kullmann et al . , 1992; Huber et al . , 1995; Grover et al . , 2009; Morgan and Teyler , 2001 ) , and mGluR blockade ( Wilsch et al . , 1998 ) . Critically , our LTP induction protocol used strong postsynaptic depolarization , which was elicited by somatic current injection , and therefore independent of synaptic activity . This circumvented the need for any glutamate receptor-dependent depolarization during paired stimulation and enabled us to directly assess the function of glutamate signalling in LTPpre , independent of its effects on postsynaptic depolarization . Based on these results , we argue that the physiological role of glutamate release in LTPpre is for driving postsynaptic spiking as opposed to conveying a synapse-specific signal; this contrasts with the role of glutamate release in postsynaptic plasticity , in which synapse-specific activation of postsynaptic NMDARs is necessary for LTPpost induction . While our approach for inducing LTP resembles that of traditional STDP protocols , which rely on NMDAR activation ( Dan and Poo , 2004 ) , there are two key differences . Firstly , in our study , postsynaptic depolarization took the form of complex spikes , which included a brief period ( 7–10 ms ) depolarization before the first spike ( see Materials and methods ) . This period of subthreshold depolarization is known to facilitate the induction of LTP , possibly by inactivating voltage-gated K+ channels within the dendrite , which otherwise impede action potential backpropagation ( Watanabe et al . , 2002; Gasparini et al . , 2007; Hoffman et al . , 1997; Johnston et al . , 1999; Migliore et al . , 1999; Sjöström and Häusser , 2006 ) . Secondly , like complex spikes recorded in vivo ( Ranck , 1973 ) , the spike trains we triggered contained broadened action potentials , which likely reflect strong depolarization in the dendrites ( Hoffman et al . , 1997; Migliore et al . , 1999 ) . Consequently , the postsynaptic waveforms used in our study were likely to generate greater levels of postsynaptic depolarization , and in a manner independent of glutamate release and NMDAR activation , than those used in traditional STDP studies . It has long been recognized that the induction of LTPpre requires a retrograde signal ( Williams et al . , 1989 ) . One promising candidate is NO ( Garthwaite and Boulton , 1995 ) . The role of NO in plasticity has been a source of much controversy , and some studies have concluded that NO signalling is not necessary in LTP induction ( for review see [Padamsey and Emptage , 2014] ) . However , given that NO is likely to be important for presynaptic strengthening , the effect of NO signalling on synaptic plasticity will depend on whether presynaptic enhancements are obtained following LTP induction ( Padamsey and Emptage , 2014 ) . Indeed , studies that actually confirm presynaptic changes following LTP induction , including our own , consistently demonstrate that presynaptic enhancements depend on the synthesis and release of NO in both acute and cultured hippocampal preparations ( Ratnayaka et al . , 2012; Nikonenko et al . , 2003; Stanton et al . , 2005; Johnstone and Raymond , 2011 ) . It has generally been assumed that NO synthesis is dependent on Ca2+ influx from postsynaptic NMDARs ( Garthwaite and Boulton , 1995 ) ; however , several studies , including our own , have demonstrated that induction of LTPpre is possible in NMDAR blockade , suggesting that a NMDAR-dependent NO signalling pathway is not required for LTPpre ( Zakharenko et al . , 2003; Bayazitov et al . , 2007; Zakharenko et al . , 2001 ) . Here , we provide direct evidence for an alternative pathway for NO synthesis that is crucial for presynaptic strengthening , and that is driven by strong postsynaptic depolarization via the activation of L-VGCCs . Why L-VGCC- , as opposed to NMDAR- , mediated NO signalling is specifically required for LTPpre is not known , but may result from differences in the magnitude , kinetics , and/or spatial extent of NO signalling associated with L-VGCC and NMDAR activation . Unfortunately , the poor sensitivity of NO-indicator dyes makes this possibility difficult to currently investigate . It has previously been shown that exogenous NO can potentiate synaptic transmission , and that this potentiation is restricted to synapses that are active during NO release ( Arancio et al . , 1996; Zhuo et al . , 1993 ) . Here , we extend these findings by showing that photolysis of NO at single synapses can directly drive increases in Pr , and that this increase can occur in the absence of glutamatergic signalling . Moreover , we demonstrate that the potentiating effects of NO are not only restricted to active synapses , but specifically at synapses whose activity precede , rather than follow , NO release; thus , the requirements of NO signalling are consistent with those of Hebbian and spike-timing dependent plasticity ( Dan and Poo , 2004 ) . These findings also suggest the existence of a Hebbian detector at the presynaptic terminal that is sensitive to the timing between presynaptic activity and NO release; at least one isoform of guanylate cyclase is sensitive to NO in a Ca2+-dependent manner , making it a potential candidate for integrating NO signalling and presynaptic activity ( Zabel et al . , 2002 ) . Although our study focussed on phasic NO signalling , LTP may additionally require a tonic , low-level of NO signalling ( Hopper and Garthwaite , 2006 ) . It will be important to examine the differential roles of tonic and phasic NO signaling in presynaptic plasticity in future studies . Moreovoer , while we provide evidence in support of NO as a retrograde signal in LTPpre , it may not be the only retrograde signal involved . Indeed , neurotrophic factors , transsynaptic signals , as well as contact-dependent processes are all known to regulate Pr ( Regehr et al . , 2009 ) ; whether such signals play a role in LTPpre induction remains to be elucidated . At active presynaptic terminals , whereas Hebbian activity drives increases in Pr , we show , unexpectedly , that glutamate release drives decreases in Pr by acting on presynaptic NMDARs . Using both pharmacological and genetic manipulations , we found that presynaptic NMDAR signalling operated both during LTPpre and LTDpre induction paradigms to reduce Pr . Our finding suggests that the potentiating effects of Hebbian activity and the depressing effects of endogenous glutamate release occur concurrently during synaptic activity . Thus , the processes underlying LTPpre and LTDpre induction do not act independently as originally believed , but operate jointly to tune synaptic function . Our results may explain why sometimes the same pairing protocol that produces LTPpre at low Pr synapses , produces LTDpre at high Pr synapses; presumably the level of Hebbian activity achieved by such protocols is not of sufficient magnitude to prevent the depressing effects of glutamate release at high Pr synapses ( Hardingham et al . , 2007; Sáez and Friedlander , 2009 ) . Our results may also explain why the locus of LTP expression , whether pre- or postsynaptic , appears to depend on initial Pr ( Larkman et al . , 1992 ) . With higher basal release probabilities , more glutamate is released for a given LTP induction protocol , meaning that LTPpost is favoured owing to greater postsynaptic NMDAR-signalling , whereas LTPpre is inhibited owing to greater presynaptic NMDAR-signalling . Thus , low Pr synapses will have a tendency to express LTP presynaptically , while high Pr synapses will have a tendency to express LTP postsynaptically ( Larkman et al . , 1992 ) . In contrast to our findings , inhibition of presynaptic NMDARs at neocortical synapses does not appear to effect LTP magnitude ( Rodríguez-Moreno et al . , 2013; Rodríguez-Moreno and Paulsen , 2008; Sjöström et al . , 2003 ) . It is possible that the low frequency ( 0 . 2 Hz ) of presynaptic stimulation used during LTP induction in these studies does not elicit sufficient glutamate release to drive decreases in Pr via presynaptic NMDAR activation . Studies using STDP protocols , however , have found a role for presynaptic NMDARs in the induction of LTDpre at neocortical synapses ( Min and Nevian , 2012; Nevian and Sakmann , 2006; Sjöström et al . , 2003; Sjöström et al . , 2007 ) . This form of LTDpre is thought to additionally require endocannabinoid receptor 1 ( CB1R ) signalling . Although we also found presynaptic NMDARs to be necessary for LTDpre induction , we found no requirement for CB1Rs . However , the protocol we used to induce LTDpre was not a STDP protocol , and did not involve postsynaptic spiking , which is thought to be necessary to drive endocannabinoid release ( Min and Nevian , 2012 ) . Instead , our protocol used presynaptic stimulation , either in the form of single or short bursts of action potentials , delivered in the absence of postsynaptic depolarization . Rodríguez-Moreno et al . , 2013 similarly found that patterned presynaptic stimulation delivered in the absence of postsynaptic spiking induced LTDpre at neocortical synapses , and in a manner independent of CB1R activity . Such findings suggest that under some experimental conditions , glutamate release from presynaptic terminals alone is sufficient to induce LTDpre by acting on presynaptic NMDARs without the additional need for endocannabinoid signalling . NMDARs can have both metabotropic and ionotropic receptor signalling capacities ( Dore et al . , 2016 ) . We found that blocking ionotropic signalling with bath , but not postsynaptic , application of MK-801 was sufficient to prevent glutamate from driving decreases in Pr . Combined with our conditional NMDAR knockout experiments , these findings suggest that ionotropic presynaptic NMDAR signalling is necessary for depressing Pr . Blockade of presynaptic NMDAR function with MK-801 similarly abolishes LTDpre in neocortex induced by STDP protocols ( Rodríguez-Moreno et al . , 2013; Rodríguez-Moreno and Paulsen , 2008; Rodríguez-Moreno et al . , 2011 ) . A recent paper by ( Carter and Jahr , 2016 ) , however , failed to find functional evidence for presynaptic NMDARs in the neocortex ( but see [Abrahamsson et al . , 2017] ) , and showed that instead , metabotropic signalling by postsynaptic NMDARs was responsible for spike-timing dependent LTD ( Carter and Jahr , 2016 ) . The locus of LTD expression , however , was not assessed in this study . Given that metabotropic receptor signalling from postsynaptic NMDARs is believed to underlie the induction of LTDpost ( Nabavi et al . , 2013 ) , and based on our current findings , we would hypothesize that ionotropic presynaptic NMDAR signalling will preferentially play a role in LTPpre and LTDpre induction . Previously we have demonstrated that presynaptic NMDARs at hippocampal synapses facilitate transmitter release during theta stimulation ( McGuinness et al . , 2010 ) . When considered with our current findings , presynaptic NMDARs appear to be important for presynaptic facilitation in the short-term , but presynaptic depression in the long-term . This is consistent with the finding that presynaptic NMDARs in the neocortex similarly mediate short-term plasticity of glutamate release , and yet are similarly implicated in LTDpre ( Min and Nevian , 2012; Sjöström et al . , 2003; Corlew et al . , 2008 ) . It may appear peculiar for a single protein to mediate seemingly disparate functions; however , another way to view the presynaptic NMDAR is as a dynamic regulator of presynaptic activity , appropriately tuning glutamate release depending on the patterns of pre- and postsynaptic activity . As such , the receptor may aid glutamate release during theta-related activity , but , triggers LTDpre when this release fails to elicit sufficiently strong levels of postsynaptic depolarization . In this study we present evidence for a novel model of presynaptic plasticity , in which changes in Pr at presynaptic terminals depend on the levels of 1 ) Hebbian signalling and 2 ) glutamate release that accompany presynaptic activity ( Figure 9 ) . Critically , the levels of glutamate release at a synapse will not only depend on basal Pr , but also on the pattern of presynaptic activity and on the state of the synapse ( e . g . facilitating or depressing ) ( Dobrunz et al . , 1997; Dobrunz and Stevens , 1997 ) , which dictates how Pr changes throughout a train of stimulation . One interpretation of the proposed model is that presynaptic plasticity , by adjusting Pr , corrects any mismatch between two variables: 1 ) the likelihood that presynaptic activity is accompanied by strong postsynaptic depolarization ( i . e . Hebbian activity ) and 2 ) the likelihood that presynaptic activity is accompanied by glutamate release . Accordingly , as we have shown in this study , increases in Pr will preferentially occur when Hebbian activity is present at the synapse , but glutamate release is absent; whereas decreases in Pr will preferentially occur when Hebbian activity is absent , but glutamate release is present . These scenarios reflect the correction of an otherwise profound mismatch that exists between the ability for presynaptic activity at a synapse to drive postsynaptic activity ( reflected by the amount of glutamate release ) , and the ability for presynaptic activity to predict postsynaptic spiking ( reflected by the amount of Hebbian signalling ) . We would hypothesize that Pr would continue to change until these mismatches are corrected . This could explain why , for a given plasticity induction protocol , Pr tends to a common equilibrium value across synapses ( Hardingham et al . , 2007 ) ; this value presumably reflects the point at which the levels of glutamate and Hebbian signalling associated with the stimulation protocol are equally matched . A key implication of our model is that the pattern of presynaptic activity will substantially impact changes in Pr . As demonstrated in our study , when high frequency bursts of presynaptic stimulation are paired with postsynaptic spiking , Pr remains low , whereas when single presynaptic stimuli are instead paired with postsynaptic spiking , Pr potentiates to higher values ( Figure 1 ) . At high Pr synapses , it is known that glutamate release is preferentially driven by single spikes , and otherwise depresses in response to high frequency bursting ( Dobrunz et al . , 1997; Dobrunz and Stevens , 1997 ) . By contrast , at low Pr synapses , glutamate release is preferentially driven by high frequency bursts , and is minimally responsive to single presynaptic spikes ( Dobrunz et al . , 1997; Dobrunz and Stevens , 1997 ) . Thus , presynaptic plasticity appears to adjust Pr such that glutamate release is preferentially driven by the pattern of presynaptic stimulation ( bursts or singe spikes ) that best predicts strong postsynaptic depolarization; Pr is set low in the case of presynaptic bursts and high in the case of single presynaptic spikes . This is particularly relevant given that different patterns and frequencies of presynaptic firing are likely to convey different information ( Butts and Goldman , 2006 ) . Consequently , presynaptic plasticity would enable the presynaptic terminal to act as a dynamic filter by preferentially tuning Pr to ensure that only information relevant for postsynaptic spiking is transmitted . Such a process would greatly enhance the signal-to-noise ratio of synaptic transmission . Unless otherwise stated in the text , cultured hippocampal slices were used for imaging and electrophysiological experiments owing to the excellent optical and electrophysiological access to cells and synapses afforded by this preparation . Cultured hippocampal slices ( 350 µm ) were prepared from male Wistar rats ( P7-P8 ) , as previously described ( Emptage et al . , 2003 ) . Slices were maintained in media at 37°C and 5% CO2 for 7–14 days prior to use . Media comprised of 50% Minimum Essential Media , 25% heat-inactivated horse serum , 23% Earl’s Balanced Salt Solution , and 2% B-27 ( ThermoFisher Scientific - Invitrogen , UK ) with added glucose ( 6 . 5 g/L ) , and was replaced every 2–3 days . During experimentation , slices were perfused with artificial cerebrospinal fluid ( ACSF; 1–2 mL/min ) , which was constantly bubbled with carbogen ( 95% O2 and 5% CO2 ) and heated to achieve near-physiological temperatures in the bath ( 31-33oC ) . ACSF contained ( in mM ) 145 NaCl , 16 NaHCO3 , 11 glucose , 2 . 5 KCl , 2–3 CaCl2 , 1–2 MgCl2 , 1 . 2 NaH2PO4 , and , to minimize photodynamic damage , 0 . 2 ascorbic acid and 1 Trolox . Acute hippocampal slices were used to confirm key findings in cultured hippocampal slices . When this preparation was used , it is clearly stated in the text and figure captions . Coronal acute hippocampal slices ( 400 µm ) were prepared from 2 to 3 week old male Wistar rats . Tissue was dissected in a sucrose-based ACSF solution ( in mM: 85 NaCl , 65 sucrose , 26 NaHCO3 , 10 glucose , 7 MgCl2 , 2 . 5 KCl , 1 . 2 NaH2PO4 , and 0 . 5 CaCl2 ) . The whole brain was sliced into coronal sections using a Microm HM 650V vibratome ( Thermo Scientific , UK ) . Hippocampal tissue were allowed to recover at room temperature in normal ACSF ( 120 NaCl , 2 . 5 KCl , 2 CaCl2 , 1 MgCl2 , 1 . 2 NaH2PO4 , 26 NaHCO3 , and 11 glucose ) , which was bubbled with 95% O2 and 5% CO2 . Slices were given at least 1 hr to recover before use . During experimentation , slices were perfused with ACSF ( 3 mL/min ) containing picrotoxin ( 100 µM; Sigma , UK ) . The ACSF was constantly bubbled with carbogen ( 95% O2 and 5% CO2 ) and heated to achieve near-physiological temperatures in the bath ( 31-33oC ) . The GluN1 NMDAR obligatory subunit was selectively knocked out of CA3 or CA1 neurons to respectively remove either pre- or postsynaptic NMDAR function at the Schaffer-collateral synapses . Hippocampal slices were cultured from Grin1fx/fx mouse pups ( P6-P8 ) ( B6 . 129S4-Grin1tm2Stl/J; Stock no . 005246; Jackson Laboratory , Bar Harbor , Maine , USA ) in which both copies of the GluN1 encoding genes are floxed . After 1–2 days in culture , Cre recombinase ( AAV1 . hSyn . Cre . WPRE . hGH; Penn Vector ) and a floxed variant of tdTomato ( AAV1 . CAG . Flex . tdTomato . WPRE . bGH; Allen Institute , Seattle , Washington , US ) were co-injected into either the CA3 or CA1 region using a sharp glass pipette ( 100–120 MΩ ) with its tip broken , coupled to a picospritzer ( Science Products , Germany ) . For dense transfection of the CA3 region , a total of 75–150 nL of virus was injected over three sites at a high titer ( Cre – 6 . 6 × 1012 GC/mL; tdTomato – 2 . 94 × 1012 GC/mL ) . For sparse transfection of CA1 cells , a single CA1 site was injected with 50 nL of virus at a lower titre . For controls , injections into CA3 or CA1 lacked Cre recombinase . Knockout was assessed by examining patch recordings of NMDAR currents at +40 mV in the presence of NBQX ( 10 µM; Abcam , UK ) and picrotoxin ( 100 µM ) . NMDAR currents were abolished by 15 days post injection . Blind patch recordings in CA3 revealed that 91% ( 21/23 ) of cells lacked NMDAR currents , suggesting that injections had successfully infected the vast majority of cells in this region . CA1 pyramidal neurons were recorded from either using low ( 4–8 MΩ ) or high resistance patch electrodes ( 18–25 MΩ ) filled with standard internal solution ( in mM: 135 KGluconate , 10 KCl , 10 HEPES , 2 MgCl2 , 2 Na2ATP and 0 . 4 Na3GTP; pH = 7 . 2–7 . 4 ) , or sharp microelectrodes ( 80–120 MΩ ) filled with 400 mM KGluconate . In some experiments ( Figure 3—figure supplement 2; Figure 5—figure supplement 1 ) low resistance ( 4–8 MΩ ) patch electrodes were used containing an ATP regenerating internal solution in order to minimize the effects of postsynaptic dialysis ( in mM: 130 KGluconate , 10 KCl , 10 HEPES , 10 NaPhosphocreatine , 4 MgATP , 0 . 4 Na3GTP and 50 U/mL creatine phosphokinase; pH = 7 . 2–7 . 4 ) ( Kullmann et al . , 1992 ) . The recording method used in a given experiment is indicated in the main text or the figure caption . A glass electrode ( 4–8 MΩ ) , filled with ACSF , was placed in stratum radiatum . Continuous basal stimulation ( 0 . 05–0 . 10 Hz ) was present for all experiments , and was only interrupted to deliver paired-pulse or tetanic stimulation . Stimulation intensity was adjusted to evoke a 5–10 mV EPSP; pulse duration was set at 100 µs . Paired-pulse stimulation , unless otherwise stated , consisted of 2 presynaptic stimuli delivered 70 ms apart . Baseline recordings were kept short ( approximately 5 min ) when recording using low resistance patch electrodes ( 4–8 MΩ ) to minimize the effects of dialysis . We found that LTPpre induction was impaired with longer baseline recordings . Indeed , we could induce LTPpre under NMDAR blockade following a 5 min baseline recording ( Figure 7A–C ) but not a 10 min baseline recording ( 5 vs . 10 min: fold ΔEPSPslope: 1 . 91 ± 0 . 13 vs 0 . 87 ± 0 . 08; ΔPPR: −0 . 48 ± 0 . 08 vs −0 . 03 ± 0 . 03; n = 12 vs . 5 cells; p<0 . 01 ) . LTP induction consisted of 60 single pulses delivered at 5 Hz each paired with postsynaptic depolarization . Postsynaptic depolarization took the form of a complex spike . To emulate a complex spike , we injected a postsynaptic current waveform ( 2–3 nA ) that was approximately 60 ms in duration and resulted in 3–6 spikes at ~100 Hz , with the first spike occurring 7–10 ms after the presynaptic stimulus . This was done by injecting a current waverform ( 2–3 nA ) with a 7–10 ms rising phase , a 20 ms plateau phase , and a 30–33 ms falling phase . However , in experiments shown in Figure 2 , the current waveform took the form of a 50 ms flat current step; although successful , this protocol led to poorer control of the start of postsynaptic bursting . LTP induction in both instances was more robust when the cell was depolarized by approximately 10–15 mV from resting membrane potential ( approx . −65 mV ) during the 12 s induction protocol; this facilitated broad spiking during postsynaptic current injeciton . Stimulating electrodes were placed within 50–70 µm of the soma to ensure that postsynaptic depolarization reached stimulated synapses without significant attenuation . In glutamate receptor blockade experiments , the two stimulating electrodes used were placed at the same depth in the slice to ensure that drug washout rates were comparable in both pathways . In these experiments , if a strong monosynaptic IPSP was present following application of full glutamate receptor blockade , the experiment was omitted . For two pathway experiments , to ensure each electrode was stimulating independent populations of axons we used the collision test ( Lipski , 1981 ) . Briefly , each pathway was successively stimulated , 1–2 ms apart . If both axonal populations are perfectly overlapping , then successive stimulation should generate a synaptic response comparable to the stimulation of either pathway alone , owing to the axonal refractory period . If however , both axonal populations are perfectly independent , successive stimulation should generate an EPSP response comparable to the sum total of the EPSP generated by stimulation of either pathway alone . LTD induction consisted of either 60 single or paired ( inter-stimulus interval of 5 ms ) presynaptic pulses delivered at 5 Hz in the absence of postsynaptic depolarization; single pulses only induced LTDpre at high Pr synapses ( Pr >0 . 5; Figures 2E and 6D ) , whereas paired pulses induced LTD at all synapses ( Figure 8F ) . During either stimulation regime , the membrane was hyperpolarized ( <-100 mV ) to prevent somatic and dendritic spiking; stimulation intensity was also kept low to avoid spiking , such that basal EPSP amplitude did not typically exceed 5 mV . All electrophysiological data was recorded using WinWCP ( Strathclyde Electrophysiology Software ) and analyzed using Clampfit ( Axon Insturments ) and Excel ( Microsoft ) . The initial EPSP slope , calculated during the first 2–3 ms of the response , was used to analyze changes in the EPSP throughout the recording . This was done to ensure only the monosynaptic component of the EPSP was analyzed . This is particularly important in cultured slices in which polysynaptic activity may confound EPSP amplitude measures . All data was normalized to the average EPSP slope recorded during baseline to yield ΔEPSP slope . Paired pulse ratio ( PPR ) was calculated as the average EPSP slope evoked by the second stimulation pulse divided by the average EPSP slope evoked by the first stimulation pulse , as previously described ( Kim and Alger , 2001 ) ; averages were calculated from 5 to 10 paired pulse trials . Decreases in PPR are thought to reflect increases in release probability ( Schulz et al . , 1994 ) . Confocal images were taken using a BioRad MRC-1000 confocal laser scanning system , controlled by LaserSharp software . A 488 nm argon laser line was used for fluorophore excitation . Images were acquired on an upright Olympus BX50WI microscope equipped with a 60x water-immersion objective ( Olympus; 0 . 9 NA ) . Bolus loading was used to fill CA1 neurons with dye or drugs whilst minimizing the amount of time the cell was patched on to . Loading was achieved by transiently patching onto cells ( 60 s ) using low-resistance patch electrodes ( 4–8 MΩ ) containing a high-concentration of drug or dye ( see relevant sections of Materials and methods for exact concentrations ) dissolved in standard internal solution . Slow withdrawal of the patch using a piezoelectric drive ensured re-sealing with no observable adverse effects to cell health . Cells were then subsequently re-patched for the purposes of delivering postsynaptic depolarization if and when required . For Ca2+ imaging , cells were bolus-loaded with OGB-1 ( 0 . 5–1 mM for 60 s ) to enable Pr measurements to be conducted in the absence of electrophysiological recordings , and associated dialysis . Cells were re-patched during LTP or LTD induction . A stimulating glass electrode ( 4–8 MΩ ) was then brought near ( 5–20 µm ) to a branch of imaged dendrite within stratum radiatum . For visualization purposes , electrode tip was coated with bovine serum albumin Alexa Fluor 488 conjugate ( ThermoFisher Scientific , Invitrogen , UK ) , as previously described ( Ishikawa et al . , 2010 ) . Briefly , a 0 . 05% BSA-Alexa 488 solution was made with 0 . 1M phosphate-buffered saline containing 3 mM NaN3 . Pipette tips were placed in the solution for 2–5 min . To find a synapse responsive to axonal stimulation , axons were stimulated with pairs of stimuli ( 2 pulses 70 ms apart ) to increase the chances of eliciting a Ca2+ response . During stimulation , laser scanning was initially restricted to a single line through a number of synapses on the dendrite to enable for rapid assessment of potentially responsive spines . Because stimulation intensity was kept low to prevent dendritic and somatic spiking , generally only one or two spines could be clearly identified as responding to stimulation; though only one spine was typically taken for experimentation since laser scanning had to be restricted to a line crossing both the spine and a region of underlying dendrite in order to determine if spine Ca2+ signals were contaminated by dendritic or somatic spikes . Responsive synapses were always found in the vicinity of the stimulating electrode , which was placed within 100 µm of the soma . Synapse selection , however , was invariably biased in favour of mushroom spines , with head diameters ranging from 0 . 3 to 1 . 0 µm , as these synapses were clearly visible and produced larger Ca2+ transients . Ca2+ images were acquired in line scan mode at a rate of 500 Hz and analyzed using ImageJ and Microsoft Excel . Increases in spine fluorescence ( ΔF/F = Ftransient–Fbaseline/Fbaseline ) following the delivery of the first stimulus is thought to reflect successful glutamate release from the presynaptic terminal ( Emptage et al . , 2003; Emptage et al . , 1999 ) . The proportion of successful fluorescent responses to the first stimulus across stimulation trials was used to calculate Pr . Pr was assessed on the basis of 15–40 trials at baseline and at 25–30 min post-tetanus . For high Pr synapses ( >0 . 8 ) the number of stimulation trials was limited to 15–20 to avoid photodynamic damage that results from imaging the frequent Ca2+ responses generated at these synapses . For all other synapses , Pr was generally assessed using 20–35 trials of stimulation . Stimulation was kept of a sufficiently low intensity to avoid somatic and dendritic spiking . When spikes did occur , as evidenced by a simultaneous Ca2+ rise in both the spine and the dendrite , a successful release event would require spine fluorescence to precede that of the dendrite , or to be of greater magnitude ( Nevian and Helmchen , 2007 ) . Synapses with initial Pr values of 0–0 . 7 were used for LTP experiments , and synapses with Pr values of 0 . 4–1 . 0 were used for LTD experiments . In experiments involving glutamate receptor blockade , Pr was measured prior to drug application at baseline , and measured post-tetanus , following drug washout . In experiments involving NMDAR blockade , using either AP5 or MK-801 , drugs were present for the duration of the experiment and , therefore , present for both the baseline and post-tetanus measurements of Pr . Experiments were excluded if the synapse became non-responsive and there was evidence of either substantial drift of the stimulation electrode or photodynamic damage ( i . e . blebbing of the dendrite or sudden increases in basal fluorescence intensity ) . Experiments involving DAF-FM ( ThermoFisher Scientific , Invitrogen , UK ) imaging were carried out in Tyrodes buffer ( in mM: 120 NaCl , 2 . 5 KCl , 30 glucose , 4 CaCl2 , 0 MgCl2 , and 25 HEPES ) containing 50 µM D-AP5 , 10 µM NBQX , 500 µM MCPG , and 100 µM LY341495 ( Abcam , UK ) to block glutamate receptors , as well as 1 µM Bay K-8644 ( Abcam , UK ) to prevent L-VGCC desensitization during K+ application as previously described ( Sattler et al . , 1999; Stanika et al . , 2012 ) . CA1 pyramidal neurons were bolus-loaded with 250 µM of DAF-FM for 60 s . Apical dendrites , often secondary or tertiary branches , within 100 µm of the soma were imaged at one focal plane , once prior to , and once 5–10 s following , the addition of a high K+ Tyrodes solution ( in mM: 32 . 5 NaCl , 90 KCl , 30 glucose , 4 CaCl2 , 0 MgCl2 , 25 HEPES , which included: 50 µM D-AP5 , 10 µM NBQX , 500 µM MCPG , and 100 µM LY341495 ) . Laser power and exposure was kept to a minimum to avoid photobleaching . In our hands , DAF-FM basal fluorescence was not quenched by intracellular addition of cPTIO . 1 , 2-Diaminoanthraquinone ( DAQ; Sigma , UK ) was used to image activity-dependent NO release under more physiological conditions . DAQ was loaded as previously described ( Chen et al . , 2001 ) . DAQ was prepared as a 5 mg/mL stock solution dissolved in DMSO . Hippocampal slices cultures were treated with 100 µg/mL of the solution for 2 hr at 37°C and 5% CO2 . Slices were then placed on the rig , and perfused with heated ( 31-33oC ) and carbogenated ( 95% O2 and 5% CO2 ) ACSF for 30 min prior to imaging to wash-off excess dye . DAQ was imaged in full glutamate receptor blockade using 488 nm excitation light and a 570 nm long-pass emission filter prior to and following stimulation of a single patched CA1 neuron with 600 complex spikes at 5 Hz ( see Stimulation protocols section ) . Control cells were left unstimulated . Following DAQ imaging , cells were re-patched , loaded with Alexa Fluor 488 ( 100 µM; ThermoFisher Scientific , Invitrogen , UK ) , and imaged . Alexa Fluor 488 fluorescence was used to determine the proportion of imaged DAQ fluorescence that co-localized to the recorded cell . DAQ fluorescence was compared before and after stimulation in the imaged cell . A 405 nm laser ( Photonics , UK ) was used for spot photolysis . The laser was focussed to a small spot ( ~1 . 2 µm diameter ) by overfilling the back aperture of a 60x water-immersion lens ( Olympus , UK ) . Electrode manipulators and recording chambers were mounted on a movable stage , which enabled a region above the spine head to be positioned beneath the photolysis spot . Laser exposure was controlled using a fast shutter ( LS6; Uniblitz ) . For glutamate photolysis , MNI glutamate ( Tocris , UK ) was focally delivered through a glass pipette ( 4–8 MΩ; 10 mM MNI glutamate ) using a picospritzer ( Science Products , Germany ) . Laser exposure was limited to ~2 ms and , in each experiment , the laser intensity ( 0 . 5–2 mW ) was adjusted to generate a Ca2+ response in the underlying spine that was comparable to the response generated by electrical stimulation . Ruthenium nitrosyl chloride ( RuNOCl3 ) , which has sub-millisecond release kinetics ( Bettache et al . , 1996 ) , was used for NO photolysis experiments . For spot photolysis , 0 . 5–1 mM RuNOCl3 ( Sigma ) was bath applied and uncaged using 30–60 laser pulses ( 25 ms; 2 mW ) delivered at 5 Hz; presynaptic stimulation either preceded or followed NO photolysis by 7–10 ms . Using the NO-indicator , DAF-FM ( Invitrogen ) , we calibrated laser power to liberate approximately 10 nM of NO per pulse . We did this by targeting the soma of DAF-FM loaded neurons ( 250 µM bolus-loaded ) for photolysis at different laser powers while recording the resulting increases in fluorescence using the confocal laser in line scan mode ( 500 Hz ) . We aimed for an increase in fluorescence of about 6–7% ( averaged across several trials ) , which based on the manufacturer’s data on the concentration-dependent fluorescence of DAF-FM , amounts to a release of approximately 10 nM of NO . For wide-field UV photolysis , 100 µM RuNOCl3 was added to the patch electrode and allowed to diffuse into the cell for 10–15 min prior to commencing the experiment . A UV Flash Lamp ( HI-TECH Scientific ) was used to deliver a 1 ms wide-field uncaging pulse ( 100 V ) that was timed to occur 7–10 ms before or after presynaptic stimulation . Because of the time required for the UV lamp to recharge between flashes , about 20 of the 60 presynaptic pulses delivered at 5 Hz were not associated with a flash . In experiments requiring both pre- and postsynaptic NMDARs to be blocked , either D-AP5 ( 50–100 µM; Abcam , UK ) or MK-801 ( 20 µM; Abcam , UK ) was added in bath for the duration of the experiment . In the case of MK-801 , slices were pre-incubated with the drug for at least 1 hr prior to experimentation . Experiments in which NMDARs were blocked with bath application of AP5 and with bath application of MK-801 produced similar results and so conditions were combined for data analysis . Postsynaptic NMDARs were blocked by bolus loading of 5 mM MK-801 for 60 s , after which 20 min was given for the drug to diffuse and take effect . This was the case for both imaging ( Figure 6 ) and electrophysiology experiments ( Figure 7 ) . In the latter , cells were re-patched 20 min after bolus loading with normal internal solution; re-patching did not result in a notable intracellular washout of MK-801 , likely reflecting the high affinity of drug binding ( approximately 37 nM ) ( Wong et al . , 1986 ) . Most electrophysiological experiments in the literature use 1 mM of MK-801 in the patch electrode to block postsynaptic NMDARs ( Rodríguez-Moreno et al . , 2013; Rodríguez-Moreno and Paulsen , 2008; Nevian and Sakmann , 2006; Rodríguez-Moreno et al . , 2011 ) . We found that using this protocol , we failed to induce LTP using paired stimulation ( ΔEPSPslope: 0 . 83 ± 0 . 13; vs . 1 . 0: p=0 . 22; ΔPPR: 0 . 13 ± 0 . 08 vs . 0; p=0 . 19 ) . However , this protocol may have resulted in more intracellular loading of MK-801 than our rapid ( 60 s ) bolus loading approach , and may have therefore resulted in off-target effects . At high concentrations ( >100 µM ) , MK-801 can inhibit voltage-gated K+ and Ca2+ channels ( Jaffe et al . , 1989; Kim et al . , 2015 ) . Indeed , we found that 1 mM patch loading of MK-801 for 5 min reduced L-VGCC-mediated Ca2+ influx ( isolated by using 10 µM mibefradil , 0 . 3 µM SNX-482 , and 1 µM ɯ-conotoxin-MVIIC , as previously described [Bloodgood and Sabatini , 2007] ) by approximately 50% ( ΔF/F: control vs . 1 mM MK-801: 1 . 15 ± 0 . 05 vs . 0 . 53 ± 0 . 06; p<0 . 01; Figure 6—figure supplement 2D-F ) . Our bolus loading procedure ( 5 mM MK-801 for 60 s ) , by contrast , produced no change in L-VGCC function ( ΔF/F: 1 . 10 ± 0 . 07; vs . control: p=0 . 99; vs . 1 mM MK-801: p<0 . 05; Figure 6—figure supplement 2D-F ) despite effectively inhibiting NMDAR function ( Figure 6—figure supplement 2A-C ) , and also failed to abolish the induction of LTPpre using our pairing protocol ( Figure 6—figure supplement 4A-C , Figure 6—figure supplement 5A-C ) . Glutamate receptor blockade was achieved using D-AP5 ( 50–100 µM; Abcam , UK ) , NBQX ( 10 µM; Abcam , UK ) , R , S-MCPG ( 500 µM; Abcam , UK ) and LY341495 ( 100 µM; Abcam , UK ) . L-VGCCs were blocked with nitrendipine ( 20 µM; Abcam , UK ) . NO synthase was inhibited by pre-incubation of slices with L-NAME ( 100 µM; Sigma , UK ) , which started at least 20 min prior to experimentation . Extracellular NO was scavenged by bath application of cPTIO ( 50–100 µM; Sigma , UK ) . Intracellular NO was scavenged by bolus loading cells with 5 mM cPTIO . Endocannabinoid signalling ( CB1 receptor ) was inhibited by bath application of the AM-251 ( 2 µM; Tocris , UK ) . Tests used to assess statistical significance are stated at the end of all Figure captions . Only non-parametric tests were used owing to small sample sizes ( Siegel , 1956 ) . For single comparisons , two-tailed Mann-Whitney or Wilcoxon matched pairs signed rank tests were used , depending on whether the data was unpaired or paired , respectively . Wilcoxon signed rank tests were also used to determine if data significantly differed from an expected value . For multiple comparisons , Kruskal-Wallis tests were used with post-hoc Dunn’s tests . Means and standard error of the mean ( S . E . M . ) are represented in the text as mean ±S . E . M . Sample sizes that were typical for the field ( n = 5–15; independent experiments/biological replicates ) were used in this study , and provided sufficient power ( >80% ) to detect the expected experimental effects reported in our study . For spine imaging , typically only one spine was imaged per cell per experiment . Samples were randomly assigned to conditions that were being concurrently run . Masking was not used for sample allocation or data collection . A single data point reflected the average of multiple measurements ( technical replicates ) within an experiment; this is detailed in the relevant Materials and methods sections .	Neurons communicate with one another at junctions called synapses . One neuron at the synapse releases a chemical substance called a neurotransmitter , which binds to and activates the other neuron . The release of neurotransmitter thus enables the electrical activity of one cell to influence the electrical activity of another . The efficiency of this communication can change over time , as is thought to occur during learning . If the neurons on both sides of a synapse are repeatedly active at the same time , the ability of the neurons to transmit electrical signals to each other increases . One way that communication between neurons can become more efficient is if the first neuron becomes more likely to release neurotransmitter . Most synapses in the brain release a neurotransmitter called glutamate , and most types of learning involve changes in the efficiency of communication at glutamatergic synapses . But glutamate release is unreliable . Active glutamatergic neurons fail to release glutamate about 80% of the time . If glutamate has a key role in learning , how does the brain learn efficiently when glutamate release is so unlikely ? To find out , Padamsey et al . studied glutamatergic synapses in slices of tissue from mouse and rat brains . When both neurons at a synapse were repeatedly active at the same time , the first neuron would sometimes become more likely to release glutamate . But this only happened at synapses in which the first neuron usually failed to release glutamate in the first place . This suggests that communication failures help to drive change at synapses . When two neurons that are often active at the same time do not communicate efficiently , this failure triggers molecular changes that make future communication more reliable . Previous results have shown that synapses can change when glutamate release occurs . The current results show that they can also change when it does not . This means that the brain can continue to learn despite frequent communication failures between neurons . Many neurological disorders , including Alzheimer’s disease , show altered glutamate signalling at synapses . Padamsey et al . hope that a better understanding of this process will lead to new therapies for these disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Glutamate is required for depression but not potentiation of long-term presynaptic function
It is well established that inducible transcription is essential for the consolidation of salient experiences into long-term memory . However , whether inducible transcription relays information about the identity and affective attributes of the experience being encoded , has not been explored . To this end , we analyzed transcription induced by a variety of rewarding and aversive experiences , across multiple brain regions . Our results describe the existence of robust transcriptional signatures uniquely representing distinct experiences , enabling near-perfect decoding of recent experiences . Furthermore , experiences with shared attributes display commonalities in their transcriptional signatures , exemplified in the representation of valence , habituation and reinforcement . This study introduces the concept of a neural transcriptional code , which represents the encoding of experiences in the mouse brain . This code is comprised of distinct transcriptional signatures that correlate to attributes of the experiences that are being committed to long-term memory . Neuronal plasticity enables cognitive and behavioral flexibility underlying the development of adaptive behaviors ( Alberini , 2009; Alberini and Kandel , 2015 ) . This neuroplasticity , induced by salient experiences , has been shown to depend on the induction of temporally-defined waves of transcription ( Alberini , 2009; Alberini and Kandel , 2015; McClung and Nestler , 2008; Flavell and Greenberg , 2008; West and Greenberg , 2011 ) . The earliest of these waves consists of the expression of immediate-early genes ( IEGs ) . IEGs have been conventionally treated as molecular markers for labeling neuronal populations that undergo plastic changes during the formation of long-term memory ( Cruz et al . , 2013; Minatohara et al . , 2015 ) . However , the literature indicates a much more significant contribution of IEGs to synaptic plasticity and memory formation ( Lanahan and Worley , 1998; Okuno , 2011 ) . It has been proposed that IEG transcription may represent the molecular signatures of long-term plastic changes underlying the formation of memory ( Alberini , 2009 ) . Thus , induced IEG transcription could represent an experience-specific neural code for long-term storage of information . The existence of a neural code embedded in transcription implies that it should be possible to decode the identity of recent experiences , and potentially derive information regarding the nature of the experience , from its transcriptional representation ( Stanley , 2013 ) . To address the existence of a neural transcriptional code , we performed a detailed analysis of IEG transcription for 13 different experiences: cocaine ( acute , repeated and challenge ) , volitional sucrose drinking ( acute and repeated ) , reinstatement of feeding following food deprivation , lithium chloride administration ( LiCl; acute and repeated ) , saline ( acute injection without habituation , acute injection after habituation and repeated administration ) , acute administration of a mild foot shock , and exposure to a novel chamber with no foot shock . The experiences were selected to enable the identification of the transcriptional representations of affective attributes , such as salience and valence ( Russell , 1980; Posner et al . , 2005 ) . As such , we chose to investigate experiences that drive robust positive or negative reinforcement . Repetition of rewarding and aversive experiences provided insight into the transcriptional representation of habituation to negative stimuli and positive reinforcement of rewarding experiences . Experiences included in this study have been previously studied using electrophysiological measures , and plasticity has been observed within individual limbic and mesolimbic brain structures ( Russo and Nestler , 2013 ) . In contrast to classic electrophysiological measurements of plasticity , which focus on measurements with synapse specificity , transcriptional analysis enables parallel investigation of the representation of experience across multiple brain structures . Assuming that the encoding of complex reinforced experiences involves coordinated neural plasticity in multiple brain regions , we analyzed transcription across structures associated with the limbic and mesolimbic systems ( Russo and Nestler , 2013; Haber and Knutson , 2010 ) . The brain structures that were analyzed include limbic cortex ( LCtx; including medial prefrontal cortex and anterior cingulate cortex ) , nucleus accumbens ( NAc ) , dorsal striatum ( DS ) , amygdala ( Amy ) , lateral hypothalamus ( LH ) , dorsal hippocampus ( Hipp ) and ventral tegmental area ( VTA ) . Our results demonstrate that the transcriptional representations of experience are robust , reliable and consistent , enabling the decoding of the recent experience of mice with high levels of accuracy from a minimal transcriptional signature . We identify transcriptional hallmarks of affective attributes of experience , prominently demonstrated in the encoding of valence . Moreover , we report opposing patterns of transcriptional modulation underlying the development of habituation to experiences of negative valence , in comparison to repeated rewarding experiences associated with positive reinforcement . We conclude with a discussion of the potential implications of a neural transcriptional code . We initiated our study with the investigation of gene expression programs induced during the development of behavioral sensitization to cocaine . Cocaine sensitization is one of the most widely applied paradigms for studying mechanisms of neural plasticity , due to the robustness of the behavioral model and the detailed insight acquired into the underlying mechanisms ( McClung and Nestler , 2008; Robbins et al . , 2008; Hyman et al . , 2006; Nestler , 2002; Robison and Nestler , 2011; Lüscher , 2016; Piechota et al . , 2010 ) . Using this paradigm , we studied the transcriptional programs induced following acute or repeated exposure to cocaine , as well as re-exposure to cocaine after a period of abstinence from repeated drug exposures ( ‘cocaine challenge’ ) ( Figure 1A , B ) ( Robison and Nestler , 2011 ) . We analyzed the transcription induced at 0 , 1 , 2 , 4 hr following each of these cocaine experiences across six brain structures ( LCtx , NAc , DS , Amy , LH , and Hipp; Figure 1—figure supplement 1 ) . Transcription was analyzed with a comprehensive set of qPCR probes against putative IEGs ( see Materials and methods and Supplementary file 1 ) . Our results demonstrate that distinct cocaine experiences ( acute , repeated , challenge ) are characterized by robust induction of a handful of genes across the different brain structures studied , with peak induction at 1 hr following cocaine administration ( Figure 1; Figure 1—figure supplement 2; transcriptional dynamics shown in Figure 1—figure supplement 3 ) . 29 genes were induced above two fold in at least one of the six brain regions ( predominantly in LCtx , NAc and DS ) , across the three cocaine experiences . We were next interested in comparing the transcription programs induced by cocaine with those induced by an experience of opposite valence , within the same experimental context . For this purpose , we performed acute , as well as repeated , administration of the pharmacological compound LiCl , which induces malaise and decreases locomotion ( Figure 1D , E ) ( Fortin et al . , 2016 ) . Similar to cocaine , LiCl drove robust induction of a small subset of IEGs ( Figure 1F ) . In the case of LiCl experiences , 30 genes were induced above 2-fold , predominantly in the LCtx , Amy and LH ( Figure 1 , Figure 1—figure supplement 4; two doses of LiCl ( 150/250 mg/kg ) induced transcriptional responses of similar magnitude - Figure 1—figure supplement 4 ) . As a reference for the transcription induced by cocaine and LiCl experiences , we characterized the transcription induced by saline in control animals ( before and after habituation , as well as following repeated exposure; Figure 1—figure supplements 5 and 6 ) . Cocaine and LiCl experiences shared a common core set of 16 genes ( Arc , Atf3 , Cyr61 , Dusp1 , Egr2 , Egr4 , Elovl1 , Enpp6 , Fos , Fosb , JunB , Ier2 , Ier5 , Nr4a1 , Ngfr and Npas4 ) of which we selected five genes for further investigation . Marker gene selection was performed by ranking genes according to the frequency of their induction ( i . e . # of appearances above two-fold induction from 30 possible appearances across 6 structures in five experiences ) , as well as ranking in inverse order of the average variance ( S2 ) of the magnitude of induction . The five genes with combined highest ranking in frequency of appearance and lowest variance in expression were selected for further analysis ( Arc , Egr2 , Egr4 , Fos and Fosb; Figure 1G ) . To test our hypothesis that experiences can be decoded from patterns of induced transcription , mice were classified based on the induction of five genes ( Arc , Egr2 , Egr4 , Fos and Fosb ) across five brain structures ( LCtx , NAc , DS , Amy and LH ) , defining 25 gene-structure ‘features’ . Classification performed according to these 25 features using the k-Nearest Neighbors algorithm ( KNN ) allowed precise allocation of individual animals based on the identity of the recent experience with 97 . 3% accuracy , such that only one mouse ( out of 37 ) was incorrectly classified ( Figure 1H ) . Taken together , these results suggest that induced transcriptional signatures , defined by the combinatorial expression of minimal subsets of IEGs across brain structures , can be derived from comprehensive gene expression programs induced following an experience . Moreover , these minimal subsets are sufficient to decode the recent salient experience of mice . To further address the existence of a transcriptional code for experience , we expanded our study , including naturalistic volitional experiences of positive valence – sucrose consumption and reinstatement of feeding , as well as foot shock , an additional experience of negative valence . To provide a birds-eye view of the transcriptional landscape , we represent the experience-specific transcriptional signatures induced by each of these experiences using radar plots ( Figure 2 ) . This representation enables immediate identification of the major transcriptional attributes of each of the experiences . Four genes ( Arc , Egr2 , Egr4 and Fos ) are shown for simplicity of presentation; for individual mice , see Figure 2—figure supplement 1 . For full data , see Supplementary file 2 . This presentation further highlights the unique nature of the transcriptional signatures characterizing each experience , and the dynamic changes in IEG induction following repeated experience . Furthermore , commonalities in the transcriptional representation of experiences with shared affective attributes are visually apparent in this presentation . To investigate the transcriptional representation of negative valence , we focused on the aversive experiences induced either pharmacologically by LiCl administration , or by acute administration of mild foot shock . It is worth noting that while LiCl and foot shock are both characterized by negative valence , they are otherwise distinct; LiCl causes visceral discomfort and reduced locomotion ( Fortin et al . , 2016 ) , while foot shock induces acute pain and fear ( Bali and Jaggi , 2015 ) . Interestingly , exposure to the experimental context ( a 18 × 20 cm perspex chamber with a metal grid floor ) was by itself sufficient to induce IEG transcription across multiple structures in naïve mice ( ‘no shock’ control; Figure 2 ) . Mice that received a foot shock within this context displayed an indistinguishable pattern of transcriptional induction compared to their ‘no shock’ controls ( Figure 2—figure supplement 2 , Supplementary file 3 - T4 ) , with the sole distinction being a robust induction of transcription in the Amy ( primarily of Egr2 and Egr4; Figure 2 , Figure 2—figure supplement 2 , Statistics Supplementary file 3 – T4 , Row 4 – Columns B , C ) . This result demonstrates transcriptional coding of negative valence in the Amy , induced by the addition of a single variable ( foot shock ) to the experience of exposure to a novel environment . This observation was supported by the transcriptional representation of acute LiCl , which drove induction of Arc , Egr2 and Fos in the Amy ( Figure 2 , Figure 2—figure supplement 3 , Statistics Supplementary file 3 – T3 , Row 4 – Columns A , B , D ) . In contrast to the experiences of negative valence , the rewarding experiences of cocaine , sucrose and feeding had a broader representation across brain structures , which was most obvious in the case of feeding , where significant gene induction was observed across all structures studied ( Figure 2—figure supplements 4 and 5; Statistics Supplementary file 3 – T6 ) . The representation of acute cocaine was primarily observed in striatal regions ( DS and NAc ) and small but significant changes were also observed in additional mesocorticolimbic structures ( VTA , Amy , LCtx; Figure 2—figure supplement 3; Statistics Supplementary file 3 – T2 ) , while the representation of acute sucrose was quite minimal , and was reinforced upon additional exposure , as discussed below ( Figure 2—figure supplement 3; Statistics Supplementary file 3 – T5 ) . Repetition of aversive or rewarding experiences drove opposing trajectories of IEG induction ( Figure 2; Figure 2—figure supplement 3 ) . Following repeated exposure to LiCl , we observed a significantly diminished transcriptional representation in the Amy , to levels similar to those observed following repeated saline experience [interaction of treatment ( LiCl vs saline ) and time ( acute vs repeated ) ; Egr2: F ( 1 , 18 ) = 8 . 47 , p<0 . 01; Fos: F ( 1 , 20 ) = 17 . 2 , p=0 . 001 , Arc: F ( 1 , 20 ) = 8 . 72 , p<0 . 01] ( Statistics Supplementary file 3 – T3 – row 4 ) . In contrast , repeated exposure to cocaine administration was associated with enhanced transcriptional induction in the LCtx , DS , and VTA ( Statistics Supplementary file 3 – T2 – rows 1 , 3 , 7 ) . This enhancement was characterized by the significant induction of Egr2 in the LCtx and DS and Fos in the LCtx , DS , VTA [interaction of treatment ( cocaine vs saline ) and time ( acute vs repeated ) ; Egr2: LCtx F ( 1 , 29 ) = 6 . 43 , p<0 . 05; DS F ( 1 , 29 ) = 4 . 58 , p<0 . 05; Fos: LCtx F ( 1 , 29 ) = 5 . 35 , p<0 . 05; DS F ( 1 , 29 ) = 4 . 21 , p<0 . 05 , VTA F ( 1 , 13 ) = 14 . 3 , p<0 . 01] ( Statistics Supplementary file 3 – T2 – rows 1 , 3 , 7 columns B , D ) . However , in the NAc , the initially robust induction of Egr2 transcription following acute cocaine decreased after repeated administration ( interaction of treatment and time , Egr2: F ( 1 , 28 ) =39 . 7 , p<0 . 0001 ) ( Figure 2 , Figure 2—figure supplement 3; Statistics Supplementary file 3 – T2 – Column B row 2 ) . Repeated exposure to sugar was also represented by significantly enhanced transcription , most prominently in the LCtx [interaction of sucrose ( sucrose vs water ) and time ( acute vs repeated ) ; Egr2: F ( 1 , 26 ) = 5 . 02 , p<0 . 05 , Fos: F ( 1 , 26 ) = 7 . 51 , p=0 . 01; Arc: F ( 1 , 26 ) = 6 . 79 , p<0 . 05] ( Figure 2 , Figure 2—figure supplement 3; Statistics Supplementary file 3 – T5 – row 1 , columns A , B , D ) . Furthermore , reinstatement of feeding was also represented by significant induction of IEGs in the LCtx , specifically Egr2 and Fos ( Egr2: F ( 2 , 28 ) = 13 . 1 , p<0 . 0001; Fos: F ( 2 , 31 ) = 41 . 5 , p<0 . 0001 ) ( Figure 2 , Figure 2—figure supplements 4 and 5; Statistics Supplementary file 3 – T6 – row 1 , columns B , D ) . The experiences of repeated cocaine , repeated sucrose and reinstatement of feeding , though quite diverse in many affective and cognitive aspects , are all characterized by positive valence and therefore positive reinforcement . Our results suggest that a hallmark of increasing salience of positively reinforcing experiences may be increased transcriptional representation , specifically in the LCtx ( Robinson and Berridge , 2008 ) . This transcriptional representation of positively reinforcing experiences contrasts with the diminished transcriptional representation associated with habituation to anticipated and unavoidable aversive experiences . Finally , we tested our capacity to decode the recent experience of mice on the full complement of experiences studied in this project . The transcriptional induction of five genes ( Arc , Egr2 , Egr4 , Fos and Fosb ) across five structures ( LCtx , NAc , DS , Amy , LH ) forms 25 gene-structure 'features’ , which were used for the decoding with the KNN algorithm . We found that these 25 features supported the decoding of the recent experience of individual mice with 90 . 7% efficiency ( Figure 3A ) . Random shuffling of the association of mice to experiences demonstrated the reliability of the classifier , and the potential for our results to generalize beyond the given dataset ( p<1e−5; Figure 3B ) . These results suggest that obtaining a reliable transcriptional representation of a recent experience requires knowledge regarding both the transcriptional induction of several genes and the identity of structures within which they are induced . To further test this hypothesis , we ran a number of permutations . We tested the capacity to decode recent experiences following averaging the data for each gene across the five tested structures ( losing spatial information; Figure 3—figure supplement 1; classification accuracy 55% ) , as well as decoding by individual structures ( the expression of 5 genes in a single structure; Figure 3C; classification accuracies 33–56% ) or individual genes ( the expression of a single gene across five structures; Figure 3D; classification accuracies 37–70% ) . Taken together , while we find that measurement of the expression of individual genes , such as Fos and Egr2 , across the five brain structures can support classification ( 67% , 70% respectively ) , the prediction is significantly improved by the measurement of multiple features ( Figure 3A ) . With the objective of identifying the individual features that provide maximal support for decoding , we performed Random K-Nearest Neighbor ( R-KNN ) feature selection ( Figure 3—figure supplement 2A ) ( Li et al . , 2011 ) . We identified that a combination of eight features ( expression of Egr2 and Fos in the LCtx , NAc and Amy , and expression of Egr2 and Fosb in the DS ) provided the highest support , with a decoding efficiency of 93 . 6% ( Figure 3—figure supplement 2B , C ) . An independent approach for feature selection ( Breiman Random Forest [Breiman , 2001] ) identified a largely overlapping set of features , with the top 10 features supporting a classification accuracy of 94 . 4% ( p<1e−5; Figure 3—figure supplement 2E–H ) . An intuitive representation of the divergence of experiences based on particular features is provided by a decision tree ( one of a number of possible trees ) , in which mice were assigned to appropriate branches according to the extent of induction of a particular gene in a given structure ( Figure 3E ) . Taken together , these results establish that a minimal set of transcriptional markers form representative signatures of recent experience , enabling precise decoding of recent salient experiences at the resolution of individual mice . The brain creates representations of the world , encoding salient information for long-term storage to support the development of adaptive behaviors . In real time , the representation of information has been shown to be correlated with neural activity in distinct brain structures ( Bialek et al . , 1991 ) . Powerful demonstrations of the potential to decode sensory experiences and correlates of emotional state have been made in both rodents and humans from neural activation patterns using in-vivo electrophysiology , fMRI , and other imaging techniques ( Horikawa et al . , 2013; Santoro et al . , 2017; Kragel et al . , 2016; Lin et al . , 2005; Reber et al . , 2002 ) . In this study we demonstrate that multiplexed IEG expression data from multiple regions of the mouse brain enables the decoding of recent salient experiences with high precision . We show that beyond mere ‘activity markers’ for labeling neurons activated during an experience , IEG expression provides a quantitative and scalable metric , representing a neural transcriptional code for recent experience . This neural transcriptional code is defined by the combinatorial expression of marker transcripts across brain regions . Interestingly , we find components of induced transcriptional signatures that are associated with affective attributes of the experiences that are being encoded . Moreover , these IEG expression patterns are modulated following repeated administration of a stimulus of positive or negative value , suggesting a role for inducible transcription in sustaining long-term plasticity underlying the development of adaptive behavior . As this code is comprised of molecular components , it also provides a rich resource for biological insight into the processes underlying the long-term encoding of experience-dependent plasticity . Transcriptional markers have been successfully utilized for the classification of developmental stages ( Matcovitch-Natan et al . , 2016 ) , diseases ( Lamb , 2007; McKinney et al . , 2010 ) , and many other aspects of contemporary biomedical science ( Collins and Varmus , 2015 ) . Here we describe the utility of transcriptional markers for classification of salient experiences characterized by diverse affective properties . While the information embedded in the expression pattern of a single gene is not sufficient , a minimal subset of transcriptional markers enable the decoding of recent experience with high accuracy . Importantly , the principles we identify likely generalize to a broader set of experiences . Furthermore , it is likely that markers we utilize in our study could be substituted by other markers genes , providing similar classification accuracy . According to the Russell circumplex model ( Russell , 1980; Posner et al . , 2005 ) , affect can be defined in two dimensions – valence and salience . Valence has been suggested to be encoded in the Amy , PFC , NAc and VTA ( Namburi et al . , 2016 ) . Our results demonstrate that experiences of negative valence are represented by a distinct transcriptional induction in the Amy . In contrast , experiences of positive valence induce transcription in the LCtx , NAc , DS and VTA . Moreover , we report that upon repetition , the transcriptional representation within these structures is dynamically modulated , potentially underlying long-term adaptations following positive and negative reinforcement . Taken together , our results suggest that inducible transcription is a rich resource for the identification of brain regions that encode properties of an experience , providing biological insight into the molecular processes underlying experience-dependent plasticity . It should be noted that in this study we focused our analysis on structures associated with limbic and mesolimbic system . It is highly likely that transcriptional signatures across other brain areas ( as well as for other experiences ) would be related to different attributes of the experience , besides affect or valence . To explain how changes in transcription could affect future behavior , we introduce the concept of ‘predictive transcriptional coding’ . Predictive transcriptional coding frames inducible transcription not as a reporter of a recent event , but rather as encoding the valuation of the experience . This experience-dependent plasticity , mediated by transcription , sets the state of the network in the context of a particular experience , priming it for prospective network plasticity , and adjusting the response of the individual to the occurrence of a similar event in the future . This notion is conceptually similar to the ‘reward prediction error’ ( Schultz , 2010 ) , but is established on prolonged time scales . In this respect , transcription also serves as a ‘salience filter’ – defining whether an experience is significant enough to induce plasticity and worthy of encoding for long-term storage . Thus , the valuation of an experience that passes the ‘salience filter’ is encoded by the identity of the neural circuits recruited by the experience and the magnitude of transcription induced within them . A crucial question arising from this concept is: how is the threshold to commit to induction of transcription determined in neurons and neural networks ? One possibility , worthy of future investigation , was proposed in a landmark treatise , in which the analogy of a ‘genomic action potential’ was drawn for mechanisms underlying inducible transcription ( Clayton , 2000 ) . According to this hypothesis , the threshold for commitment to transcription depends on the coincidence of glutamatergic and neuromodulatory inputs . Our work provides a numerical definition of the imprint of recent experience , demonstrating a quantitative and predictive approach for the analysis of neural plasticity underlying adaptive behavior . Quantitative definitions of interoceptive states are expected to have implications for drug development - providing objective metrics for comprehensive characterization of the perception and valuation ascribed to an experience by individual subjects . For example , in the context of abuse liability , an objective quantitative interoceptive metric of the hedonic potential of a compound could increase standardization , reducing the reliance on variable behavioral outcomes . While there is substantial investment being made in the development of methodologies for transcriptional profiling with deeper coverage and increasing spatial resolution , our study demonstrates that fundamental phenomena can be identified by applying simple methods with low spatial resolution and coverage . Future work , applying tools of higher resolution , could build on our observations to address additional questions – such as the spatial distribution of neuronal ensembles recruited by experience and the identity of cell types recruited by distinct experiences . Approaches for non-invasive quantitative measurement of the encoding of experience can be envisioned , utilizing fluorescent markers of inducible transcription in combination with whole-brain imaging ( Eguchi and Yamaguchi , 2009 ) . New technologies are rapidly emerging for whole-brain analyses of transcription ( Renier et al . , 2016; Sylwestrak et al . , 2016; Ye et al . , 2016 ) , as are strategies for comprehensive profiling of single neurons ( Citri et al . , 2011; Lacar et al . , 2016 ) . These technological developments , together with the novel concept we develop here , are expected to provide the foundation for a new area of neuroscience research . This discipline , of ‘Behavioral Transcriptomics’ , will apply transcriptional analysis for investigation of intricate mechanisms of neural circuit plasticity underlying cognition . We propose that the approach of behavioral transcriptomics will provide a systems-level view of the encoding of experiences to long-term memory . One could speculate that different attributes of an experience may be mediated by activation of defined signaling pathways at different cellular locations , each inducing a component of the transcriptional program . If so , taken to its extreme , deciphering this transcriptional code will enable precise decoding of synapse-specific plasticity from quantitative analysis of inducible transcriptional markers . Male C57BL/6 mice aged 6–8 weeks ( Harlan Laboratories , Jerusalem , Israel ) served as subjects for the study . Mouse body mass ranged from 18 to 35 g , while between experimental groups in each repetition of experiments , the difference in body mass between animals did not exceed four grams . Four to five mice were housed per cage in all experiments except for sucrose consumption experiments , for which animals were single-housed . Mice were maintained in 12–12 hr light/dark cycle ( 0700 on/1900 off ) , in a temperature ( 20–22°C ) and humidity ( 55 ± 10% ) controlled facility . Mice received ad libitum access to water and food , with the exception of the experiment studying reinstatement of feeding , in which they were food deprived for 18 hr before reinstatement of feeding . Mice were randomly assigned to experimental groups and tested according to Latin square design . All tests were conducted during the light phase of the circadian cycle . Each experiment was performed at least twice , by independent researchers in the group , and provided similar results . All animal protocols were approved by the Institutional Animal Care and Use Committees at the Hebrew University of Jerusalem and were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals . A table defining the number of mice ( ‘n’ ) contributing to each experiment is included as Supplementary file 2 . Mice were acclimated to the animal facility for at least 2–5 days , followed by 3–4 days of experimenter handling , before the start of an experiment . Maintenance of uniform conditions across experiments and extensive handling were essential for reducing experimental variability , enabling the identification of a robust transcriptional response specifically induced by the experience being tested and minimal contamination from contextual background . Behavioral sensitization to cocaine . Mice were subjected to three days of intraperitoneal ( i . p . ) saline injections ( 250 microliter/injection ) , prior to exposure to cocaine ( 20 mg/kg freshly dissolved in physiological saline to 2 mg/ml and injected at a volume of 10 ml/kg; cocaine was obtained from the pharmacy at Hadassah Hospital , Jerusalem ) . The acute cocaine group received a single i . p . dose of cocaine , followed by analysis of locomotor behavior for 15 min in a video-monitored open-field arena . Animals were finally taken from their home cage and sacrificed at 1 , 2 and 4 hr following the cocaine injection . The repeated cocaine group received five consecutive daily injections of cocaine and were studied ( similar to the acute cocaine group ) following the fifth cocaine injection . The challenge cocaine group were treated as the repeated cocaine group , and then made abstinent from cocaine for 21–22 days , following which they were challenged with cocaine and re-exposed to the open-field arena . All responses were normalized to baseline controls ( time 0 ) , which were interleaved with their peer group , but were not treated on the day of the experiment . Additional reference groups included acute saline without habituation , which were habituated to the open-field arena for three days after a brief period of handling , and were sacrificed 1 hr following a single injection of saline . Responses in this group were normalized to controls ( time 0 ) , which were not exposed to any saline injections . The group of acute saline without habituation served as a reference for the habituation of the acute saline group , in which animals were treated identically to the acute cocaine group ( i . e . three consecutive days of habituation to saline injections in the open-field arena ) , but received a saline injection on the day of the experiment . Following each i . p . injection , mice were placed in an open-field arena for 20 min , during which locomotion was assayed between minutes 2 to 17 . LiCl exposure . All mice were habituated to injections of saline and locomotor monitoring in an open-field arena for three days preceding onset of the experiment . Animals were subjected to either acute or repeated administration of LiCl ( Sigma-Aldrich , St . Louis , MO , USA ) . In acute LiCl experiments , mice were administered with either a single dose of LiCl ( at 150 or 250 mg/kg ) or saline . In the experiments testing repeated LiCl , mice received LiCl ( 150 mg/kg ) for five consecutive days , and following a 48 hr break were re-exposed to LiCl or saline . Mice were divided into four groups: a ) Received saline injections for five days and were not exposed to an injection on the last day ( saline-0h ) , b ) Received LiCl injections for five days and were not exposed to an injection on the last day ( LiCl-0h ) , c ) Received saline injections for five days and were subjected to saline injection on the last day ( repeated saline ) , d ) Received LiCl injections for five days and were exposed to LiCl injection on the last day ( repeated LiCl ) . In all experiments , immediately following administration of LiCl or saline , mice were placed in video-monitored open-field arenas for 30 min . Reinstatement of feeding . Mice were food deprived for 18 hr before the experiment and then re-exposed to food for 1 , 2 or 4 hr before they were sacrificed . Control animals ( 0 hr ) were sacrificed immediately after the 18 hr food restriction . An additional reference group was allowed to continuously feed . Sucrose Consumption . Mice were single-housed for at least seven days before the experiment and habituated to the addition of a second water bottle in the cage for three days before the onset of the experiment . Acute exposure to sucrose was tested by habituating mice to the bottle with 10% sucrose overnight ( 16 hr ) , and 48 hr later , re-exposing the mice to a bottle with sucrose or water ( control ) for 1 hr . Repeated exposure to sucrose was tested by exposing mice to sucrose repeatedly for eight consecutive days , 2 hr each day ( 12:00-14:00 ) , and after a 48 hr break , re-exposed to sucrose or water ( control ) for 1 hr . Mice were sacrificed 1 hr following the exposure to sucrose . Sucrose and water intake were measured as a test for sucrose preference over water . Foot Shock . Following habituation to the experimental setup , the mice were placed in the experimental chamber ( 20 × 18 cm ) for three minutes , during which time , baseline freezing behavior was measured . At three minutes , each subject received three mild foot shocks ( 2 s , 0 . 7 mA ) separated by 30 s interval and post-shock freezing behavior was assessed immediately thereafter for 30 s before return to the home cage . Freezing , defined as a lack of movement other than respiration , was measured using Ethovision software ( Noldus , Wageningen , The Netherlands ) . Locomotor activity was assessed in sound- and light-attenuated open-field chambers . Mice were placed individually in a clear , dimly lit Plexiglas box ( 30 × 30 × 30 cm ) immediately after injection of cocaine , LiCl or saline . Activity was monitored with an overhead video camera for 20 or 30 min ( in cocaine sensitization and LiCl experiments respectively ) using Ethovision software ( Noldus , Wageningen , The Netherlands ) . Performed as previously described ( Turm et al . , 2014 ) . Mice were deeply anesthetized with Isoflurane ( Piramal Critical Care , Bethlehem , PA , USA ) and euthanized by cervical dislocation , followed by rapid decapitation and harvesting of brains into ice cold artificial cerebrospinal fluid ( ACSF ) solution ( 204 mM sucrose , 26 mM NaHCO3 , 10 mM glucose , 2 . 5 mM KCl , 1 mM NaH2PO4 , 4 mM MgSO4 and 1 mM CaCl2; all from Sigma-Aldrich , St . Louis , MO ) . Coronal slices ( 400 µm ) were cut on a vibrating microtome 7000 smz2 ( Camden Instruments , Loughborough , UK ) in ice-cold artificial cerebrospinal fluid ( ACSF ) . Brain regions [Limbic cortex ( LCtx ) , Nucleus Accumbens ( NAc ) , Dorsal Striatum ( DS ) , Amygdala ( Amy ) , Lateral Hypothalamus ( LH ) , Hippocampus ( Hipp ) and Ventral Tegmental Area ( VTA ) ] were dissected from relevant slices under a stereoscope ( Olympus , Shinjuku , Tokyo , Japan ) . Samples of LCtx , NAc , DS , Amy , LH AND Hipp were obtained from 2* 400 µm thick sections , while VTA , was obtained from 2* 200 µm thick sections ( Figure 1—figure supplement 1 ) . All of the steps were performed in strictly cold conditions ( ~4°C ) and care was taken to avoid warming of the tissue sections or the ASCF at all times . The tissue pieces were immediately submerged in Tri-Reagent ( Sigma-Aldrich , St . Louis , MO ) and stored at −80°C until processing for RNA extraction . The strategy for marker selection consisted of three steps . The initial list of candidate IEGs was compiled from a whole-genome microarray analysis of transcriptional dynamics induced by cocaine experiences in the nucleus accumbens ( Illumina MouseRef-8 v2 Expression BeadChip microarrays; data not shown ) , as well as a survey of literature and databases pertaining to IEG expression . qPCR primer probes were developed for 212 genes and primer efficiency was tested , resulting in selection of 152 optimal primer pairs . Differential expression of the shortlisted IEGs was then tested on samples from multiple brain structures , dissected from mice following cocaine and LiCl experiences , utilizing microfluidic qPCR arrays . Genes that displayed at least 1 . 25-fold induction in any measurement were shortlisted , resulting in a list of 78 genes . The next round of feature selection involved ranking genes based on their frequency of induction and variance . For ranking based on frequency of induction , we counted the number of times each gene was induced above a threshold of two-fold induction across the different brain structures ( LCtx , NAc , DS , Amy , LH and Hipp ) in the cocaine ( acute , repeated and challenge ) and LiCl ( acute and repeated ) conditions ( i . e . induction in six structuresfive experiences = #/30 ) . In addition , we ranked genes in inverse order of average variance ( S2 ) of their induction across structures . The five genes that were induced most consistently ( combined highest ranking in frequency and lowest in variance ) were selected for further investigation . The ranking of these genes was as follows: Arc ( #=22/30 , S2 = 2 . 9 ) , Egr2 ( #=21/30 , S2 = 2 . 8 ) , Egr4 ( #=18/30 , S2 = 1 . 53 ) , Fos ( #=14/30 , S2 = 0 . 43 ) , Fosb ( #=11 , S2 = 0 . 6 ) . Thus , criteria for marker selection were orthogonal to the tested hypothesis , supporting unbiased analysis . RNA extraction was performed strictly in cold RNase-free conditions . Tissue was homogenized using a 25G needle attached to a 1 ml syringe or using TissueLyser LT ( Qiagen , Redwood city , CA , USA ) . The homogenate was centrifuged at high speed ( 15 k g for 10 min ) and the supernatant was mixed with chloroform ( Bio-Lab , Jerusalem , Israel ) by vigorous shaking and centrifuged ( 15 k g for 15 min ) to separate the RNA from other nucleic acids and proteins . Isopropanol ( J . T . Baker , Center Valley , PA ) and glycogen ( Roche , Basel , Switzerland ) were added to the aqueous layer and samples were placed either at −20°C for 24 hr or at −80°C for 1 hr ( producing comparable results ) . The samples were centrifuged at high speed ( 15 k g for fifteen min ) for the precipitation of the RNA . The RNA was then washed in 75% ethanol ( J . T . Baker , Center Valley , PA ) by centrifugation ( 12 k g for five min ) , dried and dissolved in ultrapure RNase free water ( Biological Industries , Kibbutz Beit Haemek , Israel ) . RNA concentration was measured with a NanoDrop 2000c spectrophotometer ( Thermo , Wilmington , DE ) and random-primed cDNA was prepared from 100 to 300 ng of RNA , with use of a High Capacity cDNA Reverse Transcription Kit ( Applied Biosystems , Foster city , CA ) , following manufacturer guidelines . cDNA was processed for qPCR analysis using qPCR primer pairs ( IDTDNA , Coralville , IA ) and SYBR Green in a Light-cycler 480 Real Time PCR Instrument ( Roche Light Cycler480 SYBR Green I Master , Roche , Basel , Switzerland ) according to manufacturer guidelines . Relative levels of gene expression ( ΔCt ) were obtained by normalizing gene expression to a housekeeping gene ( GAPDH ) . Fold induction was calculated using the ΔΔCt method , normalizing experimental groups to the average of a relevant control group . Microfluidic qPCR , querying 96 samples against 96 sets of qPCR probes was performed utilizing Fluidigm Biomark Dynamic IFC ( integrated fluidic circuit ) Arrays ( Fluidigm Corp , South San Francisco , CA ) . Briefly , samples are subjected to targeted preamplification to enrich for specific gene products , which were then assayed with dynamic array fluidic microchips . Sample preparation was performed according to previously published protocols ( Turm et al . , 2014 ) . Targeted pre-amplification ( STA ) was achieved by mixing samples with a set of diluted primer pairs in TaqMan PreAmp Mastermix ( Applied Biosystems; Foster City , CA , USA ) followed by 10 min of denaturation at 95°C and 14 cycles of amplification ( cycles of 95°C for 15 s and 60°C for 4 min ) . Primers were then eliminated by use of ExoI exonuclease ( NEB; Ipswich , MA ) , placed in a thermal cycler at 37°C for 30 min and then at 80°C for 15 min . Samples were then loaded onto a primed dynamic array for qPCR in a specialized thermal cycler [Fluidigm Biomark; Thermal mixing: 70°C for 40 min , 60°C for 30 s , 95°C denaturation for 60 s , followed by 40 cycles of PCR ( 96°C for 5 s , 60°C for 20 s ) ] . For data analysis , a reference set of genes was identified , whose expression remained constant across all experimental conditions ( Dkk3 , Tagln3 , Gars , Scrn1 , Rpl36al , Mcfd2 , Psma7 and Hpcla4 ) . In order to reduce the potential for introduction of experimental error by normalization to a single gene , a 'global-normalization' Ct value was created for each sample from the average Ct values of the genes within the reference set . Fold induction was calculated using the ΔΔCt method , normalizing each gene in a sample to the global-normalization value ( ΔCt ) , followed by normalization of the experimental groups to the average of their relevant control group . All data are presented as mean ± standard error ( s . e . m . ) . Data were analyzed using one-way or two-way analysis of variance ( ANOVA ) , as appropriate . Tukey or Dunnett test was used for post hoc analyses of significant ANOVAs to correct for multiple comparisons . Differences were considered significant at the level of p<0 . 05 . Statistical analysis was performed , and bar graphs and line graphs were created , with Prism 6 . 0 ( GraphPad , San Diego , CA ) . Heat maps were created in MATLAB R2012a ( Mathworks , Natick , MA ) . Radar plots were created in Origin 6 . 0 ( Originlab , Northampton , MA ) . Codes were written in MATLAB R2015b ( MathWorks , Natick , MA ) and confusion matrices , randomization plots were created in Python using the Matplotlib library ( http://matplotlib . org ) . The analysis was performed on data obtained from 54 mice , each of which experienced one of the experiences ( acute , repeated or challenge cocaine , acute and repeated sucrose , reinstatement of feeding , acute and repeated LiCl and foot shock and no-shock controls exposed to the same environment ) . Each mouse was represented by a vector of twenty-five features [corresponding to the induction of five genes ( Arc , Egr2 , Egr4 , Fos and Fosb ) across five structures [limbic cortex ( LCtx ) , nucleus accumbens ( NAc ) , dorsal striatum ( DS ) , amygdala ( Amy ) and lateral hypothalamus ( LH ) ] . Each gene-structure combination was defined as a ‘feature’ . The classifier used was k-Nearest Neighbors ( KNN ) , with k = 1 over the Euclidean space , unless otherwise stated . This approach was selected based on the observation that the transcriptional response of mice within an experience group formed unique clusters . We evaluated the performance of our classification by a leave-one-out method . In this approach , we iterated over each sample in our training set and classification was performed given the rest of the training set . Visualization of the accuracy of classification was performed using a confusion matrix , which conveys both mean precision and mean recall of each condition classified . Feature selections were performed using Random k-Nearest Neighbors ( RKNN ) ( Li et al . , 2011 ) or Breiman Random Forest ( RF ) ( Breiman , 2001 ) algorithm . For RKNN , the contribution of each feature for classification of individual experiences was called support . We chose large ( n = 1e6 ) , random subsets of the twenty five available features in varying sizes ( between one and twenty five ) . For each such subset we trained a classifier . Each feature f appeared in some KNN classifiers , for example , set C ( f ) of size M , where M is the multiplicity of f . In turn , each classifier c ∈ C ( f ) is an evaluator of its m features . We defined the support of a feature f as the mean accuracy of all the classifiers in C ( f ) . Namely:support ( f ) =∑cϵC ( f ) accuracy ( c ) M To further examine the effect of feature set sizes on classification performance we evaluated the classification accuracy of different subset sizes in the following manner: for each case , we chose the n features which were ranked the highest in their support , and evaluated the KNN classifier trained with those features only . For classification using Random Forest ( RF ) , we used the Breiman random forest algorithm ( Breiman , 2001; ) , according to which a large number ( n = 1e5 ) of decision trees were built , where each tree used a varying number of features ( between 1 and 25 ) . Bifurcations were chosen according to modified Gini gain . For each feature , we averaged over the decrease in the Gini gain ( MDG ) ( Han et al . , 2016 ) over the ensemble of decision trees . The selected features were then evaluated using a regularized ( pruned ) decision tree with a maximal depth of 4 , using a k-cross validation process with k = 10 , with the constraint of a minimum categorization of 3 animals per group . The decision tree was constructed using the CART decision tree construction algorithm ( Breiman et al . , 1984 ) ( Figure 3—figure supplement 2E–H ) . To provide an example of a descriptive classifier , we created a decision tree using the CART algorithm with Information Gain ( Ben-David and Shalev-Shwartz , 2014 ) . No constraints were applied while building this tree . Considering the limited size of our dataset , we wanted to ensure that the classifier was not over fitted to our training set S . For this purpose , we produced a large number N ( N = 1e5 ) of permuted versions of our training set ( si , . . . sN ) , and created KNN or decision tree as the classifiers in the same way as for the original data . The permutation was performed by shuffling the association of individual mice with experiences . For each such permuted training set we trained a classifier and evaluated its classification accuracy ( leave-one-out , see previous description ) . We calculated the empirical p value ( p<1e−5 for both conditions ) for the classification accuracy on our original training set in the following manner:p−val=1N∑i=1N𝟙acc ( Si ) >acc ( S ) The data sets generated during the current study , as well as the code used for analysis have all been uploaded as supplementary material ( supplementary file 1–4 , source code 1–11 ) .	Can we tell what important event a mouse – or even a person – has recently experienced ? The current experience of an individual can be inferred from brain imaging experiments . However , along with changing brain activity , such an experience also switches on gene activity throughout the brain . This enables neurons to produce the proteins required to form a long-term memory of the experience . Do distinct , memorable experiences trigger unique signatures of gene activity ? To answer this question , Mukherjee , Ignatowska-Jankowska , Itskovits et al . exposed mice to a variety of experiences . Some were unpleasant and induced aversion; for example , the mouse may have felt nauseous or experienced brief pain and fear . Other experiences , such as when the mouse drank sugary water , received food or was injected with cocaine , were rewarding . Each of the experiences led to the activation of unique combinations of genes in different regions of the brain . Analysing a subset of the activated genes in various brain regions led to the identification of unique and reliable gene expression signatures of experience . These signatures allowed the recent experience of mice to be decoded with nearly 100% accuracy . While these unique signatures can distinguish between recent experiences , experiences that share common features do trigger overlapping patterns of gene activation . For example , negative experiences – but not positive or neutral ones – activated similar patterns of genes in a brain region called the amygdala . In contrast , repeated rewarding experiences induced a distinct gene activity pattern that was most pronounced as increased activity in part of the brain called the frontal cortex . These findings increase our understanding of how the brain represents information . The approach described in the paper provides a strategy to measure the changes in the brain that occur when information is encoded for long-term storage . This measure could also be useful during drug development , revealing how new drug compounds affect the brain , as well as providing an objective measure of the subjective experience of an individual . For example , substances that trigger similar patterns of gene activation to addictive drugs may themselves be addictive . On the other hand , substances that induce similar activity patterns to known medications could also have similar therapeutic properties .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Salient experiences are represented by unique transcriptional signatures in the mouse brain
During development , proliferation must be tightly controlled for organs to reach their appropriate size . While the Hippo signaling pathway plays a major role in organ growth control , how it senses and responds to increased cell density is still unclear . In this study , we use the zebrafish lateral line primordium ( LLP ) , a group of migrating epithelial cells that form sensory organs , to understand how tissue growth is controlled during organ formation . Loss of the cell junction-associated Motin protein Amotl2a leads to overproliferation and bigger LLP , affecting the final pattern of sensory organs . Amotl2a function in the LLP is mediated together by the Hippo pathway effector Yap1 and the Wnt/β-catenin effector Lef1 . Our results implicate for the first time the Hippo pathway in size regulation in the LL system . We further provide evidence that the Hippo/Motin interaction is essential to limit tissue size during development . Control of cell number is a critical process during the development of an organism . While proliferation is essential for organs to reach a correct size , failure to tightly regulate proliferation can lead to organ overgrowth and tumor formation . Proliferation must therefore be tightly controlled and coordinated with other developmental processes so that organs reach their proper final size , but do not exceed it . How this is achieved , however , is still largely unknown . In the past 15 years , the Hippo signaling pathway has been identified as a major regulator of organ size during development and homeostasis , by promoting cell death and differentiation and inhibiting proliferation . First identified in Drosophila ( Pan , 2007 ) , the Hippo signaling pathway is highly conserved in vertebrates ( Halder and Johnson , 2010; Pan , 2010 ) . When Hippo signaling is active , the Hippo pathway effectors YAP1 ( Yes-associated protein 1 ) and TAZ ( transcriptional co-activator with a PDZ domain ) , the vertebrate homologs of the Drosophila Yorkie , are phosphorylated by a cascade of kinases leading to their sequestration in the cytoplasm and/or their degradation . In contrast , when the Hippo signaling pathway is inactive , YAP/TAZ can translocate into the nucleus and mediate transcription of genes that promote proliferation and inhibit apoptosis ( Zhao et al . , 2011; Barry and Camargo , 2013; Yu and Guan , 2013; Bossuyt et al . , 2014 ) . Contact inhibition of proliferation was found to be largely mediated by the Hippo signaling pathway ( Zhao et al . , 2007 ) . Downstream of cell–cell adhesion and apicobasal polarity , many junction-associated proteins including E-cadherin , α-catenin , and proteins of the Crumbs and Par complexes promote Hippo signaling ( Kim et al . , 2011; Schlegelmilch et al . , 2011; Silvis et al . , 2011; Enderle and McNeill , 2013 ) . These proteins serve as scaffold for the Hippo pathway kinases MST1/2 and LATS1/2 leading to YAP/TAZ phosphorylation and retention in the cytoplasm or degradation ( Grusche et al . , 2010; Boggiano and Fehon , 2012; Irvine , 2012; Schroeder and Halder , 2012; Gumbiner and Kim , 2014 ) . Recent studies have shown that , in addition to changes in cell density , YAP/TAZ responds to changes in cell shape , tension forces , and substrate stiffness . This , however , seems largely independent of the canonical Hippo kinase cascade but depends on actin ( Dupont et al . , 2011; Wada et al . , 2011; Halder et al . , 2012; Aragona et al . , 2013 ) . The actin cytoskeleton indeed plays an important role in integrating and transmitting upstream signals to the Hippo pathway effectors YAP and TAZ ( Gaspar and Tapon , 2014 ) . Yet , how this is achieved is not well understood . Several recent reports suggest that the Motin family of junction-associated proteins could play a central role here . AMOT , AMOTL1 , and AMOTL2 are scaffold proteins associated with tight-junctions , required for tight junction integrity ( Bratt et al . , 2002; Sugihara-Mizuno et al . , 2007; Zheng et al . , 2009 ) and endothelial cell migration ( Troyanovsky et al . , 2001; Bratt et al . , 2005; Aase et al . , 2007; Wang et al . , 2011; Hultin et al . , 2014; Moleirinho et al . , 2014 ) . Recently , Motin proteins have further been shown to interact with YAP and TAZ via their PPxY motifs and the WW motifs of YAP and TAZ ( Chan et al . , 2011; Wang et al . , 2011; Zhao et al . , 2011; Hirate et al . , 2013; Hong , 2013; Lucci et al . , 2013; Yi et al . , 2013 ) . In most cases , this physical interaction leads to the inhibition of YAP/TAZ via cytoplasmic retention , similar to , but independent of the canonical Hippo pathway . Interestingly , there is a competition between YAP/TAZ and F-actin to bind to Motin proteins ( Mana-Capelli et al . , 2014 ) . Motin proteins have thus been proposed to mediate the response of YAP/TAZ to changes in the actin cytoskeleton downstream of mechanical signals . Yet , whether Motin proteins play such a central role in regulating organ growth in vivo in developing organisms is still largely unknown . The posterior lateral line ( pLL ) system in zebrafish provides an excellent model system to address this question . The pLL is a sensory system comprised of mechanosensory organs , the neuromasts , scattered on the surface of the body . The pLL primordium ( pLLP ) consists of about 100 progenitors that delaminate from a cranial placode and migrate posteriorly towards the tip of the tail ( Metcalfe et al . , 1985; Ghysen and Dambly-Chaudiere , 2004 ) . As the pLLP migrates , small groups of cells within its trailing region undergo cell shape changes to assemble into rosette-like structures , called proneuromasts . This assembly requires Fibroblast Growth Factor ( FGF ) signaling and the downstream effectors Shroom3 and Rock2a to activate non-muscle myosin and apical-constriction ( Lecaudey et al . , 2008; Nechiporuk and Raible , 2008; Ernst et al . , 2012; Harding and Nechiporuk , 2012 ) . Once assembled , proneuromasts are deposited behind the migrating pLLP and differentiate into functional neuromasts . Proneuromast deposition comes with a significant loss of cells for the migrating pLLP . Wnt/β-catenin signaling partially compensates for this loss by promoting proliferation in the leading region ( Gamba et al . , 2010; Aman et al . , 2011; McGraw et al . , 2011; Valdivia et al . , 2011; Matsuda et al . , 2013 ) . Hereafter , we use the term ‘leading region’ for the cells in the posterior part of the pLLP that do not assemble into rosettes . In contrast , with ‘trailing region’ , we refer to the part of the pLLP where cells change their shape to assemble into rosettes and where FGF signaling is active . Here , we use the pLLP to address the mechanisms required to control proliferation and tissue size during organ development . We focus on the role of the Motin protein Angiomotin-like 2a ( Amotl2a ) that has been reported to be expressed in the migrating pLLP , to be a target of FGF signaling ( Huang et al . , 2007 ) , to inhibit Wnt/β-catenin signaling during early zebrafish development ( Li et al . , 2012 ) , and is a potential candidate to interact with the Hippo signaling pathway . Loss of Amotl2a function results in a significant increase in pLLP size due to overproliferation . In a yeast two-hybrid ( Y2H ) screen , we identified the zebrafish Hippo pathway effectors Yap1 and Taz as strong interacting partners of Amotl2a . We show that Yap1 is required for the pLLP to reach its correct size and that reducing the level of Yap1 suppresses the overproliferation in amotl2a mutant pLLP . Finally we show that , in addition to Yap1 , the Wnt/β-catenin pathway effector Lef1 also mediates the hyperplasia phenotype of amotl2a mutants . This leads us to propose that Amotl2a , possibly by forming a ternary complex with Yap1 and β-catenin , limits proliferation in the trailing part of the pLLP . Here , we report the first mechanism that limits proliferation in the pLLP . In addition , we implicate the Hippo effector Yap1 in size control in the pLLP for the first time . Altogether , our results strongly suggest that Motin proteins play a major role in limiting organ growth during development by negatively regulating Yap/Taz and Wnt/β-catenin activities in vivo . amotl2a has been shown to be expressed in the pLLP ( Huang et al . , 2007 ) . By combining in situ hybridization ( ISH ) with Green Fluorescent Protein ( GFP ) immunostaining in cldnb:gfp transgenic embryos , we showed that amotl2a was expressed throughout migration ( Figure 1A–C ) in most of the pLLP with reduced or no expression in the most leading and trailing domains ( Figure 1D–I ) . The other motin genes amot , amotl1 , and amotl2b were not expressed in the pLLP ( Figure 1—figure supplement 1A–C ) . To determine the intracellular localization of Amotl2a in the pLLP , we fused Amotl2a to the red fluorescent protein TdTomato ( TdT ) . Amotl2a-TdT was present in the cytoplasm and was strongly concentrated at the most apical , constricted part of the cells in assembling proneuromasts ( Figure 1J–K′ ) . In addition , it appeared as an apical ring in deposited neuromasts ( Figure 1L , L′ ) . 10 . 7554/eLife . 08201 . 003Figure 1 . amotl2a is expressed in the pLLP and localizes at the cell apical side . ( A–I ) cldnb:gfp embryos stained with an amotl2a antisense RNA probe and an anti-GFP antibody ( E , G , I ) at the indicated stages . Red arrows indicate the posterior lateral line primordium ( pLLP ) . ( D–I ) Close-up views of the pLLP . ( J–L′ ) Maximum intensity projection ( MIP ) of Z-stacks of the pLLP ( J–J′ ) and a recently deposited neuromast ( L–L′ ) in cldnb:gfp embryos injected with amotl2a-TdT mRNA . ( K–K′ ) Close-up views of ( J–J′ ) . White arrows indicate rosette centers . Colors have been inverted . ( M–X ) 30 hpf cldnb:gfp embryos stained with an amotl2a in situ hybridization ( ISH ) probe and an anti-GFP antibody ( O , R , U , X ) in the indicated genetic background . The right column shows the primordium at higher magnification . In all figures , scale bars correspond to 200 μm for whole-embryo images and 20 μm in close up views of the pLLP . In all figures , n is the total number of embryos/primordia analysed and N is the number of biological replicates ( Figure 1—figure supplement 1 , Figure 1—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 00310 . 7554/eLife . 08201 . 004Figure 1—source data 1 . Relative length of the amotl2a-free ( Excel sheet 1 related to panel J ) and amotl2a-expressing domain ( Excel sheet related to panel K ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 00410 . 7554/eLife . 08201 . 005Figure 1—figure supplement 1 . Only amotl2a is expressed in the pLLP and its expression is controlled by FGF signaling . ( A–C ) ISH staining with probes against amot ( A ) , amotl1 ( B ) , and amotl2b ( C ) in cldnb:gfp embryos . ( D–I ) ISH staining with an amotl2a probe in Tg ( hsp70l:dnfgfr1EGFP ) pdl embryos non-heat-shocked ( D–F ) or heat-shocked ( G–I ) . The red arrow points to the pLLP . The dotted lines indicate the pLLP contour . ( J , K ) Boxplots showing the comparison of the relative length of the amotl2a-free domain ( J ) and of the amotl2a-expressing domain ( K ) between the indicated groups ( Figure 1—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 005 We next asked if amotl2a expression in the pLLP also required FGF signaling as it has been reported during early development ( Huang et al . , 2007 ) . amotl2a expression was specifically lost in the pLLP of fgf3−/−;fgf10−/− double mutants ( Figure 1M–R ) , upon expression of a dominant negative FGFR1 receptor ( Figure 1—figure supplement 1D–I ) , and in embryos treated with the FGFR inhibitor SU5402 ( not shown ) . Conversely , ectopic activation of FGF signaling using the Tg ( hsp70l:fgf3-Myc ) zf115 transgenic line led to an expansion of the amotl2a expression domain in the pLLP as compared to controls ( Figure 1S–X and Figure 1—figure supplement 1J , K ) . These results indicated that FGF signaling is necessary and sufficient for amotl2a expression in the pLLP . To investigate the function of Amotl2a in the pLLP , we used a previously published translation-blocking morpholino ( Amotl2aMo ) ( Huang et al . , 2007; Wang et al . , 2011; Li et al . , 2012 ) . At low doses , Amotl2aMo did not lead to any obvious morphological defects ( Figure 2—figure supplement 1A–D ) ( Wang et al . , 2011 ) and efficiently blocked the translation of a fusion between the ATG region of amotl2a and GFP ( MoBS_Amotl2a-GFP , Figure 2—figure supplement 1E , F ) . Since Motin proteins are necessary for tight junction integrity in several contexts ( Bratt et al . , 2002; Sugihara-Mizuno et al . , 2007; Zheng et al . , 2009 ) , we performed ZO1 and phalloidin staining to label tight junctions and actin , respectively . While mature rosettes had similarly strong apical actin and ZO1 staining in morphant and control embryos ( white arrows in Figure 2C–F ) , the distance between the tip of the pLLP and the most recent fully assembled rosette ( yellow arrows ) , appeared increased in amotl2a morphants ( double arrowhead in Figure 2A , B ) . Yet , neuromasts were deposited normally ( Figure 2G , H ) . This result suggested that rosette assembly was slightly delayed in amotl2a morphants but that Amotl2a was neither essential for tight junction assembly nor for neuromast formation and deposition . 10 . 7554/eLife . 08201 . 006Figure 2 . Amotl2a is not essential for proneuromast assembly but for proper migration . ( A–F ) MIP of Z-stacks of pLLP stained with ZO-1 ( blue ) and GFP ( green ) antibodies and phalloidin ( red ) in control ( A , C , E ) and amotl2a morphant ( B , D , F ) cldnb:gfp embryos at 30 hpf . ( G , H ) MIP of overview images of control ( G ) and amotl2a morphant ( H ) embryos after completion of migration . ( I , J ) Snapshots of time-lapse movies at the indicated timepoints showing a delay in migration in amotl2a morphants ( J ) as compared to controls ( I ) . ( K , L ) Corresponding kymographs used to measure the migration speed . ( M ) Boxplot comparing the migration speeds ( Figure 2—source data 1 , Figure 2—figure supplements 1 , 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 00610 . 7554/eLife . 08201 . 007Figure 2—source data 1 . Migration speed in amotl2a morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 00710 . 7554/eLife . 08201 . 008Figure 2—figure supplement 1 . Amotl2aMo efficiency . ( A–D ) Overview pictures of 30 hpf cldnb:gfp embryos uninjected ( A , B ) or injected with Amotl2aMo ( C , D ) , imaged either with fluorescent ( A–C ) or transmitted light ( B–D ) . ( E , F ) cldnb:gfp embryos injected with RNA encoding the Mo-binding region of Amotl2a fused to GFP ( MoBS-amotl2a-gfp ) either alone ( E ) or with Amotl2aMo ( F ) . The bright green fluorescence in F comes from the expression of the cldnb:gfp transgene in the brain and eyes . The red arrowheads in A and C indicate the level of migration at which embryos were fixed for cell counting quantification ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 00810 . 7554/eLife . 08201 . 009Figure 2—figure supplement 2 . cxcr4b and cxcr7b expression are not affected in amotl2a morphants . Uninjected ( A–C and J–L ) , p53Mo-injected ( D–F and M–O ) , or Amotl2aMo-injected ( G–I and P–R ) cldnb:gfp embryos stained with a cxcr4b ( A–I ) or a cxcr7b ( J–R ) ISH probe and an anti-GFP antibody ( C , F , I , L , O , R ) . Red arrows point to the pLLP . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 009 In addition , the pLLP of amotl2a morphant embryos migrated 35% slower than controls ( Figure 2I–M ) . Yet , all primordia reached the tip of the tail ( Figure 2G , H ) . Expression of the two G-protein-coupled receptors cxcr4b and cxcr7b was not detectably affected in morphants ( Figure 2—figure supplement 2 ) , suggesting that a deregulation of their expression is unlikely to account for the decrease in migration speed . Altogether , these results suggested that while rosette assembly and migration are delayed in amotl2a morphants , the pLLP deposits neuromasts and eventually migrates to the tip of the tail . Interestingly , the size of amotl2a morphant pLLP appeared obviously increased throughout the migration process ( Figure 3A–C , also Figure 2I , J ) . To determine whether this was due to an increase in cell number , we developed an algorithm to automatically count cells in the pLLP based on the membrane labeling in cldnb:gfp embryos ( see ‘Materials and methods’ and Figure 3—figure supplement 1 ) . Cell counts were done in pLLP that had reached the middle of the yolk extension in both controls and amotl2a morphants ( thereafter referred to as ‘mid-migration’ , red arrowhead in Figure 2—figure supplement 1 , see also ‘Materials and methods’ ) . Cell counts indicated that the number of cells in amotl2a morphant pLLP was increased by 35–38% as compared to either uninjected or p53Mo-injected embryos ( Figure 3E , p = 1 . 58E-40 and p = 9 . 53E-31 , respectively ) . To confirm the specificity of the morphant phenotype , we co-injected Amotl2aMo with amotl2am , an RNA that was insensitive to the morpholino . Co-injection of Amotl2aMo with amotl2am RNA partially rescued the morphant phenotype ( Figure 3D , F ) . While overexpression of amotl2a at the concentration used for the rescue experiment did not significantly reduce the number of cells in the pLLP ( Figure 3F ) , injection at a higher concentration induced a moderate , but significant decrease in cell number ( Figure 3—figure supplement 2 , −15% , p = 5 . 20E-07 ) . Altogether , these results indicated that Amotl2a was essential to limit the number of cells in the migrating pLLP . 10 . 7554/eLife . 08201 . 010Figure 3 . Amotl2a is required to limit proliferation in the pLLP . ( A–D ) MIP of Z-stacks of pLLP in cldnb:gfp embryos injected as indicated . ( E , F ) Boxplots showing the number of cells in the primordia of indicated groups , normalized to the control group . ( G–I′′ ) MIP of Z-stacks of pLLP in cldnb:gfp embryos stained with EdU and DAPI showing the green ( membranes , middle ) , red ( EdU , right ) , and merge ( left ) channels . ( J , K ) Boxplot showing the comparison of the EdU index in the indicated experimental conditions in whole primordia ( J ) or separately in the leading and trailing region ( K ) ( Figure 3—source data 1–3 , Figure 3—figure supplements 1 , 2 , Figure 3—source data 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 01010 . 7554/eLife . 08201 . 011Figure 3—source data 1 . Cell counts in amotl2a morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 01110 . 7554/eLife . 08201 . 012Figure 3—source data 2 . Cell counts in rescue experiment of amotl2a morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 01210 . 7554/eLife . 08201 . 013Figure 3—source data 3 . EdU ratio in the entire pLLP ( Excel sheet 1 related to panel J ) , leading region and trailing region ( Excel sheet 2 related to panel K ) of amotl2a morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 01310 . 7554/eLife . 08201 . 014Figure 3—source data 4 . Cell counts in amotl2a-overexpressing embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 01410 . 7554/eLife . 08201 . 015Figure 3—figure supplement 1 . Successive steps of the automated ‘cell-counting’ algorithm . ( A ) Raw image of pLLP in a cldnb:gfp embryo . ( B ) Enhanced , filtered image using anisotropic diffusion . ( C ) Segmentation of individual cells . ( D ) Segmentation mask of the pLLP . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 01510 . 7554/eLife . 08201 . 016Figure 3—figure supplement 2 . Overexpression of amotl2a leads to reduced cell number in the pLLP . ( A , B ) MIP of Z-stacks of pLLP in cldnb:gfp embryos injected as indicated . ( C ) Corresponding boxplot comparing the number of cells in the indicated experimental conditions . ( Figure 3—source data 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 016 To determine the cause of the increase in cell counts in the pLLP of amotl2a morphants , we quantified proliferation rates using EdU to label proliferating cells ( Figure 3G–I′′ ) . There was a significant increase in the EdU index of about 15% in the pLLP of amotl2a morphants as compared to controls ( Figure 3J , p < 0 . 05 ) . Since amotl2a was not expressed in the most leading part of the pLLP , we determined the proliferation rates separately in the leading vs trailing region of the primordium . While there was no difference in the proliferation index in the leading region between morphants and controls , there was a 34% increase in the trailing region ( Figure 3K , p < 0 . 0001 ) . Altogether , these results indicated that Amotl2a is required to restrict the number of cells in the pLLP by limiting proliferation in the region where rosettes assemble . To confirm the specificity of the Amotl2Mo-induced phenotype , we generated amotl2a mutants using the transcription activator-like effector nuclease ( TALEN ) technique . Among several identified amotl2a alleles , two were further analyzed that introduced a frame-shift and a premature STOP codon in the third exon ( Figure 4A ) . The putative resulting truncated proteins would lack most of the coiled-coil domain and the PDZ-binding domain ( Figure 4B ) . Both alleles had identical phenotypes . amotl2a−/− homozygous mutants were morphologically indistinguishable from their siblings ( Figure 4C–F ) but had larger primordia ( Figure 4G , H ) . Cell count analyses revealed a significant increase in cell numbers of 21% in amotl2a−/− mutants as compared to their siblings ( Figure 4I ) . In addition and as in morphants , pLLP migrated slower in amotl2a−/− mutants as compared to controls ( −32% , p < 0 . 0001 , Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 08201 . 017Figure 4 . amotl2a mutants phenocopy the morphant phenotype . ( A ) Scheme showing the transcription activator-like effector nuclease ( TALEN ) target site in the amotl2a locus with the left and right TALEN-binding sites in red separated by the spacer including the restriction site used for screening ( blue ) ( top ) . Alignment of the two conserved amotl2a mutant alleles with the corresponding wild-type sequence showing the deleted nucleotides ( bottom ) . ( B ) Scheme comparing the functional domains present in the wild-type Amotl2a protein ( 721aa long ) and the putative truncated proteins ( 272aa +17 or +9 missense aa for allele fu45 and fu46 , respectively ) . ( C–F ) 36 hpf cldnb:gfp wild-type sibling ( C , D ) or amotl2a−/− mutant embryo ( E , F ) imaged with fluorescent ( C , E ) or transmitted light ( D , F ) . ( G , H , J , K ) MIP of Z-stacks of pLLP in cldnb:gfp embryos with the indicated genetic background . ( I , L ) Boxplots comparing the cell numbers between the indicated genetic backgrounds . ( M , N ) MZamotl2a−/− mutant embryo ( N ) showing an extra deposited neuromast as compared to a wild-type sibling embryo ( M ) . ( O ) Boxplot showing the corresponding quantification ( Figure 4—source data 1 , 2; Figure 4—figure supplements 1 , 2 , Figure 4—source data 3 , 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 01710 . 7554/eLife . 08201 . 018Figure 4—source data 1 . Cell counts in zygotic amotl2a mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 01810 . 7554/eLife . 08201 . 019Figure 4—source data 2 . Number of deposited neuromasts in MZamotl2a mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 01910 . 7554/eLife . 08201 . 020Figure 4—source data 3 . Migration speed in MZamotl2a mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 02010 . 7554/eLife . 08201 . 021Figure 4—source data 4 . Cell counts in neuromasts of morphants ( Excel sheet 1 related to panel A ) and MZamotl2a mutants ( Excel sheet 2 related to panel B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 02110 . 7554/eLife . 08201 . 022Figure 4—figure supplement 1 . MZamotl2a−/− mutants exhibit reduced pLLP migration speed . ( A , B ) Snapshots of time-lapse movies at the indicated timepoint showing a delay in migration speed in MZamotl2a−/− mutants ( B ) as compared to siblings ( A ) . ( C , D ) Corresponding kymographs . ( E ) Boxplot comparing the migration speeds in control vs MZamotl2a−/− mutant pLLP ( Figure 4—source data 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 02210 . 7554/eLife . 08201 . 023Figure 4—figure supplement 2 . Deposited neuromasts or interneuromast chains do not contain more cells in amotl2a morphants or mutants . ( A ) Boxplots showing cell number quantifications in the first 3 neuromasts ( n1 to n3 ) and interneuromast regions ( i1 and i2 ) at 48 hpf in the indicated experimental conditions . ( B ) Boxplot showing cell number quantification in one neuromast close to the end of yolk extension ( n3 or n4 ) at 72 hpf in the indicated experimental conditions ( Figure 4—source data 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 023 Homozygous mutants were viable , fertile , and generated maternal-zygotic ( MZ ) mutants that were also morphologically normal , viable , and fertile . Since amotl2a is expressed maternally ( Huang et al . , 2007 ) , we wanted to determine whether maternal RNA and/or proteins also contributed to limit the number of cells in the pLLP . pLLP cell counts were further increased to 40% in MZamotl2a−/− mutants as compared to controls ( Figure 4J–L , p = 1 . 18E-21 ) , an increase similar to that quantified in amotl2a morphants ( Figure 3E , F ) . We then asked where these additional cells ended up later in development . Neither deposited neuromasts nor interneuromast chains comprised more cells at 48 hpf in morphants as compared to controls ( Figure 4—figure supplement 2A ) . Similarly , MZamotl2a−/− mutant neuromasts also contained the same number of cells as controls at 72 hpf ( Figure 4—figure supplement 2B ) . Yet , there was on average one additional neuromast deposited in morphants ( not shown ) and mutants ( Figure 4M–O ) as compared to control embryos . Altogether , these results confirmed that Amotl2a is required to limit the number of cells in the pLLP and thus its size . Also , loss of Amotl2a does not affect the size of mature neuromasts but increases their number . To dissect the mechanisms by which Amotl2a was limiting the number of cells in the pLLP , we performed a Y2H screen to identify Amotl2a-interacting partners . The screen was performed with a full-length ( FL ) cDNA of amotl2a as bait and a cDNA library of 18–20 hpf zebrafish embryos as prey library ( Hybrigenics , France ) . 25 proteins were identified as potential Amotl2a interaction-partners and were given a score ranging from A ( very high confidence in the interaction ) to D ( moderate confidence in the interaction ) . The Hippo pathway effectors Yap1 and Taz were among the 10 strongest identified partners with an A and B score , respectively . Further analysis showed that in both cases , the amino acid ( aa ) sequence shared by all prey fragments ( Selected Interaction Domain ) included the evolutionarily conserved WW motifs ( 2 for Yap1 and 1 for Taz , Figure 5A ) . 10 . 7554/eLife . 08201 . 024Figure 5 . Yap1 physically interacts with Amotl2a and is required for the pLLP to have the correct number of cells . ( A ) Part of the yeast two-hybrid ( Y2H ) screen results showing the Selected Interaction Domain ( SID ) of Yap1 and Taz ( encoded by the gene wwtr1 ) and known functional and structural domain on these proteins . The SID is the amino acid sequence shared by all prey fragments matching the same reference protein , here Yap1 and Taz . ( B ) Y2H assay with Histidine ( left panel , growth control ) and without Histidine ( right panel , protein interaction assay ) showing the interactions between zebrafish Amotl2a and Yap1 or Taz ( red ) , but not with the corresponding proteins mutated in the known interaction motifs: LPTY/PPEY for Amotl2a ( green ) and WW domain for Yap1 and Taz ( pink ) . ( C , D ) MIP of Z-stacks of pLLP in cldnb:gfp embryos injected as indicated . ( E ) Boxplot comparing the number of cells in the pLLP in Yap1Mo-injected embryos and controls . ( F ) Scheme showing the TALEN target site in the yap1 locus with the left and right TALEN-binding sites in red separated by the spacer including the restriction site used for screening ( blue ) ( top ) . Alignment of the two yap1 mutant alleles with the corresponding wild-type sequence showing the deleted nucleotides ( bottom ) . ( G ) Scheme comparing the functional domains present in the wild-type Yap1 ( 442aa long ) and the putative truncated proteins ( 76aa +25 or +45 missense aa before for allele fu47 and fu48 , respectively ) . ( H , I ) Overview pictures of 36 hpf cldnb:gfp control and Zyap1−/− embryos . ( J–M ) MIP of Z-stacks of pLLP in cldnb:gfp embryos with the indicated genotype . ( N–P ) Boxplot showing cell counts in the pLLP in embryos of the corresponding genotypes ( Figure 5—source data 1 , 2; Figure 5—figure supplement 1; Figure 5—source data 3 , 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 02410 . 7554/eLife . 08201 . 025Figure 5—source data 1 . Cell counts in yap1 morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 02510 . 7554/eLife . 08201 . 026Figure 5—source data 2 . Cell counts in MZyap1 mutants at 24 hpf ( Excel sheet 1 related to panel N ) , 30 hpf ( Excel sheet 2 related to panel O ) and 36 hpf ( excel sheet 3 related to panel P ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 02610 . 7554/eLife . 08201 . 027Figure 5—source data 3 . Cell counts in Zyap1 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 02710 . 7554/eLife . 08201 . 028Figure 5—source data 4 . Number of deposited neuromasts ( Excel sheet 1 related to panel D ) , number of cells ( Excel sheet 2 related to panel E ) , and number of hair cells ( Excel sheet 3 related to panel H ) per neuromasts in MZyap1 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 02810 . 7554/eLife . 08201 . 029Figure 5—figure supplement 1 . yap1 and taz are ubiquitously expressed at 30 hpf . Zyap1−/− mutant embryos have fewer cells compared to sibling controls . ( A–B′′ ) 30 hpf cldnb:gfp embryos stained with a yap1 ( A–A′′ ) or a taz ( B–B′′ ) ISH probe and an anti-GFP antibody ( A′′ , B′′ ) . ( C–E , H ) Boxplot showing pLLP cell counts ( C ) , number of deposited neuromasts ( D ) , cell counts ( E ) , and number of hair cells ( H ) per neuromast in the indicated experimental conditions . Cell counts were performed in the neuromast closest to the end of yolk extension ( n3 or n4 ) at 72 hpf in ( E–H ) . ( F–G′′ ) MIP of Z-stacks of a deposited neuromast ( n3 or n4 ) stained with an anti-HCS1 antibody ( hair cells , red ) and DAPI ( cell nuclei , blue ) . ( I–L′ ) ISH with a prox1 probe and an anti-GFP antibody staining ( J′ , L′ ) on embryos with the indicated genetic background . Red arrows point to the pLLP ( A , B , I and K ) ( Figure 5—source data 3 , 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 029 To further test if the WW motifs were required for the physical interactions , we mutated both and tested the interactions with Amotl2a in a Y2H assay . The Y2H assay first confirmed the interaction between zebrafish Amotl2a and zebrafish Yap1 and Taz ( red in Figure 5B ) . Furthermore , mutations of the WW motifs of Yap1 and Taz abolished the corresponding interaction with Amotl2a ( pink in Figure 5B ) . Conversely , mutations of the LPTY and PPEY motifs of Amotl2a , but not of either of them alone , abolished the interaction with WT Yap1 and Taz ( green and yellow , respectively , in Figure 5B ) . This result indicated that , similar to their human orthologs , zebrafish Yap1 and Taz physically interact with zebrafish Amotl2a and that this interaction is mediated by the WW motifs of Yap1/Taz and the LPTY/PPEY motifs of Amotl2a . We therefore hypothesized that the increase in proliferation upon loss of Amotl2a could result from an upregulation of Yap1/Taz activity . To test whether an upregulation of Yap1/Taz activity could account for the increased cell number in the pLLP upon loss of Amotl2a function , we first assessed the expression of zebrafish yap1 and taz . Both genes were broadly expressed during development including in the pLLP ( Figure 5—figure supplement 1A , B ) . We then asked whether Yap1 or Taz was implicated in controlling pLLP cell number using morpholino-based knockdown . Injection of a previously published translation-blocking Mo against Taz ( Hong et al . , 2005 ) did not cause obvious differences in pLLP size ( data not shown ) . In contrast , embryos injected with a previously published , splice-blocking Yap1 morpholino ( Yap1Mo ) ( Skouloudaki et al . , 2009; Fukui et al . , 2014 ) were morphologically normal but had smaller pLLP than controls and often a rounder shape ( Figure 5C , D ) . Cell counts indicated that the pLLP of yap1 morphant embryos had about 15% less cells than uninjected controls ( p = 2 . 65E-12 ) ( Figure 5E ) . Yet , they completed migration and deposited neuromasts . In order to confirm this essential role of Yap1 in controlling cell number in the pLLP , we generated yap1 mutants using TALEN-mediated mutagenesis . Like for amotl2a , we further analyzed two yap1 alleles with a frame shift leading to a premature STOP codon in the second exon ( Figure 5F ) . The putative resulting truncated proteins would lack most of their functional domains including the Transcriptional Enhancer Activator Domain ( TEAD ) binding domain , the 2 WW motifs and the PDZ-binding domains ( Figure 5G ) . Both alleles had identical phenotypes . yap1−/− homozygous mutants embryos were morphologically normal ( Figure 5H , I ) with smaller and at times rounder pLLP ( Figure 5J , K ) , a phenotype identical to that of the yap1 morphants . Cell counts confirmed that yap1−/− pLLP also showed a decrease in cell number of 19% ( p = 1 . 25E-05 ) ( Figure 5—figure supplement 1C ) . Interestingly , these embryos again developed into normal looking and fertile adult fish . MZyap1−/− mutants also developed normally and were morphologically indistinguishable from controls . To determine at which stage Yap1 activity was required in the pLLP , we performed cell counts in MZyap1−/− mutants when the pLLP starts to migrate ( 24 hpf ) , when it reaches the middle of the yolk extension ( 30 hpf ) ( Figure 5L , M ) , and at the end of the yolk extension ( 36 hpf ) . At 24 hpf , the pLLP of MZyap1−/− mutants was already 15% smaller than related controls ( p = 4 . 12E-08 ) . This difference continued to increase with time to reach 21% at 30 hpf ( p = 2 . 15E-10 ) and 27% at 36 hpf ( p = 2 . 41E-14 ) ( Figure 5N–P ) . These results showed that Yap1 is required both before and during migration to establish the correct number of cells in the pLLP . A recent study reported that yap1 morphants show a reduced number of neuromasts ( Loh et al . , 2014 ) . Given the reduced number of cells in the pLLP of yap1 morphants and mutants , we expected a similar result . Intriguingly , neither our morphants ( not shown ) nor our MZyap1−/− mutants showed a decreased number of neuromasts ( Figure 5—figure supplement 1D ) . Furthermore , these neuromasts did not contain fewer cells ( Figure 5—figure supplement 1E ) . These results strongly suggested that compensatory mechanisms kick-in in the pLLP of yap1−/− mutants to allow them to migrate to the tip of the tail while depositing a normal number of neuromasts consisting of a normal number of cells despite their smaller size . It was also reported that neuromasts of yap1 morphants contain fewer hair cells due to loss of prox1 expression in these embryos ( Li et al . , 2012 ) . Intriguingly , neither the number of hair cells per neuromast ( Figure 5—figure supplement 1F–H ) nor the expression of prox1 in the pLLP ( Figure 5—figure supplement 1I–L′ ) was affected in our MZyap1−/− mutants . These discrepancies are discussed below . Altogether , our results indicated that Yap1 activity is required for the pLLP to reach its normal cell number and size . This further supported the idea that the increase in proliferation upon loss of Amotl2a activity could be due to an increase of Yap1 activity . We reasoned that if an increase of Yap1 activity was the cause of the hyperplasia in the pLLP of amotl2a morphants and mutants , we should be able to suppress this phenotype by reducing Yap1 levels . The pLLP size of embryos co-injected with Yap1Mo and Amotl2aMo was indeed smaller than that of amotl2a morphants alone and was comparable to that of control embryos ( Figure 6A–D ) . Automated cell counts revealed that the number of cells in the pLLP of amotl2a;yap1 double morphants was significantly reduced as compared to amotl2a morphants ( 34% reduction , p = 1 . 53E-16 ) and was not significantly different from controls ( p = 0 . 098 ) ( Figure 6E ) . To confirm this result , we generated MZamotl2a−/−;MZyap1−/− double mutants . These mutants were morphologically indistinguishable from related controls . Here also , the increase in pLLP size in MZamotl2a−/− was suppressed in the MZ double mutant embryos ( Figure 6—figure supplement 1A–E ) . These observations were confirmed by automated cell counts: the 37% increase in MZamotl2a−/− pLLP ( p = 4 . 62E-11 ) was suppressed in MZamotl2a−/−;MZyap1−/− double mutants and comparable to controls ( p = 0 . 06 ) . MZyap1−/− mutants were , however , still lower than double mutants ( −27% , p = 1 . 09E-11 ) ( Figure 6—figure supplement 1E ) . Together , these results demonstrated that the increased pLLP cell counts that resulted from the loss of Amotl2a function could be suppressed by an additional loss of Yap1 activity . This strongly suggested that the overproliferation in amotl2a−/− mutants pLLP was in part mediated by Yap1 . 10 . 7554/eLife . 08201 . 030Figure 6 . Loss of Yap1 suppresses the increased cell proliferation in amotl2a morphant . ( A–D , L–O ) MIP of Z-stacks of pLLP in cldnb:gfp embryos injected as indicated . ( E , P ) Corresponding boxplots comparing pLLP cell counts . ( F–J′′ ) MIP of Z-stacks of pLLP in cldnb:gfp embryos stained with EdU and DAPI showing the EdU ( right panels ) , cell membranes ( middle panels ) , and merged channels ( left panels ) . ( K ) Corresponding boxplot showing the EdU index in the leading and trailing regions of the pLLP in the indicated experimental conditions ( Figure 6—source data 1–3; Figure 6—figure supplement 1 , Figure 6—source data 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 03010 . 7554/eLife . 08201 . 031Figure 6—source data 1 . Cell counts in amotl2a;yap1 double morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 03110 . 7554/eLife . 08201 . 032Figure 6—source data 2 . EdU ratio in the leading region and trailing region of amotl2a;yap1 double morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 03210 . 7554/eLife . 08201 . 033Figure 6—source data 3 . Cell counts in embryos co-injected with Amotl2aMo and amotl2am or with Amotl2aMo and amotl2amLP . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 03310 . 7554/eLife . 08201 . 034Figure 6—source data 4 . Cell counts in MZamotl2a;MZyap1 double mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 03410 . 7554/eLife . 08201 . 035Figure 6—figure supplement 1 . Loss of Yap1 in MZamotl2a−/−;MZyap1−/− embryos supresses increased pLLP cell number resulting from loss of Amotl2a in MZamotl2a-−/− embryos . The mutation of Yap1-binding domains in Amotl2a does not render it unstable . MIP of Z-stacks of pLLP in cldnb:gfp embryos with the indicated genotype ( A–D ) . ( E ) Corresponding boxplot comparing pLLP cell counts . ( F ) Western blot showing the stability of wild-type Amotl2a , Amotl2aFL ( lane2 ) and Amotl2a forms carrying both PPEA-LATA ( lane3 ) and single PPEA and LATA ( lanes 4 and 5 ) , respectively . α-tubulin staining shows equal protein loading ( Figure 6—source data 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 035 Therefore , we directly tested if the increased proliferation rate of amolt2a morphants was similarly suppressed in double morphants . As previously observed , the proliferation rate in amotl2a morphants was significantly increased in the trailing region as compared to controls ( control vs Amotl2aMo: 0 . 27 ± 0 . 01 and 0 . 40 ± 0 . 01 , p < 0 . 0001 ) ; and this increase was partially suppressed in embryos co-injected with Amotl2aMo and Yap1Mo ( 0 . 32 ± 0 . 02 , p = 0 . 003 ) ( Figure 6F–K ) . Altogether , these results strongly suggest that Amotl2a is limiting proliferation and cell number in the pLLP , at least in part , by inhibiting Yap1 activity . Finally , we wanted to test whether the physical interaction between Amotl2a and Yap1 was necessary for Amotl2a function . For this purpose , we attempted to rescue the amotl2a morphant phenotype with amotl2am ( wild-type amotl2a modified to be insensitive to the Mo ) and with amotl2amLP , which was mutated in the LPTY and PPEY motifs and thus could not interact with Yap1 . While amotl2am could partially rescue the increase in cell counts of amotl2a morphants as shown above ( p = 6 . 36E-08 ) , amotl2amLP could not ( p = 0 . 18 ) ( Figure 6L–P ) . We also checked that the stability of the two proteins were comparable . Since there was no antibody recognizing zebrafish Amotl2a , we took advantage of the myc-tagged versions of the proteins that we used in our Y2H assays . The two proteins had a comparable stability in yeast ( Figure 6—figure supplement 1F ) . Altogether , these results suggested that the physical interaction between Amotl2a and Yap1 is required for the proliferation-limiting activity of Amotl2a . Our results so far indicated that the loss of Yap1 suppressed the increase in cell number in amotl2a mutants , possibly via a physical interaction between Amotl2a and Yap1 . Yet , pLLP cell counts in MZamotl2a−/−;MZyap1−/− double mutants were still higher than in MZyap1−/− suggesting that additional factors mediated the hyperplasia in amotl2a mutant pLLPs . Given that Wnt/β-catenin signaling , via Lef1 , promotes proliferation in the leading region of the pLLP ( Gamba et al . , 2010; Aman et al . , 2011; McGraw et al . , 2011; Valdivia et al . , 2011; Matsuda et al . , 2013 ) and that Amotl2a has been shown to physically interact with β-catenin and inhibit Wnt/β-catenin signaling during zebrafish gastrulation ( Li et al . , 2012 ) , we tested if Wnt/β-catenin signaling could also mediate the increase in cell number upon loss of Amotl2a activity . We first injected Amotl2aMo in tg ( 7xTCFXla . Siam:nlsmCherry ) ia5 embryos that carry a Wnt reporter ( Valdivia et al . , 2011; Moro et al . , 2012 ) . Although the intensity of red fluorescence was not significantly increased upon Amotl2aMo injection ( Figure 7—figure supplement 1A–E ) , we could not determine whether the number of cells expressing the reporter was increased because the red fluorescence was already present in most of the pLLP cells in control embryos ( Figure 7—figure supplement 1C , D ) , probably due to the stability of the cherry protein . Next , we tested whether the expression of a bona fide Wnt/β-catenin transcriptional target , such as lef1 , was increased upon loss of Amotl2a activity . Both amotl2a morphants and mutant embryos showed a slight but significant expansion of the lef1 expression domain into the trailing region of the primordium as compared to their respective controls ( Figure 7A–E and Figure 7—figure supplement 1F–J ) . On the other hand , the expression of lef1 was not modified in MZyap1−/− mutants ( Figure 7—figure supplement 1K–N ) . This suggested that an increase in Wnt/β-catenin/Lef1 activity in the trailing region of the pLLP could also partially mediate the increase in cell number . We thus tested if suppressing Lef1 activity in amotl2a mutants would suppress this phenotype . For this purpose , we injected MZamotl2a−/− mutant embryos with a previously published lef1 morpholino ( Lef1Mo ) and performed cell counts at mid-migration ( Valdivia et al . , 2011 ) . While MZamotl2a−/− mutants showed bigger primordia as expected ( 36% , p = 2 . 89E-09 ) ( Figure 7G , see ‘a’ in N ) , lef1 morphants at this stage did not show a significant reduction in pLLP cell counts ( p = 0 . 1 ) ( Figure 7F , J and ‘b’ in N ) . In contrast , cell counts in Lef1Mo-injected MZamotl2a−/− mutants were significantly reduced as compared to MZamotl2a−/− mutants ( −23% , p = 5 . 32E-08 , Figure 7N ‘c’ ) and equivalent to controls ( p = 0 . 28 , Figure 7N ‘d’ ) . Yet , they were still slightly higher than lef1 morphants ( 15% , p = 0 . 01 ) ( Figure 7G , K and ‘e’ in N ) . These results suggested that the increased cell number in amotl2a−/− mutants was Lef1-dependent . To confirm this , we tested if the difference in cell number in MZamotl2a−/−;MZyap1−/− and MZyap1−/− ( see above ) was due to Lef1 activity . Indeed when we injected MZamotl2a−/−;MZyap1−/− with Lef1Mo , the number of cells in the pLLP was significantly lower than in MZamotl2a−/−;MZyap1−/− double mutants ( −19% , p = 0 . 001 , Figure 7N ‘f’ ) and was not significantly different from that of MZyap1−/− mutants ( ‘g’ in Figure 7N , p = 0 . 07 ) . Altogether , these results strongly suggested that Amotl2a limits the number of cells and thus the size of the pLLP by repressing both Yap1 and Lef1 proliferation-promoting activities ( Figure 8 ) , possibly by physically interacting with Yap1 ( our data ) and β-catenin , as previously shown in zebrafish ( Li et al . , 2012 ) . 10 . 7554/eLife . 08201 . 036Figure 7 . Loss of Lef1 suppresses the increased cell proliferation in amotl2a mutants . ( A–D′ ) 30 hpf cldnb:gfp embryos with the indicated genetic background stained with a lef1 ISH probe and an anti-GFP antibody ( B′ , D′ ) . ( E ) Boxplot showing the expansion of lef1 expression domain in MZamotl2a−/− embryos . ( F–M ) MIP of Z-stacks of pLLP in cldnb:gfp embryos with the indicated genotype , either uninjected ( F–I ) or injected with a Lef1Mo ( J–M ) . ( N ) Corresponding boxplot comparing pLLP cell counts ( Figure 7—source data 1 , 2; Figure 7—figure supplement 1 , Figure 7—source data 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 03610 . 7554/eLife . 08201 . 037Figure 7—source data 1 . Relative lef1 expression domain area in MZamotl2a mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 03710 . 7554/eLife . 08201 . 038Figure 7—source data 2 . Cell counts in Lef1Mo-injected MZamotl2a;MZyap1 double mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 03810 . 7554/eLife . 08201 . 039Figure 7—source data 3 . Relative lef1 expression domain area in amotl2a morphants . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 03910 . 7554/eLife . 08201 . 040Figure 7—figure supplement 1 . Loss of Amotl2a leads to an expansion of lef1 expression . ( A–D ) MIP of confocal stacks of the pLLP in Tg ( 7xTCF-Xla . Sia:NLS-mCherry ) embryos injected ( B , D ) or not injected ( A , C ) with the amotl2aMo and showing the green ( membranes , left ) and red ( nuclear WNT reporter ) channels . ( E ) Boxplot comparing the relative mean intensity of fluorescence in the red channel in the corresponding pLLP . ( F–I′ ) Overview pictures ( F , H ) and close-up on the pLLP of 30 hpf cldb:gfp embryos injected ( H , I , I′ ) or not injected ( F , G , G′ ) with the Amotl2aMo and stained with a lef1 ISH probe and an anti-GFP antibody ( G′ , I′ ) . Red arrows point to the pLLP ( F , H ) . ( J ) Boxplot showing the expansion of lef1 expression domain in amotl2a morphants . ( K–N ) Close-up on the pLLP of 30 hpf control or MZyap1−/− cldb:gfp embryos stained with a lef1 ISH probe and an anti-GFP antibody ( L , N ) . The relative lef1 expression domain is not different in MZyap1−/− embryos ( Figure 7—source data 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 04010 . 7554/eLife . 08201 . 041Figure 8 . Working model . Scheme summarizing the main regulatory pathways involved in proliferation control and organ formation in the pLLP . The red part corresponds to new findings from the present study . Dotted lines indicate hypothetical links . DOI: http://dx . doi . org/10 . 7554/eLife . 08201 . 041 Embryonic development is associated with high proliferation rates ensuring proper organ formation . Yet , proliferation must remain under tight spatiotemporal control to ensure that a sufficient number of cells is produced , but not exceeded . In the pLLP , Wnt/βcatenin signaling is required to maintain proliferating progenitors in the leading region to compensate for the loss of cells resulting from neuromast deposition ( Gamba et al . , 2010; Aman et al . , 2011 ) . This is essential for the pLLP to deposit a complete set of LL organs ( McGraw et al . , 2011; Valdivia et al . , 2011 ) . However , no mechanism has been proposed that restricts proliferation and growth in the pLLP . We show here that Amotl2a is essential to limit proliferation of the cells that assemble into proneuromasts in the trailing region of the pLLP . This leads us to propose that the size of the pLLP is controlled throughout migration by a balance between Wnt/βcatenin-dependent mitogenic signals from the leading region and Amotl2a-dependent proliferation-restricting signals from the trailing region ( Figure 8 ) . We further show that amotl2a is expressed downstream of FGF signaling in the trailing region of the pLLP . Yet , embryos lacking FGF signaling in the pLLP do not have larger primordia ( Aman et al . , 2011 , and our own observations ) . This suggests that FGF has a mitogenic function in the pLLP in addition to promoting amotl2a expression , and thus , limiting proliferation . This is not surprising given that FGF signaling promotes proliferation in several contexts during vertebrate development ( Thisse and Thisse , 2005; Aman et al . , 2011 ) . We propose that in FGF-deficient embryos , an increase in Yap1 and Lef1 activity resulting from decreased amotl2a expression might partially compensate for the absence of FGF-dependent proliferation . The increase in cell number in the pLLP of amotl2a mutants does not lead to bigger neuromasts , but to one additional neuromast . While a recent study reported that the size of differentiated neuromasts depends on the canonical Wnt pathway at later stages ( Wada et al . , 2013 ) , the mechanisms that dictate how many cells participate in forming a new rosette in the migrating pLLP are not known . Intriguingly , our results suggest that this number does not depend on the size of the pLLP itself . This is in agreement with the fact that smaller primordia , for example in lef1 mutants , also deposit fewer , but apparently not smaller , neuromasts ( McGraw et al . , 2011; Valdivia et al . , 2011 ) . Finally , we show that rosette assembly is slightly delayed in amotl2a morphants and mutants . At least two mechanisms could account for this delay . First , Motin proteins have been implicated in the formation of tight junctions in different contexts ( Bratt et al . , 2002; Sugihara-Mizuno et al . , 2007; Zheng et al . , 2009 ) , although tight junctions do form in the absence of Amotl2a , their assembly might be challenged , leading to a delay in epithelial rosette assembly . Second , we show that lef1 expression is expanded upon loss of Amotl2a function . Interestingly , upon loss of Lef1 activity , rosettes assemble closer to the pLLP tip ( McGraw et al . , 2011; Matsuda et al . , 2013 ) . This has been attributed to the loss of the FGF pathway inhibitor Dusp6 downstream of Lef1 leading to a premature incorporation of leading cells into epithelial rosettes ( McGraw et al . , 2011; Matsuda et al . , 2013 ) . The expansion of lef1 expression into the trailing region upon loss of Amotl2a function could conversely explain the delay in rosette assembly . That this delay does not result in an increased distance between deposited neuromasts could be explained by the fact that one additional neuromast form in this context due to the increase in cell number . We show for the first time that the Hippo signaling pathway effector Yap1 is required for the pLLP to contain a sufficient number of cells . Since Yap1 is already required before migration starts , we propose that one of Amotl2a's functions is to maintain low Yap1 activity as rosettes assemble in the trailing region . In amotl2a mutants , Yap1 activity would fail to be repressed and would remain higher for a longer period of time leading to overproliferation . Our data do not allow us to determine , however , whether the canonical Hippo pathway is active in the pLLP to additionally regulate Yap1 activity . More work will be required to answer this interesting question . A paper recently reported that morpholino-mediated knock-down of Yap1 in zebrafish leads to a reduction of the number of neuromasts but not of the size of the pLLP ( Loh et al . , 2014 ) . They also reported a failure of hair cell differentiation that they attributed to a reduction of expression of the Wnt-target gene prox1 . Intriguingly , both our yap1 morphants and mutants showed significantly different phenotypes with a reduction of the size of the pLLP but not of the number of neuromasts deposited . In addition , neither prox1 expression nor the numbers of hair cells were affected in yap1−/− mutants . One possible explanation for these discrepancies is that our yap1 mutation does not affect all yap1 transcripts . This is , however , unlikely since it lies in the TEAD-binding domain , which is probably essential to Yap1 function . On the other hand , the authors also reported increased apoptosis and activation of the p53 pathway , two processes that have been shown to lead to a reduction in the number of neuromasts ( Aman et al . , 2011 ) . This could explain the discrepancies of phenotypes between the previously reported yap1 morphants and our newly generated yap1−/− mutants . Our findings that Amotl2a physically and genetically interacts with the Hippo pathway component Yap1 to control proliferation in the pLLP is in agreement with several studies showing that Motin proteins physically interact with YAP and TAZ , sequester them in the cytoplasm , and inhibit their proliferation-promoting transcriptional activity ( Varelas et al . , 2010; Chan et al . , 2011; Wang et al . , 2011; Zhao et al . , 2011 ) . Our results provide the first evidence that a similar mechanism could be involved in vivo during vertebrate development to control tissue and organ size . Antibodies recognizing Yap1 will be necessary to determine its localization in the pLLP cells . The commercial Yap1 antibodies we have tested did not cross-react with the zebrafish protein . Intriguingly , none of the known YAP/TAZ target genes we looked at , including ctgfs , cyr61 , myc , cyclins , and prox1 , showed differences in expression level in amotl2a morphants ( data not shown ) . Alternatively , rather than sequestering Yap1 into the cytoplasm , Amotl2a could promote Yap1 degradation . Motins have indeed been shown to promote the phosphorylation of YAP/TAZ by LATS1/2 , which would not only lead to their cytoplasmic retention but also to their degradation ( Paramasivam et al . , 2011 ) . Although there is currently no evidence that the canonical Hippo pathway is active in the pLLP , Amotl2a could be involved in controlling Yap1 stability . Our results indicate that Amotl2a limits the number of cells and thus the size of the migrating pLLP by repressing the activity of both Lef1 and Yap1 . This is in agreement with the fact that both Lef1 ( Gamba et al . , 2010; McGraw et al . , 2011; Valdivia et al . , 2011; Matsuda et al . , 2013 ) and Yap1 ( this paper ) are required for establishing and maintaining the correct number of cells in the pLLP . To our knowledge , a direct link between Motin proteins and the Wnt/β-catenin pathway has been described only once , with a physical interaction between zebrafish Amotl2a and β-catenin ( Li et al . , 2012 ) . Although we did not detect β-catenin in our Y2H screen nor identified an interaction between Amotl2a and β-catenin in our Y2H assay ( not shown ) , Amotl2a might be physically interacting with β-catenin in vivo in the pLLP cells thereby limiting its access to the nucleus . Our data suggest that Yap1 and Lef1 function at least in part independently downstream of Amotl2a since the pLLP of both yap1 single mutants and lef1 morphants has less cells than amotl2a;yap1 double mutants and Lef1Mo-injected amotl2a mutants , respectively . Yet , it is also possible that both pathways are partially interconnected downstream of Amotl2a . Several recent studies have indeed reported cross-regulations between Yap/Taz and Wnt/β-catenin signaling ( Varelas et al . , 2010; Fish et al . , 2011; Heallen et al . , 2011; Azzolin et al . , 2012; Imajo et al . , 2012; Barry et al . , 2013; Azzolin et al . , 2014; Piccolo et al . , 2014 ) . In particular , Yap1 and β-catenin have been shown to physically interact and mutually repress each other in the cytoplasm while rather promoting each other's activity in the nucleus . Further work will be required to precisely dissect potential cross-talks between Yap1 and Wnt/β-catenin in the pLLP . One particularly interesting property of YAP/TAZ is their capacity to sense and respond to changes in biophysical properties of cells including cell density , cell polarity , cell shape , tension forces , and substrate stiffness ( Dupont et al . , 2011; Halder et al . , 2012; Aragona et al . , 2013; Gaspar and Tapon , 2014; Rauskolb et al . , 2014 ) . Our results raise the question whether Yap1 responds to such changes in the pLLP , and to which extent the response depends on Amotl2a function . We show that the repressive function of Amotl2a on Yap1 coincides with the assembly of cells into rosettes . Interestingly , these cells undergo a number of changes including epithelialization , acquisition of a columnar shape , and apical constriction via acto-myosin contraction ( Lecaudey et al . , 2008; Ernst et al . , 2012 ) . In principle , Yap1 could respond to any of these changes and Amotl2a could act as a mediator . First , tight junction-associated Amotl2a in newly formed rosettes could recruit and inhibit Yap1 . In addition , Motins , including Amotl2a ( Hultin et al . , 2014 ) , can bind actin and this interaction has been shown to compete with the binding of Motin to YAP ( Ernkvist et al . , 2006; Chan et al . , 2013; Dai et al . , 2013; Gaspar and Tapon , 2014 ) . It is thus tempting to propose that Amotl2a could link actomyosin-based cell shape changes such as columnarization and/or apical constriction during rosette assembly to Yap1 activity and proliferation . Finally , at low density in culture , cells are flat and spread and this morphology promotes nuclear YAP accumulation ( Wada et al . , 2011 ) . In contrast at high density , cells are compact and tall and this is associated with cytoplasmic localization of YAP . In the pLLP , cells in the leading region are rather flat while cells in the trailing region , which assemble into rosettes , are more columnar ( Lecaudey et al . , 2008 ) . Thus , in the primordium , more than the cell density , the assembly of cells into rosettes and the associated changes in cell shape could regulate the cytoplasmic localization of Yap1 in an Amotl2a-dependent manner . Such a mechanism would allow to couple proliferation rate to organogenesis . Upon loss of Amotl2a , in contrast , this coupling would be lost so that proneuromasts assemble but proliferation is not restricted , leading to hyperplastic LLP and additional sensory organs . Adult zebrafish were maintained under standard conditions and embryos were staged according to Kimmel et al . ( 1995 ) . Transgenic lines Tg ( −8 . 0cldnb:lynEGFP ) zf106 ( cldnb:gfp ) , Tg ( hsp70l:dnfgfr1-EGFP ) pd1 , Tg ( hsp70l:fgf3-Myc ) zf115 , Tg ( 7xTCF-Xla . Siam:nlsmCherry ) ia5 and mutant lines fgf3/liat21142 and fgf10/daetbvbo have been described previously ( Herzog et al . , 2004; Lee et al . , 2005; Norton et al . , 2005; Thisse and Thisse , 2005; Lecaudey et al . , 2008; Ernst et al . , 2012; Moro et al . , 2012 ) . Heat-shock was performed at 39°C for 15 min . Capped mRNAs were transcribed with the SP6 mMessage mMachine Kit ( Ambion ) . Mo are described in the Supplementary file 1D . Amotl2aMo was co-injected with a p53Mo to overcome unspecific cell death ( Gerety and Wilkinson , 2011 ) and both uninjected and p53Mo-injected embryos were used as controls . FL amotl2a , amotl2aΔSTOP , MoBS_Amotl2a , yap1 , and taz were amplified by PCR from zebrafish embryo cDNA using the primer pairs indicated in the Supplementary file 1B . PCR products were further cloned into pCS2 , pCS2-gfp , pCS2-TdT , pGADT7 , or pGBKT7 vectors as indicated in the Supplementary file 1C . The mutant lines amotl2afu45 , amotl2afu46 , yap1fu47 , and yap1fu48 were generated using TALEN . The TALE repeat array plasmid kit was a gift from Daniel Voytas and Adam Bogdanove obtained via Addgene ( kit #1000000024 ) . TALE repeat arrays were assembled following the Golden Gate TALEN assembly protocol originally described in Cermak et al . ( 2011 ) , modified by the Voytas lab and available on the Addgene website ( https://www . addgene . org/static/cms/filer_public/98/5a/985a6117-7490-4001-8f6a-24b2cf7b005b/golden_gate_talen_assembly_v7 . pdf ) . Target sites and the corresponding repeat-variable diresidue ( RVD ) sequences were chosen using the online tool MoJo Hand ( http://talendesign . org/ ) in exon 2 and 3 of yap1 and amotl2a genes , respectively . The array plasmids were fused to the Fok1 endonuclease in the GoldyTALEN backbone . After linearization , mRNAs were transcribed using the T3 mMessage mMachine Kit ( Ambion by Life Technologies GmbH , Darmstadt , Germany ) according to the manufacturer's instructions . The two mRNAs corresponding to the left and right arms were then mixed in equal quantities and injected into embryos at the one-cell stage . 20–30 embryos at 48 hpf were collected from each clutch and gDNA was extracted . The target sites were amplified using primers generating a 350–600 bp long PCR product ( Supplementary file 1E ) . Efficiency of the TALEN pair was estimated by digesting the PCR product with the restriction site present in the spacer of the target site . If a significant amount of uncut PCR product was observed , the rest of the injected embryos were further grown to adulthood . The resulting mosaic adult fish were out-crossed and genomic DNA was prepared from 50 embryos to identify potential mutations using the same PCR and digestion as described above . F1 fish were finally genotyped by fin clipping . The uncut band ( carrying the mutation ) was further amplified and sent to sequence . To obtain MZ mutants , heterozygous carriers were incrossed and the progeny were raised to adulthood . Homozygous wild-type and mutants were identified by fin-clipping . In all experiments with MZ mutants , control embryos originate from incrosses of these homozygous wild-type and are thus related to the MZ mutants in a way that they share the same ‘grand-parents’ . ISH and immunofluorescence staining were performed according to standard procedures ( Lecaudey et al . , 2004 ) . amotl2a , taz , and yap1 ISH probes were amplified by PCR from zebrafish embryo cDNA using the primers indicated in the Supplementary file 1A . PCR products were cloned into the pGEM-T vector ( Promega GmbH , Germany ) according to manufacturer's instructions . The probes for cxcr7b , cxcr4b , lef1 , and prox1 were previously published ( Glasgow and Tomarev , 1998; Dorsky et al . , 1999; Thisse and Thisse , 2005; Valentin et al . , 2007 ) . The following antibodies were used: rabbit anti-GFP ( 1:500; Torrey Pines Biolabs , Secaucus , NJ , United States ) , mouse anti-GFP ( 1:500; JL8 , Takara Bio Europe/Clontech , France ) , mouse anti-ZO1 ( 1:500; GmbH ) , mouse anti-HCS-1 ( 1:20 , HCS-1 was deposited to the DSHB , Iowa city , IA , United States by Corwin , J ) , and Alexa dye-conjugated antibodies ( 1:500; Molecular Probes ) . Rhodamine-phalloidin ( Life Technologies GmbH ) was used at 1:100 . 28–30 hpf embryos were dechorionated , incubated with 10 mM EdU solution ( Click-iT EdU , Life Technologies GmbH ) for 20 min on ice . After washing , embryos were incubated at 28 . 5°C for 1 hr to allow EdU incorporation and treated with the Click-iT reaction solution following manufacturer's instructions . The Y2H screen was performed by Hybrigenics ( Paris , France ) with the FL coding sequence of zebrafish amotl2a ( aa 1–721 ) . 80 millions interactions were analyzed . The Y2H assay was performed using the Matchmaker system and the AH109 yeast strain ( Clontech ) according to the manufacturer's instructions . amotl2a was fused to the Gal4 DNA-binding domain in the pGBKT7 vector . Zebrafish yap1 , yap1ΔWW1 , 2 ( WQDP and WLDP motifs mutated to AQDA and ALDA ) , taz and tazΔWW ( WHDP motif mutated to AHDA ) were fused to the Gal4 activation domain in the pGADT7 vector . All constructs fused to the DNA-binding domain were tested for autoactivation . To check for protein interactions , colonies were scraped off the plates , diluted to an OD600 of 0 . 4 , patched on selective plates lacking histidine , and grown for 3–4 days . Yeast cell lysates were prepared according to the protocol from Kushnirov with some modifications ( Kushnirov , 2000 ) . Yeast cultures were grown to a logarithmic stage to reach an OD600 0 . 4–0 . 8 . Cultures of 2 ml were harvested by centrifugation . After discarding the supernatant , the pellets were resuspended in 500 μl of 0 . 1 M NaOH , transferred to 2 ml eppendorf tubes , and incubated at room temperature for 10 min followed by centrifugation at 14 , 000 rpm for 2 min at room temperature . After carefully discarding the supernatant , the pellet was resuspended in 1× sample buffer slightly modified from standard Laemmli ( 25 mM Tris pH6 . 8 , 30% glycerol , 5% Sodium Dodecyl Sulfate ( SDS ) , 1% bromophenol blue , and 5% β-mercaptoethanol ) and dissolved by full speed shaking on a Thermoblock set at 37°C for 20–30 min . The samples were then boiled at 95°C for 5 min . The cell debris was removed by centrifugation at 14 , 000 rpm for 5 min at room temperature . The supernatant was transferred to a fresh eppendorf tube and stored at −20°C until loading . 15 μl of each of the yeast lysates was electrophoresed on a 10% SDS polyacrylamide gel and transferred to polyvinylidene difluoride ( PVDF ) membrane ( Millipore , by Merck Chemicals GmbH , Germany ) . The proteins were analyzed by Western blotting with the following antibodies:monoclonal mouse antibody 9E10 ( anti-c-myc , Covance , by BioLegend , United Kingdom ) , monoclonal mouse antibody DM1A ( anti-α-tubulin , Sigma-Aldrich Chemie GmbH , Germany ) . Imaging and image analysis were performed as previously described ( Ernst et al . , 2012 ) . Quantifications of data are presented as boxplot in which the central line is the median of the group and the edges of the box are the 25th and 75th percentiles . The black vertical lines extend to the most extreme data points not considered outliers . In all cell count boxplots , each blue dot corresponds to one data point ( one primordium ) . Statistical analyses were performed with GraphPad Prism6 and Matlab using the Welch's t-test . p values are indicated on each figure and/or in the result part . Differences in cell counts between groups are differences between mean values ( and not between median values shown in the boxplots ) and are expressed in percentage . Automated cell counting was done using the GFP channel of cldnb:gfp transgenic embryos ( Figure 3—figure supplement 1A ) . To exclude that the delayed migration in amotl2a morphants would impact on the pLLP cell number , cell counts were done when the primordium had reached the middle of the yolk extension ( ‘mid-migration’ , Figure 2—figure supplement 1A , C ) both in control embryos ( about 30 hpf ) and in amotl2a morphants and mutants . Because the cell membrane is a plane-like structure in three dimensions , we used anisotropic diffusion ( Weickert , 1998 ) to enhance local consistency in structure and suppress noise ( Figure 3—figure supplement 1B ) . The diffusion tensor , which enhances plane-like structures , was constructed from the local Hessian matrix . We then computed the smallest eigenvalue of the local Hessian matrices , which is a good indicator of the cell membrane ( Sato et al . , 2000 ) . The cell segmentation was done by watershed segmentation with h-minima transform ( Soille , 2003 ) on the eigenvalue image . Finally , very small segments were filtered out: their regions are merged into neighboring segments by applying a second watershed segmentation ( Figure 3—figure supplement 1C ) . A number of publicly available tools , for example , ‘Icy’ ( de Chaumont et al . , 2012 ) or Matlab programs , can be used for these image-processing steps . To generate a segmentation mask of the pLLP , the 3D images were smoothed with a small Gaussian kernel . An optimal threshold was estimated for converting the smoothed intensity image to a binary image ( e . g . , using the ‘graythresh’ function in Matlab ) , so that the intraclass variance of the two regions ‘0’ and ‘1’ is minimized . Based on this pLLP segmentation mask ( Figure 3—figure supplement 1D ) , we obtained the cell numbers by counting each segment that had more than 75% of its volume inside the mask . Visual monitoring of two experiments validated the accuracy of the segmentation ( Figure 3—figure supplement 1 ) .	How do organs and tissues know when to stop growing ? A cell communication pathway known as Hippo signaling plays a central role as it can tell cells to stop dividing . It is activated when cells in developing tissues come into contact with each other and causes a protein called Yap1 to be modified , which prevents it from entering the cell nucleus to activate genes that are involved in cell division . In a zebrafish embryo , an organ called the lateral line forms from a cluster of cells that migrate along the embryo's length . At regular intervals , the cluster deposits small bunches of cells from its trailing end . The resulting loss of cells from the cluster is balanced by cell division at the front of the cluster , which is triggered by another signaling pathway called Wnt signaling . A protein of the ‘Motin’ family called Amotl2a is present in this migrating cluster . Motin proteins form junctions between cells and inhibit the activity of Yap1 , but it is not known whether they are involved in regulating the size of organs . Here , Agarwala et al . used the lateral line as a model to study the control of organ size in zebrafish embryos . The experiments show that when Amotl2a is absent , the migrating cell cluster becomes larger , with the highest levels of cell division occurring at its trailing end . Yap1 and a protein involved in Wnt signaling called Lef1 are also present in the cluster and are required for it to be normal in size . In zebrafish that lack Amotl2a , the additional loss of Yap1 prevents this cluster from becoming too large . From these and other results , it appears that Amotl2a regulates the size of the lateral line cell cluster by restricting the ability of Yap1 and Lef1 to promote cell division . Agarwala et al . 's findings demonstrate a role for Amotl2a in controlling the size of organs . A future challenge is to understand the details of how it restricts the activities of Yap1 and Lef1 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2015	Amotl2a interacts with the Hippo effector Yap1 and the Wnt/β-catenin effector Lef1 to control tissue size in zebrafish
The knirps ( kni ) locus encodes transcription factors required for induction of the L2 wing vein in Drosophila . Here , we employ diverse CRISPR/Cas9 genome editing tools to generate a series of targeted lesions within the endogenous cis-regulatory module ( CRM ) required for kni expression in the L2 vein primordium . Phenotypic analysis of these ‘in locus’ mutations based on both expression of Kni protein and adult wing phenotypes , reveals novel unexpected features of L2-CRM function including evidence for a chromosome pairing-dependent process that promotes transcription . We also demonstrate that self-propagating active genetic elements ( CopyCat elements ) can efficiently delete and replace the L2-CRM with orthologous sequences from other divergent fly species . Wing vein phenotypes resulting from these trans-species enhancer replacements parallel features of the respective donor fly species . This highly sensitive phenotypic readout of enhancer function in a native genomic context reveals novel features of CRM function undetected by traditional reporter gene analysis . The Drosophila wing provides an excellent model system for studying the relationship between gene expression and morphogenesis since it is a well-studied , two-dimensional , developmental system that generates invariant morphological features , such as veins and sensory organs ( Sturtevant et al . , 1993 ) . Wing veins provide structural support necessary for flight and supply nutrients to sensory organs that coordinate wing beat motions ( reviewed in [Bier , 2000] ) . The number and positioning of veins are highly selected characteristics that vary among dipterans , holding evolutionary significance pertinent to the divergence of insect species and their modes of flight ( reviewed in [Bier , 2000] ) . Drosophila wing venation is determined by an intricate gene regulatory network ( GRN ) acting during larval and early pupal stages to position the five major longitudinal veins ( L1-L5 ) and to sculpt the overall final shape of the wing . Secreted morphogens , including Hedgehog ( Hh ) and Decapentaplegic ( Dpp ) , activate the expression of primary response genes , such as spalt ( sal = salm + salr ) and optomotor-blind ( omb , also known as bifid ) , in broad domains . These patterning genes have been proposed to subsequently induce the formation of vein primordia along their borders by ‘for-export-only-signaling’ mechanisms ( reviewed in [Bier , 2000] ) and cross-repressive interactions ( Al Khatib et al . , 2017; Martín et al . , 2017 ) . The L2 vein primordium forms along the anterior border of the sal domain ( Sturtevant et al . , 1997 ) , where the zinc finger transcription factors , knirps and its neighboring homolog , knirps-related ( knrl ) are induced and direct the L2 vein development program ( Lunde et al . , 1998 ) . Genomic lesions associated with several independently generated radius incompletus ( ri ) regulatory alleles of the kni locus , include deletions or point mutations within a 4 . 8 kb fragment upstream of the kni coding region that greatly reduce or eliminate expression of reporter genes in the L2 primordium and result in L2 vein truncation phenotypes ( Lunde et al . , 2003 ) . Thus , the kniri[1] mutant ( Figure 1D ) contains a 252 bp deletion , while the kniri[53j] allele ( Figure 1E ) consists of a single point mutation ( C to A within the same region deleted in kniri[1] mutants ) , indicating the importance of a single nucleotide residue in this position . A minimal ~1 . 4 kb kni CRM fragment ( EX - Figure 1A ) was defined by deletion analysis , driving reporter gene expression in a pattern similar to that of the full 4 . 8 kb fragment . However , further truncation of that region to a 0 . 69 kb fragment ( EC - Figure 1A ) resulted in strong ectopic expression in anterior and posterior domains of the imaginal disc , suggesting that the reciprocal fragment contained negative regulatory elements responsible for repressing kni expression in peripheral regions of the wing ( Lunde et al . , 2003 ) . The prior experiments with CRM-reporter gene fusions summarized above , provided important information regarding the nature and organization of CRM sub-modules , but did not provide a clear link to gene function in the context of the endogenous kni locus . The recent advent of highly efficient CRISPR/Cas9 ( clustered randomly interspaced palindromic repeats/CRISPR-associated protein 9 ) genome editing tools has provided a method to precisely alter CRM sequences and observe the resulting effects on development of the final L2 vein structure in the adult wing , permitting more sensitive decoding of CRM functionality . The type II CRISPR/Cas9 endonuclease system evolved in bacteria as a defensive measure against viruses to cleave invading foreign DNA ( Barrangou et al . , 2007 ) . This natural defense system has been engineered into a bipartite genome editing tool , consisting of a guide RNA ( gRNA ) that directs double stranded DNA cleavage by the Cas9 endonuclease at specific locations within the genome ( Jinek et al . , 2012 ) and has been adapted for use in a myriad of organisms ( reviewed in: ( Bassett and Liu , 2014; Doudna and Charpentier , 2014; Hsu et al . , 2014; Overcash et al . , 2015; Sander and Joung , 2014; Sternberg and Doudna , 2015; Zhang et al . , 2014 ) , including Drosophila ( Bassett et al . , 2013; Gratz et al . , 2013a; Kondo and Ueda , 2013; Ren et al . , 2013; Yu et al . , 2013 ) . Following gRNA/Cas9 cleavage at a given target site , mutations can be generated at that genomic location via one of two pathways: error-prone non-homologous end joining ( NHEJ ) DNA repair , which typically generates small insertions and/or deletions ( indels ) at the cut site , or the more precise homology-directed repair ( HDR ) pathway that copies sequences from a homologous DNA template ( Gratz et al . , 2013 ) . With the advent of CRISPR technology , these methods have also been employed to generate ‘active genetic’ elements that are copied in the germline via HDR and inherited in a Super-Mendelian fashion ( reviewed in [Gantz and Bier , 2016] ) . In the current study , we employed CRISPR/Cas9 to generate mutations in the endogenous Drosophila kni L2-CRM and analyzed the phenotypic effects of these lesions on adult wing vein pattern . We generated a series of targeted sequential deletions spanning a ~ 2 kb segment encompassing the previously identified minimal 1 . 4 kb L2-CRM fragment plus additional well-conserved adjacent sequences . We also recovered and analyzed a variety of additional non-targeted mutations associated with imprecise lesions that were most likely generated by NHEJ . These studies provide a much higher resolution view of CRM function by linking CRM lesions in their native chromosomal context to precise phenotypic outputs . In addition , we validate the high efficiency of homology-directed , CRISPR-mediated , site-specific transgenesis as a viable alternative to traditional transgenesis methods ( e . g . P-element or ɸC31-based systems ) . We also introduce the use of active genetic elements we refer to as CopyCat cloning vectors that can be inserted at determined locations and are copied with high efficiency from one parental chromosome to another during germline transmission , in the presence of an unlinked source of Cas9 . Finally , we show that such active CopyCat vectors can be used for efficiently replacing the L2-CRM with homologous sequences from other fly species , resulting in alternative patterns of L2 vein placement . These new active genetics tools should greatly accelerate detailed combinatorial analysis of gene regulatory networks in a variety of experimental settings , as well as analyzing the function of exogenous DNA sequences found among diverged species . As summarized in the introductory section , well-validated CRISPR/Cas9 genome editing tools now make it routine to generate genome alterations via random mutagenesis near guide RNA ( gRNA ) cleavage sites ( via the NHEJ pathway ) or to create targeted edits based on templates carrying homologous sequences on either side of the gRNA-directed cut site ( via homology direct repair - HDR ) . We employed the latter precise approach to generate a series of deletions of the endogenous L2-CRM . We utilized two gRNAs directing Cas9 cleavage on each side of a specific region to be deleted , as well as single-stranded oligonucleotides donors ( ssODN ) with 60 bp of homology to either DNA end flanking the region targeted for deletion ( see Materials and Methods ) . We first used two gRNAs ( gRNA-A and gRNA-B ) to delete a 2023 bp regulatory region encompassing the previously identified 1 . 4 kb minimal L2-CRM . The choice of this region was based primarily on significant sequence conservation among closely related flies which extended beyond both borders of the fragment ( Figure 1A ) analyzed in our original L2-CRM reporter studies ( Lunde et al . , 2003 ) . When rendered homozygous , the kniri-AB allele resulted in complete elimination of the L2 vein ( Figure 1C ) suggesting that it represents a true , null allele of the L2-CRM function . Consistent with this phenotypic assessment , Kni protein expression was undetectable in the L2 primordium of ri-AB homozygous third larval instar wing imaginal discs ( Figure 2B ) or early pre-pupal wings ( Figure 2—figure supplement 1 ) . As diagrammed in Figure 1A , we also generated strains of flies carrying a series of four smaller deletions ( riΔ1-riΔ4 ) spanning the ~2 kb AB interval , each eliminating contiguous regions of sequence conservation . We also indicate three short regions that display a particularly high degree of sequence conservation across Drosophilids and other Schizophora flies , that we refer to as hyper-conserved identity islands ( labeled ID1 , ID2 and ID3 , respectively , magenta , red and green boxes in Figure 1A also indicated in the multispecies alignment depicted in Figure 5A , and at the DNA sequence level in Figure 5—figure supplement 1 ) . The generated deletions were tested for phenotypes in trans-heterozygosity with the riAB deletion allele and in the homozygous condition . Consistent with prior studies that identified a central region required for CRM activity ( i . e . a 252 bp region deleted in ri1 ( Figure 1D ) that contains essential Scalloped activator binding sites [ Lunde et al . , 2003] ) , the riΔ2 and riΔ3 deletions , which both partially overlap with the region deleted in ri1 , produced strong vein-loss phenotypes similar to those of ri1 when placed in trans to riAB ( Figure 1J , K ) . Also , consistent with the prior mapping of a repressor domain to distal sequences missing in the EC fragment ( Figure 1A: blue box ) , the homozygous riΔ4 deletion produced a mild ectopic vein phenotype in peripheral regions of the wing ( Figure 1I ) . The homozygous riΔ1 deletion produced a similar mild ectopic vein phenotype ( Figure 1H ) lending support to our suspicion mentioned above that conserved sequences that were not included in our previous reporter analysis , contribute to the overall CRM function . Since both the riΔ1 and riΔ4 deletions produced mild ectopic vein phenotypes , we generated a composite riΔ1+Δ4 deletion ( by generating the riΔ1 deletion in the background of riΔ4 ) to determine whether these separated domains of the L2 CRM might act in concert . We found that this was indeed the case , as the homozygous combined riΔ1+Δ4 deletion produced a much more pronounced phenotype in which large thickened veins formed around L2 and L5 ( Figure 1M ) , a phenotype that was similar albeit less pronounced in trans-heterozygotes over the AB deletion ( riΔ1+Δ4/riAB Figure 2L ) . Consistent with this phenotype , significant levels of ectopic Kni expression were observed in the third instar imaginal discs of riΔ1+Δ4 homozygous mutants in both anterior and posterior domains of the wing pouch ( Figure 2L ) , in a pattern similar to that driven by the EC-lacZ reporter construct analyzed previously ( Lunde et al . , 2003 ) . The domains of ectopic Kni expression in riΔ1+Δ4 homozygotes are largely complementary to those of the Spalt transcription factor ( Figure 2P–R ) as is the case for wild-type Kni expression ( Figure 2M–O ) , indicating that binding sites for Spalt to repress Kni expression in central regions of the wing , are likely retained in the remaining CRM sequences . Interestingly , the pattern of ectopic Kni expression in homozygous riΔ4 single mutant wing discs ( Figure 2H ) was similar in both pattern and level to that in riΔ1+Δ4 double mutant discs ( Figure 2L ) , and only mild ectopic Kni expression was observed in riΔ1 mutant discs ( Figure 2G ) . Slightly later in development during pre-pupal stages , however , narrower domains of high level Kni expression are observed in riΔ1+Δ4 double mutant ( Figure 2—figure supplement 1R ) in comparison to the persisting broader domains of Kni expression present in the riΔ1 single mutant ( Figure 2—figure supplement 1O ) . These later differences in Kni expression may help account for differences in the severity of ectopic wing vein phenotypes in the riΔ4 single mutant versus the double riΔ1+Δ4 mutant ( see Discussion ) . Thus , a combination of the Kni expression patterns and resulting adult morphological phenotypes provide more informative distinctions between these various mutants than can be ascertained by either measure alone . In addition to phenotypes described above that were consistent with and extended previous observations , we also encountered unexpected results when testing the effect of homozygous riΔ2 and riΔ3 deletions . In contrast to the strong vein-loss phenotypes associated with these deletions in trans to the complete riAB deletion described above ( Figure 1J , K ) , the phenotypes of these lesions were greatly reduced when homozygous ( Figure 1F , G ) . This disparity was particularly striking for the homozygous riΔ3 deletion , which generated an almost wild-type phenotype , recognizable only by a subtle meandering or less pigmented L2 vein ( Figure 1G ) . This very weak phenotype , which was observed in four independently derived riΔ3 mutant strains , is particularly surprising given that a single base pair change within that region associated with the ri53j allele leads to a substantial vein loss phenotype when homozygous ( [Lunde et al . , 2003]; Figure 1E ) . The recovery of these two deletions that produced surprisingly mild homozygous phenotypes suggested the possibility that a pairing-dependent process might contribute to sustaining CRM function . Consistent with this hypothesis , trans-heterozygotes of the riΔ3 deletion over the point mutant ri53j allele ( Figure 3J ) , - a configuration predicted to disrupt precise alignment of homologous chromosomes - resulted in a pronounced vein-loss phenotype . We tested the pairing hypothesis more directly using the classic chromosomal inversion comparison by placing the riΔ3 deletion in-trans to the ri1 mutant either in the context of a native third chromosome or in a multiply inverted balancer chromosome ( TM3 , ri1 ) that greatly reduces pairing-dependent phenomena such as transvection ( i . e . wherein the CRM on one chromosome acts to promote transcription from the homologous chromosome [Lewis , 1954] ) . Consistent with the pairing-based mechanism , the riΔ3/ri1 phenotype ( Figure 3B ) was significantly weaker than that observed in riΔ3/TM3 , ri1 individuals ( Figure 3F ) , an effect that was highly consistent ( see Figure 3—figure supplement 1 for comparative L2 vein arrays of these two genotypes ) . A similar , albeit less dramatic , difference was also observed between the riΔ2/ri1 ( Figure 3A ) and riΔ2/TM3ri1 ( Figure 3E ) phenotypes . Reflective of the adult wing phenotypes , Kni expression in larval wing was only modestly reduced in riΔ3 homozygotes ( Figure 2F ) , was further reduced in riΔ3/ri1 discs ( Figure 2I ) and virtually undetectable in either riΔ3/TM3ri1 ( Figure 2J ) or riΔ3/riAB trans-heterozygous discs ( Figure 2K ) . As a final test of the CRM-pairing hypothesis we crossed the riΔ2 and riΔ3 mutants to each other and observed a substantial vein-loss phenotype in the riΔ2/riΔ3 trans-heterozygotes , again consistent with a disruption of local chromosomal pairing of the remaining activating elements . Cumulatively , these data provide evidence for a pairing-dependent interaction between CRM sequences on homologous chromosomes that promotes CRM activity ( see Discussion; Figure 3M–R ) In addition to recovering strains of flies carrying the targeted deletions mentioned above , we also identified an ample collection of imprecise or fortuitously generated mutations that most likely resulted from DNA repair by the NHEJ or other break-repair pathways responding to single or double cleavage events at various locations ( Figure 1—figure supplement 1A ) . For example , a deletion ( riΔ3 . 35 ) that spans sequences removed in riΔ4 and includes small indels results in a variable loss of vein phenotype when placed in trans to riAB ( Figure 1—figure supplement 1B , C ) but has a wild-type phenotype when homozygous ( Figure 1—figure supplement 1D ) in contrast to the extra vein phenotype manifest by riΔ4 homozygotes ( Figure 1I ) . Additionally , we recovered mutants which were generated with the use of a single gRNA , gRNA-C , targeting a conserved region ( see Figure 1A and Figure 1—figure supplement 1A ) . They included deletions ( EV 5–1 and EV 7–1 ) that have very similar breakpoints but result in differing degrees of vein-loss ( Figure 1—figure supplement 1E , F ) , and the addition of a single nucleotide at the edge of the ID3 conserved region ( EV 9–2 ) which has a phenotype that varies from wild type to an ri-like vein phenotype ( Figure 1—figure supplement 1G ) . In another case ( riΔ1 . 1 ) , a 65 bp sequence that has been copied and inserted upstream causes a mild ectopic vein phenotype ( Figure 1—figure supplement 1H ) . Perhaps not surprisingly , a deletion removing sequences spanning riΔ1 to a portion of riΔ4 ( riΔ1 . 6 ) gives a composite partial vein loss and ectopic vein phenotypes ( Figure 1—figure supplement 1I , J ) , while a mutant with a deletion spanning riΔ3-Δ4 plus a 25 bp insertion ( riΔ3 . 38 ) causes an extreme vein-loss phenotype similar to that of riAB ( Figure 1—figure supplement 1K , L ) . Although further analysis of several of these mutants will be necessary to fully understand the basis for associated venation phenotypes , we note two salient features of this mutant class . First , a variety of mutations preferentially affect portions of the L2 vein along the proximo-distal ( PD ) axis , revealing a previously unappreciated role of the PD patterning system in contributing to L2 morphogenesis . Second , in several cases , apparently minor variations in the positions of chromosomal breakpoints or small indels , result in markedly altered final vein patterns that may reveal important effects of specific sequences or requirements for precise spacing between functional motifs . Our recovery of several mutants with novel unanticipated phenotypes highlights the strength of an in locus analysis as all these mutations , even those resulting in very mild phenotypes , were readily identified phenotypically . We recently demonstrated an active genetic process , which we refer to as the mutagenic chain reaction ( MCR ) , to generate both somatic mutations and meiotic gene-drive ( Gantz and Bier , 2015 ) . The key feature of this process is the integration of a vector containing a gRNA and a Cas9 transgene at the precise genomic location of the gRNA-directed cleavage site . We also have proposed a split-drive configuration we refer to as CopyCat elements ( Gantz and Bier , 2016 ) , in which the Cas9 source is provided in-trans in a standard Mendelian fashion and the CopyCat element carries one gRNA to target insertion of a gene cassette or two gRNAs to simultaneously trigger cassette insertion and deletion of the region in between the two gRNA cleavage sites . CopyCat cloning vectors could , in principle , be used as versatile elements for germline transformation , which once inserted , are passed down to the progeny in a Super-Mendelian fashion ( >>50% inheritance ) in the presence of a separate Cas9 source that can be later segregated away . This conditional control of Super-Mendelian inheritance can facilitate combinatorial genetic schemes by circumventing constraints imposed by Mendelian inheritance ( i . e . random chromosomal segregation and linkage ) ( Gantz and Bier , 2016 ) . We tested the concept of a two gRNA CopyCat element by generating an active riAB allele . A CopyCat riAB vector carrying the two gRNA-expressing genes required for generating the AB deletion , an eGFP marker gene , flanked by 1 kb homology arms corresponding to the kni locus abutting the two gRNA cut sites was inserted into the genome by co-injection with a Cas9-producing plasmid into embryonic polar plasm ( Figure 4A ) . Transformant lines carrying this and other similar CopyCat constructs described below were recovered at frequencies similar to those typical of germline transformation using either P-element vectors or the ɸC31 recombinase system , exemplifying how Cas9-mediated site-directed transgenesis can serve as a viable alternative to traditional methods of germline transformation . We combined the riCC-AB allele with the previously characterized y1-MCR element , which contained a germline-expressed Cas9 gene , a gRNA targeting the yellow locus , and homology arms that precisely flank the genomic cleavage site . Following the cross of riCC-AB males to y1-MCR females ( Figure 4B ) , resulting female progeny were crossed to w- males to evaluate the transmission of the riCC-AB allele in their progeny , as well as the generation of somatic phenotypes in female progeny . The results of nine such crosses carried out in parallel are shown in Figure 4C , D , all of which demonstrated a high rate of somatic mutagenesis of the yellow locus , indicating the efficient activity of the y-1MCR element . In four crosses , the riCC-AB allele was copied with 100% efficiency to F2 progeny , in three crosses the efficiency averaged 89% , and in two crosses the element was inherited at Mendelian frequencies . Regardless of the rate of germline transmission noted in these various crosses , the observable L2 truncation frequencies was indicative of somatic activity of the riCC-AB CopyCat and Cas9 alleles which were very similar and approximated that observed in their F1 female parent ( ~70% ) . We conclude that in this system the riCC-AB double-cut CopyCat element typically propagates via the germline with high efficiency and that the somatic and germline activities of the riCC-AB element seem to be separable events . This effect is probably due to high levels of Cas9 protein produced by the vasa driver in the egg , that can freely diffuse prior to cellularization , causing mutations affecting large regions of the organism , consistent with observations reported by Port et al . ( Port et al . , 2014 ) . As mentioned above , we identified three highly conserved sequence islands ( termed ‘ID’ for identity ) within the L2 CRM that displayed virtually no nucleotide variation across the entire Drosophilid clan ( Figure 1A magenta , red , and green blocks ID1 , ID2 , ID3; their location in the CRM in Figure 5A , and IDs sequence alignment in Figure 5—figure supplement 1 ) . We used these short conserved ID sequences alone or in combination to identify homologous regulatory regions from yet more distantly related flies in the broader Schizophora group ( Wiegmann et al . , 2011 ) including the housefly ( Musca domestica ) and the Medfly ( Ceratitis capitata ) . Notably in these later species , sequence conservation in the candidate L2-CRMs compared to D . mel . was restricted to the vicinity of the ID islands ( Figure 5A , Figure 5—figure supplement 1 ) . In lieu of the guidance that multi-sequence alignments provided within the Drosophilid clade to identify the likely bounds of the L2-CRM in other species , we included ~200–400 bp before the ID-1 sequence and ~600–1000 bp after the ID-3 sequence for M . domestica and C . capitata , respectively , with a goal of also retaining sequences conserved within the various clades . We inserted candidate orthologous genomic fragments from donor species into a slightly modified riCC-AB CopyCat vector ( Figure 5B ) . This arrangement permits cargo insertion and , if desired , CRE-mediated excision of non-cargo sequences , which also results in loss of the active genetic potential of the transgenic element ( see Figure 5—figure supplement 2 ) . We tested CRM replacements for three fly species in these proof-of-principle experiments ( Figure 5C , compare with Figure 5—figure supplement 2 for control experiments in which non-CRM sequences were deleted by CRE-mediated recombination ) , one distantly related species within the Drosophilid group ( Drosophila grimshawi ) ~45 Mya , and two Schizophora ( M . domestica ( Scott et al . , 2014 ) and C . capitata [Papanicolaou et al . , 2016] ) spanning evolutionary divergence periods of ~60–75 Mya . D . grimshawi is significantly larger than D . mel . , but has a very similar vein pattern , while in both M . domestica and C . capitata the position of the L2 vein is shifted anteriorly relative to other elements of the wing ( note that M . domestica is also much larger than D . mel . ) ( Figure 5C ) . As expected , a control strain homozygous for a CopyCat element that deletes and restores the original D . mel . L2-CRM sequences ( riCC-D . mel . ) exhibited full rescue of the L2 vein leading to Kni expression in third instar discs that was indistinguishable from wild type ( Figure 6C , D; compare to 6A , B ) . Similarly , flies homozygous for the D . grimshawi L2-CRM ( riCC-D . grim . ) exhibited a complete L2 vein in approximately the correct position and strong Kni expression in the expected pattern in wing discs ( Figure 6E , F ) , consistent with the similar relative vein spacing patterns within the Drosophilid clade . We expected that at greater phylogenetic distances beyond the Drosophilid clade that candidate L2-CRM replacements in D . mel . might fail to support Kni expression or rescue of the L2 vein . What we had not anticipated , however , was that the riCC-M . dom . or riCC-C . cap . L2-CRM replacements for M . domestica and C . capitata would result in shifts of the positions of rescued L2-vein segments . Strains homozygous for the riCC-M . dom . or riCC-C . cap . replacement alleles displayed complete ( riCC-M . dom . ) or partially ( riCC-C . cap . ) restored L2 veins that were substantially displaced toward the anterior of the wing relative to the endogenous D . mel . L2 vein ( or the control strains homozygous for the riCC-D . mel . element ) ( Figure 5C ) . In the case of riCC-M . dom . wings , the full vein coursed in a much sharper anterior angle from its normal point of branching from the L3 primordium , resulting in the rescued vein intersecting the wing margin in a position substantially proximal to that of the wild-type L2 vein ( Figure 5C , see figure legend for quantitation ) . Presaging the anterior displacement of the rescued L2 vein in adults , expression of Kni in third instar wing discs was significantly broadened in riCC-M . dom . discs ( Figure 6G , H , N-P ) and pre-pupal wings ( Figure 2—figure supplement 1S–U ) , extending further to the anterior than in wild-type wing primordia , but sharing the same posterior limit abutting the Spalt expression domain as in wild-type discs [Figure 6K–M; Figure 2—figure supplement 1G–H] ) . Kni expression in wild-type M . dom . discs is similarly broadened , suggesting that the pattern observed in the riCC-M . dom . replacement faithfully reproduces the endogenous pattern in its species of origin ( Figure 6—figure supplement 1B , C ) . The exclusion of riCC-M . dom . Kni expression from the central domain of high level Spalt expression suggests that the M . dom . CRM , like that of D . mel . , is subject to Spalt-mediated repression . The net anterior displacement of the adult L2 vein is similar to that resulting from low-level ubiquitous expression of kni in an ri1 mutant background ( Lunde et al . , 1998 ) and may reflect subsequent lateral inhibitory interactions during pupal stages that restrict vein formation to narrow stripes within a broader pro-vein domain ( Biehs et al . , 1998; Sturtevant and Bier , 1995 ) . Vein rescue was only partially complete for homozygotes carrying the riCC-C . cap CRM replacement , which consistently had a rescued central L2 vein segment running parallel to the margin at a much reduced distance than observed for the wild-type D . mel . vein ( Figure 5C , see figure legend for quantitation ) . Since little , if any , Kni expression is restored in riCC-C . cap wing discs ( Figure 6I , J ) or pre-pupal wings ( Figure 2—figure supplement 1F ) , it is likely that the partial vein rescue observed with this construct reflects activity of this CRM at a later developmental stage . This hypothesized kni expression in the L2 primordium is consistent with the observation that the endogenous kni and salm expression patterns in C . capitata ( Figure 6—figure supplement 1 ) and D . mel . are similar . As the observed anterior shifts of L2 veins observed in both the riCC-M . dom . and riCC-C . cap CRM replacements reflect the relative positions of L2 veins in M . domestica and C . capitata , we tentatively conclude that features of the genetic information for proper positioning of the L2 veins in these species reside within kni cis-regulatory sequences themselves and that these L2-CRMs are not merely executing patterning decisions imposed on them by more upstream elements of the wing gene regulatory network . The ability to generate targeted deletions and combinations of such lesions while recovering a myriad of imprecise lesions as a byproduct of CRISPR mutagenesis strategies , fundamentally transforms the process by which CRM function can be analyzed . In this study , we highlight three fundamentally new insights and attendant questions into the activity of a CRM that we had thought we already understood quite well . ( 1 ) Do CRMs on homologous chromosomes cooperate in a pairing-dependent fashion ? The vastly different phenotypes of the riΔ3 deletion ( and to a lesser extent riΔ2 as well ) when homozygous versus in-trans to the larger full CRM riAB deletion was the first major surprise of this current study . In the case of other previously identified partial CRM loss-of-function mutations such as ri1 or ri53j , the phenotypes of homozygotes are only modestly increased when placed in trans to riAB as is typical when a hypermorphic allele is placed over a deletion . The riΔ3/riΔ3 phenotype , however , is nearly wild-type , while that of riΔ3/riAB trans-heterozygotes results in significant vein loss . This type of trans-chromosomal interaction , which we refer to as CRM-synergy , is reminiscent of , the phenomenon of transvection as described originally by Ed Lewis ( Lewis , 1954 ) . In the case of transvection , a regulatory entity ( CRM in modern parlance ) on one chromosome promotes expression of a trait ( activates transcription ) on the other chromosome ( Figure 3N ) , and this trans-interaction between the CRM and basal promoter is abrogated by chromosomal inversions that interfere with chromosomal pairing ( Figure 3O ) . In contrast to transvection , for CRM-synergy we hypothesize that there are interactions between the two CRMs of different chromosomes that mutually reinforce the abilities of both CRM elements to engage either basal promoter . One possible explanation for CRM-synergy in the case of the riΔ2 or riΔ3 deletions is that two mutually reinforcing CRM sub-complexes could form on sequences present in either region deleted in riΔ2 or riΔ3 , and that these complexes normally act in concert to provide full CRM activity ( Figure 3M ) . Perhaps when either of the riΔ2 or riΔ3 deletions are homozygous , the remaining complexes present on the other remaining submodule can sustain CRM function as long as those elements on the two homologous chromosomes are well paired and can benefit from the trans-CRM interaction . When they are placed over the larger deletion , however , this pairing breaks down since the one remaining submodule can no longer interact productively with its homologous partner . This hypothesis is further supported by the observation that the phenotype of riΔ3/ri1 ( Figure 3B , modeled in Figure 3P ) is aggravated by interruption of chromosome pairing by the rearranged TM3 , ri1 balancer chromosome ( Figure 3F , modeled in Figure 3R ) and that riΔ2/riΔ3 trans-heterozygotes display a substantial vein-loss phenotype ( Figure 3C , modeled in Figure 3Q ) . It is also noteworthy that there is only a modest vein loss phenotype manifested in riΔ3/ri1 flies despite the fact that the riΔ3/riAB and ri1/riAB phenotypes are comparable . This effect may arise from the fact that deleted sequences in ri1 overlap with those missing in riΔ3 . Consistent with this hypothesis , the riΔ3/ri53j phenotype is substantially stronger than that of riΔ3/ri1 despite the fact that the ri53j/ri53j phenotype is typically less severe than that of ri1/ri1 . Since ri53j is a point mutant allele that is covered by the riΔ3 deletion , alignment of homologous chromosomes in riΔ3/ri53j trans-heterozygotes would likely be disrupted as compared to riΔ3/riΔ3 homozygotes that entirely lack the mutated nucleotide in ri53j and surrounding sequence . Further analysis will be required to determine the nature of the hypothesized pairing-dependent interaction and to exclude other potential explanations such as 1 ) the potential creation of novel transcription factor binding sites at the junction points of particular deletions which could then alter vein phenotypes by a neomorphic mechanism or 2 ) the ri53j mutation disrupts binding of an activator that normally acts by overcoming the formation of an inhibitory complex on adjacent sequences that are deleted in the riΔ3 mutant ( these most obvious alternative explanations , however , do not readily account for the inversion-dependent differences in the riΔ3/ri1 phenotype nor the failure of riΔ3 and riΔ2 deletions to complement ) . Another question that remains to be addressed is why the ri53j point mutant allele has a stronger homozygous phenotype than the riΔ3 deletion that covers it . One possibility to explore is that the riΔ3 deletes both necessary activator sites ( e . g . the ri53j site ) and repressor sites leading to a weaker homozygous phenotype than deletion of just the activator sites . ( 2 ) Split , cooperative , negative regulatory elements in the L2-CRM . In-locus deletion of sequential regions of the full L2-CRM suggested that two separate regions of the enhancer might contribute to repression of kni expression in peripheral regions of the wing primordium . Remarkably , when these two deletions were combined , a highly synergistic ectopic vein phenotype was observed . The ability to assay the effects of these deletions directly on gene function was a key feature of the experiments that reveal this strong cooperative phenotypic interaction . This observation motivates future studies to understand the mechanism by which the separated CRM sequences might interact . Do they assemble similar protein complexes that act in a redundant fashion ? Might the two domains bind different transcription factor complexes but loop in some way to compete with activators binding to central regions of the CRM or to block a pairing-dependent form of trans-chromosomal cooperative activation ? Another hypothesis to consider is that sequences in the riΔ1 region include not only sites for repressors , but also for inputs that limit anterior kni expression during early prepupal stages . Recall that the double riΔ1+Δ4 and single riΔ4 mutants exhibit comparable levels of ectopic Kni expression in larval stages ( Figure 5H , L ) , but that the zone of strong ectopic Kni expression seems narrowed during pre-pupal stages in riΔ1+Δ4 relative to riΔ4 ( Figure 2—figure supplement 1O , R ) . These observations might help explain why the riΔ+ , Δ4 double mutant has a more extreme ectopic vein phenotype than the riΔ4 single mutant since previous studies have shown that high-level ubiquitous kni expression eliminates expression of both vein and intervein markers , possibly via overly exuberant induction of lateral inhibitory processes . According to this seemingly paradoxical hypothesis , the narrower zone of high level Kni expression observed the riΔ1+Δ4 double mutant may be less subject to vein elimination by such lateral inhibitory influences than the riΔ4 single mutant with persistent high level broad ectopic Kni expression throughout the anterior domain . Further analysis will be required to test this hypothesis . Also , one puzzle that remains unanswered by the current analysis is how the Spalt locus might negatively regulate CRM function in central regions of the wing as we failed to recover any mutations that induced ectopic veins in this territory . Unfortunately , the DNA-binding sites for Spalt proteins ( Salm and Salr ) , if any , in the L2-CRM , are among the few transcription factors for which a well-defined binding site consensus remains to be determined ( de Celis and Ostale , 2017 ) . Perhaps , the failure to recover mutants in all our different experiments that induce ectopic veins in central regions of the wing reflects an interspersion of Sal repressor sites with activator sequences ( e . g . known functionally important Scalloped sites ) ? Further refined deletion analysis of activator domains should help address this question as well as studies to identify bona fide Spalt protein binding sites . Sequence comparisons with the highly diverged L2-CRMs from M . domestica and C . capitata may prove helpful in this effort , as the exclusion of Kni expression from the central Spalt expression domain suggests that at least the riCC-M . dom . CRM retains sensitivity to Spalt-mediated repression , although this effect could be indirect . We note that these new features and models of kni regulation derive from a combination of morphological and in locus gene expression data that were not revealed previously based solely on analysis of traditional L2-CRM-reporter gene fusion constructs ( e . g . [Martín et al . , 2017; Lunde et al . , 2003] ) . ( 3 ) Novel unexpected phenotypes . Approximately half of CRISPR-induced mutations we recovered were the targeted mutations we set out to isolate . The remaining fortuitously generated lesions included imprecise indels at one or both of the targeted cleavage sites or , in some cases , deletions or duplications of short sequences at a distance from the cleavage sites that may reflect an error-prone form of HDR repair based on the presence of micro-homologies ( sometimes referred to as micro-homology-mediated end joining or MMEJ ) . In some cases , these unanticipated mutations generated surprising phenotypes , including L2 deletions biased to proximal or distal regions of the vein , dramatically different phenotypes of deletions resulting from minor differences in endpoints , and in one case the addition of a single base in a non-conserved region of the CRM resulting in a vein-loss phenotype . While the basis for these various unexpected phenotypes remains to be determined , the recovery of a diverse array of ancillary mutations accompanying targeted CRISPR-based mutagenesis offers a potential treasure trove for gaining unbiased insights into CRM function . We previously demonstrated a new CRISPR-based method referred to as the Mutagenic Chain Reaction ( MCR ) that results in efficient copying of genetic elements to the homologous chromosome during meiosis resulting in strong gene drive ( Gantz and Bier , 2015 ) . Similar gene-drive systems have also been shown to transmit efficiently to offspring in mosquitoes ( Gantz et al . , 2015; Hammond et al . , 2016 ) and to copy efficiently in diploid yeast where such mutations can be transmitted subsequently during meiosis ( DiCarlo et al . , 2015 ) . Based on these initial studies , we proposed the design of a new type of transgenesis vector that we refer to as CopyCat elements ( Gantz and Bier , 2016 ) that could be copied efficiently to the homologous chromosome during meiosis in a Cas9-dependent fashion thereby bypassing traditional constraints of Mendelian inheritance , such as those associated with independent chromosomal assortment or genetic linkage . CopyCat elements can contain either a single gRNA that cuts at the site of vector insertion into the genome or two gRNAs that generate a self-propagating deletion of sequences lying between the two gRNA cut sites . In this study , we demonstrate that double cut CopyCat cloning vectors function as efficiently as conceived . In addition , CopyCat vectors can be used routinely to recover site directed genomic insertion making them an attractive alternative vehicle for transgenesis . Furthermore , when CopyCat elements are in presence of a Cas9 source , they can be passed on via the germline with super-Mendelian efficiency , a feature which should substantially simplify and shorten otherwise tortuous genetic schemes needed to assemble complex arrays of transgenes . If active genetics systems can also be developed in vertebrate and plant systems , they could revolutionize genetic manipulations in an array of organisms , in which standard Mendelian constraints pose a yet greater impediment than in fruit flies . Potential caveats for the transmission of active genetic elements: The high frequency of CopyCat transmission in the experiments presented in this study suggest that this should be a generally applicable method . We noted , however , that in two of the nine crosses , transmission of the element was only Mendelian . The basis for the failure of the active copying process in such instances remains to be further evaluated , however , based on previous analysis of MCR transmission in flies ( Gantz and Bier , 2015 ) and mosquitoes ( Gantz et al . , 2015; Hammond et al . , 2016 ) , the most likely explanation is that NHEJ repair can in some instances intervene between the time of egg fertilization and the partitioning of germ cell lineages , which takes place following seven cell divisions in the blastoderm embryo . In Drosophila , the frequency of such NHEJ events seems to be fairly low as revealed by propagation of the y1-MCR occurring at comparable frequencies via eggs that either are preloaded or not with a source of Cas9 ( e . g . eggs from females that carry the MCR element versus eggs from wild-type females fertilized by MCR males ) . In contrast , in mosquitoes there is a large difference in transmission frequency between these two scenarios ( 98–99% chromosome conversion via males versus only 12–25% conversion via females - [Gantz et al . , 2015] ) . The basis for this difference in male/female propagation between the two Dipteran species remains to be determined . However , it may be that for double cut CopyCat elements , which by their nature are more demanding of the HDR repair process than single cut MCRs , that non-HDR events are more frequent . Alternatively , the nuclease targeting of the two gRNA sites may occur asynchronously or the Cas9 enzyme may preferentially bind one gRNA thereby reducing cleavage directed by the second gRNA . In either of these scenarios , if the gRNA-A cuts before the gRNA-B , it could generate an NHEJ event at the location ‘A’ preventing successful HDR-driven chromosomal conversion of the CopyCat element at a subsequent cell cycle , when the gRNA-B has a chance of successfully guiding Cas9 to generate a dsDNA cleavage . Future experiments should resolve this question . Another factor to be kept in mind when using CopyCat elements as a genetic tool is that DNA repair mechanisms involved in copying these elements may be more error prone than standard cellular DNA replication ( Malkova and Haber , 2012 ) . CopyCat CRM-replacement elements should accelerate evo-devo analysis of gene-regulatory networks: We also present proof-of-principle experiments showing that CopyCat elements can also carry gene sequences of interest that replace those deleted in a single genetic event . Such CopyCat elements should greatly facilitate the assembly of multiplex replacements to help identify genetic sequences from divergent species that underlie the morphological diversification . Indeed , we were surprised that replacement of just the kni L2-CRM with candidate homologous regulatory sequences from the housefly or Medfly led to pronounced anterior shifts of the L2 vein typical of vein placement in those donor species , suggesting that alterations in this single down-stream regulatory element are capable of mediating significant morphological changes . The nature of the salient changes in the L2-CRM remain to be investigated . Candidates for negative regulators in peripheral regions of the wing primordium that may contribute to these anterior vein shifts include the repressors Brinker and Optix/Six3 which have been implicated as part of a cross-inhibitory network involved in L2 development ( Al Khatib et al . , 2017; Martín et al . , 2017 ) . Intriguingly , knock-down of Optix function in anterior regions of the wing results in large anterior shifts of the L2 vein ( Al Khatib et al . , 2017; Martín et al . , 2017 ) , although this phenotype results at least in part from decreased proliferation of cells in the anterior domain where Optix is normally expressed ( Martín et al . , 2017 ) . It also will be of significant interest to compare such changes that have occurred in different lineages that have yielded similar alterations in vein positions . For example , will similar or different sequences be identified as the key substrates for the anterior vein shifts we observe in M . domestica versus C . capitata ? Sequencing and selective functional testing of additional species may provide estimates for how often such alternative evolutionary trajectories are taken . Also , how many CRMs of genes acting in the wing gene regulatory network would need to be replaced to transform the D . mel . wing into one resembling a housefly or Medfly ? CopyCat-mediated CRM replacements should greatly enable studies to answer these and other frontier questions . The studies presented here highlight the phenomenal impact of CRISPR-Cas9 mediated genome editing and active genetics in Drosophila , and by extension , most likely in many other organisms . These methods that efficiently generate targeted mutations , as well as a mix of fortuitous mutations in regulatory regions , provide unprecedented genetic leverage in dissecting non-coding DNA functionality . For example , as described above , in locus analysis of CRM function based on a combination of readout of endogenous gene expression at different developmental stages and on final morphological phenotype , revealed features of kni regulation that were not evident from previous CRM-gene-fusion studies . These new insights include: proposed chromosome pairing dependent CRM interactions , identification and analysis of complex cooperative interactions between CRM sub-domains , isolation of novel CRM mutations with unexpected phenotypes , and subtle phenotypic analysis of divergent CRM function during evolution . In addition , the comparable efficiency of CRISPR-based HDR-mediated transgenesis to that of traditional methods offers the possibility of delivering genetic cargo of interest to any desired chromosomal location . It should also be possible to develop genome-spanning collections of active CopyCat elements capable of delivering transgenic elements to a broad array of chromosomal insertion sites in a variety of sexually reproducing organisms as alternatives to Mendelian recombination-based systems . Multiple CopyCat elements could then be rapidly combined in the presence of an unlinked source of Cas9 . Manipulations of this kind should significantly facilitate the import of entire gene regulatory networks from one organism into another to address important questions in evolution and promote the goals of principle-directed synthetic biology . Plasmids were purified using the Qiagen Plasmid Midi kit ( #12191 ) . The gRNA plasmids were co-injected with respective 120 nucleotide ssODNs ( also synthesized de novo in their entirety ) that contained 60 bases homologous to the sequence flanking either side of the targeted deletion . Injection mixes were assembled with two gRNA plasmids ( final concentration: 250 ng/µl each ) and the donor oligo ( final concentration: 100 ng/µl ) in a volume of 50 µl . The mixes for riΔ1-riΔ4 were sent to Best Gene Inc . for injection into their Vasa-Cas9 ( X ) stock ( BDSC #51323 ) while CopyCat constructs were injected into the w1118 stock ( BDSC #5905 ) with a transient source of pHsp70-Cas9 ( pHsp70-Cas9 was a gift from Melissa Harrison and Kate O'Connor-Giles and Jill Wildonger ( Addgene plasmid # 45945 ) . All constructs were fully sequenced prior to injection and after recovery in analyzed transgenic fly stocks . injected flies were singly crossed to the riAB stock , approximately 20 larvae were collected from each vial , and genomic DNA was prepared . The larvae were rinsed in deionized water and homogenized with a motorized pestle in homogenizing buffer prepared according to protocols by Steller et al . ( Steller and Pirrotta , 1986 ) . Genomic DNA from single adult flies were prepared according to protocols by ( Gloor et al . , 1993 ) . PCR screening of deletion mutants: Some crosses were screened by phenotype followed by sequencing of the kni CRM , while others were screened by PCR using primers that would anneal to the sequence flanking the target deletion . In the case of deletions that did not display obvious overt phenotypes across the AB deletion , DNA was prepared from 20 third instar larvae from each vial of single injected F0 flies crossed to the riAB deletion as summarized above , PCR amplified with appropriate primers to detect the expected deletion , and analyzed by gel electrophoresis . Crosses giving the expected deletion band were kept and individual F1 progeny from those crosses were crossed again to the riAB deletion and re-tested . F2 flies again testing positive were then crossed to a TM3 balancer stock , and flies lacking an overt L2 phenotype ( i . e . the non-AB/TM3 flies ) were crossed to each other to establish a stock . Stocks homozygous for the desired deletion were then selected based on the absence of the TM3 balancer . Other primers used to screen for additional deletion events were designed with sequences lying outside of the deletion sequence but inside of the kni enhancer ( AB fragment ) in order to amplify the chromosome containing the newly generated deletions . PCR reactions were assembled with Phusion High-fidelity polymerase from NEB ( #M0530S ) and the genomic DNA previously prepared from collected larvae . PCR products were purified using the QIAquick PCR purification kit ( #28104 ) before sequencing . All crosses using active genetics were performed in accordance to an Institutional Biosafety Committee ( IBC ) approved protocol in a secure ACL2 insectary as previously described ( Gratz et al . , 2013 ) consistent with currently suggested guidelines for laboratory confinement of gene drive systems ( Hammond et al . , 2016; DiCarlo et al . , 2015 ) . Non-CRM sequences flanked by Lox-P sites carried on the CRM-replacement CopyCat vectors were deleted by crossing the DsRed +CopyCat elements to a ubiquitous source of CRE-recombinase ( BDSC #851 ) . The resulting F1 progeny displayed a loss of the DsRed marker as trans-heterozygotes indicating early developmental CRE-mediated deletion of the cassette resulting in somatic as well as germline deletion of the DsRed cassette . These F1 flies were crossed to balancer stocks and isogenic lines were established from individual DsRed- F2 progeny . Molecular analysis of these deletion stocks revealed the clean predicted deletions of the DsRed cassette . Chromosomes carrying these unmarked elements were homozygosed by elimination of balancer chromosomes . Homozygous CRM-only replacements were analyzed for their phenotypes ( Figure 5—figure supplement 1 ) , which displayed little if any differences from their DsRed +parent lines ( Figure 5 ) . Drosophila wings were dissected in 90% ethanol and mounted in 100% Canada balsam . 3rd instar larvae were dissected on ice in 0 . 1% Tween-PBS and fixed with 2% PFA in Brower buffer for 1 hr at 4˚C . Discs were stained in a mix of 1:1000 guinea pig anti-Knirps , 1:200 mouse anti-Engrailed , 1:500 mouse anti-Delta , 1:1000 rabbit anti-Spalt overnight at 4˚C . Samples were mounted in Slowfade diamond anti-fade mountant ( #S36963 ) and imaged on a Leica SP5 confocal microscope .	Gregor Mendel was an Austrian monk and botanist whose work with pea plants in the 19th century founded the field of genetics . Though he was not aware of genes at the time , Mendel essentially worked out that pea plants had two copies of each gene , and that each copy had a 50% chance of being passed on to any one offspring . Yet not all genes actually follow this pattern of inheritance . In 2015 , researchers reported that they had used components of the CRISPR/Cas9 genome editing system to edit genes so that they could propagate in a “Super-Mendelian” fashion . Indeed , when it was engineered into fruit flies , any parent carrying this active genetic element passed it on to almost every offspring . Active genetic elements have potential applications in many different fields of scientific research . These include providing new ways to explore how genes control the formation and activity of different organisms . Now , Xu , Gantz et al . – including the two researchers involved in the 2015 work – have used a new active genetic element called a CopyCat element and more traditional genome editing to analyze the control of a gene that coordinates the formation of a simple structure in a fruit fly – a vein in the wing . The goal was to understand which sections of DNA controlled where and when genes are activated to result in this structure being reliably located in its correct position . First , Xu , Gantz et al . used genome editing to make mutations in a stretch of DNA that regulates the gene involved in wing vein formation . The effects of these mutations unexpectedly suggested that pairs of chromosomes might be interacting to control the activity of this gene . This was something that had not been seen before , which shows the advantage of editing a gene’s regulatory sequence at its normal location within the genome . Next , Xu , Gantz et al . used the CopyCat tool to delete the regulatory sequence and replace it with sequences from three other species of flies . When the sequence was replaced with that of a housefly , a complete vein formed but it was further forward than normal for a fruit fly , and more closely matched the position of the wing vein in a housefly . These findings show how gene activity can affect the position of a simple structure; they also suggest that this strategy could help scientists to understand how the genomes of different species have evolved . Xu , Gantz et al . hope these advances will encourage other researchers to use active genetic elements in a broad range of organisms to enable and accelerate their research . Since these tools fundamentally change the rules of genetic inheritance , they have many applications beyond research too . These applications are not without their risks and would need careful consideration , but could include engineering wild mosquito populations to combat diseases like malaria , dengue fever , chikungunya and Zika .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "tools", "and", "resources", "genetics", "and", "genomics" ]	2017	CRISPR/Cas9 and active genetics-based trans-species replacement of the endogenous Drosophila kni-L2 CRM reveals unexpected complexity
The entorhinal cortex ( EC ) is the primary site of interactions between the neocortex and hippocampus . Studies in rodents and nonhuman primates suggest that EC can be divided into subregions that connect differentially with perirhinal cortex ( PRC ) vs parahippocampal cortex ( PHC ) and with hippocampal subfields along the proximo-distal axis . Here , we used high-resolution functional magnetic resonance imaging at 7 Tesla to identify functional subdivisions of the human EC . In two independent datasets , PRC showed preferential intrinsic functional connectivity with anterior-lateral EC and PHC with posterior-medial EC . These EC subregions , in turn , exhibited differential connectivity with proximal and distal subiculum . In contrast , connectivity of PRC and PHC with subiculum followed not only a proximal-distal but also an anterior-posterior gradient . Our data provide the first evidence that the human EC can be divided into functional subdivisions whose functional connectivity closely parallels the known anatomical connectivity patterns of the rodent and nonhuman primate EC . The entorhinal cortex ( EC ) is a major hub within the medial temporal lobe that mediates hippocampal-neocortical communication ( Buzsáki , 1996; Lavenex and Amaral , 2000 ) . Together with the adjacent perirhinal cortex ( PRC ) and parahippocampal cortex ( PHC ) , these brain regions form a neural circuitry that is critical for learning and memory ( Suzuki and Eichenbaum , 2000; Eichenbaum et al . , 2007 ) . Virtually nothing is known about how hippocampal and neocortical connectivity with the EC is organized in humans . This lack of knowledge has significantly limited the development of neurobiological theories of memory and navigation and our understanding of the clinical impact of localized EC damage in the early stages of neurodegenerative conditions such as Alzheimer's disease ( AD ) . Neuroanatomical evidence from studies in rodents suggests that there are two parallel input pathways that convey spatial and non-spatial input into the hippocampus via the EC ( Knierim , 2006; van Strien et al . , 2009; Witter et al . , 2014 ) . Specifically , spatial information is conveyed from the postrhinal cortex ( POR , thought to be homologous to the PHC in primates ) , which shows preferential connectivity with ‘medial EC’ ( MEC ) . In contrast , non-spatial information is conveyed from the PRC to the ‘lateral EC’ ( LEC ) . LEC and MEC , in turn , are differentially connected with hippocampal subfields ( i . e . , subiculum and CA1 ) along the proximo-distal ( transverse ) axis . Projections of the LEC preferentially target the region close to the border between CA1 and subiculum ( distal CA1 and proximal subiculum ) , whereas the MEC preferentially projects to proximal CA1 and distal subiculum ( e . g . , Witter et al . , 2000a; Henriksen et al . , 2010 ) . These partially segregated pathways have been differentially associated with the processing of , and memory for , object and context information ( e . g . , Hunsaker et al . , 2013; for review see; Knierim et al . , 2014; Ranganath and Ritchey , 2012; Ritchey et al . , in press ) . Moreover , a recent network analysis on the rat connectome highlighted the LEC as a major cortical hub , forming the richest set of association connections of any cerebral cortical region ( Bota et al . , 2015 ) . Notably , although terminology for EC subdivisions in the rat emphasizes the lateral to medial axis , these areas do not differ solely with respect to their position in relation to the hippocampal formation and the rhinal fissure ( Witter et al . , 2000a ) . In actuality , LEC occupies the rostrolateral portion of the EC , whereas MEC occupies the caudomedial portion of the EC . In primates , ventral hippocampus and the adjacent EC are situated in a relatively more rostral position in the anterior temporal lobe . Although the position of the EC on the cortical surface and the orientation of anatomical axes differ across species , the relative topography of EC connectivity seems to be preserved in nonhuman primates . Anatomical studies suggest that the PRC is predominantly interconnected with the anterior third of the EC , whereas the PHC is predominantly interconnected with approximately the posterior two-thirds ( Suzuki and Amaral , 1994 ) . In addition , PRC/PHC connectivity with EC differs between lateral and medial domains ( Suzuki and Amaral , 1994 ) . A functional distinction between anterior and posterior EC has been substantiated by a single-unit recording study of grid-cell-like neurons in nonhuman primates ( Killian et al . , 2012 ) . To summarize , findings from rodents and nonhuman primates suggest that connectivity of the human EC might differ along the longitudinal and lateral-medial axis , but because anatomical tracing studies cannot be performed in humans , direct evidence for this idea is lacking . Recent structural and functional MRI studies have assumed a lateral-medial distinction in humans according to the rodent nomenclature of LEC and MEC . However , whether such a lateral-medial dissociation of EC connectivity exists in humans and/or whether connectivity differs along the anterior-posterior EC axis has not been assessed so far . Numerous studies have demonstrated that networks of brain regions linked by direct and indirect anatomical connections exhibit temporally coherent , low-frequency fluctuations in blood oxygenation level dependent ( BOLD ) functional magnetic resonance imaging ( fMRI ) data during both the resting and task states . Recent work has demonstrated that spatially contiguous but anatomically distinct brain regions can be reliably differentiated based on functional connectivity profiles measured with BOLD fMRI ( for a review see Fox and Raichle , 2007 ) . Using fMRI at 3 Tesla , human resting-state fMRI studies ( Kahn et al . , 2008; Libby et al . , 2012 ) have reported reliable differences in connectivity between the PRC and PHC with the hippocampus . Whereas the PRC showed higher functional connectivity with the anterior hippocampus , the PHC showed stronger connectivity with the posterior hippocampus , a dissociation that was most evident for subiculum and CA1 subfields ( Libby et al . , 2012 ) . Due to limitations in signal-to-noise ratio ( SNR ) and spatial resolution , these studies were unable to assess functional connectivity with the EC . Here , we used ultra-high resolution fMRI at 7 Tesla to characterize the functional organization of the human EC . The high SNR of MRI data collected at 7 Tesla makes it feasible to acquire BOLD fMRI data at an unprecedented level of anatomical detail ( e . g . , Maass et al . , 2014 ) . To determine the topographic organization of EC connectivity in humans , we conducted two experiments in which fMRI data were acquired with a resolution of 0 . 8 mm ( isotropic ) . Notably , this level of spatial resolution is more than six times higher than in previous high-resolution fMRI studies that investigated intrinsic connectivity within the MTL ( Lacy and Stark , 2012; Libby et al . , 2012 ) . In two independent samples , we examined correlations of activation between individually anatomically defined PRC vs PHC seeds and the EC along its anterior-posterior and lateral-medial axis . In addition , we tested whether functionally distinct EC subregions also exhibited differential connectivity with the subiculum along a transverse or longitudinal hippocampal axis . In addition , we also analyzed connectivity profiles of PRC and PHC seeds with the subiculum . The present study , which focuses on patterns of EC connectivity within the MTL is complemented by a study by Navarro Schröder et al . ( 2015 ) , which examined connectivity topography of the EC with extended cortical networks outside of the MTL . We report the first detailed topographic parcellation of the human EC on the basis of its functional connectivity with neocortical and hippocampal subregions . In two independent samples , our analyses revealed that anterior-lateral and posterior-medial EC subregions ( al-EC and pm-EC , respectively ) exhibited distinct patterns of intrinsic functional connectivity with regions in the neocortex ( PRC and PHC ) and hippocampal formation ( subiculum ) . Specifically , the al-EC region could be delineated on the basis of preferential connectivity with PRC , whereas the borders of pm-EC were derived from connectivity with PHC . Al-EC and pm-EC , in turn , were found to have preferential connectivity with proximal and distal subiculum , respectively . Moreover , the pattern of subiculum connectivity with al-EC and pm-EC was partially distinct from its connectivity with PRC and PHC . A schematic summary of functional connectivity gradients in the subiculum related to PRC/PHC seeds and EC subdivisions is illustrated in Figure 5 . These results reveal the functional topography of the human EC as a gateway between neocortex and hippocampus and show remarkable accordance with principles known from anatomical studies of rodents ( rostrolateral vs caudomedial; for reviews see Witter et al . , 2000a; van Strien et al . , 2009 ) and studies of nonhuman primates ( anterolateral vs posteromedial; see e . g . , Witter and Amaral , 1991; Suzuki and Amaral , 1994 ) . As we describe below , these data provide a link between basic and translational research on the human medial temporal lobes ( Small et al . , 2011; Ranganath and Ritchey , 2012 ) and results from detailed circuit level analyses of the rodent hippocampal formation ( e . g . , Moser and Moser , 2013 ) . 10 . 7554/eLife . 06426 . 011Figure 5 . Schematic summary of functional connectivity gradients in the subiculum related to PRC/PHC seeds and EC subdivisions . ( A ) Functional connectivity analyses revealed preferential connectivity of PRC ( red ) with the anterior-lateral EC and PHC ( blue ) with the posterior-medial EC . Regarding the subiculum , PRC showed strongest connectivity with most anterior and proximal parts , whereas PHC showed strongest connectivity with most posterior and distal parts of the subiculum . ( B ) Anterior-lateral ( red ) and posterior-medial ( blue ) EC exhibited a similar dissociation in connectivity with the subiculum along its transverse ( proximal-distal ) axis but there was no trend for a dissociation of entorhinal connectivity along the longitudinal axis of the subiculum . DOI: http://dx . doi . org/10 . 7554/eLife . 06426 . 011 Previous fMRI studies have used functional connectivity analyses on data collected at 3T to characterize topographic patterns of connectivity between the PRC , PHC , and hippocampal subfields ( Lacy and Stark , 2012; Libby et al . , 2012 ) . These studies have generally found that PRC and PHC exhibit different patterns of connectivity along the longitudinal axis of the hippocampus . Unfortunately , these studies could not address the topographic organization of connectivity within the EC , possibly due to limitations in resolution and SNR . The present results demonstrate that the enhanced resolution and sensitivity of ultra-high field fMRI can overcome these limitations and reveal fine-grained topographical patterns in connectivity . Three-dimensional plots of entorhinal connectivity preferences revealed a gradient of decreasing PRC and increasing PHC connectivity running from anterior-lateral to posterior-medial EC . It is notable that , by training a pattern classifier on the coordinates of EC voxels that showed preferential connectivity with PRC or PHC within a subset of participants , we could reliably predict these voxels in the remaining participant . This finding indicates that the topography of neocortical connectivity within the EC is highly conserved across participants , which , in turn , could indicate fundamental functional differences between the two EC subdivisions . Two recent fMRI studies reported evidence for task-related activation differences between lateral and medial sections of EC in humans ( Schultz et al . , 2012; Reagh and Yassa , 2014 ) . Schultz and colleagues ( 2012 ) reported differential activation in medial and lateral sections of EC during scene and face processing in a working memory task . Reagh and Yassa reported preferential activation in a medial section of EC during mnemonic discrimination of spatial locations and preferential activation in a lateral section of EC during mnemonic discrimination of objects . This functional dissociation was observed by splitting the EC into equally-sized lateral and medial parts according to the rodent terminology of ‘LEC’ and ‘MEC’ . Notably , they also found a trend towards a dissociation between anterior and posterior EC after a similar equal division along the longitudinal axis . Our data help to explain these findings by empirically demonstrating that al-EC and pm-EC exhibit differential functional connectivity with PRC and PHC . Numerous fMRI studies have shown that PHC is preferentially engaged in memory tasks that involve scenes , spatial or context information , whereas PRC is preferentially engaged in memory tasks that involve object or item information ( see Ranganath and Ritchey , 2012; Ritchey et al . , in press; for review ) . Thus , it makes sense that EC subregions that interact predominantly with PRC or PHC also differentially participate in item and context processing . However , our data also suggest that a simple lateral-medial distinction does not capture the functional organization of EC . Future fMRI studies ( as well as structural MRI studies , e . g . , Khan et al . , 2014 ) could more effectively study the EC by using the high-consistency pm-EC and al-EC masks derived from our data ( see also ‘Landmarks for delineation of al-EC and pm-EC’ ) , or by using functional connectivity metrics to identify subject-specific EC subregions . Concurrent with our investigation , Navarro Schröder and colleagues ( Navarro Schröder et al . , 2015 ) also studied the functional organization of the human EC . Whereas the present study focused on MTL connectivity , their study focused on functional connectivity of the EC with large-scale cortical networks . Consistent with our study , they also differentiated between anterior-lateral and posterior-medial EC subregions based on differential global network connectivity in resting state and task fMRI data ( Navarro Schröder et al . , 2015 ) . They found that the al-EC exhibited stronger connectivity with regions in an anterior-temporal cortical system including medial-prefrontal and orbitofrontal cortex , whereas the pm-EC exhibited stronger connectivity with regions in a posterior-medial system including regions in occipital and posterior-parietal cortex . Analyses of task fMRI data revealed that al-EC activity was enhanced during processing of object information and pm-EC activity was enhanced during processing of scenes . Their findings are consistent with the idea that al- and pm-EC may be related to two cortico-hippocampal networks that support distinct types of memory ( Ranganath and Ritchey , 2012; Ritchey et al . , in press ) . Although the EC is a major gateway for the hippocampus , neocortical regions such as PRC and PHC also have direct reciprocal connectivity with CA1 and subiculum ( Naber et al . , 1999 , 2001; Agster and Burwell , 2013 ) . Our analyses revealed that the topographic differences in subicular connectivity with PRC vs PHC along the hippocampal transverse axis paralleled the differences of subicular connectivity with al-EC vs pm-EC . Whereas al-EC and PRC showed stronger connectivity with proximal subiculum , pm-EC and PHC showed stronger connectivity with distal subiculum . In contrast to the transverse axis , PRC/PHC vs al-EC/pm-EC connectivity profiles differed along the longitudinal hippocampal axis . For al-EC and pm-EC , there was no evidence or trend for an anterior-posterior dissociation , compatible with connectivity of LEC and MEC in rodents ( Naber et al . , 1999 , 2001; Witter , 2006; O'Reilly et al . , 2013 ) . In contrast , the most anterior subiculum showed stronger connectivity with PRC than PHC , whereas in one data set the most posterior subiculum ( in the hippocampal body [HB] ) showed stronger connectivity with PHC than PRC . This finding replicated the direct anatomical connectivity profiles observed in rodents ( Naber et al . , 1999 , 2001; Agster and Burwell , 2013 ) . Such an anterior-posterior dissociation of hippocampal connectivity accords with findings from human resting-state fMRI studies that investigated functional connectivity profiles of PRC and PHC ( rather than EC ) with hippocampal subfields ( Libby et al . , 2012 ) . These functional connectivity data suggest that there might be two parallel cortico-hippocampal pathways in humans — one via the EC and one that is direct . The differences in the topographic organization of EC-subicular connectivity and PRC/PHC-subicular connectivity could have important functional implications . One implication is that the EC is not a simple anatomical extension of the PRC and PHC . If that were the case , we would not have observed any reliable difference between neocortical-hippocampal connectivity profiles and EC-hippocampal connectivity profiles . These results add support to the notion that the EC is more than a mere cortico-hippocampal relay ( e . g . , Lavenex and Amaral , 2000; de Curtis and Paré , 2004 ) . One possibility is that this organization might allow a comparison between EC-gated hippocampal memory signals with direct neocortical input ( e . g . , Naber et al . , 1999 ) . Furthermore , the diffuse nature of LEC/MEC projections along the anterior-posterior hippocampal axis and a structured gradient of direct PRC/POR projections that has been identified in rodents could allow for integration of information across both processing streams ( Burwell , 2000; Witter et al . , 2000b; Agster and Burwell , 2013 ) . Results from the present study may be pertinent to understanding memory impairment in clinical conditions that compromise the structural integrity of the medial temporal lobes , including neurodegenerative diseases such as AD and frontotemporal lobar degeneration , temporal lobe epilepsy , depression , schizophrenia , developmental amnesia and ischemia . In AD , for instance , tau pathology emerges in lateral regions of the EC early in the course of the disease ( Braak and Braak , 1991; Braak and Del Tredici , 2004 ) . Analyses of functional connectivity can potentially reveal how EC degeneration in the early stages of AD could impact the functional organization of distributed cortical networks ( Khan et al . , 2014; also see; La Joie et al . , 2014 ) and also shed light on the transsynaptic progression of pathology in AD . To summarize , the results of the present study provide a detailed description of the organization of functional connectivity within the human EC . Based on differential functional connectivity with PRC , PHC and subicular subregions , our data , along with those of Navarro Schröder et al . ( in review ) , demonstrate that the human EC can be reliably subdivided into anterior-lateral and posterior-medial subregions that could be critical nodes in two cortico-hippocampal processing pathways . Future studies can apply the high resolution functional connectivity analyses to differentiate the roles of al-EC and pm-EC in memory and alterations of EC connectivity in AD and other neurodegenerative diseases . Two independent samples of 21 and 22 young , healthy subjects underwent high resolution fMRI scanning ( Exp . 1: mean age 26 ± 3 . 6 yrs , 12 male; Exp . 2: mean age 28 ± 3 . 9 yrs , 7 male ) . Exclusion criteria were metallic implants ( other than standard dental implants ) , tinnitus , known metabolic disorders or a history of neurological or psychiatric disorders . Both studies were approved by the local ethics committee of the University Magdeburg . All subjects gave written informed consent and consent to publish prior to participation and received monetary compensation for participation . Six subjects from Experiment 1 and six from Experiment 2 were excluded due to strong dropouts in the PRC and/or EC or due to severe movement artifacts . Functional connectivity analyses were performed on the residuals of task data after extraction of task effects ( NExp . 1 = 15 , NExp . 2 = 14 ) . MRI data were acquired using a 7T MR system ( Siemens , Erlangen , Germany ) . A 32-channel head coil was used ( Nova Medical , Willmington , MA ) . First , a high-resolution whole head MPRAGE volume ( TE = 2 . 8 ms , TR = 2500 ms , TI = 1050 ms , flip angle = 5° , resolution 0 . 6 mm isotropic ) was acquired . Subsequently , the fMRI session was run . T2-weighted gradient echo planar images ( EPIs ) were acquired with an isotropic resolution of 0 . 8 mm ( 28 axial slices , TE = 22 ms , TR = 2000 ms , FOV 205 mm , matrix 256 × 256 , partial Fourier 5/8 , parallel imaging with grappa factor 4 , bandwidth 1028 Hz/Px , echo spacing 1 . 1 ms , echo train length 40 , flip angle 90° ) . The slices were acquired in an odd–even interleaved fashion oriented parallel to the hippocampus long axis . The fMRI session comprised 1 run ( 13 min ) with 370 EPI volumes in Experiment 1 and 4 runs ( 13 . 5 min each ) with 400 EPI volumes in Experiment 2 . EPIs were distortion corrected using a point spread function mapping method ( In and Speck , 2012 ) and motion corrected during the online reconstruction . Finally , a high-resolution partial structural volume was acquired ( T2-weighted imaging , TE = 18 . 5 ms , TR = 680 ms , resolution 0 . 33 mm × 0 . 33 mm , 45 slices , slice thickness 1 . 5 mm + 25% gap , FOV 212 mm × 179 mm , matrix 640 × 540 ) , with a slice alignment orthogonal to the hippocampus main axis . Total MRI duration was around 60 min in Experiment 1 and 100 min in Experiment 2 . FMRI data pre-processing and statistical modeling was done in SPM8 ( Wellcome Department of Cognitive Neuroscience , University College , London , UK ) . The pre-processing included slice timing correction and smoothing with a 1 . 5 mm full-width half-maximum Gaussian kernel ( FWHM < 2 × voxel size ) to keep high anatomical specificity . Outliers in average intensity and/or scan-to-scan motion were identified using the ARTRepair toolbox for SPM ( Percent threshold in global intensity: 1 . 3 , movement threshold: 0 . 3 mm/TR ) and included as spike regressors . To remove task effects , general linear models were run ( including all task conditions and the movement parameters ) and the residual images were saved for subsequent intrinsic functional connectivity analyses . Based on previous studies suggesting a linear superposition of task activity and spontaneous BOLD fluctuations , removing task-induced variance of event-related fMRI data should yield a remaining residual signal similar to ‘continuous’ resting state data ( e . g . , Fox et al . , 2006 ) . Although quantitative differences between residuals derived from task data and continuous resting state data have been reported ( Fair et al . , 2007 ) , in qualitative terms , patterns of functionally connected regions have been shown to be remarkably consistent ( Fair et al . , 2007; Lacy and Stark , 2012 ) . In order to analyze PRC vs PHC seed-to-voxel connectivity , we manually segmented PRC and PHC regions of interest ( ROIs ) for each subject on the individual high resolution MPRAGEs ( which had been bias-corrected and coregistered to the individual mean EPIs ) . Furthermore , the EC and the subiculum were labelled on the T1-group template in order to analyze PRC vs PHC connectivity topography within the EC and al-EC vs pm-EC as well as PRC vs PHC connectivity topography within the subiculum , respectively , at group level ( individual beta-maps were registered to the template ) . ROIs were traced on consecutive coronal slices bilaterally using MRIcron ( Chris Rorden , Version 4 April 2011 ) . Tracing of the EC started anteriorly at the first slice where the amygdala was visible . Caudally the EC moves along the parahippocampal gyrus until the collateral sulcus disappeared ( Fischl et al . , 2009 ) , typically at the level where also the HH ends , merging into the PHC . At anterior levels , the EC borders the amygdala nuclei medially ( Fischl et al . , 2009 ) . When the gyrus ambiens disappears and the hippocampal fissure opens , the EC borders the parasubiculum medially . This EC-subiculum boundary was located at the angle formed by the most medial extent of both subiculum and EC ( Wisse et al . , 2012 ) . The opening of the collateral sulcus typically coincides with the lateral border of the EC ( Fischl et al . , 2009 ) , and was chosen as lateral limit . The PRC , which laterally abuts the EC , was defined as the region between the medial and lateral edges of the collateral sulcus ( covering medial and lateral banks ) . Contrary to other markings schemes for the EC and PRC ( Insausti et al . , 1998 ) , we did not mark the part of the EC within medial banks of the collateral sulcus that depends on the depth of the collateral sulcus , since this border shows remarkable within and between subject variability and is also sometimes difficult to identify due to partially occurring susceptibility artefacts . For the purpose of the current study , we further deleted PRC mask voxels that were directly neighbored to the EC/PRC border ( leaving a gap of approx . 2 voxels ∼1 mm ) to avoid autocorrelations between PRC seed voxels and our target EC ROI . Segmentation of the PHC started one slice after the disappearance of the collateral sulcus , directly posterior to PRC and EC ( approx . at the level where the HB starts ) . Labeling was continued posteriorly , ending on the last slice where the inferior and superior colliculi were jointly visible . The PHC was delineated as the region between subiculum ( medial border ) and the deepest point of the collateral sulcus ( Zeineh et al . , 2001 ) . The subiculum was labeled on the T1-template ( 0 . 8 mm isotropic resolution ) in the HH and body ( until the colliculi disappeared ) according to the segmentation protocol of ( Wisse et al . , 2012 ) . Note that we did not segment the hippocampal tail as borders were difficult to identify . For division of EC and subiculum masks into anterior-posterior and lateral-medial or proximal-distal subregions , see ‘Second Level Analyses’ ( Univariate ) . Subsequently , PRC and PHC masks were coregistered and resliced to the individual mean EPI images and manually adjusted to achieve a precise overlay on the functional data . To exclude voxels susceptible to signal dropout , for each ROI , voxels with mean intensity ( across timepoints ) < 2 SD from the mean intensity ( across voxels ) in an ROI were removed from the ROI ( Libby et al . , 2012 ) . Thresholding led to the rejection of no more than 5% of voxels in any ROI . Additionally , areas of PRC were occasionally subject to distortion artifact , and these voxels were manually deleted from ROIs . These adjusted and thresholded ROIs were used as seed regions for the functional connectivity analyses . Probabilistic white matter ( WM ) and cerebral spinal fluid ( CSF ) masks were generated by automated segmentation ( SPM8 , ‘New Segment’ ) of the co-registered MPRAGE images and thresholded at p ( tissue ) > 0 . 95 . We performed seed-to-voxel correlational analyses on the native ( preprocessed , unnormalized ) residual fMRI data using the conn-toolbox ( Whitfield-Gabrieli and Nieto-Castanon , 2012 ) . First , functional connectivity patterns of PRC vs PHC seeds with the EC were analyzed . For each functional connectivity analysis , seed regions' average time series were generated as regressors of interest . As covariates of no interest , WM and CSF time series and subjects' realignment parameters ( including spike regressors ) were included to account for physiological noise and movements , respectively . Functional data were band-pass filtered for frequencies of 0 . 01–0 . 1 Hz . Bivariate correlations were computed , resulting in beta maps containing Fisher-transformed correlation coefficients . To perform group analyses , beta maps were registered to the group-specific T1 template ( see below ) and Z-standardized . In order to enable precise cross-participant alignment for hippocampal and parahippocampal regions , we used Region of Interest-Advanced Normalization Tools ( ROI-ANTS [Klein et al . , 2009; Yassa and Stark , 2009; Avants et al . , 2011] ) . First , the Oxford Centre for Functional MRI of the Brain ( FMRIB ) software library ( FSL 5 . 0 . 6 [Smith et al . , 2004] ) was used to register each single participant's structural MPRAGE and the mean functional EPI using epi_reg , a command-line program that belongs to the FMRIB's linear registration tool ( FLIRT v6 . 0 [Jenkinson and Smith , 2001] ) and was specifically written to register EPI images to structural images . Second , a study-specific template was created ( from individual MPRAGE images of Exp . 1 ) using the buildtemplateparallel . sh command-line script within ROI-ANTS ( Cross-Correlation similarity metric [Avants et al . , 2010] ) . Although the resulting alignment parameters already allow for a good registration to the template , we further improved normalization for the MTL regions by adding landmarks to the template . Therefore , the HH ( on the first slice on which it appears ) , EC ( on the first four consecutive slices , starting on the HH slice ) , the HB and the PHC ( same slices as HB ) were labeled on the T1-template as landmarks for the subsequent label-guided alignment . Similarly , subject-specific ROIs were drawn on the individual MPRAGEs to match the template priors . Third , the expectation-based point set registration ( ‘pse’; step size: SyN[0 . 5] ) was used to register the individual MPRAGEs on the T1-template based on the labeled points sets ( = MTL masks ) . The resulting transformation matrix was then applied to each participant's beta map as well as to the MTL masks in order to verify alignment precision . Finally , the aligned beta images were submitted to second-level group analyses .	In the early 1950s , an American named Henry Molaison underwent an experimental type of brain surgery to treat his severe epilepsy . The surgeon removed a region of the brain known as the temporal lobe from both sides of his brain . After the surgery , Molaison's epilepsy was greatly improved , but he was also left with a profound amnesia , unable to form new memories of recent events . Subsequent experiments , including many with Molaison himself as a subject , have attempted to identify the roles of the various structures within the temporal lobes . The hippocampus—which is involved in memory and spatial navigation—has received the most attention , but in recent years a region called the entorhinal cortex has also come to the fore . Known as the gateway to the hippocampus , the entorhinal cortex relays sensory information from the outer cortex of the brain to the hippocampus . In rats and mice the entorhinal cortex can be divided into two subregions that have distinct connections to other parts of the temporal lobe . The ‘medial entorhinal cortex’ is the subregion nearest the centre of the brain , and it predominantly connects to parahippocampal cortex , which is involved in processing visual scenes . The other subregion , the ‘lateral entorhinal cortex’ , is to the left or right of the center and has particularly strong connections to the perirhinal cortex , which is involved in the memory of objects . The two subregions are also connected to different parts of the hippocampus . For many years researchers had assumed that the connectivity of the human entorhinal cortex was quite similar to that observed in rats and mice . However , it was not possible to check this as the entorhinal cortex measures less than about 1 cm across , which placed it beyond the reach of most commonly available brain-imaging techniques . Now , two independent groups of researchers have used ultra high-resolution functional magnetic resonance imaging ( fMRI ) to reveal a more complex structure in humans . The fMRI data reveal that the entorhinal cortex is divided into an anterior-lateral ( to the front and at the side ) subregion and a posterior-medial ( to the back and at the centre ) subregion in humans . One of the groups—Maass , Berron et al . —used the imaging data to show that the anterior-lateral and posterior-medial subregions of the entorhinal cortex form distinct patterns of connections with the perirhinal cortex and the parahippocampal cortex , as well as with different parts of the hippocampus . The other group—Navarro Schröder , Haak et al . —studied functional connections across the whole neocortex to come to the same conclusions . The discovery of these networks in the temporal lobe in humans will help to bridge the gap between studies of memory in rodents and in humans . Given that the lateral entorhinal cortex is one of the first regions to be affected in Alzheimer's disease , identifying the specific properties and roles of these networks could also provide insights into disease mechanisms .	[ "Abstract", "Introduction", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Functional subregions of the human entorhinal cortex
Strict L-chiral rejection through Gly-cisPro motif during chiral proofreading underlies the inability of D-aminoacyl-tRNA deacylase ( DTD ) to discriminate between D-amino acids and achiral glycine . The consequent Gly-tRNAGly ‘misediting paradox’ is resolved by EF-Tu in the cell . Here , we show that DTD’s active site architecture can efficiently edit mischarged Gly-tRNAAla species four orders of magnitude more efficiently than even AlaRS , the only ubiquitous cellular checkpoint known for clearing the error . Also , DTD knockout in AlaRS editing-defective background causes pronounced toxicity in Escherichia coli even at low-glycine levels which is alleviated by alanine supplementation . We further demonstrate that DTD positively selects the universally invariant tRNAAla-specific G3•U70 . Moreover , DTD’s activity on non-cognate Gly-tRNAAla is conserved across all bacteria and eukaryotes , suggesting DTD’s key cellular role as a glycine deacylator . Our study thus reveals a hitherto unknown function of DTD in cracking the universal mechanistic dilemma encountered by AlaRS , and its physiological importance . D-aminoacyl-tRNA deacylase ( DTD ) is a key factor that keeps chiral errors away from the translational machinery by allowing only L-amino acids to form proteins and has therefore been implicated in perpetuation of homochirality in the protein world ( Calendar and Berg , 1967; Soutourina et al . , 1999 , 2000 ) . The design principle by which this remarkable configurational specificity is achieved by DTD involves only strict L-chiral rejection , rather than D-chiral selection . An invariant cross-subunit Gly-cisPro motif forms the structural and mechanistic basis for DTD’s enantioselection ( Ahmad et al . , 2013 ) . Thus , the architecture of DTD’s chiral proofreading site is such that it cannot prevent misediting of achiral glycine charged on tRNAGly and seems to have an inherent flaw . The glycine ‘misediting paradox’ is , however , effectively resolved through protection of the cognate achiral substrate by elongation factor thermo unstable ( EF-Tu ) ( Routh et al . , 2016 ) . While occasional chiral errors that occur during aminoacylation are cleared by DTD to ensure the accuracy of aminoacyl-tRNAs present in the cellular pool , a major role is played by editing functions associated with about half of the 20 aminoacyl-tRNA synthetases ( aaRSs ) to rectify the incorrect pairing of a similar non-cognate L-amino acid with a tRNA ( Guo and Schimmel , 2012; Ibba and Soll , 2000 ) . These aaRSs can edit the non-cognate amino acid at the aminoacylation site itself after the amino acid has been activated ( i . e . formation of aminoacyl-AMP using ATP as a substrate ) but prior to its transfer to the tRNA ( pre-transfer editing ) . Alternatively , proofreading can happen at a distinct editing site after the activated non-cognate amino acid has been esterified with the tRNA ( post-transfer editing ) . These proofreading processes are so crucial that even mild defects can lead to adverse cellular outcomes like cell growth retardation , neurodegeneration , cardiomyopathy and cell death ( Bacher et al . , 2005; Bullwinkle et al . , 2014; Karkhanis et al . , 2007; Korencic et al . , 2004; Lee et al . , 2006; Liu et al . , 2014; Lu et al . , 2014; Moghal et al . , 2016; Nangle et al . , 2002; Roy et al . , 2004 ) , although a compromise in editing can also be beneficial as it helps the organism to tide over stress conditions ( Moghal et al . , 2014 ) . Being comparatively small and similar to the cognate alanine , glycine and serine are misactivated by alanyl-tRNA synthetase ( AlaRS ) at significantly high frequencies of 1/240 and 1/500 , respectively , relative to alanine ( Tsui and Fersht , 1981 ) ( Figure 1 ) . However , these misactivation rates are much higher than the overall error rates of ~10−4–10−3 observed during protein biosynthesis ( Ogle and Ramakrishnan , 2005 ) . Once ( mis ) activated , these non-cognate amino acids are mischarged on tRNAAla by AlaRS . This creates a unique mechanistic challenge for the editing domain of AlaRS , which has to specifically remove two non-cognate amino acids—the larger serine and the smaller glycine—attached to tRNAAla without acting on the cognate alanine , which is intermediate in size between serine and glycine . This 3 . 5-billion-year-old double-discrimination problem is shown to be unavoidable for AlaRS in all forms of life ( Guo et al . , 2009 ) . It has also been shown that serine mischarging on tRNAAla is detrimental to the cell and even a mild deficiency in the proofreading activity of AlaRS leads to cell death and severe neuropathologies in mouse ( Lee et al . , 2006; Liu et al . , 2014 ) . The problem is so severe that several standalone trans-editing modules ( collectively called AlaXs ) , which are homologous to AlaRS cis-editing domain , have come into being . However , these trans-editing factors are not ubiquitously present; their distribution is more in archaea than in eukaryotes and bacteria ( Guo and Schimmel , 2012 ) . Surprisingly , only archaeal AlaXs are known to clear both Ser-tRNAAla and Gly-tRNAAla ( Ahel et al . , 2003 ) ; eukaryotic AlaXs have been shown to act as cellular redundancies to edit only Ser-tRNAAla ( Guo et al . , 2009 ) , whereas biochemical activity of bacterial AlaXs is yet to be probed ( Figure 1 ) . These findings corroborated the notion that only serine mischarging by AlaRS poses the major threat to the cell ( Guo et al . , 2009; Lee et al . , 2006; Liu et al . , 2014 ) . 10 . 7554/eLife . 24001 . 003Figure 1 . Mischarging by AlaRS . AlaRS activates and charges alanine ( A ) to form cognate Ala-tRNAAla which is routed for protein synthesis . In this process , AlaRS also misactivates glycine ( G ) and serine ( S ) at frequencies of 1 per 240 alanine and 1 per 500 alanine , respectively ( Tsui and Fersht , 1981 ) . The two non-cognate amino acids are then charged on tRNAAla to produce Gly-tRNAAla and Ser-tRNAAla species , with glycine mischarging being nearly twice that of serine . Since AlaRS does not distinguish much between Gly-tRNAAla and Ser-tRNAAla while clearing the two , higher levels of Gly-tRNAAla might accumulate in the cell . However , there are additional free-standing trans-editing factors called AlaX ( found in all domains of life but not in all organisms ) , which are known to edit mainly Ser-tRNAAla . This leads to a fundamental question as to how the problem of Gly-tRNAAla editing is solved in the cellular context . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 003 In the current study , we show that there is significant glycine mischarging by AlaRS in the presence of EF-Tu which can be equally pernicious as serine mischarging . We demonstrate that DTD plays an active and crucial role in preventing the accumulation of mischarged Gly-tRNAAla species . A cell lacking DTD in AlaRS editing-defective background displays pronounced toxicity toward even low levels of glycine which is , nevertheless , alleviated by alanine supplementation . Our data also indicate that DTD has selectivity for the G3•U70 wobble base pair that is unique to tRNAAla , suggesting that in the primordial scenario , DTD could have been recruited primarily as a glycine-removing factor . Our study thus brings to the fore three important aspects of translational fidelity , which were underappreciated or unknown so far . Firstly , glycine , like serine , can be toxic and deleterious to the cell under conditions wherein the cell is deficient in disposing of the mischarged Gly-tRNAAla species . Secondly , how the design of the active site of DTD , notwithstanding its unwarranted activity on Gly-tRNAGly , is used to efficiently decouple glycine mischarged on tRNAAla despite the presence of EF-Tu , thereby fortifying translational fidelity . Thirdly , there is a positive selection of the element ( s ) of tRNAAla by DTD , indicating for the first time the role of tRNA elements in modulating DTD’s activity . To test whether Gly-tRNAAla is accumulated due to mischarging of glycine on tRNAAla by AlaRS , we performed aminoacylation assays in the presence of EF-Tu . In comparison to alanine charging , significant glycine mischarging was observed . Furthermore , the level of glycine mischarging was about twice that of serine mischarging ( Figure 2a ) . This clearly indicated that even with full AlaRS editing potential , there can be significant accumulation of Gly-tRNAAla species in the cell . Moreover , this is in accordance with the twofold higher misactivation rate of glycine by AlaRS when compared to that of serine ( Tsui and Fersht , 1981 ) . We then checked the accumulation of Gly-tRNAAla when AlaRS editing was compromised , and it was found to be significantly high , almost equal to the level of Ala-tRNAAla formation ( Figure 2—figure supplement 1 ) . To accomplish a compromise in the proofreading activity of AlaRS , a known editing site mutation ( viz . , C666A ) in AlaRS from Escherichia coli was used ( Beebe et al . , 2003 ) . The above data lead to a couple of fundamental questions: ( a ) why does a defect in the same editing domain that edits both serine and glycine from tRNAAla cause toxicity only due to serine ? and ( b ) how does the cell tackle the problem of glycine mischarging ? Furthermore , based on structural considerations and evolutionary substitution patterns of alanine , where it is replaced more by serine than by glycine ( Betts et al . , 2003 ) , it is to be expected that substitution of glycine for alanine is more detrimental than substitution of serine . 10 . 7554/eLife . 24001 . 004Figure 2 . Misacylation of tRNAAla with glycine by AlaRS and its prevention/rectification by DTD . ( a ) Aminoacylation of tRNAAla by EcAlaRS in the presence of activated EF-Tu: L-alanine ( green square ) , L-alanine and 10 pM EcDTD ( green triangle ) , glycine ( pink square ) , glycine and 10 pM EcDTD ( pink triangle ) , L-serine ( purple square ) , L-serine and 10 pM EcDTD ( purple triangle ) . No enzyme control ( blue diamonds ) reaction had all the components of the reaction ( with L-alanine ) except for EcAlaRS . ( b ) Deacylation of Gly-tRNAAla in the presence of unactivated EF-Tu ( green diamond ) , activated EF-Tu ( blue diamond ) , 5 pM EcDTD and unactivated EF-Tu ( purple square ) , 5 pM EcDTD and activated EF-Tu ( orange square ) . Error bars indicate one standard deviation from the mean of triplicate readings . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 00410 . 7554/eLife . 24001 . 005Figure 2—source data 1 . Misacylation of tRNAAla and deacylation of Gly-tRNAAla in the presence of EF-Tu . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 00510 . 7554/eLife . 24001 . 006Figure 2—figure supplement 1 . Accumulation of Ala/Gly/Ser-tRNAAla during aminoacylation by EcAlaRS C666A in the presence of EF-Tu . Aminoacylation of tRNAAla by EcAlaRS C666A in the presence of activated EF-Tu: L-alanine ( green square ) , glycine ( pink square ) , L-serine ( purple square ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 006 The leads for the solution to this puzzle came when , surprisingly , we found that the activity of DTD on Gly-tRNAAla was ~1000-fold more than that on Gly-tRNAGly ( as discussed later ) . Moreover , although the ratio of activated EF-Tu to DTD in our assays ( viz . , ~200 nM to 5 or 10 pM ) was much higher than the cellular ratio ( viz . , ~200:1 ) ( Li et al . , 2014 ) , DTD could easily edit Gly-tRNAAla in the presence of EF-Tu ( Figure 2b ) . This was unlike the case of Gly-tRNAGly , wherein EF-Tu showed significant protection of the cognate achiral substrate from DTD ( Routh et al . , 2016 ) . The deacylation of Gly-tRNAAla by DTD was so striking that ~20 , 000 times more AlaRS ( which is the only universally occurring editing factor for Gly-tRNAAla ) as compared to DTD was required for similar kind of deacylation under identical conditions ( Figures 2b and 3b ) . In addition , unlike DTD , AlaRS showed a significant decrease in deacylation activity on Gly-tRNAAla when tested in the presence of EF-Tu ( Figure 3a , b ) . Moreover , our assays demonstrated that DTD is not only efficient in eliminating Gly-tRNAAla despite the presence of EF-Tu ( Figure 2b ) but can also very effectively prevent the accumulation of Gly-tRNAAla during aminoacylation by AlaRS in the presence of EF-Tu ( Figure 2a ) . 10 . 7554/eLife . 24001 . 007Figure 3 . DTD has higher activity than AlaRS for the editing of Gly-tRNAAla . ( a ) Deacylation of Gly-tRNAAla in the presence of unactivated EF-Tu: buffer ( blue diamond ) , 10 nM EcAlaRS ( red circle ) , 50 nM EcAlaRS ( green circle ) , 100 nM EcAlaRS ( purple circle ) , 500 nM EcAlaRS ( pink circle ) . ( b ) Deacylation of Gly-tRNAAla in the presence of activated EF-Tu: buffer ( blue diamond ) , 10 nM EcAlaRS ( red square ) , 50 nM EcAlaRS ( green square ) , 100 nM EcAlaRS ( purple square ) , 500 nM EcAlaRS ( pink square ) . ( c ) Deacylation of Gly-tRNAAla by EcDTD and increasing concentration of EcAlaRS: buffer ( blue diamond ) , 10 pM EcDTD ( orange diamond ) , 10 pM EcDTD and 10 nM EcAlaRS ( red triangle ) , 10 pM EcDTD and 50 nM EcAlaRS ( green triangle ) , 10 pM EcDTD and 100 nM EcAlaRS ( purple triangle ) , 10 pM EcDTD and 500 nM EcAlaRS ( pink triangle ) . Error bars indicate one standard deviation from the mean of triplicate readings . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 00710 . 7554/eLife . 24001 . 008Figure 3—source data 1 . Deacylation of Gly-tRNAAla in the presence of unactivated and activated EF-Tu . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 008 Considering the high activity of DTD on Gly-tRNAAla , we probed the relative efficiencies of DTD and AlaRS in editing Gly-tRNAAla . To this end , we performed competition assays involving AlaRS , DTD and EF-Tu . In both aminoacylation and deacylation conditions , we found that DTD deacylated Gly-tRNAAla even in the presence of EF-Tu and AlaRS at just 10 pM concentration of DTD ( Figures 2a and 3c ) . Considering the cellular ratio of DTD to AlaRS ( viz , ~1:5 ) ( Li et al . , 2014 ) and their relative activities on Gly-tRNAAla , it is evident that when DTD is present , it will eliminate Gly-tRNAAla more efficiently than AlaRS if the non-cognate achiral substrate is released in solution from the synthetase . In this context , it is important to note that compared to Class I synthetases , enzymes belonging to Class II , which include AlaRS , have been shown to have faster product release rates ( Zhang et al . , 2006 ) . Hence , Class II aaRSs require resampling of the released mischarged product to edit the cytosolic pool of mischarged tRNAs ( Ling et al . , 2009 ) . This makes even more sense as regards AlaRS , since our structural analysis of AlaRS in complex with tRNAAla ( PDB id: 3WQY ) suggests that it would be very difficult for the CCA-arm at the 3′ end of tRNAAla harboring the non-cognate amino acid to flip from the aminoacylation site to the editing site without undergoing major conformational changes ( Naganuma et al . , 2014 ) . Such a dynamics would naturally facilitate a faster release of the misacylated product in solution , implying that a significant fraction of Gly-tRNAAla and Ser-tRNAAla is released from AlaRS prior to their recapture for proofreading by the cis-editing domain . Moreover , our own data , in which we observed significant accumulation of Gly-tRNAAla in the presence of EF-Tu ( Figure 2a ) , corroborate the aforementioned aspects of Class II synthetases in general and AlaRS in particular . Overall , the above data suggest that the problem of glycine mischarging by AlaRS would have been so detrimental that a highly efficient factor like DTD was required to be employed for this function in addition to AlaRS editing domain . Earlier in vivo studies had shown that AlaRS editing defect causes glycine toxicity only at very high levels of glycine supplementation ( 80 mM ) as opposed to serine which causes toxicity at significantly lower levels ( 2 . 5 mM ) ( Beebe et al . , 2003 ) . It is worth noting that these studies were carried out in strains harboring DTD , hence explaining the need for supplementation with more glycine to show toxicity . To check if the absence of DTD makes E . coli susceptible to glycine , we generated an E . coli strain in which dtd ( the gene encoding DTD ) was knocked out in the background of editing-defective AlaRS . To create a strain that was completely devoid of AlaRS editing activity , the genomic copy of AlaRS gene ( alaS ) was knocked out and a triple-mutant AlaRS ( viz . , T567F/S587W/C666F ) was expressed from a plasmid . The editing site mutations were designed on the basis of a structural ( homology ) model of E . coli AlaRS cis-editing domain that was generated using the structure of Archaeoglobus fulgidus AlaRS ( PDB id: 2ZTG ) as a template . This model was then superimposed on Pyrococcus horikoshii AlaX complexed with serine ( PDB id: 1WNU ) ( the best substrate-mimicking complex for AlaRS and AlaX available so far ) ( Sokabe et al . , 2005 ) . Three residues in the proposed editing site ( Beebe et al . , 2003; Sokabe et al . , 2005 ) were supplanted by bulkier residues to occlude the pocket and prevent substrate binding ( Figure 4a , b ) . The triple-mutant was found to be inactive on both Ser-tRNAAla and Gly-tRNAAla even when the protein concentration was increased to 1500-fold that of wild-type AlaRS ( Figure 4c , d ) . It is worth mentioning here that the previously known editing-defective mutants of AlaRS ( C666A and C666A/Q584H ) ( Beebe et al . , 2003 ) , when checked for deacylation activity on both Ser-tRNAAla and Gly-tRNAAla , were found to show significant activity at just 10-fold higher concentration of the enzyme ( Figure 4c , d ) . Thus , to completely abrogate AlaRS editing activity and to see the effect of editing from only DTD , we chose to use AlaRS triple-mutant for our cell-based toxicity studies . 10 . 7554/eLife . 24001 . 009Figure 4 . E . coli AlaRS editing site mutants . Homology model of E . coli AlaRS depicting serine ( green sticks/spheres ) in the editing site . E . coli AlaRS cis-editing domain was modeled using A . fulgidus AlaRS ( PDB id: 2ZTG ) as a template , whereas the position and orientation of serine in the model corresponds to that observed in serine-bound P . horikoshii AlaX structure ( PDB id: 1WNU ) . ( a ) In the wild-type enzyme , residues selected for mutagenesis are represented as megenta sticks/spheres , showing an open pocket for substrate binding . ( b ) In AlaRS T567F/S587W/C666F , the mutated bulkier residues are depicted as blue sticks/spheres , showing occlusion of the pocket to prevent substrate binding . ( c ) Deacylation of Ser-tRNAAla by buffer ( blue diamond ) , 50 nM EcAlaRS ( pink circle ) , 500 nM EcAlaRS C666A ( green circle ) , 500 nM EcAlaRS C666A/Q584H ( purple circle ) , 75 µM EcAlaRS T567F/S587W/C666F ( orange circle ) . ( d ) Deacylation of Gly-tRNAAla by buffer ( blue diamond ) , 50 nM EcAlaRS ( pink square ) , 500 nM EcAlaRS C666A ( green square ) , 500 nM EcAlaRS C666A/Q584H ( purple square ) , 75 µM EcAlaRS T567F/S587W/C666F ( orange square ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 00910 . 7554/eLife . 24001 . 010Figure 4—source data 1 . Deacylation of Ser-tRNAAla and Gly-tRNAAla by E . coli AlaRS editing site mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 010 Cellular toxicity studies using spot dilution assays and growth curve analysis with the DTD knockout strain in the background of AlaRS editing defect showed some toxicity even without any amino acid supplementation , and the toxicity increased with glycine supplementation as low as 3 mM . At 10 mM of glycine supplementation , the cells showed severe growth defect ( Figure 5a ) . To check whether this was specifically due to mischarging caused by AlaRS , toxicity experiments were carried out in the presence of alanine , since the latter is expected to compete for the AlaRS aminoacylation site during charging of tRNAAla . It was observed that alanine supplementation completely recovered the toxicity caused by glycine and rescued the growth completely ( Figure 5b , c , d ) . Moreover , to rule out any non-specific effects due to amino acid supplementation , histidine was used as a negative control , and it was found that it failed to rescue the cells from glycine toxicity ( Figure 5c , d ) . Furthermore , DTD was found to be totally inactive on Ser-tRNAAla in our biochemical assays which confirmed that the toxicity observed in the DTD-lacking cells in the background of AlaRS editing defect was not due to serine mischarging on tRNAAla by AlaRS ( Figure 5—figure supplement 1 ) . The above observation is expected since it has been shown earlier that DTD’s chiral proofreading site rejects even L-alanine , the smallest L-chiral substrate ( Routh et al . , 2016 ) . Taken together , these experiments established that DTD acts as a key cellular factor that edits glycine mischarged on tRNAAla by AlaRS . However , no toxicity of glycine supplementation was observed in E . coli MG1655 ∆dtd strain in the background of wild-type AlaRS ( Figure 5—figure supplement 2 ) . This indicates that under normal growth conditions , AlaRS editing is sufficient for the cell to survive . The errors produced because of the absence of DTD are possibly tolerated by E . coli under laboratory conditions , as has been noted in several other cases of editing function of aaRSs ( Reynolds et al . , 2010a , 2010b ) . The real implications of editing defects are only recently being appreciated in some specific growth conditions like oxidative stress , oxygen deprivation , starvation/nutrient limiting conditions etc . ( Bullwinkle et al . , 2014; Cvetesic et al . , 2014; Kermgard et al . , 2017 ) . 10 . 7554/eLife . 24001 . 011Figure 5 . DTD knockout causes pronounced glycine toxicity in E . coli . Spot dilution assay of E . coli MG1655 ∆alaS/para: : alaS-T567F , S587W , C666F compared with that of E . coli MG1655 ∆dtd , ∆alaS/para:: alaS-T567F , S587W , C666F ( a ) in the presence of no amino acid , 3 mM glycine , or 10 mM glycine , and ( b ) in the presence of 1 mM L-alanine , or 10 mM glycine and 1 mM L-alanine . ( c ) Growth curve of E . coli MG1655 ∆alaS/para: : alaS-T567F , S587W , C666F supplemented with no amino acid ( blue diamond ) , 3 mM glycine ( orange triangle ) , 10 mM glycine ( red triangle ) , 30 mM glycine ( black triangle ) , 30 mM glycine and 10 mM L-alanine ( green triangle ) , 30 mM glycine and 10 mM L-histidine ( cyan triangle ) ( d ) Growth curve of E . coli MG1655 ∆dtd , ∆alaS/para: : alaS-T567F , S587W , C666F supplemented with no amino acid ( blue diamond ) , 3 mM glycine ( orange circle ) , 10 mM glycine ( red circle ) , 30 mM glycine ( black circle ) , 30 mM glycine and 10 mM L-alanine ( green circle ) , 30 mM glycine and 10 mM L-histidine ( cyan circle ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 01110 . 7554/eLife . 24001 . 012Figure 5—source data 1 . Growth curves of E . coli MG1655 with and without dtd knockout in AlaRS editing-defective background . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 01210 . 7554/eLife . 24001 . 013Figure 5—figure supplement 1 . DTD is inactive on Ser-tRNAAla . Deacylation of Ser-tRNAAla by buffer ( blue diamond ) , 50 nM EcDTD ( pink circle ) , 500 nM EcDTD ( purple circle ) , 5 µM EcDTD ( orange circle ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 01310 . 7554/eLife . 24001 . 014Figure 5—figure supplement 2 . Spot dilution assay of E . coli MG1655 compared with E . coli MG1655 ∆dtd with increasing concentration of glycine . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 014 Significantly high ( ~1000-fold higher ) activity of DTD on Gly-tRNAAla when compared to that on Gly-tRNAGly ( Figure 6d , Figure 6—figure supplement 1 ) indicated that tRNAAla must have some positive determinants for DTD . Since G3•U70 is unique to and one of the major identity elements of tRNAAla across all life forms ( Hou and Schimmel , 1988; 1989; McClain and Foss , 1988; Ripmaster et al . , 1995; Shiba et al . , 1995 ) , we envisaged that DTD could be positively selecting tRNAAla using the wobble base pair . To test this hypothesis , we transplanted G3•U70 at the same position in tRNAGly . We found that DTD had more than 10-fold increased activity on Gly-tRNAGly harboring G3•U70 as compared to the wild-type achiral cognate substrate ( Figure 6a , c ) . To strengthen this hypothesis , we substituted the G3•C70 in the wild-type Gly-tRNAGly with the other Watson-Crick base pair , that is A3•U70 . This substitution did not cause any increase in the activity of DTD ( Figure 6b ) . This clearly suggests that the G3•U70 wobble base pair , which is a universally conserved feature of tRNAAla , acts as a positive determinant for DTD . It also suggests the existence of other features in tRNAAla distinct from tRNAGly that accounts for the higher activity of DTD on tRNAAla when compared to tRNAGly , and this aspect requires further exploration . Since DTD is expected to act on all D-aminoacyl-tRNAs , it was assumed that there is no specificity code for its action on tRNAs . The identity determinant–switching experiment resulting in higher activity suggests for the first time an underlying tRNA-based code for DTD action . 10 . 7554/eLife . 24001 . 015Figure 6 . DTD positively selects the tRNA acceptor stem element G3•U70 . ( a ) Deacylation of Gly-tRNAGly by buffer ( blue diamond ) , 5 nM EcDTD ( purple square ) , 50 nM EcDTD ( pink circle ) . ( b ) Deacylation of Gly-tRNAGly A3•U70 by buffer ( blue diamond ) , 5 nM EcDTD ( purple square ) , 50 nM EcDTD ( pink circle ) . ( c ) Deacylation of Gly-tRNAGly G3•U70 by buffer ( blue diamond ) , 500 pM EcDTD ( green triangle ) , 5 nM EcDTD ( purple square ) , 50 nM EcDTD ( pink circle ) . ( d ) Deacylation of Gly-tRNAAla by buffer ( blue diamond ) , 5 pM EcDTD ( orange diamond ) . Error bars indicate one standard deviation from the mean of triplicate readings . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 01510 . 7554/eLife . 24001 . 016Figure 6—source data 1 . Deacylation of Gly-tRNAGly mutants and Gly-tRNAAla by DTD . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 01610 . 7554/eLife . 24001 . 017Figure 6—figure supplement 1 . DTD’s activity on the cognate achiral substrate . Deacylation of Gly-tRNAGly by buffer ( blue diamond ) , 5 nM EcDTD ( brown diamond ) . Error bars indicate one standard deviation from the mean of triplicate readings . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 017 DTD’s ubiquitous presence in bacteria and eukaryotes prompted us to test whether its role in clearing mischarged Gly-tRNAAla , in addition to its role in proofreading D-aminoacyl-tRNAs , is conserved in these two domains of life . It is important to investigate this aspect because we found significant differences in residues—which are believed to interact with the acceptor stem of tRNA—around the chiral proofreading site of DTD . To this end , we first superimposed crystal structures of DTD from various organisms and manually docked tRNAAla ( tRNAAla was taken from PDB id: 3WQY ) on the superposed structures . Since DTD is not expected to establish contacts beyond the acceptor stem of tRNA , we took into consideration only those residues of DTD that were within 6 Å from the 3′-terminal CCA-arm of tRNA ( Figure 7—figure supplement 1b ) , and looked for their conservation/variation using structure-based multiple sequence alignment ( Figure 7—figure supplement 1a ) . Notably , only 7 out of 20 residues in the selected region are invariant across bacterial and eukaryotic DTDs ( Figure 7—figure supplement 1 ) . Given the significant differences observed in a region around the active site of DTD which is likely to interact with tRNA , it is not unreasonable to anticipate that they may affect DTD’s activity on Gly-tRNAAla . To ascertain this , we tested DTDs from different eukaryotes—Leishmania major ( LmDTD ) , Drosophila melanogaster ( DmDTD ) and Danio rerio ( DrDTD ) —spanning the entire spectrum of unicellular , invertebrate and vertebrate species , in addition to DTDs from Plasmodium falciparum ( PfDTD ) and E . coli ( EcDTD ) . These DTDs were tested on glycine mischarged on both E . coli tRNAAla and D . melanogaster tRNAAla ( Figure 7 ) . All these DTDs were found to act effectively on both bacterial and eukaryotic tRNAs at 10 pM of DTD concentration . This indicates that in spite of the cross-species differences in DTD and tRNAAla , DTD’s activity on Gly-tRNAAla is most likely conserved throughout bacteria and eukaryotes . 10 . 7554/eLife . 24001 . 018Figure 7 . DTD edits Gly-tRNAAla across bacteria and eukaryotes . ( a ) Deacylation of E . coli Gly-tRNAAla by buffer ( blue diamond ) , 10 pM EcDTD ( red square ) , 10 pM PfDTD ( green triangle ) , 10 pM LmDTD ( purple cross ) , 10 pM DmDTD ( cyan star ) , 10 pM DrDTD ( orange circle ) . ( b ) Deacylation of D . melanogaster Gly-tRNAAla by buffer ( blue diamond ) , 10 pM EcDTD ( red square ) , 10 pM PfDTD ( green triangle ) , 10 pM LmDTD ( purple cross ) , 10 pM DmDTD ( cyan star ) , 10 pM DrDTD ( orange circle ) . Error bars indicate one standard deviation from the mean of triplicate readings . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 01810 . 7554/eLife . 24001 . 019Figure 7—source data 1 . Deacylation of E . coli Gly-tRNAAla and D . melanogaster Gly-tRNAAla by bacterial and eukaryotic DTDs . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 01910 . 7554/eLife . 24001 . 020Figure 7—figure supplement 1 . Variations in the tRNA-binding site of DTD . ( a ) Structure-based multiple sequence alignment of DTD from different organisms . The residues which are within a distance of 6 Å from the 3′-terminal CCA-arm of tRNA are marked by green stars , and among these , the residues showing variations are enclosed in green box . Black arrowheads indicate the organisms from which DTDs have been tested biochemically in the current study . ( b ) Surface model of E . coli DTD ( PDB id: 1JKE ) with 3′-terminal CCA-arm of tRNA placed in the active site on the basis of D-Tyr3AA from the structure of P . falciparum DTD ( PDB id: 4NBI ) . The residues colored in green show the positions showing variations among different organisms . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 020 This study provides an unprecedented solution to a fundamental and long-standing puzzle by elucidating a hitherto unknown and physiologically important function of DTD , which was till now implicated only in enforcing homochirality during translation of the genetic code . The discovery of DTD as a key cellular factor for the elimination of Gly-tRNAAla provides an elegant explanation as to why glycine mischarging by AlaRS was not encountered or considered as a cellular hazard in all the previous studies . So far , only serine mischarging on tRNAAla by AlaRS was believed to be the major threat to the cell , since cells harboring editing-defective AlaRS would show toxicity only to low levels of serine but not glycine ( Beebe et al . , 2003; Lee et al . , 2006 ) . However , this observation seemed puzzling for two reasons . Firstly , glycine misactivation by AlaRS is known to occur at about twice the rate of serine misactivation ( Tsui and Fersht , 1981 ) . Secondly and more importantly , it is unlikely that a defect in the proofreading domain that edits both serine and glycine would cause toxicity only due to serine but not glycine . Moreover , for a protein’s structure and function , the substitution of glycine for alanine is expectedly more subversive than the substitution of serine for alanine ( Betts et al . , 2003 ) . Our study thus brings forth the criticality of glycine mischarging problem , which was largely overlooked and underappreciated prior to this work . It also reveals that nature was forced to devise and retain throughout evolution a key checkpoint in the form of DTD that is more efficient than even AlaRS’s proofreading function to tackle this problem ( Figure 8 ) . However , for E . coli under laboratory conditions , knockout of DTD in AlaRS editing-proficient background did not cause toxicity on glycine supplementation . Very likely , errors caused due to defect in proofreading by DTD knockout are tolerated by E . coli , as noted in several cases of other proofreading deficiencies in E . coli ( Reynolds et al . , 2010a , 2010b ) . Moreover , defects in error correction are known to manifest in toxic effects only in special growth conditions like oxidative stress , oxygen deprivation , starvation etc . ( Bullwinkle et al . , 2014; Cvetesic et al . , 2014; Kermgard et al . , 2017 ) . 10 . 7554/eLife . 24001 . 021Figure 8 . DTD doubles as a key factor to uncouple glycine mischarged on tRNAAla . In the cell , aminoacylation by aaRSs leads to the formation of different aa-tRNAs , of which L-aa-tRNAs ( left extreme ) are not acted upon by DTD , while D-aa-tRNAs ( right extreme ) are effectively decoupled in the presence or absence of EF-Tu , thereby enforcing homochirality . Glycylated tRNAs are acted upon by DTD ( centre ) but EF-Tu offers protection to the cognate Gly-tRNAGly to prevent its misediting , while the mischarged/non-cognate Gly-tRNAAla is efficiently cleared even in the presence of EF-Tu . Thick connecting arrows indicate the cellular scenario , wherein both DTD and EF-Tu are present . DOI: http://dx . doi . org/10 . 7554/eLife . 24001 . 021 The study also highlights the necessity of keeping DTD’s active site design intact during the course of evolution , probably because removal of the mischarged Gly-tRNAAla species from the cellular pool took precedence over DTD’s unwarranted activity on Gly-tRNAGly . Nevertheless , the glycine ‘misediting paradox’ was effectively resolved by safeguarding the cognate achiral substrate using EF-Tu as well as keeping the cellular levels of DTD low and tightly regulated ( Routh et al . , 2016 ) . Thus , what seemed to be an apparent flaw in the architecture of DTD’s active site proved to be a necessity . Moreover , the dual activity of DTD on both achiral and D-chiral substrates depicts it as a plausible ‘connecting link’ or ‘bridging factor’ between D-chirality–based and the canonical L-chirality–based proofreading during protein biosynthesis ( Figure 8 ) . This view gains support from the fact that a DTD-like fold appended to archaeal threonyl-tRNA synthetase ( ThrRS ) as the N-terminal editing domain ( NTD ) is specific for editing L-serine mischarged on tRNAThr ( Ahmad et al . , 2015; Dwivedi et al . , 2005; Hussain et al . , 2006 , 2010 ) . It is also worth noting here that the DTD-like fold present in two functional contexts—as NTD in archaea , and as DTD in bacteria and eukaryotes—operates majorly through main chain-mediated contacts for substrate recognition and performs catalysis through RNA , suggesting its primordial origins ( Ahmad et al . , 2013 , 2015; Routh et al . , 2016 ) . Glycine mischarging by AlaRS is inevitable and is a classic case of error made by the aminoacylation site of aaRS , whereas serine mischarging is an offshoot of amino group selection for alanine ( Guo et al . , 2009 ) . DTD’s significantly higher activity on Gly-tRNAAla as compared to AlaRS suggests that wherever and whenever present , DTD plays the major role in clearing the non-cognate achiral substrate from the cellular pool . This probably helps the cell to overcome the double-discrimination problem that is encountered by AlaRS in all extant organisms . In the primordial scenario , DTD could have been primarily employed as a glycine-removing factor . This activity of DTD could have aided the formation of relatively rigid and stable peptide/protein scaffolds by precluding glycine misincorporation , since the latter would have been detrimental to their stability . Another interesting facet of DTD that has emerged from this study is its ability to specifically recognize G3•U70 in the acceptor arm of tRNAAla . This wobble base pair is a unique and major identity determinant of tRNAAla which marks it for both aminoacylation and deacylation by AlaRS from bacteria to humans ( Beebe et al . , 2008; Hou and Schimmel , 1988; McClain and Foss , 1988 ) . The specificity of AlaRS for G3•U70 is so robust that incorporation of this base pair into other tRNAs or minihelices converts non-alanine-accepting tRNAs to be recognized and charged by AlaRS ( Musier-Forsyth and Schimmel , 1999 ) . This primordial mode of recognition was proposed to be an acceptor stem-based genetic code that could have been operational since the pre-tRNA era . The G3•U70-based selection of tRNAAla by DTD clearly indicates towards the role of DTD in editing Gly-tRNAAla even before the recruitment of AlaRS editing domains , which have primarily evolved to remove serine mischarged on tRNAAla ( Novoa et al . , 2015 ) . Furthermore , modulation of DTD’s activity depending on tRNA elements is counter-intuitive as DTD is expected to act on multiple tRNAs with comparable efficiencies . Hence , the present work has unveiled a completely new aspect of DTD’s aminoacyl-tRNA recognition code in which the role of the amino acid as well as the tRNA component needs to be looked at separately . Interestingly , a recognition code exists for EF-Tu , wherein three successive base pairs in the T-stem of tRNA thermodynamically compensate for the differential binding affinity of EF-Tu toward different amino acids ( LaRiviere et al . , 2001; Roy et al . , 2007; Sanderson and Uhlenbeck , 2007b; Schrader et al . , 2009 ) . Thus , it becomes important to understand how DTD treats multiple aminoacyl-tRNAs using its own recognition code . It is interesting to note that in certain cellular contexts , glycine as well as some D-amino acids can be present in relatively high concentrations , wherein these amino acids play important physiological roles . For example , in neuronal tissues , D-serine and D-aspartate along with glycine are abundant and act as neurotransmitters/neuromodulators ( Hashimoto and Oka , 1997; Snyder and Kim , 2000 ) . In such instances , especially in neuronal tissues , DTD’s function and its corresponding up-regulation ( Zheng et al . , 2009 ) suggest an all-pervasive requirement of this protein from a primordial domain involved in perpetuation of homochirality to current-day proofreader in physiological context . It has been established that even a mild compromise in AlaRS editing for Ser-tRNAAla causes severe pathological conditions , such as neurodegeneration and cardiomyopathy in mouse ( Lee et al . , 2006; Liu et al . , 2014 ) . In this regard , the role and regulation of DTD in various cellular contexts , and more importantly in higher eukaryotes , will be an important aspect which needs to be probed to gain newer insights into DTD’s physiological significance . DTD genes from the genome of E . coli and cDNAs of P . falciparum , L . major , D . melanogaster ( fruit fly ) and D . rerio ( zebrafish ) were cloned , and the proteins were expressed and purified as described previously ( Ahmad et al . , 2013; Routh et al . , 2016 ) . The gene ( alaS ) encoding E . coli AlaRS ( EcAlaRS ) cloned in pET-26b vector was a gift from Prof . William H . McClain ( University of Wisconsin-Madison , USA ) . An N-terminal 6X His-tagged ( N-His ) construct was made for EcAlaRS in pET-28b using restriction-free cloning method ( van den Ent and Löwe , 2006 ) . The plasmid containing EcAlaRS gene was transformed in E . coli BL21 ( DE3 ) cells for protein overexpression . The N-His-tagged EcAlaRS protein was purified by a two-step protocol involving Ni-NTA affinity chromatography and size exclusion chromatography ( SEC ) . The cells were lysed in lysis buffer containing 50 mM Tris–HCl pH 8 . 0 , 150 mM NaCl , 5 mM 2-Mercaptoethanol ( β-ME ) , and 10 mM imidazole . The same buffer was used to pre-equilibrate Ni-NTA column on which the cell lysate was loaded . After loading , the column was first washed with lysis buffer followed by wash buffer containing 50 mM Tris–HCl pH 8 . 0 , 150 mM NaCl , 5 mM β-ME , and 30 mM imidazole . Protein was eluted with elution buffer containing 50 mM Tris–HCl pH 8 . 0 , 150 mM NaCl , 5 mM β-ME , and 250 mM imidazole . The fractions containing protein of interest were pooled , concentrated and subjected to SEC purification using Superdex-200 in a buffer containing 100 mM Tris–HCl pH 8 . 0 , 300 mM NaCl , and 10 mM β-ME . Finally , the fractions containing purified protein were pooled , concentrated and mixed with equal volume of 100% glycerol before storing as aliquots at −30°C for further use . The gene ( tufA ) encoding EF-Tu was PCR-amplified from the genomic DNA of Thermus thermophilus , and cloned in pET-28b using restriction-free cloning method ( van den Ent and Löwe , 2006 ) . The N-His-tagged EF-Tu protein from T . thermophilus was was then overexpressed in E . coli BL21 ( DE3 ) cells and purified by a two-step method involving Ni-NTA affinity chromatography and SEC . The overall purification protocol remained very similar to the one described above except for changes in the buffer composition used in both steps . For Ni-NTA affinity chromatography , all the buffers contained 50 mM 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid ( HEPES ) titrated with potassium hydroxide ( KOH ) , that is , HEPES-KOH , pH 7 . 5 , 500 mM NaCl , 10 mM magnesium chloride ( MgCl2 ) , 5% glycerol , 5 mM β-ME , and 100 µM guanosine-5′-diphosphate ( GDP ) . Additionally , lysis , wash and elution buffers contained 10 mM , 30 mM and 250 mM imidazole , respectively . Following affinity chromatography , SEC was carried out using Sephadex G-200 and a buffer containing 50 mM HEPES-KOH pH 7 . 5 , 500 mM ammonium chloride ( NH4Cl ) , 20 mM MgCl2 , 10% glycerol , 5 mM dithiothreitol ( DTT ) , and 100 µM GDP . Finally , the purified protein was processed and stored at −30°C as described above . All protein purification steps from cell lysis onwards were carried out on ice or at 4°C . The mutants were generated using QuickChange XL Site-directed kit ( Stratagene , La Jolla , CA ) . E . coli tRNAAla was charged with alanine , serine and glycine by EcAlaRS C666A mutant as described by Pasman et al . ( 2011 ) . The same protocol was followed to charge D . melanogaster tRNAAla with glycine . E . coli tRNAGly was charged with glycine by T . thermophilus GlyRS as described by Routh et al . ( 2016 ) . Deacylation assays with AlaRS and DTD were carried out as described by Pasman et al . ( 2011 ) and Ahmad et al . ( 2013 ) . EF-Tu activation was carried out as described by Routh et al . ( 2016 ) . It is to be noted that considering only 10–15% efficiency of EF-Tu activation reaction ( Cvetesic et al . , 2013; Sanderson and Uhlenbeck , 2007a ) , the effective ( activated ) EF-Tu concentration in our assay conditions was in the range of 200–300 nM when the total EF-Tu concentration used was 2 µM . Aminoacylation competition assays were performed in a solution of 100 mM HEPES pH 7 . 2 , 2 . 5 mM DTT , 2 mM adenosine-5′-triphosphate ( ATP ) , and 200 mM amino acid ( alanine/serine/glycine ) with 2 μM ( total concentration ) EF-Tu , 100 nM tRNAAla , 100 nM EcAlaRS and 10 pM EcDTD . These assays were performed at 37°C and were tracked for 15 min . Deacylation competition assays were also carried out in a solution of 100 mM HEPES pH 7 . 2 , and 2 . 5 mM DTT with 2 μM ( total concentration ) EF-Tu , 200 nM aa-tRNAAla and varying concentrations of EcAlaRS and EcDTD . Unless otherwise stated , the tRNAs used in the assays were from E . coli . Every data point denotes the mean of three independent readings . Error bars represent one standard deviation from the mean . Viability assays were performed with deletion mutant strains of ΔalaS and Δdtd∆alaS in minimal medium ( Miller , 1992 ) . Relevant cultures were grown until OD600 reached 0 . 6 and were 10-fold serially diluted ( 10−2 , 10−3 , 10−4 , 10−5 and 10−6 ) . Of each serially diluted sample , 3 μl was spotted on minimal agar plates containing 0 . 002% L-arabinose , 0 . 2% maltose as carbon source , glycine ( 3 mM or 10 mM ) and/or L-alanine ( 1 mM ) . The plates were incubated at 37°C for 20–36 hr . For growth curves , primary cultures were grown in LB medium containing 0 . 0002% L-arabinose , Kanamycin and Chloramphenicol at 37°C until OD600 reached 1 . 0 . Subsequently , 2% inoculum was used to initiate 15 ml secondary culture in 1X minimal medium containing 0 . 2% maltose as carbon source and 0 . 0002% L-arabinose . The secondary culture was grown at 37°C to obtain a cell density ( OD600 ) of ~0 . 6 . These cultures were again grown in 1X minimal medium with/without amino acids ( glycine , L-alanine and L-histidine ) of the indicated concentrations . The growth was monitored at every 2-hr interval . All the experiments were done in triplicates .	Proteins are made up of many different building blocks called amino acids , which are linked together in chains . The exact order of amino acids in a protein chain is important for the protein to work properly . When a cell makes proteins , molecules known as transfer ribonucleic acids ( or tRNAs for short ) bind to specific amino acids to guide them to the growing protein chains in the correct order . Most amino acids – except one called glycine – have two forms that are mirror images of one another , known as left-handed ( L-amino acids ) and right-handed ( D-amino acids ) . However , only L-amino acids and glycine are used to make proteins . This is because of the presence of multiple quality control checkpoints in the cell that prevent D-amino acids from being involved . One such checkpoint is an enzyme called D-amino acid deacylase ( DTD ) , which removes D-amino acids that are attached to tRNAs . Other enzymes are responsible for linking a particular amino acid to its correct tRNA . Along with mistaking D-amino acids for L-amino acids , these enzymes can also make errors when they have to distinguish between amino acids that are similar in shape and size . For example , the enzyme that attaches L-alanine to its tRNA can also mistakenly attach larger L-serine or smaller glycine to it instead . Previous research has shown that attaching L-serine to this tRNA can lead to neurodegeneration in mice , whereas attaching glycine does not seem to cause any harm . It is not clear why this is the case . Pawar et al . investigated how incorrectly attaching glycine or L-serine to the tRNA that usually binds to L-alanine affects a bacterium called Escherichia coli . The experiments show that , if the mistake is not corrected , glycine can be just as harmful to the cells as L-serine . The reason that glycine appears to be less of a problem is that the DTD enzyme is able to remove glycine , but not L-serine , from the tRNA . Further experiments show that DTD can play a similar role in a variety of organisms from bacteria to mammals . The findings of Pawar et al . extend the role of DTD beyond preventing D-amino acids from being incorporated into proteins . The next step is to understand the role of this enzyme in humans and other multicellular organisms , especially in the context of nerve cells , where it is present at high levels .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2017	Role of D-aminoacyl-tRNA deacylase beyond chiral proofreading as a cellular defense against glycine mischarging by AlaRS
GPCRs are increasingly recognized to initiate signaling via heterotrimeric G proteins as they move through the endocytic network , but little is known about how relevant G protein effectors are localized . Here we report selective trafficking of adenylyl cyclase type 9 ( AC9 ) from the plasma membrane to endosomes while adenylyl cyclase type 1 ( AC1 ) remains in the plasma membrane , and stimulation of AC9 trafficking by ligand-induced activation of Gs-coupled GPCRs . AC9 transits a similar , dynamin-dependent early endocytic pathway as ligand-activated GPCRs . However , unlike GPCR traffic control which requires β-arrestin but not Gs , AC9 traffic control requires Gs but not β-arrestin . We also show that AC9 , but not AC1 , mediates cAMP production stimulated by endogenous receptor activation in endosomes . These results reveal dynamic and isoform-specific trafficking of adenylyl cyclase in the endocytic network , and a discrete role of a heterotrimeric G protein in regulating the subcellular distribution of a relevant effector . G protein-coupled receptors ( GPCRs ) comprise nature’s largest family of signaling receptors and an important class of therapeutic targets ( Lefkowitz , 2007; Rosenbaum et al . , 2009 ) . GPCRs are so-named because a major mechanism by which they mediate transmembrane signaling is through ligand-dependent activation of heterotrimeric G proteins that act as intracellular signal transducers ( Gilman , 1987; Hilger et al . , 2018; Spiegel , 1987; Stryer and Bourne , 1986; Sunahara et al . , 1996 ) . This conserved signaling cascade invariably requires one additional component , an ‘effector’ protein which is regulated by the G protein to convey the signal downstream ( Dessauer et al . , 1996; Gilman , 1987; Rosenbaum et al . , 2009 ) . Ligand-activated GPCR - G protein - effector cascades were thought for many years to be restricted to the plasma membrane , with endocytosis considered only in the context of signal termination and homeostatic down-regulation of receptors . This view has expanded over the past several years , driven by accumulating evidence that ligand-dependent GPCR and G protein activation processes can also occur from internal membrane compartments including endosomes ( Di Fiore and von Zastrow , 2014; Irannejad et al . , 2015; Jong et al . , 2018; Lobingier and von Zastrow , 2019; Lohse and Calebiro , 2013; Lohse and Hofmann , 2015; Vilardaga et al . , 2014 ) . The beta-2 adrenergic receptor ( β2AR ) provides a clear example , and is generally considered a model for the GPCR family more broadly ( Lefkowitz , 2007; Rosenbaum et al . , 2009 ) . β2ARs initiate signaling in response to binding of an agonist ligand by activating ( 'coupling' to ) the stimulatory heterotrimeric G protein , Gs , at the plasma membrane . β2ARs then internalize through agonist-dependent accumulation into clathrin-coated pits , efficiently recycle and transit continuously between the plasma membrane and endosomes in the prolonged presence of agonist ( von Zastrow and Kobilka , 1994; von Zastrow and Kobilka , 1992 ) . Agonist-induced clustering of β2ARs into coated pits , the process initiating this cycle , is promoted by receptor phosphorylation and binding to β-arrestins at the plasma membrane ( Ferguson et al . , 1996; Goodman et al . , 1996 ) . These events were shown previously to inhibit β2AR coupling to Gs ( Lohse et al . , 1990 ) , and β2AR inactivation was recognized to precede removal from the cell surface even before this mechanistic elucidation ( Harden et al . , 1980 ) . Accordingly , it was believed for many years that β2ARs are unable to engage G proteins once internalized . This view changed with the finding that β2ARs reacquire functional activity shortly after arriving in the limiting membrane of early endosomes , and then activate Gs again from this location ( Irannejad et al . , 2013 ) . A number of GPCRs have now been shown to activate Gs after endocytosis , initiating a discrete ‘second wave’ of downstream signaling which varies in magnitude and duration depending ( in part ) on the residence time of activated receptors in the endocytic network ( Irannejad et al . , 2015; Lohse and Calebiro , 2013; Thomsen et al . , 2016; Tian et al . , 2016; Varandas et al . , 2016; Vilardaga et al . , 2014 ) . Gs transduces downstream signaling by stimulating adenylyl cyclases ( ACs ) to produce cyclic AMP ( cAMP ) , an important diffusible mediator ( Lohse and Hofmann , 2015; Sutherland , 1971; Taylor et al . , 2013 ) , and studies of cAMP signaling provided much of the initial evidence supporting the potential of GPCRs to mediate ligand-dependent signaling via G proteins after endocytosis ( Calebiro et al . , 2009; Clark et al . , 1985; Ferrandon et al . , 2009; Kotowski et al . , 2011; Mullershausen et al . , 2009; Slessareva et al . , 2006 ) . Nine transmembrane AC isoforms are conserved in mammals , each stimulated by Gs but differing in regulation by other G proteins and signaling intermediates , and multiple AC isoforms are typically coexpressed in tissues ( Sadana and Dessauer , 2009; Sunahara et al . , 1996; Wang et al . , 2019 ) . Biochemical and structural aspects of regulated cAMP production by ACs have been extensively studied but much less is known about the cellular biology of ACs . According to the present understanding , GPCR-stimulated cAMP production requires all three ‘core’ components of the signaling cascade – the GPCR , Gs and AC – to be in the same membrane bilayer ( Gilman , 1989 ) . Whereas GPCRs and Gs are well known to undergo dynamic redistribution between the plasma membrane and endocytic membranes ( Allen et al . , 2005; Hynes et al . , 2004; Irannejad et al . , 2015; Marrari et al . , 2007; von Zastrow and Kobilka , 1992; Wedegaertner et al . , 1996 ) , the subcellular localization and trafficking properties of transmembrane ACs remain largely unknown . Nevertheless , adenylyl cyclase activity was noted on intracellular membranes many years ago ( Cheng and Farquhar , 1976 ) and more recent studies have implicated several AC isoforms in GPCR-regulated cAMP production from endomembrane compartments ( Calebiro et al . , 2009; Cancino et al . , 2014; Ferrandon et al . , 2009; Inda et al . , 2016; Kotowski et al . , 2011; Vilardaga et al . , 2014 ) . Key knowledge gaps at the present frontier are how ACs localize to relevant internal membranes , if the subcellular localization of ACs is selective between isoforms , and if this localization is regulated . Here we report initial inroads into this frontier by demonstrating the dynamic and isoform-specific endocytic trafficking of AC9 , and its ligand-dependent regulation through Gs . Human embryonic kidney ( HEK293 ) cells comprise a well established model system for investigating GPCR signaling via the cAMP cascade , particularly signaling initiated by β2ARs which are endogenously expressed in these cells ( Violin et al . , 2008 ) . The β2AR stimulates cAMP production primarily from the plasma membrane , with endosomal activation contributing a relatively small but functionally significant fraction ( Irannejad et al . , 2013; Tsvetanova and von Zastrow , 2014 ) . In a similar vein , AC3 and AC6 are the most highly expressed AC isoforms and are major producers of global cAMP in these cells , while AC1 and AC9 transcripts are expressed at only moderately reduced levels and comparable to one another ( Soto-Velasquez et al . , 2018 ) . This prompted us to ask whether AC1 and/or AC9 might be relevant to generating the fraction of cellular cAMP produced by β2AR activation in endosomes . As a first step to investigate this hypothesis , we examined the subcellular localization of AC1 and AC9 using a recombinant epitope tagging strategy , beginning with AC1 because this isoform was shown previously to tolerate an N-terminal Flag tag ( Chen et al . , 1997 ) . When expressed in HEK293 cells , Flag-tagged AC1 ( Flag-AC1 ) localized primarily to the plasma membrane , similar to a co-expressed HA-tagged β2AR construct ( HA-β2AR ) . Application of the β-adrenergic agonist isoproterenol promoted HA-β2ARs to redistribute within minutes to cytoplasmic punctae , as described previously , but Flag-AC1 remained at the plasma membrane ( Figure 1A ) . While clearly resolved in confocal sections , this difference in localization was sufficiently strong to be evident in lower-magnification widefield images surveying many cotransfected cells ( Figure 1—figure supplement 1D ) . We assessed reproducibility of this effect in two ways . First , we determined the number of internal punctae per cell in which HA-β2AR and Flag-AC1 colocalized . Second , we determined the fraction of cells visualized in each microscopic field that contained at least 10 such punctae . Both metrics verified selective internalization of the β2AR but not AC1 ( Figure 1C and D , left set of bars ) . We applied a similar tagging strategy to AC9 and verified that Flag tagging also does not disrupt the functional activity of AC9 ( Figure 1—figure supplement 1A , B ) . Flag-AC9 localized predominantly in the plasma membrane in the absence of agonist , similar to Flag-AC1 . However , after application of isoproterenol , Flag-AC9 redistributed to intracellular punctae , the majority of which also contain internalized HA-β2AR ( Figure 1B ) . Punctate redistribution of both Flag-AC9 and HA-β2AR was also evident in lower-magnification widefield images ( Figure 1—figure supplement 1E ) and was verified quantitatively ( Figure 1C and D , right set of bars ) . AC9 labeled in its C-terminal cytoplasmic domain with GFP ( AC9-GFP ) also redistributed , enabling live-cell confocal imaging which revealed mobile AC9-containing endosomes ( Video 1 ) . These results indicate that AC9 traffics dynamically to endosomes containing β2ARs , this trafficking is isoform-specific because AC1 remains in the plasma membrane , and it is regulated because AC9 accumulation in endosomes is increased by β2AR activation . Isoproterenol also stimulated selective internalization of Flag-AC9 in the absence of recombinant β2AR overexpression ( Figure 1E and F , Figure 1—figure supplement 1C ) and this effect was blocked by the β-adrenergic antagonist alprenolol ( Figure 1G and H , Figure 1—figure supplement 1C ) . These results indicate that endogenous β2AR activation is sufficient to stimulate AC9 trafficking and this is not an off-target drug effect . AC9 , and ACs in general , are naturally expressed in low abundance . We were unable to reliably detect endogenous AC9 in HEK293 cells using available antibodies , but endogenous AC9 was detectable in primary human airway smooth muscle cells that naturally coexpress β2ARs ( Billington et al . , 1999 ) . Endogenous AC9 immunoreactivity localized in these cells primarily to the plasma membrane under basal conditions , and its localization to internal punctae increased after isoproterenol application ( Figure 1I ) . These results suggest that the trafficking behavior revealed by study of recombinant , tagged AC9 is relevant to the native protein . Internalized β2ARs accumulate in endosomes marked by Early Endosome Antigen 1 ( EEA1 ) and agonist-dependent activation of Gs by the β2AR occurs on this compartment ( Irannejad et al . , 2013 ) . We verified isoform-specific localization of Flag-AC9 but not Flag-AC1 to EEA1-marked endosomes by confocal microscopy ( Figure 2A and B ) and then applied anti-EEA1 immunoisolation ( Cottrell et al . , 2009; Hammond et al . , 2010; Temkin et al . , 2011 ) to purify these endosomes and probe their composition . Coexpressed HA-β2AR and Flag-AC9 were enriched in the endosome fraction prepared from isoproterenol-treated cells , as detected by immunoblot analysis and verified quantitatively across multiple experiments . In contrast , HA-β2AR but not Flag-AC1 was enriched in parallel isolations from cells co-expressing HA-β2AR and Flag-AC1 , with similar levels of overall expression across both cell populations verified in cell lysates ( Figure 2C , D , and Figure 2—figure supplement 1B ) . Documenting separation efficiency and fraction purity , the endosome fraction recovered ~34% of total cellular EEA1 but <5% of Golgi , endoplasmic reticulum , or plasma membrane markers ( Figure 2—figure supplement 1A ) . As a distinct and additional biochemical approach to verify these trafficking properties , we used cell surface biotinylation to assess protein depletion from the plasma membrane ( Flesch et al . , 1995; Whistler and von Zastrow , 1998 ) . Isoproterenol produced a marked reduction of Flag-AC9 in the surface-biotinylated fraction but Flag-AC1 was unchanged , and surface HA-β2AR decreased after isoproterenol application irrespective of whether AC1 or AC9 was coexpressed ( Figure 2E , F , and Figure 2—figure supplement 1C ) . These observations independently demonstrate isoform-specific trafficking of AC9 from the plasma membrane to endosomes , regulated coordinately with endocytosis of the β2AR . A characteristic property of the conserved , clathrin-dependent pathway mediating β2AR endocytosis is that it also requires dynamin , an endocytic GTPase which can be acutely inhibited by the cell-permeant small molecule DYNGO-4a ( Irannejad et al . , 2013; Macia et al . , 2006; McCluskey et al . , 2013 ) . Whereas isoproterenol promoted both Flag-β2AR and AC9-GFP to accumulate in endosomes in the vehicle ( 0 . 1% DMSO ) control condition ( Figure 2G , H , and Figure 2—figure supplement 1D ) , endosomal accumulation of both proteins was blocked in the presence of DYNGO-4a ( Figure 2I and J , and Figure 2—figure supplement 1E ) . These results further support the hypothesis that regulated AC9 trafficking to endosomes utilizes a shared membrane pathway and mechanism relative to regulated endocytosis of the β2AR , and many other GPCRs . Despite these similarities , AC9 and β2AR were found to traffic independently . An early clue to this distinction was that brief exposure of cells outside of the tissue culture incubator ( see Materials and methods ) inhibits AC9 trafficking but β2AR trafficking is resistant to this environmental stress . While we still do not fully understand the basis for this difference , it focused our attention on investigating the mechanism of AC9 traffic control in detail . We first asked if the β2AR is unique in its ability to stimulate endosomal accumulation of AC9 or if AC9 trafficking can be stimulated by another Gs-coupled GPCR . To do so we focused on the vasopressin-2 receptor ( V2R ) , a Gs-coupled GPCR which also undergoes agonist-induced trafficking to early endosomes but is not endogenously expressed in HEK293 cells ( Birnbaumer , 2000 ) . The V2R is of particular interest because this GPCR , unlike the β2AR , has been shown to produce a sustained cAMP response via Gs that is associated with slow recycling and binding to β-arrestin in endosomes ( Innamorati et al . , 1998 , Feinstein et al . , 2013; Klein et al . , 2001; Oakley et al . , 1999; Thomsen et al . , 2016 ) . When examined in cells not transfected with recombinant V2Rs , Flag-AC9 remained in the plasma membrane irrespective of the presence of the V2R agonist arginine vasopressin ( AVP ) as expected ( Figure 3—figure supplement 1A ) . However , in cells coexpressing HA-V2R , AVP stimulated the redistribution of both Flag-AC9 and HA-V2R to a shared population of endosomes ( Figure 3A , D , E , Figure 3—figure supplement 1D ) . AVP-stimulated endocytosis of both proteins was confirmed by surface biotinylation ( Figure 3F , G , Figure 3—figure supplement 1B ) . These results indicate that the ability of GPCR activation to promote trafficking of AC9 to endosomes is not unique to the β2AR . Rather , it appears to be a shared property of GPCRs which activate Gs . We next asked if the ability to stimulate AC9 trafficking extends to GPCRs that couple to other heterotrimeric G proteins . We focused on the µ-opioid receptor ( MOP-R or MOR ) because this GPCR transits a similar early endocytic pathway as the β2AR ( Keith et al . , 1998 ) but activates Gi rather than Gs ( Kieffer and Evans , 2009 ) , and because opioid receptors have been shown explicitly to undergo ligand-dependent activation in endosomes ( Stoeber et al . , 2018 ) . Application of the µ-opioid agonist [D-Ala2 , N-MePhe4 , Gly-ol]-enkephalin ( DAMGO ) stimulated transfected HA-MOR to accumulate in endosomes , as shown previously , while Flag-AC9 remained in the plasma membrane ( Figure 3B , D , E , Figure 3—figure supplement 1E ) . We verified selective internalization of HA-MOR , but not Flag-AC9 , by surface biotinylation ( Figure 3F , G , Figure 3—figure supplement 1C ) . These results suggest that the ability to stimulate AC9 trafficking to endosomes is a property specific to GPCRs which couple to Gs relative to Gi . We then returned to the V2R , coexpressing a mutant version truncated in its C-terminal cytoplasmic tail . This mutant V2R ( HA-V2R-T ) retains the ability to activate Gs at the plasma membrane but internalizes less efficiently after agonist-induced activation , promotes β-arrestin recruitment to endosomes less strongly and recycles more rapidly ( Innamorati et al . , 1998; Innamorati et al . , 1997; Oakley et al . , 1999 ) . Despite visibly reduced internalization of HA-V2R-T after agonist application ( Figure 3C , D , E ) , Flag-AC9 internalization was still observed ( Figure 3C , D , E and Figure 3—figure supplement 1F ) and this was confirmed by surface biotinylation ( Figure 3F , G ) . These results suggest that Gs-coupled GPCRs share the ability to stimulate AC9 trafficking , irrespective of differences in receptor trafficking kinetics or binding to β-arrestin in endosomes . A possible basis for such shared control of AC9 trafficking is through cAMP elevation that occurs downstream of Gs activation . To test this , we applied the diterpene drug forskolin ( FSK ) to stimulate cytoplasmic cAMP production independently from the receptor or Gs . While AC9 is relatively insensitive to activation by FSK , other AC isoforms that are major contributors to cAMP production in HEK293 cells ( such as AC3 and AC6 ) are sensitive , making FSK an effective stimulus of overall cAMP elevation ( Baldwin et al . , 2019; Soto-Velasquez et al . , 2018 ) . FSK did not cause detectable internalization of either HA-β2AR or Flag-AC9 assessed by imaging ( Figure 4—figure supplement 1A , C , D , I ) or surface biotinylation ( Figure 4—figure supplement 1E , F ) . Further , as expected , Flag-AC1 remained in the plasma membrane irrespective of the presence of FSK ( Figure 4—figure supplement 1B–F , J ) . This was also true in the combined presence of 3-isobutyl-1-methylxanthine ( IBMX ) , a phosphodiesterase inhibitor which enhances FSK-induced cAMP elevation in the cytoplasm ( Figure 4—figure supplement 1H ) . As an independent approach , and to consider the possibility that cAMP exerts local rather than global control , we asked if endosomal accumulation of AC9 requires its own catalytic activity . To test this , we mutated a conserved aspartic acid residue that coordinates a catalytic magnesium in the active site , and which is essential for activity of AC6 ( Gao et al . , 2011; Tesmer , 1997 ) . We verified that mutating the equivalent residue in AC9 ( Flag-AC9-D442A ) blocked cAMP production ( Figure 1—figure supplement 1B ) , but found that regulated trafficking of Flag-AC9-D442A still occurred ( Figure 4—figure supplement 1G ) . Together , these results indicate that the ability of GPCR-Gs activation to regulate AC9 trafficking is not a consequence of global cytoplasmic cAMP elevation , nor does it require local cAMP production by AC9 . We next investigated whether AC9 internalization is regulated by Gs itself , and did so by introducing a point mutation into the alpha subunit ( HA-Gs-Q227L ) that renders Gs constitutively active by reducing its rate of intrinsic GTP hydrolysis ( Masters et al . , 1989 ) . Flag-AC9 localized to the plasma membrane in the absence of agonist when coexpressed with wild type HA-Gs , but coexpression with activated HA-Gs-Q227L resulted in both proteins localizing to internal punctae ( Figure 4A , C , D , Figure 4—figure supplement 2E ) . This effect was specific to AC9 because AC1 remained at the plasma membrane when coexpressed with either HA-Gs or HA-Gs-Q227L ( Figure 4B , C , D , Figure 4—figure supplement 2F ) . We verified by immunoisolation that both HA-Gs-Q227L and Flag-AC9 accumulate in EEA1-positive endosomes when coexpressed ( Figure 4—figure supplement 2A , B ) , whereas coexpression of HA-Gs-Q227L with Flag-AC1 failed to produce endosome enrichment of either protein ( Figure 4—figure supplement 2C , D ) . Independently supporting a discrete regulatory effect of Gs , application of cholera toxin ( CTX ) to activate the endogenous cellular complement of Gs resulted in receptor-independent accumulation of Flag-AC9 , but not Flag-AC1 , in endosomes ( Figure 4—figure supplement 3 ) . The immunoisolation analysis also indicated that constitutive activation of Gs produced a degree of endosomal enrichment of AC9 similar to that produced in response to endogenous β2AR activation with isoproterenol . Further , application of isoproterenol to cells which coexpress HA-Gs-Q227L did not detectably increase the degree of endosomal enrichment observed for either Gs or AC9 ( Figure 4E , F ) . Together , these results suggest that Gs activation mediated by endogenous β2ARs is fully sufficient to stimulate AC9 trafficking to endosomes , without requiring additional effects of upstream receptor activation or downstream cAMP signaling . Because Gs activation is sufficient to stimulate accumulation of AC9 in endosomes , we next asked if it is necessary to regulate this trafficking process . To do so , we utilized previously described Gs-knockout ( GsKO ) cells which lack Gs due to CRISPR-mediated editing of the alpha subunit ( GNAS ) gene ( Stallaert et al . , 2017 ) . Flag-β2AR and AC9-EGFP localized to the plasma membrane of GsKO as well as wild type HEK293 cells . However , AC9-GFP internalization was blocked in GsKO cells while Flag-β2AR still internalized . Moreover , AC9 trafficking was rescued by expression of recombinant HA-Gs ( Figure 5A-D , Figure 5—figure supplement 1A , B ) . These results indicate that Gs is necessary for regulated trafficking of AC9 but not β2AR . Stimulation of β2AR endocytosis by agonists is known to depend on β-arrestins ( Ferguson et al . , 1996; Goodman et al . , 1996 ) . Accordingly , we asked if this is also true for AC9 . To test this , we used gene-edited HEK293 cells lacking both β-arrestin isoforms ( Arrestins 2 and 3 , or β-arrestin-1 and β-arrestin-2 ) ( O'Hayre et al . , 2017 ) . Isoproterenol-stimulated internalization of HA-β2AR was lost in β-arrestin double-knockout ( Arr DKO ) cells , as expected , but AC9-EGFP internalization was still observed . Further , expressing recombinant Arrestin 3 ( β-arrestin-2 ) rescued the HA-β2AR trafficking defect without a noticeable change in AC9-EGFP trafficking ( Figure 5E-H , Figure 5—figure supplement 1C , D ) . These results indicate that AC9 and GPCR trafficking are regulated coordinately but through distinct mechanisms– with AC9 requiring Gs but not β-arrestin , and the GPCR requiring β-arrestin but not Gs . Both AC1 and AC9 are known physiological effectors of β-adrenergic signaling ( Sadana and Dessauer , 2009; Small et al . , 2003; Tantisira et al . , 2005 ) and both are endogenously expressed in HEK293 cells , despite neither being the primary contributor to global cAMP elevation produced by β2AR activation in this cell type . Nevertheless , analysis by isoform-specific knockdown using siRNA indicated that both AC1 and AC9 make a small but statistically significant contribution to the overall cAMP elevation elicited by activation of endogenous β2ARs . AC1 but not AC9 depletion reduced the FSK-induced cAMP response ( Figure 6—figure supplement 1 ) , consistent with AC9 being relatively insensitive to stimulation by FSK ( Baldwin et al . , 2019 ) and verifying specificity of the knockdown approach . Considering that AC9 selectively accumulates in endosomes relative to AC1 , we next investigated the hypothesis that AC9 also selectively contributes to the endosome-initiated component of the β2AR-elicited cellular cAMP response . We tested this hypothesis using a pharmacological approach based on the ability of the membrane-impermeant β2AR antagonist CGP12177 ( CGP ) to access receptors selectively at the plasma membrane , whereas the membrane-permeant antagonist alprenolol accesses receptors both at the plasma membrane and endosomes ( Staehelin et al . , 1983 ) . This approach has been used successfully in previous studies to isolate effects of endosomal activation ( Irannejad et al . , 2013; Thomsen et al . , 2016 ) . We validated it in the present study using a conformational biosensor , Nb80-EGFP , which is recruited specifically and reversibly by activated β2ARs in living cells ( Irannejad et al . , 2013 ) . Isoproterenol application promoted recruitment of Nb80-GFP both to the plasma membrane and endosomes , and application of excess alprenolol rapidly reversed this activation readout at both locations ( Figure 6A , Video 2 ) . Application of CGP , in contrast , reversed Nb80-GFP recruitment only at the plasma membrane but not endosomes ( Figure 6B , Video 3 ) . We next applied this approach to probe the contribution of endosomal β2AR activation to the overall cellular cytoplasmic cAMP response . Both CGP and alprenolol markedly inhibited the isoproterenol-induced cAMP elevation measured at 37 °C in living cells . This verifies that a large fraction of the overall cAMP elevation elicited by endogenous β2AR activation in these cells is initiated from the plasma membrane . However , we also consistently observed a more pronounced inhibition of cellular cAMP elevation following application of alprenolol compared to CGP ( Figure 6C , left set of bars , and Video 4 ) . We interpret this difference as a readout of the component of cAMP production initiated from endosomes . Remarkably , this CGP-resistant ‘signal gap’ was lost after AC9 knockdown but it remained in cells depleted of AC1 ( Figure 6C , D , Figure 6—figure supplement 1C–H ) . As another test of this hypothesis , and to investigate selectivity under conditions of recombinant AC overexpression ( which were necessary for the trafficking studies ) , we asked if similar selectivity can be observed also using tagged AC isoforms . To do so we utilized gene-edited HEK293 cells lacking both AC3 and AC6 ( AC3/6 DKO ) , which were shown previously to provide a reduced background useful for assessing effects of recombinant AC expression on cellular cAMP ( Soto-Velasquez et al . , 2018 ) . The increment of isoproterenol-induced cAMP accumulation produced by overexpressing Flag-AC9 in these cells ( ‘AC9 cAMP response’ ) was blocked by alprenolol but not CGP ( Figure 6E ) . In contrast , the corresponding increment produced by overexpressing Flag-AC1 ( ‘AC1 cAMP response’ ) was blocked by both alprenolol and CGP ( Figure 6F ) . This verifies that AC9 selectively contributes to cellular cAMP production initiated by β2AR activation in endosomes using recombinant , as well as endogenous AC9 . The endocytic network is a dynamically regulated system critical for homeostatic integrity of the cell . From the point of view of GPCR-G protein signaling , this network was believed for many years to be silent , functioning only in signal termination and longer-term modulation of surface receptor number . Such homeostatic effects indeed occur , but an accumulating body of evidence supports an expanded view in which internalized GPCRs reacquire the ability to activate G proteins after endocytosis and initiate a second wave of signaling from endomembrane sites ( Irannejad et al . , 2015; Lohse and Calebiro , 2013; Vilardaga et al . , 2014 ) . Endosomal signaling depends on the presence of a G protein-regulated effector , but whether or how effectors localize to relevant internal membrane locations has remained a relatively unexplored frontier . We approached this frontier by focusing on ACs as important effectors of signaling initiated by GPCR - Gs activation . We demonstrate dynamic and regulated trafficking of AC9 to early endosomes . This compartment is known to accumulate a wide variety of GPCRs , and it has been explicitly shown to be a site of Gs activation by the β2AR ( Irannejad et al . , 2013 ) . AC9 is widely expressed ( Premont et al . , 1996 ) , is a physiologically and clinically relevant effector of β2AR-Gs signaling in particular ( Small et al . , 2003; Sunahara et al . , 1996; Tantisira et al . , 2005 ) and is endogenously expressed in the HEK293 model system used in the present study . We also show that AC9 , while contributing only a minor fraction to the overall cellular cAMP response elicited by β2AR activation in this system , is necessary to produce a specific endosome-initiated component of the endogenous β2AR-elicited cAMP response . Further , we demonstrate that AC9 is sufficient to increase cAMP production from endosomes when expressed as a recombinant protein . Moreover , we show that AC trafficking is isoform-specific because AC1 does not detectably accumulate in endosomes , nor does AC1 contribute detectably to the endosome-initiated component of cellular cAMP signaling ( Figure 6G ) . An important future goal is to identify structural and biochemical determinants of isoform-specific AC trafficking . We note that various isoform-specific protein interactions which impact other aspects of AC organization and function are already known , with AC9 being a particularly well-studied example ( Baldwin et al . , 2019 ) . Another important question for future investigation is whether regulated intracellular trafficking is unique to AC9 or more widespread . We favor the latter possibility because a distantly related AC isoform was previously localized to a multivesicular intracellular compartment in D . discoideum ( Kriebel et al . , 2008 ) . However , in this case , AC trafficking appears to occur through the biosynthetic pathway and it is not known if the AC-containing compartment also contains a relevant GPCR or G protein . We also note that several other transmembrane AC isoforms have been implicated previously in endomembrane cAMP signaling by mammalian GPCRs ( Calebiro et al . , 2009; Cancino et al . , 2014; Ferrandon et al . , 2009; Kotowski et al . , 2011; Mullershausen et al . , 2009; Vilardaga et al . , 2014 ) , and that a distinct AC isoform which lacks any transmembrane domains ( ‘soluble’ AC or AC10 ) has been implicated as well ( Inda et al . , 2016 ) . Thus we anticipate that AC9 is not the only isoform to exhibit discrete trafficking behavior , and that much remains to be learned along this line . In particular , we note that the localization and trafficking properties of AC3 and AC6– which are major contributors to overall cAMP production stimulated by β2ARs in HEK293 cells ( Soto-Velasquez et al . , 2018 ) – have yet to be delineated . One possible mechanism of AC9 trafficking to GPCR-containing endosomes is by physical association with the receptor or receptor-G protein complex , and there is previous evidence indicating that AC5 can form a complex including GPCRs ( Navarro et al . , 2018 ) . However , our results provide two lines of evidence indicating that AC9 traffics independently , despite trafficking via a similar dynamin-dependent membrane pathway as the β2AR and in a coordinated manner . First , activation of Gs is sufficient to promote the accumulation of AC9 but not β2AR in endosomes . Second , AC9 trafficking requires Gs but not β-arrestins , whereas the converse is true for trafficking of the β2AR . Accordingly , AC trafficking is likely subject to different modulatory input ( s ) relative to the trafficking of GPCRs . This is consistent with the difference in environmental sensitivity between AC9 and β2AR trafficking which initially motivated our investigations . However , additional studies will be required to fully elucidate the mechanistic basis for differential control of AC9 trafficking , and to delineate physiological inputs into regulated AC trafficking more broadly . The physiological significance of isoform-specific AC trafficking also remains to be determined , but we note that there is already significant evidence that cAMP produced internally can mediate different downstream signaling effects relative to cAMP produced from the plasma membrane ( O'Banion et al . , 2019; Tsvetanova and von Zastrow , 2014 ) . In closing , to our knowledge the present study is the first to delineate the dynamic endocytic trafficking of a functionally relevant AC isoform , and to identify a role of Gs in regulating the trafficking of a defined AC separately from its catalytic activity . The finding that such AC trafficking is isoform-specific , and regulated separately from its activating GPCR , reveals a new layer of specificity and control in the cAMP system . HEK 293 cells ( CRL-1573 , ATCC , mycoplasma-tested ) were cultured in complete growth Dulbecco’s modified Eagle’s medium ( DMEM , Gibco ) and supplemented with 10% fetal bovine serum ( UCSF Cell Culture Facility ) . HA-β2AR ( Tang et al . , 1999; von Zastrow and Kobilka , 1992 ) , HA-V2R ( Rochdi et al . , 2010 ) , HA-MOR ( Whistler and von Zastrow , 1998 ) , HA-V2R-T ( Charest and Bouvier , 2003; Rochdi et al . , 2010 ) , all described previously , were sub-cloned from Flag-tagged constructs . Nb80-EGFP was previously described ( Irannejad et al . , 2013 ) . HA-G ( alpha ) s , G ( beta-1 ) , G ( gamma-2 ) were gifts from Philip Wedegaertner . HA-G ( alpha ) s-Q227L , a previously described point mutant of Gs that is constitutively active ( Masters et al . , 1989 ) , was made from the original construct using the QuikChange Site-Directed Mutagenesis Kit ( Agilent Technologies ) with the forward primer 5’-CGATGTGGGCGGCCTGCGCGATGAACGCCGC-3’ . Flag-AC1 , Flag-AC9 from the Dessauer Lab , were originally described by Hacker et al . , 1998; Krupinski et al . , 1989; Paterson et al . , 2000; Premont et al . , 1996 . Flag-AC9-D442A ( Catalytic inactive mutant ) was also made from the original construct using QuikChange Kit with the forward primer 5’-CCACTAGTCCAGTGTGGTGGAATTCGCCATGGACTACAAAGACGATGACGAC-3’ . Transfections were carried out using Lipofectamine 2000 ( Life Technologies ) according to the manufacturer’s protocol . Cells were transfected 48 hr before experiments . siRNA knockdown of AC1 and AC9 expression in HEK293 cells was carried out using Lipofectamine RNAiMAX ( Life Technologies ) according to the manufacturer’s protocol . Cells were transfected 72 hr before experiments . Knockdown of AC1 used the siRNA CCGGGCGGTTCAGACCTTCAA and AC9 knockdown used CTGGGCATGAGGAGGTTTAAA . Primary cultures of human airway smooth muscle cells were prepared as described previously ( Tsvetanova et al . , 2017 ) . Cells were passaged no more than five times using Trypsin-EDTA ( Life Technologies ) and maintained in 10% FBS in DMEM . Gs knockout ( Stallaert et al . , 2017 ) and beta-arrestin-1/2 double knockout ( O'Hayre et al . , 2017 ) HEK293 cells were previously described . AC3/AC6 double knockout HEK293 cells were also described previously ( Soto-Velasquez et al . , 2018 ) and were provided as a generous gift by Drs . Monica Soto-Valasquez and Val Watts ( Purdue University ) . Cells were passaged using PBS-EDTA and maintained in 10% FBS in DMEM . Cholera Toxin ( Sigma ) was administered to cells for 16 hr overnight treatment at 10 ng/ml concentration in 10% FBS in DMEM . We found AC9 trafficking to be environmentally sensitive . Specifically , exposure of cells outside of the incubator for more than 2 min tended to reduce the degree of isoproterenol-stimulated internalization of AC9 , without affecting internalization of β2AR . Accordingly , this restriction was consistently adhered to in the present study . Antibodies used were rabbit anti-Flag ( Sigma ) , mouse anti-Flag M1 ( Sigma ) , mouse anti-Flag M2 ( Sigma ) , mouse anti-HA 16B12 ( Biolegend ) , rat anti-HA ( Roche ) , goat anti-AC9 ( Santa Cruz Biotech ) , mouse anti-Golgin-97 ( Thermo ) , rabbit anti-calnexin ( Cell Signaling ) , mouse anti-Sodium/Potassium ATPase ( Fisher ) . Cells were transfected with the indicated construct ( s ) and then plated on glass coverslips coated with poly-L-lysine ( 0 . 0001% , Sigma ) 24 hr later . For antibody feeding assays , cells were: ( 1 ) placed on ice and rinsed with ice-cold phosphate-buffered saline ( PBS ) , ( 2 ) labeled by the addition of antibodies diluted 1:1000 in DMEM for 10 min , and ( 3 ) rinsed with room temperature PBS and allowed to traffic for 30 min by the addition of 37°C fresh media ( DMEM + 10% fetal bovine serum ) with or without a saturating concentration of β2AR agonist ( 10 μM isoproterenol , Sigma ) , V2R agonist ( 10 µM arginine-vasopressin , Sigma ) , MOR agonist ( 10 µM DAMGO , Sigma ) , or forskolin ( 10 µM , Sigma ) . For all assays , cells were rinsed with cold PBS and fixed by incubation in 3 . 7% formaldehyde ( Fisher Scientific ) diluted in modified BRB80 buffer ( 80 mM PIPES , 1 mM MgCl2 , 1 mM CaCl2 , pH 6 . 8 ) for 20 min at room temperature . Cells were then blocked in 2% Bovine Serum Albumin ( Sigma ) in PBS with permeabilization by 0 . 2% triton X-100 ( Sigma ) . Primary antibody labeling was performed by the addition of antibodies diluted 1:1000 in blocking/permeabilization buffer for one hour at room temperature . Secondary labeling was performed by addition of the following antibodies diluted at 1:500 in blocking/permeabilization buffer for 20 min at room temperature: Alexa Fluor 555 or 488 donkey anti-mouse ( Invitrogen ) , Alexa Fluor 555 or 488 donkey anti-rabbit ( Invitrogen ) , Alexa Fluor 488 or 555 goat anti-rat ( Invitrogen ) , or Alexa Fluor 488 donkey anti-sheep ( Life Technologies ) . Specimens were mounted using ProLong Gold antifade reagent ( Life Technologies ) . Fixed cells were imaged by spinning disc confocal microscope ( Nikon TE-2000 with Yokogawa confocal scanner unit CSU22 ) using a 100X NA 1 . 45 objective . A 488 nm argon laser and a 568 nm argon/krypton laser ( Melles Griot ) were used as light sources . Spinning disc images were collected using an electron multiplying CCD camera ( Andor iXon 897 ) operated in the linear range controlled by Micro-Manager software ( https://www . micro-manager . org ) . Images were processed at full bit depth for all analysis and rendered for display by converting to RGB format using ImageJ software ( http://imagej . nih . gov/ij ) and linear look up table . The number of endosomes was quantified by thresholding images and the ParticleTracker ImageJ plugin . Live cell imaging was carried out using Yokagawa CSU22 spinning disk confocal microscope with a × 100 , 1 . 4 numerical aperture , oil objective and a CO2°C and 37°C temperature-controlled incubator . A 488 nm argon laser and a 568 nm argon/krypton laser ( Melles Griot ) were used as light sources for imaging EGFP and Flag signals , respectively . Cells expressing both Flag-tagged receptor and the indicated nanobody–EGFP were plated onto glass coverslips . Receptors were surface labelled by addition of M1 anti-Flag antibody ( 1:1000 , Sigma ) conjugated to Alexa 555 ( A10470 , Invitrogen ) to the media for 30 min , as described previously . Indicated agonist ( isoprenaline , Sigma ) or antagonist ( CGP-12177 , Tocris ) ( alprenolol , Sigma ) were added and cells were imaged every 3 s for 20 min in DMEM without phenol red supplemented with 30 mM HEPES , pH 7 . 4 ( UCSF Cell Culture Facility ) . Time-lapse images were acquired with a Cascade II EM charge-coupled-device ( CCD ) camera ( Photometrics ) driven by Micro-Manager 1 . 4 ( http://www . micro-manager . org ) . Cells were transfected with the indicated construct ( s ) 48 hr before lysis and plated onto 60 mm cell culture dishes 24 hr before lysis . Cells were allowed to traffic for 30 min by the addition of 37°C fresh media ( DMEM + 10% fetal bovine serum ) with or without a saturating concentration of the indicated agonist . Cells were then placed on ice , washed with ice-cold PBS , and scraped into an isotonic homogenization buffer ( 10 mM HEPES , 100 mM KCl , 25 mM sucrose , Complete protease inhibitor ( Roche ) , pH 7 . 2 ) and passaged 20 times through a 22 G BD PrecisionGlide Needle . Whole cell lysates were then spun down at 1000 G for 10 min at 4°C and the pellets discarded . The supernatant was then bound to Early Endosome Antigen one mouse antibody ( 1:250 , Fisher Scientific ) and anti-mouse IgG magnetic microbeads ( Miltenyi Biotech ) overnight . Endosomes were then bound to magnetic columns which were blocked with 3% BSA and washed with PBS . Proteins in the isolated fraction were eluted with 0 . 1% Triton-X and characterized by western blot . Cells were transfected with the indicated construct ( s ) 48 hr before lysis and plated onto 60 mm cell culture dishes coated with poly-L-lysine ( 0 . 0001% , Sigma ) 24 hr before lysis . Cells were allowed to traffic for 30 min by the addition of 37°C fresh media ( DMEM + 10% fetal bovine serum ) with or without a saturating concentration of the indicated agonist . Cells were then placed on ice , washed with ice-cold PBS , and then surface labeled with EZ-link Sulfo-NHS-biotin ( Pierce ) for 30 min , rocking at 4°C . Reaction was then quenched with tris buffered saline ( TBS ) twice for 10 min . Cells were then placed on ice , washed with ice-cold PBS , and scraped into an isotonic homogenization buffer ( 10 mM HEPES , 100 mM KCl , 25 mM sucrose , Complete protease inhibitor ( Roche ) , pH 7 . 2 ) and passaged 20 times through a 22 G BD PrecisionGlide Needle . Cell lysate was then bound to streptavidin agarose resin ( Thermo ) overnight . Resin was spun down and the supernatant discarded , resuspended and washed in ice-cold PBS , and characterized by western blot . Real-time analysis of cAMP elevations were carried out in living HEK293 cells and in the absence of phosphodiesterase inhibitors using a were transfected with a plasmid encoding a cyclic-permuted luciferase reporter construct , based on a mutated RIIβB cAMP-binding domain from PKA ( pGloSensor-20F , Promega ) , which produces rapid and reversible cAMP-dependent activation of luciferase activity in intact cells and is capable of detecting cAMP elevations in the absence of phosphodiesterase inhibitors . Cells were plated in 24-well dishes containing approximately 200 , 000 cells per well in 500 μl DMEM without phenol red and no serum and equilibrated to 37°C in a light-proof cabinet . An image of the plate was focused on a 512 × 512 pixel electron multiplying CCD sensor ( Hamamatsu C9100-13 ) , cells were equilibrated for 1 hr in the presence of 250 μg ml−1 luciferin ( Gold biosciences ) , and sequential luminescence images were collected every 10 s to obtain basal luminescence values . The camera shutter was closed , the cabinet opened and the indicated concentration of isoprenaline was bath applied , with gentle manual rocking before replacing in the dark cabinet and resuming luminescence image acquisition . In endocytic manipulation experiments , cells were pre-incubated with 30 μM Dyngo-4a ( abcam Biochemicals ) for 15 min . Every 10 s , sequential images were acquired using Micro-Manager ( http://www . micro-manager . org ) and integrated luminescence intensity detected from each well was calculated after background subtraction and correction for vignetting using scripts written in MATLAB ( MathWorks ) . In each multiwell plate , and for each experimental condition , a reference value of luminescence was measured in the presence of 5 μM forskolin , a manipulation that stimulates a moderate amount of receptor-independent activation of adenylyl cyclase . The average luminescence value—measured across duplicate wells—was normalized to the maximum luminescence value measured in the presence of 5 μM forskolin . A biochemical assay of cAMP accumulation was used to determine the effects of AC mutation on catalytic activity , with high sensitivity and without dependence on subcellular location due to inhibition of cellular phosphodiesterases . Briefly , cells were pre-incubated in the presence of 1 mM IBMX ( Sigma ) for 30 min at 37°C in Dulbecco's modified Eagle's medium followed , and then incubated for an additional 10 min in absence or presence of isoproterenol ( in the continued presence of IBMX ) , as indicated . Cells were quickly washed with ice-cold PBS and lysed by exposure to 0 . 1 M HCl for 10 min at room temperature . The cAMP concentration in lysates was determined using a commercial immunoassay ( Direct cAMP ELISA kit , Enzo Life Sciences , Farmingdale , NY ) according to the manufacturer’s instructions . Results are displayed as the mean of results from each experiment or data set , as indicated in figure legends . The statistical significance between conditions for experiments with two conditions was calculated using paired , two tailed t-tests . All statistical calculations were performed using Excel ( Microsoft Office ) or Prism ( GraphPad ) . The threshold for significance was p<0 . 05 and the coding for significance is reported as follows: ( n . s . ) p>0 . 05 , ( * ) p≤0 . 05 , ( ** ) p≤0 . 01 .	Cells sense changes in their chemical environment using proteins called receptors . These proteins often sit on the cell surface , detecting molecules outside the cell and relaying messages across the membrane to the cell interior . The largest family of receptors is formed of ‘G protein-coupled receptors’ ( or GPCRs for short ) , so named because they relay messages through so-called G proteins , which then send information into the cell by interacting with other proteins called effectors . Next , the receptors leave the cell surface , travelling into the cell in compartments called endosomes . Researchers used to think that this switched the receptors off , stopping the signaling process , but it is now clear that this is not the case . Some receptors continue to signal from inside the cell , though the details of how this works are unclear . For signals to pass from a GPCR to a G protein to an effector , all three proteins need to be in the same place . This is certainly happening at the cell surface , but whether all three types of proteins come together inside endosomes is less clear . One way to find out is to look closely at the location of effector proteins when GPCRs are receiving signals . One well-studied effector of GPCR signaling is called adenylyl cyclase , a protein that makes a signal molecule called cAMP . Some G proteins switch adenylyl cyclase on , increasing cAMP production , while others switch it off . To find out how GPCRs send signals from inside endosomes , Lazar et al tracked adenylyl cyclase proteins inside human cells . This revealed that a type of adenylyl cyclase , known as adenylyl cyclase 9 , follows receptors as they travel into the cell . Under the influence of active G proteins , activated adenylyl cyclase 9 left the cell surface and entered the endosomes . Once inside the cell , adenylyl cyclase 9 generated the signal molecule cAMP , allowing the receptors to send messages from inside the cell . Other types of adenylyl cyclase behaved differently . Adenylyl cyclase 1 , for example , remained on the cell surface even after its receptors had left , and did not signal from inside the cell at all . Which cell behaviors are triggered from the membrane , and which are triggered from inside the cell is an important question in drug design . Understanding where effector proteins are active is a step towards finding the answers . This could help research into diseases of the heart , the liver and the lungs , all of which use adenylyl cyclase 9 to send signals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2020	G protein-regulated endocytic trafficking of adenylyl cyclase type 9
Segmentation and tracking of cells in long-term time-lapse experiments has emerged as a powerful method to understand how tissue shape changes emerge from the complex choreography of constituent cells . However , methods to store and interrogate the large datasets produced by these experiments are not widely available . Furthermore , recently developed methods for relating tissue shape changes to cell dynamics have not yet been widely applied by biologists because of their technical complexity . We therefore developed a database format that stores cellular connectivity and geometry information of deforming epithelial tissues , and computational tools to interrogate it and perform multi-scale analysis of morphogenesis . We provide tutorials for this computational framework , called TissueMiner , and demonstrate its capabilities by comparing cell and tissue dynamics in vein and inter-vein subregions of the Drosophila pupal wing . These analyses reveal an unexpected role for convergent extension in shaping wing veins . Understanding how cells collectively shape a tissue is a long-standing question in developmental biology . We recently addressed this question by analyzing morphogenesis of the Drosophila pupal wing at cellular resolution ( Etournay et al . , 2015 ) . To understand the cellular contributions to pupal wing shape changes , we quantified the spatial and temporal distribution of both cell state properties ( e . g . cell area , shape and packing geometry ) , as well as dynamic cellular events like rearrangements , divisions , and extrusions . We quantitatively accounted for wing shape changes on the basis of these cellular events . By combining these analyses with mechanical and genetic perturbations , we were able to develop a multiscale physical model for wing morphogenesis and show how the interplay between epithelial stresses and cell dynamics reshapes the pupal wing . Researchers interested in epithelial dynamics face similar challenges in processing and analyzing time-lapse movie data . Quantifying epithelial dynamics first requires image-processing steps including cell segmentation and tracking , to digitalize the time-lapse information . Recently , software tools for segmentation and tracking have become generally available ( Aigouy et al . , 2010; Mosaliganti et al . , 2012; Sagner et al . , 2012; Barbier et al . , 2015; Cilla et al . , 2015; Wiesmann et al . , 2015; Heller et al . , 2016; Aigouy et al . , 2016 ) . However , more advanced analysis is required to quantify , interpret and visualize the information derived from segmentation and tracking . Epithelial cells share a set of core behaviors , such as division , rearrangement , shape change and extrusion , which underlie a wide variety of morphogenetic events in different tissues . Methods for analyzing these core behaviors have been developed independently in several labs ( Blanchard et al . , 2009; Bosveld et al . , 2012; Etournay et al . , 2015; Guirao et al . , 2015 ) . However , these analysis tools have not yet been made available to other users in an easy to use and well-documented form . Here , we propose a generic data layout and a comprehensive and well-documented computational framework called TissueMiner ( see Box 1 ) for the analysis of epithelial dynamics in 2D . It enables biologists and physicists to quantify cell state properties and cell dynamics , their spatial patterns and their time evolution in a fast , easy and flexible way . It also facilitates the comparison of quantities within and between tissues . To make TissueMiner accessible to a novice , we provide tutorials that guide the user through its capabilities in detail and release a workflow that automatically performs most of the analysis and visualization tasks we reported previously for Drosophila pupal wings ( Etournay et al . , 2015 ) . These tutorials operate using one small example dataset and 3 large wild-type datasets corresponding to the distal wing blade , which we also provide . The code for TissueMiner , along with tutorials and datasets , are publically available ( Box 1 ) . We illustrate the utility and power of these tools by performing a more extensive analysis of pupal wing morphogenesis focused on differences in the behavior of vein and inter-vein cells . Wing veins are specified during larval stages , but only become morphologically distinct during prepupal and pupal morphogenesis . During pupal morphogenesis , the dorsal and ventral surfaces of the wing epithelium become apposed to each other on their basal sides , except in the regions that will give rise to veins - here the basal surfaces of dorsal and ventral cells form a lumen . Vein and inter-vein cells also differ on their apical surfaces . Vein cells have a narrower apical cross-section and form corrugations that protrude from the dorsal and ventral surfaces of the wing blade . The cell dynamics underlying vein morphogenesis have never been quantitatively examined . Before conducting any analysis , the TissueMiner automated workflow reads three configuration files that contain ( 1 ) user-defined regions of interest ( ROI’s ) , ( 2 ) time offsets for movie synchronization , and ( 3 ) the rotation angle used to align the tissue to a standard orientation ( Figure 1—figure supplement 1 ) . An important step in analyzing tissue morphogenesis is to quantify cell state properties over time . These properties include cell area , shape anisotropy and packing geometry . In this section , we demonstrate the analysis and visualization tools of TissueMiner by comparing how these state properties evolve during wing morphogenesis in vein and inter-vein regions . Morphogenesis is often characterized by changes in cell area and elongation . In the TissueMiner workflow , these properties are calculated from the original segmentation masks and stored in the database ( Materials and methods ) . To visualize the evolution of the cell area pattern at the scale of the whole tissue , we map the area values of each individual cell to a gradient color scale ( see Figure 2A–A’ , Video 4 ) . Each cell contour is filled with a color that corresponds to its area . Figure 2A’ shows the pattern of cell areas in the wing at the end pupal wing blade elongation . This visualization scheme reveals that cells in the proximal hinge and in wing veins have a smaller cross-sectional area ( blue ) at this time . 10 . 7554/eLife . 14334 . 009Figure 2 . Patterned cell state properties in the developing pupal wing of Drosophila . ( A–D' ) Cell state patterns at 22 hr and 31 hr after puparium formation ( hAPF ) . ( A–A' ) Color-coded cell area . ( B–B' ) Color-coded cell elongation . The magnitude of cell elongation corresponds to the norm of the cell elongation nematic tensor . ( C–C'' ) Coarse-grained pattern of cell elongation nematics and ( C'' ) cell elongation nematics represented as bars on each individual cell . The wing was divided into adjacent square-grid elements of 33x33 microns in which cell elongation nematics were averaged . ( D–D' ) Color-coded representation of the cell neighbor number . ( E ) Time evolution of the average cell area in different regions of interest: wing blade ( Figure 1B ) , veins ( Figure 1E ) , and inter-vein regions . ( F ) Time evolution of the average cell elongation magnitude in the blade , veins and inter-vein regions . Scale bar: 50 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 00910 . 7554/eLife . 14334 . 010Video 4 . Color-coded cell area pattern . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 010 Cell elongation is characterized by a nematic tensor describing the axis and magnitude of the elongation ( Aigouy et al . , 2010 ) . As with cell area , we map the magnitude of cell elongation to a color scale ( Figure 2B–B’ , Video 5 ) . This fine-grained quantification of cell elongation highlights striking differences between inter-vein and vein cells . Inter-vein cells are more elongated than vein cells at 22 hr after puparium formation ( hAPF ) , but this pattern is reversed by 31 hAPF . 10 . 7554/eLife . 14334 . 011Video 5 . Color-coded cell elongation norm pattern . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 011 The color scale above reveals only the magnitude of the tensor . To visualize both the magnitude and direction of cell elongation , we represent the elongation nematic as a line whose length and angle correspond to the magnitude and angle of cell elongation , respectively . Nematics can also be averaged across multiple cells in a region in order to coarse-grain the patterns and highlight the main features ( Figure 2C–C’’ , Video 6 ) . For example , the coarse-grained elongation nematics shown in Figure 2C , highlight the global alignment of cell elongation in the proximal-distal direction at 22 hAPF . 10 . 7554/eLife . 14334 . 012Video 6 . Coarse-grained cell elongation pattern . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 012 Cells in the wing become progressively more hexagonal during pupal wing morphogenesis ( Classen et al . , 2005 ) . To visualize packing geometry , we map the neighbor number of each cell to a discrete color code ( Figure 2D–D’ , Video 7 ) . This makes changes in packing geometry during morphogenesis immediately obvious ( 22 and 31 hAPF ) . 10 . 7554/eLife . 14334 . 013Video 7 . Color-coded cell packing pattern . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 013 The visualization tools described above effectively reveal detailed spatial patterns of cell properties . To highlight how average cell properties change over time , and to facilitate comparison between movies and ROI’s , TissueMiner also provides tools to create plots of average quantities as a function of time . In Figure 2E and Figure 2F , we compare the time evolution of the average cell area and the average cell elongation in movies of the 3 WT wings ( blue , green , red ) used in ( Etournay et al . , 2015 ) . The plots in Figure 2 compare the time evolution of average cell elongation and area values for vein and inter-vein cells . We previously showed that average cell area in the wing blade decreases during morphogenesis , but that cell area decrease is balanced by cell divisions to maintain wing blade area . Quantifying average area values in vein and inter-vein ROI’s reveals that vein cells contract over a longer period of time than inter-vein cells , and thus have a smaller cross-sectional area at the end of morphogenesis ( Figure 2F ) . As previously described , cells in the wing blade elongate and then relax their shapes during pupal wing morphogenesis ( Etournay et al . , 2015 ) ( Figure 2E , blade part ) . Plotting elongation in vein and inter-vein ROI’s reveals that vein cells elongate more slowly and also relax their elongation more slowly than inter-vein cells . These differences suggest that vein and inter-vein cells have different mechanical properties . Oriented tissue morphogenesis may reflect the number , orientation and spatio-temporal pattern of cell divisions . TissueMiner provides several tools to visualize these events . Overlaying color-coded generation number on a pupal wing movie reveals patterns of cell divisions as they occur ( Video 8 ) , and examining the last frame of the movie ( Figure 3A ) reveals the cumulative pattern of cell divisions . This analysis is largely consistent with the cell division timing inferred from classical BrdU pulse-chase experiments ( Schubiger and Palka , 1987; Garcia-Bellido et al . , 1994; Milan et al . , 1996 ) , but also reveals unexpected additional features . The pattern of cell divisions correlates with veins: most cells in the wing blade divide only once during pupal morphogenesis , whereas in some parts of inter-vein regions they divide twice . These include the cells lying adjacent to veins L3 , L4 and L5 , and the region posterior to L5 . We estimate the median cell-cycle length between the first and second rounds of cell divisions to be ( 5 . 25 ± 1 . 50 ) hr . 10 . 7554/eLife . 14334 . 014Video 8 . Color-coded cell generation pattern . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 01410 . 7554/eLife . 14334 . 015Figure 3 . Visualization of cell generations and cell divisions . ( A ) Color-coded pattern of cell generations . The wing cartoon on the bottom right shows the names of subregions that we analyze in panel B . Scale bar 50 microns . ( B ) Cell division rate in different regions of interest . To smooth fluctuations , these rates were averaged in discrete time intervals of one hour ( TM R-User Manual , section 3 . 7 ) . We further averaged these rates amongst the three wild-type wings . Error bars depict the standard deviation between wings . Cells divide earlier in veins L2 and L4 than in L3 and L5 . Two maxima corresponding to two rounds of divisions are visible in inter-vein regions: interL2-L3 , distInterL3-L4 and postL5 . ( C–C' ) A dividing cell with its unit nematic depicting the division orientation . Scale bar 10 microns . ( D ) Coarse-grained pattern of cell division orientation ( grid size of 33x33 microns ) . Scale bar 50 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 015 To further investigate how cell divisions are patterned in the blade , we quantified the time evolution of cell division rates in each vein and inter-vein region ( Figure 3B ) . This analysis reveals differences in the timing and numbers of cell divisions in these different ROI’s . Cells in veins L2 and L4 divide before those in L3 and L5 . These divisions are followed by a second peak of division in the inter-vein regions distInterL3-L4 , interL2-L3 and postL5 ( see cartoon in Figure 3A ) . To more easily visualize the spatio-temporal pattern of divisions in veins only , the powerful tools available in TissueMiner allow us to assign vein cells a color corresponding to the time at which they divide: blue for 16–18 hAPF and red for 18–20 hAPF ( see Video 9 ) . This analysis reveals more detailed patterning in division timing . Cell divisions in vein regions that protrude ventrally ( L2 and proximal L4 ) , peak at the same time and earlier than those that protrude dorsally ( L3 , distal L4 and L5 ) . Precise correlation of cell divisions with specific vein and inter-vein regions suggests that they are autonomously controlled by signaling associated with veins . 10 . 7554/eLife . 14334 . 016Video 9 . Color-coded cell division pattern in veins and by time intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 016 To measure the orientation of cell divisions , we define a unit nematic tensor ( see Materials and methods ) . For each cell division , the orientation of this unit nematic is defined by the line connecting the centers of mass of the two daughter cells when they first appear ( see Figure 3C–C' , and TM R-User Manual section 2 . 8 ) . Each nematic is assigned a position on the tissue that corresponds to the center of combined mass of the two daughter cells . To visualize division orientation patterns , unit nematics can be added within different regions and averaged over different time intervals ( Figure 3D , Video 10 , TMR-User Manual section 2 . 9 ) . 10 . 7554/eLife . 14334 . 017Video 10 . Coarse-grained cell division pattern . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 017 Epithelial tissues can be reshaped by cell rearrangements , or T1 transitions ( for review [Walck-Shannon and Hardin , 2014] ) . In the simplest case , a T1 transition involves two pairs of cells , that exchange neighbors by disassembling one cell-cell contact and replacing it by another – bringing together two previously separated cells ( Figure 4A ) . In reality , cell contacts may undergo multiple rounds of shrinkage and regrowth before resolving ( Figure 4A' ) . Furthermore some epithelia undergo the related process of rosette formation where multiple cell junctions are disassembled before new neighbors are brought into contact . By separately quantifying the orientation with which cell contacts are gained and lost , one can reveal whether there is a net directionality to cell junction assembly and disassembly . To identify gained and lost cell contacts , we compare cell neighbor relationships between 2 subsequent frames . We exclude changes in neighbor relationships resulting from cell division , extrusion or a cell moving in and out of the field of view . The remaining neighbor relationship changes are used to define cell contacts that have appeared or disappeared . 10 . 7554/eLife . 14334 . 018Figure 4 . Visualization and quantification of T1 transitions . ( A–A' ) Cartoon depicting an effective T1 transition ( A ) that corresponds to cell-contact loss and gain in different directions . Each contact loss or gain is assigned a unit nematic describing its orientation . ( B–B' ) Pattern of cells losing contact ( green ) , gaining contact ( red ) or both ( blue ) . ( C ) Rate of neighbor change per cell and per hour in the blade , veins and inter-vein regions of interests . Rates were averaged within discrete time intervals of one hour and further averaged among the 3 WT wings ( TM R-User Manual , section 3 . 8 ) . Error bars depict the standard deviation amongst wings . ( D ) Coarse-grained pattern of neighbor exchange orientation at 17 hAPF . Cell neighbor change nematics were obtained by summing up unit nematics in each grid elements of 33x33 microns and further averaged in time using a 50 min time window . Scale bar 50 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 01810 . 7554/eLife . 14334 . 019Figure 4—figure supplement 1 . T1 and cell elongation nematic orientation . ( A ) Cell neighbor change nematics were averaged at each frame within each region of interest and are represented as bars in a circular diagram . The bar angle indicate the average T1 orientation , and its length ( nematic norm ) reflects how ordered cell neighbor change nematics are in a given region of interest . Their color depicts the developmental time in hours after puparium formation . ( B ) Cell elongtation nematics were also averaged at each frame within each region of interest . The average T1 nematic orientation starts to match the average cell elongation nematic orientation from about 22 hAPF ( peak of cell stretch ) on , when stress-induced PD-oriented T1 dominate over autonomous AP-oriented T1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 019 We characterize the orientation of contact gains and losses by assigning them a unit nematic tensor . For contact loss , the orientation of the nematic is defined by the axis intersecting the two cell centers . For contact gain , the orientation of the nematic is perpendicular to the axis intersecting the two cell centers ( Figure 4A–A’ ) . If there is a simple disappearance and reappearance of a single cell contact , corresponding nematics will cancel out . Therefore , the sum of contact gain and contact loss nematics over time and/or space will represent an effective T1 nematic describing net direction of contact assembly/disassembly . The rate of contact gain and loss can be visualized in different ways . Cell contact dynamics can be viewed directly on movies of tissue morphogenesis by assigning colors to cells as they gain ( red ) or lose ( green ) contacts . Those cells that simultaneously gain and lose different cell contacts are colored blue ( Figure 4B–B’ ) . The frequency of contact gain and loss , independent of orientation , can be plotted over time . Figure 4C compares the frequency of contact assembly/disassembly in vein and inter-vein regions . In both regions , this rate begins to decrease in the second half of morphogenesis . To visualize the pattern of orientation of T1 transitions throughout the wing , we sum contact gain and loss nematics over square grid elements , and average over a chosen time window ( about 50 min in Figure 4D , Video 11 , see TM R-User Manual section 2 . 12 ) . 10 . 7554/eLife . 14334 . 020Video 11 . Coarse-grained cell rearrangement patternDOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 020 Finally , the average orientation of effective T1 nematics in sub-regions over time can be visualized using circular diagrams , where nematics are color-coded to indicate developmental time . Figure 4—figure supplement 1A reveals that the orientation of effective T1’s is along the anterior-posterior ( AP ) axis early ( blue ) and shifts to the proximal-distal ( PD ) axis in the second half of morphogenesis ( red ) . A similar approach can be used to illustrate average cell elongation nematics over time ( Figure 4—figure supplement 1B ) . While it is useful to quantify the number and orientation of cellular events like elongation , rearrangement , extrusion and division , this by itself does not provide quantitative information about the amount of tissue shape change contributed by each type of event . We therefore devised a method to measure deformation caused by these cellular processes such that they sum to the measured tissue deformation . Tissue deformation can be decomposed into isotropic and anisotropic parts that distinguish changes in area ( compression/expansion ) from changes in aspect ratio ( pure shear , for details see also Materials and methods ) . The quantities describing area changes are scalar , whereas the quantities describing shear rate in a 2D-network are nematic tensors harboring two distinct components that describe the orientation and magnitude of the shear . Tissue area changes can be calculated based on cell area change and the number of cells gained and lost by divisions and extrusions – information that is all available in the TissueMiner database ( Etournay et al . , 2015 ) . To quantify the cellular contributions to anisotropic tissue deformation , TissueMiner uses the so-called Triangle Method , which is based on a triangular tiling of the junctional network ( Etournay et al . , 2015; Merkel et al . , 2016 ) . Triangle elongation is a proxy for cell elongation , and topological changes in the network result in redrawing of triangles ( Figure 5A–C ) . The resulting change in average triangle elongation can be used to calculate the shear due to the topological changes ( Etournay et al . , 2015 ) . In addition to contributions from divisions , cell rearrangements , extrusions and cell shape changes , the method also takes into account deformation caused by correlations between elongation and both area change and rotation . 10 . 7554/eLife . 14334 . 021Figure 5 . Visualization and quantification of anisotropic cell and tissue deformation . ( A ) Triangulation of the cell network: each triangle vertex corresponds to a cell center . ( B–B' ) Cartons depicting triangle pure shear and total tissue shear along the x axis . ( C ) Cartons depicting shear due to T1 transition , cell division and extrusion . ( D ) Pattern of local tissue shear rate obtained from the triangulation method . Scale bar 50 microns . ( E ) shows the average rate of tissue shear ( blue ) in the blade , interveins and veins , and the corresponding cellular shear contributions ( other colors ) . Shaded regions indicate the standard deviation amongst wings . ( F ) shows the accumulated tissue shear over time and the accumulated contributions of each type of cellular event . The tissue shear ( blue ) in veins is orientated along the PD axis and it is higher than in inter-vein regions during most of pupal morphogenesis . It leads to an extension along the PD axis and to a narrowing along the anterior-posterior ( AP ) direction . By the end of the movie , accumulated tissue shear ( blue ) is almost twice as high in veins as in inter-vein regions . Shaded regions represent the standard deviation amongst wings . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 02110 . 7554/eLife . 14334 . 022Figure 5—figure supplement 1 . Measurements of cell and tissue deformation from two computer-generated sheets of hexagonal cells . ( A–D ) One dataset corresponds to hexagonal cells undergoing a constant isotropic expansion rate of 3 . 50 10–2 per frame , and the other corresponds to hexagonal cells undergoing constant pure shear rate of 1 . 75 10–2 per frame . These datasets are termed iso . exp movie and shear movie respectively in graphs . There isn't any topological change . To keep consistent sets of cells in time , we filtered out cells that become in contact to the image border . We then performed our measurement on these tracked regions of about 50 cells in the shear movie and about 100 cells in the iso . exp movie . ( A ) Relative tissue area changes ( blue ) and its decomposition into cell area changes ( green ) , cell number increase by divisions ( orange ) and cell number descrease by extrusions ( cyan ) . Their corresponding cumulative sums are shown in ( B ) . ( C ) shows the average tissue shear ( blue ) and its decomposition into cellular shear contributions ( other colors ) . Their corresponding cumulative sums are shown in ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 02210 . 7554/eLife . 14334 . 023Figure 5—figure supplement 2 . Tissue isotropic deformation and cellular contributions in different regions . ( A ) Relative rates of tissue area changes ( blue ) averaged over 3 WT wings for the blade , veins and interveins , and its decomposition into cell area changes ( green ) , cell number increase by divisions ( orange ) and cell number descrease by extrusions ( cyan ) . Their corresponding cumulative sums are shown in ( B ) . ( B ) Cumulative tissue area changes and its cellular contributions . Shaded regions represent the standard deviation amongst wings . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 02310 . 7554/eLife . 14334 . 024Figure 5—figure supplement 3 . Comparison of patterns of cell event orientation with their correponding quantitative patterns of shear . ( A–A' ) Coarse-grained patterns of cell division orientation ( A ) and of shear contributed by cell division ( A' ) . The pattern shown in ( A ) was obtained by summing up cell division nematics in each grid element and by further averaging in time . The pattern shown in ( A' ) was obtained by averaging the shear nematics in each grid element and by further averaging in time . ( B–B' ) Coarse-grained patterns of neighbor-change orientation ( B ) and of shear contributed by neighbor changes ( B' ) . These patterns were obtained similarly as for cell divisions . Only the shear patterns ( A' and B' ) obtained with the triangulation method provide a quantitative measurement of the local deformation induced by each type of cellular event . Square-grid size of 26x26 microns . Time averaging covering about 55 min ( 11 frames ) in each grid element . Scale bar 50 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 024 To test the reliability of TissueMiner in calculating large cell and tissue deformations , we created two computer-generated movies of hexagonal cells sheets ( Videos 12 , 13 ) . In one movie , we imposed a constant isotropic expansion rate of 3 . 50 10−2 per frame , without any anisotropic deformation . In the second movie , we imposed a constant pure shear along the x-axis with a rate of 1 . 75 10−2 per frame , and without any isotropic expansion . The amounts of isotropic expansion and pure shear have been chosen to be at least 10 times higher than what we measure between subsequent frames of pupal wing movies . 10 . 7554/eLife . 14334 . 025Video 12 . Computer-generated hexagonal cells with an imposed shear rate . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 02510 . 7554/eLife . 14334 . 026Video 13 . Computer-generated hexagonal cells with an imposed isotropic expansion rate . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 026 We then asked if TissueMiner could quantitatively recapitulate the respectively imposed deformation rates . In each dataset , TissueMiner automatically defines a 'whole_tissue' region of interest that corresponds to a consistent set of cells that are always visible ( about 100 cells in the isotropic expansion movie and about 50 cells in the pure shear movie , green labels in Videos 12 and 13 ) . All measurements are done in this ROI to avoid measuring deformation due to inward and outward cell flows . Figure 5—figure supplement 1 shows the time evolution of the measured tissue expansion rate ( panel A ) and tissue shear rate ( panel C ) that were averaged over the 'whole_tissue' ROI , and their respective cellular contributions . Panels B and D show the corresponding cumulated curves . As expected , in the isotropic expansion movie we observe a nearly constant isotropic expansion rate , which is accounted for by the cell area change contribution . We measure an average expansion rate of ( 3 . 53 ± 0 . 04 ) 10–2 per frame , which is consistent with the value imposed when creating the movie . The measured uncertainty is the 95% confidence interval of the standard error of the mean . The pure shear rate and its cellular contributions nearly vanish in this movie ( Figure 5—figure supplement 1C , D ) . For the pure shear movie , we measure an approximately constant horizontal component of the pure shear rate of ( 1 . 74 ± 0 . 02 ) 10–2 per frame , which is consistent with the value imposed when creating the movie . This pure shear rate is entirely accounted for by cell elongation change . The isotropic expansion rate and its cellular contributions nearly vanish ( Figure 5—figure supplement 1A , B ) . Other contributions to expansion and shear rates are negligible in both movies . The pixelated nature of individual cell contours contributes to fluctuations of our measured values . Moreover , we find that these fluctuations cancel out when cumulating the deformation ( Figure 5—figure supplement 1B and D ) . Thus , the current implementation of TissueMiner captures the tissue isotropic expansion and pure shear rates as well as the corresponding cellular contributions with a good precision in these computer-generated movies . Figure 5—figure supplement 2 shows the rate of relative area change and cumulative area change of vein and inter-vein regions over time , as well as the cellular contributions to these area changes . As previously noted , the area of the blade as a whole changes very little . However sub-region analysis reveals that inter-vein expansion compensates for compression in vein regions . Vein cells not only divide less than inter-vein cells , but also decrease their area more . Next we use the Triangle Method to calculate pure shear rates in the time-lapse movies of developing pupal wings . To visualize the spatial pattern of pure shear rate in the wing , TissueMiner allows us to plot nematics corresponding to the local tissue shear rates ( Figure 5D ) and to rates of shear produced by different cellular contributions ( Figure 5—figure supplement 3 , and [Etournay et al . , 2015] ) averaged within the squares of about 26 x 26 microns . To compare the time evolution of pure shear rate between different tissue subregions we plot this rate averaged over the corresponding ROI ( Figure 5E–F and [Etournay et al . , 2015] ) . A positive sign for shear indicates an extension along the PD axis and a contraction along the AP axis , whereas a negative sign indicates an extension along the AP axis and a contraction along the PD axis . As reported previously , the wing blade as a whole shears along its PD axis between 16 and 32 hAPF . T1 transitions and cell elongation are major contributors to total PD shear , and they display complementary behavior that evolves over time . In the first phase , cells elongate in the PD axis in response to tissue stresses generated by hinge contraction , and by actively oriented T1 transitions that occur first along the AP axis . In the second phase , cell elongation causes the orientation of T1 transitions to shift 90˚ from the AP to the PD axis ( Etournay et al . , 2015 ) . These PD oriented T1 transitions both contribute to tissue shear and relax PD cell elongation . We now compare shear and cellular contributions to shear in vein and inter-vein regions . Tissue shear peaks earlier in inter-vein regions than in veins , but veins shear more overall . Examining the cellular contributions to shear suggests that increased shear in veins reflects a different relationship between cell elongation and T1 transitions . PD-oriented T1 transitions do not only produce more shear in veins , they also fail to relax PD cell elongation as much as in inter-vein regions . Quantitative image analysis of developing epithelia is a powerful approach to understanding morphogenesis , but the tools with which to tame and analyze these complex data have not been widely available in a standard and well-documented format . Here we provide an introduction to the capabilities of TissueMiner and tutorials for its use . TissueMiner provides general strategy to store and analyze large data sets of interwoven objects by combining state of the art tools for data mining . It allows quantification and visualization of epithelial morphogenesis at multiple scales – from individual cells to entire tissues . It provides both a generic database format and a multi-platform toolkit to interrogate and visualize data and quantify cellular contributions to large-scale epithelial deformations . TissueMiner has been designed to be versatile and expandable . The database format we provide standardizes the organization of tracked cell data and collects all data into a single file per movie . Such a standardized data format facilitates data sharing between different sources , thereby enhancing cross-laboratory reproducibility . As the database stores positional information about cells and cell contacts , as well as cell neighbor topology , it could also be useful for parameterizing simulations of epithelial remodeling by vertex models or other physical network models . The scheme of our relational database is expandable: additional properties of cells , bonds and vertices can be appended to the database without affecting the relationships between tables . As a consequence , our current query tools to interrogate the database remain functional , even if the database is extended with new properties of cells , bonds and vertices . TissueMiner takes advantage of the advanced graphical capabilities of R and Python to enable the visualization of patterns of deformation and cell state properties directly on the movie images or quantitatively summarized in graphs . In particular , R provides a flexible grammar with which to manipulate tables obtained from the database and to easily plot graphs ( Wickham , 2009; Francois and Wickham , 2015 ) . TissueMiner also offers multiple options for coarse-graining data in space and time through an expandable collection of scripts , which constitutes the TissueMiner library for R or Python . These two easy-to-learn programming languages give TissueMiner its great flexibility to both address general questions of epithelial morphogenesis and project-specific questions , and enable automation , parallelization and customization of user-specific workflows . The tools underlying TissueMiner were originally developed to understand the interplay of cell dynamics and epithelial tension on the developing wing of the fruit fly , where we described cellular contributions to pupal wing morphogenesis averaged throughout the entire wing blade ( Etournay et al . , 2015 ) . Here , to illustrate the utility of the TissueMiner framework , we compared the behavior of vein and inter-vein regions in the developing pupal wing . Comparing cell dynamics in veins and inter-vein regions provided an unexpected explanation for the process of 'vein refinement' . Vein refinement refers to the fact that veins become narrower during pupal morphogenesis . This had been interpreted as a signaling-dependent reduction in the number of cells assuming the vein fate ( Blair , 2007 ) . Here we show instead that vein narrowing results from a convergent extension-like process that is stronger in veins than in inter-vein regions . This elongates and narrows the veins without reducing vein cell number . It will be interesting to examine how signaling pathways involved in vein refinement influence cell dynamics in veins during morphogenesis . The standardization of analysis that TissueMiner provides will facilitate these and other comparisons critical for deciphering the molecular mechanisms underlying epithelial morphogenesis . The knock-in Ecad::GFP fly line ( Huang et al . , 2005 ) was used for live imaging of the developing pupal wing . Flies were raised and maintained at 25°C during imaging by using a temperature-controlled chamber equipped with a humidifier to prevent desiccation . Long-term time-lapse imaging was performed as previously described ( Etournay et al . , 2015 ) . After the imaging session , flies were maintained in a humid environment where they eclosed at the term of pupal development . The visualization and quantification of cell dynamics underlying tissue morphogenesis rely on the ability to extract information about cell geometry , cell neighbor topology and cell histories from time-lapse movies ( Aigouy et al . , 2010; Etournay et al . , 2015 ) . We use TissueAnalyzer to segment and track the cell network over time . This results in a series of digital images that contain this information ( Figure 6—figure supplement 1 ) . To facilitate its access and use , we developed tools in the TissueMiner framework to extract and convert this information initially stored in images into a specific database format ( see details in appendix 1 ) , which we call 'TM-DB' ( schematically outlined in Figure 6A ) . 10 . 7554/eLife . 14334 . 027Figure 6 . Construction of the relational database of TissueMiner . ( A ) Conceptual scheme of the database . Entities ( square boxes ) are related to other entities by associations ( rounded boxes ) . Each entity contains an identifier ( underlined ) that uniquely defines each record . The database can be implemented by converting entities into tables ( see appendix 1 and Figure 6—figure supplement 2 ) . ( B ) Cell lineage trees are stored in the database: upon division a mother cell identifier a gives rise to two new daughter cell identifiers b and c . {a , b , c , d , e , f , g} defines one lineage group . ( C ) A pixelated cell contour in the 2D cell network: green=bond pixels , red=vertex pixels , white=other cell network pixels . ( D ) Vectorized representation of the cell shown in ( C ) . To preserve the topology of the cell network , directed bonds ( cyan ) are defined from within a given cell alpha and ordered anticlockwisely along the cell contour . Each directed bond is complemented by a conjugated bond ( magenta ) and is linked to it next counter-clockwise follower ( dashed ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 02710 . 7554/eLife . 14334 . 028Figure 6—figure supplement 1 . Tracked cells identified by unique colors in TissueAnalyzer . ( A ) shows two consecutive frames depicting colored-tracked cells from a time-lapse movie processed with TissueAnalyzer . Each cell is assigned a color identifier that uniquely defines it in the course of the time-lapse . One pixel wide cell-cell interfaces are visible in white on the raster image . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 02810 . 7554/eLife . 14334 . 029Figure 6—figure supplement 2 . Logical scheme of the relational database . ( A ) The conceptual scheme shown in Figure 6A can be automatically converted to a logical scheme shown here by using softwares such as IntelliJDEA or MySQL workbench . The rules of conversion are briefly evoked in appendix 1 . The entities defined in the conceptual scheme are converted into tables containing one primary key ( upper part of the table ) that uniquely defines each record in the table , the properties of each record , and the foreign keys ( arrows ) . Foreign keys are properties of one table pointing to the primary key of a related table ( ex: conj_dbond_id:dbond_id means that the conj_dbond_id column is a foreign key whose values must be defined in the dbond_id column of the dbonds table ) . As a consequence of logical contraints by foreign keys , tables harbor more columns that one expected from looking at Figure 6A . This logical scheme now shows all tables and columns of the database . This scheme is implemented in physical SQLite tables can are indexed for the sake of performance ( see CreateDbFromParser . R on https://github . com/mpicbg-scicomp/tissue_miner ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 029 First , the history of each tracked cell in the movie is stored as a separate row in the cell_histories table of the TM-DB ( Figure 6A ) . This includes the movie frames in which it first appears and disappears and why , along with its lineage relationship to other cells ( see appendix 1 ) . The reason for cell appearance and disappearance is inferred by the parser . A primary reason could be a cell division , which results in the disappearance of the mother cell and in the appearance of two daughter cells . It could be a cell extrusion that results in its disappearance . It could also be that cells move in and out of the field of view of the microscope lens , resulting in gain and loss of cells . Furthermore , we use this information to establish the lineage relationship that corresponds to each group of cells related by ancestry ( Figure 6B ) . Each cell within the lineage group is assigned a generation number . The lineage group and generation number for each cell are listed in the cell_histories table . We store the time points at which the movie images were recorded into a frames table that links each movie frame to its corresponding time point . For each movie frame , we need to store geometrical and topological information about cells within the cellular network . Geometrical information includes position and shape descriptors , whereas topological information indicates the arrangement of neighboring cells around each cell . We use cell histories , geometry and topology to understand how individual cells contribute to the whole tissue deformation during morphogenesis ( Etournay et al . , 2015 ) . The geometrical information is stored in three tables of the TM-DB: cells , bonds and vertices . They correspond to the 3 generic entities - cells , cell-cell contacts and intersections between cell-cell contacts , respectively illustrated in Figure 6C . These entities are commonly used in vertex model simulations ( for review [Fletcher et al . , 2014] ) . The cells table contains cell geometrical data ( center of mass , area , shape anisotropy ) and the polarized distribution of proteins along the cell circumference , as represented by a polarity nematic tensor ( Aigouy et al . , 2010 ) . The bonds table informs about bond length , and the vertices table about vertex position in each movie frame . The directed_bonds table exclusively stores the cell neighbor topological information at each frame , i . e . how bonds are organized around each cell along with the cell neighbor relationship information . To store the cell neighbor topology in an unambiguous manner , we define for each cell a directed path of consecutive bond vectors oriented counterclockwise , which forms the oriented circumference of the cell ( Figure 6D , see also [Kachalo et al . , 2015] ) . We link each directed bond to its counter clockwise follower ( left directed bond ) in the same cell . To store the cell neighbor relationship , we link each directed bond to its corresponding directed bond ( conjugated bond ) of the neighboring cell ( Figure 6D , and appendix 1 ) . The TM-DB is relational , which means that it establishes contextual relationships between items stored in one ore more tables ( see appendix 1 ) . These relationships are outlined in rounded boxes in the conceptual scheme of the TM-DB ( Figure 6A ) . Technically , each item in a table is stored in a separate row and is given a unique number as identifier . For a relationship between two tables , one of the tables contains an additional column , which refers to items in the other table by holding their identifier number . Such additional columns for the TM-DB format are shown in blue in Figure 6—figure supplement 2 . When extracting information from a database using so-called queries , these columns serve as bridges connecting the information stored about related items . In essence , this structure creates a generic relational model to represent complex cell tracking data in 2D . In practice , the data for each movie is stored in a separate SQLite database file . Since all movie files are stored using the same database structure , automated data mining and visualization are greatly facilitated . For the same reason , usage of the TissueMiner database format encourages exchange of both movie data and analysis tools . To help the user to perform complex tissue morphogenesis analysis , we developed an automated pipeline that uses the tracked data from TissueAnalyzer as an input to build the database and perform all downstream analyses described above . To do so , we use the snakemake workflow engine developed by Koster and Rahmann ( Koster and Rahmann , 2012 ) . This engine channels the different processing steps into a well-formed workflow graph . Snakemake automatically determines the execution order , provides means for error recovery and job control , and supports High Performance Computing ( HPC ) environments . By using snakemake we enable the user to easily run and monitor TissueMiner , while maintaining a proper decoupling of tools as independent executables . Practically , the user defines a workflow definition file in which processing steps are defined as a set of execution rules , namely a list of scripts to be run along with required input ( s ) and expected output ( s ) . Snakemake automatically builds a directed graph from which the execution order of processing steps is inferred . If only one branch of the graph needs to be run , the engine will ensure that all input data are present and will automatically run upstream steps if necessary . This engine also provides the possibility to visualize a directed acyclic execution dependency and execution state graph ( DAG ) for a given workflow ( see Figure 7 ) . 10 . 7554/eLife . 14334 . 030Figure 7 . Automated workflow using snakemake . ( A ) The snakemake engine can generate a directed acyclic graph ( DAG ) where we show an example here . This graph represents both the execution dependency ( grey arrows ) and the execution state of the workflow ( solid or dashed line ) . Each box corresponds to an execution rule , namely a program to be run along with required input ( s ) and expected output ( s ) . This DAG can be generated at any time when running the workflow ( see documentation ) . Solid lines indicate the rules that have not been executed yet , whereas dashed lines depict completed jobs . The first rule to be executed is called 'prepare_movie': it prepares the tracked images from TissueAnalyzer to be converted by the parser into tables of values containing all necessary entities along with their properties ( 'parse_tables' rule ) . Then the 'make_db' rule is executed for building the database . Following the grey arrows can one navigate into the next steps of the workflow . The 'roi_tracking' rule filters out cells in contact to margin cells including user-defined regions of interest , and the 'roi_movie' rule allows us to visualize regions of interest over time . The 'deformation_movies' and 'db_elongation_movies' rules generate annotated movies to visualize the deformation of the tissue and the cell state properties ( area , elongation ) . The 'four_way' rule detects four-way vertices and performs basic statistics on vertices . The 'tri_create' rule performs the triangulation of the network for further shear calculation and visualization ( 'shear_calculate' and 'shear_movies' ) . It also enables triangle tracking and mapping to each type of cell event ( 'tri_categorize' ) . The 'topo_countT1' rule detects neighbor changes that are not due to division or extrusion , and categorizes them into gained or lost neighbors . The 'topo_movies' rule allows one to visualize the coarse-grained rates of division and neighbor changes on the tissue . The 'topo_unbalance' rule is a quality check to verify that the number of gained neighbors is similar to the number of lost neighbors . The 'polygon_class' rule performs the cell-neighbor number count . The 'lineage_colors' rule allows us to optimize the color of each lineage group such that adjacent lineage groups always have different colors . Finally , the 'lineage_movies' rule allows one to visualize lineage groups and cell generations on the tissue . The rule 'all' checks that all upstream jobs have been completed . DOI: http://dx . doi . org/10 . 7554/eLife . 14334 . 030 One major advantage of a workflow engine such as snakemake is that it can run the workflow on various architectures - from single-core workstations to multi-core servers and clusters - without the need to modify the rules , thereby facilitating reproducible research . To simplify the TissueMiner installation procedure , we provide a pre-configured system to be loaded in the docker software available at http://docker . com . The TissueMiner docker image can be run without any setup using provided example data or custom user data as detailed out on the TissueMiner GitHub project page . More advanced users can use TissueMiner directly from the command-line with or without snakemake and can thus perform simultaneous analyses of multiple movies . After applying our automated workflow to different movies , the results can be easily compared using a collection of command-line tools written in R and Python . These tools aggregate different experiments for plotting and performing comparative analysis . Here we describe the tools written in R . Python tools are described in the corresponding tutorial . The R tools are designed to be used in an integrated development environment such as RStudio , which provides a user-friendly environment to assist the user in writing and executing command lines . These command line tools are organized in the spirit of a grammar of data manipulation and they can be combined with the existing R tools like dplyr ( Francois and Wickham , 2015 ) or ggplot2 ( Wickham , 2009 ) for manipulating and visualizing data ( https://mpicbg-scicomp . github . io/tissue_miner/user_manual/Learning_the_R_basics_for_TissueMiner . html ) . We developed generic 'multi-query functions' ( mqf ) to collect specific information for individual movies . These mqf tools are organized into fine-grained and coarse-grained categories according to the type of analysis to be carried out . The fine-grained tools aggregate data at cellular level , namely individual cell properties inside regions of interest . These tools are prefixed with 'mqf_fg_' . The coarse-grained mqf tools are further separated into 'roi' and 'grid' categories to distinguish between regions moving with the tissue and static square regions tiled into a grid . They allow one to visualize and quantify average cell properties at different tissue locations and various spatial scales , and are prefixed with 'mqf_cg_roi_' and 'mqf_cg_grid_' respectively . To compare fine-grained and coarse-grained cell properties amongst movies we developed a 'multi-db-query' tool , which streamlines the application of the mqf tools to a set of movies . To use this tool , the user should first align the movies in time , using convenient morphological or cellular landmarks . As for the Drosophila wing , we align movies such that the peaks of cell elongation coincide in the different movies . The user can then apply a chosen mqf tool to multiple movies and multiple ROI’s . All mqf tools , alone or in combination with the 'multi-db-query' tool , generate a table that contains individual or averaged measurements to be visualized on the tissue ( Figure 1 A–E’ , Figure 2A–D’ , Figure 3A , D , Figure 4B , D , Figure 5D ) or in graphs ( Figure 2E–F , Figure 3B , Figure 4C , Figure 5E–F ) . This library of tools is described in detail in the TM R-User Manual , which also provides many examples . These tools can be easily extended to address project specific questions . To detect cell neighbor changes , we developed a routine in R that queries the DB and establishes the cell-neighbor relationship at each frame . By comparing the list of neighboring cell identifiers for a given cell between two consecutive frames [f , f + 1] , can one identify and count the changes in neighbor relationships . These can be subdivided into those caused by cell divisions , cell extrusions or the simple gain or loss of a cell contact ( not due to division or extrusion ) . We call these half-T1’s because they resemble the gain and loss of cell contacts that occurs during a T1 transition – although they may also be generated by other events such as rosette formation . To assign a neighbor change to the half-T1 category , the corresponding cell identifiers must be present in both frames , ruling out extrusions and cells moving in and out of the field of view . To detect half-T1’s that occur simultaneously with divisions , we mask neighbor changes due to divisions by propagating the mother cell identifier ( frame f ) to the two daughter cells ( frame f+1 ) that we fuse into one fake cell having the mother cell identifier . We iterate over each pair of consecutive frames and store the half-T1 events due to a gain and a loss of cell neighbors . We pool all lineage information ( as contained in the cell_id , left_daughter_cell_id and right_daughter_cell_id columns from the cell_histories table ) into a directed lineage graph ( Nepusz , 2006 ) from which we infer a lineage group identifier and a generation number . By definition the root of each lineage tree is considered as the F0 generation and is thus given a generation value of 0 . We follow ROI’s backward and forward in time by browsing lineage graphs that were selected based on the regions drawn by using the draw_n_get_ROIcoord . ijm Fiji macro . However cells may be lost or not detected when browsing the lineages . One primary reason is that extruding cells are not detected when browsing the lineage backward in time . Cells could also be lost due to possible tracking mistakes . To improve spatial continuity of ROI’s we have implemented a method to reassign lost cells to ROI’s when located within ROI’s . To identify lost cells for a frame within a given ROI , we first distinguish bonds that connect two cells within the ROI , only one cell within the ROI or none . All corresponding cell-pairs define an undirected graph on which a connected component analysis ( Nepusz , 2006 ) allows to identify the ROI and non-ROI regions . All cells of non-ROI regions , except for the largest one , are reassigned to become part of the ROI . By doing so , we make the assumption that the largest non-ROI component is defined by the tissue surrounding the ROI . When analyzing and visualizing single cell properties , we use the same cell elongation definition as in Aigouy et al . ( 2010 ) . For a given Cartesian xy coordinate system , the elongation of a given cell is defined by the nematic tensor ( ϵxxϵxyϵxy−ϵxx ) withϵxx=1Ac∫cos⁡ ( 2ϕ ) dAϵxy=1Ac∫sin⁡ ( 2ϕ ) dA . Here , Ac is the area of the given cell , and the integrals are carried out over all points r within the cell . The angle is the angle between the vector r − rc and the x axis , where rc is the cell center defined asrc=1Ac∫r dA . Here , the integral is again carried out over all points r within the cell . The magnitude of the elongation is given by ϵ= ( ϵxx2+ϵxy2 ) 12 and the elongation angle φ is given by the following two equationscos⁡ ( 2φ ) =ϵxxϵsin⁡ ( 2φ ) =ϵxyϵ . Note that this definition of cell elongation is different from the triangle-based definition that is also discussed in this article . However for the fruit fly wing , both cell elongation definitions yield very similar results . To characterize the axes of cell divisions and T1 transition , we introduce the unit nematic tensors n~CD , n~T1+ , and n~T1− . The orientation of a single cell division is quantified by the unit nematic n~CD defined by:n~CD= ( cos⁡ ( 2ϕCD ) sin⁡ ( 2ϕCD ) sin⁡ ( 2ϕCD ) −cos⁡ ( 2ϕCD ) ) . Here , the angle ϕ is the angle of the line connecting both cell centers with respect to the x axis , measured in counter-clockwise sense . The orientation for a half-T1 transition during which two cell lose neighborship is characterized by:n~T1+= ( cos⁡ ( 2ϕT1+ ) sin⁡ ( 2ϕT1+ ) sin⁡ ( 2ϕT1+ ) −cos⁡ ( 2ϕT1+ ) ) , where ϕT1+ is the angle of the line connecting the centers of the cells losing neighborship . The orientation for a half-T1 transition during which two cell gain neighborship is characterized by:n~T1−=− ( cos⁡ ( 2ϕT1− ) sin⁡ ( 2ϕT1− ) sin⁡ ( 2ϕT1− ) −cos⁡ ( 2ϕT1− ) ) , where ϕT1− is the angle of the line connecting the centers of the cells that gain neighborship . The axes of the nematics n~CD , n~T1+ and n~T1− roughly correspond to the axis along which the tissue extends due to the respective cell division or half-T1 transition . In particular , because of the minus sign in the definition of n~T1− , when the same two cells gain neighborship and lose it again along the same axis , the total effect adding n~T1+ and n~T1− is zero . Here we discuss the formal definitions used to characterize tissue deformation , area change , and shear . We characterize the local rate of tissue deformation by the gradient of the velocity field v ( r ) . We then define the overall deformation rate V of a given piece of tissue by the integral over the area At of this piece:V=1At∫ ( ∂vx∂x∂vy∂x∂vx∂y∂vy∂y ) dA . This 2x2 tensor can be decomposed into an isotropic part Viso characterizing the relative growth rate of tissue area , a symmetric , traceless part V~ characterizing the anisotropic part of the deformation ( pure shear rate ) , and an antisymmetric part Ω characterizing overall tissue rotation:V=VisoI2+V~+Ωe . Here , we have defined Viso=1At∫ ( ∂vx∂x+∂vy∂y ) dA , Ω=12At∫ ( ∂vx∂y−∂vy∂x ) dA , I= ( 1001 ) , V~=12At∫ ( ∂vx∂x−∂vy∂y∂vy∂x+∂vx∂y∂vy∂x+∂vx∂y∂vy∂y−∂vx∂x ) dA , ande= ( 0−110 ) . In recent work , we have shown that the overall shear rate V~ can be exactly decomposed into a sum of cellular contributions using our Triangle Method ( Merkel , 2014; Merkel et al . , 2016 ) :V~=DQ~Dt+T+C+E+D . Here , the nematic tensors Q~ is the average cell elongation defined based on triangles , and the nematic tensors T , C , E , and D are the shear contributions by T1 transitions , cell divisions , cell extrusions , and correlation effects , respectively . The corotational time derivative DQ~/Dt is defined byDQ~Dt=dQ~dt−2 ( cΩ+[1−c]dΦdt ) e⋅Q~ . The operator d/dt denotes the total derivative , c=tanh⁡ ( 2Q ) / ( 2Q ) , and the dot denotes the tensor dot product . The quantities Q and Φ denote magnitude and angle of the average cell elongation tensor Q~ . These formal definitions for Q~ , DQ~/Dt , T , C , E , and D refer to deformation rates in the limit of infinitesimal deformations . However , subsequent frames of any real tissue movie are separated by finite time intervals , i . e . finite deformations . There are different ways to adapt these definitions to finite deformations ( Etournay et al . , 2015; Merkel et al . , 2016 ) . The current implementation of TissueMiner uses the finite-deformation definitions presented in detail in Etournay et al . ( 2015 ) .	Cells interact , divide , rearrange and change shape to build an organ during development . Modern microscopy and computer technology can follow each individual cell of an entire organ in a living organism . However , to understand how the complex choreography of cells leads to well-shaped organs , researchers need tools to help the store and analyze the large amounts of data generated . Tools are also needed to visualize and quantify the complex cell behaviors in an easy and flexible manner . During its development , a fruit fly’s wings become divided into distinct regions separated by tubular supports called veins . Early on in development , the vein cells are indistinguishable from their neighbors , but at late stages , vein cells become a different shape . Veins also become narrower , which is assumed to be due to the number of vein cells falling . However , the way in which cells behave to bring about these changes has not been studied in detail . Etournay , Merkel , Popović , Brandl et al . have now developed a toolkit called TissueMiner that enables users to store large amounts of data about cells and analyze how cells collectively shape an organ . TissueMiner was then used to identify vein cells at late stages of wing development and follow them backward in time to reveal their position at early stages . This showed that veins become narrower and more elongated because the cells that make up the veins shrink more than cells in other regions . TissueMiner was then used to show that vein cells specifically rearrange and elongate to produce thinner regions , while the number of cells increases slightly because the cells divide . These results suggest that the cell behaviors responsible for making veins elongate and narrow are likely to be different from what had previously been assumed . TissueMiner can be used in future studies to help understand the molecule signals that influence how cells behave in veins during wing development . The toolkit could also now be used to explore the changes involved in the development of other organs in other organisms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "cell", "biology", "tools", "and", "resources" ]	2016	TissueMiner: A multiscale analysis toolkit to quantify how cellular processes create tissue dynamics
Phenotypic screens allow the identification of small molecules with promising anticancer activity , but the difficulty in characterizing the mechanism of action of these compounds in human cells often undermines their value as drug leads . Here , we used a loss-of-function genetic screen in human haploid KBM7 cells to discover the mechanism of action of the anticancer natural product ophiobolin A ( OPA ) . We found that genetic inactivation of de novo synthesis of phosphatidylethanolamine ( PE ) mitigates OPA cytotoxicity by reducing cellular PE levels . OPA reacts with the ethanolamine head group of PE in human cells to form pyrrole-containing covalent cytotoxic adducts and these adducts lead to lipid bilayer destabilization . Our characterization of this unusual cytotoxicity mechanism , made possible by unbiased genetic screening in human cells , suggests that the selective antitumor activity displayed by OPA may be due to altered membrane PE levels in cancer cells . Natural products are an important source for the development of pharmaceutical drugs , especially in oncology; half of all anticancer drugs developed since the 1940s are natural products or derivatives of natural products ( Newman and Cragg , 2012 ) . Compounds with anticancer activity can be readily identified in cytotoxicity assays and other phenotypic screens ( Harvey et al . , 2015; Eggert , 2013 ) . To use these small molecules as anticancer drug leads or to identify new chemotherapy molecular targets , it is essential to characterize the molecular mechanism of action ( MOA ) that underlies cytotoxicity ( Schenone et al . , 2013; Bunnage et al . , 2015 ) . However , as unraveling the MOA of bioactive small molecules remains challenging and time consuming ( Schenone et al . , 2013; Ziegler et al . , 2013 ) , the MOA of many natural products that display promising anticancer activities in phenotypic screens remains uncharacterized ( Shoemaker , 2006 ) . An example of such a natural product is ophiobolin A ( OPA ) , a plant toxin isolated from pathogenic fungi of the Bipolaris genus which displays cytotoxicity at nanomolar concentrations against a range of cancer cell lines ( Au et al . , 2000; Bury et al . , 2013 ) . OPA induces paraptosis , a form of non-apoptotic cell death , in glioblastoma cells and displays antitumor activity in a mouse glioblastoma model ( Bury et al . , 2013; Dasari et al . , 2015 ) . The toxicity of OPA to plants is believed to involve calmodulin inhibition via formation of a covalent adduct between OPA and specific lysine side chains ( Leung et al . , 1984 ) . More recently , it has been shown in synthetic studies that primary amines react with the 1 , 4-dicarbonyl moiety of OPA to form covalent adducts and that this moiety is critical for animal cell cytotoxicity , leading the authors to suggest that the MOA of OPA in animal cells is through covalent modification of an unknown intracellular target protein ( Dasari et al . , 2015 ) . In conclusion , OPA represents an interesting candidate for the treatment of glioblastomas that are resistant to classical pro-apoptotic therapeutic approaches , but the lack of information on cellular targets of OPA impedes any further development . Genetic screens represent an unbiased genome-wide approach to identify molecular targets involved in small molecule MOA but have been mainly limited to application in genetically tractable organisms such as Saccharomyces cerevisiae ( Roemer et al . , 2012; Lee et al . , 2014 ) . Recent technical breakthroughs , such as insertional mutagenesis in haploid cells ( Carette et al . , 2009 , 2011 ) and CRISPR-Cas9 genome editing ( Wang et al . , 2014; Shalem et al . , 2014; Gilbert et al . , 2014; Smurnyy et al . , 2014 ) have revolutionized the use of genetic screens in human cell lines to facilitate the study of the MOA of bioactive molecules in model systems more relevant to human disease ( Nijman , 2015 ) . To unravel the MOA of OPA , we took advantage of a genome-wide strategy in human cells to identify genes that are required for OPA to exert its cytotoxic effect . We used insertional mutagenesis in the near-haploid human cell line KBM7 to generate loss-of-function mutants and then selected for growth of cell lines resistant to OPA treatment ( Carette et al . , 2011 ) . We discovered that inactivation of the pathway for de novo synthesis of phosphatidylethanolamine ( PE ) , also named the Kennedy pathway , confers resistance to OPA . Increased OPA resistance was correlated with decreased cellular PE levels . Surprisingly , we determined that the molecular target of OPA is PE itself; OPA forms a covalent adduct with PE in human cells . This work illustrates the power of unbiased genetic screens in human cells in discovering novel MOAs of compounds identified in phenotypic screens . Loss-of-function genetic screens in human KBM7 cells have been used to identify cellular factors that are necessary for entry of viruses and bacterial toxins , and transporters of small molecules , but rarely to identify the molecular targets of small molecule drugs ( Carette et al . , 2011; Nijman , 2015; Winter et al . , 2014; Reiling et al . , 2011; Birsoy et al . , 2013 ) . Before applying the screen to studies of compounds of unknown MOA displaying anticancer activity , we first determined whether such screens can robustly identify genes involved in the cytotoxicity of anticancer drugs with well-characterized MOAs . Briefly , we used a retroviral gene-trap approach to generate approximately 75 million insertions in the near-haploid human cell line KBM7 , covering more than 95% of all expressed genes ( Carette et al . , 2011 ) . This library of loss-of-function cell lines was treated with a toxic dose of anticancer drug and resistant mutants were allowed to grow over 3 weeks . Retroviral insertion sites were identified by amplification of the genomic sequence flanking the insertion site , high-throughput sequencing , and mapping to the human genome . Insertions in exonic regions or in the sense orientation of intronic regions are typically expected to cause gene inactivation ( Carette et al . , 2009 ) . For each gene locus , we calculated an enrichment p-value by comparing the number of inactivating insertions in the pooled drug-resistant cells to the number of such insertions in mutagenized cells before selection . This enrichment p-value allows the identification of genes whose inactivation renders cells resistant to the toxic effects of the small molecule tested ( Figure 1—figure supplement 1 ) . We performed screens with anticancer drugs including topoisomerase inhibitors ( topotecan , etoposide , and doxorubicin ) , a proteasome inhibitor ( bortezomib ) , an antimetabolite ( gemcitabine ) and a platinum-based DNA crosslinking agent ( oxaliplatin ) . As expected , we observed enrichment of inactivating insertions in genes known to play a role in the MOA of these anticancer compounds ( Figure 1—figure supplement 2 ) . For instance , doxorubicin and etoposide induce cytotoxicity by forming a ternary complex with DNA and the enzyme topoisomerase IIA ( Pommier et al . , 2010 ) . In both screens , we detected a significant enrichment of inactivating insertions in gene TOP2A , which encodes topoisomerase IIA ( Figure 1—figure supplement 2a–b ) . Bortezomib kills cancer cells by proteasome inhibition ( Adams , 2004 ) and , accordingly , we observed a significant enrichment of inactivating insertions in genes encoding proteasome subunits ( Figure 1—figure supplement 2d ) . In addition to targets of anticancer drugs , the screens also identified transporters and genes known to metabolize the drug tested ( Figure 1—figure supplement 2 ) . Thus , loss-of-function screens in KBM7 cells are a powerful way to initiate MOA studies of anticancer compounds . We then used the KBM7 screening platform to investigate the mechanism of cytotoxicity of OPA , isolating human cells containing insertions that rendered cells resistant to OPA treatment . Three genes had a significant enrichment of retroviral insertions: ethanolamine kinase 1 ( ETNK1 , p = 7 . 2 × 10−12 ) , phosphate cytidylyltransferase 2 , ethanolamine ( PCYT2 , p = 4 . 0 × 10−7 ) , and ethanolaminephosphotransferase 1 ( EPT1 , p = 4 . 0 × 10−7 ) ( Figure 1a ) . These three genes encode the three enzymes required for the de novo synthesis of PE , also known as the Kennedy pathway ( Gibellini and Smith , 2010 ) ( Figure 1b ) . To test the robustness of this result , we repeated screens at different concentrations of OPA; at least one gene in the Kennedy pathway was enriched above background at every concentration tested ( Figure 1—figure supplement 3 ) . 10 . 7554/eLife . 14601 . 003Figure 1 . Identification of a genetic interaction between ophiobolin A ( OPA ) and the Kennedy pathway using a loss-of-function genetic screen in the near-haploid human cell line KBM7 . ( a ) A collection of loss-of-function mutants generated in KBM7 cells using retroviral insertional mutagenesis was treated with 388 nM OPA . Resistant clones were allowed to expand for 3 weeks and retroviral insertion sites were identified by high-throughput sequencing . For each gene , an enrichment factor ( p-value ) was calculated to quantify the enrichment of inactivating insertions in the pool of resistant clones compared to the number existing before selection . Each bubble represents a gene and the diameter of the bubble is proportional to the number of unique insertion sites in the pool of resistant clones ( for ETNK1 , N = 11 ) . Genes are ordered on the x axis by chromosomal location ( Figure 1—source data 1 ) . ( b ) The Kennedy pathway: de novo synthesis of phosphatidylethanolamine . ( c–e ) Characterization of KBM7 clonal cell lines resistant to OPA treatment with inactivating mutations in either PCYT2 or ETNK1 , referred to as PCYT2GT and ETNK1GT . ( c ) Quantification of relative PCYT2 and ETNK1 mRNA levels by RT-qPCR , normalized to levels in wild-type KBM7 ( WT ) . ( d ) Cell viability measurement after 72 hr of treatment with OPA ( or DMSO vehicle ) using a luciferase-based assay quantifying ATP content . The viability of each vehicle-treated cell line was normalized to 1 . ( e ) Determination of cellular phosphatidylethanolamine ( PE ) content by total lipid extraction , separation of phospholipids by thin layer chromatography and quantification of phospholipids by phosphorus content analysis . PE content is displayed as a percentage of total phospholipids . ( f ) Expression of PCYT2 in PCYT2GT cells restores OPA sensitivity . Constructs expressing either PCYT2 or GFP ( control ) were delivered to WT or PCYT2GT cells by lentiviral transduction . The viability of each cell line was assayed using a luciferase-based assay quantifying ATP content after 72 hr of treatment with OPA ( or DMSO vehicle ) . '—' denotes non-transduced cell lines . ( c–f ) Results were obtained from three independent experiments ( c and e ) or from assays performed in triplicate ( d and f ) and data represent mean values ± standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 00310 . 7554/eLife . 14601 . 004Figure 1—source data 1 . Source data for the ophiobolin A ( OPA ) loss-of-function KBM7 screen . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 00410 . 7554/eLife . 14601 . 005Figure 1—figure supplement 1 . Illustration of loss-of-function genetic screens in haploid human KBM7 cells ( Carette et al . , 2009 , 2011 ) . Near-haploid KBM7 cells are infected with a gene-trap retrovirus to generate random insertions . Insertions in exonic regions or in the sense orientation in intronic regions typically cause loss-of-function mutations . Mutagenized KBM7 cells are treated with a toxic dose of a molecule of interest and resistant clones are allowed to expand into colonies . Retroviral insertion sites in the pooled resistant population are identified by amplification of flanking genomic DNA by inverse PCR , high-throughput sequencing , and mapping to the human genome . For each gene locus , an enrichment p-value is calculated using Fisher's exact test by comparing the number of unique inactivating insertions in the pool of resistant cells to the number of unique inactivating insertions in mutagenized cells before selection . The location of retroviral insertion sites was analyzed in 5 million mutagenized cells before selection ( representing 5% of the number of cells used per screen ) and 142 , 800 unique insertions were identified , 38 , 628 of which were gene inactivating mutations ( Materials and methods ) . This collection of 38 , 628 unique inactivating insertions was used in the calculation of enrichment p-values ( Figure 1—figure supplement 1—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 00510 . 7554/eLife . 14601 . 006Figure 1—figure supplement 1—source data 1 . Source data for the characterization of mutagenized KBM7 cells before selection ( control library ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 00610 . 7554/eLife . 14601 . 007Figure 1—figure supplement 2 . Validation of KBM7 loss-of-function screens using anticancer drugs with well-characterized mechanisms of action . ( a–f ) Loss-of-function genetic screens were performed and displayed as described in Figure 1a . Cells were treated with anticancer drugs at the concentration specified in each panel . Green bubbles represent genes known to be involved in the mechanism of action of the drug tested and the diameter of the bubble is proportional to the number of unique insertion sites in the pooled resistant clones ( doxorubicin , TOP2A: N = 4; etoposide , TOP2A: N = 5; topotecan , TOP1: N = 10; bortezomib , PSMC6: N = 8; oxaliplatin , SLC43A2: N = 5; gemcitabine , DCK: N = 161 ) . Genes with a total of less than 50 sequencing reads were not displayed in the bubble plots . ( a ) The screen with doxorubicin uncovered TOP2A , which encodes topoisomerase IIA , the direct target of doxorubicin ( Tewey et al . , 1984 ) . ( b ) The screen with etoposide uncovered TOP2A , consistent with its known mechanism of action ( Montecucco and Biamonti , 2007 ) . ( c ) The screen with topotecan identified its known direct target ( Pommier , 2006 ) , type I topoisomerase ( TOP1 ) , as well as ABCG2 . It is known that overexpression , not inactivation , of this transporter causes resistance to topotecan ( Mao and Unadkat , 2015 ) . In the pooled resistant clones , retroviral integration sites clustered to the beginning of the ABCG2 transcript and most were actually found upstream of the transcription start site ( TSS ) . A topotecan-resistant clone ( ABCG2GT ) with a retroviral site 400 bp upstream of the TSS of ABCG2 was isolated and the expression level of ABCG2 was determined by RT-qPCR ( Materials and methods ) . ABCG2GT displayed a △△CT of 6 . 2 ± 0 . 3 compared to KBM7 indicating an increase in ABCG2 mRNA levels of up to 70-fold compared to KBM7 ( mean values of three independent experiments ± standard deviation ) . ( d ) The screen with bortezomib , a proteasome inhibitor ( Adams , 2004 ) , identified proteasome subunits ( PSMC6 , PSMD12 , PSMC5 , PSMD7 , PSMD2 ) . ( e ) The screen with oxaliplatin , a DNA crosslinker , identified SLC43A2 , a solute carrier that transports neutral amino acids into cells ( Bodoy et al . , 2013 ) , suggesting that SLC43A2 facilitates the transport of oxaliplatin into cells consistent with the known role of solute carriers in the transport of anticancer drugs ( Li and Shu , 2014 ) . ( f ) The screen with gemcitabine , a deoxycytidine analog , identified DCK and CDADC1 . Deficiency of DCK is associated with gemcitabine resistance ( Bergman et al . , 2002 ) and CDADC1 is a cytidine and dCMP deaminase known to be involved in resistance to deoxycytidine analogs ( Cai et al . , 2008 ) ( Figure 1—figure supplement 2—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 00710 . 7554/eLife . 14601 . 008Figure 1—figure supplement 2—source data 1 . Source data for anticancer drug loss-of-function KBM7 screens . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 00810 . 7554/eLife . 14601 . 009Figure 1—figure supplement 3 . Loss-of-function genetic screens in KBM7 cells performed at three different concentrations of ophiobolin A ( OPA ) consistently identify genes in the Kennedy pathway . ( a–c ) Screens were performed and analyzed as in Figure 1a , at the indicated concentrations of OPA . Each bubble represents a gene and the diameter of the bubble is proportional to the number of unique insertion sites in the pool of resistant clones ( for OPA 358 nM , ETNK1: N=9; for OPA 365 nM , ETNK1: N=10; for OPA 418 nM , ETNK1: N=6 ) ( Figure 1—figure supplement 3—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 00910 . 7554/eLife . 14601 . 010Figure 1—figure supplement 3—source data 1 . Source data for additional ophiobolin A ( OPA ) loss-of-function KBM7 screens . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 01010 . 7554/eLife . 14601 . 011Figure 1—figure supplement 4 . Titration of the toxicity of ophiobolin A ( OPA ) towards KBM7 wild-type and gene-trapped cell lines , and additional data for complementation assays . ( a ) Viability of wild-type KBM7 ( WT ) , PCYT2GT , and ETNK1GT cell lines after treatment with OPA . Cells grown in standard conditions were treated with OPA ( or DMSO vehicle ) at the indicated concentration for 72 hr . Cell viability was quantified using a commercial luciferase-based assay measuring ATP content and the viability of each vehicle-treated cell line was normalized to 1 . ( b , c ) Additional characterization of complemented PCYT2GT cells . Constructs expressing either PCYT2 or GFP ( control ) were delivered to WT or PCYT2GT cells by lentiviral transduction . '—' denotes non-transduced cell lines . ( b ) Quantification of relative PCYT2 mRNA levels by RT-qPCR in cell lines expressing the indicated cDNA , normalized to the level of PCYT2 mRNA in WT . ( c ) Determination of cellular phosphatidylethanolamine ( PE ) content in complemented cell lines by total lipid extraction , separation of phospholipids by thin layer chromatography , and quantification of phospholipids by phosphorus content analysis . The PE content of WT was normalized to 100% . ( a–c ) Results were obtained from assays performed in triplicate and data represent mean values ± standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 011 We isolated and characterized KBM7 clonal cell lines carrying inactivating insertions in the genes PCYT2 ( in first intron ) and ETNK1 ( in first exon ) ( named PCYT2GT and ETNK1GT , respectively ) . As expected , these cell lines had strongly reduced levels ( <2% ) of either PCYT2 or ETNK1 mRNA , as quantified by RT-qPCR ( Figure 1c ) and , consistent with the screening results , both clones were less sensitive to OPA ( Figure 1d and Figure 1—figure supplement 4a ) . Since human cells can synthesize PE through multiple mechanisms ( the two major ones being the Kennedy pathway and decarboxylation of phosphatidylserine in mitochondria; Gibellini and Smith , 2010 ) , and it is known that cells employ mechanisms to maintain homeostasis of phospholipid levels ( Hermansson et al . , 2011 ) , we tested whether inactivation of the Kennedy pathway in KBM7 cells decreases total cellular PE levels . We extracted total lipids , separated phospholipids by thin layer chromatography , and quantified phospholipids by phosphorus content analysis . Both PCYT2GT and ETNK1GT cell lines showed reduced PE levels compared to KBM7 cells , by 24% and 16% , respectively ( Figure 1e ) . Since gene-trapping of PCYT2 renders KBM7 cells slightly more resistant to OPA and reduces PE levels to a larger extent than inactivation of ETNK1 , our further studies focused on PCYT2 . To validate that inactivation of the Kennedy pathway causes OPA resistance , we transduced PCYT2GT cells with a lentiviral construct driving expression of PCYT2 and observed that this complementation rescues OPA sensitivity ( Figure 1f ) . As expected , both PCYT2 mRNA levels and total PE content were also restored to wild-type levels in complemented cells ( Figure 1—figure supplement 4b–c ) . We next assessed the generality of our results across cell lines by testing the effect of silencing PCYT2 in HEK293T cells on viability during OPA treatment . Cells expressing two shRNA ( short hairpin RNA ) constructs had 85% and 70% reduced PCYT2 mRNA levels , respectively ( Figure 2a ) , exhibited increased OPA resistance ( Figure 2b ) , and had reduced cellular PE levels ( Figure 2c ) . In summary , these data show that a reduction of the Kennedy pathway activity , and thus PE levels , in human cells leads to an increase in OPA resistance . 10 . 7554/eLife . 14601 . 012Figure 2 . Interaction between the activity of the Kennedy pathway and ophiobolin A ( OPA ) cytotoxicity . ( a–c ) shRNA knockdown of the Kennedy pathway in HEK293T cells leads to increased resistance to OPA toxicity . Constructs enabling stable expression of shRNAs against PCYT2 ( kd1 and kd2 ) , a scrambled shRNA ( Scr . ) , or an empty vector control ( Control ) were delivered to HEK293T cells by lentiviral transduction . Results were obtained from assays performed on three independent transduced cell lines and data represent mean values ± standard deviation . ( a ) Quantification of relative PCYT2 mRNA levels by RT-qPCR , normalized to the level of PCYT2 mRNA in the control . ( b ) Cell viability measurement after 72 hr of treatment with OPA ( or DMSO vehicle ) using a luciferase-based assay quantifying ATP content . The viability of each vehicle-treated cell line was normalized to 1 . ( c ) Determination of cellular phosphatidylethanolamine ( PE ) levels by total lipid extraction , separation of phospholipids by thin layer chromatography ( TLC ) , and quantification of phospholipids by phosphorus content analysis . PE content is displayed as a percentage of total phospholipids . ( d ) OPA treatment activates the Kennedy pathway and increases PE content in HEK293T and HCT116 cells . Cells were treated with OPA , meclizine , or DMSO vehicle for 5 hr , then ethanolamine [1 , 2-14C] was added and the treatment was prolonged for an additional 24 hr . Total phospholipids were extracted and separated by silica gel TLC . PE contents were quantified as in ( c ) and the PE content of vehicle-treated cells was normalized to 100% . 14C-PE levels were quantified by liquid scintillation counting of silica scrapings and were normalized to total phospholipid content . Results were obtained from three independent experiments and data represent mean values ± standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 012 Having established a clear link between the activity of the Kennedy pathway and the cytotoxicity of OPA , we next investigated the underlying molecular mechanism . There is precedent for small molecules that target the Kennedy pathway: the antihistamine and antiemetic drug meclizine directly inhibits PCYT2 , reducing the average flux through the Kennedy pathway ( Gohil et al . , 2013 ) . To explore the possibility that OPA exerts cytotoxicity by inhibiting or activating one of the enzymes in the Kennedy pathway , we measured the activity of the pathway after OPA treatment in the two commonly used human cell lines , HEK293T and HCT116 . Cells were treated with OPA ( or vehicle ) and then incubated with ethanolamine [1 , 2-14C] , the substrate of the first enzyme in the Kennedy pathway . Interestingly , when either cell line was treated with concentrations of OPA that induced mild cytotoxicity , both the average flux through the Kennedy pathway ( measured by accumulation of the radiolabel into PE ) and the steady state level of PE were increased compared to vehicle-treated cells ( Figure 2d ) . In contrast , when we treated cells with meclizine we observed the expected decrease in the flux through the pathway and reduced PE content ( Figure 2d ) . Since the change in enzyme activity induced by OPA treatment is small , we do not believe that OPA acts directly on an enzyme in the Kennedy pathway to cause cytotoxicity . The known reactivity of OPA with primary amines ( Au et al . , 2000 ) and the observation that resistance to OPA toxicity correlates with lower PE content ( Figures 1e and 2c ) led us to propose that OPA might directly target PE through covalent modification of its ethanolamine head group . If this hypothesis is correct , adding exogenous PE along with OPA to the growth medium of cells should result in the formation of PE-OPA covalent adducts in the medium . The majority of these adducts should not partition efficiently into cells due to limited aqueous solubility , and thereby adding exogenous PE should reduce the number of OPA molecules available to react with endogenous PE and kill cells . In agreement with this prediction , addition of a commercial preparation of PE to cells treated with a cytotoxic amount of OPA rescued cell viability , whereas addition of the phospholipids phosphatidylcholine ( PC ) or phosphatidylserine ( PS ) had no effect ( Figure 3a ) . PE extracts from diverse sources were all similarly able to rescue cellular viability , suggesting that small quantities of impurities in these preparations are not likely responsible for OPA inactivation ( Figure 3—figure supplement 1a ) . To provide additional evidence that the difference in the effects of PE and PC on OPA toxicity derive solely from differences in the head group ( and not differences in fatty acid composition or impurities ) , we assayed a pair of commercial lipid preparations: one that consists of an extract of PC , and an identical PC extract in which the choline head group of PC has been exchanged for ethanolamine to yield a transphosphatidylated PE extract . When tested in the OPA inactivation assays , transphosphatidylated PE rescued cell viability whereas PC did not , substantiating the claim that PE is the molecule responsible for OPA inactivation ( Figure 3—figure supplement 1a ) . In addition , we observed that constituents of the head group of PE , ethanolamine and O-phosphorylethanolamine , could inactivate OPA , although much less potently than PE , and that triethanolamine ( lacking a primary amine ) had no effect ( Figure 3b and Figure 3—figure supplement 1b ) . 10 . 7554/eLife . 14601 . 013Figure 3 . Ophiobolin A ( OPA ) reacts with the ethanolamine ( Etn ) head group of phosphatidylethanolamine ( PE ) via a Paal-Knorr reaction . ( a ) Exogenous PE , but not phosphatidylcholine ( PC ) or phosphatidylserine ( PS ) , added to growth medium quenches the cytotoxicity of OPA . Commercially available phospholipids extracted from chicken egg ( PE and PC ) or bovine brain ( PS ) ( or vehicle ) were added to 20 ng/μL ( ~25 μM ) to the growth medium of HEK293T cells . Cells were subsequently treated with OPA ( or DMSO vehicle ) for 72 hr and cell viability was then quantified using a luciferase-based assay measuring ATP content . The viability of vehicle-treated cells in the absence of OPA was normalized to 1 . ( b ) The primary amine of Etn is essential for OPA inactivation . A cell viability assay was performed as in ( a ) with Etn , O-phosphorylethanolamine ( Phospho-Etn ) and triethanolamine ( Trietn ) . Only viabilities in 300 nM OPA are displayed and the viability of vehicle-treated cells in 300 nM OPA was normalized to 100% . Full plots are available in Figure 3—figure supplement 1b . ( c ) OPA was incubated with an excess of Etn ( or ethanol control ) in aqueous buffer . HEK293T cells grown in standard conditions were treated with the reaction product for 72 hr and cell viability was then quantified using a luciferase-based assay measuring ATP content . ( a–c ) Results were obtained from assays performed in triplicate and data represent mean values ± standard deviation . ( d ) Liquid chromatography-mass spectrometry ( LC-MS ) analysis in positive ion mode of the in vitro reaction of OPA ( exact mass = 400 . 2614 ) with Etn ( exact mass = 61 . 0528 ) shows formation of a single product at an m/z corresponding to an addition reaction minus two molecules of H2O . Both OPA and Etn were used as reactant controls and the total ion chromatograms of the three samples are displayed overlaid . The m/z of the most abundant ion is displayed above corresponding peaks . ( e ) OPA reacts with Etn to form a pyrrole-containing product detectable using Ehrlich's reagent . An in vitro reaction of OPA with Etn was mixed with Ehrlich's reagent and the absorbance of the resulting solution was measured between 450 and 700 nm . Reactions with DMSO ( vehicle ) instead of OPA and Trietn instead of Etn were used as negative controls . ( f ) Proposed reaction between PE and OPA . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 01310 . 7554/eLife . 14601 . 014Figure 3—figure supplement 1 . Additional data for ophiobolin A ( OPA ) inactivation assays with exogenously added small molecules . ( a ) Commercially available phospholipids ( or vehicle ) were added at 20 or 40 ng/μL ( ~25 or 50 μM ) to the growth medium of HEK293T cells . Cells were subsequently treated with OPA ( or DMSO vehicle ) for 72 hr and cell viability was then quantified using a luciferase-based assay measuring ATP content . The viability of vehicle-treated cells in the absence of OPA was normalized to 1 . For clarity , only cell viability data in the presence of OPA are displayed . ( b ) Full plots of the data displayed in Figure 3b . The viability of vehicle-treated cells in the absence of OPA was normalized to 1 . 0 . ( c ) Assay performed as in ( a ) using synthetic DOPC ( dioleoyl-PC ) , DOPE ( dioleoyl-PE ) , and DOPS ( dioleoyl-PS ) . ( d ) Assay performed as in Figure 3b , but using ethanolamine and serine . ( e ) OPA inactivation assays with 1 , 4-dicarbonyl scavengers , lysine , and ethanolamine . Assays were performed and displayed in a similar way as in ( b ) except that the exogenous molecules were preincubated with OPA in phosphate-buffered saline at 5× their final concentration for 2 . 5 hr before adding to the growth medium of HEK293T cells . ( a–e ) Results were obtained from assays performed in triplicate and data represent mean values ± standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 01410 . 7554/eLife . 14601 . 015Figure 3—figure supplement 2 . Proposed mechanism of covalent modification of phosphatidylethanolamine ( PE ) by ophiobolin A ( OPA ) through a Paal-Knorr-like reaction pathway ( Bernoud-Hubac et al . , 2004; Amarnath et al . , 1991 , 1995 ) . Addition of the amino group of PE onto the aldehyde moiety of OPA leads to formation of a hemiaminal . Based on the mechanism of the Paal-Knorr reaction and the specific reactivity of OPA , the reaction between PE and OPA is believed to proceed through two potential mechanisms: ( i ) elimination of a water molecule leading to formation of an imine and subsequent rearrangement to an enamine . Cyclization of the enamine followed by dehydration yields the final pyrrole-containing adduct . ( ii ) Cyclization of the hemiaminal leads to formation of a dihydroxypyrrolidine . A double elimination of water and double bond shift leads to the final pyrrole-containing adduct . In studies involving the formation of pyrrole adducts from the reaction of 4-oxohexanal and a primary amine , the reaction is known to proceed through the mechanism involving cyclization of the hemiaminal ( Amarnath et al . , 1995 ) . Under certain conditions , pyrrole adducts are known to be oxidized and the two main expected products are the lactam and hydroxylactam derivatives ( Sullivan et al . , 2010 ) . In addition , OPA contains an unsaturated aldehyde moiety which is known to readily react with thiols in a Michael addition reaction ( Grimsrud et al . , 2008 ) . 4-Oxo-2-nonenal , a product of cellular lipid oxidation which has similar reactive functionalities as OPA , has been shown to readily react with N-acetylcysteine ( Amarnath and Amarnath , 2015 ) . Interestingly , initial Michael adduct formation of OPA with a thiol would not prevent a subsequent Paal-Knorr reaction with PE and furthermore these two sequential reactions could lead to protein-lipid crosslinking , a potential cause of cytotoxicity . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 015 The serine head group of PS also contains a primary amine that could react with OPA . However , we observed no inactivation of OPA using either a natural extract of PS ( Figure 3a ) or synthetic dioleoyl-PS ( Figure 3—figure supplement 1c ) . In contrast to ethanolamine , adding serine to the growth medium of cells does not lead to OPA inactivation , suggesting that the primary amine in the serine head group is less reactive with OPA than that in the ethanolamine head group ( Figure 3—figure supplement 1d ) . This observation is consistent with previous reports of inefficient adduct formation between PS and reactive aldehydes from fatty acid peroxidation ( Guichardant et al . , 1998 , 2002 ) . In conclusion , these results suggest that OPA is specifically inactivated by PE via reaction with the primary amine on its head group . Synthetic studies exploring the chemical reactivity of OPA have shown that OPA can react with primary amines in a Paal-Knorr reaction to yield pyrrole-containing adducts ( Dasari et al . , 2015 ) . To test the hypothesis that OPA also reacts with PE through its primary amine , we explored the product of the reaction between OPA and ethanolamine . Incubation of OPA with an excess of ethanolamine abolished cytotoxicity and analysis by LC-MS/MS revealed that the main product of this reaction is consistent with formation of a covalent adduct with two dehydration reactions ( Figure 3c–d ) . Incubation of the reaction product of OPA and ethanolamine with Ehrlich's reagent ( Amarnath et al . , 2004 ) yielded a purple solution ( Amax = 580 nm ) , characteristic of pyrroles ( Figure 3e ) . Addition of salicylamine and pentyl-pyridoxamine , potent scavengers of 1 , 4-dicarbonyls , to OPA led to its inactivation at concentrations 30- and 300-fold lower than for ethanolamine , consistent with the known kinetics of these scavengers in Paal-Knorr reactions ( Amarnath et al . , 2004 , 2015 ) ( Figure 3—figure supplement 1e ) . These findings , together with the results from the OPA inactivation assays , are consistent with OPA reacting with PE according to a Paal-Knorr reaction mechanism ( Figure 3f and Figure 3—figure supplement 2 ) . Based on our observations that OPA can be inactivated by exogenous PE in vitro and that it forms a pyrrole-containing adduct with ethanolamine , we hypothesize that the bioactivity of OPA arises from the formation of covalent adducts with PE . To facilitate detection of such adducts in human cells after treatment with OPA , we utilized phospholipase D ( PLD ) from Streptomyces chromofuscus ( Sullivan et al . , 2010 ) to release modified PE head groups from heterogeneous populations of lipids differing in fatty acid composition ( Figure 4a ) . We first confirmed the efficacy of this approach by synthesizing PE-OPA covalent adducts in vitro and hydrolyzing these adducts with PLD to release ethanolamine-OPA ( Etn-OPA ) ( Figure 4b ) . Etn-OPA was unequivocally characterized by its exact mass , retention time , and MS/MS fragmentation pattern using the reaction product of OPA with ethanolamine as a standard ( Figure 3d ) . The detection of Etn-OPA was dependent on the presence of PE and OPA , and PLD treatment , demonstrating that OPA forms a pyrrole-containing covalent adduct with PE which PLD is able to hydrolyze into Etn-OPA ( Figure 4b ) . 10 . 7554/eLife . 14601 . 016Figure 4 . Ophiobolin A ( OPA ) forms a pyrrole-containing covalent adduct with phosphatidylethanolamine ( PE ) in human cells . ( a ) Formation of PE-OPA adducts was detected by measuring the abundance of ethanolamine-OPA ( Etn-OPA ) after hydrolysis by phospholipase D from Streptomyces chromofuscus ( PLD ) . ( b ) Extracted ion chromatograms ( m/z = 426 . 2982–426 . 3024 ) of the liquid chromatography-mass spectrometry ( LC-MS ) analysis of in vitro reactions of PE with OPA and subsequent digestion with PLD . Control reactions include systematic replacement of each reagent by vehicle and replacement of PE with phosphatidylcholine ( PC ) . ( c–d ) Extracted ion chromatograms ( m/z = 426 . 2982–426 . 3024 ) showing the detection of PE-OPA adducts in lipids extracted from cells treated with OPA . ( c ) HEK293T cells grown in standard conditions were incubated with 250 nM OPA for 24 hr . Total cellular lipids were extracted in the presence of pentyl-pyridoxamine to quench unreacted OPA . Lipids were incubated with PLD and analyzed by LC-MS for the presence of Etn-OPA . Negative controls include replacement of OPA by DMSO vehicle or absence of PLD treatment . ( d ) Same as ( c ) but for HCT116 cells treated with 450 nM OPA . Full chromatograms and replicate experiments are available in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 01610 . 7554/eLife . 14601 . 017Figure 4—figure supplement 1 . Raw data and replicate experiment for data represented in Figure 4c–d . Extracted ion chromatograms ( m/z = 426 . 2982–426 . 3024 ) showing the detection of phosphatidylethanolamine-ophiobolin A ( PE-OPA ) adducts in lipids extracted from cells treated with OPA . ( a ) HEK293T cells grown in standard conditions were incubated with 250 nM OPA for 24 hr . Total cellular lipids were extracted in the presence of pentyl-pyridoxamine to quench unreacted OPA and prevent any post-lysis reaction . Lipids were then incubated with phospholipase D ( PLD ) and analyzed by LC-MS for presence of Etn-OPA . Negative controls include replacement of OPA by DMSO vehicle ( - OPA ) or absence of PLD treatment ( - PLD ) . ( b ) Same as ( a ) but for HCT116 cells treated with 450 nM OPA . Extracted ion intensity levels ( 'NL' ) were normalized to the peak of highest intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 01710 . 7554/eLife . 14601 . 018Figure 4—figure supplement 2 . Higher-energy collisional dissociation ( HCD ) MS/MS fragmentation spectra of ethanolamine-ophiobolin A ( Etn-OPA ) . ( a–c ) Etn-OPA was unequivocally characterized by its exact mass , liquid chromatography retention time , and MS/MS fragmentation pattern . Ions at m/z = 426 . 30 and at a retention time of 12 . 5 min were fragmented by collision-induced dissociation to ensure an MS/MS spectrum consistent with that of Etn-OPA prepared by incubation of OPA with ethanolamine . [M+H]+ ions were targeted using dynamic exclusion at a normalized collision energy of 35 eV . ( a ) HCD spectrum of ion at m/z = 426 . 30 and at a retention time of 12 . 5 min in the experiment in Figure 3d involving in vitro reaction of OPA with ethanolamine . ( b ) HCD spectrum of ion at m/z = 426 . 30 and at a retention time of 12 . 5 min in the experiment in Figure 4b involving in vitro reaction of OPA with phosphatidylethanolamine ( PE ) and hydrolysis with phospholipase D ( PLD ) . ( c ) HCD spectrum of ions at m/z = 426 . 30 and at a retention time of 12 . 5 min in the experiments in Figure 4c–d and Figure 4—figure supplement 1 involving PLD hydrolysis of phospholipids extracted from cells treated with OPA . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 018 We next used this analytical strategy to query the presence of PE-OPA adducts in the pool of phospholipids extracted from human cells treated with OPA . HEK293T and HCT116 cells were treated , total cellular phospholipids were extracted , and residual OPA was rapidly quenched with pentyl-pyridoxamine to prevent any reaction of PE with residual OPA after cell lysis ( Amarnath et al . , 2015 ) . The quenched extracted phospholipids were digested with PLD and analyzed by LC-MS/MS , which revealed the presence of Etn-OPA dependent on OPA and PLD treatment ( Figure 4c–d and Figure 4—figure supplements 1 and 2 ) . Importantly , the absence of Etn-OPA in the control lacking PLD treatment indicates that OPA does not react with cellular ethanolamine ( Figure 4c–d ) . These findings demonstrate that OPA forms pyrrole-containing covalent adducts with PE in human cells . Modification of PE to PE-OPA adducts substantially changes the biophysical properties of PE by modifying its head group from small and polar to bulky and hydrophobic . We thus hypothesized that PE-OPA adduct formation could lead to destabilization of cellular lipid bilayers and be the main cause of OPA cytotoxicity . To determine whether OPA induces membrane leakiness due to adduct formation , we used artificial liposomes whose PE content we could modulate . We prepared large unilamellar lipid vesicles ( LUVs ) composed of a variable ratio of dioleoyl-PE to dioleoyl-PC and loaded with the fluorescent probe calcein ( Allen and Cleland , 1980; Zhang et al . , 2001 ) . As calcein fluorescence becomes dequenched upon release from LUVs , an increase in fluorescence can be used as an indicator of liposome leakage . We observed that OPA treatment caused extensive liposome leakage , and that the extent of leakage is directly dependent on both the PE content of the liposomes and on the OPA concentration ( Figure 5a ) . Importantly , we observed that liposomes composed only of PC or with very low PE content remain intact even at high OPA concentration , indicating that the leakage is likely specifically due to the formation of PE-OPA adducts . These results clearly show that OPA destabilizes lipid bilayers , supporting our hypothesis that membrane permeabilization is the main cause of OPA cytotoxicity . 10 . 7554/eLife . 14601 . 019Figure 5 . Ophiobolin A ( OPA ) induces leakage from liposomes . ( a ) Effect of 60 min OPA treatment on the leakage of the fluorescent dye calcein from artificial liposomes composed of various ratios of phosphatidylethanolamine ( PE ) to phosphatidylcholine ( PC ) . Large unilamellar vesicles ( LUVs ) composed of dioleoyl-PE and dioleoyl-PC , and encapsulating calcein were prepared by extrusion . Leakage assays were initiated by addition of OPA and were monitored by quantification of the fluorescence of calcein , which dequenches upon release from liposomes . OPA-induced liposome leakage was determined by normalization of fluorescence to a DMSO vehicle control ( 0% leakage ) and to detergent-treated liposomes ( 100% leakage ) . Results were obtained from assays performed in triplicate and data represent mean values ± standard error of the mean . Inset plots display examples of the kinetics of leakage after OPA addition for two liposome preparations . ( b ) Proposed model of the mechanism of action of OPA in human cells . When applied to cells , OPA accumulates in the phospholipid bilayer of the plasma membrane as it is a lipophilic compound . Due to high local concentrations or the hydrophobic environment ( or both ) , OPA efficiently reacts with the primary amine head group of phosphatidylethanolamine ( PE ) in a Paal-Knorr-like reaction . Formation of PE-OPA adducts changes the biophysical properties of PE by modifying its head group from small and polar to bulky and hydrophobic , leading to membrane permeabilization and ultimately cell death . DOI: http://dx . doi . org/10 . 7554/eLife . 14601 . 019 Our work illustrates the utility of loss-of-function screens in human cells to identify genes involved in the MOA of promising anticancer compounds such as OPA . The Kennedy pathway genes identified in the screen of OPA led to the identification of a phospholipid molecule as the cellular target of OPA . This discovery was possible due to the unbiased nature of the screen and would have been difficult to achieve using alternative methods such as affinity chromatography followed by mass spectrometry . Few studies have yet taken advantage of the recent development of novel genetic tools in human cells to identify the target of small molecules , and this is the first study to uncover a non-protein based target ( Nijman , 2015 ) . Here we show that OPA forms a pyrrole-containing covalent adduct with PE in human cells . Reduction of PE levels through inactivation of the Kennedy pathway results in a reduction in OPA cytotoxicity . We also show that PE-OPA adduct formation leads to destabilization of lipid bilayers in vitro . Collectively , this study indicates that PE is the main target of OPA in human cells and leads to the hypothesis that formation of PE-OPA adducts directly causes the observed cytotoxicity of OPA through membrane destabilization ( Figure 5b ) . OPA is reactive towards primary amines , but the chemical reactivity of PE alone does not explain why OPA would selectively react with PE over ethanolamine , lysine side chains of proteins , or any other abundant primary amine in human cells ( Dasari et al . , 2015 ) . While we cannot exclude the possibility that covalent modification of cellular proteins contributes to the observed toxicity of OPA , our experimental results show that the concentration of PE required to inactivate OPA in vitro is at least 2000-fold lower than that of ethanolamine or lysine ( Figure 3a and Figure 3—figure supplement 1 ) . This selectivity may arise because , at the concentrations tested , ethanolamine and lysine are freely soluble in aqueous buffers whereas PE forms insoluble lipid aggregates . Considering the lipophilic nature of OPA , it is likely that OPA accumulates in the lipid aggregates and efficiently reacts with PE due to high local concentrations or to the hydrophobic environment , or both . In a comparable way , we believe that the selectivity in living cells is the result of accumulation of OPA in lipid bilayers and efficient reaction with PE , the most abundant primary amine in human lipid bilayers ( Vance and Tasseva , 2013 ) . The observation that OPA does not react with ethanolamine in human cells , despite efficient reaction with PE , supports this claim ( Figure 4c–d ) . The increase in OPA resistance conferred by inactivation of the Kennedy pathway in our haploid genetic screen strongly suggests that PE-OPA covalent adduct formation is a major determinant of OPA cytotoxicity . We surmise that the observed cytotoxicity of OPA reflects the amount of covalent PE-OPA adducts formed in cells , which , based on the mechanism of the Paal-Knorr reaction , should be dependent on the concentrations of both OPA and PE . Accordingly , we observed that a mild reduction in PE levels ( 16–24% ) was accompanied by a mild increase in OPA resistance ( IC50 from 43 nM in wild-type cells to 70–85 nM in the mutants ) ( Figure 1e and Figure 1—figure supplement 4a ) . As the amplitude of changes in the observed cytotoxicity of OPA can be accounted for by changes in PE levels , we believe that PE represents the main target of OPA in human cells . Several members of the ophiobolin family of fungal metabolites have been isolated and evaluated for cytotoxicity in a panel of human cell lines ( Au et al . , 2000; Dasari et al . , 2015 ) . Notably , the C-6 epimer of OPA ( 6-epi-OPA ) is 40-fold less cytotoxic than OPA ( Dasari et al . , 2015 ) , although both compounds would form the same pyrrole adduct upon reaction with PE , as the stereocenter at position 6 is lost with pyrrole formation . This observation must mean that the reaction of PE with 6-epi-OPA is on the order of 40-fold slower than the reaction of PE with OPA . The ring cyclization step is the rate limiting step of the Paal-Knorr reaction ( Figure 3—figure supplement 2 ) and this step is strongly influenced by the stereochemistry of the 1 , 4 diketone; rates of pyrrole formation can be up to 57-fold different between stereoisomers ( Amarnath et al . , 1991 ) . Thus , the 40-fold difference in potency between OPA and 6-epi-OPA can likely be attributed to differences in the rate of pyrrole adduct formation , and this observation supports our model that cytotoxicity depends on the amount of PE-OPA adduct formed . We hypothesize that formation of PE-OPA adducts disrupts the lipid bilayer of human cells to induce cell death . As seen in the LUVs leakage experiments , OPA induces strong membrane permeabilization in model membranes and the extent of permeabilization is dependent on PE content ( Figure 5a ) . Interestingly , this observation is consistent with our expectation that the rate of adduct formation is dependent on PE content and suggests that OPA might display higher potency against cells or tissues containing high PE contents . Also , the dependence of membrane leakiness on PE content provides a mechanistic basis for our observation that Kennedy pathway KBM7 mutants are resistant to OPA due to lower PE levels . Cytotoxicity through modification of PE has previously been reported for polygodial , a 1 , 4-dialdehyde antifungal compound , which has been shown to form a pyrrole-containing adduct with PE on the cell surface that is directly linked to its antifungal activity ( Fujita and Kubo , 2005 ) . Furthermore , it has been previously shown that covalent modification of PE with isoketals alters membrane curvature because the newly formed bulky hydrophobic head group partitions to the lipid bilayer and increases lateral pressure ( Guo et al . , 2011 ) . We believe such a mechanism may contribute to the ability of PE-OPA adducts to disrupt lipid bilayers , and suspect that the extent of lipid bilayer destabilization is highest in the plasma membrane as it is the first source of PE encountered by OPA . We were initially surprised to find that treatment of human cells with OPA led to activation of the Kennedy pathway and an increase in PE content ( Figure 2d ) . However , phospholipid homeostasis in mammalian cells is not well understood , but it is generally accepted that mechanisms are in place to tightly regulate membrane lipid composition ( Hermansson et al . , 2011 ) . The activation of the pathway seen upon OPA treatment may be a cellular response to the formation of PE-OPA adducts due to mechanisms regulating membrane PE homeostasis . In addition to its activity in human cells , OPA is cytotoxic towards a broad range of organisms ( Au et al . , 2000 ) . Early studies on the mechanism of cytotoxicity of OPA in plant cells suggested that it causes non-specific damage to membranes ( Chattopadhyay and Samaddar , 1976 ) or even "covalent modification of some membrane component" ( Tipton et al . , 1977 ) . Considering our work in human cells and the fact that PE is ubiquitously found in nature ( Vance and Tasseva , 2013 ) , PE may be the main target of OPA in plants and potentially other organisms . OPA has been shown to have an antitumor effect in a mouse glioblastoma model ( Bury et al . , 2013; Dasari et al . , 2015 ) . The basis for the tumor selectivity of OPA may be due to altered distribution or higher abundance of PE in cancer cells . Indeed , it has recently been shown that PE is found in higher quantities on the outer leaflet of cancer cells ( Stafford and Thorpe , 2011 ) . Furthermore , host defense peptides have been shown to display selectively against cancer cells based on the difference in surface phospholipid composition compared to normal cells ( Leite et al . , 2015; Riedl et al . , 2011 ) . Using the changes in lipid composition of cancer cells as a biomarker represents an interesting approach to chemotherapeutics development ( Leite et al . , 2015 ) and our findings raise the exciting possibility that OPA will prove an effective chemotherapy tool for multidrug-resistant glioblastoma . Unless otherwise mentioned , all chemicals were from Sigma-Aldrich ( St Louis , MO ) . Ophiobolin A ( >95% ) was from Enzo Life Sciences ( Farmingdale , NY ) . Gemcitabine , oxaliplatin , bortezomib , topotecan , and doxorubicin were from LC Laboratories ( Woburn , MA ) . Ethanolamine [1 , 2-14C] HCl was from American Radiolabeled Chemicals ( St Louis , MO ) . Phospholipids were from Sigma-Aldrich ( PE , chicken egg; PC , chicken egg; PS , bovine brain ) and from Avanti Polar Lipids ( Alabaster , AL ) ( transphosphatidylated PE , chicken egg; PE , porcine brain; DOPE; DOPC; DOPS ) . Salicylamine and pentyl-pyridoxamine were a kind gift of V . Amarnath ( Vanderbilt ) ( Amarnath et al . , 2004 , 2015 ) . Stock solutions , unless otherwise mentioned , were prepared in DMSO ( 99 . 9% ) at 500× concentration to yield a final concentration of 0 . 2% DMSO . KBM7 cells were obtained from Thijn Brummelkamp ( Carette et al . , 2009 , 2011 ) and were cultured at 37°C in 5% CO2 in Iscove's Modified Dulbecco Medium ( IMDM ) ( Gibco , Thermo Fisher Scientific , Waltham , MA ) supplemented with 10% heat inactivated fetal bovine serum ( Gibco ) , and penicillin/streptomycin at final concentrations of 100 U/mL and 100 μg/mL , respectively ( P/S ) ( Corning Inc . , Corning , NY ) . HEK293T and HCT116 were obtained from ATCC ( Manassas , VA ) . HEK293T were cultured at 37°C in 5% CO2 in Dulbecco’s Modified Eagle Medium ( DMEM ) ( ATCC ) supplemented with 10% fetal bovine serum ( FBS ) ( ATCC ) , and P/S . HCT116 were cultured at 37°C in 5% CO2 in McCoy's 5A ( ATCC ) supplemented with 10% FBS , and P/S . Cell lines were used at low passage numbers from primary stocks and were not further authenticated or tested for mycoplasma . Mutagenized KBM7 cells were prepared as in Birsoy et al . ( 2013 ) . For each screen , 100 million cells were diluted in 200 mL growth medium and the small molecule of interest was added from a 500× stock solution in DMSO . Then 100 , 000 cells per well were aliquoted in 96-well plates . Plates were incubated at 37°C in 5% CO2 until colonies were visible ( about 3 weeks ) . All resistant cells were pooled , washed with Dulbecco's phosphate buffer saline ( PBS ) and genomic DNA was prepared from 30 million cells using the QIAamp DNA mini kit ( Qiagen , Hilden , Germany ) . Genomic DNA was first digested in separate reactions with NlaIII and MseI and then self-ligated under dilute conditions using T4 DNA ligase ( New England Biolabs , Ipswich , MA ) . After clean-up of the reactions using the MiniElute PCR purification kit ( Qiagen ) , self-ligated products were amplified in a PCR reaction using 10 μM LTRSolexaI , 1 μM either NlaIII or MseI adaptor ( depending on the enzyme used for DNA digestion ) , and 10 μM of index primer , and using Phusion Hot Start Flex polymerase ( New England Biolabs ) . PCR products were cleaned up using the MiniElute PCR purification kit and the presence of amplified products was verified by agarose gel electrophoresis . The PCR reactions of up to 20 screens with unique barcodes were pooled and sequenced using Illumina's ( San Diego , CA ) HiSeq 2500 platform ( 50 bp , single read ) and using primer SolexaSeqFlank . About 5–10 million reads were obtained for each screen . Reads containing MseI or NlaIII sites flanked by vector DNA sequence were trimmed after the restriction site ( to allow potential alignment of fragments shorter than 50 bp ) . Using Bowtie ( Langmead et al . , 2009 ) , reads were aligned to the human genome hg19 with no mismatch allowed and a single alignment site . A list of unique genomic alignment sites was compiled and sites separated by only 1 or 2 bp were discarded . Additionally , alignment sites represented by only one sequencing read were discarded . The insertion sites were next compared to a list of all annotated human introns and exons ( Roche Nimblegen Exon-Intron table , July 2010 , hg19 ) . For each human gene , the total number of unique insertions in exonic regions and those in intronic regions in the sense orientation were counted . Finally , an enrichment p-value was calculated using Fisher's exact test for each annotated gene by comparing the number of inactivating insertion sites after selection to the number of inactivating insertions in that gene in a control library . Genes with less than 10 total reads were not displayed in the bubble plots . A control library was prepared by extraction of genomic DNA from mutagenized KBM7 cells collected prior to initiating loss-of-function screens . PCR products were prepared and analyzed in the same way as above except that genomic DNA from a total of 5 million cells was used as template and 35 million sequencing reads were obtained . Drug-resistant cell lines were isolated from loss-of-function KBM7 screens from wells containing single colonies . Cell lines with retroviral insertions in genes of interest were identified by screening for altered gene expression by RT-qPCR ( see below ) . To ensure clonality of the gene-trapped cell lines for subsequent experiments , single colonies were isolated by serial dilution and propagated in standard growth medium . The location of the retroviral insertion sites in the clonal cell lines was identified using a similar strategy as in haploid screens except that the PCR products were sequenced by Sanger sequencing using primer CCseq . HEK293T or KBM7 cells were grown in 24-well plates and total RNA was extracted from 2 million cells using the RNeasy kit ( Qiagen ) . cDNA was synthesized from 0 . 5 μg total RNA using Superscript III reverse transcriptase ( Invitrogen , Carlsbad , CA ) and oligo ( dT ) 20 primers ( Invitrogen ) , following the manufacturer's instructions . Reactions were diluted two-fold with H2O and 10 μL qPCR reactions were set up using 2× SYBR Green PCR Master mix ( Thermo Fisher Scientific ) , 2 μL diluted cDNA preparation , and 0 . 2 μM of primers . Reactions were monitored using the Stratagene MX3000P qPCR system ( Agilent , Santa Clara , CA ) . Primer pairs CC085/CC086 and CC097/CC098 were used for quantification of ETNK1 expression levels . Primer pairs CC089/CC090 and CC133/CC134 were used for quantification of PCYT2 expression levels . Primer pairs CC042/CC043 and CC044/CC045 were used for quantification of ABCG2 expression levels . Primer pair GAPDH7-8f/GAPDH7-8r was used for quantification of GAPDH expression levels . Relative expression levels were quantified by the △△CT method using GAPDH as reference gene . Values reported are the average △△CT values calculated with the two primer pairs used . All measurements were performed in triplicate for each primer pair . For KBM7 cell lines , 2000 cells were seeded per well in a 96-well plate in 90 μL growth medium . OPA was diluted in 10 μL growth medium and added to wells . After 72 hr incubation at 37°C and 5% CO2 , 100 μL of CellTiter-Glo reagent ( Promega , Madison , WI ) was added to each well . After homogenization , 100 μL of the resulting solution was transferred to a black opaque 96-well plate and luminescence was recorded on a Perkin Elmer ( Waltham , MA ) TopCount NXT system . For HEK293T cell lines , cultures were grown to 80–90% confluence . Cell lines were subsequently seeded in wells of a 96-well plate at a dilution of 1:75 in 90 μL medium . After 16–18 hr incubation at 37°C and 5% CO2 , the remaining steps of the assay were performed as described above . Cells were washed with Tris buffer saline ( TBS ) ( 20 mM Tris , 150 mM NaCl , pH 7 . 6 ) and total lipids were extracted using the Folch method ( Folch et al . , 1957 ) . Briefly , cells were resuspended in 20 vol of CHCl3/MeOH 2:1 ( v/v ) , homogenized for 20 min at room temperature and extracted using 4 vol of NaCl 0 . 9% . After drying the lower phase using a stream of nitrogen , lipids were resuspended in 2 vol of CHCl3/MeOH 2:1 ( v/v ) and separated on Silica Gel 60 thin layer chromatography ( TLC ) plates ( EMD Millipore , Darmstadt , Germany ) according to the method of Skipski et al . ( Skipski et al . , 1964 ) . Plates were developed using CHCl3/MeOH/AcOH/H2O ( 50:30:8:3 , v/v/v/v ) . Phospholipids were visualized by iodine staining and their identity was determined using standards . Phospholipids were then quantified by phosphate determination on scraped silica gel spots using the 'micro' assay as described by Zhou and Arthur ( 1992 ) . For each experiment , the spot corresponding to PE was scraped into a tube . All other spots in the lane visualized by iodine staining were scraped into a second tube . The quantity of phosphate ( Pi ) in each sample was determined using standards consisting of known amounts of KH2PO4 . The relative cellular PE content was estimated using the following formula: PE content = Quantity of Pi in PE tube/ ( Quantity of Pi in PE tube + Quantity of Pi in tube containing all other phospholipids ) . PCYT2 was re-expressed in PCYT2GT cells using the lentiviral vector pLJM1 ( Sancak et al . , 2008 ) . cDNA encoding PCYT2 was amplified by PCR from cDNA prepared from total HEK293T RNA using primers CC109 and CC110 . The plasmid Flag pLJM1 RagB wt ( Addgene #19313 ) was digested with SalI and EcoRV . Amplified PCYT2 cDNA was inserted into pLJM1 by Gibson assembly and the constructed plasmid propagated in stbl2 cells ( Invitrogen ) . Plasmid pLJM1-EGFP ( Addgene #19319 ) was used as a negative control . Lentivirus was produced in HEK293T cells using pLJM1-based vectors together with packaging vector psPAX2 ( Addgene #12260 ) and envelope vector pCMV-VSV-G ( Addgene #8454 ) . KBM7 cell lines were infected with lentiviral particles in the presence of 8 μg/mL polybrene and selected in 0 . 4 μg/mL puromycin for 6 days . Transduced cell lines were maintained in growth medium supplemented with 0 . 3 μg/mL puromycin during subsequent assays . Constructs expressing shRNAs targeting PCYT2 were based on pLKO . 1-TRC ( Addgene #10878 ) ( Moffat et al . , 2006 ) . Five independent shRNAs ( TRCN0000236037 , TRCN0000035648 , TRCN0000236039 , TRCN0000236038 , TRCN0000236040 ) designed by the RNAi consortium ( Broad Institute ) were used to construct HEK293T knockdown cell lines as described by Addgene . Briefly , shRNA oligos were cloned into pLKO . 1-TRC and the constructed plasmids were transfected into HEK293T cells together with psPAX2 and pCMV-VSV-G to produce lentivirus . HEK293T cells were then infected with lentiviral particles and selected in 2 . 0 μg/mL puromycin for 6 days . Cell lines stably expressing shRNAs of interest were maintained in growth medium supplemented with 1 . 5 μg/mL puromycin during subsequent assays . Scrambled shRNA in pLKO . 1 ( Addgene #1864 ) and empty vector pLKO . 1-TRC were used as negative controls . The two shRNA sequences that achieved the highest knockdown efficiency were: kd1 ( TCACGGCAAGACAGAAATTAT , TRCN0000035648 ) and kd2 ( ACTAGAGACCCTGGACAAATA , TRCN0000236039 ) . HEK293T and HCT116 cells were grown to 30–40% confluence in 100 mm dishes in 20 mL medium . OPA or meclizine dihydrochloride ( >97% ) was diluted to 500 μL in growth medium and added to the dishes . After 5 hr incubation at 37°C and 5% CO2 , 0 . 5 μCi ethanolamine [1 , 2-14C] was added to each dish and cells were incubated for an additional 24 hr . Cellular PE content was quantified as described above . For determination of [14C]-PE levels , phospholipids were separated by TLC as described above . After iodine staining , silica spots corresponding to PE were scraped into scintillation vials containing 5 mL of Ready-Solv HP scintillation cocktail ( Beckman Coulter , Brea , CA ) and [14C] counts per minute were measured on a Beckman Coulter LS 6500 system . Levels of [14C]-PE were normalized to the total amount of phospholipids in each sample determined by phosphate determination . This assay was performed as the cell viability assay described above with the following modifications . HEK293T cells were seeded in 96-well plates at a dilution of 1:75 in 80 μL medium . Phospholipids were prepared from 5 or 10 mg/mL stocks in CHCl3 and were first diluted fivefold in MeOH and then fivefold in growth medium . Organic solvents were degassed at 37°C and 10 μL was added to the wells . OPA was diluted in 10 μL growth medium and added to the wells . The rest of the assay was performed as above . Ethanolamine , triethanolamine , O-phosphorylethanolamine , and serine were prepared as 1 M solutions in H2O , adjusted to pH 7 with HCl or NaOH , and diluted with H2O in order to prepare 20 stock solutions . These solutions were diluted to 10 μL with growth medium and added to the wells . For experiments with scavengers of 1 , 4-dicarbonyls , lysine·HCl was prepared as a 1 M stock in H2O and adjusted to pH 7 with NaOH , and salicylamine and pentyl-pyridoxamine were prepared as 50 mM stocks in H2O . All solutions were diluted with H2O in order to prepare 20x stock solutions and used as above . OPA was adjusted to 100 μM in 100 μL PBS and to a final concentration of 2% DMSO . Then 10 μL of either neat ethanolamine ( or ethanol as control ) was added . After 2 hr at 37°C , OPA was extracted using 400 μL of CHCl3/MeOH 2:1 ( v/v ) . The lower phase was dried using a stream of nitrogen and resuspended in 100 μL ethanol . The recovery of OPA was assumed to be 100% and the reactions were tested in cell viability assays . For LC-MS/MS analysis , OPA was adjusted to 0 . 5 mM in 50 μL H2O ( 4% DMSO ) and 3 μL of neat ethanolamine was added to the solution . For the 'ethanolamine only' control , OPA was replaced by DMSO . For the 'OPA only' control , no ethanolamine was added . After 2 hr at 37°C , the reactions were diluted 50-fold in MeOH before LC-MS/MS analysis . LC-MS/MS analysis was performed on a Thermo q-Exactive Plus mass spectrometer coupled to a Thermo Ultimate 3000 uHPLC ( Thermo Fisher Scientific ) . The HPLC method used a Phenomenex ( Torrance , CA ) Kinetex C18 column ( 2 . 6 µm particle size , 10 nm pore size , 150 mm length , and 2 . 1 mm internal diameter ) at a constant flow rate of 0 . 2 mL/min . Mobile phase A was 0 . 1% formic acid in H2O ( v/v ) and mobile phase B was 0 . 1% formic acid in CH3CN ( v/v ) . A 10 μL sample was injected onto the column at 0% B and washed at this solvent composition for 3 min . The gradient was first increased to 10% B in 0 . 1 min and then to 100% B over the next 26 . 9 min . Detection on the q-Exactive Plus mass spectrometer was performed in positive mode between 300 and 2000 m/z , using an acquisition target of 3E6 , and a maximum ion injection time of 200 ms at a resolution of 70 , 000 for MS and 35 , 000 for MS/MS data . For MS/MS experiments , [M + H]+ ions were targeted for isolation and fragmentation at a normalized collision energy of 35 eV . A 10 μL sample of OPA 5 mM in DMSO and 5 μL of ethanolamine 1 M ( pH 7 , prepared above ) were added to 85 μL PBS . Control reactions were prepared by either replacing OPA by DMSO or ethanolamine by triethanolamine . After 3 hr at 37°C , OPA was extracted with CHCl3/MeOH 2:1 ( v/v ) . The lower phase was dried and resuspended in 35 μL ethanol . Ehrlich's reagent was prepared according to Amarnath et al . ( 2004 ) ( 80 mM 4- ( dimethylamino ) benzaldehyde in MeOH/0 . 6 M HCl 1:1 ( v/v ) ) . The resuspended reactions were diluted to 0 . 5 mL with H2O and then to 1 mL with Ehrlich's reagent . After 2 min at 68°C , the solutions were cooled , transferred to a quartz cuvette , and the absorbance was measured between 450 and 700 nm ( SpectraMax Plus 384 , Molecular Devices , Sunnyvale , CA ) . A 5 . 2 μL sample of OPA 5 mM in DMSO and 2 μL of transphosphatidylated chicken egg PE ( 13 mM in CHCl3 ) were added to 44 . 8 μL reaction buffer ( 1 M triethylammonium acetate/CHCl3/MeOH 1:1:3 ( v/v/v ) , as described by Sullivan et al . [2010] ) . Control reactions were performed by replacing either OPA by DMSO , PE by CHCl3 , or PE by chicken egg PC ( 13 mM in CHCl3 ) . After 3 hr at 37°C , the reactions were diluted to 500 μL with MeOH and 50 μL of the diluted reactions was dried using a stream of nitrogen . After resuspension in 25 μL MeOH , 225 μL PBS was added and the suspension was sonicated in a water bath for 2 min . Then 5 μL ( 275 U ) phospholipase D ( PLD ) from Streptomyces chromofuscus ( Enzo Life Sciences ) was added and after 16 hr at 37°C , the reaction was extracted with 1 mL CHCl3/MeOH 2:1 ( v/v ) . The lower phase was dried , resuspended in 1 mL CHCl3 , and analyzed by LC-MS/MS . Cells were seeded at 10% confluence in 100 mm dishes and were grown to 60% confluence in standard conditions . OPA was added to 250 nM for HEK293T and 450 nM for HCT116 in 15 mL complete growth medium . After 24 hr at 37°C and 5% CO2 , the cells were washed with 15 mL TBS and then resuspended in 20 vol of CHCl3/MeOH 2:1 ( v/v ) and 0 . 6 vol of pentyl-pyridoxamine 50 mM . After homogenization for 20 min at room temperature , the suspension was extracted using 4 vol of NaCl 0 . 9% . The lower phase was dried using a stream of nitrogen and resuspended in 100 μL MeOH by sonication . A 50 μL sample of the suspension was diluted to 500 μL in PBS and further sonicated . Then 15 μL PLD ( 825 U ) was added and the suspension was incubated at 37°C for 14 hr . The reaction was extracted with 2 mL CHCl3/MeOH 2:1 ( v/v ) and the lower phase was dried and resuspended in 0 . 5 mL CHCl3/MeOH 2:1 ( v/v ) . A second extraction was performed by addition of 125 μL NaCl 0 . 9% . The lower phase was dried , resuspended in 180 μL CHCl3 , and analyzed by LC-MS/MS . Solutions containing a total of 1 mg of dioleoyl-PE and dioleoyl-PC ( Avanti Lipids ) were prepared from 10 mg/mL stock solutions in CHCl3 . The solvent was removed from these solutions using first a stream of nitrogen and then by evaporation under vacuum . Lipid films were rehydrated at 37°C for 5 hr with 0 . 4 mL 100 mM calcein pH 7 . 4 , and homogenized by vortexing and five freeze-thaw cycles . Liposome suspensions were next extruded 20 times through 100 nm polycarbonate filters ( Avanti Mini-Extruder ) to generate LUVs encapsulating calcein . LUVs were separated from free calcein by gel filtration over Sephadex G-50 using 20 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM EDTA as buffer . Leakage experiments were started immediately after chromatography . A 50 μL sample of calcein-containing LUVs diluted in 20 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM EDTA was dispensed in 96-well plates . Then 50 μL OPA solution ( diluted from 200 stocks in the same buffer as the LUVs ) was added to start the leakage assay . The fluorescence of calcein was monitored each 30 s on a Spectramax i3 ( Molecular Devices ) using excitation at 493 nm and emission at 518 nm over 65 min and at room temperature . Background fluorescence was subtracted for each OPA concentration by using control wells in which calcein-containing LUVs were replaced by buffer . A control using DMSO vehicle instead of OPA was used as the 0% leakage reference and a solution of 0 . 2% Triton X-100 was used instead of the OPA solution in the 100% leakage reference . For each time point , the fluorescence data were normalized to these two reference samples .	Many of the medications that are available to treat cancer are either collected from natural sources or inspired by molecules existing in nature . While it is often challenging to understand how these natural compounds selectively kill cancer cells , characterizing these mechanisms is essential if researchers are to develop new anticancer drugs and treatments based on these compounds . Ophiobolin A is a compound naturally made by a fungus in order to attack plant cells . It is also able to potently kill cancer cells from humans . In particular , ophiobolin A is a promising candidate for treatment of a type of brain tumor called glioblastomas , which are notoriously difficult to treat with existing medications . Using a newly developed method , Chidley et al . have now tested which components of human cancer cells are important for ophiobolin A to exert its killing effect . The method revealed that ophiobolin A was less able to kill cancer cells if the cells had lower levels of a molecule called phosphatidylethanolamine in their surface membranes . This observation led Chidley et al . to show that ophiobolin A enters the membrane of human cancer cells and combines chemically with phosphatidylethanolamine to form a new composite molecule . Further experiments showed that the formation of this composite molecule disrupted a model membrane , which suggests that ophiobolin A kills cancer cells by breaking their membranes . The next challenge is to understand exactly how the composite molecule kills cancer cells via membrane disruption . It also remains unclear if the anticancer activity of ophiobolin A results from cancer cells having a membrane composition that is different from normal cells , and why this difference arises in the first place .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2016	The anticancer natural product ophiobolin A induces cytotoxicity by covalent modification of phosphatidylethanolamine
Two classes of riboswitches related to the ykkC guanidine-I riboswitch bind phosphoribosyl pyrophosphate ( PRPP ) and guanosine tetraphosphate ( ppGpp ) . Here we report the co-crystal structure of the PRPP aptamer and its ligand . We also report the structure of the G96A point mutant that prefers ppGpp over PRPP with a dramatic 40 , 000-fold switch in specificity . The ends of the aptamer form a helix that is not present in the guanidine aptamer and is involved in the expression platform . In the mutant , the base of ppGpp replaces G96 in three-dimensional space . This disrupts the S-turn , which is a primary structural feature of the ykkC RNA motif . These dramatic differences in ligand specificity are achieved with minimal mutations . ykkC aptamers are therefore a prime example of an RNA fold with a rugged fitness landscape . The ease with which the ykkC aptamer acquires new specificity represents a striking case of evolvability in RNA . RNA has diverse functional capabilities , which has driven speculation that the first organisms may have been RNA-based ( Breaker , 2012; Crick , 1968; Gilbert , 1986; Orgel , 2004; 1968; Strobel , 2001; Woese et al . , 1966 ) . For this hypothesis to be plausible , RNA must be adaptable; that is , capable of acquiring new functions through mutation . In the field of evolutionary biology , this trait is described as evolvability ( Kirschner and Gerhart , 1998; Wagner and Altenberg , 1996 ) . Evolvability is the propensity of a system to produce a mutated genotype that yields a beneficial phenotype under new selective pressures ( Ancel and Fontana , 2000; Kirschner and Gerhart , 1998; Wagner , 2008; Wagner and Altenberg , 1996 ) . Often , this occurs through mutation of an existing gene through divergent evolution . For example , bacterial β-lactamases demonstrate significant evolvability through mutations in the Ω-loop . This loop determines substrate specificity , but mutation or outright deletion of the loop does not dramatically affect the overall structure of the protein ( Banerjee et al . , 1998; Hujer et al . , 2001; Kurokawa et al . , 2000; Wachino et al . , 2004 ) . This locus of evolvability allows the protein to adapt to the selective pressures of novel antibiotics . The concept of evolvability has also been studied in RNA , including a notable paper by Draghi et al . ( Draghi et al . , 2010 ) . This study found that adaptation rate is hastened when the build-up of some phenotypically neutral mutations occurs and the web of accessible phenotypes becomes broad . The speed at which an organism adapts is determined by this , as well as the ruggedness of the fitness landscape , which is related to the number of mutations required to reach a new fitness maximum . Variant riboswitches yield insight into the evolvability of RNA . Riboswitch variants are naturally occurring riboswitches with a conserved overall fold but altered ligand specificity . Examples include the guanine/adenine riboswitches and the cyclic-di-GMP/cyclic-GMP-AMP riboswitches ( Kellenberger et al . , 2015; Mandal et al . , 2003; Mandal and Breaker , 2004; Ren et al . , 2015; Serganov et al . , 2004; Smith et al . , 2009; Sudarsan et al . , 2008 ) . Bioinformatic and structural studies of the guanine/adenine and cyclic-di-GMP/cyclic-GMP-AMP aptamers showed that the altered specificity occurs simply by changing base pairing between the RNA and ligand . Recently , the ykkC RNA motif was identified as binding to multiple , chemically dissimilar ligands , which makes this specific scaffold a compelling target for structural studies of RNA evolvability ( Sherlock et al . , 2018a; 2018b; Nelson et al . , 2017 ) . The ykkC RNA was discovered in 2004 and its ligand ( s ) remained unknown for over a decade ( Barrick et al . , 2004; Nelson et al . , 2017 ) . In 2017 , Nelson et al . published two pivotal discoveries regarding this motif: ( 1 ) the majority of these RNAs bind specifically to the guanidinium cation and ( 2 ) the ykkC riboswitch class can be divided into at least two subtypes . Subtype 1 , which has approximately 1500 known examples , is the major class now known as the guanidine-I riboswitch . Subtype 2 was defined as all variants of this motif that do not recognize guanidine . The subtype 2 variants are overall quite similar to guanidine-I riboswitches . They retain the same overall fold , but possess a few characteristic differences at nucleotides crucial for guanidine binding . Notably , most of these differences are centered around a classic S-turn motif that forms the binding pocket of the guanidine-I riboswitch . A similar overall architecture with key differences in binding pocket nucleotides is a signature characteristic of a riboswitch variant ( Weinberg et al . , 2017 ) . Variant ykkC RNAs are found upstream of a variety of genes , although two major groups are apparent . One major group regulates amino acid synthesis and transport genes , which are upregulated during the stringent response . The other regulates de novo purine biosynthesis , which produces purine nucleotides from smaller metabolites under conditions where intact nucleobases are not available ( Sherlock et al . , 2018a; 2018b; Ebbole and Zalkin , 1987; 1989 ) . These riboswitches were designated as ykkC subtype 2a and 2b , respectively . When compared to guanidine riboswitches , subtypes 2a and 2b harbor systematic changes to residues directly involved in guanidine binding , which led to the suggestion that they may have different ligand specificity . For example , where guanidine riboswitches have a conserved adenosine residue ( A46 in the guanidine-I structure solved by Reiss et al . ) , subtypes 2a and 2b have a pyrimidine ( C49 in the present study ) ( Battaglia et al . , 2017; Reiss et al . , 2017 ) . Sorting the entire ykkC class by the identity of this position alone results in a strikingly complete segregation of guanidine-related gene contexts from those that are incongruent with mitigation of guanidine toxicity ( Nelson et al . , 2017 ) . Alignment of subtype 1 , 2a , and 2b sequences also shows an extension of conservation at both the 5′ and 3′ ends of the 2a and 2b aptamer subtypes . These key differences in conserved residues and gene contexts suggested that these ykkC variants have altered ligand specificity while retaining the same overall architecture . Subtype 2a and 2b ykkC riboswitches do not retain the ligand specificity of their parent riboswitch . Using transcription termination and in-line probing assays , Sherlock et al . found that neither subtype is responsive to guanidine ( Sherlock et al . , 2018a; 2018b ) . Instead , subtype 2a is responsive to guanosine tetra/pentaphosphate ( ( p ) ppGpp , hereafter referred to as ppGpp ) , an alarmone that regulates the stringent response ( Cashel and Gallant , 1969; Dalebroux and Swanson , 2012; Gaca et al . , 2015 ) . Subtype 2b is responsive to phosphoribosyl pyrophosphate ( PRPP ) , a precursor in purine biosynthesis . Like the guanidine riboswitch , both function as ON switches . The consensus motifs for subtypes 2a and 2b are remarkably similar to each other , even relative to other ykkC RNAs ( Sherlock et al . , 2018a; 2018b ) . The most apparent difference is a highly-conserved guanosine ( G96 in this study ) in subtype 2b that is not conserved in subtype 2a . This residue is equivalent to G89 in the guanidine-I riboswitch and is a conserved part of its S-turn motif . Although bioinformatic data suggest that variation in G96 is central to the structural differences between subtype 2a and 2b riboswitches , its precise role in this context remains uncertain . Unlike guanidine , the biological roles of PRPP and ppGpp are both well-documented . PRPP is an activated form of ribose 5-phosphate , and a major macromolecular building block ( Hove-Jensen et al . , 2017 ) . It is a central metabolite used in biosynthesis of purine and pyrimidine nucleotides , the amino acids histidine and tryptophan , nicotinamide adenine dinucleotide , thiamine diphosphate , flavins , and pterins ( Hove-Jensen , 1988; Jiménez et al . , 2008; White , 1996 ) . The centrality of PRPP within metabolism makes it an appealing target for regulation . ppGpp is an alarmone that initiates the stringent response , a global reaction to nutrient starvation in bacteria ( Cashel and Gallant , 1969; Cashel and Kalbacher , 1970; O'Farrell , 1978; Potrykus and Cashel , 2008 ) . Amino acid starvation triggers synthesis of ppGpp and it binds to a variety of effector molecules to initiate sweeping changes in the cell’s transcriptional profile , including a reduction in tRNA and rRNA synthesis and an increase in transcription of amino acid biosynthesis genes ( Cashel , 1970; Paul et al . , 2005; Ryals et al . , 1982; van Ooyen et al . , 1976 ) . Consistent with a role in the stringent response , the ppGpp riboswitch turns on transcription of amino acid biosynthesis and transport genes in response to alarmone binding . Although the tree topology is unknown , a common ancestral RNA likely diverged to recognize guanidine , PRPP , and ppGpp in spite of the chemical and structural diversity among these ligands . PRPP and ppGpp are more similar to each other than either is to guanidine , which reflects the greater similarity in their aptamers . While guanidine harbors a single delocalized positive charge , PRPP and ppGpp harbor multiple separate loci of negative charge . Guanidine is small and achiral with three-fold rotational symmetry , while PRPP and ppGpp are larger , chiral , asymmetric molecules . PRPP and ppGpp both contain ribose sugars and pyrophosphate moieties , but ppGpp has an entire guanine base that PRPP lacks . Bioinformatic evidence suggests that the 2a and 2b aptamers represent an especially concise solution to a central biophysical problem: biologically relevant switching entails recognition of a cognate ligand and rejection of structurally similar alternatives . We set out to determine how three RNA elements with a common scaffold could recognize such dissimilar ligands with high specificity . Central questions include how a polyanionic macromolecule differentially recognizes two distinct small polyanions , and how the presence or absence of the guanine base changes the RNA’s recognition strategy . To address these questions of molecular recognition by RNA , we report the near-atomic resolution structure of a native ykkC 2b riboswitch in complex with PRPP via X-ray crystallography . We also convert this construct into a ppGpp aptamer with a single G96A mutation and present the structure of the mutant bound to ppGpp . This structural and biochemical information reveals how the ykkC RNA differentiates between ppGpp and PRPP . This study showcases the functional plasticity of RNAs and the evolvability of RNA function from a single structural scaffold . To understand the basis of ligand recognition by the PRPP riboswitch , we determined the crystal structure of the aptamer domain of the ykkC 2b riboswitch from Thermoanaerobacter mathranii at 2 . 5 Å resolution in the presence of its native ligand , PRPP ( Supplementary file 1 ) . PRPP is an activated metabolic intermediate . As a result , it is highly unstable . It degrades on a time course of minutes to hours via several mechanisms in the presence of divalent metal ions , acidic or basic pH , and/or elevated temperatures ( Dennis et al . , 2000; Hove-Jensen et al . , 2017; Khorana et al . , 1958; 1955; Meola et al . , 2003; Remy et al . , 1955 ) . However , binding to the PRPP riboswitch aptamer domain protects PRPP on a time scale of hours to days ( Figure 1—figure supplement 1 ) . The stabilizing effect of the aptamer permitted crystals of the intact complex to be observed after two days . Once formed , unfrozen crystals disappeared after approximately five to ten days , underscoring the need for prompt crystallization and cryogenic preservation in this study . The structure was solved by molecular replacement using the guanidine-I aptamer as an initial model . After model building and refinement , the model fit the data with an Rwork of 0 . 216 and an Rfree of 0 . 253 . Like its parent aptamer , the PRPP riboswitch contains two adjacent helical stacks ( Figure 1 ) . P3 forms a large portion of the binding pocket , and a conserved loop at the end of P3 docks into P1a . This allows conserved nucleotides from P1a to participate in ligand recognition . P1 , P1a , P1b , and P2 together form a continuous coaxial stack adjacent to P3 . However , unlike the guanidine aptamer , the PRPP aptamer has structured tails at the 5′ and 3′ ends that are not conserved in the guanidine riboswitch . The ends pair to form an additional short helix that we have termed P0 , resulting in a four-way junction between P0 , P1 , P2 , and P3 . P0 coaxially stacks with P3 and extends the binding pocket for recognition of the larger PRPP ligand . The overall architecture of the PRPP aptamer reveals that it is a rather conservative adaptation of the guanidine aptamer with key differences that allow for PRPP recognition . Although PRPP is unstable in solution , it has high occupancy in this crystal structure . PRPP is modeled with an occupancy of 1 , and its B factors refined similarly to those of nearby residues . The quality of the fit between the electron density data and this model shows that a combination of protection by the riboswitch and a vast molar excess of ligand permitted a high degree of aptamer saturation when data were collected . PRPP is a potentially challenging ligand for RNA to recognize; it has three negatively charged phosphate groups and lacks a moiety resembling a nucleobase . PRPP is known to interact with two divalent metal ions per molecule in solution . The 5-phosphate associates weakly with one metal and the pyrophosphate moiety more strongly coordinates a second metal ( Thompson et al . , 1978 ) . In the current model , these two metals are present in the complex with the riboswitch ( Figure 2 ) . One metal ( M1 ) associates with the 5-phosphate , and the second metal ( M2 ) associates with the pyrophosphate . Both metals form contacts bridging PRPP and the RNA aptamer . A third metal ion , M3 , forms a water-mediated coordination to the 5-phosphate . The same water molecule also coordinates M1 . The three phosphate groups are major elements of recognition via interactions with nucleobase amines and divalent metal ions . This construct crystallizes in the presence of BaCl2 , so both Ba2+ and Mg2+ are present in the crystallization condition . M1 and M3 are modeled as Ba2+ due to the appearance of large positive peaks in the electron density map when they are modeled as Mg2+ . M2 is modeled as Mg2+ , but exhibits coordination distances higher than expected for this species ( Figure 3—figure supplement 1 ) . The aptamer binds PRPP with nearly equal affinity in the presence of either Ba2+ or Mg2+ alone ( 2 . 0 ± 0 . 4 and 2 . 0 ± 0 . 3 µM , respectively ) . Given that both metals support binding , we expect that there may be partial occupancy of these two species that cannot be resolved at this resolution . The 5-phosphate of PRPP experiences recognition by a metal ion and the amino groups of conserved nucleotides ( Figure 3A ) . The N1 and N2 of G48 form hydrogen bonds with two phosphate oxygens , while the N4 of C78 hydrogen bonds to the third non-bridging phosphate oxygen . The 5-phosphate also coordinates M1 , which is held in place by coordination interactions with a non-bridging phosphate oxygen of C77 and the O2 of C49 . The residue equivalent to C49 is conserved as an adenosine in the guanidine-I riboswitch but is a pyrimidine in PRPP and ppGpp riboswitches , and the identity of residue 49 was used as a marker to distinguish between these two variants ( Sherlock et al . , 2018a; 2018b ) . The O6 of G48 coordinates M3 , but M3 is too distant from the 5-phosphate to be directly coordinated by it . The ribose moiety of PRPP also makes extensive interactions with the RNA aptamer ( Figure 3B ) . The sugar edge of G96 forms hydrogen bonds with the 2- and 3-hydroxyl groups . The N4 of C77 donates a hydrogen bond to the ribose oxygen , and the N1 group of G104 donates a hydrogen bond to the 2-hydroxyl group . These three residues are all highly conserved in the consensus sequence of this aptamer . At 2 . 5 Å resolution , conclusive determination of the sugar pucker is not possible , but a C2-endo pucker is the most likely conformation in this complex and it fits the electron density data well . This conformation avoids a steric clash between the 2-hydroxyl and the β-phosphate and allows the 3-hydroxyl to coordinate M2 . This conformation is also consistent with previously reported structures of PRPP in complex with macromolecules ( Evans et al . , 2014; González-Segura et al . , 2007; Héroux et al . , 2000 ) . The P0 region of the aptamer extends below P3 and permits a suite of interactions with the pyrophosphate group of PRPP ( Figure 3C–D ) . The β-phosphate of PRPP is more extensively recognized than the α-phosphate . The O6 of G6 coordinates M2 , which in turn forms several interactions with the pyrophosphate group ( Figure 3C ) . The N6 group of the weakly conserved A101 ( >75% conserved as a purine ) contacts a non-bridging oxygen of the α-phosphate ( Figure 3D ) . The N6 group of A5 and the N1 groups of G6 and G105 make direct contacts with non-bridging oxygens of the β-phosphate . An abrupt deformation in the local backbone conformation positions A103 under G105 , allowing a lone pair-π interaction to form between the O6 atom of G105 and the six-membered ring of A103 ( Chawla et al . , 2017; Egli and Sarkhel , 2007; Ran and Hobza , 2009; Sarkhel and Desiraju , 2003; Singh and Das , 2015 ) . The present results show that the PRPP aptamer recognizes its ligand through a shifted and extended helical ligand-binding region , allowing for the retention of bound metal ions and extensive hydrogen bond donation to phosphate groups . The intracellular PRPP concentration in bacteria is estimated to be in the millimolar range ( Hove-Jensen et al . , 2017; Jendresen et al . , 2011; Jensen et al . , 1979; Nygaard and Smith , 1993; Saxild and Nygaard , 1991; Schneider and Gourse , 2004; Yaginuma et al . , 2015 ) . However , enzymes and protein regulatory elements that sense PRPP concentrations in bacteria typically have micromolar dissociation ( Kd ) or Michaelis ( KM ) constants ( Bera et al . , 2003; Hove-Jensen et al . , 2017; Jørgensen et al . , 2008 ) . Sherlock and colleagues recently found that the T50 ( the ligand concentration that produces half-maximal effect ) of a PRPP riboswitch in transcription termination assays is 90 μM ( Sherlock et al . , 2018b ) . We determined the Kd of the riboswitch aptamer domain for PRPP ( Table 1 , see also Figure 1—figure supplement 2A ) by equilibrium dialysis using radiolabeled [β-33P]-PRPP . This assay yields a Kd of 2 . 0 ± 0 . 3 μM . There are two notable differences between the present experimental system and that employed by Sherlock et al . First and most importantly , the present study examines binding affinity in an isolated aptamer domain , while Sherlock et al . focused on the ability of the full riboswitch to terminate transcription . The full system is governed by the kinetics of ligand association and RNA folding , while the present experimental system only measures the thermodynamics of ligand binding . Also , in this study , [β-33P]-PRPP was used in trace quantities and the amount of intact PRPP remaining in each sample was carefully measured to deconvolute the counts obtained from intact PRPP and the counts obtained from breakdown products . Sherlock et al . used unlabeled PRPP and could not quantify the extent of degradation , likely resulting in some underestimation of PRPP’s ability to terminate transcription . The present data show that the affinity of the complex is at least of low micromolar affinity , placing it well within the range observed for complexes of PRPP with protein elements ( Bera et al . , 2003; Jørgensen et al . , 2008 ) . In parallel with structural inquiries into the PRPP riboswitch , crystallization of native ppGpp aptamers was pursued . However , crystallization was unsuccessful with the subset of ppGpp aptamers tested . Considering the evident versatility of the ykkC motif and the overt similarity between the consensus sequences of ykkC RNA subtypes 2a and 2b , a specificity switch of the PRPP aptamer to a ppGpp aptamer was pursued via mutation as an alternative strategy . Close examination of the consensus motifs of the PRPP and ppGpp riboswitch aptamers revealed that the ppGpp aptamer consensus sequence was almost entirely a subset of the PRPP aptamer consensus sequence , with the PRPP aptamer generally having more stringent requirements than the ppGpp aptamer . The most salient difference between the two consensus sequences is at position 96 . In the PRPP aptamer , this position is >97% conserved as a guanosine , but this conservation is lost in the ppGpp aptamer . In the ppGpp aptamer , the lack of conservation in this region complicates the process of sequence alignment . However , it appears that this nucleotide is not always present and , when it is , it appears to be conserved as A , C or U , but not G ( Sherlock et al . , 2018a ) . The dramatic difference in conservation at this site suggested that it may be critical for differential recognition of PRPP and ppGpp . We mutated position 96 in the T . mathranii PRPP aptamer from guanosine to adenosine , generating the G96A mutant . The wild-type aptamer shows low affinity for ppGpp ( Kd = 91 ± 3 μM ) and 46-fold greater affinity for PRPP ( Kd = 2 . 0 ± 0 . 3 μM ) ( Table 1 ) . Conversely , the G96A mutant binds ppGpp with an affinity equivalent to that of wild-type for PRPP ( Kd = 1 . 8 ± 0 . 1 μM ) , but PRPP binding is abolished in the mutant up to 400 μM RNA ( estimated Kd = 1600 ± 200 μM ) . The G96A mutant has approximately 900-fold higher affinity for ppGpp than PRPP . The G96A mutation thus strikingly resulted in approximately a 40 , 000-fold switch in ligand specificity from PRPP to ppGpp . The mutant’s affinity for ppGpp is well within the range of native aptamers tested ( data not shown ) . Having shown that the G96A mutant is a ppGpp aptamer , we solved its crystal structure in the presence of ppGpp to 3 . 1 Å resolution . The crystallization conditions that reproducibly gave rise to co-crystals of the wild-type PRPP aptamer did not yield comparable results for co-crystals of the G96A mutant . However , the G96A mutant was found to crystallize in a separate condition that also produced crystals of the wild-type aptamer . The crystallization reagent used for G96A lacks barium , which was the most abundant divalent metal ion in the wild type crystallization condition . Potassium chloride , sodium chloride , and magnesium chloride were present in the crystallization drops . K+ and Mg2+ ions are observed in the mutant crystal structure . The best mutant crystal diffracted to a resolution of 3 . 1 Å and its structure was solved by molecular replacement using chain A of the PRPP riboswitch as an initial model . The asymmetric unit contained four aptamer molecules . Molecular replacement and refinement revealed robust density for the electron-dense pyrophosphate groups of ppGpp as well as its guanine base . In the initial solution and throughout refinement , the quality of the electron density was worse in chain D compared to chains A-C . The model of chain D is consistent with that of chains A-C , but is excluded from discussion in the text . Overall , the architecture of the G96A mutant is very similar to that of the wild-type aptamer ( Figure 1D ) . Notably , the 2FO−FC map generated directly by molecular replacement showed no electron density in the former location of the ribose and phosphate of G96 . Additional lack of electron density for the ribose of G95 and the phosphate of G97 immediately suggested that the G96A mutation caused major conformational rearrangement in this region . The orientation of the ppGpp ligand was determined by examining an FO−FC map where the input model lacked ppGpp . The positions of the 5′ and 3′ pyrophosphate groups of ppGpp are easily inferred from the available electron density data , which clearly show that the 5′ pyrophosphate occupies the former position of the pyrophosphate of PRPP . In this orientation , there is high electron density at the phosphates and lesser electron density at the 4′ and 5′ carbons , as expected ( Figure 4A ) . This results in the ppGpp ribose having the opposite orientation of the PRPP ribose . The 5′ pyrophosphate is oriented toward P0 in the ppGpp structure , but the 5-phosphate is oriented away from P0 in the PRPP structure ( Figure 1C , D , 2A and 4A ) . Several metal ions appear to associate with the pyrophosphate moieties . These were initially assigned as magnesium ions or water molecules , and subsequently assigned as more electron dense potassium ions due to implausibly low B factors after refinement . The positioning of these entities is highly variable among the molecules in the asymmetric unit , suggesting that they do not make essential contributions to ligand recognition , but may provide general charge stabilization . The guanine base of ppGpp is modeled in the syn conformation ( Figure 4—figure supplement 1 ) . At 3 . 1 Å resolution , it is essential to inform this decision with the expected behavior of the chemical constituents in addition to the available electron density data . The shape of the electron density appears visibly more consistent with the syn conformation than the anti conformation . The chemical environment is also more plausible . In the syn conformation , the guanine base of the ligand forms three hydrogen bonds with C75 in a Watson-Crick base pair . In the anti conformation , the Hoogsteen face of the guanine base would form just one hydrogen bond with the Watson-Crick face of C75 . Refinement of the ligand in the anti conformation created steric clashes or very short hydrogen bonds between the O6 of ppGpp and the N4 of C75 , while simultaneously yielding unusually long hydrogen bonds ( >3 . 5 Å ) between the N7 of ppGpp and the N3 of C75 . Modeling a Watson-Crick base pair ( syn conformation ) is consistent with a recent study showing that the equivalent of a C75U mutant in a native ppGpp riboswitch confers specificity to adenine-containing ligands over guanine-containing ligands ( Sherlock et al . , 2018a ) . The syn conformation of ppGpp was previously observed in a 2 . 0 Å X-ray crystal structure of an E . coli lysine decarboxylase , LcdI ( Kanjee et al . , 2011 ) . Finally , a structural overlay of the wild-type and G96A structures at C75 shows that in the syn conformation , the base of ppGpp in the G96A structure occupies the same position as the base of G96 in the wild type structure . The 3′ pyrophosphate of ppGpp consistently sits in a pocket lined with hydrogen bond donors ( Figure 5A ) . The N4 of C77 , the N1 and N2 of G48 , and the 2′OH of A76 all make hydrogen bonds to the phosphate oxygens . A76 and G48 form a type I A-minor-like interaction in which the Watson-Crick edge of G48 interacts with the ligand , rather than being involved in a canonical base pair . While the position of the 5′-pyrophosphate of ppGpp is relatively invariable , the 3′-pyrophosphate occupies a slightly different position in each molecule of the asymmetric unit . Consistent with this model , the 3′ pyrophosphate atoms have slightly higher B factors than the rest of the ligand ( ~138 Å2 for the 3′ pyrophosphate compared to ~119 Å2 for the 5′ pyrophosphate ) . In chain A , the 3′-β-phosphate has one oxygen that accepts a hydrogen bond from the N1 of G48 , a second oxygen that accepts a hydrogen bond from the N4 of C77 , and a third , unrecognized oxygen . The recognition strategy is slightly different for chains B and C . While these observations may suggest genuine variation in recognition of the 3′ pyrophosphate , definitive interpretation is confounded by the comparatively lower resolution of this data set . The guanine base of ppGpp is buried in the RNA and is the focal point of ligand recognition . In the PRPP aptamer , the highly conserved C75 forms a Watson-Crick base pair with G96 . The G96A mutant ppGpp aptamer recognizes its ligand through a similar Watson-Crick base pair between the G of ppGpp and C75 ( Figure 5B ) . The guanine base of the ligand is also recognized via stacking with G6 and is 56% buried , compared to 38% of ppGpp overall . Such extensive recognition of ppGpp’s nucleobase suggests a likely mechanism for the mutant’s observed discrimination for ppGpp over PRPP . In the native PRPP aptamer , C75 is in the same location near the binding pocket , poised to form this interaction with ppGpp . However , the highly conserved G96 is also available to form this base pair and its spatial proximity to C75 raises its effective concentration , making it potentially able to outcompete ppGpp for this base pairing interaction . This model is consistent with the observation of low-affinity ppGpp binding ( Kd = 91 ± 3 μM ) in the wild type PRPP aptamer and explains why a single mutation at position 96 renders this aptamer capable of recognizing ppGpp with high affinity . The ribose is not recognized by the aptamer , leaving the guanine base and pyrophosphates as the major points of recognition . Recognition of the 5′-pyrophosphate of ppGpp is extensive; its phosphate oxygens accept several hydrogen bonds from amino groups of conserved nucleobases ( Figure 5C ) . The 5′-β-phosphate has three oxygen atoms that can accept hydrogen bonds from the aptamer . One of these oxygens accepts a hydrogen bond from the N6 group of A5 . The second oxygen can accept hydrogen bonds from the N6 of A5 and N1 and N2 of G6 , although it is not expected that these would all form simultaneously . The third oxygen can accept hydrogen bonds from the N1 and N2 of G105 and the N1 of G104 . As with the previous oxygen , it is not expected that these would all form simultaneously . The 5′-α-phosphate appears to be unrecognized , consistent with its similar position to the poorly recognized α-phosphate of PRPP in the native structure . Nucleotide A74 appears to play a conserved structural role in the PRPP and ppGpp aptamers . In all three ykkC subtypes it forms a noncanonical base pair with G6 , which directly contacts PRPP and ppGpp , suggesting that it plays a role in positioning G6 ( Figure 6—figure supplement 1 ) . In the guanidine-I riboswitch , this nucleotide is not conserved . However , A6 in the guanidine-I crystal structure flips out to form the same non-canonical base pair with A68 ( equivalent to A74 ) that is observed in the present study ( Battaglia et al . , 2017; Reiss et al . , 2017 ) . The lack of conservation at this position does not support a role in guanidine recognition , but this conserved interaction is observed in all three aptamers ( Nelson et al . , 2017 ) . The guanidine-I and PRPP ykkC aptamers each have an S-turn motif in the P3 helix . In the guanidine-I aptamer , the orientation of G88 is reversed relative to its stacking partners and G89 flips out of the helix . These are classic features of the S-turn . The guanidine-I riboswitch also possesses a cross-strand purine stack , a characteristic backbone kink on the opposite strand from the S-turn , and stabilizing hydrogen bonds , all of which were first observed in the S-turn of the conserved sarcin-ricin loop in the 23S rRNA ( Correll et al . , 1999 ) . In the PRPP riboswitch , a similar S-turn motif exists at the equivalent position ( Figure 6B ) . Equivalent to G89 in guanidine-I , G96 flips out and base pairs with C75 while also hydrogen bonding to PRPP . Notably , the cross-strand purine stack is absent in the PRPP riboswitch , but other S-turn characteristics are preserved . Conversely , the S-turn motif is abolished in the G96A mutant , and no contacts are observed between A96 and other nucleotides . Even more significantly , G95 does not possess the reverse ribose orientation that defines an S-turn . Rather , this region resembles a standard A-form helix with a single nucleotide bulge . The guanine of ppGpp replaces the flipped out guanosine of the former S-turn motif ( G89/G96 ) ( Figure 6 ) , revealing that the S-turn is a key center of functional plasticity in the ykkC RNAs . Taken together , the present structural and biochemical data shed light on the evolvability of RNA as a whole and of the ykkC motif in particular . Just as residue C49 was previously used to distinguish guanidine aptamers from subtype 2 ykkC RNAs , here we show that G96 is the residue that differentiates PRPP and ppGpp aptamers . Clearly , the sequence space of the ykkC motif is rugged with potential functionality . The existence of ykkC RNAs with other gene contexts and unknown ligand specificity further reinforces the diversity of functions that this single RNA structural motif achieves with very small variations in consensus sequence ( Nelson et al . , 2017 ) . Three-dimensional structural models of the wild type and G96A mutant aptamers reveal that the mechanism of specificity switching is recruitment of C75 as a primary effector of ligand recognition ( Figure 6A ) . The presence or absence of the S-turn motif governs whether an RNA base or the ppGpp base can pair with C75 , and therefore controls the specificity of the aptamer . The wild-type aptamer featured in the current study binds PRPP at a location very near , but distinct from the binding pocket of the guanidine-I riboswitch . The P0 region , which is not present in the guanidine-I riboswitch , recognizes a portion of the larger ligand; metal ion M3 binds in the location where its parent motif binds guanidine ( Figure 6C; see also Figure 3A ) . In the S-turn of the sarcin-ricin loop , the bulged G re-inserts into its helix to form a base triple . In an overlay of the S-turns of the sarcin-ricin loop , the guanidine-I riboswitch , and the PRPP riboswitch , the guanidino group of the bulged guanosine in the sarcin-ricin loop overlays almost exactly with the guanidinium cation , and both roughly overlay with metal M3 in the PRPP riboswitch . The common binding site of M3 in the PRPP riboswitch and guanidine in the guanidine riboswitch may be a case of molecular exaptation ( the co-option of an existing feature for a new purpose ) . This is similar to a case documented in a ribozyme created by SELEX , suggesting that structured RNAs are functionally versatile and can readily adapt to new selection pressures ( Lau et al . , 2017 ) . However , the evolutionary relationship of these two aptamers remains uncertain . The present structural data shed additional light on the potential mechanism of switching in tandem guanine-PRPP aptamers ( Sherlock et al . , 2018b ) . The PRPP riboswitch ( an ON switch ) , is often found immediately downstream of a guanine riboswitch ( an OFF switch ) , in an IMPLY two-input logic gate ( Figure 6—figure supplement 2 ) . In these tandem systems , transcription proceeds in all cases except when guanine is present and PRPP is not . This suggests that PRPP binding disrupts formation of the guanine aptamer , allowing transcription to proceed when both ligands are present ( Figure 6—figure supplement 2D ) . The T . mathranii PRPP aptamer studied in the present work is part of one of these tandem aptamer systems . Its predicted secondary structure shows that formation of the P0 stem of the PRPP aptamer and the P1 stem of the guanine aptamer are mutually exclusive . The present data reveal that the 5′ tail of the PRPP aptamer participates in P0 and plays a central role in PRPP recognition . In the proposed model , PRPP binding stabilizes P0 and disrupts the P1 helix of the guanine aptamer . The IMPLY character of this two-input gate may depend on the relative stabilities of the two helices , which in turn suggests that alternative logic gates could be constructed through mutation of P0 or P1 . The ppGpp and T-box riboswitches are also often found in tandem . In contrast with the PRPP/guanine tandem system , the ppGpp and T-box riboswitches each maintain their own expression platform , suggesting that they fold independently . This is consistent with the AND behavior of this logic gate and the unimportance of the order of the two aptamers within the molecular circuit , but future studies in vivo are needed to confirm this . The observed ability of the ykkC scaffold to reach new ligand specificities via mutation of a few key residues is reminiscent of other accounts of adaptability in both proteins and RNAs . In a clinically relevant contemporary example , β-lactamases evolve to expand their catalytic repertoire through mutations in a flexible loop . These mutations preserve the overall architecture of the protein while enabling it to metabolize new variations on a common antibiotic scaffold , contributing to the worldwide threat of antibiotic resistance ( Banerjee et al . , 1998; Hujer et al . , 2001; Kurokawa et al . , 2000; Wachino et al . , 2004 ) . The repurposing of protein scaffolds for the development of new catalysts has also been exploited in the design of novel enzymes including a Diels-Alderase ( Siegel et al . , 2010 ) . In a related RNA example , the Tetrahymena ribozyme scaffold supports catalytic activities including self-cleavage , RNA polymerization , and peptide bond hydrolysis , though this is likely due to the placement of the substrates into physical proximity with each other ( Kruger et al . , 1982; Lau and Ferré-D’Amaré , 2016; Piccirilli et al . , 1992; Zaug and Cech , 1986 ) . Natural riboswitch aptamers subjected to directed evolution switch specificity , but maintain their overall fold ( Porter et al . , 2017 ) . The existence of riboswitch variants that use the same scaffold but bind slightly different ligands , including the adenine/guanine and cyclic-di-GMP/cyclic-GMP-AMP riboswitch classes , has previously hinted at the adaptability of RNA elements ( Kellenberger et al . , 2015; Mandal et al . , 2003; Mandal and Breaker , 2004; Ren et al . , 2015; Serganov et al . , 2004; Smith et al . , 2009; Sudarsan et al . , 2008 ) . However , the present finding that a single conservative point mutation in the PRPP aptamer can dramatically alter both ligand specificity and tertiary structure reveals a striking example of RNA plasticity and speaks to the macromolecular evolvability of RNA . A final key observation in this study is the direct visualization that an RNA element has evolved to specifically recognize PRPP . PRPP is a central metabolite , and likely has played that role since before the metabolic pathways of life’s last universal common ancestor ( LUCA ) were fully developed ( Glansdorff et al . , 2008 ) . It is possible that PRPP was used for the synthesis of nucleotide precursors on the prebiotic Earth ( Akouche et al . , 2017 ) . The finding that an extant RNA specifically recognizes PRPP lends credence to the hypothesis that RNA elements may have been capable of recognizing PRPP before the advent of coded protein synthesis . RNA was prepared essentially as in Reiss et al ( Reiss et al . , 2017 ) . Plasmids containing ykkC PRPP riboswitch DNA from T . mathranii downstream of the T7 promoter were obtained from GeneArt at Thermo Fisher Scientific . The aptamer domain was extended at the 5′ end by one nucleotide to aid transcription by T7 polymerase ( Salvail-Lacoste et al . , 2013 ) . Plasmid DNA was prepared using a QIAgen MaxiPrep kit and the accuracy of the sequence was verified using Sanger sequencing ( Sanger et al . , 1977 ) . Template DNA for transcription was made using PCR with Phusion polymerase and primers 5′-TAATACGACTCACTATAGTGAAAGTGTACC-3′ and 5′-TACGAGTGAAACCTATCCTCCCG-3′ . G96A transcription template was generated using the primers 5′-TAATACGACTCACTATAGTGAAAGTGTACC-3′ and 5′-TACGAGTGAAACCTATCCTCTCGGGCTTTTGTCC-3′ . Template was purified using the Zymo Research DNA Clean and Concentrator 500 kit . RNA was transcribed from 20 ng/μL PCR template using T7 polymerase in the presence of 80 mM HEPES-Na pH 7 . 5 , 5 mM DTT , 1 mM spermidine , 0 . 12 mg/mL bovine serum albumin , 6 mM NTPs , 44 mM MgCl2 , and 1 U/nL inorganic pyrophosphatase ( Hartmann , 2009 ) . Transcription reactions proceeded for approximately 4 hr at 37°C . Monomeric RNA was exchanged into gel filtration buffer ( 50 mM MES pH 6 . 2–6 . 3 , 100 mM KCl , 10 mM MgCl2 ) , filtered , and purified natively on a HiLoad 26/600 Superdex 75 pg gel filtration column in a cold room ( 6 ± 2°C ) . Monomers eluted at ca . 0 . 6 column volumes , and were pooled and concentrated to >100 μM . Crystals were grown using the microbatch-under-oil method with 2:1 paraffin:silicon oil . In all cases , crystals appeared within two days . Initial crystallization screening was performed using Hampton Research Natrix HT at 23 and 30°C . To produce the wild type crystals used for data collection , 2 μL of 150 μM RNA in 10 mM MgCl2 , 10 mM KCl , 10 mM HEPES-KOH pH 7 . 5 , and 10 mM PRPP ( Millipore Sigma ) was mixed with 1 μL of a solution of 80 mM sodium chloride , 20 mM barium chloride dihydrate , 40 mM sodium cacodylate trihydrate pH 5 . 6 , 45% v/v ( +/- ) −2-methyl-2 , 4-pentanediol ( MPD ) , and 12 mM spermine tetrahydrochloride and incubated at 30°C . To produce the G96A crystals , 150 μM RNA in 10 mM MgCl2 , 10 mM KCl , 10 mM HEPES-KOH pH 7 . 5 , and 1 mM ppGpp was mixed with a solution of 80 mM sodium chloride , 40 mM sodium cacodylate pH 7 . 0 , 30% MPD , and 12 mM spermine ( 1 μL RNA solution plus 0 . 8 μL reagent ) and incubated at 23°C . Crystals were flash-frozen without further preparation . For the wild type aptamer , a solution was generated using molecular replacement with the ykkC guanidine riboswitch as an initial model ( PDB ID: 5T83 ) ( Reiss et al . , 2017 ) . For the G96A mutant , a solution was generated using molecular replacement with chain A of the PRPP riboswitch structure presented in this study as an initial model . Data were processed using HKL-2000 ( Otwinowski and Minor , 1997 ) . Model building was performed in Coot ( Emsley and Cowtan , 2004 ) . The wild type aptamer crystallized in space group P21 with two molecules present in the asymmetric unit . Discussion is for the most part limited to chain A as there is better structural information for this entity than for chain B . The first component modeled was the RNA . Further unaccounted-for electron density was assigned to metal ions and water molecules . This process yielded a structure in which one significant area of electron density in each chain was unaccounted for . One molecule of PRPP and its two associated metal ions fit well in this area of density . The G96A aptamer crystallized in space group P1 with four molecules in the asymmetric unit . Discussion in the manuscript is limited to chains A-C , due to chain D yielding generally poorer density . Overall , chain D is consistent with chains A-C , but more subject to error in individual atom positions . Regions disagreeing with the wild type ( mainly in the S-turn ) were deleted and re-modeled . Very large σ peaks in the difference Fourier map , corresponding to the very electron dense pyrophosphate moieties of ppGpp , were used to identify the ppGpp binding pocket . Refinement of the two structures was performed with Refmac and Phenix ( Adams et al . , 2010; Winn et al . , 2011 ) . Refinement was concluded when no more entities could be modeled into the electron density and computational refinement ceased to produce improvements in Rwork and Rfree . Metal ions were identified by first modeling a magnesium ion and evaluating coordination geometry , B factors , and unaccounted-for density using difference Fourier methods , followed by reassignment where appropriate . The figures of the crystal structure were made in PyMOL ( Schrödinger , n . d . ) . The ligand interaction map was made in ChemDraw . ( β-33P ) PRPP was synthesized using E . coli ribose-phosphate pyrophosphokinase ( RPPK ) obtained from Abbexa . 17 . 7 μg/mL RPPK was incubated at 37°C for two hours in the presence of 50 mM potassium phosphate dibasic pH 8 , 10 mM ribose 5-phosphate , 5 mM MgCl2 , and trace quantities of ( γ-33P ) ATP essentially as in Switzer and Gibson ( Switzer and Gibson , 1978 ) . PRPP and ATP were separated on a native 20% acrylamide gel at 4°C . PRPP was distinguished from the substrate by its faster rate of migration and eluted overnight in 400 μL dH2O at 4°C . SAS1 enzyme was expressed and purified based on the protocol in Steinchen et al . ( Steinchen et al . , 2015 ) . Briefly , the SAS1 protein from B . subtilis was amplified by colony PCR , cloned into a pET-28aM vector , and transformed into E . coli BL21 ( DE3 ) cells . A 15 mL starter culture was used to inoculate 1 . 5 L Terrific Broth plus 50 μg/mL kanamycin and grown at 37°C . At OD600 ~0 . 8 , expression was induced with 0 . 5 mM IPTG and the culture was shaken overnight at 18°C . Cells were then pelleted and lysed using a microfluidizer ( lysis buffer: 50 mM Tris , pH 8 . 0 , 300 mM NaCl , 20 mM imidazole , 20 mM MgCl2 , 20 mM KCl ) and the lysate was run on a nickel column . The protein was eluted from the column with 400 mM imidazole ( elution buffer: 50 mM Tris , pH 8 . 0 , 400 mM NaCl , 400 mM imidazole , 0 . 2 mM TCEP ) . A band running between 25 and 30 kDa was seen on an SDS-PAGE gel , indicating SAS1 was successfully eluted . The eluted protein was diluted in 50 mM Tris , pH 8 . 0 and run on a Q column ( HiTrap Q column , 5 mL ) to remove contaminants . Finally , the Q column fractions were pooled and run on a gel filtration column ( Superdex 200 , running buffer: 20 mM HEPES-Na , pH 7 . 5 , 200 mM NaCl , 20 mM KCl , 20 mM MgCl2 ) , and a peak eluted consistent with the relevant tetrameric assembly of the protein . The protein was concentrated and frozen at −80°C in aliquots for storage . The SAS1 protein accepts GDP ( or GTP ) and ATP as substrates and catalyzes the transfer of the β and γ phosphates from ATP onto the 3′ end of GDP or GTP to form ppGpp or pppGpp , respectively . To make unlabeled ppGpp for crystallography , a reaction setup based on the protocol of Steinchen et al ( Steinchen et al . , 2015 ) was used . Briefly , 5 mM GDP , 5 mM ATP , and 5 μM SAS1 were combined in reaction buffer ( 100 mM HEPES-Na , pH 7 . 5 , 200 mM NaCl , 20 mM MgCl2 , and 20 mM KCl ) and incubated at 37°C for two hours . A chloroform extraction was performed to remove SAS1 , followed by 10-fold dilution in ddH2O and purification by Q column ( HiTrap Q HP , 5 mL column volume ) , where buffer A ( 10 mM HEPES-KOH , pH 7 . 5 ) was used to bind nucleotides to the column and a gradient of buffer B ( 2 M NaCl ) was used to elute the nucleotides . Nucleotides eluted from the column such that the number of phosphates positively correlated with %B . ppGpp eluted last at ~15% buffer B ( approximately 300 mM NaCl ) ( Figure 6—figure supplement 3 ) . ppGpp was then precipitated by lithium chloride ( LiCl ) precipitation . Eluate from the Q column was brought to 1 M LiCl , 4 volumes of ethanol were added , and the tubes were frozen at −20°C before centrifuging at 6000 rpm in an Eppendorf F-45-18-11 fixed-angle centrifuge rotor at 4°C for 10 min to pellet the precipitate . The supernatant was discarded and the pellet was washed twice with cold ( −20°C ) ethanol , repeating the freezing and pelleting steps between each wash step . After pouring off the ethanol of the final wash , pellets were completely dried . A dry , white powder resulted . Concentration was calculated by measuring UV absorbance at 252 nm ( ε252 = 13600 L mol−1 cm−1 ) . A reaction mixture resembling that in the previous section was made , substituting 5 mM ATP for 150 μCi [γ-32P]-ATP ( Perkin Elmer ) . ( 3′-β-32P ) -ppGpp was purified using a 20% denaturing polyacrylamide gel to separate it from [γ-32P]-ATP . The band was soaked in 300 μL ddH2O overnight at 4°C . The gel slice was then filtered off and the solution containing ( 3′-β-32P ) -ppGpp was frozen at −20°C for use in binding assays . The dissociation constants of the PRPP-RNA and ppGpp-RNA complexes were determined by equilibrium dialysis using cassettes with a 10 kDa cutoff obtained from Harvard Apparatus , essentially as in Reiss et al ( Reiss et al . , 2017 ) . Trace quantities of radiolabeled ligand were dissolved in equilibrium dialysis buffer ( 50 mM HEPES-KOH , pH 7 . 5 , 200 mM KCl , 20 mM MgCl2 ) and were added to one side of the cassette , while varying concentrations of RNA dissolved in the same buffer were added to the other side of the cassette . The cassettes were incubated at room temperature overnight with gentle shaking and recovered by centrifugation . For ppGpp , which experiences negligible amounts of degradation overnight , 20 μL of the recovered material was directly subjected to scintillation counting . For PRPP , 10 μL of the recovered material was subjected to scintillation counting , while another 10 μL was electrophoresed on a denaturing 20% acrylamide gel containing 7 . 5 M urea . The latter step allowed determination of the amount of PRPP remaining in each sample after overnight incubation at room temperature in the presence of magnesium . The fraction of ligand bound in each cassette was determined using the following equation:F= ( CPMR∗PR ) − ( CPML∗PL ) CPMR∗PR Where F is the fraction of PRPP bound . CPML and CPMR are the counts per minute measured via scintillation counting of the ligand and RNA sides of the cassette , respectively . PL and PR are the percentages of intact PRPP remaining as determined by gel electrophoresis for the ligand and RNA sides of the cassette , respectively . Fitting was performed in GraphPad Prism using the following equation:F=Fmax∗[RNA]Kd+[RNA] Where F is the fraction of PRPP bound , Fmax is the maximum fraction of PRPP bound , [RNA] is the concentration of RNA , and Kd is the dissociation constant . All binding data consist of three technical replicates , of which the arithmetic mean and standard deviation are represented in Figure 1—figure supplement 2 . All replicates were performed using a single stock of RNA from the same round of in vitro transcription and purification . Each data point in each replicate was collected using a different equilibrium dialysis cassette with independently diluted RNA solutions . Data were fit to a single-binding hyperbolic curve , with Bmax floating or constrained as follows . TmaWT binding to PRPP and TmaG96A binding to ppGpp fully reached saturation and Bmax was allowed to float . These Bmax values fit here should represent the fraction of radiolabeled ligand available for binding . TmaWT binding to ppGpp and TmaG96A binding to PRPP did not reach saturation , so the Bmax was constrained to equal the Bmax values for TmaG96A binding to ppGpp and TmaWT binding to PRPP , respectively . Values shown in Supplementary file 1 are the dissociation constant ( Kd ) from the hyperbolic fit plus or minus the standard error of the fit calculated by GraphPad Prism . Secondary structure predictions were obtained from mFold using the default settings ( Zuker , 2003 ) . NCBI Reference Sequence: NC_014209 . 1 , location 657606 to 657788 was used to predict the secondary structure of the aptamers in the presence of guanine and absence of PRPP . Location 657617 to 657800 was used to predict the secondary structures of the aptamers in the absence of guanine , in the presence of PRPP , and in the presence of both ligands . The secondary structure of the transcription terminator was predicted using location 657606 to 657841 . The secondary structure deemed most likely by the program was used for interpretation . Bound guanine aptamer and bound PRPP aptamer secondary structure predictions are consistent with X-ray crystallography data . Coordinates have been deposited in the Protein Data Bank ( PDB ) with accession numbers 6CK5 for the wild-type PRPP aptamer and 6CK4 for the G96A ppGpp aptamer .	DNA’s iconic double helix has made it possibly the most widely recognized biological molecule . The closely related RNA , however , is less well known but just as vital . In contrast with DNA’s typical rigid structure , RNA is more flexible and can fold into a wide range of shapes; this allows RNA molecules to have many jobs . Some RNA molecules form structures called riboswitches . As the name suggests , these act as molecular switches that help cells to respond to the presence of important small molecules . When a riboswitch encounters the right molecule , it changes shape , which in turn changes how the cell behaves . It is very difficult , if not impossible , to predict how a riboswitch recognizes its preferred small molecule . To address this , scientists use a technique called X-ray crystallography to directly examine the riboswitch’s structure . Knappenberger , Reiss and Strobel have now determined the structures of two recently discovered riboswitches . The two switches detect molecules called PRPP and ppGpp , respectively . These riboswitches are structurally similar to one that binds to a very different type of chemical called guanidine . The aim was to understand how similar switches respond to different signals . The results reveal that a PRPP riboswitch could become a ppGpp riboswitch just by making a single change to the RNA sequence . Many scientists believe RNA preceded DNA and proteins in some of the earliest organisms on Earth . Understanding how RNAs have evolved and diversified could thus help to understand how early life developed . The results may also help to design synthetic riboswitches for a variety of uses . Since many riboswitches are unique to bacteria , this work could also contribute to the search for new antibiotics .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2018	Structures of two aptamers with differing ligand specificity reveal ruggedness in the functional landscape of RNA
During the development , tight regulation of the expansion of neural progenitor cells ( NPCs ) and their differentiation into neurons is crucial for normal cortical formation and function . In this study , we demonstrate that microRNA ( miR ) -128 regulates the proliferation and differentiation of NPCs by repressing pericentriolar material 1 ( PCM1 ) . Specifically , overexpression of miR-128 reduced NPC proliferation but promoted NPC differentiation into neurons both in vivo and in vitro . In contrast , the reduction of endogenous miR-128 elicited the opposite effects . Overexpression of miR-128 suppressed the translation of PCM1 , and knockdown of endogenous PCM1 phenocopied the observed effects of miR-128 overexpression . Furthermore , concomitant overexpression of PCM1 and miR-128 in NPCs rescued the phenotype associated with miR-128 overexpression , enhancing neurogenesis but inhibiting proliferation , both in vitro and in utero . Taken together , these results demonstrate a novel mechanism by which miR-128 regulates the proliferation and differentiation of NPCs in the developing neocortex . Neurogenesis , the process by which functionally integrated neurons are generated from neural progenitor cells ( NPCs ) , involves the proliferation and neuronal fate specification of NPCs and the subsequent maturation and functional integration of the neuronal progeny into neuronal circuits ( Gupta et al . , 2002 ) . Given its importance in the development of the nervous system , neurogenesis is tightly regulated at many levels by both extrinsic and intrinsic factors ( Heng et al . , 2010 ) , and its disruption has been associated with various pathologies , including autism spectrum disorders ( ASDs ) , Treacher Collins syndrome , and various neural tube defects ( Sun and Hevner , 2014 ) . Therefore , uncovering the molecular mechanisms that underlie neurogenesis is crucial to understand the functions and plasticity of brain development and to prevent such pathologies ( Sun and Hevner , 2014 ) . MicroRNAs ( miRNAs ) are small noncoding RNA molecules that function in the transcriptional and post-transcriptional regulation of gene expression in a variety of organisms ( Kawahara et al . , 2012 ) . Encoded by eukaryotic nuclear DNA , miRNAs function through imperfect base-pairing with complementary sequences in target mRNA molecules , typically triggering their targeted degradation or translational repression by the RNA-induced silencing complex RISC ( Bartel , 2009; Shi et al . , 2010 ) . The role of miRNAs in neuronal development and function has recently received increased attention , and the specific spatiotemporal expression of these molecules may be essential for brain morphogenesis and neurogenesis ( Volvert et al . , 2012; Zhang et al . , 2014 ) . However , the specific miRNAs that regulate the proliferation and differentiation of NPCs during early cortical development are not well established . In this study , we demonstrate that miR-128 , which has previously been shown to play a crucial role in cortical migration ( Franzoni et al . , 2015 ) , inhibits self-renewal and promotes neuronal differentiation in mouse NPCs by targeting pericentriolar material 1 ( PCM1 ) , a critical protein for cell division , during early cortical development ( Ge et al . , 2010 ) . Specifically , ectopic overexpression of miR-128 reduces the proliferation of NPCs but promotes the differentiation of NPCs into neurons both in vivo and in vitro . Conversely , knockdown of miR-128 enhances proliferation but inhibits neuronal differentiation . Knockdown of endogenous PCM1 mimics the cellular phenotype of miR-128-overexpressing NPCs . Furthermore , the concomitant overexpression of both PCM1 and miR-128 in NPCs rescues the cellular phenotype associated with the overexpression of miR-128 in NPCs , indicating that PCM1 lies downstream of miR-128 and regulates the proliferation and neural fate specification of NPCs in vitro and in utero . Taken together , our data indicate that miR-128 is an important regulator of neurogenesis in the embryonic cortex and suggest that aberrant miR-128 expression may account for the abnormal cortical development that underlies the pathophysiology of certain neuropsychiatric disorders , including autism . To determine the spatial distribution of miR-128 in the developing embryonic cortex , we performed in situ hybridization ( ISH ) using digoxigenin ( DIG ) -labeled locked nucleic acid ( LNA ) detection probes targeted to the mature form of miR-128 ( Figure 1A ) . As previously reported ( Tan et al . , 2013 ) , miR-128 was found to be predominantly expressed in the brains and spinal cords of E14 . 5 mice , and no signal was detected with a scrambled miRNA probe ( Figure 1A ) . As an alternative method , we performed quantitative real-time PCR ( qPCR ) and found spatial expression patterns of miR-128 that were similar to those observed using ISH ( Figure 1B ) . Within the E14 . 5 forebrain , miR-128 was clearly detectable in the cortical layers , and high-magnification images of cortical slices at E14 . 5 revealed the expression of miR-128 in cells within the ventricular/subventricular zone ( VZ/SVZ ) ( Figure 1C ) . To further confirm this , we performed fluorescence ISH in combination with immunostaining using the NPC marker NESTIN in cortical slices at E14 . 5 and found that NPCs within the VZ/SVZ expressed miR-128 ( Figure 1—figure supplement 1 ) . Furthermore , NPCs isolated from the E14 . 5 forebrain co-expressed miR-128 and NESTIN ( Figure 1D ) , indicating the potential functional role of miR-128 in regulating the proliferation and/or differentiation of NPCs . 10 . 7554/eLife . 11324 . 003Figure 1 . miR-128 expression in the developing cerebral cortex . ( A ) ISH was performed in an E17 . 5 mouse embryo with a miR-128 LNA detection probe . A sagittal section shows that miR-128 is expressed in the CNS . The left-side section was probed with a miR scramble control . Scale bars , 1 mm . ( B ) Real-time qPCR analyses of miR-128 from various tissues at E17 . 5 , showing that miR-128 is highly expressed in the brain and spinal cord . miR-128 expression was measured and normalized to that of U6 and is shown relative to the liver expression level . The values represent the mean ± s . d . ( n = 3 ) . ( C ) miR-128 expression in the cortex . ISH was performed in an E14 . 5 mouse embryo brain coronal section with a miR-128 LNA detection probe . The left-side section was probed with a miR scramble control . Scale bars , 50 µm , and 5 µm in the higher magnification image ( right-most panel ) . ( D ) miR-128 expression in NPCs in vitro . miR-128 LNA in situ hybridization followed by immunofluorescence analysis of the neural stem cell marker NESTIN in NPCs . DAPI staining indicates the location of cell nuclei . Scale bars , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 00310 . 7554/eLife . 11324 . 004Figure 1—figure supplement 1 . miR-128 expression in the developing CNS . miR-128 expression in the NPCs of the VZ at E14 . 5 . miR-128 LNA in situ hybridization followed by immunofluorescence analysis of the neural stem cell marker NESTIN on E14 . 5 mouse embryo brain coronal sections . The micrographs in the bottom panels show slices probed with a miR scramble control . White dotted line outlines a single cell within the VZ . DAPI staining indicates the location of cell nuclei . Scale bars , 5 µm in left-most panel , 2 µm in all other panels . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 004 To examine whether miR-128 regulates the proliferation and/or differentiation of embryonic NPCs , we designed constructs for gain-of-function and loss-of-function studies . A dual-promoter expression construct carrying CMV-driven mouse miR-128-1 precursor and EF1α-driven copGFP ( Hollis et al . , 2009 ) ( miR-128 ) was used to overexpress miR-128 , while a dual-promoter expression construct containing H1-driven anti-sense miR-128 short-hairpin RNA and CMV-driven copGFP ( miR-Zip-128 ) ( Guibinga et al . , 2012 ) was used to knock down endogenous miR-128 ( Figure 2—figure supplement 1A and B ) . When we transfected primary embryonic NPCs that were isolated from E14 . 5 mouse forebrains with these constructs , the miR-128-1 precursor was efficiently processed to generate the mature miRNA , as shown by a 15-fold upregulation of miR-128 in transfected cells ( Figure 2—figure supplement 1A ) . In addition , miR-Zip-128 transfection resulted in efficient knockdown of endogenous miR-128 ( Figure 2—figure supplement 1B ) . NPCs that were isolated from E14 . 5 mouse forebrains were electroporated with the aforementioned constructs and were pulse-labeled with 5-bromo-2’-deoxyuridine ( BrdU ) for 6 hr to label dividing cells of a heterogeneous group of NPCs from all phases of the cell cycle ( Bez et al . , 2003; Zhang et al . , 2014 ) ( Figure 2A ) . Overexpression of miR-128 inhibited NPC proliferation , as indicated by a 56% reduction in BrdU incorporation in transfected cells ( arrowheads ) compared with the incorporation in those transfected with the miRNA mimic control ( Figure 2B and C ) . Conversely , miR-128 knockdown led to a 50% increase in the number of GFP-BrdU double-positive cells compared to the scramble control ( Figure 2D and E ) . 10 . 7554/eLife . 11324 . 005Figure 2 . miR-128 modulates the proliferation and differentiation of NPCs in vitro . ( A ) Schematic representation of the cell proliferation assay procedures . ( B-E ) Ectopic expression of miR-128 decreases proliferation , while that of miR-Zip-128 increases NPC proliferation . NPCs were electroporated with the indicated plasmids and pulse-labeled with BrdU for 6 hr . NPCs were immunostained with an antibody against BrdU . The arrowheads indicate BrdU and GFP double-positive cells . Scale bars , 10 µm . Quantification of the number of GFP-BrdU double-positive cells relative to the total number of GFP-positive cells ( C , E ) . ( F ) Schematic representation of the cell differentiation assay procedures . ( G-J ) Ectopic expression of miR-128 increases neurogenesis , while that of miR-Zip-128 decreases the neurogenesis of NPCs . NPCs were electroporated with the indicated plasmids and immunostained with antibodies against TUJ1 . The arrowheads indicate TUJ1+GFP+ cells . Scale bars , 50 µm . Quantification of the number of the GFP-TUJ1 double-positive cells relative to the number of GFP-positive cells ( H , J ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at *p<0 . 001 and p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 00510 . 7554/eLife . 11324 . 006Figure 2—figure supplement 1 . miR-128 overexpression and miR-128 knockdown constructs and their expression efficiency . ( A ) Premature miR-128-1 ( pre-miR-128 ) was cloned into a dual expression construct . The corresponding control construct contained a random miR mimic sequence . ( B ) qPCR quantification of miR-128 levels in NPCs following transfection with the miR-128 constructs . ( C ) A shRNA sequence targeting mature miR-128 ( anti-miR-128 ) was cloned into a dual expression construct . The corresponding control construct contained a scrambled shRNA sequence . ( D ) qPCR quantification of miR-128 levels in NPCs following transfection with the miR-Zip-128 constructs . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 00610 . 7554/eLife . 11324 . 007Figure 2—figure supplement 2 . miR-128 overexpression and knockdown does not affect NPC apoptosis . ( A-D ) Ectopic expression of miR-128 and miR-Zip-128 does not affect TUNEL staining . Quantification of the number of GFP-TUNEL double-positive cells relative to the total number of GFP-positive cells ( B , D ) . ( E-H ) Ectopic expression of miR-128 and miR-Zip-128 does not affect cleaved caspase-3 ( C-caspase 3 ) staining . Scale bars , 20 μm . Quantification of the number of GFP-C-caspase 3 double-positive cells relative to the total number of GFP-positive cells ( B , D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 00710 . 7554/eLife . 11324 . 008Figure 2—figure supplement 3 . Ectopic expression of miR-128 increases neurogenesis , while that of miR-Zip-128 decreases neurogenesis . ( A-D ) NPCs were electroporated with the indicated plasmids and immunostained with antibodies against MAP2 . The arrowheads indicate MAP2+GFP+ cells . Scale bars , 50 µm . Quantification of the number of GFP-MAP2 double-positive cells relative to the number of GFP-positive cells ( B , D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 00810 . 7554/eLife . 11324 . 009Figure 2—figure supplement 4 . Lentivirus-mediated transduction of miR-128 and miR-Zip-128 increases and decreases neurogenesis , respectively . ( A-D ) NPCs were transduced with lentiviruses carrying the indicated plasmids and immunostained with antibodies against TUJ1 . The arrowheads indicate TUJ1+GFP+ cells . Scale bars , 50 µm . Quantification of the number of GFP-TUJ1 double-positive cells relative to the number of GFP-positive cells ( B , D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 00910 . 7554/eLife . 11324 . 010Figure 2—figure supplement 5 . Ectopic expression of miR-128 decreases gliogenesis , while that of miR-Zip-128 increases gliogenesis . ( A-D ) NPCs were transduced with lentiviruses carrying the indicated plasmids and immunostained with antibodies against GFAP . The arrowheads indicate GFAP+GFP+ cells . Scale bars , 50 µm . Quantification of the number of GFP-GFAP double-positive cells relative to the number of GFP-positive cells ( B , D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 01010 . 7554/eLife . 11324 . 011Figure 2—figure supplement 6 . Optimizing the duration of BrdU pulse labeling for in vitro NPC proliferation assay . ( A ) NPCs were pulsed with BrdU for the indicated duration before fixation . Quantification of BrdU-positive cells over total DAPI-positive cells . ( B ) Following electroporation with either miR-128-ZIP or the control constructs NPCs were pulsed with BrdU for the indicated duration . Quantification of BrdU- GFP-double-positive cells over GFP-positive cells shows that after 4 hr of BrdU exposure , NPCs electroporated with miR-128-ZIP show significantly increased BrdU incorporation compared to control . At least 1500 cells were counted for each condition . The values represent the means ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 001 , p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 011 To determine whether modulating the levels of miR-128 affected cell death , we performed a TUNEL assay ( Cai et al . , 2000 ) ( Figure 2—figure supplement 2A–D ) and antibody staining against cleaved caspase-3 ( Chen and Dong , 2009 ) ( Figure 2—figure supplement 2E–H ) . These results showed that upregulation or downregulation of the miR-128 level specifically affected cell proliferation without affecting apoptosis . Next , to determine the fate of the NPCs following cell-cycle exit , NPCs were transfected with the aforementioned constructs and were subsequently induced to differentiate in vitro by withdrawing growth factors from the culture medium for 5–6 days ( Ma et al . , 2008; Zhang et al . , 2014 ) ( Figure 2F ) . Neuronal differentiation was assayed by immunostaining with TUJ1 , a specific antibody against beta-III-tubulin ( Ferreira and Caceres , 1992 ) . An increase in the number of GFP and TUJ1 double-positive cells upon treatment would indicate increased neuronal differentiation , whereas a decrease would indicate the inhibition of neurogenesis . Intriguingly , miR-128 overexpression increased the neuronal differentiation of NPCs , as shown by a significant increase ( of approximately 100% ) in the number of GFP and TUJ1 double-positive cells ( Figure 2G and H ) . In contrast , miR-128 knockdown resulted in a 40% decrease of GFP-TUJ1 double-positive cells compared with treatment with the scrambled control miRNA ( Figure 2I and J ) . When we assayed neuronal differentiation using immunostaining with the marker of mature neurons MAP2 , we observed a similar effect on neuronal differentiation ( Figure 2—figure supplement 3A–D ) . Moreover , lentiviral transduction of miR-128 or miR-Zip-128 , as an alternative gene delivery approach , resulted in similar effects on neurogenesis ( Figure 2—figure supplement 4A–D ) . Taken together , these results indicate that miR-128 overexpression enhances neuronal differentiation of NPCs following cell-cycle exit , while miR-128 knockdown shows the opposite effect on neuronal differentiation . To further test whether the enhanced differentiation was restricted to a neuronal fate , we immunostained NPCs with a specific antibody against GFAP , a marker for glial cells , and observed that miR-128 overexpression significantly decreased the number of GFP and GFAP double-positive cells , indicating that NPCs that had exited the cell cycle following miR-128 overexpression may have over-committed to neuronal differentiation , preventing NPCs from becoming glial cells . The opposite effect was observed upon miR-128 knockdown in NPCs ( Figure 2—figure supplement 5A–D ) . To rule out the potential bias in the in vitro cell proliferation assay we monitored the time-course of BrdU incorporation using control NPCs ( Figure 2—figure supplement 6A ) as well NPCs electroporated with miR-ZIP-128 . We found that the effects of miR-128 knockdown in NPCs ( increased cell proliferation compared to control ) remained unchanged at most of the time points ( Figure 2—figure supplement 6B ) . During development , NPCs in the VZ maintain proliferative capacity and the ability to self-renew ( apical progenitors , APs ) . These APs give rise to intermediate progenitors ( basal progenitors , BPs ) within the SVZ and IZ that can proliferate and become neurons ( Martinez-Cerdeno et al . , 2006; Pontious et al . , 2008 ) . To examine the role of miR-128 during neocortical development in vivo , we introduced miR-128 and miR-Zip-128 into the lateral ventricular wall of E13 . 5 mouse brains by in utero electroporation and analyzed the electroporated brains at E14 . 5 . First , to detect changes in NPC proliferation , we monitored the mitotic spindle orientation ( Huttner and Kosodo , 2005; Wang et al . , 2011 ) of APs within the VZ that were undergoing mitosis using an antibody against phosphorylated histone H3 ( PH3 ) ( Postiglione et al . , 2011 ) , which labels dividing nuclei ( Figure 3A ) . We found a significant decrease in the percentage of horizontal divisions upon miR-128 overexpression ( Figure 3A and B , 10% ± SD ) , while miR-128 knockdown led to a significant increase in this percentage ( Figure 3A and C , 20% ± SD ) . Based on these observations , we performed further experiments to identify the fate of AP progeny following miR-128 overexpression and miR-128 knockdown . 10 . 7554/eLife . 11324 . 012Figure 3 . miR-128 regulates the proliferation and differentiation of NPCs in vivo . ( A-C ) miR-128 regulates the symmetric division of apical progenitors in the VZ/SVZ . Mouse embryos were electroporated at E13 . 5 with the indicated plasmids and sacrificed at E14 . 5 . The nuclei of mitotic cells were labeled using antibodies against PH3 . The orientation of the mitotic spindle relative to the ventricular surface was determined and categorized into horizontal ( 0–30° ) , oblique ( 30–60° ) , or vertical ( 60–90° ) as pictured . Scale bars , 5 µm . ( A ) . Percentage of GFP and PH3 double-positive cells with the indicated mitotic spindle orientation following miR-128 overexpression ( B ) and miR-128 knockdown ( C ) . ( D-G ) miR-128 regulates the proliferation of apical progenitors in the VZ/SVZ . Mouse embryos were electroporated at E13 . 5 with the indicated constructs . Twenty-four hours post-electroporation , dividing cells were marked by EdU pulse-labeling for 2 hr and sacrificed . The arrowheads indicate EdU+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-EdU double-positive cells relative to the number of GFP-positive cells ( E , G ) . ( H–K ) Ectopic expression of miR-128 decreases apical progenitors , while that of miR-Zip-128 increases apical progenitors . Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E14 . 5 . Brain sections were immunostained with antibodies against PAX6 . The arrowheads indicate PAX6+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-PAX6 double-positive cells relative to the number of GFP-positive cells ( I , K ) . ( L–O ) Ectopic expression of miR-128 increases basal progenitors , while that of miR-Zip-128 decreases basal progenitors . Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E14 . 5 . Brain sections were immunostained with antibodies against TBR2 . The arrowheads indicate TBR2+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-TBR2 double-positive cells relative to the number of GFP-positive cells ( M , O ) . ( P-S ) miR-128 promotes cell cycle exit , and miR-Zip-128 inhibits cell cycle exit . Mouse embryos were electroporated at E13 . 5 with the indicated constructs . Twenty-four hours post-electroporation , dividing cells were marked by EdU pulse-labeling for 24 hr and sacrificed . The brain sections were immunostained with antibodies against Ki67 . The blue arrowheads indicate GFP+EdU+Ki67- cells that had exited the cell cycle . Scale bars , 10 µm . Quantification of the number of GFP-EdU double-positive , Ki67-negative cells relative to the number of GFP-BrdU double-positive cells ( Q , S ) . ( T-W ) miR-128 promotes the differentiation of NPCs in vivo . Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E17 . 5 . Brain slices were immunostained for NEUN . Arrowheads indicate NEUN+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-NEUN double-positive cells relative to the number of GFP-positive cells ( U , W ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 001 , p<0 . 01 , and p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 01210 . 7554/eLife . 11324 . 013Figure 3—figure supplement 1 . Ectopic expression of miR-128 decreases proliferation , while that of miR-Zip-128 increases proliferation in vivo . ( A-D ) Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E14 . 5 . Brain sections were immunostained with antibodies against Ki67 . The arrowheads indicate Ki67+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-Ki67 double-positive cells relative to the number of GFP-positive cells ( B , D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 01310 . 7554/eLife . 11324 . 014Figure 3—figure supplement 2 . Ectopic expression of miR-128 decreases apical progenitors , while that of miR-Zip-128 increases apical progenitors . ( A-D ) Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E14 . 5 . Brain sections were immunostained with antibodies against SOX2 . The arrowheads indicate SOX2+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-SOX2 double-positive cells relative to the number of GFP-positive cells ( B , D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 001 and p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 01410 . 7554/eLife . 11324 . 015Figure 3—figure supplement 3 . Model of the miR-128 in vivo phenotype . miR-128 overexpression in embryonic neural stem cells leads to two phenotypes that when combined together , ultimately results in amplified neurogenesis . First , an increase in the asymmetric cell division of apical progenitors occurs , which results in a relatively higher number of basal progenitors and a lower number of apical progenitors . Second , an increase in cell cycle exit occurs in the expanded basal progenitor pool , which leads to a greater number of neurons . miR-128 knockdown results in the opposite phenotypes , namely an increase in the symmetric cell division of apical progenitors , resulting in a relatively higher number of apical progenitors and a lower number of basal progenitors . In addition , the basal progenitors do not exit the cell cycle as readily , ultimately leading to a smaller number of neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 015 Furthermore , miR-128 overexpression led to a marked decrease in the percentage of cells that were positively labeled for the incorporation of 5-ethynyl-2'-deoxyuridine ( EdU ) ( Ishino et al . , 2014 ) ( ~12% ) ( Figure 3D and E ) , reduced cell division , as indicated by immunostaining with Ki67 ( Figure 3—figure supplement 1A and B ) , and a marked decrease in the number of cells that were positive for the AP marker PAX6 ( 28% reduction in the number of VZ/SVZ cells that were positive for both GFP and PAX6 ) ( Figure 3H and I ) and SOX2 ( Figure 3—figure supplement 2A and B ) . These data indicate that miR-128 overexpression decreased the number of proliferating APs within the VZ/SVZ . In contrast , miR-128 knockdown had the opposite effects on EdU incorporation and on Ki67 , PAX6 and SOX2 immunostaining in APs ( Figure 3F , G , J and K ) ( Figure 3—figure supplement 1C and D , 2C and D ) . Next , given that we observed increased numbers of obliquely dividing cells upon miR-128 overexpression ( Figure 3A and B ) , indicating potential expansion of BPs ( Huttner and Kosodo , 2005; Wang et al . , 2011 ) , we monitored TBR2 expression upon miR-128 overexpression . MiR-128 overexpression led to an increase in the number of TBR2-positive cells ( 60% increase in GFP-TBR2 double-positive cells ) ( Figure 3L and M ) , while miR-128 knockdown resulted in a decrease in the number of TBR2-positive cells ( 30% decrease in GFP-TBR2 double-positive cells ) ( Figure 3N and O ) . Taken together , these data indicate that miR-128 regulates NPC proliferation by promoting intermediate basal progenitors at the expense of apical progenitors within the VZ/SVZ . Because BPs will generate the bulk of cortical neurons ( Tan and Shi , 2013 ) , we tested whether miR-128 regulates the neuronal differentiation of NPCs in vivo . First , we analyzed the number of BPs that exited the cell cycle ( Ge et al . , 2010; Yang et al . , 2012 ) upon miR-128 manipulation . Following in utero electroporation , dividing cells were labeled with EdU for 24 hr and subsequently immunostained for Ki67 . Compared to the control treatment , miR-128 overexpression increased the number of cells that were positive for both GFP and EdU but negative for Ki67 cells ( by ~40% ) , indicating an increase the number of BPs that had exited the cell cycle ( Figure 3P and Q ) . Conversely , miR-128 knockdown had the opposite effect ( Figure 3R and S ) . To further examine whether the observed changes in cell cycle exit ultimately led to a difference in neurogenesis , we analyzed brains at E17 . 5 , four days after electroporation , and assayed neuronal differentiation by immunostaining with a specific antibody against the neuronal marker NEUN ( Zhang et al . , 2014 ) . miR-128 overexpression significantly increased ( by ~25% ) the number of NEUN-positive neurons in the CP zone compared with the number in the control condition ( Figure 3T and U ) . Moreover , miR-128 knockdown had the opposite effect on neurogenesis ( Figure 3V and W ) . Taken together , these results indicate that miR-128 may act on two different stages of NPC development: first , by regulating symmetric/asymmetric division of APs , miR-128 promotes BP production; and second , by enhancing the exit of BPs from the cell cycle , miR-128 promotes overall neurogenesis ( Figure 3—figure supplement 3 ) . In contrast , downregulation of miR-128 in early neuronal precursors impeded their developmental progression by causing them to be retained in a more primitive , proliferative stage , resulting in the expansion of a pool of the NPC pool . Intriguingly , this early expansion of NPC pools upon miR-128 knockdown did not result in a net increase in neuronal numbers at E17 . 5 , suggesting that further investigation is necessary to delineate whether the observed phenomena is due to compromised neurogenic capability of NPCs or simply due to a delay in neurogenesis which could eventually be overcome at a later postnatal stage ( Figure 3—figure supplement 3 ) . miRNAs normally regulate the translation and/or degradation of multiple target mRNAs ( Bartel , 2009 ) . To further characterize the molecular mechanisms that underlie the changes in NPC competence , we utilized two widely used in silico microRNA target prediction algorithms ( TargetScan and miRanda ) ( Mi et al . , 2013; Witkos et al . , 2011 ) , which identified 800 and 940 potential targets of miR-128 , respectively ( Figure 4—source data 1 ) . Among the 77 overlapping targets identified using two algorithms , only 53 genes were annotated to have known biological processes and thus selected for further testing ( Figure 4—source data 1 ) . qPCR analysis of cultured NPCs that overexpressed miR-128 revealed 21 genes that were downregulated ( Figure 4—source data 2 ) . Eleven out of the 53 genes tested exhibited consistent reduction in mRNA levels upon miR-128 overexpression in cultured mouse NPCs ( Pcm1 , Lmbr1l , Foxo4 , Sh2d3c , Nfia , Pde8b , Sec24a , Pde3a , Fbxl20 , Ypel3 and Kcnk10 , Figure 4—source data 2 , highlighted in yellow ) . The 11 genes were further tested for reciprocal upregulation when miR-128 was inhibited . qPCR analysis following miR-128 inhibition showed that only Pcm1 , Nfia , Foxo4 , and Fbxl20 were consistently upregulated among which Pcm1 displayed the greatest change ( Figure 4—source data 3 ) . We further validated Pcm1 as a target of miR-128 using a luciferase assay . First , we cloned the 3’-UTR of Pcm1 ( WT-Pcm1 ) into a dual-luciferase reporter construct , pmirGLO , to assess translation of the target protein based on the luciferase activity ( Krishnan et al . , 2013 ) ( Figure 4A ) . In this assay , co-transfection of miR-128 with the WT-Pcm1 reporter construct markedly suppressed the luciferase activity ( by 58% , Figure 4B ) . However , co-transfection of miR-128 with random 3’-UTR sequences ( Control , Figure 4B ) did not affect the luciferase activity . To further determine whether the targeting of PCM1 by miR-128 was specific , we introduced three mismatched nucleotides to the predicted seed region of the miR-128 binding site ( MT-Pcm1 ) ( Figure 4A , red underlines ) . Mutating these seed sequences abolished the miR-128-mediated suppression of PCM1 luciferase activity and restored the luciferase activity to the control level ( Figure 4B ) , indicating the specificity of miR-128 targeting of the 3’-UTR of Pcm1 . 10 . 7554/eLife . 11324 . 016Figure 4 . miR-128 regulates PCM1 expression in NPCs . ( A ) TargetScan analysis identified a miR-128 target site in the mouse Pcm1 3’-UTR region ( highlighted in green ) . The mutant PCM1 is shown , with the seed binding sites highlighted in red . ( B ) PCM1 luciferase activity is suppressed by miR-128 . HEK293T cells were co-transfected with miR-128 and the 3’-UTR of Pcm1 containing either the miRNA binding site ( WT ) or mutant ( MT ) versions of the Pcm1 seed binding sites for 2 days . The cells were harvested and lysed , and a luciferase activity assay was then performed . miR-128-mediated suppression of PCM1 luciferase activity was relieved upon mutation of the Pcm1 seed binding sites . ( C , D ) miR-128 overexpression in NPCs led to reduced endogenous Pcm1 mRNA levels , as determined by qPCR ( C ) , and PCM1 protein expression , as demonstrated via densitometry analysis of western blots ( D ) . ( E , F ) anti-miR-128 leads to increased endogenous Pcm1 mRNA levels , as demonstrated by qPCR ( E ) , and protein expression of PCM1 ( F ) . ( G , H ) LCM was used to isolate RNA from three specific cortical layers of E14 . 5 embryonic brains: the VZ/SVZ , IZ , and CP . qPCR quantification of miR-128 levels ( G ) and Pcm1 mRNA levels ( H ) . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 001 , p<0 . 01 , and p<0 . 05 for all panels in the figure . ANOVA , differences were considered significant at p<0 . 001 and p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 01610 . 7554/eLife . 11324 . 017Figure 4—source data 1 . Gene Ontology ( GO ) of the miR-128 target gene list . 53 genes with annotated GO terms which were predicted as targets of miR-128 by both TargetScan and miRanda algorithms , were listed . The genes were sorted in alphabetical order . GO information was obtained from the PANTHER website ( http://www . pantherdb . org ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 01710 . 7554/eLife . 11324 . 018Figure 4—source data 2 . Relative expression of 53 predicted miR-128 targets upon overexpression of miR-128 in NPCs . Cultured NPCs were transfected with either miR-128 or miR-mimic control for 48 hr and were harvested for qPCR analysis . 11 genes ( Pcm1 , Lmbr1l , Foxo4 , Sh2d3c , Nfia , Pde8b , Sec24a , Pde3a , Fbxl20 , Ypel3 and Kcnk10 ) showed consistent and significant reduction in mRNA levels upon miR-128 expression in NPCs ( highlighted in yellow ) . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 01810 . 7554/eLife . 11324 . 019Figure 4—source data 3 . Relative expression of miR-128 targets upon downregulation of miR-128 in NPCs using miR-Zip-128 . NPC primary cultures were transfected with either miR-Zip-128 or miR-Zip scramble control , and mRNAs of 11 target genes were measured via qPCR analysis . Only PCM1 was significantly upregulated upon anti-miR-128 expression in NPCs . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 01910 . 7554/eLife . 11324 . 020Figure 4—source data 4 . List of qPCR primers . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 02010 . 7554/eLife . 11324 . 021Figure 4—figure supplement 1 . Schematic diagram outlines rationale of gene selection process . Two widely used in silico microRNA target prediction databases ( TargetScan and miRanda ) predicted 800 and 940 potential targets of miR-128 respectively . Among the 77 overlapping targets only 53 were annotated to have known biological function and only 11 out of the 53 genes exhibited consistent reduction in mRNA levels upon miR-128 overexpression in cultured mouse NPCs and were further tested for reciprocal upregulation when miR-128 was inhibited . Lastly , upon miR-128 inhibition in NPCs , only Pcm1 , Nfia , Foxo4 , and Fbxl20 were consistently upregulated . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 02110 . 7554/eLife . 11324 . 022Figure 4—figure supplement 2 . miR-128 inhibitor knockdown efficiency . qPCR quantification of miR-128 levels in NPCs following transfection with 2 µg miR-128 inhibitor ( anti-miR-128 ) compared to the scramble control ( anti-miR-control ) . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 02210 . 7554/eLife . 11324 . 023Figure 4—figure supplement 3 . Inverse relationship between the temporal expression patterns of miR-128 and PCM1 . ( A ) Real-time PCR analyses of miR-128 from brain tissues at various developmental stages . miR-128 expression was measured and normalized to that of U6 and is shown relative to the E12 . 5 expression level . The values represent the mean ± s . d . ( n = 3 ) . ( B ) Western blot of PCM1 expression in brain lysates prepared at various developmental stages with densitometry analysis . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 023 Next , we examined whether miR-128 downregulated the endogenous expression of PCM1 at the mRNA and protein levels by overexpressing either miR-128 or a scrambled control in NPCs and then performing qPCR and western blot analyses using a specific antibody against PCM1 . We observed that miR-128 overexpression significantly reduced PCM1 mRNA ( Figure 4C ) and protein levels ( Figure 4D ) . Conversely , miR-128 knockdown using a specific inhibitor of miR-128 ( anti-miR-128 ) ( Smrt et al . , 2010 ) , in comparison to using a scrambled anti-miR ( anti-miR-control ) ( Figure 4—figure supplement 2 ) in NPCs produced the opposite effect on the expression of PCM1 mRNA ( Figure 4E ) and protein ( Figure 4F ) . Taken together , these data suggest that miR-128 targets PCM1 expression in NPCs , which in turn controls NPC proliferation and differentiation in vitro . In addition , qPCR analyses of tissue samples that were isolated from the VZ/SVZ , IZ , and CP using laser capture microdissection ( LCM ) ( Wang et al . , 2009 ) revealed an inverse relationship between miR-128 and PCM1 mRNA ( Figure 4G and H ) . An inverse relationship between miR-128 and PCM1 mRNA levels was also evident temporally , given that the expression of miR-128 gradually increased starting from E12 . 5 through P0 , whereas PCM1 protein expression was found to gradually decrease over this time period ( Figure 4—figure supplement 3A–B ) , suggesting that miR-128 might regulate PCM1 levels to control NPC proliferation and differentiation in the developing cortex . If the effect of miR-128 on the proliferation and differentiation of NPCs is mediated through the suppression of endogenous PCM1 , then PCM1 downregulation should mimic the cellular effects of miR-128 overexpression . To test this hypothesis , we generated two small hairpin RNA ( shRNA ) vectors that expressed shRNAs that targeted mouse PCM1 and validated the efficiency of these shRNAs in knocking down PCM1 in mouse neuroblastoma ( N2A ) cells ( Figure 5—figure supplement 1A ) . The expression of the shRNA vectors #1 and #2 led to a reduction in endogenous PCM1 of approximately 60 and 70% , respectively . Based on these data , we used shRNA #2 to examine the role of PCM1 in early neurogenesis . PCM1 has been shown to affect the proliferation and neurogenesis of NPCs; knockdown of PCM1 inhibits NPC proliferation but promotes NPC differentiation in the developing cortex ( Ge et al . , 2010 ) . To further confirm the loss of PCM1 function in NPCs , we electroporated NPCs with PCM1 shRNA and assessed the proliferation of NPCs using BrdU pulse-labeling for 6 hr . The reduction of endogenous PCM1 via shRNA led to significant inhibition of NPC proliferation , as indicated by a 35% reduction in BrdU incorporation by transfected cells compared with the BrdU incorporation by NPCs that were transfected with a control scrambled shRNA ( Figure 5A and 5B ) ; this reduction was comparable to that caused by the overexpression of miR-128 in NPCs ( Figure 2B and C ) . PCM1 knockdown did not affect cell survival , as indicated by TUNEL assay and antibody staining against activated caspase-3 ( Figure 5—figure supplement 2A–D ) . 10 . 7554/eLife . 11324 . 024Figure 5 . PCM1 knockdown decreases NPC neurogenesis . ( A , B ) PCM1 knockdown in NPCs decreases neural proliferation . NPCs were electroporated with a plasmid expressing PCM1 shRNA or with a control vector and pulse-labeled with BrdU for 6 hr . NPCs were immunostained with an antibody against BrdU . The arrowheads indicate BrdU and GFP double-positive cells . Scale bars , 10 µm . Quantification of the number of GFP-BrdU double-positive cells relative to the total number of GFP-positive cells ( B ) . ( C , D ) PCM1 knockdown in NPCs increases neurogenesis . NPCs were electroporated with a plasmid expressing PCM1 shRNA or with a control vector and then immunostained with antibodies against TUJ1 . The arrowheads indicate TUJ1+GFP+ cells . Scale bars , 50 µm . Quantification of the number of GFP-TUJ1 double-positive cells relative to the number of GFP-positive cells ( D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 02410 . 7554/eLife . 11324 . 025Figure 5—figure supplement 1 . Efficient knockdown of PCM1 . ( A ) PCM1 knockdown in N2A cells demonstrated by western blots . ( B ) Densitometry analysis of western blots presented in ( A ) . PCM1 shRNA2 was used for further proliferation and neurogenesis analysis in NPCs . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 02510 . 7554/eLife . 11324 . 026Figure 5—figure supplement 2 . PCM1 knockdown in NPCs did not trigger apoptotic cell death . ( A , B ) Ectopic expression of PCM1 shRNA does not affect TUNEL staining . Quantification of the number of GFP-TUNEL double-positive cells relative to the total number of GFP-positive cells ( B ) . ( C , D ) Ectopic expression of PCM1 shRNA does not affect cleaved caspase-3 ( C-caspase 3 ) staining . Scale bars , 20 μm . Quantification of the number of GFP-C-caspase 3 double-positive cells relative to the total number of GFP-positive cells ( D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 02610 . 7554/eLife . 11324 . 027Figure 5—figure supplement 3 . PCM1 knockdown in NPCs increases neurogenesis . ( A- , B ) NPCs were electroporated with a plasmid expressing PCM1 shRNA or with a control vector and then immunostained with antibodies against MAP2 . The arrowheads indicate MAP2+GFP+ cells . Scale bars , 50 µm . Quantification of the number of GFP-MAP2 double-positive cells relative to the number of GFP-positive cells ( B ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 02710 . 7554/eLife . 11324 . 028Figure 5—figure supplement 4 . Overexpression of PCM1 increases NPC proliferation . ( A , B ) Overexpression of PCM1 increases NPC proliferation . NPCs were electroporated with a PCM1 overexpression construct or a control vector and pulse-labeled with BrdU for 6 hr . NPCs were immunostained with an antibody against BrdU . The arrowheads indicate BrdU and GFP double-positive cells . Scale bars , 10 µm . Quantification of the number of GFP-BrdU double-positive cells relative to the total number of GFP-positive cells ( B ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 02810 . 7554/eLife . 11324 . 029Figure 5—figure supplement 5 . Overexpression of PCM1 decreases NPC neuronal differentiation . ( A , B ) Ectopic expression of PCM1 decreases TUJ1 staining . NPCs were electroporated with a PCM1 overexpression construct or a control vector and immunostained with antibodies against TUJ1 . The arrowheads indicate TUJ1+GFP+ cells . Scale bars , 50 µm . Quantification of the number of GFP-TUJ1 double-positive cells relative to the number of GFP-positive cells ( B ) . ( C , D ) Ectopic expression of PCM1 decreases MAP2 staining . NPCs were electroporated with the indicated plasmids and immunostained with antibodies against MAP2 . The arrowheads indicate MAP2+GFP+ cells . Scale bars , 50 µm . Quantification of the number of GFP-MAP2 double-positive cells relative to the number of GFP-positive cells ( D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 02910 . 7554/eLife . 11324 . 030Figure 5—figure supplement 6 . Overexpression of PCM1 in NPCs did not trigger apoptotic cell death . ( A , B ) Ectopic overexpression of PCM1 does not affect TUNEL staining . Quantification of the number of GFP-TUNEL double-positive cells relative to the total number of GFP-positive cells ( B ) . ( C , D ) Ectopic overexpression of PCM1 does not affect cleaved caspase-3 ( C-caspase 3 ) staining . Scale bars , 20 μm . Quantification of the number of GFP-C-caspase 3 double-positive cells relative to the total number of GFP-positive cells ( D ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 03010 . 7554/eLife . 11324 . 031Figure 5—figure supplement 7 . PCM1 regulates NPC proliferation and differentiation in vivo . ( A , B ) PCM1 regulates the symmetric division of apical progenitors in the VZ/SVZ . Mouse embryos were electroporated at E13 . 5 with the indicated plasmids and sacrificed at E14 . 5 . The nuclei of mitotic cells were labeled using antibodies against PH3 . The orientation of the mitotic spindle relative to the ventricular surface was determined and categorized into horizontal ( 0–30° ) , oblique ( 30–60° ) , or vertical ( 60–90° ) as pictured . Scale bars , 5 µm ( A ) . Percentage of GFP and PH3 double-positive cells with the indicated mitotic spindle orientation following PCM1 overexpression ( B ) . ( C , D ) Overexpression of PCM1 increases NPC proliferation in vivo compared to vector-only control treatment . Mouse embryos were electroporated at E13 . 5 with the indicated constructs . Twenty-four hours post-electroporation , dividing cells were marked by EdU pulse-labeling for 2 hr and sacrificed . The arrowheads indicate EdU+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-EdU double-positive cells relative to the number of GFP-positive cells ( D ) . ( E , F ) Overexpression of PCM1 increases apical progenitors compared to vector-only control treatment . Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E14 . 5 . Brain sections were immunostained with antibodies against PAX6 . The arrowheads indicate PAX6+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-PAX6 double-positive cells relative to the number of GFP-positive cells ( F ) . ( G , H ) Overexpression of PCM1 decreases the number of basal progenitors compared to vector-only control treatment . Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E14 . 5 . Brain sections were immunostained with antibodies against TBR2 . The arrowheads indicate TBR2+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-TBR2 double-positive cells relative to the number of GFP-positive cells ( H ) . ( I , J ) Overexpression of PCM1 decreases neurogenesis compared to vector-only control treatment . Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E17 . 5 . Brain slices were immunostained for NEUN . Arrowheads indicate NEUN+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-NEUN double-positive cells relative to the number of GFP-positive cells ( J ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . Student’s t-test , differences were considered significant at *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 031 Next , to determine whether knocking down endogenous PCM1 could trigger the neuronal differentiation of NPCs , NPCs were transfected with PCM1 shRNA or a scrambled control shRNA , induced to differentiate , and immunostained for TUJ1 ( Figure 5C and D ) . Similar to miR-128 overexpression , PCM1 shRNA expression significantly increased the neuronal differentiation of NPCs ( by approximately 60% compared with that of scrambled shRNA-transfected NPCs , Figure 5C and D ) . Consistent with this finding , MAP2 immunostaining revealed a similar increase in the neuronal differentiation of PCM1 shRNA-expressing NPCs ( Figure 5—figure supplement 3A and B ) . Taken together , these results indicate that inhibiting PCM1 , which is a target of miR-128 , mimics the cellular effect of miR-128 on NPC proliferation and differentiation . To examine whether PCM1 overexpression would exert the opposite effect of PCM1 knockdown in NPCs , we co-expressed PCM1 ( without its 3’-UTR ) in NPCs using a GFP expression construct and found that PCM1 overexpression increased the proliferation of NPCs compared to the vector-only control treatment ( 70% increase in BrdU and GFP double-positive cells ) ( Figure 5—figure supplement 4A and B ) . However , PCM1 overexpression decreased neurogenesis , as determined by immunostaining for TUJ1 ( 35% decrease in the number of TUJ1 and GFP double-positive cells ) ( Figure 5—figure supplement 5A and B ) and MAP2 ( 45% decrease in the number of MAP2 and GFP double-positive cells ) ( Figure 5—figure supplement 5C and D ) . TUNEL and activated caspase-3 staining demonstrated that PCM1 overexpression had no effect on apoptosis ( Figure 5—figure supplement 6 ) . We confirmed these in vitro findings in vivo by electroporating PCM1 into NPCs in the VZ of E13 . 5 mouse brains . First , we monitored the mitotic spindle orientation of APs within the VZ that were undergoing mitosis by labeling the dividing nuclei with PH3 ( Figure 5—figure supplement 7A ) . We observed a significant increase in the percentage of horizontally dividing cells upon PCM1 overexpression ( Figure 5—figure supplement 7B , 20% ± SD ) . In addition , the expression of exogenous PCM1 in vivo led to a marked increase in the number of cells that incorporated EdU ( 40% increase in EdU and GFP double-positive cells ) ( Figure 5—figure supplement 7C and D ) . Furthermore , we observed a marked increase in the number of PAX6-positive apical progenitor cells ( 60% increase in PAX6 and GFP double-positive cells ) ( Figure 5—figure supplement 7E and F ) and a decrease in the number of TBR2-positive intermediate progenitor cells in brains in which NPCs had been electroporated with PCM1 ( 40% decrease in TBR2 and GFP double-positive cells ) ( Figure 5—figure supplement 7G and H ) . We monitored the neuronal differentiation of PCM1-expressing NPCs in vivo using NEUN staining and observed a 25% decrease in the neuronal differentiation of the NPCs ( Figure 5—figure supplement 7I and J ) . To determine whether the effect of miR-128 on neuronal differentiation is mediated through PCM1 , we co-electroporated cells with miR-128 and a PCM1 expression vector lacking the 3’-UTR . A vector-only construct was used as the corresponding control construct for PCM1 overexpression . We found that PCM1 overexpression rescued the decrease in NPC proliferation ( 103% versus 62% for BrdU and GFP double-positive cells , Figure 6A and B ) , and reversed the observed increase in neuronal differentiation induced by miR-128 overexpression ( 150% versus 225% for TUJ1 and GFP double-positive cells , Figure 6C and D; 115% versus 160% for MAP2 and GFP double-positive cells; Figure 6—figure supplement 1A and B ) in vitro . 10 . 7554/eLife . 11324 . 032Figure 6 . PCM1 is a functional target of miR-128 in vitro . ( A , B ) PCM1 antagonizes the effects of miR-128 on NPC proliferation in vitro . NPCs were electroporated with a miRNA control vector or with miR-128 . Either a PCM1 expression construct or vector only control was co-electroporated in miR-128-expressing NPCs to examine the rescue effect of PCM1 in these cells . The cells were pulse-labeled with BrdU for 6 hr , and a proliferation assay was performed by immunostaining for BrdU ( A ) . Scale bars , 10 µm . ( B ) Quantification of BrdU and GFP double-positive cells , demonstrating that PCM1 overexpression in the miR-128 cells rescued the reduced proliferation in the miR-128-expressing cells . ( C , D ) miR-128 antagonizes the function of PCM1 in neurogenesis in vitro . A neuronal differentiation assay was performed by immunostaining for TUJ1 . The arrowheads indicate cells that are double-positive for TUJ1 and GFP ( C ) . Scale bars , 50 µm . ( D ) Quantification of TUJ1 and GFP double-positive cells , demonstrating that PCM1 overexpression in the miR-128-expressing cells reverses the increased neurogenesis phenotype of the miR-128-expressing cells . More than 2000 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . ANOVA , differences were considered significant at p<0 . 001 , p<0 . 01 and p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 03210 . 7554/eLife . 11324 . 033Figure 6—figure supplement 1 . PCM1 is a functional target of miR-128 in vitro . NPCs were electroporated with a miRNA control vector or with miR-128 . Either a PCM1 expression construct or vector only control was co-electroporated in miR-128-expressing NPCs to examine the rescue effect of PCM1 in these cells . A neuronal differentiation assay was performed by immunostaining for MAP2 . The arrowheads indicate cells that are double-positive for MAP2 and GFP ( A ) . Scale bars , 50 µm . ( B ) Quantification of MAP2 and GFP double-positive cells , demonstrating that PCM1 overexpression in the miR-128-expressing cells rescues the enhanced neurogenesis phenotype of the miR-128-expressing cells . More than 2000 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . ANOVA , differences were considered significant at *p<0 . 001 and p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 033 To further verify these results in vivo , we co-electroporated the PCM1 expression vector with miR-128 in utero . Similarly , we found that co-expression of PCM1 with miR-128 successfully reversed the effects of miR-128 on neural stem cell proliferation ( 100% versus 75% for EdU and GFP double-positive cells , Figure 7A and B ) , PAX6 ( 78% versus 60% for PAX6 and GFP double-positive cells , Figure 7C and D ) , and TBR2 expression ( 120% versus 160% for TBR2 and GFP double-positive cells , Figure 7E and F ) as well as on neuronal differentiation ( 100% versus 123% for NEUN-positive cells , Figure 7G and H ) . Taken together , these results strongly suggest that miR-128 regulates the proliferation and differentiation of NPCs in the murine embryonic cortex by targeting PCM1 expression though a direct interaction with its 3’UTR . 10 . 7554/eLife . 11324 . 034Figure 7 . PCM1 is a downstream target of miR-128 during NPC proliferation and differentiation in vivo . ( A , B ) PCM1 antagonizes the effects of miR-128 on NPC proliferation in vivo . Mouse embryos were electroporated at E13 . 5 with a miRNA control vector or with miR-128 . Either a PCM1 expression construct or vector only control was co-electroporated in miR-128-expressing NPCs to examine the rescue effect of PCM1 in these cells . Twenty-four hours post-electroporation , dividing cells were marked by EdU pulse-labeling for 2 hr and sacrificed . The arrowheads indicate EdU+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-EdU double-positive cells relative to the number of GFP-positive cells ( B ) . ( C , D ) PCM1 overexpression rescues the miR-128-mediated decrease in the number of apical progenitors . Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E14 . 5 . Brain sections were immunostained with antibodies against PAX6 . The arrowheads indicate PAX6+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-PAX6 double-positive cells relative to the number of GFP-positive cells ( D ) . ( E , F ) PCM1 overexpression rescues the miR-128-mediated increase in the number of basal progenitors . Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E14 . 5 . Brain sections were immunostained with antibodies against TBR2 . The arrowheads indicate TBR2+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-TBR2 double-positive cells relative to the number of GFP-positive cells ( F ) . ( G , H ) PCM1 overexpression rescues the miR-128-mediated increase in neurogenesis . Mouse embryos were electroporated at E13 . 5 with the indicated constructs and sacrificed at E17 . 5 . Brain slices were immunostained for NEUN . Arrowheads indicate NEUN+GFP+ cells . Scale bars , 10 µm . Quantification of the number of GFP-NEUN double-positive cells relative to the number of GFP-positive cells ( H ) . More than 1500 GFP-positive cells were counted for each condition . At least three sets of independent experiments were performed . The values represent the mean ± s . d . ( n = 3 ) . ANOVA , differences were considered significant at *p<0 . 001 and p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 11324 . 034 The cerebral cortex , which is the most complex structure of the brain , is responsible for cognitive , motor and perceptual behaviors ( Volvert et al . , 2012 ) . The generation of cortical neurons depends on NPCs exiting the cell cycle , migrating , and subsequently partially maturing into neurons . These processes are orchestrated by multiple gene products that ultimately converge on the cytoskeleton to support morphological remodeling ( Sun and Hevner , 2014 ) . Therefore , for all of these processes to be correctly executed , the timing and abundance of specific gene products must be precisely regulated , and a lack of regulation may result in the cortical malformations that are associated with certain neuropsychiatric disorders ( Zhang et al . , 2014 ) . miRNAs are abundant , short-lived double-stranded RNAs of ~20-25 nucleotides that are derived from endogenous short-hairpin transcripts ( Bartel , 2009 ) . miRNAs contribute to various developmental processes by acting as post-transcriptional regulators; thus , they introduce an additional level of intricacy to gene regulation in neurogenesis . Recent data obtained by several groups support a primary role of miRNAs in fine-tuning signaling pathways that control the synchronized steps of cortical development ( Kawahara et al . , 2012; Shi et al . , 2010 ) . Abu-Elneel et al . investigated the expression of 466 human miRNAs from postmortem cerebellar cortical tissue from individuals with ASDs and identified twenty-eight miRNAs that were expressed at significantly different levels in the ASD brains compared with the non-autism control brains ( Abu-Elneel et al . , 2008 ) . Interestingly , of these dysegulated miRNAs , only three of them ( miR-7 , miR-128 , and miR-132 ) were exclusively expressed in the brain ( Li and Jin , 2010 ) . miR-128 is transcribed from two distinct loci , miR-128-1 and miR-128-2 , as two primary transcripts that are processed into identical mature miRNA sequences ( Adlakha and Saini , 2014 ) . miR-128-1 and miR-128-2 reside in the intronic regions of genes on two different chromosomes ( Tan et al . , 2013 ) . Previously , downregulation of miR-128 has been reported in several brain cancers , including glioblastoma and medulloblastoma ( Adlakha and Saini , 2013; Peruzzi et al . , 2013 ) . Consistent with these findings , allelic loss of chromosome 3p , where miRNA-128-2 is encoded , has also been associated with the most aggressive forms of neuroblastoma , indicating that miR-128 may play an important role in the cell cycle as well as growth and differentiation ( Adlakha and Saini , 2014 ) . In identifying the targets of miR-128 , we eventually narrowed our search of potential miR-128 targets in NPCs to Pcm1 , Nfia , Foxo4 , and Fbxl20 ( Figure 4—source data 1 ) . Among them , Foxo4 , which encodes for an insulin/IGF-1 responsive transcription factor that regulates cell cycles ( Furukawa-Hibi et al . , 2005; Schmidt et al . , 2002 ) , was ruled out as a probable functional target of miR-128 based on a recent study that reported the loss of FOXO4 reduces the potential of human embryonic stem cells ( hESCs ) to differentiate into neural lineages ( Vilchez et al . , 2013 ) , which is opposite from miR-128 overexpression effects that we observed . Nfia ( Nuclear Factor I/A ) encodes for a protein that functions as a transcription and replication factor for adenovirus DNA replication ( Qian et al . , 1995 ) , while Fbx120 , encodes for a F-box protein which is involved in synaptic plasticity of neuronal networks ( Takagi et al . , 2012 ) . As such , both genes were ruled out since their known biological functions were not relevant to the neurogenesis phenotype observed in miR-128 manipulation . Based on these rationales , we sought to focus on PCM1 as our primary gene of interest and investigate the functional relationship between PCM1 and miR-128 in early neurogenesis . Furthermore , based on our expression analysis , miR-128 expression is specific to neural stem cells and gradually increases during cortical development ( Figure 1 ) . In contrast , the expression level of its target protein PCM1 is higher during early developmental stages and lower during late developmental stages ( Figure 4—figure supplement 3 ) . The inverse expression patterns of PCM1 and miR-128 indicate that miR-128 may function by turning off PCM1 expression , indicating that PCM1 may be the primary regulator by which miR-128 governs NPC proliferation and differentiation . In this study , we sought to recapitulate the miR-128 upregulation observed in some ASD brains by overexpressing miR-128 specifically in NPCs . Importantly , we observed a dramatic change in the proliferation and differentiation of NPCs . The observed effects of miR-128 are consistent with a previous study by Ge et al . that demonstrated that the loss of PCM1 triggered NPCs to exit the cell cycle early and promoted the premature differentiation of NPCs to neurons ( Ge et al . , 2010 ) . In this study , knockdown of PCM1 resulted in impaired interkinetic nuclear migration of NPCs , which leads to the overproduction of neurons and to premature depletion of the NPC pool in the developing neocortex . Consistent with this description , our results showed that miR-128 overexpression produced a phenotype similar to that previously reported for PCM1 knockdown ( Figures 2 and 3 ) . More importantly , we found that this phenotype could be rescued by the co-expression of a miR-128-resistant PCM1 variant ( Figures 6 and 7 ) . In contrast , knockdown of miR-128 resulted in increased levels of PCM1 in NPCs ( Figures 4 ) and showed the opposite phenotype of miR-128 overexpression in terms of NPC proliferation and differentiation ( Figures 2 and 3 ) . Our results suggest that miR-128 regulates NPC proliferation and differentiation by fine-tuning endogenous PCM1 levels , which serve as a primary regulator that is required for proper neurogenesis to occur . Intriguingly , we recently identified 9 novel ( i . e . , previously unreported ) missense mutations in the Pcm1 gene in ASD patients ( H . S . J . and S . G . R . , unpublished observations ) , indicating that PCM1 misregulation might be a core mechanism in some ASD patients with disrupted cortical development . Other recent studies using miR-128-2 knockout mice indicate that miR-128 levels regulate the excitability of adult neurons ( Tan et al . , 2013 ) . By selectively inactivating miR-128-2 in forebrain neurons using Camk2a-Cre and floxed miR-128-2 , Tan et al . found that reduced miR-128 expression triggered the early onset of hyperactivity , seizures , and death ( Tan et al . , 2013 ) . Based on their bioinformatics network and pathway analyses of miR-128 target genes , those authors found that miR-128 may regulate the expression of numerous ion channels and transporters as well as genes that contribute to neurotransmitter-driven neuronal excitability and motor activity ( Tan et al . , 2013 ) . Because NPCs are not excitable due to a lack of active sodium channels ( Li et al . , 2008 ) , it is unlikely that the cellular effects of miR-128 observed here resulted from changes in the expression of ion channels or transporters . However , it will be interesting to follow neurons derived from NPCs with misregulated miR-128 to characterize how these neurons integrate into and function in cortical circuits . Moreover , it will be interesting to generate miR-128-1 and miR-128-2 double knockout mice and inducible miR-128-overexpressing transgenic mice to monitor the proliferation and differentiation of NPCs and their effects on behavior . Taken together , our results suggest that miR-128 is an important regulator of cortical development through PCM1 . Future studies to further elucidate specific aspects of the roles of miR-128 and PCM1 in neuronal development and function will be of great interest to this field . All studies were conducted with protocols that were approved by the Institutional Animal Care and Use Committee ( IACUC , protocol number: 2013/SHS/809 ) of the Duke-NUS Graduate Medical School and National Neuroscience Institute . Time-mated C57BL/6 mice were purchased ( InVivos , Singapore ) at E13 . 5 and E14 . 5 for in utero electroporation and culturing of NPCs . Mouse embryos were harvested at E14 . 5 , and the dorsolateral forebrain was dissected and enzymatically triturated to isolate a population of cells enriched in NPCs as previously described . NPCs isolated from a single brain were suspension-cultured in a T25 tissue culture flask in proliferation medium containing human EGF ( 10 ng ml-1 ) , human FGF2 ( 20 ng ml-1 ) ( Invitrogen , Carlsbad , CA ) , N2 supplement ( 1% ) ( GIBCO ) , penicillin ( 100 U ml-1 ) , streptomycin ( 100 mg ml-1 ) , and L-glutamine ( 2 mM ) for 5 days and were allowed to proliferate to form neurospheres . DIV 5 neurospheres were dissociated into single cells using accutase , yielding 4–6 × 106 cells per T25 flask . For each electroporation reaction , 1 × 106 cells were mixed with 2 µg DNA and electroporated using the Neon electroporator ( Invitrogen ) according to the manufacturer’s protocol . Immediately following electroporation , cells were suspension-cultured in a 6-well tissue culture plate in proliferation medium and were allowed to re-form neurospheres for 24 hr . Twenty-four hours post-electroporation , the cells were pulsed with 1 mM 5-bromo-2’-deoxyuridine ( BrdU , Roche ) for 6 hr . The neurospheres were then gently dissociated by pipetting and seeded onto 60 mm coverslips coated with poly-L-lysine and laminin , at a density of 2 × 104 cells/coverslip . After waiting 30 min to allow the cells to attach , the cells were fixed with 4% paraformaldehyde for 30 min at room temperature . To detect BrdU , NPCs were pre-treated with 2 M HCl for 15 min at 37°C , washed with borate buffer ( pH 8 . 5 ) for 30 min and immunostained using a mouse antibody against BrdU ( #8039 , Abcam ) . Twenty-four hours post-electroporation , neurospheres were gently dissociated by pipetting and seeded onto 60 mm coverslips coated with poly-L-lysine and laminin , at a density of 2 × 104 cells/coverslip . Subsequently , NPCs were cultured as monolayer in differentiation medium containing N2 ( 1% ) in DMEM/F12 and were maintained for 5–6 days . The primary antibodies included the following: rabbit anti-Ki67 ( #15580 , Abcam ) , mouse anti-beta III tubulin ( TUJ1 , #1637 , Millipore ) , mouse anti-GFAP ( #N206A/8 , NeuroMab ) , mouse anti-NEUN ( #MAB377 , Millipore ) , chicken anti-MAP2 ( #5392 , Abcam ) , rabbit anti-PH3 ( #9701 , Cell Signaling ) , rabbit anti-TBR2 ( #23345 , Abcam ) , rabbit anti-cleaved caspase-3 ( #9661 , Cell Signaling ) , and rabbit anti-PAX6 ( #PRB-278P , Covance ) . The secondary fluorochrome-conjugated antibodies were diluted 1:400 ( donkey anti-mouse , donkey anti-rabbit , goat anti-mouse and goat anti-rabbit , goat anti-chicken ( Invitrogen] ) . Nuclear counterstaining was performed with 4’ , 6-diamidino-2-phenylindole dihydrochloride ( DAPI ) ( Sigma-Aldrich , #B2261at 0 . 25 μg/μl ) . TUNEL staining was carried out using a kit purchased from Roche ( #12156792910 ) . Images were obtained using an LSM710 confocal microscope ( Zeiss ) . For LCM , unfixed , fresh E14 . 5 brains were embedded in Tissue-Tek O . C . T . compound ( Sakura ) at -24°C . Ten-micrometer cryosections were mounted on PEN Membrane slides ( Leica Microsystems , Germany ) and stored at -80°C until ready for dissection . LCM of the VZ/SVZ , IZ , and CP was performed under direct visualization of the unstained tissue based on tissue morphology using an inverted microscope and PALM Robo software ( Carl Zeiss , Germany ) . The isolated tissue samples were attached onto an Adhesive Cap-500 tube ( Carl Zeiss ) for direct lysis and total RNA extraction . Lentiviral miR-128 expression plasmid and miR-128-Zip knockdown plasmid was obtained from System Biosciences . miR-128 inhibitor was obtained from Genepharma . The PCM1 expression construct was a kind gift from A . Kamiya ( Johns Hopkins , US ) . To generate luciferase reporter constructs , the 3’-UTR of Pcm1 was produced by PCR using murine genomic DNA library . Following primers were used: Pcm1 ( 1865bp ) , forward primer: 5’-GAACCTGAAACAGTGGGAGC-3’; reverse primer: 5’-ACGGTTGCATGTTCCCAATC-3’ . The resulting PCR products were cloned into the pmirGLO Dual-Luciferase miRNA Target Expression Vector ( Promega ) . To generate the pmirGLO-PCM1 3’-UTR mutant construct , miR-128 targeting sequences ( CACTGTG ) were mutated to ( GAGTCTG ) by the Quick Change Site-Directed Mutagenesis Kit ( Stratagene ) using following primers: forward primer: 5'-CCTGGACAGATTCAAACCTTGACAGAGTCTGGGATTTTTCTTTTGC-3' and reverse primer: 5'-GCAAAAGAAAAATCCC‍AG‍ACTCTGTCAA‍G‍G‍TTTGAATCTGTCCAGG-3' . We generated two Pcm1 shRNA vectors with two different published targeting sequences ( Kamiya et al . , 2008 ) #1: 5’-AGCTACTTAATACAGACTA-3’ and #2: TCAGCTTCGTGATTCTA using pCDH-U6-nucGFP-Puro vector which was modified from the lentiviral miR-128 expression plasmid , MMIR-128-1-PA-1 . HEK293 cells were transfected with either the miR-128 mimic or control miRNA in conjunction with the luciferase reporter constructs . Forty-eight hours after they were transfected , the cells were lysed and subjected to luciferase assays using the Dual Luciferase Reporter Assay System ( Promega ) according to the manufacturer’s protocol . Total RNA was extracted using the miRNeasy kit ( Qiagen ) from tissue samples or NPCs . The extraction procedure was then followed by cDNA synthesis using a cDNA synthesis kit ( Promega ) . PCRs were performed on three independent sets of template , and the cycling parameters were as follows: 94°C for 15 s , 55°C for 30 s , and 70°C for 30 s for 40 cycles using the CFX96 real-time PCR detection system ( Bio-Rad , Hercules , CA ) . For each assay , PCR was performed after a melting curve analysis . To reduce variability , we ran each sample in duplicate or even triplicate and included control qPCR reactions without template for each run . The qPCR primers are listed in Figure 4—source data 4 . Timed-pregnant mice ( E13 . 5 ) were anesthetized with isoflurane ( induction , 3 . 5%; surgery , 2 . 5% ) , and the uterine horns were exposed by laparotomy . The DNA ( 3–5 µg µl-1 in water ) together with the dye Fast Green ( 2 mg ml-1; Sigma Aldrich , St . Louis , MO ) was injected through the uterine wall into one of the lateral ventricles of each embryo using a 30-gauge Hamilton syringe . Approximately 2 μl of DNA and dye solution was delivered using a pressure injector ( Picospritzer III; General Valve , Pine Brook , NJ ) . For electroporation , five electrical pulses ( amplitude , 35 V; duration , 50 ms; intervals , 950 ms ) were delivered with a BTX square-wave electroporation generator ( Harvard Apparatus , Holiston , MA ) . The uterine horns were placed back into the abdominal cavity after electroporation , and the embryos were allowed to continue their normal development . Electroporated mouse brains were harvested at E14 . 5 and E17 . 5 for the proliferation and differentiation analyses , respectively . The mitotic spindle orientation of dividing cells was determined on micrographs by measuring the angle of chromosomes relative to the ventricular surface . Nuclei were counterstained with DAPI , and mitotic cells were immunostained with PH3 . GFP-PH3 double-positive cells within the VZ were identified on micrographs , and approximately 150 cells per experimental condition were used for quantification . For proliferation assays , E14 . 5 pregnant dams were administered 5-ethynyl-2’-deoxyuridine ( 5 mg per kg body weight dissolved in 0 . 9% saline ) ( EdU , ThermoFisher Scientific ) by intraperitoneal injection , and the embryos were harvested 2 hr later . For cell cycle exit assays , E13 . 5 pregnant dams were IP injected with 5 mg per kg body weight EdU immediately following in utero electroporation , and the embryos were harvested 24 hr later . Visualization of EdU was performed in accordance with the manufacturer’s protocol . A 5’- digoxigenin ( DIG ) -labeled locked nucleic acid ( LNA ) miR-128 ISH detection probe ( Exiqon ) was used to detect miR-128 expression in the brain . The sequence of the probe was 5’-AAAGAGACCGGTTCACTGTGA-3’ . Briefly , E12 . 5 to P0 perfused brains were dehydrated in 30% sucrose , embedded with Tissue-Tek , and sectioned at 12 µm ( coronal sections ) or 14 µm ( sagittal sections ) . Next , the brain slices were treated with 10 μg ml-1 Proteinase K at 37°C for 15 min , followed by incubation with the LNA-DIG-labeled miR-128 detection probes ( 50 nM ) at 50°C overnight . The sections were blocked with PBS containing 0 . 1% Triton X-100 , 10% normal goat serum ( NGS ) , and 0 . 2% bovine serum albumin ( BSA ) and were incubated with an AP-conjugated anti-DIG antibody ( 1:1000 ) in PBS at 4°C overnight . The NBT+CIP substrate was then added ( #K2191020 , BioChain Institute , Newark , CA ) . For simultaneous miR-128 ISH and immunostaining , the brain slices were incubated with primary antibodies ( peroxidase-conjugated anti-DIG , 1:200 [Roche]; mouse anti-NESTIN , 1:1000 [Sigma] ) in Tris-Buffered Saline ( TBS ) containing 0 . 1% Tween-20 , 20% NGS and 0 . 1% BSA . Immunostaining was detected using Alexa 488 , Alexa 555 or Alexa 647 fluorescent secondary antibodies . Slices were mounted with Vectashield containing DAPI ( Vector Labs , Burlingame , CA ) and examined with confocal microscopy . For the analysis of electroporated brain areas , volumetric z-stacks were acquired for each experimental condition using an LSM710 laser confocal microscope and the Z-series images were collapsed with a maximum intensity projection to a 2D representation . To quantify NPC neurogenesis , the Z-series images were taken using a 40x oil immersion objective ( NA 1 . 3 ) in the CP zone . To quantify the proliferation of NPCs , the Z-series images were taken using a 60x oil immersion objective ( NA 1 . 3 ) in the SVZ . Approximately 1500–2000 GFP-positive cells were counted for each condition , and at least three sets of independent experiments were performed and manually quantified in a blind manner . Cells or brain tissues were lysed using RIPA buffer containing phosphatase and protease inhibitors . Proteins were separated by SDS–PAGE under reducing conditions and transferred to polyvinyl difluoride ( PVDF ) membranes ( Millipore , Billerica , MA ) . Antibody incubations were performed in 5% BSA in TBS buffer using following antibodies: anti-PCM1 , 1:1000 ( Cell Signaling , Danvers , MA ) ; anti-β-actin , 1:1000 ( Santa Cruz , Dallas , TX ) ; and HRP-conjugated rabbit or mouse antibodies , 1:3 , 000 ( GE Life Science , Pittsburgh , PA ) . Recombinant lentiviral particles were produced by co-transfection of HEK293T cells with 6 μg of pMD2G-VSVG , 6 μg of PAX2 packaging plasmids , and 0 μg of pCDH-U6-shRNA-nucGFP-Puro plasmid construct using Lipofectamine 2000 ( Life Technologies ) . Lentiviral particles were collected from supernatant after 72 hr by using PEG-it kit ( System Biosciences ) . Lentiviral particles were concentrated up to 2 × 105 TU/μL . At least 3 experiments were performed independently under each experiment condition , and similar results were obtained . Statistical analyses were performed using ANOVA and Student’s t-test . All data were presented as mean and standard deviation ( mean ± SD ) . Statistical significance was defined when *p<0 . 001 , p<0 . 01 , *p<0 . 05 compared with the control .	The neurons that transmit information around the brain develop from cells called neural progenitor cells . These cells can either divide to form more progenitor cells or to become specific types of neurons . If these carefully regulated processes go wrong – for example , if progenitors fail to stop dividing in order to mature – a range of neurodevelopmental conditions may develop , including autism spectrum disorders . Small RNA molecules called microRNAs control gene activity and protein formation by targeting certain other RNA molecules for destruction . One such microRNA , called miR-128 , helps newly formed neurons to move to the correct region of the cortex – the outer layer of the brain , which is essential for many cognitive processes including thought and language . However , it was not clear whether miR-128 plays any other roles in the development of neurons . Zhang , Kim et al . have now analysed the role of miR-128 in the developing cortex of mice . The findings suggest that miR-128 prevents cortical neural progenitor cells from dividing and supports their development into more specialized cells . Causing miR-128 to be over-produced in the progenitor cells caused the cells to divide less often and encouraged them to mature into neurons . Conversely , removing miR-128 from the progenitor cells caused them to divide more and resulted in fewer neurons forming . Further investigation revealed that miR-128 works by causing less of a protein called PCM1 to be produced . Without this protein , cells cannot divide properly . Future studies could now investigate in more detail how miR-128 and PCM1 affect how the neurons in the cortex develop and work .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2016	MiRNA-128 regulates the proliferation and neurogenesis of neural precursors by targeting PCM1 in the developing cortex
Studies of the genetic basis and evolution of complex social behavior emphasize either conserved or novel genes . To begin to reconcile these perspectives , we studied how the evolutionary conservation of genes associated with social behavior depends on regulatory context , and whether genes associated with social behavior exist in distinct regulatory and evolutionary contexts . We identified modules of co-expressed genes associated with age-based division of labor between nurses and foragers in the ant Monomorium pharaonis , and we studied the relationship between molecular evolution , connectivity , and expression . Highly connected and expressed genes were more evolutionarily conserved , as expected . However , compared to the rest of the genome , forager-upregulated genes were much more highly connected and conserved , while nurse-upregulated genes were less connected and more evolutionarily labile . Our results indicate that the genetic architecture of social behavior includes both highly connected and conserved components as well as loosely connected and evolutionarily labile components . The main conclusion of a decade of sociogenomic research with a range of solitary and social animal species is that highly conserved genes underpinning core physiological processes can also influence behavioral state ( Amdam et al . , 2004 , 2006; Toth and Robinson , 2007; Toth et al . , 2007 , 2010; Woodard et al . , 2011; Woodard et al . , 2014 ) . For example , the insulin signaling pathway , which mediates an organism's response to its internal nutritional state , also influences its behavior ( Ament et al . , 2008 ) . The genetic toolkit hypothesis and related hypotheses propose that a conserved set of genes and gene pathways involved in core physiological processes such as metabolism and reproduction has been repeatedly used in the evolution of complex social behavior in diverse lineages ( West-Eberhard , 1996; Amdam et al . , 2004 , 2006; Toth and Robinson , 2007; Toth et al . , 2007 ) . This hypothesis stems from findings in Evolutionary Developmental Biology that morphological innovation in disparate lineages often involves the convergent use of a conserved set of genes ( e . g . , Hox genes ) ( Carroll et al . , 2001; Toth and Robinson , 2007; Wilkins , 2013 ) . However , social behavior and other social traits are commonly viewed as having unique genetic features and evolutionary dynamics , including especially rapid evolution ( West-Eberhard , 1983; Tanaka , 1996; Moore et al . , 1997; Wolf et al . , 1999; Nonacs , 2011; Bailey and Moore , 2012; Van Dyken and Wade , 2012 ) . Could the molecular mechanisms underlying social interactions ( e . g . , social signal production and response ) and social behavior , together with the process of social evolution result in distinct genetic architectures for social traits compared with other traits ? Recent comparative transcriptomic and genomic studies find low overlap in genes associated with social behavior in different highly social animals and instead highlight the importance of novel genes and rapid evolution of social traits ( Johnson and Tsutsui , 2011; Ferreira et al . , 2013; Simola et al . , 2013; Wissler et al . , 2013; Feldmeyer et al . , 2014; Harpur et al . , 2014; Sumner , 2014; Jasper et al . , 2015 ) , in accordance with recent studies emphasizing the ubiquity of taxonomically restricted genes ( Domazet-Loso and Tautz , 2003; Khalturin et al . , 2009; Tautz and Domazet-Loso , 2011 ) . Perhaps social evolution does not consistently use sets of highly conserved genes to the same degree as morphological evolution ? The novel social genes hypothesis proposes that genes underlying social behavior are often novel socially acting genes or are genes with novel social functions not present in solitary ancestors ( Johnson and Linksvayer , 2010; Johnson and Tsutsui , 2011; Sumner , 2014 ) . Research supporting the genetic toolkit hypothesis has stressed the significant signal of highly conserved genes affecting core physiological processes in transcriptomic data sets for social behavior ( Robinson et al . , 2008; Toth et al . , 2010; Fischman et al . , 2011; Woodard et al . , 2011 , 2014; Toth et al . , 2014 ) . In contrast , research supporting the novel social genes hypothesis has stressed the overall low proportional overlap of genes underlying social behavior in divergent lineages as well as the apparently general low degree of transcriptomic and genomic conservation in divergent lineages ( Johnson and Tsutsui , 2011; Ferreira et al . , 2013; Simola et al . , 2013; Wissler et al . , 2013; Feldmeyer et al . , 2014; Harpur et al . , 2014; Jasper et al . , 2015; Sumner , 2014 ) . We sought to build on these previous results by examining how transcriptional regulatory context influences evolutionary conservation for genes associated with ant social behavior , and further whether genes associated with ant social behavior exist in distinct regulatory and selective contexts compared to the rest of the genome . Research in a range of model organisms demonstrates that the degree of a gene's connectivity to the rest of the regulatory network and its level of expression is often negatively correlated with its rate of molecular evolution ( Krylov et al . , 2003; Hahn and Kern , 2005; Jovelin and Phillips , 2009; Ramsay et al . , 2009 ) . For example , highly connected ‘hub’ genes are often highly expressed and evolutionarily conserved . Previous research has compared rates of molecular evolution for genes associated with reproductive division of labor in social insects ( Hunt et al . , 2010 , 2013; Harpur et al . , 2014 ) , as well as other conditionally expressed phenotypes in other organisms ( Brisson and Nuzhdin , 2008; Leichty et al . , 2012; Purandare et al . , 2014 ) , indicating that genes associated with the expression of worker traits experience elevated rates of molecular evolution . However , the relationships among molecular evolution , connectivity , and expression have been little explored in social insects and are generally little understood for genes associated with social behavior . As a result , it is unclear if observed differences in rates of molecular evolution are caused by differences in regulatory architecture , expression , or perhaps result from distinct evolutionary mechanisms such as kin selection , which may operate differentially on genes associated with social behavior relative to the rest of the genome ( Linksvayer and Wade , 2009; Hall and Goodisman , 2012 ) . We further sought to identify modules of co-expressed genes that may be composed of both conserved and novel genes and may contribute to the expression and evolution of social complexity . We studied the genetic basis and evolution of a fundamental aspect of social insect behavior , age-based division of labor ( age polyethism ) . Age polyethism involves the progression of workers from in-nest tasks such as nursing to outside-nest tasks such as foraging . Because age polyethism is a trait expressed by the functionally sterile worker caste , it is expected to be shaped primarily through kin selection ( Hamilton , 1964 ) . While age polyethism plays a central role in the functioning of many eusocial systems ( Hölldobler and Wilson , 2009 ) , the molecular underpinnings have only been well studied in the honey bee Apis mellifera ( Whitfield et al . , 2006; Ament et al . , 2008; Chandrasekaran et al . , 2011 ) , so that the genetic and evolutionary basis of age polyethism is not generally understood outside of honey bees . We identified transcriptional modules of co-regulated genes associated with worker age polyethism in the pharaoh ant Monomorium pharaonis; we identified the degree that these genes overlap with genes involved in age polyethism in two other social insects ( Alaux et al . , 2009; Manfredini et al . , 2014 ) ; and we studied the relationship between expression level , connectivity and rates of molecular evolution at these genes compared to the rest of the genome . We tracked cohorts of age-marked workers and recorded their behavior and location inside and outside the nest . In order to identify differentially expressed genes associated with age-based division of labor , we collected age-marked workers and workers observed performing specific behaviors . The observed location of workers from different age classes changed with both nest location and behavior ( glm with quasipoisson errors and log link , both p < 0 . 01 ) ( Figure 1 , Figure 1—figure supplements 1 , 2 ) . In concordance with the expected pattern of age polyethism , the average age of workers observed in the different locations increased as distance from the brood area increased ( Figure 1—figure supplement 2 ) . Of the 15 behaviors observed more than 15 total times ( Supplementary file 1 ) , the likelihood of observing workers performing the behaviors ‘nurse’ , ‘groom’ , ‘rest’ , ‘trophallaxis’ , ‘walk’ , and ‘forage’ depended on worker age ( Figure 1A; glm with binomial errors and logit link , all nominal p < 0 . 0002 , α = 0 . 003 , controlling for multiple testing ) . Nursing and foraging were at the two extremes: the average age of workers observed nursing was 6 . 94 days and the average age of workers observed foraging ( i . e . , in the act of collecting food ) was 13 . 04 days . There appeared to be a marked transition from nursing to foraging between 9 and 12 days of age ( Figure 1A ) , with 75% of nursing observations made for workers less than 10 days old and 75% of foraging observations made for workers over 10 days old ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 04775 . 003Figure 1 . Behavioral and transcriptional changes associated with worker age and behavior . Numbers along the x-axis represent ages of marked worker cohorts , starting at worker eclosion as an adult ( day 0 ) . In all plots dark green represents greater values , while white represents lower values of the measure being plotted . ( A ) Behavioral results . Workers showed an age-dependent progression of activity , progressing from tasks such as nursing and grooming in the nest to outside tasks such as walking and foraging . ( B ) Heat map of expression levels over the course of worker aging ( higher expression in darker green ) , for 25 genes most differentially expressed between nurses and foragers . The red line separates the samples classified as ‘nurses’ by K-nearest neighbor classification on the left , from ‘foragers’ on the right , suggesting a distinct transition between the two categories . ( C ) Correlation between patterns of expression in the 14 identified modules across worker age and behavior . The colors of the boxes are scaled with the value of correlation coefficients , ranging from white to dark green . On the right side of the heat map are the numbers of genes in each module and a dendrogram showing the inferred relationships among modules . The modules show complex patterns of expression , for example with some most highly expressed at age 0 , some showing decreasing expression over time , and some increasing expression over time . DOI: http://dx . doi . org/10 . 7554/eLife . 04775 . 00310 . 7554/eLife . 04775 . 004Figure 1—figure supplement 1 . The behaviors performed by age-marked workers changed as the workers aged , from nursing to foraging . Boxplots show the distribution of age in days for each behavior; white diamonds and the printed number show the mean age for each behavior; and the number of observations for each behavior is shown at the bottom of the graph . DOI: http://dx . doi . org/10 . 7554/eLife . 04775 . 00410 . 7554/eLife . 04775 . 005Figure 1—figure supplement 2 . The location of age-marked workers also changed as the workers aged , from the nest area over the brood to outside the nest . Boxplots show the distribution of age with the mean and sample size for each category . DOI: http://dx . doi . org/10 . 7554/eLife . 04775 . 00510 . 7554/eLife . 04775 . 006Figure 1—figure supplement 3 . The identified modules vary in expression pattern , composition of nurse-upregulated and forager-upregulated genes , and the proportion of conserved genes with identified fire ant orthologs . The total number of genes , number of nurse-upregulated genes , and forager-upregulated genes are shown , along with the proportion of identified fire ant orthologs and prominent gene ontology terms enriched for each module ( see Supplementary file 4 for the full GO enrichment profiles for each module ) . Both the module number and associated module color are shown on the left . DOI: http://dx . doi . org/10 . 7554/eLife . 04775 . 006 There was a trade-off in the assemblies between N50 and overall assembly lengths , as a function of kmer size . We chose k = 69 as a compromise between these two metrics , resulting in a scaffolded assembly of 284 mb , with a N50 of 19 . 0 kb . Although there is no M . pharaonis genome size estimate , the assembly is in the range of genome sizes typical of other myrmicine ants ( Tsutsui et al . , 2008 ) . CEGMA analysis ( Parra et al . , 2009 ) found complete sequences for 92% of the ultra-conserved eukaryotic genes , and partial sequences for 97% . Most reads ( 97 . 6% ) could be re-mapped to the genome assembly , resulting in a coverage estimate of 40× . Cufflinks assembly identified 22 , 385 transcribed loci . 74 . 9 ± 18% ( median 85 . 1% ) of the reads for each sample could be re-mapped to predicted transcripts extracted from the reference . After the reads were re-mapped to the assembled transcripts using the RSEM pipeline , each library had 10 , 602 , 832 ± 2 , 925 , 898 expected counts . The complete analysis of gene expression data , including R code and output , is available in the Supplementary file 2 ( with the complete R markdown script as Source code 1 ) , and it is summarized below . We wished to examine which of the four worker behavioral samples ( nursing larvae , foraging , grooming larvae , and worker–worker trophallaxis [i . e . , exchanging liquid food] ) had distinct expression profiles vs all of the others . We used linear contrasts to determine the number of differentially expressed genes between the focal behavioral category and the other behaviors . Of these contrasts , only foragers and nurses had significantly different gene expression patterns , when compared to the rest , that is , there was no evidence that workers engaged in grooming and trophallaxis had distinct transcriptional states . Consequently , we focused subsequent analysis on nurse and forager behavioral categories , except in the construction of the co-expression networks , where all behavioral category and age class samples were used ( see below ) . There were 1217 forager-upregulated , 1247 nurse-upregulated transcripts , and 14 , 907 transcripts that were not differentially expressed . Qualitatively , gene expression patterns mirrored the behavioral transition from nursing to foraging that we observed around day 10 ( Figure 1A , B ) . To quantify these observations , we used a supervised learning approach ( K-nearest neighbor classifier or KNN ) to check whether genes differentially expressed in nurses and foragers could be used to differentiate the age class data into two clusters . After the KNN was trained on nurse and forager profiles , it clearly separated workers into two distinct classes based on age , assigning those younger than 12 days into the nurse class , and the rest into the forager class ( Supplementary file 2 pages 14–15 ) , suggesting a fairly discrete transcriptomic transition between the two behaviors . The proportion of genes with identified orthologs in the fire ant Solenopsis invicta differed between behavioral categories ( Manfredini et al . , 2014 ) , with forager-upregulated genes having a higher proportion ( 0 . 54 ) relative to nurse-upregulated ( 0 . 43 ) and non-differentially expressed ( 0 . 43 ) ( multiple comparison Kruskal–Wallis , p < 0 . 05 ) . Similarly , the proportion of genes with identified honey bee A . mellifera orthologs was higher for forager-upregulated genes ( 0 . 50 ) , relative to nurse-upregulated ( 0 . 38 ) , and non-differentially expressed genes ( 0 . 38 ) ( multiple comparison Kruskal–Wallis , p < 0 . 05 ) ( note we used a less conservative BLAST threshold for the honey bee so that the proportions of honey bee and fire ant orthologs are not directly comparable , see ‘Materials and methods’ ) . Furthermore , approximately half of non-differentially expressed ( 0 . 51 ) and nurse-upregulated ( 0 . 50 ) genes did not have orthologs identified in either the fire ant or honey bee genomes , but this proportion was lower for forager-upregulated genes ( 0 . 39 ) ; correspondingly , the proportion of forager-upregulated genes with orthologs identified from both fire ants and honey bees was higher ( 0 . 43 ) compared to nurse-upregulated and non-differentially expressed genes ( 0 . 32 ) ( X2 = 71 . 42 , df = 6 , p < 10−13 ) . Genes previously detected as upregulated in nurses and foragers of S . invicta were more likely to have identified M . pharaonis orthologs up-regulated in these contexts as well ( p = 0 . 0022 and p = 0 . 040 , respectively ) . However , the actual percentage of genes differentially expressed in the same context in these two ant data sets was small: 3 . 8% ( 47/1247 ) of nurse genes and 3 . 2% ( 39/1217 ) of forager genes; or if only considering genes with orthologs identified in both species , 8 . 6% ( 47/549 ) nurse genes and 5 . 9% ( 39/657 ) forager genes . While there was low overlap in the lists of differentially expressed genes , there could still be stronger overlap in genome-wide expression profiles when comparing nurse and forager samples between S . invicta and M . pharaonis . Thus , we estimated the correlation in the change of expression between nurse and forager samples ( i . e . , log fold change ) between the S . invicta and M . pharaonis datasets for all genes with identifiable homologs . There was a significant correlation in the change of expression for nurse and forager samples , but one that explained only 2% of the variance ( Spearman's rho = 0 . 14 , 6324 genes , p < 10−16 ) . In contrast to the fire ant and pharaoh ant comparison , previously identified forager- and nurse-upregulated honey bee A . mellifera genes ( Alaux et al . , 2009 ) were not more likely to have M . pharaonis orthologs expressed in the same context ( p = 0 . 99 , p = 0 . 98 , respectively ) , consistent with a previous comparison between S . invicta and A . mellifera ( Manfredini et al . , 2014 ) . The actual overlap in honey bee and pharaoh ant gene lists was higher ( 71 nurse-upregulated genes and 46 forager-upregulated genes ) due to the less conservative BLAST threshold we used for identifying honey bee orthologs , but the honey bee lists were also larger ( Alaux et al . , 2009 ) and the overlap was not significant . Nurse-upregulated genes were strongly enriched for a range of GO terms associated with metabolism ( nearly 50 metabolism-related terms with p < 10−5; Supplementary file 3 ) . Forager-upregulated genes had a more diffuse signal , being relatively more weakly enriched for various GO terms , for example , associated with signal transduction and gland morphogenesis . Forager-upregulated genes showed a more consistent signal for underrepresented terms , for example , GO terms associated with metabolic processes and chromatin modification ( Supplementary file 3 ) . The number of modules produced by WGCNA can vary based on several thresholding parameters , which we left as defaults ( Supplementary file 2 , pages 20–21 ) . These settings resulted in 14 co-expression modules , ranging in size from 83 to 4218 genes ( Figure 1C; Figure 1—figure supplement 3 ) . A module's overall expression can be characterized by its eigengene . Correlations between eigengenes and traits in the original data suggest the involvement of corresponding modules in these traits . Eigengenes in two of the modules—1 and 14 , which contained the most nurse and forager genes , respectively—were strongly correlated with worker age , although in opposite directions , suggesting their role in aging and age-based division of labor ( r = −0 . 95 , r = 0 . 91 and with FDR-adjusted p-values 0 . 0038 , 0 . 023 , respectively ) ( Supplementary file 2 , page 24 ) . Other modules showed complex patterns of age and behavior specific expression , with most of them showing a peak in expression once or twice during the lifetime of a worker ( Supplementary file 2 , page 26 ) . Interestingly , most module eigengenes switched signs during the period between 9 and 12 days , corresponding to the behavioral transition from nursing to foraging . In other words , there appeared to be a major reprogramming step , where modules initially showing low expression became up-regulated , while modules initially showing high expression were down-regulated . Forager-upregulated genes were concentrated in just a few modules , with only two modules containing more than 100 forager-upregulated genes ( Figure 1—figure supplement 3 ) . In contrast , nurse-upregulated genes were more widely distributed , with five modules having more than 100 nurse-upregulated genes ( Figure 1—figure supplement 3 ) . These five modules were mainly enriched for GO terms associated with metabolism and development ( Figure 1—figure supplement 3; Supplementary file 4 ) . Module 5 , which contained 116 nurse-upregulated genes , was also enriched for terms associated with female gonad development , which is surprising given that M . pharaonis workers lack ovaries and are completely sterile . The modules containing forager-upregulated genes were enriched for a broad range of GO terms , for example associated with regulation of signaling , development and neurogenesis , and gene expression ( Figure 1—figure supplement 3; Supplementary file 4 ) . The proportion of module genes with identified S . invicta orthologs ranged from 0 . 28 to 0 . 53 ( Figure 1—figure supplement 3 ) , suggesting that in addition to being involved in different functions , the modules are composed of different proportions of conserved and taxonomically restricted genes . Forager-upregulated genes were much more connected than nurse or non-differentially expressed genes , while nurse-upregulated genes were less connected than non-differentially expressed genes ( Figure 2A ) ( multiple comparison Kruskal–Wallis , p < 0 . 05 ) . There was a small but significant difference in evolutionary rate dN/dS ( Figure 2C ) , with nurse-upregulated genes evolving more rapidly than non-differentially expressed genes ( multiple comparison Kruskal–Wallis , p < 0 . 05 ) . Nurse and forager genes were also more highly expressed ( Figure 2B ) than non-differentially expressed genes ( Kruskal–Wallis , p < 0 . 05 ) , although this last comparison is likely biased because differential expression is more easily detected in highly expressed genes . 10 . 7554/eLife . 04775 . 007Figure 2 . Connectivity , expression , and evolutionary rate for nurse-upregulated ( blue ) , forager-upregulated ( red ) , and non-differentially expressed genes ( gray ) . Overall , connectivity and expression are positively correlated ( F ) and negatively associated with evolutionary rate ( D and E ) , as expected . At the same time , forager-upregulated genes are much more strongly connected while nurse-upregulated genes are more loosely connected compared to non-differentially expressed genes ( A ) ; Nurse-upregulated genes have a small but significant increase in evolutionary rate ( C ) ; and both forager- and nurse-upregulated genes are more highly expressed than non-differentially expressed genes ( B ) . The top panels show results for all data , while the bottom panels show results only for genes with S . invicta orthologs that had estimated evolutionary rates . DOI: http://dx . doi . org/10 . 7554/eLife . 04775 . 007 Co-expression network connectivity and expression level were overall negatively associated with evolutionary rate , such that highly connected and highly expressed genes had decreased rates of molecular evolution ( Figure 2D , E; evolutionary rate and connectivity , r = −0 . 15 , p < 2 × 10−16; evolutionary rate and expression , measured in terms of transcriptional abundance , fragments per million reads mapped , FPKM , r = −0 . 12 , p < 2 × 10−16 ) ; and connectivity and expression were positively correlated ( r = 0 . 30 , p < 2 × 10−16 ) . In a full model considering how a gene's rate of molecular evolution depended on its gene expression level , network connectedness , and behavioral category , the largest effects were main effects of expression ( z = −7 . 42 , p = 1 . 29 × 10−13 ) and connectivity ( z = −3 . 69 , p = 0 . 00023 ) . We also studied the effects of gene category ( i . e . , upregulated in nurses or foragers , or not differentially expressed ) , expression level , and connectivity on whether a given M . pharaonis gene had an identifiable fire ant S . invicta and honey bee A . mellifera orthologs . Overall , genes with orthologs in the fire ant or honey bee had greater connectivity and expression ( Figure 3 , Figure 3—figure supplement 1 ) . In considering a model with both main and interaction effects of behavioral category , expression level , and connectivity , connectivity had the strongest effect ( glm with quasibinomial residuals: t = 24 . 5 , p < 10−16 , for the presence of S . invicta orthologs; t = 32 . 2 , p < 10−16 , for the presence of A . mellifera orthologs ) , with more highly connected genes being more likely to have an ortholog . There were also much smaller interaction effects indicating that nurse-upregulated genes had fewer orthologs than expected given their connectivities ( i . e . , connectivity had a weaker effect on nurse-upregulated genes than other genes , Figure 3 and Figure 3—figure supplement 1; t = −3 . 17 , p = 0 . 0015 for S . invicta orthologs; t = −2 . 76 , p = 0 . 0057 for A . mellifera orthologs ) , and forager-upregulated genes had fewer orthologs than expected given their expression ( t = −2 . 33 , p = 0 . 02 for S . invicta orthologs; t = −2 . 58 , p = 0 . 0098 for A . mellifera orthologs; Figure 3 and Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 04775 . 008Figure 3 . Genes with identified fire ant orthologs were more highly connected and expressed , but this relationship also depended on whether the gene was nurse-upregulated ( blue ) , forager-upregulated ( red ) , or non-differentially expressed ( NDE , gray ) . As shown in Figure 2 , forager-regulated genes were much more highly connected , and overall , forager-upregulated genes had a higher proportion of identified fire ant orthologs ( 0 . 54 ) relative to nurse-upregulated and non-differentially expressed genes ( 0 . 43 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04775 . 00810 . 7554/eLife . 04775 . 009Figure 3—figure supplement 1 . Very similarly to Figure 3 , genes with identified honey orthologs were more highly connected and expressed , but this relationship also depended on whether the gene was nurse-upregulated ( blue ) , forager-upregulated ( red ) , or non-differentially expressed ( NDE , gray ) . Forager-upregulated genes had a higher proportion of identified honey bee orthologs ( 0 . 50 ) relative to nurse-upregulated and non-differentially expressed genes ( 0 . 38 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04775 . 009 Pharaoh ant workers showed a clearly defined age-based transition from nursing to foraging , in terms of both behavioral and transcriptional patterns , with nurses and foragers having strongly differentiated sets of upregulated genes ( Figure 1 ) . We recovered the commonly observed genome-wide relationship between a gene's rate of molecular evolution , its expression level , and its network connectivity ( Krylov et al . , 2003; Hahn and Kern , 2005; Jovelin and Phillips , 2009; Ramsay et al . , 2009 ) . Specifically , the rate of molecular evolution ( dN/dS ) as well as the likelihood a gene had identified fire ant and honey bee orthologs was negatively correlated with its expression level and connectivity within co-expression networks , while expression and connectivity were positively correlated ( Figures 2 , 3 ) . In addition to these genome-wide patterns , nurse- and forager-upregulated genes had distinct regulatory and evolutionary patterns relative to each other and to the rest of the transcriptome ( Figures 2 , 3 ) . Most strikingly , forager-upregulated genes were much more highly connected and correspondingly more conserved , while nurse-upregulated genes were less connected , and more rapidly evolving and less conserved . Previous studies of the evolutionary genetic basis of social behavior have focused on the overlap of genes lists associated with social traits in different lineages . We found significant but seemingly low ( <4% ) overlap in lists of differentially expressed genes and the correlation in genome-wide expression profiles ( r = 0 . 14 ) when comparing gene expression in nurse and forager samples between the pharaoh ant and fire ant , S . invicta . Such low overlap seems surprising , given that these two ants are in closely related ant genera , having diverged on the order of 50 Mya ( Ward et al . , 2014 ) . However , the comparison is not perfect , given substantial differences between the two studies in methodology used to characterize the behaviors , and in the technology used to measure gene expression ( i . e . , microarray vs RNA sequencing ) ( Manfredini et al . , 2014 ) . We did not find significant overlap between lists of honey bee and pharaoh ant genes associated with age polyethism , consistent with results reported by the earlier fire ant study ( Manfredini et al . , 2014 ) . While we expected decreased overlap given that honey bees and ants diverged longer ago , ∼170 Mya ( Ronquist et al . , 2012 ) , and represent independent origins of eusociality , the ant-honey bee comparison is also more problematic because the honey bee data are based on brain gene expression profiles whereas the fire ant and pharaoh ant data are based on whole body gene expression profiles . Past studies have often interpreted significant but similarly low overlap in lists of genes associated with social behavior from different lineages as supporting the genetic toolkit hypothesis ( Toth et al . , 2010 , 2014; Woodard et al . , 2014 ) . In contrast , other authors have recently interpreted low overlap as being consistent with the novel social genes hypothesis , which emphasizes the importance of taxonomically restricted genes ( Ferreira et al . , 2013; Feldmeyer et al . , 2014; Sumner , 2014 ) . The contrasting emphasis of authors on either conserved or novel genes begs the question: what degree of conservation in gene lists is necessary for confirmation or rejection of these two hypotheses ? For example , the fact that nurse-upregulated genes in M . pharaonis are more rapidly evolving than the rest of the genome and that 50% of nurse-upregulated genes do not have identifiable fire ant or honey bee orthologs suggests that novel genes may have important nurse-specific functions . At the same time , the significant overlap of fire ant and pharaoh ant gene lists and the strong enrichment of nurse-upregulated genes for gene ontology terms associated with metabolism and development suggests that conserved genes involved in core physiological processes also play important roles in nurse function and the evolution of division of labor . Thus , our results are generally consistent with both hypotheses . We suggest that neither of these two hypotheses has yet been formulated in a way that is readily tested , in part because it is unclear what precise genes are expected to be included or excluded from a genetic toolkit ( Wilkins , 2013 ) . Furthermore , these hypotheses are not mutually exclusive , since both conserved and novel genes likely play roles in the evolution of all new traits ( Johnson and Linksvayer , 2010; Woodard et al . , 2011 ) . We suggest that shifting the focus , from lists of genes to modules of co-expressed genes in the context of genome-wide transcriptional and evolutionary patterns , can help to elucidate how social evolution has produced social complexity . In this way , one question we can ask is whether we see any simple molecular signature of social evolution , for example due to kin selection ? As Monomorium ant workers are obligately sterile , all worker traits are expected to be shaped exclusively by indirect selection ( i . e . , kin selection ) ( Hamilton , 1964 ) . All-else-equal , such indirect selection is weaker than direct selection , proportional to relatedness ( Hamilton , 1964 ) , and a priori is expected to produce relaxed selective constraint and elevated rates of molecular evolution for all genes associated with worker traits ( Linksvayer and Wade , 2009 ) . Past studies have found different rates of molecular evolution for worker-biased and queen-biased genes , with most studies finding that worker-biased genes are more rapidly evolving ( Ferreira et al . , 2013; Feldmeyer et al . , 2014; Harpur et al . , 2014; but see; Hunt et al . , 2010 ) . Some researchers have interpreted different patterns between lineages as being consistent with simple kin selection predictions based on differences in within-colony relatedness ( Hall and Goodisman , 2012 ) , but most studies have emphasized the association between conditional expression and relaxed selection ( Hunt et al . , 2011 , 2013 ) , as well as genes associated with worker traits simply experiencing stronger positive selection ( Hunt et al . , 2010; Ferreira et al . , 2013; Feldmeyer et al . , 2014; Harpur et al . , 2014 ) . We observed weakly elevated rates of molecular evolution at nurse-upregulated genes compared to the rest of the genome , but much more notable was the distinct connectivity and corresponding differences in gene conservation for forager-upregulated genes relative to nurse-upregulated and non-differentially expressed genes . These results suggest that social evolution does not just have simple genome-wide effects such as relaxed effective selection associated with kin selection , but instead shapes complex social traits while acting within general systems-level constraints imposed by regulatory architecture . The common perception that social evolution often involves rapid evolutionary dynamics ( West-Eberhard , 1983; Tanaka , 1996; Moore et al . , 1997; Wolf et al . , 1999; Nonacs , 2011; Bailey and Moore , 2012; Van Dyken and Wade , 2012 ) may result from the fact that genes influencing many key social traits are not only conditionally expressed , but are also located peripherally within regulatory networks , and so are relatively unconstrained . For example , we expect that traits associated with social signal production ( e . g . , pheromone and glandular secretions ) are often located peripherally within regulatory networks and as a result may be evolutionarily labile ( Jasper et al . , 2015 ) , as is the case more generally with secreted proteins ( Julenius and Pedersen , 2006; Liao et al . , 2010; Nogueira et al . , 2012 ) . More core and conserved components are also certain to be important to the expression of these traits , but their contribution to trait evolution may be minimized by virtue of the fact that they are highly connected . These arguments suggest how both conserved , toolkit genes , as well as rapidly evolving and taxonomically restricted novel genes , likely play important roles in the evolution of social novelty , with novel genes being added peripherally to regulatory networks . Our results are consistent with this interpretation , because M . pharaonis age-based division of labor seems to have a complex genetic basis with some components that are highly connected and conserved , and other components that are more loosely connected and evolutionarily labile . Our findings that nurse-upregulated genes are more rapidly evolving and less conserved among social insect lineages relative to forager-upregulated genes suggest that nurse traits have been a major focus of evolutionary innovation between social insect lineages . This result seems surprising given that foragers of different lineages experience diverse environments outside the nest compared to the relatively constant within-nest environment experienced by nurses and could be expected to experience more diverse selective pressures . One explanation is that the physiological mechanisms associated with metabolically costly foraging activities and older adult life ( M . pharaonis workers usually only live several weeks [Peacock and Baxter , 1950] , so that foragers which start right before their second week of age may already be senescing ) may be relatively conserved and simple . Nursing behavior , occurring during very early adult life , may involve more diverse physiological and developmental processes , and nursing itself may also involve more diverse behaviors and physiological processes , such as food processing and the synthesis of glandular secretions that are fed to larvae . Perhaps the relatively more complex genetic architecture ( less tightly connected , involving more modules , and diverse processes ) has served as less of a constraint and facilitated more evolutionary change for nurse-related genes . If so , we predict that nurse-specific functions and functions for early adult life may be generally more evolutionarily labile as well as more physiologically and behaviorally labile within and across lineages than forager-specific functions . Note that this prediction is opposite the typical expectation that genes acting early in development have more pleiotropic effects and are thus especially constrained ( Roux and Robinson-Rechavi , 2008; Piasecka et al . , 2013 ) , but obligate sterility may , in part , release workers from these constraints on the evolution of genes acting early in worker development . We identified two discrete sets of genes with distinct genetic architecture associated with age-based division of labor . The majority of forager-upregulated genes were contained within a single gene module ( module 14; Figure 1—figure supplement 3 ) that was significantly positively associated with age . Another module with expression negatively associated with age contained the largest number of nurse genes , but nurse genes were also broadly spread out across a number of other modules with complex expression patterns across age and behavioral groups . Interestingly , the modules differed in the proportion of constituent genes which had identifiable S . invicta and A . mellifera orthologs , indicating that modules vary in the degree to which they are composed of conserved genes and gene networks vs rapidly evolving genes with unknown function . That said , the modules were enriched for various gene ontology terms , providing some insight into their putative functional importance ( Supplementary file 4 ) . By explicitly studying regulatory architecture and inferring modules of tightly connected genes in other species as well as M . pharaonis , it will be possible to further identify what network components contribute to the expression of social traits , how rapidly these components are evolving within populations , and how they have contributed to phenotypic differences between divergent lineages . Building on the genetic toolkit conceptual framework , it will be possible to ask to what degree diverse lineages repeatedly use the same modules , and importantly approaches already exist for quantifying module overlap in the absence of functional information ( Oldham et al . , 2006; Langfelder et al . , 2011 ) . Similarly , after finding non-significant overlap in lists of genes associated with queen- and worker-caste development in paper wasps and honey bees , Berens et al . ( 2014 ) recently invoked a ‘looser’ version of the genetic toolkit hypothesis by examining the overlap of inferred functional enrichment of gene lists ( i . e . , via gene ontology analysis ) . Focus on co-expressed modules may actually improve the feasibility of inferring the function of co-expressed genes based on observed expression patterns together with standard functional information inferred from the subset of conserved annotated genes with identifiable orthologs from model systems . It will also be possible to determine the relative contribution of conserved vs taxonomically restricted genes to co-expression modules . Two replicate M . pharaonis observation colonies were established , each with 10 queens , approximately 4000 workers , and 1000 brood , representing a random subsample of a larger source colony . Each colony was established from a separate source colony , which came from a stock of approximately 40 colonies that have been repeatedly mixed across generations so that they are genetically similar . Observation nests were constructed of two pieces of 5 × 15 cm glass separated by 1 . 5 mm thick plastic sheeting . Colonies were given water in cotton-plugged test tubes , 50% honey solution , beef liver , egg yolk , and mealworms ad libitum , replaced twice a week . Colonies were maintained at 27 C and 65% relative humidity in climate controlled rooms at the University of Pennsylvania . Every 3 days , 600 newly eclosed callow workers , which were inferred to be approximately 0–1 days old , were collected from 8–10 stock colonies . These callow workers were briefly anesthetized with CO2 and individually paint marked on the gaster with a unique age cohort color dot using a Sharpie extra fine oil based paint pen , and then 300 were added to each of the observation colonies . Five uniquely marked age cohorts were thus added to the colonies on days 1 , 4 , 7 , 10 , and 13 of the study . Nestmate recognition is at most weak and transient in M . pharaonis ( Schmidt et al . , 2010 ) , and callows in particular are readily accepted . We also set up a camera to automatically take images of the nest areas of each colony once every 20 min for the entire period of the study , although we do not further discuss these images . Previous literature indicates that M . pharaonis workers are expected to live 9–10 weeks ( Peacock and Baxter , 1950 ) , but our preliminary trials with our setup indicated that workers tend to die or lose their paint marks after several weeks . We ran the study for 1 month , expecting to capture the major age-based transitions in worker behavior ( e . g . , the nursing to foraging transition observed in other species ) , but it is possible that we missed late behavioral transitions that occurred towards the end of workers' lives . In practice , such late transitions are difficult to detect as sample size necessarily declines as increasing numbers of workers die . A behavioral scan of each colony was completed once each day for the duration of the month-long study by recording the instantaneous behavior and location observed for every visible paint-marked worker . Each behavioral scan was performed at 20× magnification with a Nikon SMZ800 stereomicroscope . We recorded 30 distinct behaviors , but only 15 were observed more than 15 total times during the study period ( Supplementary file 1 ) . We defined an individual as foraging if it was observed on a food or water source or actually carrying food ( i . e . , foraging included the behaviors ‘on honey’ , ‘on liver’ , ‘on water’ , or ‘carrying food’; Supplementary file 1 ) . Each experimental colony contained four identifiable locations that were redefined prior to each behavioral scan: brood area , brood periphery , remaining nest area , and foraging area . The brood area was defined as the central area within the nest containing all brood and queens ( Edwards , 1991 ) . The nest periphery was defined as the region directly adjacent to the brood area , where workers were dense in aggregation but not in contact with any of the brood . The nest area was defined as the sparsely occupied remainder of the space within the nest , not including the brood area and nest periphery . The foraging area included all areas outside of the nest . Analyses of behavioral data were conducted in R ( www . r-project . org ) . Every 3 days , whole bodies of five individuals from each available uniquely paint marked age cohort were collected , flash frozen in liquid nitrogen , and stored at −80°C . This sampling scheme resulted in seven groups of individuals of known age ( 0 , 3 , 6 , 9 , 12 , 15 , and 18+ days old ) . 20 individuals of each of these age category were pooled for whole body RNA extraction for each of the two replicate observation colonies . In addition , for each of the two replicate observation colonies , we collected and pooled 20 non-paint marked workers in the act of the following five behaviors: nursing larvae , grooming larvae , engaged in trophallaxis with other workers , foraging for protein ( collecting egg , mealworm , or liver ) , and foraging for carbohydrates ( collecting honey solution ) . RNA was extracted from pools of worker samples of known age or observed behavior using Qiagen RNeasy kits with standard protocols . RNA sequencing libraries were constructed at the University of Arizona Genetics Core ( UAGC ) with RNA TruSeq library construction kits following standard protocols . In total there were 24 libraries: 2 colony replicates × ( 7 age groups + 5 behavioral groups ) . RNA sequencing was conducted at the University of Arizona Genetics Core on an Illumina HiSeq2000 with 100 bp paired ends reads , with six samples multiplexed per lane , randomly distributed across four lanes . Sequences were post-processed by cutadapt ( Martin , 2011 ) to remove Illumina adapter sequences and ConDeTri ( Smeds and Künstner , 2011 ) to remove low-quality bases . DNA from a single haploid male ( 183 ng ) was used to prepare a TruSeq library , which was sequenced in multiplex on an Illumina HiSeq 2000 , yielding 70 , 894 , 179 million 100 bp read pairs . Raw genomic reads were quality and adaptor trimmed using ConDeTri and cutadapt ( Martin , 2011; Smeds and Künstner , 2011 ) , producing 57 , 002 , 951 read pairs and 8 , 361 , 560 single reads ( 12 . 3 Gb total ) . The assembly was carried out using ABYSS , with a range of kmers from 53 to 91 ( Simpson et al . , 2009 ) . We then chose the assembly with the longest N50 as the reference for transcriptome assembly . Genome assembly quality was evaluated using the CEGMA pipeline ( Parra et al . , 2009 ) , and by re-mapping the paired end trimmed reads using bowtie2 ( Langmead and Salzberg , 2012 ) . The transcriptome was mapped to the reference using Tophat 2 , and assembled into transcripts using Cufflinks 2 . 1 ( Roberts et al . , 2011; Kim et al . , 2013 ) . Gene expression data were obtained by re-mapping the transcript reads to the extracted transcripts using RSEM and calculating the expected counts at the gene level ( Li and Dewey , 2011 ) . When multiple isoforms of a single locus were found , only the longest transcript was used for gene annotation . Assembled transcripts were annotated using BLASTX from the non-redundant NCBI database with expectation values of E = 10−5 . These results were used to assign Gene Ontology ( GO ) profiles with Blast2go ( Conesa et al . , 2005 ) . Transcript counts were filtered by abundance , removing those with less than 1 fragment per kilobase mapped ( FPKM ) in more than half of the libraries ( Mortazavi et al . , 2008 ) . Differential gene expression analysis was carried out in edgeR , using a GLM fit to the count data and identifying differentially expressed genes using planned linear contrasts ( Robinson et al . , 2010 ) . In order to infer co-expression modules and gain an insight into network structure of gene interactions , we performed a weighted gene co-expression network analysis ( WGCNA ) on the count data ( Langfelder and Horvath , 2008 ) . WGCNA was conducted on the entire transcript set , after filtering out the low-abundance transcripts . This analysis relies on patterns of gene co-expression , but has been shown to reconstruct protein–protein interaction networks with reasonable accuracy ( Zhao et al . , 2010; Allen et al . , 2012 ) . We used total connectivity as a measure of gene interaction strength , because it is not as sensitive to module assignments , and most likely reflects the overall selective pressures acting on the gene , beyond those imposed by its role in age polyethism . As with most gene expression analysis , WGCNA provides better estimates for highly abundant genes , and in particular for genes showing variation in their expression levels . Consequently , low-abundance and invariant genes will show lower connectivity . GO term enrichment analysis was performed using the R package GOstats ( Falcon and Gentleman , 2007 ) . We report GO terms as enriched when p < 0 . 05 . Fire ant ( S . invicta ) orthologs for each gene were determined using reciprocal best BLASTP , using cutoffs of 10−10 . This parameterization allowed for high specificity , though at the cost of sensitivity , since paralogs were ignored ( Chen et al . , 2007 ) . These results were used to predict the M . pharaonis coding sequence using ORFPredictor ( Min et al . , 2005 ) . We then computed the pairwise dN/dS ratios for each gene using the branch model in PAML ( v . 4 . 6 ) . Using the list of differentially expressed genes in foragers vs nest workers in the fire ant ( Manfredini et al . , 2014 ) , Fisher's exact tests were used to examine whether genes differentially expressed in these categories of workers were more likely conserved , than expected by chance . We repeated the analysis above using honey bee ( A . mellifera ) genes , except that the BLAST cutoff was lowered to 10−5 to increase the chance of identifying orthologs in the more divergent honey bee . To initially study whether evolutionary rate ( dN/dS ) , connectivity ( kTotal ) , and expression ( FPKM ) differed between behavioral categories ( nurse-upregulated , forager-upregulated , and non-differentially expressed ) , we used a Kruskal–Wallis test , adjusted for multiple comparisons ( kruskalmc function in the R package pgirmess ) . Finally , to study the main and interaction effects of connectivity , expression , and behavioral category on evolutionary rate , we used a linear model log transformed rate as the dependent variable , log transformed connectivity and expression as continuous predictors , and behavioral category as a categorical predictor . Statistical analysis was performed with R . Means are presented ± their standard deviations . p-value cutoffs of 0 . 05 were used throughout the analysis . In the case of differential gene expression , data analyses were corrected for multiple comparisons using the Benjamini-Hochberg ( FDR ) procedure ( Benjamini and Hochberg , 1995 ) . Scripts for the bioinformatic analyses , and a README explaining the workflow can be found at https://github . com/mikheyev/monomorium-polyethism . Most of the workflow and output is shown in Supplementary file 2 , with the corresponding R script shown in Source code 1 . All behavioral and gene expression data , including a MySQL database for the gene expression data have been deposited to Dryad , doi:10 . 5061/dryad . cv0q3 ( Mikheyev and Linksvayer , 2014 ) . Raw reads and the genome assembly are available at the DNA Data Bank of Japan , DDBJ BioProject PRJDB3164 .	Animal species vary widely in their degree of social behavior . Some species live solitarily , and others , such as ants and humans , form large societies . Many researchers have tried to understand the genetic changes underlying the evolution of social behavior . Some researchers suggest that it involves recycling existing genes that also have other conserved functions . Others propose that the evolution of social behavior involves completely new genes that are not found in related but solitary species . Ants are one of the best-studied social animals . An established colony can contain many 1000s of individuals that live and work together and perform different roles . The queen's job is to lay eggs , while the worker ants do everything else , including collecting food , caring for the young , and protecting the colony . In some species of ant—including the pharaoh ant—a worker's role changes as it ages . Younger workers tend to stay in the nest and nurse the brood , while older workers tend to leave the nest and forage for food . Mikheyev and Linksvayer asked: which genes are responsible for this age-based division of labor ? And how did this aspect of social behavior evolve ? First , after observing pharaoh ants from two colonies set up in the laboratory , they confirmed that workers nursing the brood were on average almost a week younger than those seen collecting food . Next Mikheyev and Linksvayer identified which genes were expressed in ants of different ages , or ants engaged in different tasks . Specific sets of genes were expressed more ( or ‘up-regulated’ ) in nurse workers , while others were up-regulated in foraging workers . Mikheyev and Linksvayer then investigated how rapidly these genes had evolved by comparing them to related genes found in other social insects ( fire ants and honey bees ) . They also determined the ‘connectivity’ of these genes by asking how many other genes showed similar expression patterns . In many organisms , how rapidly a gene evolves depends on how tightly connected its expression is to the expression of other genes; highly connected genes evolve more slowly . The genes that were expressed more in the older foraging workers were both more highly connected and more evolutionarily conserved in the other social insects . Genes that were up-regulated in the younger nurse workers were more loosely connected and rapidly evolving . Mikheyev and Linksvayer's findings show that the evolution of social behavior in animals involves both new genes , which tend to be loosely connected , and conserved genes , which tend to be more highly connected .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2015	Genes associated with ant social behavior show distinct transcriptional and evolutionary patterns
Aging stem cells lose the capacity to properly respond to injury and regenerate their residing tissues . Here , we utilized the ability of Drosophila melanogaster germline stem cells ( GSCs ) to survive exposure to low doses of ionizing radiation ( IR ) as a model of adult stem cell injury and identified a regeneration defect in aging GSCs: while aging GSCs survive exposure to IR , they fail to reenter the cell cycle and regenerate the germline in a timely manner . Mechanistically , we identify foxo and mTOR homologue , Tor as important regulators of GSC quiescence following exposure to ionizing radiation . foxo is required for entry in quiescence , while Tor is essential for cell cycle reentry . Importantly , we further show that the lack of regeneration in aging germ line stem cells after IR can be rescued by loss of foxo . In tissues with continuous cellular turnover , homeostasis is maintained by resident populations of adult stem cells . These cells both self-renew to maintain a constant pool of pluripotent cells and differentiate into a variety of cell types to replace cells that are lost to either natural wear and tear or to acute injury and insult ( Fuchs et al . , 2004 ) . As tissues age , the ability of adult stem cells to replenish tissues is impaired ( Schultz and Sinclair , 2016 ) . As a result , tissue function declines , leading to a number of different age-related deficits: grey hair is a result of impaired melanocyte maintenance ( Nishimura et al . , 2005 ) , decreased immunity results from reduced hematopoietic stem cell populations ( Linton and Dorshkind , 2004 ) , and decreases in neuron production has been implicated in the pathogenesis of a number of different neurodegenerative disorders , such as Alzheimer’s Disease ( Donovan et al . , 2006 ) . However , the mechanisms that govern the regenerative competence of aging adult stem cells remain unclear . Of particular importance is the period when age-related declines first begin to manifest – when baseline stem cell function is preserved , yet , the ability to recover from injury may be impaired . One of the most prevalent causes of injury in adult stem cells is genotoxic stress , such as that induced by exposure to ionizing radiation ( IR ) . The fly is a particularly interesting model organism with which to examine stem cell survival post IR because recent work has demonstrated that there are several cell populations that display differing levels of resistance to ionizing radiation . Previous work in the young fly has shown a remarkable ability of Drosophila germline stem cells ( GSCs ) to survive IR , even when their progeny undergo rapid apoptosis . GSCs are resistant to the apoptotic effects of ionizing radiation ( Xing et al . , 2015 ) : when flies are exposed to low doses of ionizing radiation GSCs survive , while their progeny , the transiently amplifying cells , do not . Dying GSC daughter cells secrete the ligand Pvf1 , which signals via the Tie receptor and microRNA bantam to inhibit the apoptotic machinery in GSCs ( Bilak et al . , 2014; Xing et al . , 2015 ) . After a period of quiescence , the GSCs re-enter the cell cycle and , ultimately , regenerate the germline . Knockdown of Pvf1 , a Tie ligand , in differentiating daughter cells rendered stem cells sensitive to IR , suggesting that differentiating daughter cells send survival signals to protect stem cells for future repopulation . Similar pools of IR-resistant cells have also been identified in other tissues . For example , in the larval imaginal disc , there is a population of IR-resistant cells that are able to generate viable adult tissues , even when exposed to high levels of radiation ( Verghese and Su , 2016 ) . Today , however , the ability of aging adult stem cells to maintain their resistance to ionizing radiation remains unexamined . Gaining a better understanding of stem cells' ability to recover from ionizing radiation will provide valuable insight into a wide range of physical phenomena , ranging from development of cancer therapeutics to improved aging remedies . Stem cells in Drosophila melanogaster are a versatile system with which to study age related changes in regenerative potential ( Lucchetta and Ohlstein , 2017; Fabian and Brill 2012; Resende et al . , 2017; Resnik-Docampo et al . , 2017 ) . Defects in GSC function in aged flies have been identified and are in line with hypothesized defects in aging human stem cells: decreased proliferative capacity , accumulation of DNA damage , and eventual loss of stem cells ( Zhao et al . , 2008; Kao et al . , 2015 ) . However , the initiation of the aging process , and , particularly , how GSCs early in the aging process respond to injury , remains an open area of investigation . Furthermore , the ability of aging GSCs to regenerate their resident tissue following injury has not been fully elucidated . Since a hallmark of aging stem cells is the inability to properly regenerate tissue following injury and insult ( Sharpless and DePinho , 2007 ) , it is critical to understand the relationship between the initiation of aging and the ability of stem cells to recover from injury , such as following exposure to ionizing radiation . Here , we identify and mechanistically dissect a regeneration defect in aging GSCs following exposure to ionizing radiation . Aging GSCs survive exposure to radiation , but exhibit a defect in cell cycle reentry upon completion of DNA repair . We further show that young GSCs enter a 24 hr period of quiescence following exposure to ionizing radiation before reentering the cell cycle and beginning to regenerate the germline . In our investigation of the mechanisms governing this process , we identify the foxo-encoded transcription factor and the human mTOR ortholog , Tor as important regulators for GSC entry and exit of quiescence following exposure to ionizing radiation , respectively . Lastly , we show that the regeneration defect of aging GSCs can be rescued by knockdown of foxo , suggesting that misregulation of foxo may underlie the regenerative decline with age . In the Drosophila ovary , at the apical tip of each germarium are two to three germline stem cells ( GSCs ) in direct contact with their somatic niche ( Spradling 1993 ) . These GSCs undergo asymmetric rounds of self-renewing divisions to give rise to a new stem cell and to a transiently amplifying cell ( cystoblast ) that undergoes four incomplete divisions generating an interconnected 16 cell cyst , of which one cell will eventually become an oocyte . Intricate interactions between GSCs and somatic cells allow for GSC maintenance in the niche ( Fuchs et al . , 2004; Ward et al . , 2006 ) . Germline stem cells can be identified by their proximity to the cap cells in the niche and the prominent foci of adducin staining , labeling the subcellular structures called spectrosomes , while progeny can be identified by a branched focus of adducin , known as the fusome , in cells that do not reside within the niche ( Figure 1A . ) . As GSCs progress through the cell cycle , they alternately have an elongated or a round spectrosome , the morphology of which can be used to identify dividing GSCs ( Figure 1A , ( de Cuevas and Spradling , 1998 ) . GSCs in the female Drosophila ovariole lose replicative capacity with age ( Pan et al . , 2007; Zhao et al . , 2008; Kao et al . , 2015 ) , however , the early steps in aging and GSC ability to survive exposure to ionizing radiation during the early aging process has not been probed . We first asked whether aging GSCs survive radiation exposure . Since it has been shown that young GSCs survive exposure to IR and are able to regenerate the germline by one week following exposure to IR ( Xing et al . , 2015 ) , we probed the system to identify the earliest time points where we could observe a defect in the aging GSCs’ ability to recover from exposure to IR . We found that at 4 weeks , recovery from IR and regeneration was normal , while a defect could be observed when 6 week old animals were irradiated . We exposed 4- and 6 week old wild type flies to 50 Grays of radiation and quantified the number of GSCs in unirradiated flies and compared them to the number of GSCs in germaria of flies one week following irradiation . We found that , although 4 and 6 week old germaria lose a small number of GSCs one week post IR the majority of 4 and 6 week old germaria still had 1 to 2 GSCs one week following irradiation ( Figure 1B ) , indicating that the Tie-mediated protective mechanism remains mainly intact in aging germaria . Next , we assayed the level to which the GSCs were able to regenerate the germarium following exposure to ionizing radiation . We visualized GSC progeny with adducin staining and compared the number of germaria that had four or more progeny to those that had fewer than four progeny in unirradiated flies and in germaria of flies one week following exposure to irradiation . We found that , while at 4 weeks , the number of germaria with progeny was not significantly different before and after exposure to irradiation , at 6 weeks , the number of germaria with progeny one week following irradiation was significantly lower than in the germaria of unirradiated flies ( Figure 1C ) . This suggests that , while aging GSCs are able to survive exposure to irradiation , they are unable to re-enter the cell cycle and regenerate the germarium in a timely manner . We confirmed that regeneration was impaired by assaying the percentage of GSCs with elongated spectrosomes , which is an indication of GSC division . We found that levels of spectrosome elongation were similar before and after irradiation in the 4 week old animals , however , in the 6 week old animals , there was a significant decrease in the percentage of GSCs with elongated spectrosomes one week after irradiation ( Figure 1D ) . We further confirmed that regeneration was impaired in 6 week old animals by comparing the number of adults produced by 4 and 6 week old irradiated animals . We found that while irradiated 4 week old animals produced less adults than unirradiated animals of the same age , this defect was much more pronounced in irradiated 6 week old animals ( Figure 1—figure supplement 1A , B ) . Taken together , our data suggest that aging GSCs are able to survive exposure to low IR , but are unable to reenter the cell cycle and regenerate the germline . To assay the general fitness of the 6 week old flies , we compared the survival rates of 6 week old flies , both irradiated and unirradiated . We found no significant difference in the life span of the irradiated and unirradiated flies ( Figure 1E ) . This indicates that we have found a time when GSCs have begun to age and show deficits in regenerative capacity , but dramatic aging phenotypes are not yet detectable at the organismal level . Hence , our analysis will allow us to understand the earliest processes in stem cell aging . DNA damage can inhibit cell cycle progression ( Bunz et al . , 1998; Reinhardt and Schumacher , 2012 ) , and increases in levels of DNA damage have been reported in aged GSCs ( Kao et al . , 2015 ) . To assay whether the observed delay of cell cycle reentry in aging GSCs following irradiation was due to delays in DNA damage repair , we exposed 6 week old flies to 50 Gys of ionizing radiation . We then dissected ovaries from flies 30 min , 24 hr , and 7 days following irradiation and compared levels of DNA damage in GSCs to those of unirradiated flies , as visualized by γH2AV staining ( Figure 2A–C ) . We compared the number of GSCs with high , moderate , or minimal levels of DNA damage at these time points . We found that DNA damage peaked 30 min following IR , with a majority of GSCs showing high levels of γH2AV staining ( Figure 2B , D ) . However , by 24 hr following IR , levels of DNA damage had returned to baseline levels , similar to those in unirradiated flies ( Figure 2C–E ) . Additionally , by 7 days post-irradiation , there was no significant difference in the level of DNA damage compared to unirradiated flies ( Figure 2D ) . This indicates that DNA damage repair has concluded , even though the aging GSCs remain unable to regenerate the germline , suggesting that additional mechanisms must be responsible for the aging defect we identified in 6 week old GSCs . Having identified a regeneration defect in aging GSCs , we next investigated the timing of IR induced cell cycle exit and reentry in young , healthy flies . We exposed 2–7 day old flies to 50 Gys of ionizing radiation and compared levels of GSC division and regeneration to unirradiated flies at 24 hr intervals . We visualized branched fusomes and spectrosomes via adducin staining ( Figure 3B–E ) . In order to assay the rates of GSC division , we compared the morphology of the spectrosomes in GSCs from flies that had been irradiated to unirradiated flies . We quantified the percentage of GSCs with elongated spectrosomes , as an indicator of GSC cellular division ( Figure 3B , D , yellow arrow ) . We observed a significant decrease in the percentage of GSCs with elongated spectrosomes one day post-IR ( Figure 3F ) . By two days following irradiation , the percentage of GSCs with elongated spectrosomes had returned to baseline ( Figure 3F ) . This suggests that when well fed , young animals are exposed to low doses of irradiation , GSCs enter a brief , approximately 24 hr period of quiescence . Similarly , we quantified the number of regenerated germaria by quantifying the number of germaria with germ line cysts containing branched fusomes ( Figure 3G ) . Unlike GSCs , transiently amplifying cells do not survive exposure to ionizing radiation and the number of new daughter cells can , therefore , be used as an indirect measure of GSCs’ regeneration capacity following irradiation damage ( Xing et al . , 2015 ) . There was a significant decline in the percentage of germaria with GSC daughters containing branched fusomes one and two days post-IR ( Figure 3G ) . By 3 days , post-IR , the percentage of regenerated germaria had dramatically increased , with complete recovery achieved by 4 days post-IR . This suggests that GSCs give rise to progeny by 3 days post-IR , which is in line with our observation that GSCs begin dividing around two days post-IR . Taken together , our data indicate that GSCs enter an approximately 24 hr period of quiescence after exposure to ionizing radiation before returning to the cell cycle and regenerating the germline . Notch signaling also plays an essential role in the development and maintenance of the Drosophila germline stem cell niche . Niche cells and GSCs communicate with one another via the Delta and Serrate Notch ligands to regulate various niche features , including niche size and GSC number ( Ward et al . , 2006; Song et al . , 2007 ) . Abrogation of Notch signaling by expressing a nos-Gal4-inducible RNAi construct against neuralized ( neur ) , a ubiquitin ligase which mediates the internalization and subsequent activation of the Delta and Serrate Notch ligands in the germline , resulted in a complete loss of GSCs , even before exposure to ionizing radiation ( data not shown ) , confirming the essential role of Notch signaling in GSC maintenance ( Ward et al . , 2006; Song et al . , 2007 ) . To study if supernumerary GSCs follow the wild type GSC kinetics of post-IR quiescence , we drove overexpression of Delta in the germline using the Gal4 system . Nos-Gal4 > Delta germaria showed an increased number of spectrosome marked cells , which we confirmed were GSCs via expression of the TGFβ target , Dad ( Figure 3H–I ) . This indicates that the expanded TGFβ signaling from niche induced extranumerary GSCs , as seen previously ( Ward et al . , 2006; Song et al . , 2007 ) . We exposed Delta overexpression flies to ionizing radiation ( 50 Gys ) and dissected their ovaries 1 , 2 and 7 days post IR . GSCs in the expanded niche enter and exit quiescence in a timely manner ( Figure 3H–L ) . However , while the somatic niche remained large , GSC number was reduced one day after IR ( Figure 3—figure supplement 1 ) , suggesting that the protective signal from daughter cells cannot penetrate to protect all the supernumerary GSCs after exposure to ionizing radiation . To confirm that DNA damage repair kinetics are not substantially different in young flies and old flies , we assayed levels of DNA damage via γH2AV staining in the germaria of 2–7 day old flies exposed to ionizing radiation ( Figure 4A–H ) . We quantified the percentage of GSCs with high , moderate , or no/minimal levels of γH2AV at 30 min , 12 hr , and 24 hr after exposure to ionizing radiation and compared this to levels of γH2AV in unirradiated germaria ( Figure 4I ) . We found that there was a significant increase in the percentage of GSCs with high levels of DNA damage 30 min post-IR ( Figure 4C , D , I ) . By 12 hr post-IR , a majority of the germaria had repaired DNA damage to a moderate amount: only 8% showed high levels of DNA damage , while 83% had moderate levels of DNA damage ( Figure 4E–F , I–J ) . By 1 day following radiation exposure , only 34% of GSCs had moderate levels of yH2AV staining , with 66% of GSCs returned to baseline levels of DNA damage ( Figure 4G , H , I , J ) . This suggests that DNA damage repair kinetics in young flies resemble those of the aging fly , supporting our previous findings that alterations in DNA damage repair kinetics alone cannot account for the regeneration defect in aging GSCs . We next asked what mechanisms are involved in regulating IR-induced quiescence in GSCs . We first probed the role of the G1 checkpoint in IR-induced quiescence by manipulating levels of the p21 ortholog , dacapo . We found that , while overexpression of dacapo was sufficient to prolong IR-induced quiescence ( Figure 5—figure supplement 1B ) , there was no significant difference in the ability of GSCs to enter quiescence when dacapo levels were reduced ( Figure 5—figure supplement 1E , F ) . This suggests that dacapo is not required for GSCs to enter quiescence after a radiation challenge , suggesting that the G1 checkpoint is not where GSCs arrest following exposure to IR . Additionally , we examined the role of the DNA damage sensing machinery in regulating IR-induced quiescence . We found that when the CHK2 ortholog , loki , was knocked down via RNAi , it impaired the ability of GSCs to enter quiescence ( Figure 5—figure supplement 1G and H ) , consistent with recent work demonstrating the vital role of loki in regulating GSC survival following exposure to high levels of IR ( Ma et al . , 2016 ) . Thus , we worked to identify the functional machinery downstream of CHK2 that regulates the stem cell quiescence . foxo is a key player in the cellular response to IR ( Chung et al . , 2012; Xing et al . , 2015 ) . To probe whether foxo was required for IR-induced GSC cell cycle exit and reentry , we knocked down foxo in the germline by crossing UAS-Dcr-2; nos-Gal4 flies to two independent UASp-foxo RNAi lines . We then assayed the morphology of GSCs' spectrosomes at 1 and 2 days post-IR in nos-GAL4 > foxo RNAi flies and compared them to unirradiated GSCs . We found that , while in UAS-Dcr-2; nos-Gal4 control flies , there is a dramatic decrease in the percentage of GSCs with elongated spectrosomes 1 day post-IR , foxo deficient GSCs in both RNAi lines kept dividing at a normal rate ( Figure 5A , B ) . Additionally , the percentage of germaria with branched fusomes 1 day post-IR is increased in foxo RNAi flies ( Figure 5C; Supplementary file 1A , B ) , further strengthening our finding that knockdown of foxo eliminates IR-induced quiescence . This suggests that foxo is required for GSCs to initiate IR-induced quiescence and withdraw from the cell cycle . Foxo has been shown to regulate Tor in C . elegans , Drosophila , and mammalian systems ( Puig et al . , 2003; Jia et al . , 2004; Chen et al . , 2010 ) . Since Tor signaling is known to modulate both Drosophila longevity and GSC division ( Bjedov et al . , 2010; LaFever et al . , 2010 ) we analyzed its potential role in cell cycle regulation following exposure to IR . To probe whether Tor is required for IR-induced GSC cell cycle exit or reentry , we knocked down Tor in the germline using a nos-GAL4 driver to express a Tor RNAi construct under UAS control ( nos-GAL4 > Tor RNAi ) . We then assayed Tor mutant GSC division capacity by analyzing the morphology of spectrosomes and the number of daughters produced at 1 and 2 days post-IR and compared them to control and unirradiated GSCs . We found that when Tor is knocked down , there is an even larger decrease in the percent of GSCs with elongated spectrosomes one day post-IR than in control animals , suggesting a higher penetrance in cell cycle exit ( Figure 5A , B ) . Furthermore , the percentage of Tor RNAi GSCs with elongated spectrosomes and the number of GSC daughters remained decreased two days post-IR , when control GSCs have reentered the cell cycle ( Figure 5C; Supplementary file 1A , B ) , suggesting a dramatic delay in the reentry to the self-renewing cell cycle and regenerative capacity in Tor mutant GSCs . We also probed the role of Tor signaling pharmacologically with rapamycin . Rapamycin is a potent inhibitor of the TORC1 complex , preventing phosphorylation of Tor’s downstream targets ( Sabatini et al . , 1995 ) . Following irradiation , wild type flies were fed grape juice with either rapamycin ( 200 µM ) or vehicle for two days . There was a significant decrease in the percentage of GSCs with elongated spectrosomes 2 days post-IR with rapamycin treatment ( Figure 6A–C ) . Taken together , these data suggest that Tor is required for GSC exit from quiescence and cell cycle reentry post-IR . Finally , we probed the question of whether IR-induced quiescence is protective to GSCs . When nos-Gal4 > foxo RNAi flies were exposed to a secondary dose of ionizing radiation 24 hr following the initial dose ( Figure 5—figure supplement 2A ) , we found that there was a decrease in the number of GSCs per germaria in foxo RNAi flies ( Figure 5—figure supplement 2C ) . This difference cannot be attributed to foxo reduction alone , since unirradiated nos-Gal4 > foxo RNAi ovaries have a normal number of GSCs per germarium . This suggests that foxo-mediated IR-induced quiescence is important for GSC survival . Since we identified opposing roles for foxo and Tor in regulating IR-induced quiescence , we next asked whether these two signaling components operated independently or in conjunction with each other . To visualize foxo activity , we stained for Foxo protein and to assay levels of Tor activity , we stained for phosphorylated ribosomal protein S6 ( p-S6 ) , a downstream effector of TORC1 . We compared levels of Foxo and p-S6 staining in young , wild type files following exposure to ionizing radiation . We observed a dramatic increase in the level of Foxo in GSCs’ nuclei 1 day post-IR ( Figure 7A ) . Levels of Foxo staining returned to baseline ( Figure 7D ) by 2 days post-IR . Phospho-S6 staining showed a complimentary pattern to Foxo staining: while foxo is highly expressed at the anterior tip of germaria and the GSCs , p-S6 levels are high in 8- and 16 cell cysts towards the posterior end of germaria , suggesting a possible regulatory role of Tor activity by foxo ( Figure 7B ) . To test this , we reduced foxo levels and measured Tor activity by analyzing p-S6 patterns . When foxo is depleted via nos-Gal4-induced RNAi , the level of p-S6 staining increases and is observed closer to the anterior tip of the germaria and GSCs , which is not observed in wild type animals ( Figure 7C ) . p-S6 staining was completely absent in germaria of nos-Gal4 > Tor RNAi flies ( Figure 7F ) , confirming that p-S6 is a reliable measure of Tor activity . Together , this suggests that Tor and Foxo activity are spatially segregated due to an antagonistic relationship between the activity of these two proteins ( Figure 7G ) . In particular , these data show that foxo can repress the TORC1 target , p-S6 in the Drosophila ovary . Since we identified foxo as a critical regulator of IR- quiescence , we next asked whether knockdown of foxo in the aging GSC could rescue the observed aging regeneration defect . We aged nos-Gal4 > foxo RNAi flies to 6 weeks and exposed them to 50 Gys of ionizing radiation . We quantified the number of GSCs per germaria , as well as the number of germaria with four or greater progeny in unirradiated and one week post-IR flies . We found that , compared to unirradiated flies , there was no significant difference in the number of GSCs/germaria in 6 week old nos-Gal4 > foxo RNAi flies one week following irradiation ( Figure 8A , C ) . Strikingly , we also found that one week following exposure to IR , nos-Gal4 > foxo RNAi flies showed evidence of germline regeneration , with equal numbers of germaria with greater than four progeny when compared to their unirradiated counterparts ( Figure 8B , D ) . We also observed large 8 cell cysts one week post-IR in 6 week old nos-Gal4 > foxo RNAi flies ( Figure 8B ) indicating robust and extensive regeneration of the germline . This developmental stage is never observed in 6 week-old wild type flies one week post-IR . These findings suggest that knockdown of foxo is sufficient to relieve the aging regeneration defect: aging flies with reduced levels of foxo are able to regenerate the germline within a week , while wild type flies cannot ( Figure 8F ) . To study foxo’s mode of function in the context of aging , we probed Tor signaling , a Foxo target repressed post-injury in young animals . Aging flies expressing a UASp RNAi construct against foxo showed a dramatic increase in germline Tor activity , as measured by p-S6 antibody staining ( Figure 8E ) . This suggests that Foxo represses Tor activity during aging and that overactivation of Foxo may account for the inability of aging GSCs to regenerate following exposure to IR ( Figure 9 ) , as evidenced by the ability of the aging germline to regenerate with decreased levels of Foxo . Adult stem cells experience a decrease in regenerative potential with age that results in a decrease in the ability of adult tissues to repair themselves following injury or insult . We have now identified the earliest time at which aging Drosophila germline stem cells lose the ability to appropriately recover from exposure to sublethal doses of ionizing radiation ( IR ) and dissect the mechanism for this process . Following exposure to IR , most aging GSCs survive , but fail to reenter the cell cycle and regenerate the germline , a process that is activated in young flies post IR . This is not due to a defect in DNA damage repair , as DNA damage repair concludes in a timely manner , even though the aging GSCs fail to return to the cell cycle . We have now identified two key regulators for IR induced quiescence: foxo and Tor . These two genes have opposing roles in regulating GSC cell cycle , exit and reentry after IR , respectively . Furthermore , Tor inactivation by RNAi or Rapamycin treatment induces a premature GSC aging phenotype , impairing Tor-dependent regeneration post injury . Conversely , knocking down foxo in aging animals rescues the aging phenotype , allowing GSCs to regenerate the germline , as observed in young flies . Finally , we show that foxo and Tor have opposing patterns of expression in the germarium and depletion of foxo leads to increases in Tor activity . This suggests that foxo regulates post-IR quiescence and cell cycle reentry by regulating Tor activity . Importantly , we show that loss of foxo rescues the GSC age-related regeneration phenotype due to IR . Overall , this study shows that IR induced quiescence is regulated by foxo and the mTOR ortholog , Tor , and suggests that upregulation of foxo and misregulation of Tor signaling in aging adult stem cells may be responsible for the decline in regenerative capacity following injury or insult ( Figure 9 ) . Aging adult stem cells are unable to regenerate injured tissue as effectively and efficiently as young stem cells ( Schultz and Sinclair , 2016 ) . However , it has remained an open area of investigation as to whether this is due to a loss of adult stem cells with age or whether this is due to a decrease in the ability of adult stem cells to regenerate appropriately . Our work shows that the anti-apoptotic protective mechanisms ( Xing et al . , 2015 ) that shield adult stem cells from death remain mainly intact , but the aging GSCs are unable to reenter the cell cycle following IR-induced quiescence . Aging is a complex process , involving the cumulative decline of multiple cell types . Defects in the replicative potential of old GSCs have been reported by other groups ( Zhao et al . , 2008; Tseng et al . , 2014; Kao et al . , 2015; Rauschenbach et al . , 2015 ) . However , our work expands our understanding of the onset of aging in a unique way . Here , we identify the earliest time point at which defects can be detected in GSC proliferation in an injury model . Before the induction of IR-mediated quiescence in our aging flies , rates of GSC division , as well as the number of GSCs per germaria were similar to that seen in young , healthy flies . Defects were only readily observed following exposure to IR . This suggests that baseline levels of GSC function remain unperturbed , however , the GSCs are unable to recover successfully from insult . This leads us to believe that we have identified a defect early in the initiation of the aging process . Therapeutically , this is a very important window , as it allows us to identify times when an intervention may be useful in helping to slow the progression of aging , or prevent it from initiating in the first place , rather than attempting to reverse it late in the process . High doses of irradiation have been shown to lead to GSC loss ( Ma et al . , 2016 ) . We specifically utilized a relatively low dose of ionizing radiation , in order to induce damage , but not lead to GSC loss ( Xing et al . , 2015 ) and to probe the ability of aging stem cells to recover from an injury that should be surmountable were the cells functioning properly . We were able to identify critical roles for two known proteins involved in tissue homeostasis: Foxo for cell cycle withdrawal and Tor for cell cycle reentry . foxo has been well documented as a regulator of stem cell self-renewal and quiescence ( Demontis and Perrimon , 2010; Xing et al . , 2012; Eijkelenboom and Burgering , 2013; Gopinath et al . , 2014; Xing et al . , 2015 ) . Notably , foxo tends not to be active during normal physiology , but rather during stressful conditions , when it responds to and counteracts a stressor in order to maintain homeostasis ( Kenyon , 2010; Eijkelenboom and Burgering , 2013 ) . Here , we show that foxo activity is required in Drosophila GSCs in order for them to withdraw from the cell cycle following exposure to ionizing radiation . There are multiple ways that Foxo may be able to sense the damage caused to the cell by irradiation . In response to the presence of reactive oxygen species , JNK-mediated phosphorylation of Foxo can cause its translocation to the nucleus ( van den Berg and Burgering , 2011 ) . Foxo can also be the target of multiple pathways that are responsive to DNA damage: Foxo is a target of phosphorylation by ATM ( Matsuoka et al . , 2007 ) and the MAPK pathway ( Kress et al . , 2011 ) both of which have been shown to be activated by DNA damage . Lastly , Foxo is capable of directly sensing cellular redox status via oxidation and reduction of amino acids , particularly cysteine ( Dansen et al . , 2009 ) . CHK2 , a highly conserved checkpoint kinase , controls DNA repair , cell cycle arrest and apoptosis following DNA damage . The fly CHK2 ortholog , loki , has been shown to mediate GSCs’ self-renewal and differentiation following high doses of ionizing radiation ( Ma et al . , 2016 ) . Here we show that depletion of loki in the germline prevents GSCs from entering quiescence following exposure to low doses of ionizing radiation . Loki’s ability to sense DNA damage and interact with Foxo via the ATM-CHK2-p53 complex ( Chung et al . , 2012 ) could explain how GSCs know to activate Foxo and withdraw from the cell cycle following IR-induced double stranded breaks . Notably , p53 , another component of the ATM-CHK2-p53 complex , has also been shown to regulate GSC irradiation-induced quiescence ( Wylie et al . , 2014 ) although how p53 interacts with Foxo in this context remains unclear . It is possible that any of these , or the combination of multiple of these systems sense the damage to the GSCs caused by the ionizing radiation and translocate Foxo to the nucleus , initiating IR-induced quiescence . Mechanistic target of rapamycin ( mTOR ) signaling has been implicated in a number of different age-related functions , from extension of lifespan ( Vellai et al . , 2003; Harrison et al . , 2009; Bjedov et al . , 2010; Laplante and Sabatini , 2012; Bitto et al . , 2016 ) to germline stem cell self-renewal ( LaFever et al . , 2010; Sun et al . , 2010 ) , induction of a diapause like quiescent state ( Bulut-Karslioglu et al . , 2016 ) and muscle satellite cell activation following injury ( Rodgers et al . , 2014 ) . We found that Tor signaling was required in order for GSCs to reenter the cell cycle and regenerate the germline following exposure to IR . The sensitivity of wild type GSC proliferation to treatment with rapamycin after IR indicates that this could be mediated via the Tor complex 1 ( TORC1 ) since rapamycin preferentially targets TORC1 . We cannot completely rule out a role for Tor complex 2 ( TORC2 ) in GSCs’ quiescence since rapamycin treatment has been shown to affect TORC2 activity by keeping Tor associated with TORC1 ( Sarbassov et al . , 2006; Lamming et al . , 2012 ) . Further studies will focus on investigating the roles of both TORC1 and TORC2 downstream effectors in GSC quiescence . GSCs with decreased levels of Tor activity are unable to reenter the cell cycle post-IR , which is unlikely to be a general consequence of Tor inhibition inhibiting GSC division . In a number of different experiments , we observed a more pronounced defect in GSC proliferation in the context of recovery from injury post-IR than at baseline . This indicates that , while Tor might play a role in regulating stem cell division and self-renewal under normal physiological conditions , it likely has an additional injury-specific role in helping to replenish adult tissues that have been damaged , either by natural wear and tear or due to disease or injury . Given Tor’s ability to regulate translation , nucleotide synthesis , autophagy , lipid synthesis , and proteasome assembly , ( Laplante and Sabatini , 2012 ) it will be important to dissect which of these or other cellular processes are required for GSCs’ exit from quiescence . It is also quite striking that inhibition of Tor resembles the defect observed in aging GSCs , while at an organismal level , inhibition of Tor increases lifespan , suggesting a slowing of the aging process . This would indicate that Tor inhibition , albeit beneficial at an organismal level , may damage stem cells’ capacity to regenerate tissue after injury . This is a particularly important implication of our findings , given the increasing number of anti-aging studies involving rapamycin ( Fan et al . , 2015; Bitto et al . , 2016 ) . Mutations in insulin receptor ( InR ) in Drosophila and insulin-like growth factor ( IGF1 ) in mice , result in Foxo activation and significant lifespan extension ( Clancy et al . , 2001; Tatar et al . , 2001; Bluher et al . , 2003; Holzenberger et al . , 2003; Webb et al . , 2016 ) . In humans , single-nucleotide polymorphisms ( SNPs ) in the FOXO3 locus have been associated with extraordinarily long lifespans ( Morris , 2005 ) , though the mechanism for this remains elusive . Our study identifies a novel foxo-dependent stem cell defect in aged animals in which elevated foxo activity prevents GSCs from re-entering the cell cycle and regenerating the germline after a challenge . In contrast to other studies showing the benefits of high levels of foxo activity , we show , for the first time , that elevated levels of foxo activity , albeit beneficial in terms of lifespan extension , are detrimental to stem cell function in the context of tissue regeneration during aging . There are several reasons why pathologically high levels of foxo might prevent tissue regeneration in old animals . A meta-analysis of mouse Foxo targets that change with age has revealed that several cell cycle genes , such as the evolutionarily conserved cyclin-dependent kinase 4 ( Cdk4 ) , which controls the G1 to S transition , and several ribosomal proteins , which are directly involved in protein translation , are misregulated in aging ( Webb et al . , 2016 ) . In our study , we show how , after IR exposure , foxo and Tor have opposing patterns of expression in young animals . We also demonstrate how reducing foxo levels via RNAi increases p-S6 levels in young and aging animals . This strengthens the idea that foxo and Tor signaling interact with one another to regulate GSC division following injury and that misregulatin of this crosstalk might contribute to stem cell aging . Our study shows how Foxo misregulation may impair aging GSCs’ regeneration potential . foxo’s ability to repress Tor could shed light on aging GSCs’ inability to resume division following insult . Though the mechanism with which foxo and Tor interact in the context of aging remains elusive , previous studies have already probed the relationship between these signaling pathways . Foxo has been shown to repress Tor signaling by allowing TSC ( Tuberous Sclerosis Complex ) to localize to the lysosome ( Menon et al . , 2014 ) . At the lysosomal membrane , TSC is then able to inhibit Rheb , an essential activator of mTORC1 . Other studies have shown that Foxo is able to inhibit mTORC1 by reducing Raptor levels ( Jia et al . , 2004 ) or by promoting the transcription of Sestrin 3 and Rictor ( Chen et al . , 2010 ) . Notably , Tor signaling can also inhibit Foxo activity by upregulating SGK ( Saxton and Sabatini , 2017 ) , an AGC-kinase shown to inhibit Foxo . This suggests the possibility of a negative feedback loop between these signaling pathways . In the future , it will be of vital importance to dissect the crosstalk between foxo and Tor signaling to understand why GSCs lose their regeneration potential with age . The following stocks were obtained from the Bloomington Drosophila Stock Center at Indiana University: w[1118] ( RRID:BDSC_3605 ) , P[UAS-Dcr-2 . D]1 , w1118; P[GAL4-nos . NGT]40 ( RRID:BDSC_25751 ) , UASp-foxoRNAi ( RRID:BDSC_32427 and RRID:BDSC_32993 ) , UASp-TorRNAi ( RRID:BDSC_35578 ) , UASp-ThorRNAi ( RRID:BDSC_36815 ) , UASp-dmRNAi ( RRID:BDSC_43962 ) , UASp-dapRNAi ( RRID:BDSC_36720 ) , UASp-LokiRNAi ( RRID:BDSC_64482 ) . The following stocks were previously generated for and described in Ward et al . , 2006: UASp-Delta/CyO , UASp-Delta/CyO; Dad-GFP/TM3 , UASp-Delta/CyO; Ly/TM3 . The following stocks were previously generated for and described in Yu et al . , 2009: pin/CyO;hs-dap-7-7 , hsFLP; FRT42B GFP/CyO , FRT42B/CyO , FRT42B dap4 w1118 flies were used as a control , unless noted otherwise . Flies were cultured at 25° C on standard cornmeal-yeast-agar medium , augmented with wet yeast . In aging experiments , flies were transferred to fresh vials without wet yeast every 2–3 days . Young and old flies were given wet yeast two days prior to irradiation . After feeding on standard cornmeal-yeast-agar medium augmented with wet yeast paste for two days , young and old flies were transferred to empty vials and treated with 50 Gγs of gamma-irradiation . A Cs-137 Mark I Irradiator was used to administer the proper irradiation dosage , according to instructed dosage chart . Post-treatment animals were transferred back to fresh food with wet yeast and maintained at 25° C until dissection . Following irradiation , flies were place in an empty vial with filter paper soaked in grape juice with either 200 µM rapamycin or DMSO dissolved in it . Following irradiation , 10 females were placed in a new vial with 5 young , unirradiated wild type male flies . Flies were transferred to new vials every 2–3 days and the death of any flies was noted . Vials from flies 5–7 days post-IR were collected and the number of progeny hatched per female was calculated . GSCs clones were induced via the heat shock FLP-FRT system . Young flies ( 2–3 days old ) of the following genotypes hsFLP; FRT42B GFP/FRT42B , hsFLP; FRT42B GFP/FRT42B dap4 , were heat shocked in a 37° C water bath for 45 minutes hour once a day for two consecutive days Heat shocked flies were given fresh food and yeast paste every other day until dissection and stored at 25° C for the duration of the experiment . Ovaries were fixed in 4% paraformaldehyde for 15 min , rinsed in PBT ( PBS containing 0 . 2% Triton X-100 ) , and blocked in PBTB ( PBT containing 0 . 2% BSA , 5% normal goat serum ) for at least one hour at room temperature . Samples were stored up to 72 hr at 4° in PBTB . The following primary antibodies were used: mouse anti-adducin ( RRID:AB_528070 1:30 ) , mouse anti-Lamin C ( RRID:AB_528339 1:30 ) rabbit anti-γH2AV ( RRID:AB_828383 1:200 ) , rabbit anti-p-S6 ( RRID:AB_916156 1:200 ) , rabbit anti-foxo ( generous gift from Pierre Léopold 1:200 ) . Ovaries were incubated with primary antibodies for either 1 . 5 hr at room temperature or overnight at 4° . After washes with PBT , secondary fluorescence antibodies were utilized including anti-rabbit Alexa 488 ( RRID:AB_221544 1:250 ) and anti-mouse 568 ( RRID:AB_2535773 1:250 ) for 1 . 5–2 hr at room temperatures in the dark . DAPI was added to one of the final washes to visualize cells’ nuclei . The samples were mounted in glycerol and analyzed on a Leica SPE5 confocal laser-scanning microscope . All data are presented as the mean of at least three independent experiments ( n ≥ 3 ) with the standard error of the mean ( SEM ) indicated by error bars , unless otherwise indicated . Statistical significance was determined using Student's t test ( for two groups ) or ANOVA with the appropriate post hoc test ( for more than two groups ) . Data were compiled using Excel 2013 software and analyzed using Excel ( version 2013 for Windows; Microsoft , Seattle , WA , USA ) or the Astatsa Online Web Statistical Calculator ( astatsa . com , Philadelpha , PA , USA ) .	Stem cells are unspecialized cells that have the unique ability to replace dead cells and repair damaged tissues . To give rise to new cells , stem cells need to divide . This process , known as the cell cycle , includes several stages and is regulated by many different genes . For example , in many organisms , a gene called foxo helps cells respond to stress and to regulate the cell cycle and cell death . Defects in this gene have been linked to age-related diseases , such as cancer and Alzheimer’s disease . Previous research has shown that foxo can also regulate Tor – a gene that helps cells to divide and grow . As we age , stem cells become less efficient at regenerating tissues , especially after exposure to toxins and radiation . However , until now , it was not known how stem cells control their division after injury and during aging , and what role these two genes play in injured and aging stem cells . Now , Artoni , Kreipke et al . used germline stem cells from fly ovaries to investigate how young and old stem cells respond to injury . In young flies , foxo paused the cell cycle of the damaged stem cells . After 24 hours , Tor was able to overcome the action of foxo , and the stem cells resumed dividing and regenerating the damaged tissue . However , in old stem cells , foxo and Tor were misregulated and the stem cells could not restart dividing or repairing tissue after injury . When the levels of foxo in old stem cells were experimentally reduced , their ability to regenerate the tissue was restored . These discoveries provide new insights into how stem cells respond to injury and suggest that stem cell aging may be a reversible process . A next step will be to investigate why foxo and Tor are misregulated during aging and how these two genes interact with each another . In future , this could help develop new anti-aging therapies that can restore the body’s natural ability to repair itself following injury . Moreover , since cancer cells can become resistant to conventional cancer treatment by withdrawing from the cell cycle , developing new treatments that target foxo and Tor could help beat cancer and prevent its reoccurrence .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "cell", "biology" ]	2017	Loss of foxo rescues stem cell aging in Drosophila germ line
Persistent reservoirs remain the major obstacles to achieve an HIV-1 cure . Prolonged early antiretroviral therapy ( ART ) may reduce the extent of reservoirs and allow for virological control after ART discontinuation . We compared HIV-1 reservoirs in a cross-sectional study using polymerase chain reaction-based techniques in blood and tissue of early-treated seroconverters , late-treated patients , ART-naïve seroconverters , and long-term non-progressors ( LTNPs ) who have spontaneous virological control without treatment . A decade of early ART reduced the total and integrated HIV-1 DNA levels compared with later treatment initiation , but not reaching the low levels found in LTNPs . Total HIV-1 DNA in rectal biopsies did not differ between cohorts . Importantly , lower viral transcription ( HIV-1 unspliced RNA ) and enhanced immune preservation ( CD4/CD8 ) , reminiscent of LTNPs , were found in early compared to late-treated patients . This suggests that early treatment is associated with some immunovirological features of LTNPs that may improve the outcome of future interventions aimed at a functional cure . A reservoir of long-lived latently HIV-1 infected cells is established early in the course of the infection . It persists despite suppressed viremia in patients undergoing effective antiretroviral therapy ( ART ) and fuels viral rebound upon treatment discontinuation ( Wong et al . , 1997; Finzi et al . , 1997; Chun et al . , 1997; Finzi et al . , 1999; Fernandez et al . , 2005; Alexaki et al . , 2008 ) . Not only is this reservoir present in blood , but also in tissues such as lymphoid organs , the gut and potentially the central nervous system ( Chun et al . , 2008; Sturdevant et al . , 2015; Bednar et al . , 2015 ) . The mechanisms underlying HIV-1 persistence have not been fully elucidated . Although an initial decay of these reservoirs is observed after ART intervention , it is assumed that replenishment may occur through clonal proliferation of infected CD4 T cells during ART ( Chomont et al . , 2009; Josefsson et al . , 2013; Maldarelli et al . , 2014; Murray et al . , 2014 ) or through residual virus production despite suppressive ART ( Chun et al . , 2008; Buzon et al . , 2010; Hatano et al . , 2013a ) possibly in sanctuary sites where ART penetration is suboptimal ( Yukl et al . , 2010; Fletcher et al . , 2014 ) . Low levels of viral reservoirs have been associated with an absence of viral rebound after treatment discontinuation in several case reports and the Visconti cohort , suggesting the possibility of post-treatment virological control even in the presence of viral reservoirs ( Salgado et al . , 2011; Van Gulck et al . , 2012; Saez-Cirion et al . , 2013; Kinloch-de Loes et al . , personal communication ) . Achieving such a long-term control of HIV-1 replication in the absence of ART is widely defined as a functional cure ( Saag and Deeks , 2010; Fauci and Folkers , 2009 ) . A low saturation of viral reservoirs facilitated by early treatment initiation might be a necessary condition , although not in itself sufficient for post-treatment virological control ( Saez-Cirion et al . , 2013; Van Gulck et al . , 2012 ) . Recent evidence indicates that the interplay between virological and immunological parameters is likely to be fundamental to achieve this goal ( Cellerai et al . , 2011 ) . A sustained remission from viremia rebound seems a more realistic prospect in terms of HIV-1 cure research in the short term ( Katlama et al . , 2013 ) . Early treatment initiation with ART will likely become the standard clinical practice in HIV care . This is supported by the recent outcome of the first large-scale international 'Strategic Timing of AntiRetroviral Treatment' ( START ) study , showing a considerably lower risk of developing AIDS and other serious conditions when compared to later treatment initiation ( INSIGHT START Study Group , 2015 ) . Interestingly , early treatment initiation during HIV-1 seroconversion is also the most effective intervention to limit the extent of viral reservoirs ( Ananworanich et al . , 2012; Hoen et al . , 2007; Hocqueloux et al . , 2013; Ananworanich et al . , 2015 ) . Very low or even undetectable HIV-1 DNA has been described when treatment is initiated during the very early stages of primary HIV-1 infection ( PHI ) ( Ananworanich et al . , 2012; Laanani et al . , 2015 ) . In addition , a lower level of HIV-1 transcription has been described in ART-treated patients who initiated treatment during seroconvertion ( Schmid et al . , 2010 ) . Elite controllers and long-term non-progressors ( LTNPs ) represent an important group as a comparator . These HIV-1 infected individuals display low or undetectable blood reservoirs , and are able to control viremia over the long-term with limited CD4 T cell loss in the absence of treatment . Consequently , LTNPs have been extensively studied in an attempt to unravel the underlying mechanisms of spontaneous virological control ( Autran et al . , 2011; Deeks and Walker , 2007 ) . Although their viral reservoirs have been shown to be low , replication-competent viruses can still be found in these individuals ( Blankson et al . , 2007; Buzon et al . , 2014 ) . LTNPs display strong HIV-1-specific T cells responses with polyfunctionality , thereby suggesting a role of T cell immunity in viremia control ( Cellerai et al . , 2011 ) . In the present study , we have assessed whether undetectable or low levels of reservoirs in blood and tissue could be achieved with very prolonged therapy initiated at PHI or during chronic infection using newly-developed polymerase chain reaction ( PCR ) -based virological assays for in-depth measurement of the size of the HIV-1 reservoir in blood ( total and integrated HIV-1 DNA ) and its dynamics ( episomal 2-long terminal repeat ( LTR ) circles and cell-associated unspliced RNA [usRNA] ) as well as total HIV-1 DNA burden in the rectal mucosa . These patients were compared to LTNPs and to untreated seroconverters . We have analyzed whether 1 ) a decade of ART or an LTNP status was associated with the absence of detectable HIV-1 DNA in the blood and rectal mucosa; 2 ) long-term treated seroconverters could reach levels of virological reservoirs , residual replication , and transcription comparable to those of LTNPs; 3 ) a similar period of aviremia with ART initiation during the chronic phase of HIV-1 infection could achieve levels of reservoirs , residual replication , and transcription comparable to long-term treated seroconverters; 4 ) immune reconstitution , as measured by CD4/CD8 ratio , was enhanced with early treatment intervention; and 5 ) a correlation was present between the various virological and immunological parameters used in this study . Eighty-four patients were included in this cross-sectional study from four different cohorts: patients who had undergone a decade of successful ART , initiated either during seroconversion ( SRCV on ART; n = 25 ) or during the chronic phase of the infection ( Chronic ART , n = 32 ) , LTNPs ( n = 17 ) , and recently infected ART-naïve seroconverters ( Recent SRCV; n = 10 ) ( Figure 1; Table 1 ) . The CD4 nadir was significantly different between each of the cohorts ( p < 0 . 001 ) except between SRCV on ART and Recent SRCV ( p = 0 . 623 ) . The patients were sampled at a single time point ( blood and rectal biopsies ) to perform PCR-based assays and characterize viral reservoirs and its dynamics ( total and integrated HIV-1 DNA , 2-LTR circles , and HIV-1 usRNA ) . 10 . 7554/eLife . 09115 . 003Figure 1 . Patient cohorts in the cross-sectional study . SRCV on antiretroviral therapy ( ART ) : patient cohort with ART initiated at the time of HIV-1 seroconversion; LTNP: long-term non-progressors; Chronic ART: patients with ART initiated during the chronic phase of HIV-1 infection; Recent SRCV: recent ART-naïve seroconverters . In total , 84 patients were included in this study , 25 in SRCV on ART , 17 LTNPs , 32 Chronic ART patients , and 10 Recent SRCV . Blue arrows represent time of sampling . PHI: primary HIV-1 infection . DOI:http://dx . doi . org/10 . 7554/eLife . 09115 . 00310 . 7554/eLife . 09115 . 004Table 1 . Clinical and laboratory characteristics of the four patient cohorts . DOI:http://dx . doi . org/10 . 7554/eLife . 09115 . 004Value for cohortSRCV on ARTLTNPChronic ARTRecent SRCVn = 25n = 17n = 32n = 10Clinical characteristicsAge ( years ) 44 ( 34–53 ) 49 ( 31–51 ) 48 ( 31–53 ) 39 ( 30–46 ) Number of females ( % ) 0 ( 0 ) 7 ( 41 . 2 ) 5 ( 15 . 6 ) 1 ( 10 ) Total ART duration ( years ) 10 . 8 ( 4 . 2–11 . 9 ) 09 . 8 ( 4 . 9–14 . 7 ) 0Viremia zenith ( log10HIV-1 c/ml ) 5 . 5 ( 2 . 4–5 . 9 ) 2 . 5 ( 1 . 6–2 . 8 ) 4 . 9 ( 1 . 9–5 . 5 ) 6 . 2 ( 5 . 2–6 . 4 ) CD4 count , nadir ( cells/mm3 ) 390 ( 107–466 ) 624 ( 507–693 ) 155 ( 0–266 ) 440 ( 284–495 ) CD4 count at sampling ( cells/mm3 ) 714 ( 476–977 ) 793 ( 414–1010 ) 624 . 5 ( 172–889 ) 440 ( 284–604 ) CD4/CD8 ratio1 . 10 ( 0 . 52–1 . 35 ) 0 . 91 ( 0 . 36–1 . 47 ) 0 . 74 ( 0 . 23–0 . 93 ) 0 . 62 ( 0 . 36–0 . 94 ) Virological markersTotal HIV-1 DNA ( c/106 PBMCs ) 92 ( 9 . 8–127 . 7 ) 48 ( 5 . 4–56 . 5 ) 137 ( 8 . 6–219 . 2 ) 1901 . 3 ( 602 . 4–4786 . 9 ) Integrated HIV-1 DNA ( c/106 PBMCs ) 211 . 3 ( 0–589 . 3 ) 28 . 2 ( 0–158 . 4 ) 586 . 7 ( 131 . 6–918 . 2 ) 1802 . 7 ( 367 . 9–5890 . 6 ) HIV-1 usRNA ( c/106 PBMCs ) 1 . 6 ( 0–3 . 7 ) 0 . 4 ( 0–3 . 51 ) 6 . 1 ( 0–10 . 1 ) 15 . 5 ( 0 . 9–100 . 6 ) 2-LTR circles ( c/106 PBMCs ) 2 . 1 ( 0–4 . 3 ) 0 . 8 ( 0–2 . 7 ) 1 . 3 ( 0–2 . 2 ) 13 . 3 ( 5 . 1–21 . 6 ) Total HIV-1 DNA ( c/106 cells ) in rectal biopsies27 . 2 ( 22 . 2–61 . 7 ) 21 . 3 ( 16 . 7–34 . 5 ) 35 . 1 ( 16–77 . 5 ) Values are reported as median ( IQR ) ; c: copies; PBMCs: peripheral blood mononuclear cells; usRNA: unspliced RNA; ART: antiretroviral therapy; SRCV on ART: early treated seroconverters; LTNP: long-term non-progressors; Chronic ART: late treated patients during chronic HIV-1 infection; Recent SRCV: acute ART-naïve seroconverters; LTR: Long terminal repeat . Total HIV-1 DNA represents the most commonly used virological marker for the assessment of the size of the proviral HIV-1 reservoir and is predictive of viral rebound when tested at the time of treatment interruption in early-treated patients ( Williams et al . , 2014 ) . We have assessed the impact of the temporal treatment initiation ( during early versus chronic infection ) in the context of very prolonged ART treatment ( e . g . a decade ) on this marker and compared results to those of LTNPs and acute seroconverters before ART initiation to assess the size of HIV-1 DNA using digital PCR . All patient cohorts had detectable levels of total HIV-1 DNA in peripheral blood mononuclear cells ( PBMCs ) . Differences in the reservoir size in terms of total HIV-1 DNA were observed between patient cohorts . Median total HIV-1 DNA was: 92 ( interquartile range ( IQR ) : 9 . 8–127 . 7 ) , 48 ( IQR: 5 . 4–56 . 5 ) , 137 ( IQR: 8 . 6–219 . 2 ) and 1901 . 3 ( IQR: 602 . 4–4786 . 9 ) copies ( c ) /106 PBMCs in SRCV on ART , LTNPs , Chronic ART , and Recent SRCV , respectively . Lower total HIV-1 DNA was detected in the SRCV on ART compared to the Chronic ART cohort ( p = 0 . 041; Figure 2A ) . The LTNP cohort showed the lowest total HIV-1 DNA levels when compared to SRCV on ART ( p = 0 . 015 ) and other patient cohorts ( p < 0 . 001; Figure 2A ) . These results demonstrate that , the total HIV-1 DNA remains detectable in all patients even in the setting of effective early and prolonged ART , although at lower levels than in patients with later ART initiation , but not reaching the levels found in LTNPs . 10 . 7554/eLife . 09115 . 005Figure 2 . Total HIV-1 DNA ( A ) and integrated HIV-1 DNA ( B ) in four patient cohorts . Data is shown as log10 copies/million ( c/M ) PBMCs and significant p-values are indicated by * . Differences between the cohorts were determined by Wilcoxon Signed Rank test . DOI:http://dx . doi . org/10 . 7554/eLife . 09115 . 005 In addition to total HIV-1 DNA levels , we also measured integrated HIV-1 DNA levels . This marker is not biased by the presence of variable quantities of unintegrated HIV-1 DNA produced after reverse transcription of newly infecting HIV-1 or dead-end DNA products of failed integration such as 1- and 2-LTR circles . Some authors have suggested that integrated HIV-1 DNA represents a better surrogate marker of viral burden , especially in patients off ART in whom HIV-1 replication might be ongoing ( Graf et al . , 2011 ) . Median integrated HIV-1 DNA levels were: 211 . 3 ( IQR: 0–589 . 3 ) , 28 . 2 ( IQR: 0–158 . 4 ) , 586 . 7 ( IQR: 131 . 6–918 . 2 ) , and 1802 . 7 ( IQR: 367 . 9–5890 . 6 ) c/106 PBMCs for SRCV on ART , LTNPs , Chronic ART , and Recent SRCV , respectively . A lower level of integrated HIV-1 DNA was found in SRCV on ART compared to the Chronic ART cohort ( p = 0 . 003 ) and in LTNPs compared to both the SRCV on ART ( p = 0 . 021 ) and Chronic ART cohorts ( p < 0 . 001; Figure 2B ) . These results confirm the low levels of integrated DNA found in LTNPs as described previously ( Graf et al . , 2011 ) . Interestingly , levels of integrated HIV-1 DNA were not significantly different between the ART-naïve Recent SRCV and the Chronic ART cohort ( p = 0 . 104 ) . The Recent SRCV cohort displayed higher integrated HIV-1 DNA levels compared to SRCV on ART ( p = 0 . 006 ) and LTNPs ( p = 0 . 002; Figure 2B ) . Of note , the absolute values derived from total and integrated HIV-1 DNA measurements in this study cannot be directly compared with each other . This is due to the difference in quantification methods , the absolute quantification of integrated HIV-1 DNA being corrected by using an integration standard as calibrator as previously described ( Liszewski et al . , 2009; De Spiegelaere et al . , 2014 ) . This calibrator does not alter the relative differences of integrated HIV-1 DNA between patient samples , but may bias the absolute quantitative outcome . In contrast , total HIV-1 DNA is reported by direct absolute quantification and represents a better estimate of the amount of HIV-1 DNA c/106 cells present in patients . Due to the lower limit of detection compared to that of the total HIV-1 DNA assay , integrated HIV-1 DNA was undetectable in 12 patients; 5 were SRCV on ART , 6 LTNPs , and 1 Recent SRCV . Residual viral replication is one of the likely mechanisms through which the viral reservoir is replenished , even with effective ART ( Hong and Mellors , 2015 ) . Markers that reflect such a phenomenon provide insight into the reservoir dynamics . Episomal 2-LTR circles can be used as a marker of viral replication and have been shown to be labile end-products of failed proviral integrations ( Sharkey et al . , 2005; 2011 ) . They are likely to be particularly elevated in acutely infected patients because of the intense level of viral replication . Median levels of 2-LTR circles in the various cohorts were: 2 . 1 ( IQR: 0–4 . 3 ) , 0 . 77 ( IQR: 0–2 . 7 ) , 1 . 3 ( IQR: 0–2 . 2 ) and 13 . 3 ( IQR: 5 . 1–21 . 6 ) c/106 PBMCs in SRCV on ART , LTNPs , Chronic ART , and Recent SRCV , respectively . As expected , 2-LTR levels were significantly higher in the Recent SCRV compared to the other cohorts ( p ≤ 0 . 002; Figure 3A ) , but did not differ between SRCV on ART and Chronic ART ( p = 0 . 259 ) or LTNPs ( p = 0 . 595 ) or between the Chronic ART and LTNP cohorts ( p = 0 . 743;Figure 3A ) . Of note , 2-LTR circles were undetectable in a number of patients . They were detected in 17/25 ( 68% ) of SRCV on ART , 13/17 ( 76% ) of LTNPs , 25/32 ( 78% ) of late-treated patients and in all recent ART-naïve seroconverters 10/10 ( 100% ) . The absence of detection in about one-quarter of patients on ART in contrast to the Recent SRCV cohort suggests that residual replication is substantially halted with ART . 10 . 7554/eLife . 09115 . 006Figure 3 . 2-LTR circles ( A ) , cell-associated HIV-1 usRNA ( B ) and CD4/CD8 ratio ( C ) in four patient cohorts . Total HIV-1 DNA in rectal biopsies ( D ) in three patient cohorts ( SRCV on ART , LTNP , and Chronic ART ) . Data is shown as log10 copies/million ( c/M ) PBMCs ( A , B ) , ratio ( C ) or log10 c/M cells in rectal biopsies ( D ) and significant p-values are indicated by * . Differences between the cohorts were determined by Wilcoxon Signed Rank test . DOI:http://dx . doi . org/10 . 7554/eLife . 09115 . 006 Levels of cell-associated HIV-1 usRNA are associated with recent HIV-1 transcriptional activity and indicate an active proviral reservoir ( Pasternak et al . , 2013 ) . Median levels were: 1 . 6 ( IQR: 0–3 . 7 ) , 0 . 4 ( IQR: 0–3 . 5 ) , 6 . 1 ( IQR: 0–10 . 1 ) and 15 . 5 ( IQR: 1–100 . 6 ) c/106 PBMCs in SRCV on ART , LTNPs , Chronic ART , and Recent SRCV , respectively . The patients with the highest levels of HIV-1 usRNA were found in the Recent SRCV cohort , indicating that this cohort is not similar to the others , but this study was underpowered to reach statistical significance between the Recent SRCV and the other cohorts . Of note , HIV-1 usRNA was not detected in 8/84 samples ( 10% , two samples in each patient cohort ) . Three samples ( one in SRCV on ART and two in Chronic ART ) were excluded from the final analysis due to missing values for the reference genes . Higher levels of HIV-1 usRNA were detected in the Chronic ART cohort compared to the SRCV on ART ( p = 0 . 007 ) and LTNPs ( p = 0 . 027; Figure 3B ) . The SRCV on ART cohort was not significantly different from LTNPs based on HIV-1 usRNA levels ( p = 0 . 615; Figure 3B ) . Not only have we used the CD4/CD8 ratio as a measure of immune preservation/reconstitution in terms of T cell count but also as an indirect marker of residual immune activation as shown recently ( Serrano-Villar et al . , 2014 ) . The CD4/CD8 ratio was higher in SRCV on ART ( median = 1 . 10 , IQR: 0 . 52–1 . 35 ) compared to the cohorts of Chronic ART ( median = 0 . 74 , IQR: 0 . 23–0 . 93 ) , ( p = 0 . 009 ) and Recent SRCV ( median = 0 . 62 , IQR: 0 . 36–0 . 94 ) , ( p = 0 . 017 ) , and was comparable to that of LTNPs ( median = 0 . 91 , IQR: 0 . 36–1 . 47 ) , ( p = 0 . 978; Figure 3C ) . The LTNP cohort had a higher CD4/CD8 ratio compared to late-treated patients ( p = 0 . 036 ) and Recent SRCV ( p = 0 . 048; Figure 3C ) . Of note , CD4 T cell counts at sampling did not differ significantly between the cohorts of SRCV on ART ( median = 714 cells/mm3 , IQR: 476–977 ) and Chronic ART ( median = 625 cells/mm3 , IQR: 172–889 ) , ( p = 0 . 066 ) . Conflicting results have been published regarding the impact of ART on the HIV-1 DNA reservoir in the gut compartment when using rectal biopsies ( Yukl et al . , 2010; Chun et al . , 2008; Anton et al . , 2003 ) . It remains unclear whether HIV-1 DNA decays more substantially after a decade of ART compared to a shorter intervention and whether some LTNPs may have undetectable levels in rectal biopsies . We have measured total HIV-1 DNA in rectal biopsies from 51 patients who had consented to sampling: 14 SRCV on ART , 8 LTNPs , and 29 Chronic ART patients . Five patients had undetectable total HIV-1 DNA; one SRCV on ART , one LTNP , and three Chronic ART patients . Median HIV-1 DNA levels were: 27 . 2 ( IQR: 22 . 2–61 . 7 ) , 21 . 3 ( IQR: 16 . 7–34 . 5 ) , and 35 . 1 ( IQR: 16–77 . 5 ) c/106 cells in SRCV on ART , LTNP , and Chronic ART , respectively . No difference was found between SRCV on ART and Chronic ART ( p = 0 . 604 ) or LTNPs ( p = 0 . 375; Figure 3D ) or between the Chronic ART cohort and the LTNPs ( p = 0 . 337; Figure 3D ) . The median number of cells assayed per patient was 125 , 390 ( IQR: 101 , 308–168 , 156 ) . Of note , we did not find any correlation in terms of total HIV-1 DNA levels between the blood ( PBMCs ) and the gut mucosa ( rectal biopsies ) compartments ( R2 = 0 , p = 0 . 919; Figure 4A ) . 10 . 7554/eLife . 09115 . 007Figure 4 . Correlation of total HIV-1 DNA in rectal biopsies and blood ( A ) . Correlation was assessed in 46 patients in whom total HIV-1 DNA was detected both in the blood and rectal biopsies , representing patients from three cohorts: SRCV on antiretroviral therapy ( ART ) , Chronic ART , and long-term non-progressors ( LTNP ) . Data is shown as log10 copies/million ( c/M ) cells in rectal biopsies and log10 c/M peripheral blood mononuclear cells ( PBMCs ) in blood . Correlation of blood HIV-1 usRNA and total HIV-1 DNA ( B ) , HIV-1 usRNA and integrated HIV-1 DNA ( C ) , integrated HIV-1 DNA and total HIV-1 DNA ( D ) and CD4/CD8 ratio and integrated HIV-1 DNA ( E ) . Data is shown as log10 c/M PBMCs in detectable patients from three cohorts: SRCV on ART , Chronic ART , and LTNP ( B–E ) . To assess correlations between the markers , a linear regression was performed . DOI:http://dx . doi . org/10 . 7554/eLife . 09115 . 007 In order to assess whether we could observe correlations between the various viroimmunological markers used in this study , we have performed a linear regression analysis using combined data from the cohorts of patients on ART and LTNPs . The Recent SRCV cohort was excluded from this analysis because of its high level of active replication , which would have biased the HIV-1 usRNA , 2-LTR , and total HIV-1 DNA measurements . For this analysis , patient-derived samples with detectable markers were included . This was confirmed by Spearman’s rank correlation , which includes all samples ( data not shown ) . We found a positive correlation between HIV-1 usRNA and both total HIV-1 DNA ( R2 = 0 . 19 , p < 0 . 001; Figure 4B ) and integrated HIV-1 DNA ( R2 = 0 . 18 , p < 0 . 01; Figure 4C ) . Total HIV-1 DNA correlated with integrated HIV-1 DNA levels ( R2 = 0 . 31 , p < 0 . 001; Figure 4D ) . No correlation was observed between 2-LTR circles and HIV-1 usRNA , total or integrated HIV-1 DNA ( data not shown ) . A negative correlation was found between the CD4/CD8 ratio and integrated HIV-1 DNA ( R2 = 0 . 14 , p < 0 . 01; Figure 4E ) . No correlation was found between the CD4/CD8 ratio and the other virological markers assessed ( total HIV-1 DNA , 2-LTR circles and usRNA in PBMCs and total HIV-1 DNA in rectal biopsies ) , ( data not shown ) . In the present study , we have shown that a decade of ART , initiated during seroconversion , decreases the HIV-1 DNA reservoir size and viral transcription level in blood , and benefits immunological restoration in comparison to later treatment initiation of the same duration . These data support the notion that early treatment initiation at the time of seroconversion may favor post-treatment viral control by limiting the establishment of an extended viral reservoir . However , we have also found that seeding of the viral reservoir occurs very rapidly after acquisition of the infection and , indeed , our data show that the small total and integrated HIV-1 DNA reservoir in early treated patients is still significantly larger despite early treatment initiation when compared to LTNPs . As the majority of viral reservoirs is located in tissues , we sampled rectal tissues to assess the reservoir size in the gut mucosa . Our results do not show a higher total HIV-1 DNA levels in this compartment in both early and late ART-treated patients than in LTNPs . We did not find a correlation between the blood and gut reservoirs . This is in accordance with one previous study ( Di Stefano et al . , 2001 ) , but not with others ( Anton et al . , 2003; Avettand-Fenoel et al . , 2008 ) . A less efficient clearance of total HIV-1 DNA in rectal tissues compared to PBMCs has previously been described after early therapy initiation ( Chun et al . , 2008; Ananworanich et al . , 2012 ) . Low level cryptic HIV-1 replication may be a result of sub-optimal drug penetration in this compartment , immunological impairment , or caused by another unknown mechanism . Our results do not confirm such a hypothesis , although we have observed a non-significant trend towards a lower HIV-1 DNA level in gut in LTNPs compared to ART-treated patients . Lower levels of total HIV-1 DNA in gut were also previously detected in elite controllers compared to ART-suppressed patients ( Hatano et al . , 2013b ) . It remains unclear how the HIV-1 reservoir in the gut compartment might be influenced by prolonged early therapy initiation , as we did not have longitudinal samples , and whether it may contribute to viral rebound after therapy interruption . Furthermore , sampling the gut mucosa remains an invasive procedure compared to blood sampling , it cannot be performed as frequently and standardization to the exact sampling location is difficult . Residual viral replication is a topic of intense debate in the HIV-1 cure field . Evidence is accumulating that the reservoirs may consist of truly latent provirus maintained in long-lived quiescent CD4 memory T cells , seeded soon after acute infection or maintained through homeostatic proliferation ( Maldarelli et al . , 2014; Wagner et al . , 2014; Cohn et al . , 2015 ) . Low level residual viral replication may also represent an important mechanism in sustaining a replication-competent viral reservoir ( Sharkey et al . , 2005; 2011; 2013; Hong and Mellors , 2015 ) . The presence of 2-LTR circles is considered indicative of viral replication and , as expected in this study , its levels were highest in recent ART-naïve seroconverters who display a high amount of ongoing replication . Yet no difference was observed between ART-treated patients and LTNPs . This suggests that ART effectively suppresses viral replication , regardless of the timing of treatment initiation . However , the value of 2-LTR circles as a marker for replication remains unclear as some reports suggest that 2-LTR circles are stable and long-lived ( Brussel et al . , 2003; Pierson et al . , 2002 ) . Notably , the amount of 2-LTR circles was low in a substantial number of ART-treated patients and LTNPs and undetectable in 26% of treated patients . Levels of 2-LTR circles did not correlate with total or integrated HIV-1 DNA or HIV-1 usRNA , indicating that factors other than the size of the viral reservoir may determine the presence of 2-LTR circles . We have used cell-associated HIV-1 usRNA to reflect proviral DNA transcription . This marker may predict the replicative-competence of the viral reservoir , and previous studies have shown a correlation of usRNA levels with virological failure ( Pasternak et al . , 2013 ) and markers of immune activation in elite controllers ( Hunt et al . , 2011 ) . In addition , recent trials using histone deacetylase inhibitors to stimulate viral production in the reservoir have used HIV-1 usRNA as a marker to assess viral transcription ( Archin et al . , 2012; Elliott et al . , 2014; Rasmussen et al . , 2014; Wei et al . 2014 ) . An earlier report has shown that early-treated patients have lower HIV-1 usRNA compared to late-treated patients after a short period of ART ( Schmid et al . , 2010 ) . With a long-term follow-up , Buzon et al . observed a trend towards lower levels of HIV-1 usRNA in elite controllers and early-treated patients compared to late-treated patients ( Buzon et al . , 2014 ) . Here , we can confirm that early-treated patients have lower HIV-1 usRNA compared to patients who started therapy with onset of chronic HIV-1 infection . This observation persists after a decade of successful ART . The levels of HIV-1 usRNA in early-treated seroconverters were not different from those of LTNPs , suggesting that early treatment initiation reduces viral transcriptional activity to levels reminiscent of LTNPs , which may facilitate further intervention to control such low level of viral replication . However , it must be noted that HIV-1 usRNA levels positively correlated with those of total and integrated HIV-1 DNA . This may indicate that lower levels of HIV-1 usRNA are a consequence of a smaller pool of HIV-1 DNA reservoir rather than a low transcriptional activity . Cell-associated HIV-1 usRNA levels were shown to be very low in LTNPs , thus supporting its use as a reference parameter in future studies aimed at an HIV-1 cure . T cell activation often remains elevated in chronic HIV-1 infection in spite of ART and is linked to a lower rate of CD4 T cell count recovery ( Goicoechea et al . , 2006 ) and higher mortality rates ( Serrano-Villar et al . , 2014 ) . Lower levels of T cell activation were shown recently in early-treated patients within 6 months of infection with a shorter ART duration ( median: 2 . 8 years ) compared to later ART initiation ( Jain et al . , 2013 ) . Here , we show that a higher CD4/CD8 ratio is found in long-term early-treated seroconverters with levels comparable to those observed in LTNPs , confirming observations from a previous study ( Cellerai et al . , 2011 ) . We were also able to show that the increased CD4/CD8 ratio was associated with a smaller integrated HIV-1 DNA reservoir , but not with total HIV-1 DNA . These two markers should therefore be further studied for their accuracy in reflecting the state of the HIV-1 reservoir in cure studies . We have shown that the virological markers used in this study are affected by early treatment initiation . They could potentially represent predictors of a functional cure or of the time to viral rebound after ART interruption , where treatment interruption studies are necessary to validate these predictors . Although , the inter-patient variability of results may prevent the use of reservoir markers on an individual basis , a longitudinal follow-up of these markers may increase the rationale for their use in such studies . Recently , an association with time to viral rebound and HIV-1 DNA was found in early-treated patients within the SPARTAC study of seroconverters ( Williams et al . , 2014 ) , and with integrated HIV-1 DNA in a study using pegylated Interferon alfa-2a to purge the viral reservoir ( Azzoni et al . , 2013 ) . Hence , future studies should emphasize the longitudinal profiling of HIV-1 reservoir markers to validate their use in cure studies . Some of the limitations of our study include the lack of pre-treatment and longitudinal follow-up in terms of viral reservoirs , which could provide information on their dynamics over the long-term and indicate whether a shorter period of treatment could achieve the same reservoir levels in both blood and tissues . The present study used PCR-based HIV-1 markers , which cannot differentiate between the presence or absence of replicative-competent reservoirs , and we did not use ex vivo assays such as the viral outgrowth assay ( VOA ) in blood , which can quantify replication-competent proviral HIV-1 DNA , the fraction of the reservoir that matters in terms of viral eradication . Both types of methods either overestimate or underestimate the size of the replicative-competent reservoir ( Bruner et al . , 2015 ) , and VOA remains a cumbersome and expensive assay that is not widely available and will need to be replaced by easier and cheaper assays . In conclusion , our study provides important data on the blood HIV-1 reservoir size and dynamics after a decade of successful treatment with ART together with gut mucosal HIV-1 DNA . Our results support early treatment initiation in terms of achieving lower levels of viral reservoirs , when compared to later treatment . The levels of HIV-1 DNA that are likely to be associated with a functional cure remain to be determined together with other factors such as protective immunological responses involved in virological control . Eighty-four HIV-1 infected patients from four pre-defined cohorts under clinical follow-up were enrolled into the study in two clinical centers ( The Ian Charleson Day Centre , Royal Free Hospital , London , United Kingdom and the AIDS Reference Center , Ghent University Hospital , Ghent , Belgium ) : long-term-treated patients with ART initiated during seroconversion or chronic infection , LTNPs , and recent ART-naïve seroconverters . The first three cohorts were recruited using databases at the clinical centers ( the Royal Free Center Research Database and Ghent University Hospital Database ) . The fourth cohort consisting of acute seroconverters was enrolled prospectively in both clinical centers . The study was approved by the Ethics Committee of Ghent University Hospital ( Reference number: B670201317826 ) and the Royal Free Hospital ( Reference number: 13/LO/0729 ) . The patient cohorts are described in Figure 1 . The first cohort consisted of HIV-1 seroconverters on long-term ART initiated and uninterrupted since PHI ( SRCV on ART; n = 25 ) . These patients were selected on the basis of the following inclusion criteria: ( a ) ART for ≥4 years; ( b ) long-term aviremia ( <50 HIV-1 RNA copies ( c ) /ml ) ; and ( c ) an absence of treatment failure defined by a viral load ( VL ) ≥400 HIV-1 RNA c/ml . Seroconversion to HIV-1 was defined by: ( a ) negative HIV-1 antibody by enzyme-linked immunosorbent assay ( ELISA ) and evidence of HIV-1 viremia ≥5000 HIV-1 RNA c/ml plasma and/or ( b ) incomplete HIV-1 Western blot with ≤3 bands and/or a detuned assay with a value of ≤0 . 6 for HIV-1 clade B patients . The second cohort of LTNPs ( n = 17 ) were therapy-naïve patients , with ≥7 years of documented HIV-1 infection , viremia ≤1000 HIV-1 c/ml and a CD4 T cell count ≥500 cells/mm3 during follow-up . Exception was made for temporary ART to prevent mother-to-child transmission . The third cohort consisted of HIV-1 infected patients successfully treated by ART initiated during chronic infection ( Chronic ART; n = 32 ) with: ( a ) a treatment duration of ≥4 years and ( b ) long-term undetectable VL ( <50 HIV-1 c/ml for ≥4 years ) . ART failure during follow-up was defined by a VL ≥400 HIV-1 c/ml . The fourth cohort consisted of recent ART-naïve seroconverters ( Recent SRCV; n = 10 ) . Criteria for the diagnosis of seroconversion and enrollment into the study were the following: ( a ) negative or indeterminate HIV-1 antibody result by fourth generation ELISA and evidence of HIV-1 viremia ≥5000 HIV-1 c/ml plasma and/or ( b ) positive HIV-1 by fourth generation ELISA and two negative confirmatory tests ( negative Vidas and Immunocomb ) and/or ( c ) positive HIV-1 by fourth generation ELISA and negative InnoLIA . Baseline characteristics , clinical and laboratory parameters including total duration of ART , VL zenith , CD4 T cell count at blood sampling and nadir , as well as CD4/CD8 ratio at blood sampling were collected and are summarized in Table 1 . Both the SRCV on ART and Chronic ART cohorts had a comparable uninterrupted ART duration with a median of 10 . 8 ( IQR: 4 . 2–11 . 9 ) and 9 . 8 years ( 4 . 9–14 . 7 ) , ( p = 0 . 936 ) , respectively . PBMCs and rectal biopsies were collected on one occasion for each patient ( Figure 1 ) . Blood was drawn in 6 EDTA 9 ml tubes and 10 rectal biopsies were sampled after obtaining written informed consent from the patients . PBMCs were isolated within 4 hr of blood sampling by using Lymphoprep centrifugation ( ELITech Group , Zottegem , Belgium ) . Cells were manually counted using a hemocytometer counting grid , aliquoted in 106PBMCs as dry pellets or in fetal calf serum + 7 . 5% dimethyl sulfoxide and stored at −80°C . Flexible sigmoidoscopy was performed and 10 gut biopsies ( volume around 1 mm3 of each biopt ) were taken from the mucosa of the upper rectum ( 10 cm from the anal margin ) using single-use biopsy forceps . Intact rectal biopsies were immediately frozen and stored at −80°C until further processing . PBMCs were collected from all patients ( n = 84 ) included in the study and rectal biopsies were obtained from 51 patients , 14 in SRCV on ART , 8 in LTNP , and 29 in Chronic ART cohorts . Methods used to quantify virological parameters relating to HIV-1 reservoirs , ongoing replication , and transcription in this study have been recently published and include total HIV-1 DNA and 2-LTR circles ( Malatinkova et al . , 2014 ) , cell-associated HIV-1 usRNA ( Kiselinova et al . , 2014 ) with the use of droplet digital PCR ( ddPCR ) and integrated HIV-1 DNA measured by a repetitive sampling Alu-HIV PCR ( Liszewski et al . , 2009; De Spiegelaere et al . , 2014 ) . Total HIV-1 DNA , 2-LTR circles , and HIV-1 usRNA were measured in triplicates on ddPCR with the QX100Droplet Digital PCR platform ( Bio-Rad , Hercules , CA ) . The ddPCR mix was made by adding 2 µl of sample ( restricted genomic DNA [gDNA] or plasmid DNA ) or 4 µl of sample ( cDNA ) to 10 µl 2× ddPCR™ supermix for probes ( Bio-Rad ) , 500 nM of primers and 250 nM of probe in a final volume of 20 µl . ddPCR amplification reactions consisted of initial denaturation at 95°C for 5 min , followed by 40 cycles of 95°C for 30 s denaturation and assay-specific annealing/elongation temperature ( Supplementary file 1 ) for 60 s with a ramp rate of 2 . 5°C/s . Droplets were read by the QX100 droplet reader ( Bio-Rad ) and the data was analyzed with the QuantaSoft analysis software ( Bio-Rad ) . Primers and probes used for each quantification assay are summarized in Supplementary file 1 and were purchased from IDT DNA Technologies ( Integrated DNA Technologies , Leuven , Belgium ) . Total gDNA was isolated from 106 PBMCs using DNeasy Blood & Tissue Kit ( Qiagen , Venlo , The Netherlands ) and eluted in 75 µl elution buffer , kept at 56°C for 10 min in order to maximize the DNA yield . Three intact rectal biopsies were pooled per patient and used to isolate gDNA using DNeasy Blood & Tissue Kit ( Qiagen ) . Total gDNA was eluted in 40 µl elution buffer to concentrate the sample and kept at 56°C for 10 min . To measure total HIV-1 DNA , an enzyme restriction digestion with EcoRI ( Promega , Leiden , The Netherlands ) was performed on gDNA from both PBMCs and rectal biopsies with the use of 17 . 3 µl gDNA in a total volume of 20 µl of restriction mix . This step is preferred for ddPCR as the fragmented DNA will be more uniformly distributed in all droplets compared to full-length chromosomal DNA . To quantify integrated HIV-1 DNA , gDNA isolated from PBMCs was subjected to a repetitive sampling Alu-HIV PCR , according to a recently described protocol ( De Spiegelaere et al . , 2014 ) . Briefly , a method based on Poisson statistics was used to analyze the binomial data of positive and negative reactions from a 40-replicate Alu-HIV PCR ( De Spiegelaere et al . , 2014 ) . First , total HIV-1 DNA was measured in gDNA samples by ddPCR and based on these measures , each sample was diluted to approximately five copies of total HIV-1 DNA per PCR replicate and distributed in replicates . Alu-HIV PCR has been described previously ( Liszewski et al . , 2009; Yu et al . , 2008; De Spiegelaere et al . , 2014 ) and was performed by using an HIV-1-specific reverse primer in Gag and a human Alu-specific forward primer ( Supplementary file 1 ) in 40 replicates . In parallel , 20 replicates were run for background quantification by using only the HIV-1 Gag primer . The PCR mix was made by adding 10 µl of diluted gDNA sample to 10 µl PCR mix consisting of 5× Go Taq G2 master mix , 0 . 2 µl Go Taq G2 DNA Polymerase , 4 mM of dNTP mix ( Promega ) , 200 nM of Alu primer and 1200 nM of HIV-1 primer in a final volume of 20 µl . PCR amplification reactions consisted of initial denaturation at 95°C for 2 min , followed by 40 cycles of 95°C for 15 s denaturation , 50°C for 15 s annealing , and 70°C for 3 . 5 min elongation . 2 µl of the PCR product were processed in the nested qPCR ( Light Cycler 480 System , Roche Applied Science , Penzberg , Germany ) , qPCR mix contained 2× LightCycler 480 Probes Master mix ( Roche Applied Science , Vilvoorde , Belgium ) , 400 nM primers , and 200 nM probe ( Supplementary file 1 ) , and qPCR consisted of initial denaturation at 95°C for 5 min , followed by 45 cycles of 95°C for 15 s denaturation , and 60°C for 1 min annealing/elongation . The quantities of total and integrated HIV-1 DNA c/106 PBMCs or total HIV-1 DNA c/106 cells in rectal biopsies were normalized to a reference gene RPP30 measured by ddPCR . The number of cells assayed per patient was measured by RPP30 in all PBMCs and rectal biopsies samples . Episomal HIV-1 2-LTR circles were measured in plasmid DNA isolated by QIAprep Spin Miniprep kit ( Qiagen ) from dry pelleted 106 PBMCs . A known number of pSIF1-H1-Puro non-HIV plasmid was spiked to the samples ( System Biosciences , Mountain view , CA ) as an internal control for copy number normalization and plasmid DNA was eluted in 25 µl in order to increase DNA concentration , as described previously ( Malatinkova et al . , 2014 ) . The internal reference plasmid was quantified by detection of the woodchuck hepatitis virus posttranscriptional regulatory element ( WPRE ) ( Lizee et al . , 2003 ) . The 2-LTR assay is designed to span over the 2-LTR junction ( Buzon et al . , 2010; Supplementary file 1 ) . RNA was isolated from 106 PBMCs by using RNeasy mini kit ( Qiagen ) subjected to DNase treatment by RNase-Free DNase Set ( Qiagen ) and eluted in 30 µl nuclease-free water . Samples were measured by NanoDrop 2000 ( Thermo Fisher Scientific , Waltham , MA ) and 1 . 5 mg of RNA was processed by the iScript cDNA Synthesis Kit ( Bio-Rad ) 5 min at 25°C , 30 min at 42°C , and 5 min at 85°C . The cDNA was used to measure HIV-1 usRNA on ddPCR . Normalization of input cDNA was performed by quantifying gene expression of stably expressed internal reference genes as described earlier ( Ceelen et al . , 2014; Messiaen et al . , 2012 ) . Briefly , the three most stably expressed reference genes ( from total of nine genes tested ) were selected over all patient samples by geNorm analysis ( Beta-2-microglobulin: B2M , TATA box binding protein: TBP , and Ubiquitin C: UBC ) ( Vandesompele et al . , 2002 ) . Normalization factors were determined per patient as the geometric mean of the three most stable reference genes . Subsequently , raw ddPCR values for HIV-1 usRNA were divided by the normalization factors to reach normalized data and reported as c/106 PBMCs for each patient sample . Previously described primers and probe sets for HIV-1 usRNA quantification were used ( Kiselinova et al . , 2014; Palmer et al . , 2008 ) , as summarized in Supplementary file 1 . Total HIV-1 DNA , integrated HIV-1 DNA , 2-LTR circles and cell-associated HIV-1 usRNA levels as well as immunological data ( CD4/CD8 T cell ratios and CD4 T cell counts at sampling and nadir CD4 T cell counts ) were described using median values and IQR . Statistical analysis was performed using R ( RStudio , Inc . , Boston , MA ) . Standard non-parametric test ( Wilcoxon Signed Rank Test ) was performed to assess statistically significant differences between patient cohorts . A p-value of <0 . 05 was considered significant . Linear regression was used to assess the correlations .	Many people with HIV infections are able to live relatively normal lives thanks to major advances in drug therapies . A cure , however , remains elusive . One reason for this is that the virus can hide in certain types of human cells , where it is protected from the immune system and the effects of “antiretroviral” drugs . This creates reservoirs of virus particles in the body that can quickly multiply and spread if treatment stops . Some people who become infected with HIV are able to contain the virus without the help of drug treatments . These individuals – known as long-term non-progressors – do not become ill and only have low numbers of HIV particles in reservoirs . People who receive treatment early in the course of an HIV infection also have fewer viruses in reservoirs and are less likely to develop severe illness . Therefore , it might be possible to develop a “functional” cure that may not completely eliminate the virus from the body , but would prevent illness and allow the individuals to eventually stop taking antiretroviral drugs . Now , Malatinkova , De Spiegelaere et al . studied samples from 84 patients with HIV-1 to find how much effect an early start to treatment has on the amount of the virus in reservoirs . People who started treatment soon after infection had lower levels of HIV-1 in their blood than people who started treatment later ( even after 10 years of treatment ) . However , patients that started treatment early had higher levels of HIV-1 in the blood than the patients who were long-term non-progressors . All the patients had similar levels of HIV-1 in tissue samples taken from the rectum , regardless of when they started treatment . The experiments suggest that HIV-1 reservoirs form very soon after infection . Malatinkova , De Spiegelaere et al . found that in addition to reducing reservoirs of HIV-1 , an early start to drug treatment reduced the ability of the virus to make copies of its genetic code . People who started treatment earlier also had healthier immune cells . Together , the experiments support the benefits of starting drug treatments as soon as possible after a person is infected with HIV-1 . It is important to further characterize thoroughly the viral reservoir in patients with limited HIV-1 reservoirs and to look for other immune factors involved in virus control , in the search for a functional cure of HIV .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2015	Impact of a decade of successful antiretroviral therapy initiated at HIV-1 seroconversion on blood and rectal reservoirs
Endogenous retroviral sequences provide a molecular fossil record of ancient infections whose analysis might illuminate mechanisms of viral extinction . A close relative of gammaretroviruses , HERV-T , circulated in primates for ~25 million years ( MY ) before apparent extinction within the past ~8 MY . Construction of a near-complete catalog of HERV-T fossils in primate genomes allowed us to estimate a ~32 MY old ancestral sequence and reconstruct a functional envelope protein ( ancHTenv ) that could support infection of a pseudotyped modern gammaretrovirus . Using ancHTenv , we identify monocarboxylate transporter-1 ( MCT-1 ) as a receptor used by HERV-T for attachment and infection . A single HERV-T provirus in hominid genomes includes an env gene ( hsaHTenv ) that has been uniquely preserved . This apparently exapted HERV-T env could not support virion infection but could block ancHTenv mediated infection , by causing MCT-1 depletion from cell surfaces . Thus , hsaHTenv may have contributed to HERV-T extinction , and could also potentially regulate cellular metabolism . When the targets of a given retrovirus include host cells that will become part of the germ-line , viral sequences can become inherited and sometimes fixed in host populations . Indeed , about 8% of the human genome is composed of inactive or fragmented endogenous retroviruses ( ERVs ) ( Lander et al . , 2001 ) . A few retroviral genes and regulatory DNA elements have been exapted to perform diverse host functions , such as syncytial trophoblast fusion ( Lavialle et al . , 2013 ) , and transcriptional regulation ( Chuong et al . , 2016; Fort et al . , 2014; Lu et al . , 2014; Macfarlan et al . , 2012; Rebollo et al . , 2012; Ting et al . , 1992 ) . In particular , a number of endogenous viral elements have been shown to exhibit antiviral activity . Indeed , gag ( Best et al . , 1996; Murcia et al . , 2007; Yan et al . , 2009 ) , env ( Gardner et al . , 1991; Ito et al . , 2013; Kozak , 2015; McDougall et al . , 1994; Robinson and Lamoreux , 1976; Spencer et al . , 2003 ) , and accessory genes ( Czarneski et al . , 2003; Frankel et al . , 1991 ) from endogenous viruses have been shown to contribute to host defense against exogenous retroviruses , through a variety of mechanisms . Although some functions have been ascribed to exapted retroviral fragments , most provide no obvious advantage to their host . Nonetheless , endogenous proviruses represent a fossil record of past infections , enabling the study of ‘paleovirology’ ( Emerman and Malik , 2010 ) . In particular , the synthesis of consensus or deduced ancestral sequences based on retroviral fossils has enabled the functional analyses of reconstructed proteins from ancient , presumptively extinct , retroviruses ( Dewannieux et al . , 2006; Goldstone et al . , 2010; Kaiser et al . , 2007; Lee and Bieniasz , 2007; Perez-Caballero et al . , 2008; Soll et al . , 2010 ) . Numerous members of the gammaretrovirus genus currently circulate in mammalian species , but exogenous gammaretroviruses appear absent from modern humans ( Bénit et al . , 2001 ) . A key question in virology , that is potentially answerable using paleovirological analyses , is what caused the elimination of extinct viral lineages ? Here , we address this question for an extinct lineage of gammaretroviruses that replicated in ancient primates ( HERV-T ) ( Blusch et al . , 1997 ) . While no functional HERV-T genes have been found in available modern primate genomes , we show that a single HERV-T env gene was apparently exapted in hominids . We find that the product of this HERV-T env gene can block retroviral infection mediated by a reconstructed , functional ancestral HERV-T env , by depleting the HERV-T receptor from cell surfaces . Thus , this exapted env gene may have driven the extinction of HERV-T from hominids . Using similarity search-based approaches , we constructed a comprehensive catalog of HERV-T fossils in old world monkey ( OWM ) and ape genomes ( Figure 1—source data 1 ) . Phylogenetic analysis of near-complete proviral elements revealed three major HERV-T clades ( T1–T3 ) , each of which was associated with phylogenetically distinct LTR sequences , ( Figure 1—figure supplement 1A ) . Two clades ( HERV-T3 and HERV-T2 ) were previously annotated ( LTR6A and LTR6B in RepBase [Bao et al . , 2015] ) while a third ( HERV-T1 ) was novel . Other HERV-T-like sequences ( found by similarity to HERV-T protein-coding genes ) are present in new world monkeys ( platyrrhini primates ) and documented in Repbase ( Bao et al . , 2015 ) . The presence of orthologous HERV-T loci in various primate species , and integration dates estimated from LTR divergence ( Figure 1—source data 1 ) , indicated that the infectious ancestor of HERV-T entered primate germlines between ~43 and ~32 MY ago ( MYA ) . Additional germline integration events occurred in various primate lineages for the ensuing 25 MY , with the most recent integrations that became fixed in primates having occurred ~11 MYA in hominids and ~8 MYA in macaques ( Figure 1—source data 1 ) . Comparison of recovered HERV-T fossils indicated that the HERV-T2 clade derives from a single , hypermutated HERV-T1 element ( Figure 1—figure supplement 1B ) , and likely expanded by complementation in-trans . HERV-T3 represented the oldest HERV-T clade in catarrhini primates , with integration dates ranging from ~8 to ~30 MYA , and had the highest overall copy number ( Figure 1A and Figure 1—source data 1 ) . There was some support for the division of HERV-T3 into two sub-clades ( Figure 1—figure supplement 1A ) . Using a maximum likelihood approach we inferred the sequence of an infectious ancestral HERV-T3 from near complete proviruses ( Figure 1A and Figure 1—source data 1 ) . Unlike fossilized viral elements , the reconstructed ancestral sequence had gag , pol and env genes with full coding potential , and revealed the likely presence of an additional open reading frame ( ORF ) of unknown function , located 5’ to gag ( Figure 1B and Figure 1—figure supplement 2 ) . We tested the functionality of the reconstructed ancestral HERV-T envelope gene ( ancHTenv , Supplementary file 1 ) by generating murine leukemia virus ( MLV ) particles containing a gfp reporter vector and pseudotyped with ancHTenv . MLV-ancHTenv virions were able to infect many cell lines at high titer ( 105–106 infectious units ( IU ) /ml , Figure 1C ) while some rodent ( NIH3T3 and Rat2 ) and chicken ( DF-1 ) cells were comparatively resistant to MLV-ancHTenv ( titers ≤ 103 IU/ml ) . 10 . 7554/eLife . 22519 . 003Figure 1 . Functional reconstruction of a ~32 MY old HERV-T envelope protein . ( A ) Phylogenetic tree of HERV-T3 proviral sequences in OWM and apes . Orthologous sequences are bracketed and estimated integration times indicated . * >90% bootstrap support . Red circle = ancestral root node . ( B ) Ancestral HERV-T3 genome with ORFs indicated . ( C ) Infectivity of MLV particles containing a GFP reporter and pseudotyped with the ancHTenv protein ( Mean ± SD , n = 3 replicates ) . See also Figure 1—figure supplements 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 00310 . 7554/eLife . 22519 . 004Figure 1—source data 1 . HERV-T elements in OWM and ape genomes . Type: HERV-T1 , HERV-T2 or HERV-T3 . Start and end: coordinates in the genome scaffold . Strand: orientation of the provirus . Assembly: identifier of the genome assembly used for the screening procedure . Scaffold: identifier of the nucleotide sequence where the provirus lies . Proviral structure: coding and non-coding features present in each provirus , shown in the order they occur ( LEA=leader sequence ) . Divergence: substitutions per site . Age: estimate of element age ( i . e . time since integration ) in millions of years , based on comparison of 3’ and 5’ LTRs ( for proviruses that encoded paired LTRs ) or comparison to a clade-specific LTR consensus sequence ( for solo LTRs ) . ND: non-determined . Proviruses highlighted in green correspond to the sequences used in the ancestral reconstruction of HERV-T3 . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 00410 . 7554/eLife . 22519 . 005Figure 1—figure supplement 1 . HERV-T proviruses cluster into four monophyletic clades . ( A ) Maximum likelihood phylogenetic tree of 44 HERV-T proviruses . Three distinct clades in OWM and apes ( HERV-T1 , HERV-T2 and HERV-T3 ) are indicated by colored blocks . HERV-T1 had LTRs that shared 64% and 55% identity with HERV-T2 and HERV-T3 s respectively . The tree was rooted using a related outgroup sequence from the squirrel monkey ( Saimiri boliviensis ) . Bootstrap support for internal nodes is indicated by asterisks: ( * ) >80% , ( *** ) >99% ( 1000 bootstrap replicates ) . The temporal appearance of these HERV-T-like integrations and the geographical separation of the platyrrhini and catarrhini primate lineages , suggest that platyrrhini and catarrhini HERV-T-like sequences most likely result from independent germ line invasion events . ( B ) Proviral sequences were aligned to a consensus HERV-T sequence and analysed using Hypermut 2 . 0 with default parameters . Image corresponds to nucleotides 6877 to 7876 spanning the junction of the pol and env genes . Lines in red , cyan green and magenta represent ( GG to AG , GA to AA , GC to AC and GT to AT transitions , respectively ) . Asterisks indicate statistical significance in a fisher-exact test for APOBEC3-mediated hypermutation . ( * ) p-value<0 . 05 , ( ) p-value<0 . 01 , ( * ) p-value<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 00510 . 7554/eLife . 22519 . 006Figure 1—figure supplement 2 . Deduced sequence of a 32MY old ancestral HERV-T3 . Nucleotide and translated amino acid sequence of the reconstructed ancestral HERV-T3 sequence . LTRs are shown in deep blue and bold . The leader sequence is shown in italics and underlined . The pre-gag , gag , pol and env ORFs are indicated in pink , red , green and purple color , respectively . Underlined in green is the sequence overlap between pol and env . The predicted signal peptide and furin cleavage sites are indicated by orange and red triangles , respectively . The predicted fusion peptide , immunosuppressive and transmembrane domains are highlighted in cyan , green and yellow , respectively . The CX6CC motif is indicated with a red box , and the cysteine residues involved in the disulfide-bonded loop are highlighted in red . Cleavage sites were predicted using ProP 1 . 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 006 When challenged with an MLV-ancHTenv expressing a neo resistant gene , even DF-1 and NIH3T3 cells could be infected at low efficiency ( titers of ~4 × 102 and ~1 × 103 G418-resistant colonies ( G418-RC ) /ml ) . Therefore , DF-1 cells , which exhibited the lowest susceptibility to MLV-ancHTenv , were selected to identify candidate HERV-T receptors as they would give the lowest ‘background’ level of infection in the context of a cDNA library screen . We generated a lentivirus-based human cDNA library from human 293T cells , which were among the most highly permissive cell lines tested , and then iteratively challenged and selected library-transduced DF-1 cells with MLV-ancHTenv virions containing antibiotic resistance or fluorescent protein genes ( Figure 2A ) . This procedure progressively enriched MLV-ancHTenv-susceptible DF-1 cells ( Figure 2—figure supplement 1A ) . Lentiviral-vector directed PCR primers were then used to identify a human cDNA in the selected cells , that encoded human monocarboxylate transporter 1 ( hMCT1 , also known as SLC16A1 , Figure 2—figure supplement 1B ) , a 12-pass transmembrane protein that mediates the transport of monocarboxylates such as lactate and pyruvate across the plasma membrane , and is upregulated in some cancers ( Halestrap , 2012 , 2013 ) . Reintroduction of an hMCT1 cDNA into naïve DF-1 cells rendered them highly sensitive to infection with MLV-ancHTenv ( Figure 2B and C ) . Moreover , MCT1 expression in DF-1 cells also conferred dramatically enhanced ability to bind to MLV Gag-GFP virus-like particles ( VLPs ) pseudotyped with ancHTenv , but had no effect on the low-level binding of VLPs that lacked an Env protein or incorporated an ecotropic MLV envelope ( Figure 3 ) . Thus , MCT1 was likely the receptor used by HERV-T to infect ancient old world primates . 10 . 7554/eLife . 22519 . 007Figure 2 . MCT-1 functions as a receptor for ancestral HERV-T . ( A ) Scheme of the receptor screening strategy . DF-1 cells were transduced with a lentiviral cDNA library . Two days later , the cells were challenged with MLV-ancHTenv containing a neo gene . After a further two days , cells were placed in G418 selection . After another 10 days , G418-resistant cells were replated and challenged with MLV-ancHTenv containing a hygromycin resistance gene . Two days later cells were placed in hygromycin selection . After a further 10 days , Hygromycin-resistant cells were replated and challenged with MLV-ancHTenv containing RFP and were found to be highly susceptible to infection ( Figure 2—figure supplement 1A ) . Genomic DNA ( gDNA ) was extracted from this cell population and subjected to PCR using primers specific to the lentiviral vector ( Figure 2—figure supplement 1B ) . ( B ) Fluorescent micrographs of 293T and DF-1 cells , expressing hMCT1 or a control protein , following infection with MLV-ancHTenv expressing GFP as reporter . Scale bar = 200 μm . ( C ) Titers of MLV-ancHTenv/GFP on 293T and DF-1 cells expressing hMCT1 or a control protein ( Mean ± SD , n = 2 replicates , one of two separate experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 00710 . 7554/eLife . 22519 . 008Figure 2—figure supplement 1 . HERV-T receptor identification . ( A ) Relative MLV-ancHTenv/RFP sensitivity of library transduced ( Mean ± SD , n = 3 separate pools of cells assayed once each ) , and untransduced DF1 cells subjected to the infection selection strategy outlined in ( Figure 2A ) . ( B ) Products of PCR reactions from three RFP positive DF-1 cell populations , and control DF-1 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 00810 . 7554/eLife . 22519 . 009Figure 3 . Binding of ancHTenv-pseudotyped MLV particles to DF1 cells expressing MCT1 . ( A ) Fluorescent micrographs of MLV Gag-GFP VLPs pseudotyped with ancHTenv or ecotropic MLV , bound to DF-1 cells expressing hMCT1 or an empty vector . Scale bar = 5 μm . ( B ) Enumeration of MLV Gag-GFP VLPs bearing the indicated Env proteins bound to DF-1 cells expressing MCT-1 or an empty vector . Each data point represents an individual cell ( n = 20 for each condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 009 Notably , the modern human genome harbors a single HERV-T provirus that includes an env gene with nearly full coding potential ( de Parseval et al . , 2003 ) . This env ORF lacks only five amino acids at the C-terminus , shares 86% amino acid identity with the reconstructed ancHTenv and is expressed mainly in healthy thyroid tissue ( de Parseval et al . , 2003 ) . The provirus is also present at the orthologous site in gorilla and orangutan genomes , but absent in chimpanzees due to a segmental deletion at that locus . This provirus originated from bona-fide retroviral integration ( presence of a 4 bp target site duplication ) into the internal sequence of another HERV ( Repbase: HERVIP10FH ) , in a locus surrounded by other repetitive sequence elements . Sequence similarity searches using the env region suggested the absence of an orthologous insertion in non-hominid primate genomes . Moreover , the divergence between the paired LTRs of this provirus gave an estimated integration date of ~7–19 MYA ( Table 1 ) . Adjustment of this estimate to account for species distribution leads to the conclusion that this provirus was inserted into the germline of the ancestor of modern hominids at least ~13–19 MYA . 10 . 7554/eLife . 22519 . 010Table 1 . Molecular evolution analyses of the provirus containing hsaHTenv in its orthologs . ( 1 ) Divergence measured in substitutions per site . ( 2 ) Integration dates inferred from the divergence of the paired LTRs for human , gorilla and orangutan proviruses . Age is calculated assuming a human neutral substitution rate of 2 . 2 × 10−9 substitutions per site per year . ( 3 ) Pairwise dN/dS ratios for the env genes calculated using codeml ( CodonFreq = F3 × 4 , Kappa and Omega estimated ) . ( ) Significantly different from dN/dS = 1 ( p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 01010 . 7554/eLife . 22519 . 011Table 1—source data 1 . Likelihood ratio tests on dN/dS estimates for hsaHTenv and its orthologs . dN/dS ratios ( ω ) were estimated on a pairwise basis using codeml . Likelihood ratio tests were performed comparing the log likelihood of the estimated ω ( L1 ) to the log likelihood when ω was fixed to 1 ( L0 , neutral selection ) . The probability ( P ) of twice the difference ( D = 2 ( L1−L0 ) ) was calculated using a chi-squared distribution ( degrees of freedom = 1 ) . The single nucleotide frame-shift insertion in the gorilla sequence was artificially deleted in order to compare orthologous codons . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 011LocusLTR divergence ( 1 ) Age ( MY ) ( 2 ) Env dN/dS ratio ( 3 ) GorillaOrangutanHuman0 . 03177 . 201 . 48600 . 5184Gorilla0 . 04149 . 41−0 . 7592Orangutan0 . 085519 . 430 . 7592− Unlike the reconstructed ancHTenv , the Env protein product of the modern human HERV-T provirus ( hsaHTenv ) was not able to generate infectious pseudotyped MLV particles ( Figure 4A ) . Moreover , unlike ancHTenv-HA , the hsaHTenv-HA protein was not correctly processed and was not incorporated into viral particles ( Figure 4B ) . Furthermore , while ancHTenv expression was able to drive syncytium formation , hsaHTenv had no fusogenic activity ( Figure 4C and D ) . Inspection of the hsaHTenv and orthologous env sequences in gorillas and orangutans revealed mutations at the site where proteolytic cleavage by furin-like proteases generates surface ( SU ) and transmembrane ( TM ) subunits ( Hallenberger et al . , 1992 ) ( Figure 4E ) . Insertion of these cleavage site mutations into ancHTenv abolished cleavage and pseudotype infection , while reversion of mutations in hsaHTenv did not correct the cleavage defect and did not yield infectious MLV particles ( Figure 4—figure supplement 1 ) , Thus , loss of HERV-T envelope function by hsaHTenv was a multi-step process , including loss of furin cleavability , similar to findings with feline endogenous retroviral Env proteins ( Ito et al . , 2016 ) . 10 . 7554/eLife . 22519 . 012Figure 4 . The human genome encodes a HERV-T Env ORF that does not function as a retroviral envelope . ( A ) Infectiousness of MLV particles pseudotyped with untagged or C-terminally HA-tagged ancHTenv or hsaHTenv ( Mean ± SD , n = 3 replicates , one of two experiments ) . ( B ) Western blot analyses ( α-CA and α-HA ) of cell lysates and MLV virions generated following expression of C-terminally HA-tagged ancHTenv or hsaHTenv M: markers . ( C ) Examples of cell fusion in 293T cell cultures expressing ancHTenv or hsaHTenv linked to IRES-GFP . Scale bar = 100 μm . ( D ) Percentage of GFP+ multinucleated cells ( >5 nuclei/cell ) in 293T cell cultures expressing ancHTenv or hsaHTenv linked to IRES-GFP ( Mean ± SD , n = 3 groups of ten microscopic fields , one of two experiments ) Blue: DAPI . ( E ) Alignment of ancHTenv and intact or nearly intact HERV-T Env protein sequences encoded by hominid genomes , proximal to the furin cleavage site . See also Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 01210 . 7554/eLife . 22519 . 013Figure 4—figure supplement 1 . Effects of mutations at the furin cleavage site on ancHTenv and hsaHTenv processing . Western blot ( anti-CA and anti-HA ) analyses of cell lysates and virions following expression of MLV Gag-Pol and HA-tagged ancHTenv , hsaHTenv , or furin cleavage site-modified derivatives . AncHTenv-FurinMut contains the furin cleavage site residues from hsaHTenv . HsaHTenv-FurinFix contains the furin cleavage site residues from ancHTenv . M= molecular weight markers . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 013 Although the hsaHTenv did not maintain function , it has retained a complete open reading frame ( ORF ) since its deposition in an ancestral hominid genome . To determine the degree to which this ORF preservation should be expected , we deduced an ancestral env gene sequence for this particular hominid provirus ( ancHTenv16MYA ) and simulated its neutral evolution using the human neutral substitution rate ( Lander et al . , 2001 ) . Only 6% of sequences maintained the env ORF following simulated neutral evolution for 13–19 MY ( Figure 5A ) . Note that 6% represents an overestimate as indels were not simulated . Strikingly , inspection of orthologous human and orangutan proviruses revealed highly selective preservation of the env ORF , while numerous stop codons ( 6 to 16 ) and frameshifts ( 3 to 5 ) were present in gag and pol genes of the same provirus ( Figure 5B and Table 2 ) . For the gorilla provirus , a single nucleotide frameshift was the only lesion in env , suggesting recent inactivation . Pairwise sequence comparisons revealed that the human and orangutan env genes show signatures of purifying selection ( dN/dS ≈ 0 . 5 , p<0 . 01 ) , whereas comparisons with the gorilla env suggest a relaxed selection ( Table 1 ) . Overall , selective pressures appear to have preserved this HERV-T env ORF despite the loss of its retroviral envelope function , and numerous inactivating mutations in the accompanying gag and pol genes . 10 . 7554/eLife . 22519 . 014Figure 5 . Preservation of hsaHTenv and its orthologs in hominids . ( A ) Monte-Carlo simulations of ancHTenv16MYA evolution for 13 . 45–19 . 68 MY using a human neutral substitution rate . The percentage of 10 , 000 simulated sequences of each type is plotted with error bars indicating maximum and minimum estimates . ( B ) Distribution of stop codons ( colored circles ) in the Gag , Pol and Env coding sequence of proviruses orthologous to the HERV-T provirus containing hsaHTenv . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 01410 . 7554/eLife . 22519 . 015Table 2 . Analysis of inactivating mutations of hsaHTenv encoding provirus in human , gorilla and orangutan genomes . Proviral sequences were aligned to the ancestral HERV-T3 sequence . Nonsense mutations and frameshift indels relative to the ancestral sequence were quantified . ‘Indel’ = insertion or deletion events compared to the ancestral HERV-T3 sequence that resulted in a reading frame change . ‘Stop’ = mutations that resulted in a stop codon . ( * ) Indicates a single nucleotide insertion that results in the truncation of the last five amino acids compared with ancHTenv . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 015LocusGag ( length: 519 codons ) Pol ( length: 1206 codons ) Env ( length: 631 codons ) IndelStopIndelStopIndelStopHuman46316 ( 1* ) 0Gorilla464151 ( 1* ) 0Orangutan59315 ( 1* ) 0 We tested the antiviral potential of the hsaHTenv and its direct ancestor in hominids ( ancHTenv16MYA , Supplementary file 1 ) by expressing them into 293T cells , as well as DF-1 clones expressing hMCT1-HA ( Figure 6A ) . Strikingly , hsaHTenv and ancHTenv16MYA inhibited MLV-ancHTenv infection by >10 fold in both cell lines ( Figure 6B and C and Figure 6—figure supplement 1 ) . Cell clones ( 293T ) stably expressing hsaHTenv were ~50–100-fold less susceptible to MLV-ancHTenv as compared to control cells , but fully susceptible to amphotropic MLV infection ( Figure 6D ) . Western blot analyses showed that hsaHTenv , ancHTenv16MYA or ancHTenv expression caused depletion of endogenous MCT1 in 293T cells ( Figure 7A ) . In DF-1 cells , expression of hsaHTenv-HA , ancHTenv-HA or ancHTenv16MYA-HA clearly depleted Flag-tagged hMCT1 from the surface of nearly all cells examined ( Figure 7B and Figure 7—figure supplement 1 ) . In the majority of hsaHTenv-HA , ancHTenv-HA or ancHTenv16MYA-HA expressing DF-1 cells , MCT1-FLAG immunofluorescence was nearly undetectable while a few cells showed low levels of intracellular staining ( a mixture of diffuse and punctate MCT1-FLAG staining ) ( Figure 7—figure supplement 1 ) . Thus , these data collectively suggest that ancHTenv16MYA and hsaHTenv exhibit antiviral activity specifically against HERV-T by causing a reduction in the levels of MCT1 at the cell surface . 10 . 7554/eLife . 22519 . 016Figure 6 . The hsaHTenv protein specifically inhibits HERV-T infection . ( A ) Scheme of the antiviral assay . Cells , 293T or DF-1 expressing hMCT1 , were transduced with lentiviral vectors expressing HERV-T Env proteins or an unrelated protein ( No Env ) together with a GFP reporter gene to monitor expression . Cells were challenged with MLV particles pseudotyped with ancHTenv that expresses RFP upon infection . Cell populations were analyzed by FACS 2 days after infection . ( B ) Representative experiment based on the scheme depicted in ( A ) conducted using 293T cells expressing the indicated Env proteins . ( C ) Infectivity of MLV-ancHTenv on 293T cells or two clones of MCT1-expressing DF1 cells ( #14 , #17 ) , expressing HERV-T envelope proteins according to the scheme described in ( A ) . Plots describe the percentage of RFP positive cells ( infected ) after gating on the GFP positive ( Env-expressing ) cell population . ( Mean ± SD , n = 3 independent experiments ) . ( D ) Susceptibility of clones of 293T cells expressing an empty vector or hsaHTenv-HA to infection by MLV-ancHTenv/GFP or amphotropic MLV/GFP ( Mean ± SD , n = 3 three independent single cell clones assayed once each ) . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 01610 . 7554/eLife . 22519 . 017Figure 6—figure supplement 1 . Antiviral activity of HERV-T Env proteins . Single cell clones of DF-1 cells that expressed hMCT1-HA , or naïve DF-1 cells , were transduced with lentiviral vectors expressing HERV-T Env proteins or an unrelated protein ( Control ) together with a GFP reporter gene to monitor expression . Cell populations were analyzed by FACS 2 days after infection . A represesentative experiments is shown in which the percentage of RFP positive ( infected ) cells is determined after gating on the on the GFP positive ( Env-expressing ) cell population . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 01710 . 7554/eLife . 22519 . 018Figure 7 . The hsaHTenv protein causes depletion of hMCT1 from the cell surface . ( A ) Western blot analyses ( α-HA , α-hMCT1 and α-hHsp90 ) of 293T cell lysates generated following transduction with lentiviral vectors encoding C-terminally HA-tagged HERV-T envelopes . ( B ) Immunofluorescent micrographs of hMCT1-Flag-expressing DF-1 cells transduced with vectors expressing hsaHTenv-HA or an empty vector . Green: α-Flag , red: α-HA , Blue: DAPI . Scale bar = 5 μm . See also Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 01810 . 7554/eLife . 22519 . 019Figure 7—figure supplement 1 . Additional examples of MCT-1 depletion and releocalization following expression of HERV-T Env proteins . Immunofluorescent micrographs of hMCT1-Flag-expressing DF-1 cells transduced with lentiviral vectors expressing various HA-tagged HERV-T envelopes or an empty vector . Green: α-Flag , red: α-HA , Blue: DAPI . Scale bar = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 22519 . 019 These findings suggest a scenario in which HERV-T began to infiltrate primate germ lines , using MCT1 as a receptor ~43–32 MYA . Later , ~19–13 MYA , a HERV-T Env sequence ( represented approximately by ancHTenv16MYA ) was exapted by ancestral hominids and to serve as an antiviral gene . AncHTenv16MYA apparently functioned by interacting with MCT1 , either at the cell surface promoting its internalization , or in the secretory pathway , blocking its transport to the plasma membrane , leading to MCT1 degradation and its depletion from the cell surface . This process is analogous to the specific receptor interference phenomena observed in retroviral infections ( Nethe et al . , 2005 ) and exploited by endogenous env genes that exhibit antiviral activity in other species ( Gardner et al . , 1991; Ito et al . , 2013; Kozak , 2015; Robinson and Lamoreux , 1976; Spencer et al . , 2003 ) . Given the fusogenic potential of ancHTenv , it is likely that the progenitor of hsaHTenv would have imposed a fitness cost on its ancestral hominid host . We speculate that the progenitor of hsaHTenv acquired mutations either concurrent with , or after , its integration that resulted in loss of its fusogenic potential ( e . g . furin cleavage site mutations ) . This sequence ( represented approximately by ancHTenv16MYA ) retained full length open reading frame and MCT1 binding activity , and became fixed in the ancestral hominid population . Indeed , it is plausible that this exapted gene might have contributed to the apparent extinction of HERV-T from hominids at some point after the deposition of the most recent fixed hominid germline insertion , ~11 MYA . The selection pressures acting on hsaHTenv , its orthologs and progenitors might have been complex . Loss of fusogenicity but retention of receptor binding activity are , in a sense , opposing influences in terms of maintenance of the ORF and its function . Moreover , although some purifying selective pressure ( low dN/dS ) can be detected in hsaHTenv and its orangutan ortholog , a relaxation of those forces might be expected to have occurred after HERV-T extinction , perhaps leading to reduced antiviral activity observed in modern hsaHTenv . Finally , as MCT-1 is a moncaboxylate transporter that is upregulated in human cancers ( Halestrap , 2013 ) , the ability of hsaHTenv to deplete MCT1 from cell surfaces may suggest an additional or alternative metabolism-related cellular or even anti-tumor functions and selective pressures . Such an activity might have continued to shape hsaHTenv sequence independent of anti-HERV-T activity . In either case , this study highlights the potential importance of ERV proteins as raw material for the innovation of new functions in human ancestors . Screening for HERV-T elements was performed using a BLAST-based strategy implemented using the Database-Integrated Genome Screening ( DIGS ) tool ( http://giffordlabcvr . github . io/DIGS-tool/ ) . A HERV-T genome consensus sequence ( HERV-Tcons , constructed from previously identified HERV-T sequences [Bénit et al . , 2001] ) was used as query to mine old world monkey ( OWM ) and ape genomes . Specifically , amino-acid and nucleotide sequences of the consensus HERV-T genome were used as probes for tBLASTn ( RRID: SCR_011822 ) ( gag , pol and env ) and BLASTn ( RRID: SCR_001598 ) ( LTR ) searches of a genome target database containing complete and low coverage primate genome sequences that were retrieved from publicly available databases: Human ( GenBank: GCF_000002125 . 1 ) , chimpanzee ( GenBank: GCF_000001515 . 6 ) , bonobo ( GenBank: GCA_000258655 . 2 ) , gorilla ( GenBank: GCA_000151905 . 1 ) , orangutan ( GenBank: GCF_000001545 . 4 ) , gibbon ( GenBank: GCA_000146795 . 1 ) , baboon ( UCSC: papHam1 ) , and rhesus macaque ( GenBank: GCA_000002255 . 2 ) . To avoid the identification of sequences from related but distinct retroviruses , significant BLAST hits ( e-value < 1E-50 ) were used as probes for a second round of BLASTx ( RRID: SCR_001653 ) ( translated coding hits ) or BLASTn ( non-coding hits ) searches against a sequence reference library containing exogenous and endogenous gammaretroviruses and other class-I ERVs ( including HERV-Tcons ) . Exogenous retroviral sequences were retrieved from the RefSeq database ( RRID: SCR_003496 ) ( Pruitt et al . , 2014 ) , and ERV sequences were based either on consensus sequences of previously published ERV sequence data ( Bénit et al . , 2001; Sverdlov , 2005; Tristem , 2000; Villesen et al . , 2004 ) , or previously inferred consensus sequences ( Jern et al . , 2005 ) . BLAST hits assigned to HERV-T were first ordered by genome scaffold and orientation , and adjacent or overlapping entries were assembled into proviral loci by comparison with HERV-Tcons , allowing for insertions no longer than 10 , 000 nt . Due to the high numbers of mutations , especially indels , accumulated in ERV sequences , 44 nearly-complete HERV-T sequences were aligned to HERV-Tcons using MUSCLE ( RRID: SCR_011812 ) ( Edgar , 2004 ) in a pair-wise fashion , followed by the creation of a ‘gapped’ multiple sequence alignment ( MSA ) using the profile alignment function . Insertions relative to HERV-Tcons were removed from the MSA , but saved in a separate file allowing for the restoration of the removed insertions if required . Maximum likelihood ( ML ) phylogenetic trees were constructed from a nucleotide MSA ( described above ) using raxML ( RRID: SCR_006086 ) ( Stamatakis , 2006 ) with the following parameters: rapid bootstrap analysis with 1000 replicates under the GTRCAT model followed by a ML search under GTRGAMMA model to evaluate the final tree topology ( -m GTRCAT -# 1000 -x 13 k -f a ) . Phylogenetic trees were then analyzed using FigTree v1 . 4 . 2 ( RRID: SCR_008515 ) ( Rambaut , 2008 ) . Thereafter , the tree was rooted using a platyrrhini HERV-T-like sequence as outgroup . The resulting tree showed three monophyletic HERV-T clades ( corresponding to HERV-T1 , HERV-T2 and HERV-T3 in old world primates ) whose LTR sequences were clearly distinct . Consensus sequences were derived from each clade and used to re-classify the results from the primate genome screening . Dates of integration for HERV-T elements were calculated using PAUP* ( RRID: SCR_014931 ) ( Swofford , 2002 ) by determining the divergence ( K ) from: ( i ) the corresponding consensus sequence ( for soloLTRs ) , or ( ii ) to their cognate LTR ( for proviral loci flanked by paired LTRs ) . The resulting K was divided by two times the neutral substitution rate ( r ) in order to obtain the estimated integration date ( K/2r ) ( Lebedev et al . , 2000; Subramanian et al . , 2011 ) . The human neutral substitution rate used was 2 . 2 × 10−9 substitutions per site per year ( Lander et al . , 2001 ) . APOBEC3-derived hypermutation analysis and statistics were performed using Hypermut 2 . 0 ( RRID: SCR_014933 ) ( Rose and Korber , 2000 ) on 49 gag-pol-env HERV-T containing sequences in OWM and ape genomes using default parameters . HERV-T sequences with p-value<0 . 05 in a Fisher exact-test , were treated as significant for APOBEC3- derived hypermutation . dN/dS ( ω ) ratios were calculated on a pairwise basis , by comparing the aligned codon sequences of the human , gorilla ( after artificially removing a 1nt frame-shift insertion ) and orangutan env genes , using codeml from the PAML package of programs ( RRID: SCR_014932 ) ( Yang , 1997 ) ( runmode = −2 , CodonFreq = F3 × 4 , Kappa and Omega estimated ) . Likelihood ratio tests were performed by comparing the log likelihood of the estimated ω ( L1 ) to the log likelihood when ω was fixed to 1 ( L0 , neutral selection ) . The probability ( P ) of twice the difference ( D = 2 ( L1−L0 ) ) was calculated using a chi-squared distribution ( degrees of freedom = 1 ) . For the ancestral reconstruction of HERV-T3 , we first analyzed the 5’ and 3’ flanking sequences of 22 near complete HERV-T3 proviruses retrieved from old world monkey and ape genomes . These analyses resulted in fourteen unique HERV-T3 integration events ( orthologous groups ) . Ancestral nodes for each orthologous group were used for the ancestral reconstruction . Initial ML ancestral reconstructions were guided by a MSA together with a phylogenetic tree using baseml ( PAML ( RRID: SCR_014932 ) ( Yang , 1997 ) , model: REV , branch lengths were used as initial values ) . Initial values of alpha and kappa were calculated on the MSA by jmodeltest ( Darriba et al . , 2012 ) . A correction for the effect of methylation-induced mutations at CpG dinucleotides was applied on both strands of all ancestral reconstructed sequences as previously described ( Goldstone et al . , 2010 ) . The resulting ancestral HERV-T3 sequence was further refined in two locations . A single nucleotide insertion , relative to the ancestral HERV-T1 sequence , was eliminated in the 5’ pre-gag ORF resulting in the suppression of a frame-shift mutation and an expansion of 43 codons relative to the original ancestral HERV-T3 sequence . Additionally , the state of particularly polymorphic sites in the pre-gag ORF was hand curated by a combination of their frequency and the phylogenetic relationships between the corresponding HERV-T3 sequences . The variation present in HERV-T1 sequences was used to break possible ties . A similar procedure was performed on the initial HERV-T3 ancestral env sequence with the additional consideration that if the CpG reversion procedure , at a particular position , resulted in the change in chemical nature of the amino acid , the residue at this position was reverted to its ‘pre-CpG reversion’ state . The final revised sequence corresponded to ancHTenv ( Supplementary file 1 ) . The ancestral reconstruction for ancHTenv16MYA ( Supplementary file 1 ) was performed manually by selecting residues based on the phylogenetic relationships of the three Env sequences present in human , gorilla and orangutan . Polymorphic sites were resolved by comparison with the ancHTenv sequence . Characterization of the hsaHTenv orthologs in other primate species and the ~196 Kb segmental deletion in the Chimpanzee genome , was done using the UCSC genome browser ( RRID: SCR_005780 ) ( Kent et al . , 2002 ) and tools implemented therein . Stop codon analysis was performed by translating the proviral sequences in all three forward reading frames and the location of their stop codons was plotted using R ( RRID: SCR_001905 ) . Frameshift mutations were identified by aligning proviral sequences to the ancestral HERV-T3 sequence . Monte-Carlo simulations of in silico neutral evolution on the ancHTenv16MYA were performed using seq-gen ( RRID: SCR_014934 ) ( Rambaut and Grassly , 1997 ) under the GTR model ( 10 , 000 iterations ) as previously described ( Katzourakis and Gifford , 2010 ) . Expected branch lengths were calculated for the minimum and maximum estimates of the origin of hominids ( 13 . 45 and 19 . 68 MY respectively ) ( Perelman et al . , 2011 ) using the neutral substitution for humans . The simulated 10 , 000 sequences were then evaluated for the presence of a 5’ methionine and premature stop codons . Analysis for the features of HERV-T envelope sequences was performed using tmhmm 2 . 0 ( RRID: SCR_014935 ) ( Krogh et al . , 2001 ) for transmembrane and hydrophobic domains , and ProP1 . 0 ( RRID: SCR_014936 ) ( Duckert et al . , 2004 ) for signal peptide and propeptide cleavage sites . The codon-optimized sequences for expression in human cells of ancHTenv , hsaHTenv ( GenBank: AB266802 ) and ancHTenv16MYA were synthesized ( Genewiz , South Plainfield , NJ ) and subsequently inserted into the pCAGGS expression vector ( Niwa et al . , 1991 ) using EcoRI and XhoI ( NEB , Ipswich , MA ) restriction enzymes . Furin cleavage site modified HERV-T envelopes were generated by exchanging the furin cleaveage sites of ancHTenv and hsaHTenv , and vice versa . Specifically , mutagenic PCR primers that annealed to sequences encoding the respective furin cleavage sites were used in overlapping PCR reactions , resulting in the generation of ancHTenv-FurinMut ( SRFRRAA to PRLHQAV ) or hsaHTenv-FurinFix ( PRLHQAV to SRFRRAA ) . The PCR reactions were treated with DpnI ( NEB , Ipswich , MA ) restriction enzyme for an hour at 37°C to eliminate plasmid DNA . The complete PCR fragment was inserted into pCAGGS using EcoRI and XhoI restriction sites encoded in the outer primers . HA-tagged HTenv expression plasmids were generated by introducing two copies of an HA-tag at the C-termini of all HERV-T envelopes using PCR and primers containing the tag DNA sequence and a 15nt linker sequence . The complete PCR fragments were inserted back into pCAGGS using EcoRI and XhoI restriction sites contained in the PCR primers . HA-tagged HERV-T envelope constructs were also subcloned into pCCIB ( a lentiviral expression vector containing a CMV-promoter and IRES-blasticidin resistance cassette ) , using SfiI ( NEB , Ipswich , MA ) restriction sites contained in the PCR primers , to generate stably-expressing cell lines and clones . HA-tagged HERV-T envelope constructs or a control protein ( GFP ) were also subcloned into pCCIGW ( lentiviral expression vector containing a CMV-promoter and IRES-GFP ) , using SnaBI and BstXI ( NEB , Ipswich , MA ) restriction sites contained in the PCR primers . The sequence encoding human MCT1 ( hMCT1 , UniProt: P53985 ) was amplified from the gDNA of DF-1 cells that had been transduced with a 293T cDNA library and selected as described below . The amplified hMCT1 sequence was inserted into pCCIB utilizing SfiI sites , and the resulting vector used to generate stable DF-1 cell lines expressing hMCT1 . C-terminally HA-tagged or Flag-tagged versions of hMCT1 were generated using PCR with primers containing two HA-tag or three Flag-tag encoding sequences and a 15nt linker sequence . The complete PCR fragment was inserted back into pCCIB using SfiI restriction sites contained in the PCR primers . All cells were purchased from ATCC ( Manassas , VA ) , except MT2 ( RRID: CVCL_2631 ) and NIH3T3 ( RRID: CVCL_0594 ) that were obtained through the NIH AIDS Reagent Program , and Huh7 . 5 ( RRID: CVCL_7927 ) that were a gift from Charles M Rice , where they were derived . Cells were assumed to be authenticated by their respective suppliers and were not further characterized . Cells were maintained in Dulbecco’s Modified Eagle Medium ( DMEM ) ( 293T ( RRID: CVCL_0063 ) , DF-1 ( RRID: CVCL_0570 ) , HT1080 ( RRID: CVCL_0317 ) , K562 ( RRID: CVCL_0004 ) , Huh7 . 5 , FRhK4 ( RRID: CVCL_4522 ) , MusDunni ( RRID: CVCL_9125 ) , NIH3T3 and Rat2 ( RRID: CVCL_0513 ) ) , Eagle’s Minimum Essential Medium ( EMEM ) ( MRC5 ( RRID: CVCL_0440 ) , CV-1 ( RRID: CVCL_0229 ) , CRFK ( RRID: CVCL_2426 ) and MDCK ( RRID: CVCL_0422 ) ) , Roswell Park Memorial Institute medium ( RPMI ) ( MT2 ) , or Ham’s F-12 media ( CHO ( RRID: CVCL_0214 ) ) supplemented with 10% FBS or BCS ( NIH3T3 ) , 1 mM of L-glutamine ( CHO ) and gentamycin ( 2 µg/ml ) ( Gibco , Waltham , MA ) according to ATCC instructions . Mycoplasma testing was not specifically performed , but many cell lines were used in immunofluorescence assays with DAPI staining that should reveal presence of mycoplasma . All cells were incubated at 37°C except DF1 that were incubated at 39°C . To generate stable cell lines , 293T cells were transfected with plasmids expressing HIV-1 Gag-Pol ( pCRV1 [Zennou et al . , 2004] ) , VSV glycoprotein and pCCIB plasmids encoding hMCT1 or HA-tagged/untagged versions of HERV-T envelopes using polyethylenimine . In every case virus/vector containing supernatants were harvested and filtered two days after transfection and were used to transduce naïve 293T or DF-1 cells . Transduced cells were expanded in 10 cm dishes with media supplemented with 5 µg/ml ( 293T ) or 20 µg/ml ( DF-1 ) blasticidin S ( Thermo Fisher Scientific Inc . , Waltham , MA ) and were monitored from 3 to 10 days before performing experiments or isolating single cell clones . MLV particles pseudotyped with MLV amphotropic ( MLV-A ) or ecotropic ( MLV-E ) , and different HERV-T envelopes were generated by co-transfecting 293T cells with ( i ) the corresponding envelope plasmids , plasmids ( ii ) expressing MLV gag-pol polyprotein ( MLVgp ) and ( iii ) an MLV vector encoding a neo gene and/or GFP/RFP ( pCNCG/pCNCR ) ( Soneoka et al . , 1995 ) using polyethylenimine . Viruses were harvested and filtered ( 0 . 22 µm ) two days after transfection . Viral stocks were concentrated using the Amicon Ultra-15 filters ( 10 kDa ) ( Millipore , Billerica , MA ) before freezing . Cells of interest were infected with serial dilutions of the corresponding virus supplemented with 4 µg/ml of polybrene . Viral titers were calculated by measuring the percentage of infected cells expressing GFP or RFP , 2 days post infection ( dpi ) using the Guava EasyCyte flow cytometer ( Millipore ) . For MLV particles expressing neo resistance genes , viral titers were calculated by expanding the infected cells in 10 cm dishes with media supplemented with 1 mg/ml ( NIH3T3 ) or 2 mg/ml ( DF-1 ) of G418 and monitoring for 10 days before resistant colonies were counted . Total RNA was isolated from a confluent 10 cm dish of 293T cells using Trizol ( Invitrogen , Carlsbad , CA ) . mRNA transcripts were enriched using Oligotex polyA+ resin ( Qiagen , Hilden , Germany ) . Polyadenylated RNA was used to construct a cDNA library using the SMART cDNA Library Construction Kit ( Clontech , Mountain View , CA ) . Briefly , cDNA containing SfiI restriction sites was synthesized using the SMARTScribe MMLV reverse transcriptase ( Clontech ) with SMART primers . The resulting cDNA was further amplified for 15 cycles using Phusion High-Fidelity DNA Polymerase ( Thermo Fisher Scientific Inc . ) using SMART primers . PCR products were treated with Proteinase K ( Clontech ) for 20 min at 45°C before they were digested by SfiI restriction enzyme for 2 hr at 50°C . Digested cDNA was then size fractionated using CHROMA SPIN-400 columns ( Clontech ) . The cDNA containing fractions ( >500 bp ) were selected to ligate into a pCCIB plasmid encoding corresponding SfiI sites . Overnight ligation was performed using T4 DNA ligase ( NEB ) and the library was expanded by transformation into electrocompetent DH5α bacteria cells . The 293T cDNA library had a complexity of 3 . 5 × 106 colony forming units . Transformed bacteria were cultured as suspended colonies in 6 liters of SeaPrep soft agarose ( Lonza , Basel , Switzerland ) diluted in 2x LB media at 30°C for 2 . 5 days . Thereafter plasmid DNA was extracted and used for receptor cloning . Lentiviral vector stocks carrying the cDNA library were produced by transfecting 6 × 106 293T cells with plasmids expressing HIV-1 gag-pol polyprotein ( pCRV1 ) , VSV glycoprotein and the cDNA library plasmid ( pCCIB ) . Library containing supernatants were harvested , filtered ( 0 . 22 µm ) , concentrated ( Amicon Ultra-15 filters 10 kDa , Millipore ) and frozen two days after transfection . DF-1 cells were transduced with the library at an MOI of ~8 . Transduced DF-1 cells were then challenged with MLV particles pseudotyped with ancHTenv and containing a neo resistance gene ( pCNCG ) two days later . Infected DF-1 cells were placed in G418 selection at 2 dpi and resistant colonies were collected after 10 days . G418 resistant DF-1 cells were then challenged with ancHTenv pseudotyped MLV particles containing a hygromycin resistance gene ( pLHCX ) . Cells were placed in hygromycin selection at 2 dpi and resistant colonies were collected after 10 days . Hygromycin resistant DF-1 cells were infected with ancHTenv pseudotyped MLV particles expressing RFP ( pCNCR ) and found to be susceptible to infection . Genomic DNA ( gDNA ) was extracted from this DF-1 cell population and possible receptor candidates were amplified using PCR and primers directed to the pCCIB vector sequences flanking the SfiI cDNA insertion site . A non-coding isoform of human atg12 and hMCT1 were amplified from gDNA of transduced and selected DF-1 cells . For all PCRs performed in this study Phusion High-Fidelity DNA Polymerase was used . Fluorescent MLV VLPs were generated by co-transfecting 293T cells with a plasmid ( pCAGGS ) expressing MLV Gag-GFP and a plasmid ( pCAGGS ) expressing ancHTenv , MLV-E or an empty vector , using polyethylenimine . VLPs were harvested and filtered ( 0 . 22 µm ) two days after transfection and concentrated using the Amicon Ultra-15 filters ( 10 kDa , Millipore ) . The VLP binding assay was performed as previously described ( Soll et al . , 2010 ) using naïve DF-1 cells or a DF-1 clone expressing untagged hMCT1 . Fluorescent microscopy images of the cells were acquired using a DeltaVision deconvolution microscope ( GE Healthcare , Port Washington , NY ) . A Z-series of images ( capturing the entire thickness of the cell monolayer ) were flattened onto a single image and the number of fluorescent particles associated with each cell was counted for 20 distinct cells . To determine the number of fluorescent VLPs added to the cells , 80 µl of the VLP-containing supernatant was was layered onto 1 ml of 20% sucrose in PBS followed by centrifugation at 20 , 000 x g for 90 min at 4°C . Pelleted VLPs were resuspended in 300 µl of PBS and filtered ( 0 . 22 µm ) . Two-fold serial dilutions of the VLPs ( in PBS ) were layered ( 100 µl ) to poly-D-lysine coated glass coverslips and left overnight at 4°C . Fluorescent microscopy images of the VLPs were analyzed using the DeltaVision software ( GE Healthcare ) to determine the number of VLPs/ml for each condition . The number of fluorescent particles bound to cells was then calculated as a proportion of the number of fluorescent VLPs added to the cells . Cell lysates and pelleted virions ( 600 µl of supernatant pelleted through 20% sucrose in PBS ) were resuspended in SDS-PAGE loading buffer , with the addition of 0 . 5% β-mercaptoethanol , and resolved on NuPAGE Novex 4–12% Bis-Tris Mini Gels ( Invitrogen ) in MOPS running buffer . Proteins were blotted onto nitrocellulose membranes ( HyBond , GE-Healthcare ) in transfer buffer ( 25 mM Tris , 192 mM glycine ) . The blots were then blocked with Odyssey blocking buffer and probed with rat monoclonal anti-MLV capsid ( RRID: CVCL_9230 , ATCC , R187 ) , rabbit polyclonal anti-HA ( RRID: AB_217929 , Rockland Immunochemicals , Pottstown , PA , 600-401-384 ) , mouse monoclonal anti-HA ( RRID: AB_2314672 , Covance , Princeton , NJ , MMS-101P ) , mouse monoclonal anti-hMCT1 ( RRID: AB_10841766 , Santa Cruz Biotechnology Inc . , CA , sc-365501 ) , rabbit polyclonal anti-hHsp90 ( RRID: AB_2121191 , Santa Cruz Biotechnology Inc . , CA , sc-69703 ) or mouse monoclonal anti-Flag ( RRID: AB_262044 , Sigma-Aldrich , St . Louis , MO , F1804 ) . The bound primary antibodies were detected using a fluorescently labeled secondary antibody ( IRDye 800CW or 680W Goat Anti-Mouse or anti-Rabbit Secondary Antibody , LI-COR Biosciences , Lincoln , NE , RRIDs: AB_621842 , AB_621840 , AB_10796098 , AB_621841 ) . Fluorescent signals were detected using a LI-COR Odyssey scanner ( LI-COR Biosciences ) . Stable cell populations or single cells clones expressing a particular HA-tagged HERV-T envelope and/or Flag-tagged/untagged hMCT-1 were seeded one day prior to immunofluorescence assay . Cells were fixed with 4% PFA for 30 min followed by treatment with 10 mM glycine ( diluted in PBS ) for another 30 min . Cells were permeabilized with a PBS containing 0 . 1% Triton X-100% and 5% goat serum for 15 min . Cells were then washed twice with PBS before being incubated with rabbit anti-HA and mouse anti-Flag antibodies diluted in PBS containing 0 . 1% Tween-20% and 5% goat serum for 2 hr at room temperature . Cells were washed three times with PBS before being treated with goat anti-mouse and/or anti-rabbit secondary antibody ( Alexa Fluor 488 and/or 594 dye , Life technologies , CA , RRIDs: AB_2534091 , AB_2534088 , AB_2534095 , AB_2576217 ) diluted in PBS containing 0 . 1% Tween-20% and 5% goat serum for 1 hr at room temperature . DNA was stained with 50 μg/ml of DAPI ( diluted in PBS ) ( Invitrogen ) for 5 s . Cells were washed three more times with PBS and fluorescent microscopy images were analyzed using the a DeltaVision microscope ( GE Healthcare ) or the EVOS FL Cell Imaging System ( Thermo Fisher Scientific Inc . ) . Viruses were generated in 293T cells transfected with plasmids expressing HIV-1 gag-pol polyprotein ( pCRV1 ) , VSV glycoprotein and pCCIGW ( IRES-GFP ) plasmids and expressing the various HERV-T envelopes described above , or an unrelated protein as control ( SIVgab Nef ) using polyethylenimine . Viruses were harvested and filtered ( 0 . 22 µm ) two days after transfection and were used to transduce 2 × 105 naïve 293T cells or 1 × 105 DF-1 clones expressing HA-tagged hMCT1 . Transduced cells were expanded in 10 cm dishes until reaching confluency . Afterwards , serial dilutions of MLV expressing RFP ( pCNCR ) pseudotyped with ancHTenv were used to infect 1 × 104 transduced cells in 96 well plates . Two days post infection the number of GFP+ , RFP+ and double positive cells were counted by FACS using a CyFlow space cytometer ( Partec , Görlitz , Germany ) . The resulting data was analyzed with the FlowJo analysis software . The percentage of RFP positive cells was calculated after gating on the GFP positive population . 293T cells were transduced with a lentivirus vector ( CCIGW ) expressing various HERV-T envelopes or a control protein ( SIVgab Nef ) ( described above ) and then fixed and stained with DAPI ( described above ) . Multinucleated cells ( >5 nuclei ) and cells expressing the GFP reporter ( indicative of the expression of the corresponding HERV-T envelope ) were counted in three groups of ten microscopic fields and the number of multinucleated GFP+ cells as a percentage of the total number GFP+ cells was calculated for each condition .	Over millions of years , viruses and the animals that they infect have been locked in a battle for survival , where each has needed to evolve ways to counteract the effects of the other . While the evolution of ancient animals can be studied by looking at the fossilized remains of their extinct relatives , studying how ancient viruses have evolved is more difficult as they usually do not leave behind physical traces of their existence . However , a family of viruses called retroviruses is a notable exception to this rule . Retroviruses have a step in their life cycle in which their genetic material is integrated into the genome ( the name for an organism’s complete set of genetic material ) of the cell that they have infected . In rare cases , when that cell is a precursor of a sperm or egg cell , then the viral genes may then be passed on to the animal’s offspring , ultimately leaving genetic traces that can be studied in modern animals . This acts as a genetic ‘fossil record’ of extinct viruses . HERV-T was a retrovirus that spread among our primate ancestors for about 25 million years before its extinction roughly 11 million years ago . Blanco-Melo et al . have now analyzed the genetic remains left by HERV-T in the genomes of humans and related primates , and were able to use this information to recreate a protein that made up the outer envelope that surrounded the virus . Further experiments showed that this viral protein helped HERV-T to infect human cells by interacting with a protein called MCT1 on the cell surface . Blanco-Melo et al . also found a particular HERV-T gene that was unexpectedly well preserved in the human genome . The gene retained its ability to produce an envelope protein for about 13 to 19 million years . It is likely that ancient primates ‘hijacked’ the viral gene and used the protein it produced to remove the MCT1 protein from the surface of their own cells . Without MCT1 on the surface , HERV-T was unable to infect the cells . Thus , these findings present an example of how viruses themselves can provide the genetic material that animals use to combat them , potentially leading to their extinction .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "microbiology", "and", "infectious", "disease" ]	2017	Co-option of an endogenous retrovirus envelope for host defense in hominid ancestors
Huntington disease ( HD ) is a neurological disorder caused by polyglutamine expansions in mutated Huntingtin ( mHtt ) proteins , rendering them prone to form inclusion bodies ( IB ) . We report that in yeast , such IB formation is a factor-dependent process subjected to age-related decline . A genome-wide , high-content imaging approach , identified the E3 ubiquitin ligase , Ltn1 of the ribosome quality control complex ( RQC ) as a key factor required for IB formation , ubiquitination , and detoxification of model mHtt . The failure of ltn1∆ cells to manage mHtt was traced to another RQC component , Tae2 , and inappropriate control of heat shock transcription factor , Hsf1 , activity . Moreover , super-resolution microscopy revealed that mHtt toxicity in RQC-deficient cells was accompanied by multiple mHtt aggregates altering actin cytoskeletal structures and retarding endocytosis . The data demonstrates that spatial sequestration of mHtt into IBs is policed by the RQC-Hsf1 regulatory system and that such compartmentalization , rather than ubiquitination , is key to mHtt detoxification . The Huntington disease ( HD ) is predominantly inherited , with a single gene , HTT , encoding the Huntingtin protein , at its origin ( MacDonald , 1993 ) . Mutated and aggregation-prone poly-glutamine-expanded ( Poly ( Q ) ) Huntingtins ( mHtt ) are causing HD by toxic gain-of-functions and , possibly , dominant-negative mechanisms , which are typically manifested in aged individuals ( Ross and Tabrizi , 2011 ) . While the formation of mHtt inclusion bodies ( IBs ) correlates with toxicity and disease , such formation might , in effect , be a protective response to limit proteotoxicity ( Ross and Tabrizi , 2011; Arrasate et al . , 2004 ) : For example , IB formation predicts improved survival in neurons ( Arrasate et al . , 2004 ) and the IB-forming mHtt103QP protein ( Figure 1a; exon-1 with 97Q repeats ) are not , or only mildly , cytotoxic even when produced at high levels in young yeast cells ( Dehay and Bertolotti , 2006; Duennwald et al . , 2006 ) . In contrast , when the innate proline-rich region adjacent the poly ( Q ) stretch of exon-1 is removed , the protein , mHtt103Q , forms multiple small , highly cytotoxic aggregates/oligomers ( Figure 1a ) ( Dehay and Bertolotti , 2006; Duennwald et al . , 2006; Meriin et al . , 2002 ) . These aggregates are associated with the actin cytoskeleton ( Song et al . , 2014 ) and interfere with the cytosolic ubiquitin-proteasome-system ( UPS ) by sequestering the Hsp40 chaperone Sis1 ( Park et al . , 2013 ) . Chaperones , peptides , and prion-like proteins that either prevent/modify oligomer production ( Behrends et al . , 2006; Dehay and Bertolotti , 2006; Krobitsch and Lindquist , 2000; Muchowski et al . , 2000; Gokhale et al . , 2005 ) or convert small aggregates/oligomers into IBs ( Kayatekin et al . , 2014; Wolfe et al . , 2014 ) can suppress the toxicity of the proline-less exon-1 , suggesting that small aggregates and oligomers are likely culprits in mHtt103Q-derived toxicity ( Arrasate et al . , 2004; Miller et al . , 2011 ) . 10 . 7554/eLife . 11792 . 003Figure 1 . Screen approach and mHtt IB-forming mutants . ( a ) Aggregation of different mHtt reporters as indicated . ( b ) Morphology of mHtt103QP aggregates ( red ) in young and old ( 1 . 6 and 12 . 6 bud scars ( white ) , respectively ) cells . Scale=2 μm . Bar graph shows percentages of Class3 cells in young and old cells . Mean ± s . d . ( c ) Schematic description of the HCM-based screen . ( d ) Htt103QP aggregation 0 , 60 , 120 and 180 min after HTT103QP induction . ( e ) Representative pictures of Class 0 , 1 , 2 and 3 cells . ( f ) Mutants displaying increased% of Class 3 cells , grouped according to functions . Y-axis shows fold increase relative to wild type . ( g ) Physical ( red ) and genetic ( green ) interaction between Class 3 genes/proteins and their quantitative interaction ( thickness of grey lines ) with mHtt103QP as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 003 Ubiquitination is another process suggested to prevent mHtt toxicity in both mammals ( Steffan , 2004 ) and yeast ( Willingham et al . , 2003 ) . IBs of mHtt contain ubiquitin in mice ( Davies et al . , 1997 ) and the human ubiquitin-conjugating enzyme , hE2-25K , interacts with mHtt , which has been shown to be ubiquitinated in both humans and flies ( Kalchman et al . , 1996; Steffan , 2004 ) . However , an E3 ubiquitin ligase directly responsible for mHtt ubiquitin-tagging , IB formation , and detoxification has not been identified . We approached mHtt toxicity by a different route than recent mHtt103Q toxicity-suppression screens ( Kayatekin et al . , 2014; Mason et al . , 2013; Wolfe et al . , 2014 ) by asking if the non-toxic , IB-forming mHtt103QP carrying the innate proline-rich stretch of exon-1 , requires trans-acting factors to form IBs and if such factors convert mHtt103QP into non-toxic conformers . This approach was prompted also by our finding that the ability to form large and single mHtt103QP IBs was lost upon mother cell aging and the mHtt proteins accumulated instead in multiple , three or more smaller aggregates per cell , referred to as Class 3 cells ( Figure 1b; Class 1 cells contain one aggregate and Class 2 cells contain two aggregates ) . To identify trans-acting factors required for IB formation in an unbiased genome-wide manner , we used high content microscopy ( HCM ) and a galactose-regulated version of mHtt103QP , which we introduced into the ordered yeast deletion library ( SGA-V2 ) ( Tong , 2001 ) of S . cerevisiae ( Figure 1c ) . Upon galactose-induction , mHtt103QP formed aggregates in about 50% of the cells within 180 min ( Figure 1d ) and 70% of these cells contain one large IB . HCM was used to identify mutants that formed multiple aggregates/oligomers rather than a big IB ( Class 3 mutants; Figure 1e ) , which revealed that IB formation requires proteasome/chaperone and ubiquitination functions , Golgi-vesicle trafficking , mRNA transport/metabolism , and cell cycle control ( Figure 1f&g , see Supplementary file 1 for a list of confirmed mutants ) . Among these factors , Ltn1 and Rqc1 are especially interesting as they are both partners of the ribosome quality control complex ( RQC ) ( Brandman et al . , 2012 ) and Ltn1 is the yeast homologue of the E3 RING ubiquitin ligase Listerin of mammalian cells ( Bengtson and Joazeiro , 2010 ) , which reduced activity causes premature neurodegeneration in mice ( Chu et al . , 2009 ) . Complementation analysis revealed that the ubiquitin E3 ligase activity of Ltn1 was required for both mHtt103QP IB formation ( Figure 2a ) and ubiquitination ( Figure 2b ) . It’s been reported that the absence of Ltn1 , but not Rqc1 , results in the failure to tag non-stop protein with ubiquitin ( Brandman et al . , 2012 ) . Contrasting such data on non-stop proteins , both Ltn1 and Rqc1-deficieny resulted in a failure of cells to tag also full-length mHtt103QP properly with ubiquitin ( Figure 2b , Figure 2—figure supplement 1 ) and to form IBs , even though the effect of rqc1∆ was markedly smaller than ltn1∆ on IB formation ( Figure 2a ) . Moreover , both soluble and aggregated mHtt103QP was stable in the absence and presence of Ltn1 ( Figure 2c , Figure 2—figure supplement 2 ) , and the levels of soluble and aggregated mHtt103QP was somewhat lower in ltn1∆ cells ( Figure 2—figure supplements 2 & 4 ) . These data suggest that Ltn1 is involved in mHTT103QP sequestration into IBs rather than its decay . 10 . 7554/eLife . 11792 . 004Figure 2 . Role of RQC in mHtt103QP IB formation ubiquitination and toxicity . ( a , d ) Htt103QP aggregate numbers ( % Class 1 , 2&3 cells; see Figure 1 ) in mutants as indicated . W1542E encodes a ubiquitin-ligase-defect Ltn1 protein . HSF1-R206S encodes a hyper-active Hsf1 . The hsf1-848 is a conditional ts mutant while HSF1ΔCAD lacks the c-terminal trans-activating domain . Scale=2 μm . Bar graphs show % of Class 1 , 2 and 3 cells in each strain . Mean ± s . d . ( b ) Ubiquitination of mHtt103QP in strains from ‘a’ . ( c ) Htt103QP stability in WT and ltn1Δ cells after a block in protein synthesis . Mean ± s . d . e-g . Fitness ( see Materials and methods ) of strains carrying pYES2-mHtt103QP-GFP compared to pYES2-GFP . Results from Galactose ( mHtt induced ) and Glucose ( mHtt repressed ) are shown . Ratios were calculated from the mean of three repeats ( error bars are 95% confidence intervals ) for WT , RQC , and rnq1∆ mutants ( e ) HSF1-R206S ( f ) and hsf1-848 ( g ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 00410 . 7554/eLife . 11792 . 005Figure 2—figure supplement 1 . Western blot of His-Ub pull-down mHtt103QP in RQC mutants . mHtt103QP-GFP ubiquitinated by His-tagged ubiquitin was pulled-down by Ni-beads and detected by GFP antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 00510 . 7554/eLife . 11792 . 006Figure 2—figure supplement 2 . FRAP assay of mHtt103QP aggregate in Wt and RQC mutants . ( a ) Representative images of mHtt103QP-GFP aggregate before and after laser bleach . ( b ) Relative fluorescence of the bleached region . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 00610 . 7554/eLife . 11792 . 007Figure 2—figure supplement 3 . Ltn1-GFP co-localize with mHtt103QP-mRFP . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 00710 . 7554/eLife . 11792 . 008Figure 2—figure supplement 4 . mHtt levels chase after cycloheximide treatment . ( a ) Representative Western blots of soluble and aggregated mHtt103QP from Wt and ltn1Δ strains . ( b ) Quantitafication of three repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 00810 . 7554/eLife . 11792 . 009Figure 2—figure supplement 5 . mHtt103QP aggregate in ltn1Δtae2Δ is also co-localized with dense actin structures . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 009 Ltn1- , and to a lesser extent , Rqc1-deficieny results in hyper-activation of the heat shock transcription factor Hsf1 through the RQC component Tae2 and such activation can thus be suppressed by removing the TAE2 gene ( Brandman et al . , 2012 ) . We found that deleting TAE2 in ltn1∆ or rqc1∆ cells restored IB formation ( Figure 2a , Figure 2—figure supplements 2 & 5 ) but did not restore ubiquitination ( Figure 2b ) , demonstrating that ubiquitination is not an absolute requirement for the formation of mHtt103QP IBs . Moreover , overproducing a hyperactive Hsf1 ( HSF1-R206S [Hou et al . , 2013] ) alone was sufficient to reduce IB formation , as was reducing Hsf1 activity using the hsf1-848 ( ts ) allele ( Figure 2d ) . demonstrating that maintaining a proper , intermediate , range of Hsf1 activity is required to efficiently sequester mHtt103QP into IBs . In support of this notion , a deletion in the C-terminal trans-activation domain of Hsf1 resulted in defects in IB formation that could not be further abrogated by an ltn1 deletion ( Figure 2d ) . The mHtt103QP protein displays no obvious toxicity in yeast ( Dehay and Bertolotti , 2006; Duennwald et al . , 2006 ) but we found that it became detrimental in the absence of Ltn1 , and to a somewhat lesser extent , Rqc1 ( Figure 2e ) , supporting the idea that IB formation protects the cell against Huntingtin toxicity . Consistently , a tae2∆ mutation completely suppressed the toxicity of mHtt103QP in the ltn1∆ cells ( Figure 2e ) . Since the TAE2 deletion did not restore mHtt103QP ubiquitination , we conclude that IB formation is more important than ubiquitination for the detoxification of mHtt103QP , at least in the yeast model system . Contrasting the LTN1 data , the absence of TAE2 failed to fully suppress toxicity in rqc1∆ cells indicating that the roles of Ltn1 and Rqc1 in RQC are overlapping ( Brandman et al . , 2012 ) but not identical . Consistent with small mHtt103QP aggregates/conformers being toxic , both overactive and diminished Hsf1 activity rendered mHtt103QP toxic ( Figure 2f&g ) . Since the proline-less , intrinsically noxious , mHtt103Q protein requires the presence of the prion-forming protein Rnq1 to display cytotoxicity in yeast ( Meriin et al . , 2002 ) , we tested whether the toxicity of mHtt103QP in Ltn1-deficient also required the presence of Rnq1 and found that this was not the case ( Figure 2e ) . The small cytotoxic mHtt103Q aggregates have been shown to associate with the actin cytoskeleton ( Song et al . , 2014 ) , and we , therefore , investigated if mHtt103QP in wild type and ltn1∆ cells likewise interacted with and affected actin cytoskeletal structures . First , using co-staining with the misfolded protein Ubc9ts-mCherry , we confirmed that the mHtt103QP proteins of wild type cells were deposited in IBs adjacent to the Ubc9ts-associated insoluble-protein-deposit , IPOD ( Kaganovich et al . , 2008 ) ( Figure 3a ) . Super resolution , three-dimensional structured illumination microscopy ( SIM ) revealed that these mHtt103QP IBs were associated with dense actin cytoskeletal structures ( Figure 3b , Video 1 ) . Moreover , the actin cytoskeleton appears to harness latent mHtt103QP toxicity as a screen for conditional ts mutations causing synthetic sickness/lethality with mHtt103QP ( Figure 3d&e ) revealed that cells carrying ts mutations in genes encoding actin itself ( act1 ) , profiling ( pfy1 ) involved in actin polymerization , cofilin ( cof1 ) regulating assembly/disassembly of actin filaments , Arp3 of the actin-nucleation center , Las17 , an activator of Arp2/3 and actin assembly factors , and Mss4 , a phosphatidylinositol-4-phosphate 5-kinase involved in actin cytoskeleton organization , were drastically sensitized to mHtt103QP ( Figure 3d&e , also see Supplementary file 2 for a list of alleles ) . The multiple mHtt103QP aggregates formed in ltn1∆ cells also co-localized with actin cytoskeletal structures ( Figure 3c , Video 2 ) , akin to those of the toxic mHtt103Q aggregates reported previously ( Song et al . , 2014 ) . Actin-mHtt103QP-associated structures were more abundant in Ltn1-deficient cells than in wild type cells whereas the number of aggregate-free forms of actin structures , including actin patches , was reduced ( Figure 3f ) . Because the actin cytoskeleton is required for proper endocytosis , we tested the effect of mHtt103QP and an ltn1 deletion on the rate of endocytic internalization of the dye FM4-64 , and found that Htt103QP retarded endocytosis and that such retardation was more pronounced in cells lacking Ltn1 ( Figure 3g; Figure 3—figure supplement 1 ) . In contrast , Ltn1 deficiency did not by itself cause actin cytoskeleton defects or endocytosis retardation ( Figure 3g , Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 11792 . 010Figure 3 . Role of actin in Htt103QP detoxification . ( a ) Co-localization of mHtt103QP IBs and UBC9ts IPODs . ( b , c ) Actin structures ( Red; phalloidin ) and mHtt103QP ( Green; GFP ) aggregates in WT and ltn1Δ analyzed by 3D-SIM . Scale=1 μm . ( d ) Essential ts-alleles increasing toxicity of mHtt103QP , grouped according to biological processes . ( e ) Functional enrichment analysis of mHtt103QP-sensitive ts mutants . ( f ) Number of actin-associated aggregates and aggregate-free actin structures in WT and ltn1Δ cells . Mean ± s . d . g . Endocytotic activity in WT and ltn1Δ cells analyzed by FM4-64FX uptake to vacuoles . Mean ± s . d . ( h ) A model of the regulation of mHtt103QP IB formation and toxicity by RQC components and Hsf1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 01010 . 7554/eLife . 11792 . 011Figure 3—figure supplement 1 . FM4-64FX stained cells . Images of FM4-64FX stained cells corresponding to Figure 3g . Representative cells carrying pYES2-GFP ( 'Vector' ) or pYES2-mHtt103QP-GFP ( '103QP' ) were shown . ltn1Δ . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 01110 . 7554/eLife . 11792 . 012Figure 3—figure supplement 2 . Actin integrity of Wt and a . Actin staining of Wt and ltn1Δ cells , b . Quantification of actin depolarization of Wt and ltn1Δ cells , see Materials and methods for details . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 01210 . 7554/eLife . 11792 . 013Video 1 . 3D structures of mHtt103QP aggregate and actin in WT . mHtt103QP aggregates ( green ) and actin ( red ) structures of a WT cell shown in Figure 3b . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 01310 . 7554/eLife . 11792 . 014Video 2 . 3D structures of mHtt103QP aggregate and actin in ltn1Δ . mHtt103QP aggregates ( green ) and actin ( red ) structures of a ltn1Δ cell shown in Figure 3c . DOI: http://dx . doi . org/10 . 7554/eLife . 11792 . 014 The conserved Listerin ( Ltn1 ) E3 ligase is a key factor involved in targeting protein products derived from defective mRNA or aborted translation for degradation by the 26S proteasome ( Bengtson and Joazeiro , 2010; Brandman et al . , 2012 ) . Upon translation stalling , ribosome recycling factors dissociate 80S ribosome-nascent chain complexes to 60S ribosome-nascent chain-tRNA complexes , which are recognized by Ltn1 and Tae2 ( Shen et al . , 2015; Shao et al . , 2015; Shao et al . , 2013 ) . Both nascent chains and , for example , K12- and R12-arrested polypeptides are substrates for Ltn1-dependent ubiquitin tagging , which signal their destruction by the 26S proteasome ( Bengtson and Joazeiro , 2010; Brandman et al . , 2012; Preissler et al . , 2015 ) . Herein , we report on another pivotal role of Ltn1 in protein quality control – detoxification of mutant Huntingtin through a Tae2/Hsf1-dependent sequestration of mHtt103QP into actin-associated inclusions ( Figure 3h ) . As depicted in Figure 3h , the effect of Ltn1 on mHtt103QP aggregation appears to act through Tae2 , which in turn is known to negatively control Hfs1 activity ( Brandman et al . , 2012 ) . Thus , the presence of Tae2 is known to cause hyperactivation of Hsf1 when LTN1 is deleted ( Brandman et al . , 2012 ) , which could be enough to inhibit IB formation . On the other hand , mutations reducing Hsf1 activity also inhibited IB formation suggesting that maintaining a proper , intermediate , range of Hsf1 activity is required to efficiently sequester mHtt103QP into IBs ( Figure 3h ) . In worms , elevated production of small heat shock proteins through Hsf1 activity has been shown to delay the onset of polyglutamine-expansion protein aggregation ( Hsu , 2003 ) and reducing hsf-1 activity accelerates aging ( Hsu , 2003; Morley and Morimoto , 2004 ) . Reciprocally , hsf-1 overexpression extends worm lifespan ( Hsu , 2003 ) . ( Baird et al . , 2014 ) . The data presented here , however , demonstrate that both Hsf1 elevation and Hsf1 deficiency in cells expressing the Huntingtin disease protein is detrimental ( Figure 3h ) , suggesting , again , that a fine balance of Hsf1 activity has to be maintained to assuage proteotoxicity . This notion might explain why alterations in Hsf1 levels in mammalian cells have been shown to either inhibit mHtt IB formation ( Fujimoto et al . , 2005 ) or lower the concentration threshold at which HTT forms IB ( Bersuker et al . , 2013 ) . These results raise the question of whether age-dependent penetrance of HD could be due to a reduced Hsf1 activity in aging tissues or a malignant hyperactivation of Hsf1 . The latter scenario could be the result of an age-dependent increase in translational processivity errors , which could titrate the RQC complex eliciting a Tae2-dependent activation of Hsf1 ( Figure 3h ) , possibly through Tae2-directed tagging of incomplete translation products with carboxyl-terminal Ala and Thr extensions . ( Shen et al . , 2015 ) . The exact mechanism behind Hsf1-dependent modulation of mHtt IB formation might be complex in that Hsf1 targets other genes than heat shock genes . It has been shown in worms that over-expression of hsf-1 , with or without the C-terminal trans-activation domain , elevates the levels of pat-10 , a troponin-like protein , that increase actin cytoskeleton integrity leading to lifespan extension and resistance to proteotoxic stress ( Baird et al . , 2014 ) . Thus , it is possible that Hsf1 may regulate mHtt IB formation/toxicity in the yeast model system through the regulation of actin cytoskeleton dynamics since we found that mHtt103QP is associated with dense actin structures and that genes involved in actin dynamics are required to harness the latent toxicity of mHtt103QP . In addition , our data cannot rule out the possibility that the expression of mHtt in general raises proteostasis stress in the cell leading to Hsf1 activation and that such activation is epistatically affecting the effect of Ltn1-deficieny . Plasmids and yeast strains used in each assay and figure were specified in Supplementary file 3A and B . Yeast cells were grown at 30°C if not specified , in YPD ( BY4741 background ) , YPAD ( W303 background ) or corresponding synthetic drop-out media with antibiotics . For all galactose induction experiments , yeast cells were pre-cultured , diluted , and re-grown in media with 2% raffinose until mid-log phase ( OD600=0 . 5 ) . 2% galactose was then added to induce expression for desired time . For temperature sensitive strains ( except Ubc9ts , see below ) , cells were pre-cultured at 22°C and switched to 30°C during experiments . HSF1 and HSF1ΔCAD in the W303-1A background ( as described in [Eastmond and Nelson , 2006] ) was a gift from Dr . H Nelson ( University of Pennsylvania , USA ) . The mHtt103QP plasmid pYES2-103QP-GFP ( as described in [Meriin et al . , 2007] ) was a gift from Dr . M Sherman ( Boston University , USA ) . Plasmid pYES2-GFP ( as described in [Preveral et al . , 2006] ) was a gift from Dr . C Forestier ( CEA , France ) . Plasmids pGAD-HA-Ltn1 and pGAD-HA-Ltn1-1542E ( as described in [Bengtson and Joazeiro , 2010] ) were gifts from Dr . CJoazeiro ( The Scripps Research Institute , USA ) . Plasmid pRS416-TEF1-Hsf1M ( as described in [Hou et al . , 2013] ) was a gift from Dr . J Nielsen ( Chalmers University of Technology , Sweden ) . Plasmid pADH-His-Ub ( Lu et al . , 2014 ) was a gift from Dr . S Jentsch ( Max Planck Insititute of Biochemistry , Germany ) . The pYES2-mHtt103QP-GFP plasmid was transformed to SGA-V2 single gene knock-out collection by a robotic SGA procedure to generate the strain collection SGA-V2-pYES2-mHtt103QP-GFP ( S2Y103QPG ) for HCM-based screen ( Tong , 2001; Tong , 2004 ) . A control plasmid pYES2-GFP was also transformed to SGA-V2 collection to build SGA-V2-pYES2-GFP ( S2YG ) collection as negative controls for toxicity assays . ltn1Δ::natMX4 in BY4741 , W303 and W303 HSF1ΔCAD , SGA rnq1Δ backgrounds and tae2Δ::natMX4 in SGA rqc1Δ and SGA ltn1Δ backgrounds were all generated by PCR-mediated gene deletion . The coding sequence of URA3 in pYES2-mHtt103QP-GFP and pYES2-GFP were replaced by hphMX4 cassette via PCR-mediated gene deletion , to generate pY2H-mHtt103QP-GFP and pY2H-GFP plasmids to make them compatible with URA3 plasmids . The template used to amplify hphMX4 is plasmid pAG32 ( Goldstein and McCusker , 1999 ) . Isolation of old cells was carried out via the biotin-streptavidin magnetic beads binding system as previously described ( Sinclair and Guarente , 1997 ) . Old cells ( 'Old' in Figure 1b ) were labeled with EZ-Link NHS-Biotin ( Thermo Fisher Scientific , Waltham , MA ) , first aged in glucose media for two overnights and then in raffinose media for one overnight before harvesting; young cells ( 'Young' in Figure 1b ) were the progenies of the old cells generated in the last overnight culturing in raffinose media . Both young and old cells were induced for mHtt103QP-GFP expression for 3 hr and then fixed . Mean ages of samples were assessed by counting bud scars stained by Calcofluor white ( Sigma-Aldrich , St . Louis , MO ) . Three parallel repeats were performed . Each strain from the S2Y103QPG collection was pre-cultured , induced for mHtt103QP-GFP expression as described earlier and fixed with 3 . 7% formaldehyde at room temperature for 30 min in 96-well plates . For image capturing , appropriate amount of fixed cells were transferred to new 96-well plates and imaged with the ImageXpress MICRO ( Molecular Devices , Sunnyvale , CA ) , an automated cellular imaging system . Customized sub-program of the software MetaXpress ( Molecular Devices ) was applied on the obtained images for quantification . All mutants that showed statistically significant increase larger than three times the variance of the wild type were restreaked and re-tested individually and analyzed manually to confirm the phenotypic differences observed in the screen . At least 300 cells were counted in the manual confirmation . Cell images ( except for 3D-SIM images in Figure 3b&c ) were obtained via Zeiss Axio Observer . Z1 inverted microscope and Zen Pro 2012 software ( Carl Zeiss AG , Germany ) . Filter sets used are: 38 HE GFP for mHtt103QP-GFP , 43 HE DsRed for Ubc9ts-MCherry , 45 Texas Red for FM4-64FX and 49 DAPI for DAPI and Calcofluor white . Images in Figure 1b and Figure 3a were deconvolved by ImageJ software and plugin 'Iterative deconvolve 3D' , maximum number of iterations set to 15 and 10 respectively . IB morphology tests were performed three times for each strain in Figure 1b , Figure 2a , d; 100 cells with aggregates were analyzed and quantified for each repeat . Whole cell protein extracts were obtained via mild alkali treatment and IPs by anti-FLAG M2 affinity gels ( Sigma ) were carried out following previously published protocols ( Bengtson and Joazeiro , 2010 ) . mHtt103QP-GFP was expressed for 3 hr in all samples . Western blotting was done as described before ( Molin et al . , 2011 ) using an XCell SureLock MiniCell ( LifeTechnologies ) and Immobilon-FL PVDF membranes ( Millipore , Billerica , MA ) . Ubiquitination signals were detected by a rabbit polyclonal anti-ubiquitin antibody ( ab19247; AbCam , United Kingdom ) . The mHtt103QP-GFP was detected by a chicken polyclonal anti-GFP antibody ( ab13970; AbCam ) . The stability of in vivo mHtt103QP-GFP by FACS was determined by the change of GFP fluorescent signal strength after inhibition of protein synthesis by cycloheximide , as described previously ( Song et al . , 2014 ) . Soluble protein and protein aggregates were separated by ultracentrifugation as described in ( Song et al . , 2014 ) and then quantified by Western blotting . mHtt protein levels were standardized to total protein levels determined by Coomassie Brilliant Blue staining of the membrane . Doubling time was determined by the Bioscreen Assays as described ( Warringer et al . , 2003 ) , in media with either 2% glucose or 2% galactose after overnight pre-culturing in media containing 2% raffinose as the only carbon source . Three parallel replicates were run for each strain . Actin structures were stained by Alexa568-phalloidin ( Thermo Fisher Scientific ) as described ( Liu et al . , 2010 ) . For quantifications in Figure 3f , Z-stack serial images were analyzed . To avoid possible bias caused by different distributions of cells at different cell cycle stages , only mother cells in budding events with undivided nucleus ( determined by DAPI staining ) were counted ( Anderson et al . , 1998 ) . Both mHtt103QP-GFP and Ubc9ts-mCherry were expressed for 3 hr at 28°C . The cells were then incubated at 37°C for 30 min to trigger Ubc9ts aggregate formation . Cells were fixed and washed immediately after the 37°C treatment . 3D-SIM microscopy images were obtained as previously reported ( Song et al . , 2014 ) . SGA analysis of the ts-allele collection was performed and scored as previously described ( Wagih et al . , 2013; Costanzo et al . , 2010; Li et al . , 2011 ) . The cut-off for the screen was -0 . 5 in score from the screen . The functional enrichment analysis of Htt103QP essential synthetic sick interactors was based on the result from Gene Ontology Term Finder ( Boyle et al . , 2004 ) using the SGA ts-V5 array ( 787 ts alleles , covering 497 essential genes ) as the background list . Cytoscape 3 . 2 . 0 ( Saito et al . , 2012 ) was used for interaction network analysis of hits with increased class 3 aggregates . The physical interactions between the hits were obtained from BioGRID interaction database ( Breitkreutz et al . , 2008 ) using GeneMANIA plugin ( Warde-Farley et al . , 2010 ) Actin depolarization was quantified as described in ( Anderson et al . , 1998 ) . Endocytosis was assessed by tracking FM4-64FX ( Thermo Fisher Scientific ) internalization in live cells as described ( Baggett et al . , 2003 ) with minor modifications . Yeast cells were strained on ice for 30 min with FM4-64FX after 3 hr expression of mHtt103QP-GFP . Cells were then incubated in YPD at 30°C in dark . Z-stack images of samples taken after 0 , 15 , 30 , 45 and 60 min incubation at 30°C were captured and analyzed . FRAP of mHtt103-QP aggregates was carried out on LSM 700 Axio Observer . Z1 ( Carl Zeiss ) . Images were captured every second for 90 s after photobleaching . Fluorescent intensities of the bleached region were quantified via ImageJ . His-Ub pull-down assay was carried out as described in ( Tansey , 2006 ) with minor modifications . His-tagged Ub was expressed from pADH-His-Ub and pulled down via Dynabeads His-tag ( Thermo Fisher Scientific ) . For bar graphs in Figure 1b , 2a , 2d , 3f , 3g , data shown are mean of three replicates ± s . d . , unpaired two-tailed t-test was used to compare mean values . Statistical significance was indicated as p<0 . 05; p<0 . 01; ** P<0 . 001 . For the bar graph in Figure 2 e-g , data shown is the ratio of means ± 95% confidence interval . The confidence intervals were calculated based on Fieller’s theorem ( Fieller , 1940 ) by an online-calculator http://www . graphpad . com/quickcalcs/ErrorProp1 . cfm ( GraphPad Software , La Jolla , CA ) .	Huntington’s disease is a neurological disease that is caused by mutations in the gene that encodes a protein called Htt . Individuals with this mutation gradually lose neurons as they age , resulting in declines in muscle coordination and mental abilities . The mutant Htt proteins tend to form clumps inside cells , but it is not clear if these clumps are the cause of the disease symptoms or whether they have a protective effect . Yang et al . used yeast as a model to investigate whether the mutant Htt proteins need other molecules to allow them to form clumps . The experiments identified several new molecules that are required for mutated Htt to form clumps . Some of these are components of a system called the Ribosome Quality Control ( RQC ) complex , which monitors newly made proteins and labels abnormal ones for destruction . However , Yang et al . ’s findings suggest that the RQC complex regulates the formation of Htt clumps through a different pathway involving a protein called heat shock factor 1 . In this case , cells would need to fine-tune heat shock factor 1 activity to make mutant Htt proteins clump together to protect cells from damage . Future experiments should expand Yang et al . ’s findings to animal models of Huntington’s disease and identify which other molecules contribute to the formation of Htt clumps . One challenge will be to find out why older neurons fail to form clumps of Htt proteins , and whether this can be overcome by drugs that boost the activity of the molecules that Yang et al . identified .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "cell", "biology" ]	2016	Spatial sequestration and detoxification of Huntingtin by the ribosome quality control complex
Double-stranded DNA ( dsDNA ) binding and cleavage by Cas9 is a hallmark of type II CRISPR-Cas bacterial adaptive immunity . All known Cas9 enzymes are thought to recognize DNA exclusively as a natural substrate , providing protection against DNA phage and plasmids . Here , we show that Cas9 enzymes from both subtypes II-A and II-C can recognize and cleave single-stranded RNA ( ssRNA ) by an RNA-guided mechanism that is independent of a protospacer-adjacent motif ( PAM ) sequence in the target RNA . RNA-guided RNA cleavage is programmable and site-specific , and we find that this activity can be exploited to reduce infection by single-stranded RNA phage in vivo . We also demonstrate that Cas9 can direct PAM-independent repression of gene expression in bacteria . These results indicate that a subset of Cas9 enzymes have the ability to act on both DNA and RNA target sequences , and suggest the potential for use in programmable RNA targeting applications . Prokaryotic clustered regularly interspaced short palindromic repeat ( CRISPR ) -CRISPR-associated ( Cas ) systems provide immunity against plasmids and bacteriophage by using foreign DNA stored as CRISPR spacer sequences together with Cas nucleases to stop infection ( Wright et al . , 2016; Mohanraju et al . , 2016 ) . One such nuclease , Cas9 of the type II systems , employs a CRISPR RNA ( crRNA ) and a trans-activating crRNA ( tracrRNA ) to target spacer-complementary regions ( protospacers ) on the foreign genetic element to guide double-stranded DNA cleavage ( Jinek et al . , 2012 ) . A protospacer adjacent motif ( PAM ) must also be present for the Cas9-RNA complex to bind and cleave DNA ( Jinek et al . , 2012; Gasiunas et al . , 2012; Anders et al . , 2014; Szczelkun et al . , 2014 ) . Combining the crRNA and tracrRNA into a chimeric , single-guide RNA ( sgRNA ) simplified the system for widespread adoption as a versatile genome editing technology ( Jinek et al . , 2012 ) . To date , both genetic and biochemical data support the conclusion that in vivo , Cas9 is exclusively a DNA-targeting enzyme . Nonetheless , multiple studies have harnessed Cas9 for RNA targeting under specific circumstances . For example , the S . pyogenes Cas9 ( SpyCas9 ) can be supplied with a short DNA oligo containing the PAM sequence ( a PAMmer ) to induce single-stranded RNA ( ssRNA ) binding and cutting ( O'Connell et al . , 2014; Nelles et al . , 2016 ) . More recently , it was demonstrated that SpyCas9 could be used to target repetitive RNAs and repress translation in certain mRNAs in the absence of a PAMmer ( Liu et al . , 2016; Batra et al . , 2017 ) . A different Cas9 homolog from Francisella novicida ( FnoCas9 ) has been implicated in degradation of a specific mRNA but through a mechanism independent of RNA-based cleavage ( Sampson et al . , 2013 ) . Together with evidence that some Cas9 homologs can target single-stranded DNA substrates under some conditions ( Ma et al . , 2015; Zhang et al . , 2015 ) , these studies raised the possibility that certain Cas9 enzymes might have intrinsic RNA-guided RNA cleavage activity . To determine whether evolutionarily divergent Cas9 homologs have a native capacity for programmable RNA targeting , we compared biochemical behavior of enzymes from the three major Cas9 subtypes . This analysis revealed that certain type II-A and II-C Cas9s can bind and cleave single-stranded RNA sequences with no requirement for a PAM or PAMmer . Furthermore , we found that this activity can inhibit gene expression and confer moderate protection against infection by ssRNA phage through a mechanism reminiscent of RNA-guided DNA targeting . These results establish the utility of Cas9 for facile RNA-guided RNA targeting and suggest that this activity may have biological relevance in bacteria . To assess whether divergent Cas9 enzymes can catalyze binding to and cleavage of RNA substrates by a mechanism distinct from that of double-stranded DNA cleavage , we tested homologs from the three major subtypes of Cas9 proteins for their ability to cleave single-stranded RNA in vitro ( Figure 1A , B; Figure 1—figure supplement 1A–C ) . When programmed with a cognate sgRNA , S . aureus Cas9 ( SauCas9 ) and C . jejuni Cas9 ( CjeCas9 ) direct cleavage of RNA in the absence of a PAMmer ( Figure 1; Figure 1—figure supplement 1 ) . No RNA cleavage was detected using SpyCas9 , which requires a PAMmer for efficient RNA cleavage in vitro ( O'Connell et al . , 2014 ) , or using F . novicida Cas9 ( FnoCas9 ) . While the cleavage efficiencies for both SauCas9 and CjeCas9 are indistinguishable ( Figure 1—figure supplement 1D ) , we focused on the activity of SauCas9 due to the abundance of mechanistic and structural data for this enzyme ( Nishimasu et al . , 2015; Friedland et al . , 2015; Ran et al . , 2015; Kleinstiver et al . , 2015 ) . RNA cleavage activity and products were similar to those of canonical Cas9-mediated DNA cleavage activity in vitro . RNA targeting by SauCas9 requires the presence of a guide RNA and a catalytically-active protein , as both apo protein lacking the guide RNA and a catalytically inactive mutant ( D10A and N580A ) do not cleave RNA ( Figure 1—figure supplement 2A ) . Furthermore , addition of EDTA to chelate divalent metal ions abolished RNA cleavage , verifying that divalent metal ions are necessary for catalysis . As with DNA substrates ( Sternberg et al . , 2014 ) , incubation of SauCas9 with an excess of RNA target demonstrated that cleavage is single-turnover ( Figure 1—figure supplement 2B , C ) . Hydrolysis mapping of the cleavage product revealed that the predominant RNA cleavage site is shifted by one nucleotide compared to the site of DNA cleavage ( Garneau et al . , 2010; Jinek et al . , 2012; Gasiunas et al . , 2012 ) ( Figure 1—figure supplement 2D , E ) . The shift is consistent with that observed for PAM-dependent SpyCas9 RNA-cleavage ( O'Connell et al . , 2014 ) and is likely due to the more compact geometry of an RNA-RNA helix relative to an RNA-DNA hybrid helix ( Wang et al . , 1982 ) . SauCas9 targets ssRNA in the absence of a PAMmer , a contrast to SpyCas9 targeting of ssRNA ( O'Connell et al . , 2014 ) . Testing SauCas9 in vitro ssRNA cleavage in the presence of a PAMmer ( 30x molar excess over ssRNA target ) revealed that turn-over was two-fold slower than the reaction with only target ssRNA ( Figure 1C , Figure 1—figure supplement 3C ) . SauCas9 ssRNA cleavage conducted in the presence of a non-complementary , control DNA oligo did not yield a similar reduction in cleavage rate ( Figure 1—figure supplement 3C ) , indicating that the complementary PAMmer impairs RNA cleavage activity . Consistent with cleavage being guide-dependent , single-stranded RNA that is not complementary to the sgRNA is not cleaved ( Figure 1 and Figure 1—figure supplement 3 ) . Double-stranded RNA ( dsRNA ) is also not a substrate for SauCas9 . Given that Cas9 proteins are active with different length guide RNA segments ( ~20–24 nt ) ( Chylinski et al . , 2013; Ran et al . , 2015; Friedland et al . , 2015; Kim et al . , 2017 ) , we tested whether longer guide segments might enhance ssRNA targeting activity . Increasing the length of the targeting region of the guide up to 23 nt results in tighter binding and more efficient cleavage ( Figure 1—figure supplement 4 ) , mirroring the preference for longer guides for DNA cleavage ( Ran et al . , 2015; Friedland et al . , 2015 ) . Extending the guide strand complementarity to the target beyond 23 nt did not increase RNA target binding or cleavage efficiency , indicating that 23 nt is the optimal length for in vitro binding and targeting applications . The apparent dissociation constant ( Kd , app ) of the SauCas9-sgRNA complex ( 23 nt targeting region ) for the ssRNA target is 1 . 8 ± 0 . 09 nM ( Figure 1—figure supplement 4D ) , which is ~5 x weaker than the 0 . 34 ± 0 . 03 nM binding affinity measured for a dsDNA substrate of the same sequence . We noted that SauCas9-catalyzed ssRNA cleavage is limited to ~30% fraction cleaved ( Figure 1—figure supplement 3 ) , compared to >80% fraction cleaved for ssDNA and dsDNA targets . Greater thermodynamic stability of RNA secondary structures , relative to those in ssDNA ( Bercy and Bockelmann , 2015 ) , might occlude SauCas9-sgRNA binding to an ssRNA target sequence , a possibility that we tested using a panel of partially duplexed RNA substrates ( Figure 2 ) . Previously , introduction of a short segment of mismatched base pairs to mimic partially unwound dsDNA substrates was shown to enhance the ability of type II-C Cas9s ( including CjeCas9 ) to unwind and cleave dsDNA ( Ma et al . , 2015 ) . Here , we found that duplex-RNA substrates containing a 2- or 6-base pair mismatched segment located near the 5’ or 3’ end of the 23 nt guide RNA region of the sgRNA could not be cleaved ( Figure 2A–C , substrates 5 , 6 , 10 , and 11 ) . However , when the unpaired region was increased to 12-base pairs , SauCas9 was able to cleave the target strand . There was a slight cleavage preference for RNA substrates in which the 12-base pair mismatched segment is located near the 5´ end of the guide sequence of the sgRNA ( Figure 2A–C , substrates 7 and 12 ) . Interestingly , the 23-base pair mismatched segment RNA substrates ( ‘Bubble’ substrates 8 and 9 ) are targeted more efficiently than their ssRNA counterparts ( substrates 1 and 2 ) ( Figure 2C ) . We measured the binding affinity of all substrates and found that both the 23-base pair mismatched segment RNA and ssRNA substrates are bound with similar affinity ( Figure 2D ) . Furthermore , the apparent difference in cleavage efficiency was not due to the presence of a double-stranded PAM sequence , as mutating the PAM region does not impair cleavage ( Figure 2C , compare substrates 8 and 9 ) . We hypothesize that RNA containing a mismatched segment presents a more accessible substrate to the Cas9-sgRNA complex due to stable annealing between the ends of the non-target and target strands , whereas the ssRNA substrate alone has ends that are predicted to stabilize a conformation that is partially structured and therefore inaccessible ( Figure 2—figure supplement 1A ) . An alternative hypothesis to explain the limited cleavage of ssRNA substrates is that SauCas9 enzyme inactivation occurs over the course of the reaction , even with SauCas9 protein-sgRNA ( ribonucleoprotein , RNP ) present in 10-fold excess relative to the ssRNA substrate . To test this , we spiked reactions with fresh SauCas9 protein alone or SauCas9 RNP after reactions reached equilibrium; however , we did not observe an increase in the amount of ssRNA cleavage ( Figure 2—figure supplement 1B , C ) . We also tested whether the SauCas9 RNP was able to cleave a second ssRNA substrate that was added to the reaction after it reached completion ( Figure 2—figure supplement 1D , E ) . After 1 hr of incubation , the addition of a second target ssRNA complementary to the guide RNA resulted in a burst of cleavage activity , whereas a non-complementary ssRNA substrate did not stimulate cleavage . The second target ssRNA is cleaved to a comparable extent to that observed when this second target was the only substrate in the reaction ( Figure 2—figure supplement 1D , E , compare reactions 1 and 3 ) . These observations suggest that SauCas9 RNP is still competent and available for cleavage at the end of the reaction and that a property intrinsic to the ssRNA substrate is the limiting factor . We propose that the observed difference in cleavage extents for various RNA substrates reflects the fraction of molecules that are structurally accessible for cleavage by the SauCas9 RNP . Based on the biochemical ability of SauCas9 RNP to bind and cleave ssRNA substrates , we wondered whether this activity might provide protection against RNA phage infection in bacteria . To test this , we generated a plasmid library encoding sgRNAs containing guide sequences complementary to the genome of MS2 , a single-stranded RNA phage that can infect E . coli . A subset of these sgRNAs contained scrambled guide sequences that should not target MS2 , providing negative controls . Another sgRNA subset included single-nucleotide mismatches introduced at each position of a target sequence to test for mismatch sensitivity in ssRNA recognition . This plasmid library , comprising 18 , 114 sgRNAs , was co-transformed into E . coli along with a vector encoding a catalytically active version of SauCas9 and the population of transformants was subjected to infection by bacteriophage MS2 ( Figure 3A ) . The experiment was performed in biological triplicate and included an untreated control population and two experimental conditions ( multiplicities of infection ( MOIs ) of 10 and 100 ) . After selection , plasmids were recovered from surviving colonies and sequenced ( Figure 3A ) . We identified between 131 and 166 sgRNAs that were significantly enriched ( false discovery rate ( FDR ) -adjusted p-value<0 . 05 ) in the two different MS2 infection conditions ( Figure 3B ) . The majority of these sgRNAs were perfectly complementary to the MS2 genome , and only three and five control sgRNAs ( out of 708 total control sgRNAs ) for the MOI-10 and −100 conditions , respectively , were enriched ( Figure 3B ) . The lengths of enriched guide sequences were skewed toward shorter targeting lengths ( Figure 3—figure supplement 1A , left ) ; however , this likely reflects bias in the cloned input library since the ratio between the enriched guide sequences and those of the library without phage selection are similar ( Figure 3—figure supplement 1A , right ) . When comparing the degree of enrichment between the different guide lengths , the 23-nt guide segment sgRNAs were preferentially enriched over those of shorter length ( Figure 3C ) , consistent with the in vitro observation that longer guides are more efficient for directing ssRNA cleavage ( Figure 1—figure supplement 4C ) . To assess whether there was any sequence bias within the enriched guides , we aligned guide sequences of all lengths at their 3’ end . These alignments showed no specific sequence bias in the enriched guides relative to those in the unselected library ( Figure 3—figure supplement 1B ) . This is consistent with the crystal structure of an SauCas9-sgRNA-DNA-bound complex which revealed the absence of base-specific contacts of Cas9 to the target strand ( Nishimasu et al . , 2015 ) . Strikingly , mapping enriched guide sequences onto the MS2 genome showed that enriched sgRNAs were clustered at specific regions , which were consistent across both experimental conditions ( Figure 3D; Figure 3—figure supplement 1C , D ) . Together with our biochemical data suggesting that SauCas9 cannot bind or cleave structured RNAs ( Figure 3 ) , we interpret these targeting ‘hotspots’ to be regions of low structural complexity . It is important to note that sgRNAs containing different guide segment lengths overlap at these regions , possibly indicating that increases in targeting efficiency due to guide length are secondary to target accessibility to the Cas9 RNP . We mapped the enriched guide sequences onto the published secondary structure of the MS2 genome determined through cryoelectron microscopy ( Dai et al . , 2017 ) ( Figure 3—figure supplement 2 ) . Guides targeted not only single-stranded , accessible regions but also those that form apparently stable secondary structures . The structure of the MS2 genome was determined on the intact phage particle , however , and may not represent the RNA structure ( s ) relevant to the infection stage during which SauCas9-mediated protection is crucial . Highly enriched sgRNAs from the screen were confirmed for their ability to confer protection against MS2 phage infection through a soft-agar plaque assay . Reconstitution of SauCas9 with a targeting guide confers approximately a ten-fold protection against the RNA phage ( Figure 3E , F ) . No protection was observed in the absence of an sgRNA or SauCas9 protein . Scrambling the sequence of the guide also abrogates protection , confirming that sequence complementary is necessary for phage elimination . Guide segments of all lengths tested ( 20–23 nts ) conferred protection to a similar level ( Figure 3—figure supplement 3A , B ) , consistent with the result from the MS2 screen that guide segments of all lengths were enriched in ‘hotspot’ regions ( Figure 3D; Figure 3—figure supplement 1C ) . Two ‘control’ guides were enriched in both the MOI-10 and −100 treatments . Interestingly , both guides conferred protection but their scrambled counterparts did not ( Figure 3—figure supplement 3C , D ) . Whereas a possible off-target binding site was found for one guide ( #14238 ) within the MS2 genome ( Figure 3—figure supplement 3E ) , it remains unclear how guide #14210 confers protection . Possibly this sgRNA acts by targeting an E . coli host factor that is necessary for infection . Screening against the MS2 genome was also used to test the effect of single-nucleotide mismatches on SauCas9’s targeting ability . We computed an average fold change ( between phage treated and untreated samples ) for all sgRNAs that contained a mismatch at the same position , and obtained average values for mismatches at each position across the guide . We observed a pronounced gradient of increasing guide stringency with length . On average , short guides were less sensitive to mismatches , while mismatches in longer sgRNAs led to decreased recovery compared to control samples ( Figure 3—figure supplement 4A , B ) . Previous work and models suggest that shorter guide segments should be more sensitive to mismatches and lead to higher fidelity Cas9 targeting ( Fu et al . , 2014; Bisaria et al . , 2017 ) . Further study is needed to thoroughly examine this unexpected pattern of RNA-targeting stringency , as one shortcoming of this experiment is that mismatched guides were not designed , a priori , to recognize accessible parts of the MS2 genome . Nevertheless , despite potential noise introduced in this analysis due to guide segments that target inaccessible MS2 regions , we observe an interesting correlation between mismatches in the MS2 screen and in vitro biochemical cleavage assays for the sgRNA with a 23 nt guide segment sequence ( Figure 3—figure supplement 4C , D ) . The first few nucleotides in the ‘seed’ region ( guide 3´ end proximal ) are sensitive to mismatches , while a central region of sensitivity is also observed , similar to previously demonstrated regions of sensitivity for SpyCas9 DNA cleavage ( Cong et al . , 2013; Jiang et al . , 2013; Fu et al . , 2016; Gorski et al . , 2017 ) . An efficient RNA-targeting Cas9 could serve as an important tool in regulating gene expression in vivo . To test the ability of SauCas9 to mediate repression of host gene expression , we targeted dSauCas9 and dSpyCas9 RNPs to a GFP reporter sequence encoded in the E . coli chromosome ( Qi et al . , 2013 ) . Catalytically inactive versions of Cas9 were used to prevent cleavage of the bacterial chromosome when targeting a site adjacent to a PAM . As expression of Cas9 and sgRNA exerts metabolic stress on E . coli , GFP fluorescence values were normalized by the OD600 value to account for differences in cell growth between cultures ( Oakes et al . , 2016 ) . When using sgRNAs designed to recognize a sequence in the GFP gene adjacent to the appropriate PAM for SauCas9 ( NNGRRT ) or SpyCas9 ( NGG ) , GFP expression is significantly reduced ( Figure 4A ) consistent with CRISPR-interference ( CRISPRi ) ( Qi et al . , 2013; Gilbert et al . , 2014 ) . When sgRNAs were designed to recognize GFP sequences not flanked by a PAM , dSauCas9 but not dSpyCas9 was able to repress GFP expression . The SauCas9-mediated GFP repression was dependent on sgRNAs that target the coding strand; sgRNAs that recognize the non-coding strand did not result in reduced GFP expression ( Figure 4—figure supplement 1A ) . The length of the targeting sequence in vivo corroborates in vitro data , with longer guides working more efficiently ( Figure 4B ) . Different guide sequences display variable efficiencies of targeting . We tiled sgRNAs across the GFP mRNA sequence to test the robustness of dSauCas9 to repress GFP expression ( Figure 4C ) . As no sites are adjacent to PAM sequences , all repression presumably occurs on the mRNA level . The efficiency of dSauCas9-mediated GFP repression varied according to the target sequence , with some dSauCas9 RNPs reducing GFP signal to 15–30% of that observed in the presence of the sgRNA alone ( Figure 4C , GFP2 and 6 ) and others showing no ability to repress GFP expression ( GFP7 and 9 ) . Electrophoretic mobility shift assays support the conclusion that repression is not occurring at the dsDNA level by promiscuous PAM binding ( Figure 4—figure supplement 1B ) . Repression is largely equivalent between catalytically active and inactive forms of SauCas9 ( Figure 4—figure supplement 1C ) , suggesting that binding of the Cas9-sgRNA complex to the mRNA is sufficient for repression and consistent with in vitro data showing that the enzyme does not catalyze multiple-turnover RNA cleavage . While we speculate that the Cas9-RNP blocks the ribosome directly ( either at initiation or during elongation ) , our data do not rule out the possibility that Cas9 is otherwise destabilizing the mRNA transcript through an unknown mechanism . Together our biochemical and in vivo data support a model in which SauCas9 can readily bind and cleave bacteriophage RNA and mRNA sequences that are exposed and unstructured ( Figure 4D ) . Regions that form strong structures are inaccessible to SauCas9 RNP binding , thereby preventing cleavage or repression activity . As Cas9 cleavage activity is limited by target accessibility , we expect that RNA occluded by RNA-binding proteins would also be recalcitrant to cleavage . Investigation of CRISPR-Cas9 has focused on its function as a double-stranded DNA endonuclease , while the ability of diverse homologs to cleave natural RNA substrates has remained unexplored . Here , we present evidence that type II-A and type II-C Cas9 enzymes can catalyze programmable and PAM-independent single-stranded RNA cleavage . Focusing on SauCas9 , we show that this enzyme can be employed both biochemically and in cells to cleave RNA and regulate genes on both the transcriptional and translational level in parallel by accounting for target site PAM proximity . Importantly , SauCas9 ssRNA scission requires only an sgRNA and does not need a PAMmer , thereby simplifying applications ( Nelles et al . , 2015 ) and facilitating delivery to cells as a pre-assembled RNP ( Zuris et al . , 2015; Mout et al . , 2017 ) The RNA-targeting capability of SauCas9 and related Cas9 enzymes offers the advantage of repressing viruses whose lifecycles do not involve a DNA genome or intermediate , thereby rendering them inaccessible to Cas9-mediated DNA cleavage . We demonstrated that SauCas9 could be programmed to confer protection to E . coli against MS2 , an RNA bacteriophage with no DNA intermediate . Whether RNA-based viral repression by Cas9 occurs in natural systems is not known , but seems possible based on our results . DNA cleavage by SauCas9 remains more rapid than RNA cleavage , indicating that DNA-targeting is probably the biologically preferred method for phage and plasmid interference . However , Cas9 activity on RNA is PAM-independent and may mitigate the effects of PAM-escape mutants that would evade DNA-level interference ( Deveau et al . , 2008 ) , thus acting as an additional line of defense . Intriguingly , ‘hotspots’ of preferential targeting emerged when tiling guides across the genome , but these sites were devoid of sequence bias . In conjunction with in vitro cleavage data of partially structured RNAs , we suggest that SauCas9 cleavage efficiency is inversely related to structural complexity of the RNA target . As an alternative to the current approach of screening multiple sgRNAs for activity , experimental knowledge about RNA structure , such as SHAPE-seq data ( Loughrey et al . , 2014 ) , would simplify target identification for viral targeting and repression experiments . Nevertheless , future work will concentrate on understanding the structural constraints on RNA targeting and methods to improve Cas9 access to duplex RNA regions . SauCas9 holds promise for a range of RNA targeting applications . We showed that SauCas9 could repress gene expression in E . coli . Repression of the reporter occurs in the absence of the PAM and is specific for targeting of the coding strand . Recently , the Type VI CRISPR-Cas system effector , Cas13 , has been proposed and demonstrated to target RNA ( Shmakov et al . , 2015; Abudayyeh et al . , 2016; East-Seletsky et al . , 2016 ) . ‘Activated’ Cas13 exhibits robust trans cleavage of RNAs ( Abudayyeh et al . , 2016; East-Seletsky et al . , 2016; Smargon et al . , 2017 ) . While RNA-cleavage by SauCas9 is single-turnover and kinetically less robust than that of Cas13 , Cas9 does not cleave RNAs indiscriminately and lends itself to targeting of specific transcripts . A programmable Cas9 capable of repressing genes on the RNA level has potential advantages over CRISPRi DNA-based techniques ( Qi et al . , 2013; Gilbert et al . , 2014 ) . For example , isoform-specific targeting of different transcripts originating from the same transcription start site or resulting from alternative splicing events might be possible . More broadly , due to its intrinsic ssRNA-binding activity , SauCas9 may have utility as a platform for directing other effector proteins to specific RNA molecules , such as proteins or domains that up-regulate translation or RNA base-modifying enzymes for site-specific epigenetic modification of RNAs . Cas9 homolog sequences were obtained from Chylinski and colleagues ( Chylinski et al . , 2014 ) . A structure-guided alignment was produced using PROMALS3D ( Pei et al . , 2008 ) and a maximum-likelihood tree was inferred using PHYML3 . 0 ( Guindon et al . , 2010 ) . The structure of the pUC ssRNA target was predicted using Mfold ( Zuker , 2003 ) . All proteins were expressed as His-MBP fusions ( Addgene vector #29706 ) in E . coli strain BL21 ( DE3 ) . Cells were grown to an OD600 of 0 . 6–0 . 8 , induced with 0 . 4M IPTG , and then incubated overnight at 16˚C with shaking . Proteins were purified using Superflow Ni-NTA affinity resin ( Qiagen , Valencia , CA ) , followed by a HiTrap HP Heparin column ( GE Healthcare , Pittsburgh , PA ) and gel filtration on a Superdex S200 ( GE Healthcare , Pittsburgh , PA ) , as previously described ( Jinek et al . , 2012 ) . Cas9 protein sequences can be found in Supplementary file 1 . DNA oligonucleotides were synthesized by IDT ( Coralville , IA ) . Target RNAs and sgRNAs were transcribed in vitro as previously described ( Sternberg et al . , 2012 ) . DNA targets and in vitro transcribed RNAs were gel purified by 7M urea denaturing PAGE . Target RNAs and DNAs were 5´ end-labeled with [γ-P32-ATP] by treatment with PNK ( NEB , Ipswich , MA ) . T1 sequencing and hydrolysis ladders were prepared according to manufacturer’s directions ( Ambion , Grand Island , NY ) . A list of all sgRNAs and targets can be found in Supplementary file 1 . Cas9 was reconstituted with equimolar sgRNA in 1x cleavage buffer ( 20 mM Tris-HCl – pH 7 . 5 , 200 mM KCl , 1 mM TCEP , 5% glycerol , 5 mM MgCl2 ) for 10 min at 37˚C , then immediately placed on ice . Cleavage reactions were conducted with 1 nM target and 10 nM reconstituted Cas9-sgRNA in 1x cleavage buffer unless otherwise noted . Structured RNA substrates were prepared by annealing two separate in vitro transcribed RNAs . The target strand was annealed with 10-fold excess of the non-target strand to ensure that all target is complexed prior to the cleavage reaction . Reactions were incubated at 37˚C for the indicated time and quenched in Heparin-EDTA buffer ( 10 µg/ml heparin , 25 mM EDTA ) at 25˚C for 5 min . Reactions were diluted with 2x Formamide loading buffer and incubated at 95˚C for 5 min prior to separation on a 15% denaturing 7M urea PAGE gel . Gels were dried overnight and exposed to a phosphor imaging screen ( Amersham/GE Healthcare , Pittsburgh , PA ) . Results were visualized on a Typhoon ( GE Healthcare , Pittsburgh , PA ) and quantified in ImageQuantTL ( v8 . 1 , GE Healthcare , Pittsburgh , PA ) . The cleaved fraction of total signal was calculated independently for three separate experiments and were fit with a one-phase exponential decay model in Prism7 ( GraphPad Software , La Jolla , CA ) . Binding reactions consisted of 750 nM catalytically inactive SauCas9 reconstituted with sgRNA to the final concentrations indicated . Radiolabeled target RNA was added to a final concentration of 1 nM and the reactions were incubated at 37˚C for one hour . Bound probe was separated from unbound using a three-filter system on a vacuum manifold ( Rio , 2012 ) . Membranes were allowed to dry prior to phosphor imaging and quantification . EMSAs were performed in the presence of 300 nM dSauCas9 and 1 nM radiolabeled target strand DNA pre-annealed in the presence of 10x non-target strand . Complexes were incubated at 37˚C for 1 hr prior to separation on 6% non-denaturing PAGE . Gels were dried prior to phosphor imaging . Three independent experiments were performed and the fraction of bound out of total signal was calculated in ImageQuantTL . Binding isotherms were determined in Prism7 using a one-site binding model . All guides of length 20–23 nt antisense to the MS2 bacteriophage genome were synthesized ( CustomArray Inc . , Bothell , WA ) and cloned into a guide expression vector ( Oakes et al . , 2016 ) modified with the SauCas9 sgRNA scaffold . XL1-Blue E . coli cells with a vector containing a tetracycline-inducible wtSauCas9 construct were made electrocompetent and transformed with the MS2-guide plasmid library in triplicate . Approximately 1 × 106 transformants were grown for 30 min at 37˚C with shaking prior to addition of antibiotics and 10 nM anhydrotetracycline ( aTc ) ( Sigma , St . Louis , MO ) for protein induction . After an additional 30 min of growth , cultures were split into three equal pools and treated with none , 3 . 3 × 106 , or 3 . 3 × 107 MS2 bacteriophage . After 3 hr of infection , cells were plated on LB-agar supplemented with antibiotics and incubated at 37˚C for 16 hr . Plates were scraped with LB and plasmids were isolated using a MidiPrep kit ( Qiagen , Valencia , CA ) , according to the manufacturer’s protocol . High-throughput sequencing libraries were prepared by PCR amplification of the variable region of the guide plasmid . Dual unique-molecular identifiers ( UMIs ) , included to separate true single-nucleotide mismatches , as well as duplicates , from PCR artifacts ( Kou et al . , 2016 ) , were incorporated during a single round of PCR . Excess UMIs were removed by ExoI digestion ( Thermo Scientific , Waltham , MA ) prior to library amplification and barcoding . Individual guides ( Supplementary file 1 ) were cloned using oligonucleotides synthesized by IDT and co-transformed into XL1-Blue E . coli cells with the SauCas9 vector . Resistance to MS2 bacteriophage was conducted using a soft-agar overlay method ( Abudayyeh et al . , 2016 ) and plaque forming units ( PFUs ) were calculated . To minimize variability in plaquing efficiency , the same phage dilutions were used for all experiments . After applying a low-pass filter , reads were trimmed using cutadapt v . 1 . 14 ( Martin , 2011 ) and paired-end overlapping reads were merged using pandaseq ( Masella et al . , 2012 ) for error correction . Reads were mapped to the MS2 genome with bowtie2 v2 . 3 . 0 ( Langmead and Salzberg , 2012 ) using the ‘very-sensitive’ option and de-duplicated based on the dual-UMI ( Smith et al . , 2017 ) . Feature counts were obtained using HTSeq-count ( Anders et al . , 2015 ) . Differential expression was calculated using standard pipelines implemented in ‘edgeR’ ( Robinson et al . , 2010; McCarthy et al . , 2012 ) . Significantly enriched guides were defined as those with an FDR-corrected p-value<0 . 05 . Guides with a positive fold-change compared to the control were mapped to the MS2 genome and visualized using the ‘Sushi’ package ( Phanstiel et al . , 2014 ) . To examine for nucleotide composition bias , sequences of guides with a significant positive enrichment were aligned at the 3´ end ( PAM-proximal ) and motifs were analyzed using the WebLogo server ( Crooks et al . , 2004 ) . The distribution of log2 fold-change values of significantly enriched guides plotted as box and whisker plots in Prism . The secondary structure of the MS2 genome was obtained from ( Dai et al . , 2017 ) and reads were mapped and visualized in Forna ( Kerpedjiev et al . , 2015 ) . Log2 fold-change values of single-nucleotide mismatch ( SNP ) guides for each treatment were partitioned by length and averaged at each position . High-throughput sequencing data accompanying this paper are available through the Sequencing Read Archive under the BioProject accession number PRJNA413805 . Based on the system outlined previously , SauCas9 was cloned into a tetracycline-inducible vector , while individual guides are under control of a constitutive promoter ( Oakes et al . , 2016 ) . Plasmids were transformed into an E . coli strain with a GFP reporter gene integrated into the chromosome ( Qi et al . , 2013 ) . Cultures were grown in M9 medium supplemented with 0 . 4% w/v glucose to mid-log phase and diluted to an OD600 of 0 . 05 prior to transfer to a Tecan Microplate reader ( Tecan Systems , San Jose , CA ) . Protein expression was induced with 10 nM aTc . GFP and OD600 were measured every ten minutes for at least 18 hr . Curves of GFP expression over time were fit with a logistic growth model in Prism . At 80% of the maximum value , or at least after 16 hr of growth , the GFP signal was normalized by cell density at OD600 . To account for effects of guide and protein expression , GFP/OD600 was normalized to a null guide or null protein culture , respectively . As expression of different guides change GFP expression levels , the ratio between normalized RNP and guide values was taken to allow comparison of RNP-based repression across different guides . All experiments were conducted in triplicate and all graphing and quantitative analyses were conducted in Prism . Guide and target sequences can be found in Supplementary file 1 .	Similar to humans , bacteria use an immune system known as the CRISPR-Cas system to protect themselves against invading pathogens such as viruses . CRISPRs are specialized stretches of DNA that guide Cas9 to the right location , while Cas9 proteins act like scissors that can cut foreign DNA . When a virus infects a bacterium , the bacterium steals a piece of DNA from the virus and stores it in its CRISPR region . The bacterium then produces a small RNA template that matches the stolen DNA of the virus and adds a specialized protein to it . When the virus infects the cell again , the protein-RNA complex can recognize the virus and stop the infection . Researchers have successfully adapted this system as a gene-editing tool to target and modify specific DNA sequences in different organisms . Cas9 can target and cut DNA , but until now , it was not clear whether this protein could also efficiently target RNA – the ‘genetic middleman’ between DNA and proteins . RNA is essential to make proteins , and being able to target RNA would allow researchers to answer many important questions about RNA biology . To investigate this further , Strutt et al . used three different subtypes of Cas9 proteins and small RNA sequences in a test tube . The results showed that two of the protein subtypes could target RNA efficiently , and one of which was able to target any RNA sequence . Strutt et al . then used one Cas9 to target specific RNA sequences in bacteria and were able to reduce the amount of protein made from that gene . Moreover , the Cas9 protein helped to protect the bacteria against an RNA virus . This work lays the foundation for using this Cas9 protein as a tool for researchers to study RNA in cells . A next step will be to test if Cas9 can cut RNA in human cells . If this works , it could allow direct targeting of RNA viruses , such as West Nile and Dengue , to stop them from infecting human cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2018	RNA-dependent RNA targeting by CRISPR-Cas9
Recent technological advances now allow for the collection of vast data sets detailing the intricate neural connectivity patterns of various organisms . Oh et al . ( 2014 ) recently published the most complete description of the mouse mesoscale connectome acquired to date . Here we give an in-depth characterization of this connectome and propose a generative network model which utilizes two elemental organizational principles: proximal attachment ‒ outgoing connections are more likely to attach to nearby nodes than to distant ones , and source growth ‒ nodes with many outgoing connections are likely to form new outgoing connections . We show that this model captures essential principles governing network organization at the mesoscale level in the mouse brain and is consistent with biologically plausible developmental processes . The network of physical connections among neurons in the brain provides the medium for neural communication . Investigations of these anatomical networks are typically categorized as macro- , meso- , or microscale , depending on the spatial resolution of the techniques used . The mesoscale , which describes the connectivity among local populations ( hundreds to thousands ) of neurons , is an attractive intermediate between the two more extreme scales: it has higher granularity than macroscale data , which details connectivity between large anatomically defined brain areas , but it has a broader lens than the microscale , which is concerned with synaptic level connections , often in relatively small volumes of tissue ( see ( Sporns et al . , 2005 ) for review ) . These properties make the mesoscale tractable enough for whole-brain ( i . e . connectomic ) studies with current technological and analytical tools . The Allen Institute for Brain Science recently constructed a mesoscale connectome for the mouse ( the Allen Mouse Brain Connectivity Atlas ) , which was the first complete connectivity dataset of a mammalian brain at the mesoscale ( Oh et al . , 2014 ) . Using injections of an anterograde fluorescent viral tracer and serial two-photon microscopy , Oh et al . ( 2014 ) comprehensively mapped both intra- and interhemispheric axonal tracts and estimated the directed connectivity structure among 213 non-overlapping anatomical regions . The authors also conducted a preliminary graph theoretic analysis and showed that basic network properties of the mouse connectome could not be explained by any one standard network model ( Oh et al . , 2014 ) . Rubinov et al . ( 2015 ) extended this analysis by identifying small-worldness , a hierarchical modular structure , and non-optimal wiring in the connectome ( Rubinov et al . , 2015 ) . As the mouse is one of the most pervasive model organisms in biomedical science , a deeper characterization of its connectome is likely to provide pertinent groundwork for future studies of brain development and function as well as yield insights into the broader organizational principles of the mammalian brain . Graph theory provides a mathematical framework for investigating the organization of networks and has been increasingly applied in neuroscience over the last 15 years . Graphs are mathematical objects that consist of nodes and connections between the nodes , called edges ( Bullmore and Sporns , 2009; Rubinov and Sporns , 2010 ) . Edges can be either directed or undirected , as well as binary or weighted . In the Allen mouse connectome , nodes and edges correspond to brain regions and axonal tracts , respectively . Conventionally , real-world networks ( such as the World Wide Web or social networks ) are compared to binary undirected graphs such as small-world ( Watts and Strogatz , 1998 ) and scale-free graphs ( Barabasi and Albert , 1999 ) . These models have been critical for understanding the conditions under which various network properties , such as small-worldness ( high clustering among nodes combined with short average path lengths between node pairs ) and scale-freeness ( defined by power-law degree distributions ) , emerge in the brain and other systems . Modeling brain networks with these standard graphs , however , requires some limiting simplifications . First , because they are not embedded in physical space , these networks ignore the biological cost of constructing physical fiber tracts , as well as spatial constraints imposed by the surrounding tissue . Second , small-world and scale-free models ignore directionality , intrinsically discarding information about differences in incoming and outgoing connection patterns . Both properties are crucial considerations when modeling the brain ( Laughlin and Sejnowski , 2003; Song et al . , 2014; Kaiser et al . , 2009; Ercsey-Ravasz et al . , 2013 ) . Third , these models were either not developed with neural data ( Barabasi and Albert , 1999 ) or used data from simple model organisms , such as Caenorhabditis elegans , at a particular spatial scale ( Watts and Strogatz , 1998 ) , which constrains their relevance to understanding connectomes of other organisms . For these reasons , the field has recently turned to exploring non-standard network models to elucidate generative principles of real brain networks . Recent work in developing plausible generative network models for the brain has primarily addressed spatial embedding ( Song et al . , 2014; Kaiser et al . , 2009; Ercsey-Ravasz et al . , 2013; Klimm et al . , 2014; Kaiser and Hilgetag , 2004 ) . While the exact approaches differ in their implementation and scale of the networks being modeled , a common theme is that a node is more likely to connect to nearby nodes than distal ones . This organizational principle has been able to capture a range of properties observed in the cortex , including the distribution of connection lengths ( Song et al . , 2014; Kaiser et al . , 2009; Ercsey-Ravasz et al . , 2013 ) , the inverse relationship between degree and clustering coefficient ( Watts and Strogatz , 1998; Song et al . , 2014; Betzel et al . , 2015; Mitra , 2014 ) , and the relative frequency of three-node motifs ( Ercsey-Ravasz et al . , 2013 ) . Additional generative rules have been explored by Klimm et al . ( 2014 ) , and while the resulting models have captured many properties of cortical networks , the authors note that these rules likely do not reflect the underlying generative principles of cortical networks . Similarly , Betzel et al . , ( 2015 ) , recently reported that individual human macroscale connectomes are well-fitted by generative network models , which use both spatial proximity and homophilic attraction ( i . e . nodes with similar graph theoretic properties are more likely to form connections ) . However , the homophilic rules employed by Betzel et al . also do not lend themselves to straightforward biophysical interpretations . Indeed , the difficulty of developing biophysically interpretable rules is a recurring challenge in generative network models ( Watts and Strogatz , 1998; Song et al . , 2014; Klimm et al . , 2014; Betzel et al . , 2015; Vértes et al . , 2012 ) . Here , we provide an in-depth analysis of the mouse connectome’s properties and use the findings to develop a generative network model of the mesoscale connectome . We characterized the directed and undirected degree distributions , clustering coefficient distribution , reciprocity , global efficiency , physical edge length distribution , nodal efficiency , and the characteristic path length of the connectome ( Table 2 in 'Materials and methods' for definitions or Bullmore and Sporns , 2009 and Rubinov and Sporns , 2010 for review ) . Informed by these data , we developed a spatially embedded directed network model . This model uses two simple generative principles: proximal attachment ( PA ) ‒ outgoing connections are more likely to attach to nearby nodes than distal ones , and source growth ( SG ) ‒ nodes with many outgoing connections are more likely to develop new outgoing connections . We show that this simple model , parameterized only by a length constant and the number of nodes and edges , can capture directed , undirected , and spatial properties of the mouse connectome . This work supports the existing literature on the importance of spatial embedding and provides strong evidence that SG is a major phenomenological rule that shapes connectivity patterns in the mouse brain . Lastly , we propose biological mechanisms that might account for these two generative principles . We first compared the undirected structure of the mouse connectome with that of three well-characterized standard graphs commonly used in the literature: a degree-controlled random network ( Maslov and Sneppen , 2002 ) , a small world network ( Watts and Strogatz , 1998 ) , and a scale-free network ( Barabasi and Albert , 1999 ) . The mouse connectome is characterized by a degree distribution with many low-degree nodes and a long tail of high-degree nodes ( Figure 1a; ( Oh et al . , 2014 ) ) . The degree distribution was not well replicated by any standard graph ( Figure 1a and b ) , nor was the clustering coefficient distribution ‒ a finding also shown in Oh et al . ( 2014 ) . Although the scale-free network’s degree distribution most closely resembles that of the connectome , by construction , it cannot capture the distribution of low-degree nodes ( as all nodes are instantiated with a minimum number of edges ) . 10 . 7554/eLife . 12366 . 003Figure 1 . Standard graph models vary in their ability to recreate the mouse connectome’s degree distribution and relationship between degree and clustering coefficient . ( a-b ) Probability density of degree distributions for the mouse connectome , and an average over 100 repeats of scale-free , small-world , and ( degree-controlled ) random networks plotted with ( a ) linear and ( b ) log-linear scales . ( c-f ) Clustering coefficient as a function of degree for each node in ( c ) the mouse connectome , ( d ) a degree-controlled random network , ( e ) a small-world network , and ( f ) a scale-free network . Each plot shows data from 426 nodes and the best fitting power law function ( dashed line ) . ( a ) is similar to Figure 3c from Oh et al . ( 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 003 The structural discrepancy between the mouse connectome and standard networks was further exposed by examining the relationship between degree and clustering coefficient ( Figure 1c–f ) . In the connectome , nodes with lower degree tend to be more clustered . This relationship was well-fitted with a power law: Ci∝kiγ , where Ci and ki denote clustering coefficient and degree , respectively , and with the best-fit γ=−0 . 44 . Previous studies have shown that such a relationship ( with varying γ < 0 ) is common to many real-world networks ( Ravasz and Barabási , 2003 ) , including the human connectome ( Klimm et al . , 2014 ) . While the small-world network exhibits a clustering coefficient distribution similar to that of the connectome ( Oh et al . , 2014 ) , as well as an inverse relationship between clustering coefficient and degree , these similarities are superficial: inspecting Figure 1e reveals that the small-world network’s degree distribution is much more homogeneous . While the scale-free network partly captures the connectome’s degree distribution , its nodes have a much lower clustering coefficient , and there is a weaker relation between these two metrics ( Figure 1f ) . The degree-controlled random model shows that shuffling the connectome’s edges yields a graph resembling the scale-free network ( Figure 1d ) . Thus , these graphs fail to capture key aspects of the connectome’s undirected structure . Previous modeling efforts have shown that several properties of brain networks can be captured by simple models , which assume that spatially nearby nodes are more likely to connect than distal ones ( Song et al . , 2014; Kaiser et al . , 2009; Ercsey-Ravasz et al . , 2013; Kaiser and Hilgetag , 2004 ) . Therefore , we explored an undirected spatially embedded network model . We first assigned to each node a spatial position randomly sampled from a 7 mm x 7 mm x 7 mm cube , which gave an inter-nodal distance distribution similar to the connectome ( Figure 5—figure supplement 1 ) . Edges were then added between pairs of nodes i and j by choosing node i at random and node j with probability Pij∝exp ( −dij/L ) . That is , the probability of choosing a target node j decayed with the Euclidean distance dij between nodes i and j according to length constant L ( see 'Materials and methods' ) . We call this rule proximal attachment ( PA ) . Relative to the connectome , the purely geometric model exhibited relatively narrow Gaussian-like degree distributions across several values of L ( Figure 2a ) . Thus , the PA rule fails to generate low- and high-degree nodes . This geometric model does , however , exhibit an inverse relationship between degree and clustering ( Figure 2b ) as in the connectome . However , Figure 2b demonstrates that there is no value of L for which this model adequately captures the joint degree-clustering distribution seen in the connectome; the PA model misses both low-degree nodes with high clustering and high-degree nodes with low clustering . This suggests that the purely geometric rules explored in previous studies ( Song et al . , 2014; Kaiser et al . , 2009; Ercsey-Ravasz et al . , 2013; Kaiser and Hilgetag , 2004 ) are not sufficient to recreate the mouse connectome’s properties . Therefore , we explored the hypothesis that topological rules , which are based on properties of individual nodes , play an important role in forming the connectivity patterns of the mouse connectome . 10 . 7554/eLife . 12366 . 004Figure 2 . Example networks generated using the ( purely geometric ) proximal attachment ( PA ) rule where connection probability between two nodes decreased with distance . ( a ) The degree distribution for networks grown with three values of L ( in mm ) , each averaged across 100 repeats . The mouse connectome is shown for comparison . ( b ) Clustering coefficient as a function of degree for representative networks grown with the same values of L used in ( a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 004 Conventionally , topological rules in generative network models have been applied to undirected networks . However , the mouse connectome contains directed edges , allowing us to probe whether the directionality of connections in the mouse connectome plays a role in shaping connectivity patterns . In a directed graph , edges point from source nodes to target nodes , so one can consider each node’s in-degree and out-degree ( i . e . the total number of incoming and outgoing connections for a node , respectively ) . We found a surprisingly asymmetric relationship between these distributions in the connectome ‒ while the in-degree of the network was approximately normally distributed , the out-degree distribution exhibited a peak near zero and a long tail ( Figure 3a ) , much like an exponential distribution . There was no significant correlation between in- and out-degree ( ρ = 0 . 127 , p = 0 . 065 , Spearman rank correlation ) . Figure 3b shows the proportion of incoming edges as a function of total degree for all nodes in the mouse connectome . As the total degree of a node increases , the proportion of incoming edges decreases . 10 . 7554/eLife . 12366 . 005Figure 3 . Directed analysis of the mouse connectome reveals different distributions for in- and out-degree . ( a ) Out-degree as a function of in-degree for all nodes in the mouse connectome . Margins show in- and out-degree distributions with the blue lines and axis labels corresponding to a logarithmic scale . In-degree is approximately normally distributed while the out-degree approximately follows an exponential distribution . ( b ) Proportion in-degree as a function of total degree ( i . e . in-degree divided by the sum of incoming and outgoing edges ) . Low-degree nodes tend to have mostly incoming edges , whereas high-degree nodes are characterized by mostly outgoing edges . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 00510 . 7554/eLife . 12366 . 006Figure 3—figure supplement 1 . Distribution of proportion in-degree in the mouse connectome . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 006 To model the directed mouse connectome , we extended the purely geometric PA model by considering two mathematically symmetrical , yet phenomenologically distinct frameworks: target attraction and source growth ( TAPA and SGPA ) . When adding an edge in the TAPA model , a target node is chosen with a probability proportional to its in-degree , while the source node is chosen with a probability that decreases with its distance from the target node ( Figure 4 , top ) . In the SGPA model , a source node is chosen with a probability proportional to its out-degree , while the target node is chosen with a probability that decreases with its distance from the source node ( Figure 4 , bottom ) . The TAPA and SGPA models can be considered directed spatial variants of the preferential attachment algorithm introduced by Barabasi and Albert ( 1999 ) and lead to “rich-get-richer” patterns of either incoming or outgoing connection formation . In both cases , all nodes were initialized upon network instantiation ( unlike in the preferential attachment algorithm where nodes are added iteratively ) , and directed edges were added iteratively until the number of edges matched the mouse connectome . Since these models also incorporated the geometric PA property , both models were parameterized by a single free parameter L exactly as in the PA model . 10 . 7554/eLife . 12366 . 007Figure 4 . Target attraction and source growth network generation algorithms . The numbers inside each node indicate the probability of growth or attachment . For illustration , the most probable node ( dashed ) is selected in both diagrams ( T and S , corresponding to target and source , respectively ) . Top row: target attraction proximal attachment ( TAPA ) model . A target node is selected with a probability proportional to its in-degree ( left ) , while the source node is chosen with a probability that decreases exponentially with the node’s Euclidean distance from the target ( right ) . Two nodes have zero probability of forming an edge since the target is already receiving projections from these nodes . The dashed red line shows the most probable edge . Bottom row: source growth proximal attachment ( SGPA ) model . A source node is selected with a probability proportional to its out-degree ( left ) , while the target node is chosen with a probability that decreases exponentially with the node’s Euclidean distance from the source ( right ) . The dashed cyan line shows the most probable edge . In both algorithms , we assume that each node begins with a self-connection ( corresponding to an outgoing and incoming edge ) to avoid zero-valued probabilities , though self-connections are not shown here or included when calculating any metrics . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 007 The SGPA model exhibits an in-degree distribution that is approximately normal and an out-degree distribution that is approximately exponential ( Figure 5a , cyan ) . This is matches the mouse connectome ( Figure 3a ) but is exactly opposite for the TAPA model ( Figure 5a , red ) . Figure 5b shows proportion in-degree as a function of total degree ( in-degree + out-degree ) for both network models . In the SGPA model ( and the mouse connectome; Figure 3b ) , high-degree nodes tend to have a large proportion of outgoing connections . Again , this is opposite of the TAPA model , where high-degree nodes tend to have a large proportion of incoming connections . These results suggest that the directed graph theoretic properties of the mouse connectome are best captured by a SG model of network generation . 10 . 7554/eLife . 12366 . 008Figure 5 . Directed analysis of single representative TAPA and SGPA network models and reciprocity comparison with the connectome . ( a ) Out-degree as a function of in-degree for both algorithms with L = 0 . 725 mm , which was chosen to match the connectome’s reciprocity ‒ see e ) . Margins show in- and out-degree distributions . ( b ) Proportion in-degree as a function of total degree for both algorithms . The SGPA model qualitatively captures the connectome’s directed degree distributions and proportion in-degree ( cf . Figure 2 ) . ( c ) Edge length distribution for the connectome , shown for both reciprocal ( blue ) and non-reciprocal edges ( black ) . ( d ) Same as ( c ) but for the SGPA model . ( e ) Reciprocity coefficient as a function of the length parameter for both TAPA ( red ) and SGPA ( cyan ) , with shading indicating standard deviation over 100 repeats . Both models intersect the connectome at L = 0 . 725 mm . For reference , the reciprocity coefficient for the connectome ( magenta ) and a corresponding degree-controlled random graph ( gold ) are also shown . SGPA and TAPA models overlap . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 00810 . 7554/eLife . 12366 . 009Figure 5—figure supplement 1 . Directed degree distributions and proportion in-degree for a directed Erdos-Renyi graph . ( a ) In- and out-degree are independent and both are approximately normally distributed . ( b ) Proportion in-degree is independent of total degree . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 00910 . 7554/eLife . 12366 . 010Figure 5—figure supplement 2 . Directed degree distributions and proportion in-degree for a purely topological source-growth or target-attraction directed graph ( SG only , TA only , respectively ) , with no proximal attachment ( equivalent to L = ∞ in SGPA or TAPA ) . The qualitative properties of the in- and out-degree distributions are the same as in SGPA and TAPA with L = 0 . 725 . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 01010 . 7554/eLife . 12366 . 011Figure 5—figure supplement 3 . Directed degree distributions and proportion in-degree for an SGPA model where the network is grown one node at a time . Hotter colors indicate more recently added nodes . ( a ) Out- and in-degree relationships for one representative model instantiation with L = 0 . 725 mm . ( b ) Proportion in-degree as a function of total degree . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 01110 . 7554/eLife . 12366 . 012Figure 5—figure supplement 4 . Inter-nodal distance distribution for the mouse brain ( magenta bars ) and a 7 mm³ cube ( black line ) . The use of a 7 mm x 7 mm x 7 mm cube was justified on the basis that its inter-nodal distance distribution closely mimicked that of the connectome . The results did not change considerably when coordinates from the mouse brain were used instead of the 7 mm³ cube . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 01210 . 7554/eLife . 12366 . 013Figure 5—figure supplement 5 . Directed degree distributions and proportion in-degree for a graph in which source selection was proportional to total degree ( in-degree + out-degree ) in ( a ) and ( b ) , and in which source selection was proportional to total degree raised to the power γ = 1 . 67 in ( c ) and ( d ) . In both ( a ) and ( c ) , there are significant correlations between the in- and out-degrees ( ρ = 0 . 45 , p< 10-10 and ρ = 0 . 60 , p < 10-10 for ( a ) and ( b ) , respectively; Spearman rank correlation ) . In all graphs , L = 0 . 725 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 013 Previous work has shown that both the reciprocity coefficient and average clustering coefficient of the brain are well above chance ( Oh et al . , 2014; Rubinov and Sporns , 2010; Felleman and Van Essen , 1991; Kaiser and Varier , 2011 ) . We found that the mouse connectome has a reciprocity coefficient of 0 . 13 , which is high compared to that expected from chance ( 0 . 03 ) . Reciprocal edges in the connectome are on average shorter than nonreciprocal edges ( Figure 5c; t ( 8818 ) = 13 . 25 , p << 10-10 , independent samples t-test; Cohen’s d = 0 . 47 ) . As shown in Figure 5e , the reciprocity coefficient of our model networks decreases with increasing length constant , approximately matching that of the mouse connectome ( 0 . 13 ) when L = 0 . 725 mm for both the SGPA and TAPA models . However , fitting the length constant to reciprocity ( e . g . in the SGPA model ) underestimates the physical edge lengths of the connectome ( compare Figure 5c to d ) . Using a larger length constant improved the fit to the reciprocal and nonreciprocal edge length distributions ( not shown ) , but reduced the model’s reciprocity . Incorporating SG into the PA model increases the width of the degree distribution , allowing the model to also capture the joint clustering-degree distribution seen in the connectome ( Figure 6a ) . A network grown solely with the PA rule and a random degree-controlled network both showed a joint clustering-degree distribution that differed from the connectome’s , again suggesting that both topological and geometric rules are important when modeling the connectome . The inset in Figure 6a quantifies this by showing that the SGPA model’s fitted power law relationship between clustering coefficient and degree is the most similar to the connectome . Additionally , this increased clustering is spatially localized: In the connectome , there is a negative correlation between a node’s clustering coefficient and its mean edge length . That is , nodes with short average edge lengths tend to be highly clustered ( ρ = -0 . 238 , p < 10-6 , Spearman rank correlation ) . A similar , but stronger relationship occurs in the SGPA model ( median ρ = -0 . 482 , p < 10-10 , Spearman rank correlation ) . 10 . 7554/eLife . 12366 . 014Figure 6 . Clustering and nodal efficiency for the connectome and other directed models . ( a ) Clustering-degree joint distribution for connectome and one representative instantiation of each model graph . Inset: best-fit power-law exponent γ ( see 'Materials and methods' ) for the connectome ( γ = -0 . 44 , R2=0 . 58 ) and random ( mean γ ± std . = -0 . 17 ± 0 . 02 , median R2=0 . 16 ) , SGPA ( mean γ ± std . = -0 . 48 ± 0 . 04 , median R2=0 . 59 ) , and geometric PA ( mean γ ± std . = -0 . 80 ± 0 . 06 , median R2=0 . 42 ) models . ( b ) Distributions of nodal efficiencies ( see 'Materials and methods' ) with mean ± standard deviation ( line and shaded regions , respectively ) for model networks . Mean ± standard deviation of mean nodal efficiency ( averaged over nodes ) is 0 . 409 ± 0 . 001 for the random model , 0 . 393 ± 0 . 004 for the SGPA model , and 0 . 426 ± . 001 for the pure geometric model . The connectome’s average nodal efficiency is 0 . 375 . The PA model’s histogram peaks at 279 at a nodal efficiency of 0 . 44 . ( b ) and the inset in a ) both used 100 sample instantiations of each model . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 01410 . 7554/eLife . 12366 . 015Figure 6—figure supplement 1 . Clustering and nodal efficiency for Erdos-Renyi ( ER ) and TAPA models . ( a ) Clustering-degree joint distribution for connectome and one representative instantiation of each model graph . The inset shows best-fit power-law exponents , as in Figure 6 ( ER: mean γ ± std . = -0 . 01 ± 0 . 04 , median R2 = 0 . 00; TAPA: mean γ ± std . = -0 . 47 ± 0 . 035 , median R2 = 0 . 60 ) . ( b ) Distributions of nodal efficiencies ( see 'Materials and methods' ) . Solid lines are means and shaded regions are standard deviations . ( b ) and the inset in ( a ) both used 100 sample instantiations of each model . Mean ± standard deviation ( taken over 100 graph instantiations ) of mean nodal efficiency ( averaged over nodes ) is 0 . 467 ± 0 . 002 for ER , 0 . 395 ± 0 . 003 for TAPA . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 01510 . 7554/eLife . 12366 . 016Figure 6—figure supplement 2 . Undirected degree distribution for the SGPA model and the mouse connectome in ( a ) linear and ( b ) logarithmic scales . For the SGPA model , the line and shaded region represent the mean and standard deviation across 100 graph instantiations . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 01610 . 7554/eLife . 12366 . 017Figure 6—figure supplement 3 . Clustering vs . degree ( a ) and nodal efficiency ( b ) for the node-by-node SGPA network ( L = 0 . 725 ) used in Figure 5—figure supplement 3 . Hotter colors indicate more recently added nodes . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 017 Our previous analyses of degree , in-degree , out-degree , and clustering coefficient distributions describe the connectivity patterns of a node in the context of its immediate neighbors . To examine the role played by each node in the context of the entire network , we calculated distributions of nodal efficiency . A node’s nodal efficiency is defined as the mean inverse directed shortest path length between itself and all other nodes in the network and quantifies the ease with which that node can theoretically transmit information to all other nodes ( see 'Materials and methods' ) . As shown in Figure 6b , there was a close match between the nodal efficiency distributions for the connectome and the SGPA model ( with L = 0 . 725 mm ) : both exhibited an approximately normal distribution , save for a small selection of nodes with zero-valued nodal efficiency ( corresponding to nodes with no outgoing connections ) . A degree-controlled random graph also showed approximately normally distributed nodal efficiency distributions , but the mean of the distribution was slightly higher than either the SGPA model or the mouse connectome . In contrast , the purely geometric PA model showed a sharply peaked nodal efficiency distribution , similar to what one would expect for a directed Erdos-Renyi graph ( Figure 5—figure supplement 1 , Figure 6—figure supplement 1 ) . This analysis suggests that the SGPA model captures the statistical connectivity patterns of individual nodes in relation to the whole network . Prior work has shown that a network’s resilience to the removal of nodes can provide insight into its structural composition ( Newman , 2010; Kaiser et al . , 2007 ) . These studies typically explore the structure of undirected networks , so for our analyses we converted the directed SGPA model to an undirected one by ignoring directionality of edges ( see 'Materials and methods' ) . We then simulated a lesioning ( or percolation ) process to compare the undirected SGPA model and other standard undirected models to the mouse connectome . When nodes ( and the edges connected to them ) were removed in order of decreasing degree ( i . e . targeted attack ) , the SGPA model’s global efficiency ( akin to nodal efficiency; see 'Materials and methods' ) and largest ( giant ) component size both decreased in a manner more similar to the mouse connectome than any standard graph ( Figure 7 ) . However , Figure 7a shows that the mouse connectome disintegrates the fastest , and Figure 7b shows that the global efficiency falls more rapidly in the connectome than any model . Thus , the connectome appears more vulnerable to targeted attack than the model networks explored here . 10 . 7554/eLife . 12366 . 018Figure 7 . Response of undirected networks to targeted lesions where nodes are removed in order of highest degree . Mean values ( lines ) ± standard deviations ( shaded regions ) are plotted for each model after 100 repeats . ( a ) Size of the largest ( giant ) connected component in response to targeted attack . The randomly shuffled connectome ( Random ) is obscured by the scale-free graph . ( b ) Global efficiency for the mouse connectome and network models in response to targeted attack . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 01810 . 7554/eLife . 12366 . 019Figure 7—figure supplement 1 . Network connectivity patterns ( and not just connection density ) affect global efficiency . ( a ) Connection density as a function of nodes removed ( for comparison with Figure 7 ) . Except for the small-world network , the connection density was changed similarly throughout the lesioning process . Mean values ( lines ) ± standard deviations ( shaded regions ) are plotted for each model after 100 repeats . ( b ) Global efficiency and connection density at each point during the lesioning process for representative networks grown in the same manner as Figure 7 . Note that global efficiency varies across graphs even at similar connection densities . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 019 The density of edges can influence some metrics ( like global efficiency ) . To test if our lesioning results were a product of connection density ( which is itself determined by degree distribution ) , rather than connectivity patterns specific to each model , we also investigated the connection density throughout the lesioning process ( Figure 7—figure supplement 1a ) . We found that , throughout the lesioning process , the connectome’s connection density was similar to that of all models except the small world network ( and necessarily matched the degree-controlled random graph ) . We also found that while global efficiency and connection density were correlated , their specific relationship depended on the model ( Figure 7—figure supplement 1b ) . Interestingly , the way these two variables were related in the connectome most resembled how they were related in the SGPA model , relative to all the others . This shows that the results in Figure 7 are not simply due to differences in degree distribution and the dependence of global efficiency on connection density . Therefore , we propose that these lesioning results expose differences in higher order connectivity structure between the connectome and the examined models . Average global efficiency and other undirected metrics are shown in Table 1 . 10 . 7554/eLife . 12366 . 020Table 1 . Average undirected metrics for the connectome , standard , and model networks . For the random , small-world , scale-free , and SGPA models , the standard deviation for 100 repeats is also shown . See Table 2 for metric definitions . Note that properties for the original ( single-hemisphere ) connectome are presented in Oh et al . ( 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 020ClusteringCharacteristic path lengthGlobal efficiencyConnectome0 . 3612 . 2260 . 492Random0 . 140 ± 0 . 00132 . 002 ± 0 . 00200 . 528 ± 0 . 0003Small-world0 . 359 ± 0 . 00582 . 129 ± 0 . 00580 . 507 ± 0 . 0010Scale-free0 . 159 ± 0 . 00371 . 998 ± 0 . 00260 . 527 ± 0 . 0004SGPA0 . 343 ± 0 . 00942 . 166 ± 0 . 01590 . 501 ± 0 . 0027 The fact that in- and out-degree were differentially distributed in the connectome highlights an important limitation of undirected graphs: they do not discriminate between in- and out-degree distributions and may therefore fail to reveal key connectivity properties arising from such directed structure . Interestingly , the in- and out-degree relationships observed at other scales and in other organisms differ from our findings here . For instance , the C . elegans connectome ( for the full nervous system ) ( White , 1985 ) has nearly identical in- and out-degree ( exponential ) distributions ( Amaral et al . , 2000 ) . While numerous studies have found common properties across scales ( e . g . small-worldness ) , more pronounced differences in network structure , such as different in- and out-degree distributions , suggest that a single set of generative rules cannot capture the brain’s network structure across different scales and/or organisms . Functional segregation and integration comprise two components of a framework commonly used to interpret brain network architecture ( Rubinov and Sporns , 2010; Tononi et al . , 1994 ) . Functional segregation asserts that the brain carries out specialized computations in anatomically localized and highly interconnected regions . As high clustering coefficient is thought to indicate potential for participation in this sort of computational unit ( Rubinov and Sporns , 2010 ) , our finding that nodes with a small number of neighbors tend to be highly clustered suggests that these nodes are candidates for specialized processing . However , functional data would be required to validate this conjecture since high clustering coefficient does not necessarily imply grouping of nodes into specialized clusters . Functional integration predicts that some high-degree “hub” brain regions consolidate the results of these specialized computations for higher function ( e . g . as in behavior and perception ) ( Friston , 2002 ) . Surprisingly , connections associated with such hub nodes were primarily outgoing in the directed connectome ( Figure 3b ) . Assuming again that anatomical connectivity is indicative of functional relationships , our results suggest that hubs play a stronger role in distribution , rather than integration , of information . However , these ideas are not necessarily contradictory ‒ it is possible that many brain regions are involved in both integration and distribution to varying extents . In fact , the proportion in-degree distribution roughly resembles a uniform distribution ( Figure 3—figure supplement 1 ) suggesting a continuum of these properties rather than discrete categories . Above-chance reciprocity , as we found in the mouse connectome , is also hypothesized to be important for functional integration and segregation ( Tononi et al . , 1994 ) . Previous work has suggested that bidirectional connections between brain regions constrain the dynamics in such a way that a balance arises between the two ( Tononi et al . , 1992; Sporns et al . , 1991 ) . Considering the propensity for feedback within the brain , however , perhaps the relatively high reciprocity is not surprising . Like the asymmetric degree distributions , the connectome’s high reciprocity also highlights the importance of using directed graph models ( as opposed to undirected ones ) when analyzing brain networks . The generative rules explored here follow from a series of recent publications attempting to develop generative network models of animal connectomes ( Song et al . , 2014; Kaiser et al . , 2009; Ercsey-Ravasz et al . , 2013; Klimm et al . , 2014; Betzel et al . , 2015; Lim and Kaiser , 2015 ) . Several papers have documented the importance of PA in generative network models of animal connectomes ( Song et al . , 2014; Kaiser et al . , 2009; Ercsey-Ravasz et al . , 2013; Klimm et al . , 2014; Kaiser and Hilgetag , 2004 ) . For example , Kaiser et al . ( 2009 ) noted that if axonal outgrowth occurs in a straight line and the axon attaches to the first node it physically encounters , then the probability of connecting to a target neuron depends exponentially on the target’s distance from the source neuron ( Kaiser et al . , 2009 ) . Previous research has also shown that some chemical gradients responsible for axon guidance decay exponentially with distance ( Murray , 1993; Isbister et al . , 2003 ) and have been modeled as such ( Mortimer et al . , 2009 ) . Similarly , Rubinov et al . ( 2015 ) , recently reported that distance-dependent inter-areal connection strengths in the mouse connectome were best captured by a power law . We explored a purely geometric model based on PA and found that while this model induces clustering between low-degree nodes , it fails to account for the broad degree distribution of the connectome . This motivated the exploration of two mathematically symmetric generative models that included an additional topological rule: target attraction proximal attachment ( TAPA ) and source growth proximal attachment ( SGPA ) . Only when incorporating the source-growth rule could we capture the in- and out-degree distributions of the mouse connectome . Note that these distributions arise independently of the PA rule , but other characteristics require it , as discussed previously ( see Figure 5—figure supplement 2 for the in- and out-degree distributions of a graph with source-growth but not PA ) . To explore whether source selection based specifically on out-degree was a necessary property , we also examined a network grown with source selection probability proportional to the total degree raised to a power . While this model also exhibited many of the connectome’s properties , it makes predictions that are not observed in the connectome ( e . g . it predicts that the number of incoming and outgoing connections should be correlated; Figure 5—figure supplement 5 ) ; the model itself also requires an additional parameter . This indicates that a SG rule specifically depending on out-degree is a better candidate mechanism for generating the properties of mouse connectome . Our results indicate that much of the organizational structure of the mouse connectome is captured by the geometric and topological generative rules employed by the SGPA model . The ability of these simple rules to closely capture the mouse connectome’s network structure raises the possibility that brain organization at the mesoscopic scale does not require precise specification of connectivity ( e . g . through genetic or transcriptional factors ) , but might instead be largely based on a set of relatively simple instructions . Many generative schemes proposed previously , such as the homophilic rules by Betzel et al . ( 2015 ) ( nodes with similar graph theoretic properties are more likely to connect ) , or the minimal wiring length networks by Klimm et al . ( 2014 ) , are also based on simple connectivity rules , but these rules are not readily interpretable in terms of biophysical processes . We previously discussed possible biological underpinnings of the geometric PA rule , but the topological SG rule also lends itself to speculation on a set of possible biological mechanisms . The source growth rule might be realized during brain development by the actions of a family of proteins known as neurotrophins , which play a major role in promoting the survival of innervating neurons ( Huang and Reichardt , 2001; Chao , 2003 ) and the growth and branching of their axons ( Tessier-Lavigne and Goodman , 1996 ) . Found throughout both the central and peripheral nervous systems , these proteins are typically secreted by a target ( postsynaptic ) neuron at functional synapses , endocytosed by a source ( presynaptic ) neuron , and retrogradely transported to the source neuron’s soma , where they trigger the above-mentioned processes ( Huang and Reichardt , 2001; Korsching , 1993 ) . Because neurotrophins are available in limited quantities , they are hypothesized to cause competitive interactions among growing neuronal populations ( van Ooyen and Willshaw , 1999 ) . Indeed , such an interaction resembles the way in which a single source node is probabilistically selected for each edge addition in the SGPA model ( as opposed to allowing all nodes to generate outgoing connections simultaneously ) . Most importantly , since populations that maintain many functional outgoing connections will tend to receive more neurotrophins , they should on average be more fit for survival and new connection generation , and their existing connections should be more likely to branch . This would in turn increase their probability of establishing new connections to novel targets . Such a biological mechanism would lead to the “rich-get-richer” phenomenon for outgoing connections that was fundamental in modeling the network properties of the mouse connectome . A final insight provided by the SGPA model concerns the undirected degree distribution in the mouse connectome . Degree distributions in the brain have often been characterized by power laws , where p ( k ) ∝kγ for some γ<0 ( Bullmore and Sporns , 2009; Kaiser et al . , 2007 ) . The SGPA model provides an alternative because the in- and out-degree distributions are driven by different rules . In-degree is approximately Gaussian because of the central limit theorem ( connections are random in the sense that they only depend on distance and not degree or other topological properties ) , and out-degree is approximately exponential , owing to the SG process . If in- and out-degree are independent , as they are in our model and seem to be in the connectome , the distribution of their sum ( i . e . of total degree ) will be the convolution of the two individual distributions . Thus , we would expect the distribution over undirected degree ( which is approximately the sum of in- and out-degree , the approximation arising since we do not count reciprocal edges twice ) to approach the convolution of a Gaussian and an exponential , that is , an exponentially modified Gaussian ( Figure 6—figure supplement 2 ) . Additionally , the SGPA model is able to capture the low-degree portion of the distribution ( Figure 1b ) unlike the scale-free graph . This serves as a more biologically motivated alternative to the power-law degree distribution often used to describe the brain’s network properties . While the SGPA model captured a number of the connectome’s characteristics , there are also limitations worth highlighting . Our choice to instantiate all the nodes at the start of the network’s generation is somewhat biologically implausible . This is different from the Barabasi-Albert algorithm used to generate scale-free graphs where , aside from an initial group of nodes , all the nodes are added sequentially over time . However , a version of our SPGA model in which nodes are added one-by-one ( see 'Materials and methods' ) yielded a qualitatively similar network ( see Figure 5—figure supplement 3 and Figure 6—figure supplement 3 ) . The only notable differences were that the youngest nodes had an extremely low degree and a clustering coefficient of either zero or one , and that there was a positive correlation between in- and out-degree . This deviation from the properties observed in the mouse connectome could potentially be corrected by “pruning” the final network or by modifying the probability of generating new connections as the final nodes are added . In our simulations , we chose to randomly draw node positions from a 7 mm³ cube to spatially embed our model . The dimensions of this cube were chosen to match the inter-nodal distance distribution of the connectome ( Figure 5—figure supplement 4 ) . All the simulations we explored , however , yielded comparable results when run with node coordinates determined by the centroids of brain regions in the connectome . One interesting feature of the SGPA model is that it predicts that edge lengths should be significantly shorter than those observed in the connectome . This is a consequence of fitting the length constant to the connectome’s reciprocity . A more satisfactory fit to the edge length distribution can be obtained by using a larger length constant , but this substantially reduces the reciprocity and degrades the inverse relationship observed between clustering and degree ( Figure 6a ) . Equivalently , we can scale the dimensions of the model , which allows us to match the reciprocity and the edge length distribution but not the inter-nodal distance distribution . Thus , our model is unable to account for both the edge length distribution and the reciprocity of the connectome while also maintaining the appropriate dimensionality . Within the lesioning analysis , we found that the connectome is more susceptible to targeted attack than our generative SGPA model network ( or any other model ) . This suggests that the mouse connectome is less resilient than the synthetic networks explored here . More broadly , addressing whether the SGPA model can reproduce the macroscale hierarchical modularity recently reported by Rubinov et al . ( 2015 ) is also a target for future research . Previous work on targeted attacks on the macaque and cat macroscale connectome has shown different patterns of results ( Kaiser et al . , 2007 ) , but it is difficult to assess whether this is due to differences in the organism ( mouse vs . monkey or cat ) , scale ( meso vs . macro ) , or both . Lastly , we treated the entire connectome as a network grown with homogeneous growth rules . The success of our approach lends some merit to this assumption , but it is nevertheless likely that cortico-subcortical connections follow different generative rules than cortico-cortical ones , for example . A similar point was raised by Kaiser and Varier ( 2011 ) and Oh et al . ( 2014 ) ‒ both noted that cortico-cortical projections have higher reciprocity than cortico-subcortical ones in macaque macroscale and mouse mesoscale connectomes , respectively . Uncovering the differences in generative rules employed by subnetworks in the brain , as well as those at different scales or time points , is a target for future research . Dynamically changing the physical scale of the model ( e . g . to simulate physical growth ) is another promising avenue for future work which some have begun to explore ( Ozik et al . , 2004 ) , and may account for our inability to capture the edge length distribution in the connectome . Here , our network was embedded within a physical space that maintained a constant size . We did find that a node’s age affected its final properties ( Figure 5—figure supplement 3 , Figure 6—figure supplement 3 ) , but the growth of neural tissue typically occurs in an expanding physical space which may stretch or alter the oldest connections as development proceeds . Continued use of real data to both inspire and evaluate network models will be crucial for elucidating the principles that govern the network organization of connectomes . We have characterized the network properties of the mouse mesoscale connectome , a system that highlights the importance of using spatial , directed graphs when modeling brain networks . A model that uses two simple organizational principles ‒ source growth and proximal attachment ‒ can capture a large number of directed , undirected and spatial network properties of the mouse connectome . Importantly , these rules have biologically plausible connections to developmental mechanisms and wiring properties in real brains . This model not only serves as a simple mathematical tool that can be used to model and understand mesoscopic brain organization , but also provides a parsimonious framework for informing future investigations of brain network formation . To generate a connectivity graph from empirical data , we used the mouse connectome developed by the Allen Brain Institute . To compile this dataset , anterograde tracers were injected in a single hemisphere and the authors traced projections into both hemispheres . Thus , the connectome probed connections originating in the right hemisphere and terminating in either the right or left hemisphere . The original work computed projection strengths and associated p-values ( the probability that the observed projection would arise by chance ) between 213 pairs of ipsi- and contralateral anatomically defined brain regions ( Oh et al . , 2014 ) . To generate a binary adjacency matrix , we set all connection strengths with p<0 . 01 to one and all other elements to zero . This yielded a graph with a reasonable connection density for analysis . We then assumed that all connectivity projections were bilaterally symmetric in order to construct a 426 x 426 binary adjacency matrix , which completely defines a ( single component ) graph . The results remained qualitatively the same for alternative connection thresholds ( p<0 . 05 and p<0 . 001 ) . The following sections describe the random graphs we included in our analysis . For each random graph , the number of nodes was set to 426 , and the number of edges to 7804 ( for undirected models ) or 8820 ( for directed models ) . Standard random graphs were generated in Python using the NetworkX module ( https://networkx . github . io ) ( Hagberg et al . , undefined ) . The small-world graph ( Watts and Strogatz , 1998 ) is parameterized by the number of nodes n , the number of initial connections for each node kSW , and the probability of rewiring , p . Here , we used kSW = 18 to match the mean degree of the connectome , and p=0 . 23 to approximately match the mean clustering coefficient of the connectome . The scale-free graph ( Barabasi and Albert , 1999 ) is parameterized by the number of nodes n and the number of connections kSF formed by each node as it is added to the network . We used kSF = 18 to match the number of edges in the connectome . We generated the undirected degree-controlled random network by shuffling the mouse connectome’s edges while holding the degree distribution constant ( similar to Maslov and Sneppen , 2002 ) ; this generates a control graph with random connectivity but identical degree distribution . The directed degree-controlled random was generated in a similar way , except both the in- and out-degree distributions were held nearly fixed ( “nearly” because the algorithm often converts a small number of edges to self- or double-connections , which are ignored in our analysis; however , these only represent about 5% of all connections ) . These standard graphs are similar to those used in Oh et al . ( 2014 ) . The directed Erdos-Renyi graph was generated using NetworkX , with an edge probability of 0 . 0487 , which on average yielded the number of directed edges present in the mouse connectome . The following sections describe the algorithms used to generate the custom random graphs examined in this study . For the purely geometric PA graph , 426 nodes were randomly assigned centroids within a 7 mm x 7 mm x 7 mm cube , with no edges connecting the nodes . We then generated each undirected edge by first selecting a source node i at random and subsequently selecting a target node j with probability Pij∝exp ( −dij/L ) , where dij denotes the distance between node i and node j . If an undirected edge already existed between the source and target , the probability of selecting that target was set to zero . Edges were added until the number of ( undirected ) edges matched that of the connectome ( K=7804 ) . The directed PA algorithm was identical , except that directionality of edges was retained . For the graphs incorporating topological rules ( i . e . SGPA and TAPA ) , the growth algorithm was initialized by instantiating n nodes with no edges connecting them , except for one self-connection per node to prevent zero-valued connection probabilities for nodes with no outgoing or incoming edges ( for SGPA and TAPA , respectively ) . However , these self-connections were ignored when calculating all metrics for the final graph . In all the graphs we generated ( except for node-by-node SGPA ) , we matched n and K to the number of nodes and directed edges , respectively , in the empirical connectome ( n = 426 , K = 8820 ) . At each step in the growth process an edge is added by ( 1 ) selecting the target node j from all nodes without maximum in-degree according to the in-degree of j: , and ( 2 ) selecting the source node i with a probability that decreases with distance from the target: P ( source=i\|target=j ) ∝exp ( −dij/L ) . If a connection already existed from node i to j , this probability was set to zero to avoid a duplicate edge . dij is the Euclidean distance between nodes i and j , and L is a parameter governing the strength of distance-dependence . For the nonspatial TA graph ( Figure 5—figure supplement 2 ) , the algorithm was as the TAPA algorithm , except the source was chosen with a uniform probability for all nodes ( i . e . no distance-dependence ) . At each step in the growth process an edge is added by ( 1 ) selecting the source node i according to its out-degree as P ( source=i ) ∝kiout from all nodes that do not have maximal out-degree , and ( 2 ) selecting the target node j with a probability that decreases with distance from the source: P ( target=j\|source=i ) ∝exp ( −dij/L ) . If a connection already existed from node i to j , this probability was set to zero to avoid a duplicate edge . For the nonspatial SG graph ( Figure 5—figure supplement 2 ) , the algorithm was as the SGPA algorithm , except the target was chosen with a uniform probability for all nodes ( i . e . no distance-dependence ) . Both the target attraction and source growth rules can be conceptualized as directed spatial variants of the preferential attachment rule introduced by Barabasi and Albert ( 1999 ) . This graph was identical to the SGPA graph mentioned above , except that probability of selecting the source node i was proportional to its total degree ( in-degree + out-degree ) raised to a power γ: P ( source=i ) ∝ ( kiout+kiin ) γ . Target selection was as in the SGPA model . In this algorithm , nodes are added one at a time with positions sampled uniformly from within a 7 mm x 7 mm x 7 mm cube until the graph contains 426 nodes . Upon each node addition ( starting with the second node addition ) , each existing edge “branches” with a probability p=0 . 016 ( which was chosen so as to yield approximately the same number of edges as in the connectome ) . When an edge “branches” a new edge is formed that has the same source node and whose target node is chosen with probability P ( target=j\|source=i ) ∝exp ( −dij/L ) ( i . e . in the same way as in the standard SGPA model ) . If an edge already exists between the source node and the selected target node , no new edge is added . When all the edges and nodes have been added , any nodes that are not connected to the largest ( giant ) graph component ( typically only three or four , representing < 1% of the network ) are removed from the graph . See Table 2 for metric definitions . We calculated undirected metrics ( e . g . clustering coefficients and undirected degree ) and carried out a lesioning study for directed graphs by first casting the directed graph to an undirected one . This yielded a graph which contained an undirected edge between every pair of nodes that had been connected by at least one directed edge in the directed graph . 10 . 7554/eLife . 12366 . 021Table 2 . Graph theoretical metrics used for analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 12366 . 021MetricBrief interpretationDefinitionDegreeNumber of edges connected to a node . This generalizes to in- or out-degree in directed graphs , describing the number of incoming or outgoing connections for a node , respectively . ki=∑j∈naij N=set of all nodes aij=1 , if edge from node i to j exists0 , otherwise Directed versions: kiin=∑j∈Naji‍ ‍ ‍ ‍ ‍‍ ‍ ‍ ‍ ‍ ‍ kiout=∑j∈Naij aij=1 , if directed edge from node i to j exists0 , otherwiseClustering coefficient ( Watts and Strogatz , 1998 ) Level of connectivity among nearest neighbors of node i2tiki ( ki-1 ) ti = number of triangles that include node iCharacteristic path length ( Watts and Strogatz , 1998 ) Mean shortest undirected path length over all pairs of nodes1n ( n−1 ) ∑i , j∈N , j≠idij n = number of nodes dij = shortest undirected path from i to jGlobal efficiency ( Latora and Marchiori , 2001 ) Mean inverse shortest undirected path length over all pairs of nodes1n ( n−1 ) ∑i , j ∈N , j≠i1dij dij= shortest undirected path from i to jNodal efficiency ( generalized from [Achard and Bullmore , 2007] ) Mean inverse shortest directed path length from a single node to all other nodes1 ( n−1 ) ∑j∈N , j≠i1dij dij = shortest directed path from i to jReciprocity coefficientProportion of edges from node i to node j that have a reciprocal connection from node j to node i ( when i≠j ) 1Ne∑i , j∈N , j≠iaijaji aij =1 , if directed edge from i to j exists0 , otherwise Ne = total number of undirected edges	Within the brain , neurons organize themselves into extensive networks . The physical connections between neurons determine which groups of neurons are able to communicate with one another . Recently , researchers mapped out the neural circuits within the entire brain of an adult mouse . The resulting wiring diagram , or ‘connectome’ , provides an opportunity to study these brain networks in unprecedented detail . One key question is how these networks acquire their structure . Henriksen , Pang and Wronkiewicz wondered whether the patterns of connections in the adult mouse brain might provide clues to how this connectivity emerges . Analyzing the adult mouse connectome revealed a number of unexpected properties . For example , a brain region’s incoming connections were often different from its outgoing connections . This suggests that it is important to consider the direction of connections between groups of neurons , and that different mechanisms may govern how incoming and outgoing connections are formed . Furthermore , if a brain region was connected to only a few others , those regions tended to also be connected among themselves . Using this information , Henriksen , Pang and Wronkiewicz attempted to ‘grow’ a virtual mouse connectome from scratch using two simple rules deduced from the properties of the adult connectome . The first rule was that brain regions with many outgoing connections are more likely to form more outgoing connections . The second was that outgoing connections are more likely to connect to nearby brain regions than to distant ones . The resulting model successfully reproduced a number of key properties of the mouse brain connectome . This suggests that relatively simple principles help to determine at least some of the structure of networks within the adult brain . The next challenge is to identify the exact relationships between these principles and the biological mechanisms that support brain development .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2016	A simple generative model of the mouse mesoscale connectome
Hemodynamic forces regulate vascular functions . Disturbed flow ( DF ) occurs in arterial bifurcations and curvatures , activates endothelial cells ( ECs ) , and results in vascular inflammation and ultimately atherosclerosis . However , how DF alters EC metabolism , and whether resulting metabolic changes induce EC activation , is unknown . Using transcriptomics and bioenergetic analysis , we discovered that DF induces glycolysis and reduces mitochondrial respiratory capacity in human aortic ECs . DF-induced metabolic reprogramming required hypoxia inducible factor-1α ( HIF-1α ) , downstream of NAD ( P ) H oxidase-4 ( NOX4 ) -derived reactive oxygen species ( ROS ) . HIF-1α increased glycolytic enzymes and pyruvate dehydrogenase kinase-1 ( PDK-1 ) , which reduces mitochondrial respiratory capacity . Swine aortic arch endothelia exhibited elevated ROS , NOX4 , HIF-1α , and glycolytic enzyme and PDK1 expression , suggesting that DF leads to metabolic reprogramming in vivo . Inhibition of glycolysis reduced inflammation suggesting a causal relationship between flow-induced metabolic changes and EC activation . These findings highlight a previously uncharacterized role for flow-induced metabolic reprogramming and inflammation in ECs . Atherosclerotic cardiovascular disease remains the leading cause of morbidity and mortality in the United States ( National Heart , Lung , and Blood Institute , 2013 ) . Altered blood flow characteristics ( unidirectional vs . disturbed flow ) and associated changes in flow-generated mechanical forces ( hemodynamics ) play a critical role in the focal nature of atherosclerosis ( Davies et al . , 2013; Gimbrone and García-Cardeña , 2016; Zhou et al . , 2014 ) . Unidirectional blood flow ( UF ) is associated with high time-averaged shear stress and is ‘athero-protective’ promoting a healthy endothelium , characterized by endothelial cell quiescence and maintenance of vascular barrier integrity . In contrast , disturbed flow ( DF ) , which occurs in areas where atherosclerosis develops , is associated with low time-averaged shear stress and is ‘athero-susceptible’ promoting EC ‘activation’ characterized by inflammation and reduced vascular barrier integrity ( Davies et al . , 2013; Zhou et al . , 2014; Dai et al . , 2004; Chiu and Chien , 2011; Liao , 2013; Abe and Berk , 2014; Xiao et al . , 2013 ) . These areas where ECs are subjected to DF are the branched and curved parts of the arterial tree and underscores the focal , flow-dependent nature of endothelial ‘activation’ . These flow-dependent harbingers of atherosclerosis develop before any visible signs of disease ( Hajra et al . , 2000; Won et al . , 2007 ) . Endothelial cells are plastic and their phenotypes are tightly regulated by hemodynamic changes . When cultured under static conditions , ECs exhibit increased expression of pro-inflammatory cytokines , reduced nitric oxide production ( Kizhakekuttu et al . , 2012; Chen et al . , 2010; Eelen et al . , 2015 ) , and a shift towards glycolysis ( Eelen et al . , 2015; De Bock et al . , 2013a ) . Compared to static conditions , UF reduces glycolysis in a Krüppel-like Factor 2 ( KLF2 ) -dependent manner ( Dekker et al . , 2002; Doddaballapur et al . , 2015 ) and increases mitochondrial biogenesis ( Chen et al . , 2010; Kim et al . , 2014 ) but whether UF alters mitochondrial function is controversial ( Doddaballapur et al . , 2015 ) . Furthermore , the effect of physiological DF , as opposed to no flow , on cellular metabolism is unclear and the mechanisms behind these DF-induced changes remain under-explored ( De Bock et al . , 2013a; Doddaballapur et al . , 2015; Cucullo et al . , 2011; Wilhelm et al . , 2016 ) . Here , we employed RNA-seq , pathway analyses and complementary in vitro and in vivo systems to study the effects of DF and UF on cellular metabolism of human aortic endothelial cells ( HAECs ) . We show that DF induces a metabolic phenotype of increased glycolysis and reduced oxidative phosphorylation , compared to UF . Our RNA-seq analyses predicted HIF-1α as a major transcriptional regulator controlling the DF-induced changes in endothelial phenotypes . Mechanistically , we demonstrate that DF activates hypoxia-inducible factor-1α ( HIF-1α ) via induction of NAD ( P ) H Oxidase-4 ( NOX4 ) and consequent production of ROS . Activation of HIF-1α was required for DF-induced metabolic reprogramming characterized by increased glycolysis and repression of mitochondrial oxidative phosphorylation due to inhibition of pyruvate dehydrogenase ( PDH ) via increased pyruvate dehydrogenase kinase-1 ( PDK1 ) expression . Reversal of DF-induced metabolic reprogramming reduced endothelial activation . These new molecular insights identify a previously uncharacterized role of disturbed hemodynamics in stabilizing endothelial HIF-1α to dynamically regulate endothelial metabolic plasticity , and consequently , endothelial activation and vascular health . Human aortic endothelial cells ( HAECs ) were subjected for 24 hr to ‘athero-susceptible’ DF mimicking the hemodynamics measured in human carotid sinus or ‘athero-protective’ UF representing the wall shear stress in human distal internal carotid artery ( Dai et al . , 2004 ) . RNA-seq whole-genome transcriptome profiling and multiple pathway analyses were used to globally determine genes and gene networks that are regulated in endothelium exposed to either DF or UF ( Wu et al . , 2017b ) . Analysis of RNA sequencing identified a total of 3757 differentially expressed genes ( DEGs ) using a false discovery rate cut off value of q < 0 . 05 . Gene set enrichment analysis ( GSEA ) was performed on all DEGs to identify overrepresented biological pathways ( Subramanian et al . , 2005 ) . The DEGs are available at the publisher’s website ( Figure 1—source data 1 ) . Among these pathways , the cellular response to hypoxia and glycolytic metabolism were the most up-regulated gene sets under disturbed flow ( Figure 1A ) . The enrichment plot for hypoxia and glycolysis gene sets under DF are provided in Figure 1—figure supplement 1A and B , respectively , and list of enriched genes under DF contributing to the hypoxia and glycolytic gene sets are provided in Figure 1—source data 2 , respectively . As an example , genes traditionally associated with hypoxia ( e . g . , VEGFA ) and glycolysis ( e . g . , SLC2A1 and LDHA ) are positively enriched under DF . Functional analysis of the RNA-seq data using DAVID ( the Database for Annotation , Visualization , and Integrated Discovery ) ( Huang et al . , 2009a ) , showed that glucose metabolism and response to hypoxia were among the top 10 gene ontology ( GO ) biological processes induced by DF ( Figure 1B ) . In contrast , UF had no effect on these pathways ( Figure 1—figure supplements 2 and 3 ) . A third pathway analysis was conducted by the bioinformatics tool METASCAPE ( Tripathi et al . , 2015 ) , which also showed that DF induced upregulation of DEGs enriched in glycolytic metabolism ( Figure 1—figure supplement 4 ) . Hypoxic and glycolytic signature genes are persistent up to 3 days , as shown in Figure 1—figure supplement 5 . 10 . 7554/eLife . 25217 . 003Figure 1 . Disturbed flow induces a glycolytic phenotype and inhibits mitochondrial function . Primary HAECs were subjected to either unidirectional flow ( UF ) or disturbed flow ( DF ) for 24 hr before cell lysates were collected . ( A ) Gene set enrichment analysis of RNA-seq dataset of top 10 pathways enriched under DF compared to UF . ( B ) Top 10 gene ontology pathways of differentially expressed genes as calculated via DAVID . Source data of differentially expressed genes available online . ( C ) HAECs are treated with UF or DF for 48 hr before staining with 2-NBDG . Under DF , the cells are markedly brighter and are quantified D ( five random fields , 100 cells per condition ) . ( E ) Intracellular glucose uptake can also be quantified with an uptake assay using 2-DG6P . Under DF , there is increased glucose uptake ( n = 4 ) . ( F ) HAECs are treated with either UF or DF for 48 hr before re-plating in a Seahorse XFe24 analyzer and assessed with a glycolysis stress test or ( H ) mitochondrial stress test . In ( F ) , arrows indicate injection of glucose , oligomycin A , and 2-deoxyglucose ( 2DG ) . ( G ) Under glycolysis stress test , HAECs treated with DF demonstrate increased glycolysis and glycolytic capacity when compared against UF ( n = 10 ) . ( H ) Arrows denote addition of oligomycin , carbonyl cyanide-4- ( trifluoromethoxy ) phenylhydrazone ( FCCP ) , and rotenone/antimycin . ( I ) Under mitochondrial stress test , HAECs treated with DF have decreased basal respiration , maximal respiration , and respiratory capacity when compared against UF ( n = 7 ) . p<0 . 05; p<0 . 005; *p<0 . 0005 as determined by Student’s t-test . Data represent mean ± SEM . Bar is 10 microns . Source data for RNA-seq differentially expressed genes can be found online: DOI: https://doi . org/10 . 5281/zenodo . 260122 ( Wu et al . , 2017b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 00310 . 7554/eLife . 25217 . 004Figure 1—source data 1 . Differentially expressed genes after UF or DF and RNAseq . HAECs were treated with either UF or DF for 24 hr prior to cell lysis and total RNA sequencing . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 00410 . 7554/eLife . 25217 . 005Figure 1—source data 2 . Gene sets for hypoxia and glycolysis as ranked by GSEA of HAECs under disturbed flow . ( A ) The gene set ‘HALLMARK_HYPOXIA’ ranked by GSEA of HAECs under disturbed flow . ‘Rank Metric Score’ is the signal-to-noise ratio for each gene used to position the gene in the ranked list . ‘Rank in Gene List’ refers to the position of the gene in the ranked list of all genes present in the RNAseq dataset . The ‘Running Enrichment Score’ is the enrichment score for this set at this point in the ranked list of genes . ( B ) The gene set ‘HALLMARK_GLYCOLYSIS’ ranked by GSEA of HAECs under disturbed flow . ‘Rank Metric Score’ is the signal-to-noise ratio for each gene used to position the gene in the ranked list . ‘Rank in Gene List’ refers to the position of the gene in the ranked list of all genes present in the RNAseq dataset . The ‘Running Enrichment Score’ is the enrichment score for this set at this point in the ranked list of genes . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 00510 . 7554/eLife . 25217 . 006Figure 1—figure supplement 1 . Gene set enrichment analysis ( GSEA ) enrichment plots for the hypoxia and glycolysis gene sets in unidirectional vs disturbed flow in HAECs . HAECs were subjected to UF or DF for 24 hr before RNA isolation and total RNA sequencing ( RNAseq ) . Genome-wide expression analysis of the RNAseq data was performed using gene set enrichment analysis ( GSEA ) software version 2 . 2 . 4 available from the Broad Institute ( http://www . broadinstitute . org/gsea/downloads . jsp ) . The GSEA algorithm calculates an enrichment score reflecting the degree of overrepresentation at the top or bottom of the ranked list of the genes included in a gene set in a ranked list of all genes present in the RNAseq dataset . ( TOP ) A positive enrichment score ( ES ) indicates gene set enrichment at the top of the ranked list; a negative ES indicates gene set enrichment at the bottom of the ranked list . The final enrichment score for a set is the maximum deviation from zero encountered for that set . ( MIDDLE ) The location of the hypoxia or glycolytic genes in the ranked set of all genes . The ‘Rank Metric Score’ is the signal-to-noise ratio for each gene used to position the gene in the ranked list . ( BOTTOM ) The distribution of the rank metric score across all genes present in the expression dataset . The gene set collection used was the h . all . v5 . 1 . symbols . gmt [Hallmarks] gene sets database . The analysis demonstrates that known ( A ) hypoxia genes and ( B ) many glycolysis enzymes are positively enriched in the disturbed flow sample . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 00610 . 7554/eLife . 25217 . 007Figure 1—figure supplement 2 . Gene set enrichment analysis of RNA-seq dataset of top 10 pathways enriched under unidirectional flow compared to disturbed flow . Primary HAECs were subjected to either UF or DF for 24 hr before cell lysates were collected . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 00710 . 7554/eLife . 25217 . 008Figure 1—figure supplement 3 . Gene ontology analysis of top 10 biological processes of differentially expressed genes under unidirectional flow using DAVID . Primary HAECs were subjected to either UF or DF for 24 hr before cell lysates were collected . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 00810 . 7554/eLife . 25217 . 009Figure 1—figure supplement 4 . Gene ontology analysis of top biological processes of differentially expressed genes under DF by Metascape . Primary HAECs were subjected to either UF or DF for 24 hr before cell lysates were collected . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 00910 . 7554/eLife . 25217 . 010Figure 1—figure supplement 5 . Persistence of glycolytic and hypoxia genes under disturbed or unidirectional flow . HAECs were subjected to either DF or UF for 24 , 48 , or 72 hr prior to cell lysis , RNA purification , and qRT-PCR . Fold change of expression of HK2 , SLC2A1 , EGLN3 , and VEGFA , relative to cells not treated with flow at each time point , are given above according to the color-code legend ( n = 3 for each time point and each condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 01010 . 7554/eLife . 25217 . 011Figure 1—figure supplement 6 . Mitochondrial ATP production is dependent on flow . HAECs were subjected to either DF or UF for 48 hr . 20 , 000 cells were then reseeded in either media , media with 2-deoxyglucose , or rotenenone and antimycin . The ATP production was then quantified with a luciferase assay ( n = 4 for each condition , technical replicates ) . DF ATP production comes almost exclusively from glycolysis , compared to UF , which uses mitochondria for a portion ( ~17% ) of ATP production . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 01110 . 7554/eLife . 25217 . 012Figure 1—figure supplement 7 . Glycolysis stress test and mitochondria stress test for HAEC after 48 hr of static , unidirectional flow , and disturbed flow . In ( A ) , arrows indicate injection of glucose , oligomycin A , and 2-deoxyglucose ( 2DG ) ( n = 6 for each data point ) . ( B ) Arrows denote addition of oligomycin , carbonyl cyanide-4- ( trifluoromethoxy ) phenylhydrazone ( FCCP ) , and rotenone/antimycin ( n = 6 for each data point ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 01210 . 7554/eLife . 25217 . 013Figure 1—figure supplement 8 . Comparison of glycolytic gene expression under disturbed flow , unidirectional flow , and static conditions . HAECs were subjected to either no flow ( static ) , DF or UF for 48 hr prior to cell lysis , RNA purification , and qRT-PCR . Fold change of expression of HK2 , SLC2A1 , and PDK1 are shown ( relative to static ) ( n = 3 ) . p<0 . 05; *p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 013 To determine the metabolic phenotypes of HAECs mediated by athero-relevant flow waveforms , we performed glucose uptake assays and Seahorse bioenergetics measurements ( Pike Winer and Wu , 2014 ) . Glucose uptake was determined by two approaches: ( a ) 2- ( N- ( 7-Nitrobenz-2-oxa-1 , 3-diazol-4-yl ) Amino ) -2-Deoxyglucose ( 2-NBDG ) fluorescence measurement ( Zou et al . , 2005 ) and ( b ) 2-deoxyglucose-6-phosphate ( 2-DG6P ) , a non-degradable glucose-like substrate which utilizes a colorimetric output ( Han et al . , 2015 ) . Cells subjected to DF exhibited increased glucose uptake as evidenced by increased 2-NBDG fluorescence ( Figure 1C and D ) and 2-DG6P uptake ( Figure 1E ) as compared to UF . To functionally characterize the DF-induced endothelial metabolic phenotype , HAECs were subjected to a glycolysis stress test or a mitochondrial stress test following 48 hr of exposure to either DF or UF . The glycolysis stress test measures extracellular acidification rate ( ECAR , an indicator of glycolytic lactate production ) after the addition of glucose ( basal glycolysis ) and after addition of the mitochondrial inhibitor oligomycin A ( glycolytic capacity ) ( Pike Winer and Wu , 2014; Nigdelioglu et al . , 2016 ) . DF increased both basal glycolysis and glycolytic capacity in HAECs as compared to UF ( Figure 1F , G ) . The mitochondrial stress test measures oxygen consumption rate ( OCR ) at baseline levels , and after addition of oligomycin A ( coupled respiration ) , the mitochondrial uncoupler Carbonyl cyanide-p-trifluoromethoxyphenylhydrazone ( FCCP ) ( maximum respiration ) , and respiratory inhibition with antimycin A and rotenone ( non-mitochondrial oxygen consumption ) ( Pike Winer and Wu , 2014 ) . DF reduced basal respiration , maximal respiration , and spare respiratory capacity in HAECs ( Figure 1H , I ) . To support these results , we measured the relative dependence of cellular ATP levels on glycolysis and mitochondria as a function of flow type was measured via luminescence ( Borowski et al . , 2013 ) . Whereas under UF mitochondrial oxidative phosphorylation accounted for ~17% of total ATP production , under DF mitochondrial oxidative phosphorylation accounted for less than 1% of the total ATP ( Figure 1—figure supplement 6 ) . We also performed bioenergetics measurements to compare DF and UF to static , no flow conditions . As shown in Figure 1—figure supplement 7A and B , HAECs under static conditions show an intermediate glycolytic and oxidative phenotype . Glycolytic gene expression of static HAECs was also compared against UF and DF at 48 hr in Figure 1—figure supplement 8 . Static HAECs expressed glycolytic transcriptional responses more similar to UF than DF . Collectively , these results demonstrate that DF increases endothelial glycolysis and reduces mitochondrial oxidative phosphorylation as compared to UF . Ingenuity Pathway Analysis ( IPA ) was performed to probe major putative upstream regulators that control the endothelial transcriptomes under DF or UF ( Calvano et al . , 2005 ) . Based on the flow-sensitive 3757 DEGs and a priori knowledge of upstream gene regulation , IPA predicted Krüppel-like Factor 2 ( KLF2 ) as a major driver of transcriptional changes induced by UF , and HIF-1α driving transcriptional changes induced by DF ( Figure 2A , IPA network genes for top 10 predicted upstream regulators are provided in Figure 2—source data 1 ) . HIF-1α protein expression was increased in DF-subjected HAECs ( Figure 2B , Figure 2—figure supplement 1 ) . However , DF had no effect on HIF-1α mRNA levels ( Figure 2C ) suggesting that DF increases HIF-1α protein through a post-transcriptional mechanism . Furthermore , the DF-induced increase in HIF-1α protein levels was dynamic and reversible . The DF-induced increase in HIF-1α protein could be reversed by switching to UF , whereas the UF-induced reduction could be reversed by switching to DF ( Figure 2D ) . 10 . 7554/eLife . 25217 . 014Figure 2 . Disturbed flow induces HIF-1α expression , which accounts for a major portion of the differentially expressed genes under DF . Primary HAECs were subjected to either UF or DF for 24 hr before samples were collected for RNA-seq . ( A ) Using Ingenuity Pathway Analysis , the RNA-seq data was analyzed for transcription factor predictions . The top two predicted transcription factors are KLF2 ( upregulated under UF ) and HIF-1α ( upregulated under DF ) , as judged by activation z-score . ( B ) Western blot of time course of HAECs subjected to UF or DF . HIF-1α starts to appear around 8 hr under disturbed flow . ( C ) qPCR quantification of HIF-1α under 48 hr of DF or UF ( n = 4 ) . ( D ) HAECs were subjected to either DF or UF for 24 hr , then either UF or DF , respectively , for 24 hr , before lysates were collected . The expression of HIF-1α is reversible . ( E ) HAECs were treated with either non-targeting siRNA or siHIF-1α for 24 hr before being subjected to AS flow for 48 hr . The cell lysates were then collected and sent for total RNA sequencing . Gene set enrichment analysis of pathways shows that hypoxia and glycolysis are among the top three pathways that are modulated by HIF-1α knockdown . ( F ) Gene ontology pathways of the differentially expressed genes downregulated by HIF-1α knockdown also demonstrate that metabolic changes via glycolysis is one of the most downregulated pathways . Source data of differentially expressed genes available online . Significance determined by Student’s t-test . Data represent mean ± SEM . Source data for RNA-seq differentially expressed genes can be found online: DOI: https://doi . org/10 . 5281/zenodo . 260120 ( Wu et al . , 2017a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 01410 . 7554/eLife . 25217 . 015Figure 2—source data 1 . Target genes in RNAseq dataset of predicted transcription factors for HAECs treated with UF and DF , by IPA . Differentially expressed genes from the UF/DF RNAseq data set were used . The Upstream Regulator is the predicted molecule or transcription factor which regulates the target molecules . The Predicted Activation State is the direction in which the Upstream Regulator is expressed under UF . The activation z-score statistic is a weighted sum of activating and inhibiting interactions on a given gene set . The overlap p‐value measures whether there is a statistically significant overlap between the dataset genes and the curated genes that are regulated by the Upstream Regulator . It is calculated using Fisher’s Exact Test . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 01510 . 7554/eLife . 25217 . 016Figure 2—source data 2 . Differentially expressed genes after DF and siHIF-1α and RNAseq . HAECs were treated with siHIF-1α or non-targeting siRNA ( SC ) for 24 hr prior to 48 hr of DF prior to cell lysis and total RNA sequencing . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 01610 . 7554/eLife . 25217 . 017Figure 2—source data 3 . Gene sets for hypoxia and glycolysis as ranked by GSEA of HAECs under disturbed flow and treated with siRNA targeted against HIF-1α . ( A ) The gene set ‘HALLMARK_HYPOXIA’ ranked by GSEA of HAECs under disturbed flow and treated with control siRNA . ‘Rank Metric Score’ is the signal-to-noise ratio for each gene used to position the gene in the ranked list . ‘Rank in Gene List’ refers to the position of the gene in the ranked list of all genes present in the RNAseq dataset . The ‘Running Enrichment Score’ is the enrichment score for this set at this point in the ranked list of genes . ( B ) The gene set ‘HALLMARK_GLYCOLYSIS’ ranked by GSEA of HAECs under disturbed flow and treated with control siRNA . ‘Rank Metric Score’ is the signal-to-noise ratio for each gene used to position the gene in the ranked list . ‘Rank in Gene List’ refers to the position of the gene in the ranked list of all genes present in the RNAseq dataset . The ‘Running Enrichment Score’ is the enrichment score for this set at this point in the ranked list of genes . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 01710 . 7554/eLife . 25217 . 018Figure 2—figure supplement 1 . HIF-1α under static , unidirectional flow , and disturbed flow . HAECs were treated with either UF or DF , or no flow for 8 , 24 , or 48 hr prior to cell lysis and Western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 01810 . 7554/eLife . 25217 . 019Figure 2—figure supplement 2 . Gene set enrichment analysis ( GSEA ) for the hypoxia and glycolysis gene sets in disturbed flow with siRNA targeted towards HIF-1α or control in HAECs . HAECs were treated with control siRNA or siRNA targeted towards HIF-1α for 24 hr prior to 48 hr of DF , followed by RNA isolation and total RNA sequencing ( RNAseq ) . Genome-wide expression analysis of the RNAseq data was performed using gene set enrichment analysis ( GSEA ) software version 2 . 2 . 4 available from the Broad Institute ( http://www . broadinstitute . org/gsea/downloads . jsp ) . The GSEA algorithm calculates an enrichment score reflecting the degree of overrepresentation at the top or bottom of the ranked list of the genes included in a gene set in a ranked list of all genes present in the RNAseq dataset . ( TOP ) A positive enrichment score ( ES ) indicates gene set enrichment at the top of the ranked list; a negative ES indicates gene set enrichment at the bottom of the ranked list . The final enrichment score for a set is the maximum deviation from zero encountered for that set . ( MIDDLE ) The location of the hypoxia or glycolytic genes in the ranked set of all genes . The ‘Rank Metric Score’ is the signal-to-noise ratio for each gene used to position the gene in the ranked list . ( BOTTOM ) The distribution of the rank metric score across all genes present in the expression dataset . The gene set collection used was the h . all . v5 . 1 . symbols . gmt [Hallmarks] gene sets database . The analysis demonstrates that known ( A ) hypoxia genes and ( B ) many glycolysis enzymes are positively enriched in the control siRNA sample . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 019 To determine the key biological pathways controlled by HIF-1α in endothelium under DF , HAECs were treated with either HIF-1α targeting siRNA or control siRNA and then were subjected to DF for 48 hr before performing RNA sequencing ( Wu et al . , 2017a ) . Transcriptome analyses identified 2989 DEGs regulated by HIF-1α in HAECs under DF ( FDR < 0 . 05 ) . GSEA ( Figure 2E ) and DAVID ( Figure 2F ) identified that HIF-1α-regulated DEGs are enriched in hypoxia and glycolysis , indicating that HIF-1α significantly contributes to the glycolytic gene signature detected in DF-treated HAECs . The DEGs are available at the publisher’s website ( Figure 2—source data 2 ) . The enrichment plot for hypoxia and glycolysis ( Figure 2—figure supplement 2A and B , respectively ) , as well as list of enriched genes under disturbed flow and HIF-1α knockdown contributing to the hypoxia and glycolytic gene sets are provided ( Figure 2—source data 3 ) . Genes known to be associated with hypoxia and glycolysis were positively enriched under control siRNA . We then sought to determine the mechanism ( s ) underlying DF-induced HIF-1α stabilization . Generation of reactive oxygen species ( ROS ) stabilizes HIF-1α in a wide range of cells ( Wheaton et al . , 2014; Weinberg et al . , 2010; Semenza , 2009 ) . HAECs exposed to DF had much higher ROS levels as measured by CellRox , a cytoplasmic ROS indicator ( Figure 3A , B ) . Treating cells with an antioxidant , EUK134 , a synthetic superoxide dismutase/catalase mimetic ( Rong et al . , 1999 ) , reduced HIF-1α levels in DF-treated HAECs ( Figure 3C ) . Treatment with antioxidant EUK134 had no effect on UF-treated HAECs ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 25217 . 020Figure 3 . Generation of ROS and NOX4 are required for disturbed flow-induced stabilization of HIF-1α . ( A ) HAECs were subjected to either UF or DF for 48 hr before staining for reactive oxygen species with CellROX orange dye ( 5 μM ) . ( B ) The cells were segmented using the fluorescence channel and the background intensity subtracted before calculating the average fluorescent intensity per cell ( n = 50 per condition ) . ( C ) HAECs were subjected to DF for 24 hr and simultaneously treated with EUK134 or DMSO . The cells were then lysed and Western blotted for HIF-1α and β-actin . Treatment with EUK134 under DF reduces HIF-1α ( n = 4 ) . ( D ) After 24 hr of either UF or DF , cell lysates were collected for Western blot for NOX4 ( n = 4 ) . HAECs were treated with siRNA targeted towards NOX4 ( siNOX4 ) or non-targeting control ( SC ) prior to 48 hr of DF followed by either ( E ) staining for ROS ( CellRox , 5 μM ) , quantified in ( F ) ( n = 50 ) , or cell lysis and Western blotting ( G ) . NOX4 knockdown reduces HIF-1α under DF ( n = 4 ) . p<0 . 05; p<0 . 005; p<0 . 0005 as determined by Student’s t-test . Data represent mean ± SEM . Bar is 10 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 02010 . 7554/eLife . 25217 . 021Figure 3—figure supplement 1 . EUK134 has no effect on HIF-1α under UF . HAECs were subjected to UF or DF for 24 hr , with either DMSO or EUK134 at 1 mM . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 02110 . 7554/eLife . 25217 . 022Figure 3—figure supplement 2 . HIF-1α and NOX4 kinetics under DF . ( A ) HAECs were subjected to UF for 24 hr prior to switching to DF for the indicated times . HIF-1α reaches maximal induction at 4 hr before stabilizing at a slightly lower level . NOX4 increases steadily with increasing duration of disturbed flow . ( B ) HAECs were treated with either non-targeting siRNA ( - ) or siRNA targeted towards NOX4 ( + ) for 24 hr before subjecting to DF for 4 hr . Reducing NOX4 at early time points , even when NOX4 is not abundant , will reduce HIF-1α expression . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 022 The RNA-seq data revealed increased expression of NOX4 in HAECs under DF . As an ROS generating oxidase , NOX4 is a critical mediator of endothelial inflammation under DF ( Lassègue and Griendling , 2010; Hwang et al . , 2003; Schröder et al . , 2012 ) . Western blot analysis showed that exposure to DF increased the expression of NOX4 protein ( Figure 3D ) . Inhibition of NOX4 with siRNA reduced DF-induced ROS ( Figure 3E , F ) and HIF-1α levels ( Figure 3G ) in HAECs subjected to DF . To further explore the kinetics of HIF-1α accumulation and NOX4 induction , we conditioned HAECs with 24 hr of UF prior to decreasing durations of DF and measured the expression of these two proteins . As shown in Figure 3—figure supplement 2A , HIF-1α is maximally induced at 4 hr before stabilizing at the levels seen at 24 hr . NOX4 , on the other hand , steadily accumulates over time . In order to answer whether NOX4 levels are upstream of HIF-1α at these early time points , or if early HIF-1α induction is independent of NOX4 , we knocked down NOX4 with siRNA for 24 hr before subjecting these HAECs to decreasing amounts of DF . As can be seen in Figure 3—figure supplement 2B , even at early time points , prior to maximal amounts of NOX4 , knocking down NOX4 reduces HIF-1α . Thus , NOX4 activation by DF is required for initial induction of HIF-1α . The combinatorial analyses of the two above mentioned RNA-seq experiments identified a cohort of glycolytic activators that are increased by DF and induced by HIF-1α . Transcripts of 18 glycolytic enzymes were significantly increased ( FDR < 0 . 05 ) in HAECs under DF when compared to UF and 13 of these enzymes were significantly reduced when DF induced HIF-1α was inhibited by siRNAs . Shown in Figure 4A and B are the biological and technical replicates for the mRNA fold changes for these genes . Data for combined biological and technical replicates is available in Figure 2—source data 2 . Among these DF-activated , HIF-1α-dependent genes , we found that hexokinase-2 ( HK2 ) , and glucose transporter-1 ( SLC2A1/GLUT1 ) were the most differentially expressed at the transcription level by DF . Confirming the RNA-seq data , we found that disturbed flow increased the mRNA and protein expression of endothelial SLC2A1 and HK2 ( Figure 4C , D ) . Transfecting HAECs with HIF-1α-targeting siRNAs reduced or prevented DF-induced SLC2A1 and HK2 mRNA and protein expressions ( Figure 4E , F ) . We then determined whether HIF-1α expression is sufficient to induce SLC2A1 and HK2 . To this end , we mutated the specific proline residues that regulate HIF-1α stability ( Ke and Costa , 2006 ) to generate a stabilized HIF-1α mutant ( mHIF-1α ) . mHIF-1α transfected HAECs showed high expression of HIF-1α ( Figure 4—figure supplement 1 ) . Compared to control , mHIF-1α transfected HAECs had higher mRNA expression of SLC2A1 and HK2 ( Figure 4G ) . 10 . 7554/eLife . 25217 . 023Figure 4 . Disturbed flow-induced HIF-1α stabilization is required for the glycolytic phenotype . ( A ) Expression profile of all significantly regulated glycolytic enzymes in the RNAseq data set of UF vs DF ( flow-seq ) . Four biological replicates for each condition , and three averaged technical replicates . SLC2A1 and HK2 are the top two enzymes that are upregulated under DF . Relative expression is normalized to UF , last row . HAECs were treated with siRNA for HIF-1α ( siHIF-1α ) or non-targeting control ( SC ) for 24 hr before DF for an additional 48 hr . Cell lysates were then collected and sent for total RNA sequencing ( siHIF-1α/DF-seq ) . ( B ) Expression profile of all significantly regulated glycolytic enzymes in siHIF-1α/DF-seq . Three biological replicates for each condition , and two averaged technical replicates . Relative expression is normalized to SC , first row . ( C ) SLC2A1 and HK2 upregulated in flow-seq were confirmed with qPCR ( n = 4 ) . ( D ) Western blot and quantification of select glycolytic enzymes ( SLC2A1 and HK2 ) under differential flow after 24 hr ( n = 4 ) . HAECs were treated with siHIF-1α or SC for 24 hr before DF for an additional 48 hr before cell lysis and ( E ) qPCR for HK2 and SLC2A1 ( n = 13 ) or ( F ) Western blotting/quantification of HIF-1α , HK2 , SLC2A1 and β-actin ( n = 4 ) . ( G ) An overexpression vector containing HIF-1α with mutated known prolyl hydroxylase binding amino acids was expressed in HAECs using in vitro transcription for 6 hr before cell lysis and analysis of HIF-1α , SLC2A1 ( β-actin serves as a loading control ) and qRT-PCR analysis for glycolytic genes SLC2A1 and HK2 at 50 ng dose ( n = 3 ) . ( H ) HAECs were first treated with either SC or siHIF-1α before being subjected to DF for 48 hr . The cells then underwent a glycolysis stress test . Arrows indicate injection of glucose , oligomycin , and 2-deoxyglucose ( 2DG ) . ( I ) Glycolysis and glycolytic capacity obtained during glycolysis stress test are significantly downregulated by HIF-1α knockdown ( n = 10 ) . p<0 . 05; p<0 . 005; p<0 . 0005 as determined by Student’s t-test . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 02310 . 7554/eLife . 25217 . 024Figure 4—figure supplement 1 . Disturbed flow stabilization of HIF-1α precedes NOX4 abundance . An overexpression vector containing HIF-1α with mutated known prolyl hydroxylase binding amino acids was expressed in HAECs using in vitro transcription for 6 hr before cell lysis and analysis of HIF-1α , SLC2A1 ( β-actin serves as a loading control ) showing a dose-dependent effect . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 02410 . 7554/eLife . 25217 . 025Figure 4—figure supplement 2 . Disturbed flow stabilization of EPAS1 does not contribute to glycolytic gene transcription . ( A ) HAECs were subjected to 24 hr of either UF or DF prior to cell lysis and Western blotting . Both HIF-1α and EPAS1 are increased under DF . HAECs were subjected to DF for 24 hr after 24 hr of treatment with either control siRNA or siRNA targeted towards EPAS1 prior to cell lysis , followed by ( B ) Western blotting or ( C ) RNA isolation and qRT-PCR . ( B ) The siRNA is able to reduce EPAS1 under disturbed flow . ( C ) SLC2A1 and HK2 are unchanged following siEPAS1 treatment and DF . EPAS1 levels are significantly lower ( n = 6 ) . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 025 To determine whether HIF-1α is required for DF-induced increase in glycolysis , we performed a glycolysis stress test in HAECs under DF in combination with inhibition of HIF-1α by siRNA . HIF-1α knockdown significantly reduced the basal glycolysis and glycolytic capacity in HAECs subjected to DF for 48 hr ( Figure 3I , J ) . To assess whether EPAS1 ( HIF-2α ) contributes to the glycolytic phenotype under DF , as EPAS1 is also predicted to be a transcriptional regulator from IPA ( Figure 2A ) , we first verified that EPAS1 was increased under DF ( Figure 4—figure supplement 2A ) , before treating HAECs with siRNA targeted towards EPAS1 ( siEPAS1 ) ( Figure 4—figure supplement 2B ) ; however , there was no statistically significant reduction in glycolytic genes HK2 and SLC2A1 ( Figure 4—figure supplement 2C ) as the result of EPAS1 inhibition in cells under DF . Together , these results demonstrate that HIF-1α but not EPAS1 ( HIF-2α ) is required for the DF-induced increase in two of the most highly transcriptionally regulated glycolytic enzymes . We next sought to determine the contribution of HIF-1α in DF-induced reduction in mitochondrial respiration ( Figure 1H , I ) . We performed a mitochondrial stress test in HAECs treated with control or HIF-1α-targeting siRNAs and then were subjected to DF for 48 hr . HIF-1α knockdown reversed the DF-induced reduction in maximal respiratory capacity ( Figure 5A ) , and increased the basal oxygen consumption and reserve respiratory capacity ( Figure 5B ) . The amount of maximal respiration recovered via HIF-1α knockdown is substantial , suggesting that HIF-1α reduces respiratory capacity in HAECs under DF . 10 . 7554/eLife . 25217 . 026Figure 5 . Disturbed flow inhibits mitochondrial respiration via HIF-1α . ( A ) HAECs were subjected to either DF for 48 hr following 24 hr treatment with either non-targeting siRNA ( SC ) or HIF-1a targeted siRNA ( siHIF-1α ) , followed by a mitochondrial stress test . ( B ) Respiratory parameters ( A ) are all significantly higher under DF cells treated with siHIF-1α ( n = 8 ) . ( C ) HAECs were subjected to either UF or DF for 48 hr . Glucose oxidation was measured during sequential treatment with UK5099 and BPTES/Etoxomir . Under DF , HAECs use glucose as a larger fraction of the total mitochondrial oxidation ( n = 4 ) . HAECs were subjected either 24 hr UF or DF . Western blot ( D ) and qRT-PCR ( E ) shows increased PDK1 expression under DF ( n = 4 for both ) . ( F ) HAECs were subjected to DF for 48 hr following 24 hr treatment with either SC or siHIF-1α . Cell lysates demonstrate reduced PDK1 under siHIF-1α ( n = 4 ) . ( G ) HAECs were subjected to 48 hr of either UF or DF before undergoing a PDH activity dipstick assay . DF reduces PDH activity ( n = 4 ) . ( H ) Under combined siHIF-1α treatment and DF , there is increased PDH activity ( n = 4 ) . ( I ) Under combined siPDK1 treatment and DF , there is increased PDH activity ( n = 4 ) . ( J ) HAECs are treated with either DMSO or DCA ( 4 mM ) and simultaneous DF for 48 hr before mitochondrial stress test . Respiratory parameters are shown in K . There is a signficant increase in maximal respiration and respiratory capacity with DCA treatment ( n = 4 ) . ( L ) HAECs were treated with SC or siPDK1 for 24 hr before DF for 48 hr and subsequent mitochondrial stress test . Respiratory parameters are shown in M ( n = 8 ) . p<0 . 05; p<0 . 005; *p<0 . 0005 as determined by Student’s t-test . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 02610 . 7554/eLife . 25217 . 027Figure 5—figure supplement 1 . Mitochondrial biogenesis is the same in differential flow . HAECs were subjected to either DF or UF for 48 hr . ( A ) After flow , the cells were washed and incubated in mitotracker for 30 min before imaging ( n = 50 per condition ) . ( B ) After flow , the cells were lysed and RNA harvested before performing qRT-PCR for mitochondrial genes ND1 and ND5 ( n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 02710 . 7554/eLife . 25217 . 028Figure 5—figure supplement 2 . Differential flow regulates PDK1 . HAECs were subjected to either UF or DF for 24 hr prior to cell lysis , RNA purification , and qRT-PCR for PDK1 ( n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 02810 . 7554/eLife . 25217 . 029Figure 5—figure supplement 3 . siRNA targeted against PDK1 . HAECs were treated with either control siRNA ( - ) or siRNA targeted against PDK1 ( + ) for 24 hr prior to disturbed flow for 48 hr , followed by cell lysis and Western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 029 We then determined whether changes in mitochondrial mass contribute to the reduced endothelial respiratory capacity under DF . In HAECs subjected to either DF or UF for 48 hr , no difference was noted in mitochondria density by MitoTracker staining , a surrogate for mitochondria mass ( Figure 5—figure supplement 1A ) ( Agnello et al . , 2008 ) . Furthermore , there was also no difference in transcript levels of mitochondrially encoded genes such as NADH-ubiquinone oxidoreductase chain 1 ( ND-1 ) and 5 ( ND-5 ) under athero-relevant flows ( Figure 5—figure supplement 1B ) . Substrate limitations or substrate import into mitochondria could account for the decreased mitochondrial respiration in HAEC under DF . To delineate these possibilities , oxygen consumption rates ( OCR ) were monitored in HAECs while inhibiting pyruvate entry into mitochondria with UK5099 , which inhibits the mitochondrial pyruvate carrier , glutamine entry into mitochondria via BPTES , and fatty acid entry into mitochondria via etoxomir . As demonstrated in Figure 5C , glucose oxidation ( as estimated by UK5099 treatment ) contributes to 43% of the total oxygen consumption rate under UF but confers 62% of total OCR under DF . These findings suggest that decreased pyruvate availability or pyruvate dehydrogenase activity might therefore limit the maximal OCR under UF . We then sought to determine how DF affects pyruvate availability and pyruvate dehydrogenase ( PDH ) , which converts pyruvate into acetyl-CoA . Activity of PDH is tightly regulated by pyruvate dehydrogenase kinase-1 ( PDK1 ) , which inhibits PDH . Among the 4 PDK genes ( PDK1-4 ) expressed in HAECs , PDK1 was the only one that is upregulated by DF in the RNA-seq data . Consistent with these data , DF increased PDK1 mRNA ( Figure 5—figure supplement 2 ) and protein expression ( Figure 5D ) . Loss of HIF-1α in HAECs attenuated DF-induced upregulation of PDK1 mRNA ( Figure 5E ) and protein ( Figure 5F ) . To determine whether the DF-induced increase in PDK1 affects PDH activity , we measured PDH activity ( Hong et al . , 2013 ) . Consistent with the increased PDK1 expression , DF increased PDH activity compared to UF ( Figure 5G ) . Studies in HAECs with HIF-1α knockdown and PDK1 knockdown showed partial ( Figure 5H ) and complete reversal ( Figure 5I ) of the DF-induced reduction of PDH activity , respectively . To test the putative role of PDK1 in inhibiting endothelial mitochondrial respiration , we subjected HAECs to DF in the presence or absence of dichloroacetate , a known inhibitor of PDK1 ( Xie et al . , 2011 ) , and performed a mitochondrial stress test . PDK1 inhibition partially rescued the reduced maximal respiration and respiratory capacity induced by DF ( Figure 5J–K ) . Similar results were obtained with genetic inhibition of PDK1 expression by siRNA ( Figure 5L–M , Figure 5—figure supplement 3 ) . Collectively , these findings suggest that under DF , HIF-1α promotes PDK1 expression to reduce the entry of glucose-derived pyruvate into the TCA cycle , and consequently limits mitochondrial reserve function . To determine how DF-induced changes in cellular metabolism affect inflammatory gene expression in endothelial cells , we measured the expression of pro-inflammatory cytokines in DF-activated HAECs . Exposure of HAECs to DF for 48 hr resulted in increased expression of pro-inflammatory genes including VCAM-1 , IL8 and CCL2 compared to UF ( Figure 6A ) . Inhibition of HIF-1α by siRNA reduced DF-induced expression of IL8 , VCAM-1 , and CCL2 ( Figure 6B ) . Knocking down HIF-1α under disturbed flow also reduced the binding of monocytes ( THP-1 ) to the HAECs ( Figure 6—figure supplement 1 ) . Conversely , ectopic expression of stabilized HIF-1α , like DF , increased IL8 , VCAM-1 , and CCL2 expression in HAECs ( Figure 6C ) . Moreover , siRNA-mediated NOX4 inhibition , which prevented the DF-induced HIF-1α , reduced the expression of pro-inflammatory genes in DF-treated HAECs ( Figure 6D ) . 10 . 7554/eLife . 25217 . 030Figure 6 . Disturbed flow-induced metabolic reprogramming is required for expression of inflammatory markers . ( A ) HAECs were treated with either UF or DF for 24 hr prior to lysis , RNA purification , and qRT-PCR for IL8 , VCAM1 , and CCL2 . IL8 , VCAM1 , and CCL2 are all increased under DF ( n = 4 ) . ( B ) HAECs were treated with either non-targeting siRNA ( SC ) or siRNA targeting HIF-1α ( siHIF-1α ) for 24 hr before subjected to DF for 48 hr prior to lysis , RNA purification , and qRT-PCR for IL8 , VCAM1 , and CCL2 . IL8 , VCAM1 , and CCL2 are all reduced under HIF-1α knockdown ( n = 4 ) . ( C ) HAECs were transfected with either a blank or with stabilized HIF-1α mRNA transcript for 6 hr prior to lysis , RNA purification , and qRT-PCR for IL8 , VCAM1 , and CCL2 . IL8 , VCAM1 , and CCL2 are all increased under HIF-1α transcription delivery ( n = 4 ) . ( D ) HAECs were treated with either non-targeting siRNA ( SC ) or siRNA targeting NOX4 ( siNOX4 ) for 24 hr before subjected to DF for 48 hr prior to lysis , RNA purification , and qRT-PCR for IL8 , VCAM1 , and CCL2 . IL8 , VCAM1 , and CCL2 are all reduced under NOX4 knockdown ( n = 4 ) . ( E ) HAECs were treated with either non-targeting siRNA ( SC ) or siRNA targeting SLC2A1 ( siSLC2A1 ) for 24 hr before subjected to DF for 48 hr prior to lysis , RNA purification , and qRT-PCR for IL8 , VCAM1 , and CCL2 . IL8 , VCAM1 , and CCL2 are all reduced under SLC2A1 knockdown ( n = 4 ) . ( F ) HAECs were treated with either non-targeting siRNA ( SC ) or siRNA targeting PDK1 ( siPDK1 ) for 24 hr before subjected to DF for 48 hr prior to lysis , RNA purification , and qRT-PCR for IL8 , VCAM1 , and CCL2 . IL8 , VCAM1 , and CCL2 are all reduced under PDK1 knockdown . ( G ) HAECs were treated with either UF or DF for 24 hr prior to lysis and Western Blotting ( n = 4 ) . ( H ) siHIF-1α treated HAECs were subjected to DF for 48 hr prior to lysis . siHIF-1α significantly reduces phospho-p65 ( n = 4 ) . ( I ) HAECs were treated with 10 μM of NBD for 1 hr prior to HIF-1α overexpression for 6 hr prior to RNA purification and qRT-PCR ( n = 4 ) . p<0 . 05; p<0 . 005; p<0 . 0005 as determined by Student’s t-test . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 03010 . 7554/eLife . 25217 . 031Figure 6—figure supplement 1 . HIF-1α knockdown reduces leukocyte adhesion . HAECs were subjected to DF for 48 hr after either control siRNA ( SC ) or siRNA targeted towards HIF-1α treatment for 24 hr . After removal from DF , 5-fold THP-1 cells loaded with 5 µM calcein AM dye for were incubated with the HAECs for 1 hr . The samples were subsequently washed three times with PBS prior to ( A ) fluorescence counting . ( B ) Biological replicates and fluorescence intensities of plates are shown ( n = 3 ) . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 03110 . 7554/eLife . 25217 . 032Figure 6—figure supplement 2 . SLC2A1 controls glycolysis in HAECs . Control siRNA or siRNA targeted against SLC2A1 ( 50 nM ) were incubated with HAECs for 72 hr prior to cell lysis , ( A ) RNA purification and qRT-PCR for SLC2A1 ( n = 4 ) or ( B ) Western blotting for SLC2A1 . HAECs were treated with either control siRNA or siRNA targeted against SLC2A1 ( 50 nM ) for 24 hr prior to disturbed flow treatment for 48 hr and ( C ) glycolysis stress test and quantification of glycolysis stress test parameters ( D ) ( n = 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 03210 . 7554/eLife . 25217 . 033Figure 6—figure supplement 3 . Mitochondrial inhibition increases inflammation . HAECs were treated with DMSO or rotenone ( 1 µM ) and antimycin ( 1 µM ) and subjected to UF for 24 hr prior to cell lysis , RNA purification , and qRT-PCR ( n = 4 ) . p<0 . 05; p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 03310 . 7554/eLife . 25217 . 034Figure 6—figure supplement 4 . HIF-1α is upstream of NF-κB-induced inflammatory gene transcription . HAECs were subjected to DF for 24 hr and simultaneously treated with DMSO or NBD prior to cell lysis and Western blotting . There is no reduction in HIF-1α by inhibiting the NF-κB pathway with either of the two compounds . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 03410 . 7554/eLife . 25217 . 035Figure 6—figure supplement 5 . Disturbed flow stabilization of HIF-1α reduces KLF2 . HAECs were treated with either control siRNA or siRNA targeting HIF-1α for 24 hr prior to DF for 48 hr , followed by cell lysis , RNA purification , and qRT-PCR ( n = 4 ) . ( A ) KLF2 is increased under siHIF-1α and disturbed flow . ( B ) KLF4 is not significant ( n = 4 , p=0 . 057 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 035 We then investigated whether modulation of metabolism influences endothelial inflammatory gene expression . To inhibit DF-induced glycolysis , we treated HAECs with siRNA against SLC2A1 ( siSLC2A1 ) , which was the most highly upregulated glycolytic gene under DF . SLC2A1 knockdown cells had reduced SLC2A1 mRNA and protein levels , and exhibited reduced ECAR when compared with control knockdown cells ( Figure 6—figure supplement 2 ) . Inhibition of SLC2A1 ( or glycolysis ) reduced DF-induced VCAM-1 and CCL2 gene expression ( Figure 6E ) . To rescue the DF-induced reduction in mitochondrial oxidative phosphorylation , we treated HAECs with siRNA against PDK1 . Inhibition of PDK1 also significantly reduced inflammatory gene expression ( Figure 6F ) . To further explore whether the inhibition of oxidative phosphorylation by itself would further increase inflammation , we treated HAECs with a combination of rotenone and antimycin which together halt electron flow in the mitochondria ( Chen et al . , 2003 ) . We found that inhibition of the electron transport chain with rotenone and antimycin increased IL8 , VCAM1 , and CCL2 in UF ( Figure 6—figure supplement 3 ) . These results are consistent with siHIF-1α and siPDK1 improving mitochondrial capacity , as well as reducing inflammation , under DF . In summary , these findings suggest that DF-induced metabolic reprogramming; namely induction of glycolysis and inhibition of mitochondrial oxidation , is required for the DF-induced pro-inflammatory phenotype . NF-κB has been reported to be an important regulator of endothelial inflammation under DF ( Hajra et al . , 2000 ) . Indeed , we found that DF causes increased phosphorylation of the p65 subunit of NF-κB ( Figure 6G ) . Knock down HIF-1α reduced DF-induced NF-κB activation ( Figure 6H ) . To further confirm the role of HIF-1α in NF-κB activation , we treated HAECs with NEMO-binding domain peptide ( NBD ) , which blocks association of NEMO with IκB kinase that phosphorylates IκB , thereby preventing NF-κB activation ( May et al . , 2002 ) . We then over-expressed HIF-1α ( mHIF-1α ) in these NBD-treated HAECs . Pretreatment of NBD prevented the HIF-1α-induced increase in inflammation ( Figure 6I ) . We next examined whether NF-κB activation was required for HIF-1α stabilization , as has been suggested by others ( van Uden et al . , 2008 , 2011 ) . Pretreatment of HAECs with NBD prior to 24 hr of DF did not affect DF-induced increase in HIF-1α levels , suggesting that NF-κB activation is not responsible for HIF-1α stabilization in HAECs under DF ( Figure 6—figure supplement 4 ) . DF is also associated with decreased KLF2 expression compared to UF , and KLF2 is a known inhibitor of NF-κB activation ( SenBanerjee et al . , 2004 ) . In our RNAseq dataset , KLF2 was downregulated 11-fold under DF when compared to UF ( FDR < 0 . 000345 ) . Furthermore , KLF2 is known to inhibit HIF-1α under hypoxic conditions ( Kawanami et al . , 2009 ) . We thus asked whether HIF-1α induction decreases KLF2 levels under DF . As shown in Figure 6—figure supplement 5A , knockdown of HIF-1α under DF increased KLF2 levels ( but not KLF4 ) ( Figure 6—figure supplement 5B ) . Of note , in the RNAseq DEG dataset of siHIF-1α under DF , both KLF2 and KLF4 were significantly upregulated , compared to scrambled control . Together , these results suggest that DF-induced metabolic changes driven by HIF-1α cause increased inflammation through an NF-κB-dependent mechanism . To investigate in vivo relevance of DF-mediated regulation of endothelial metabolism , we measured the expression of glycolytic enzymes and ROS in endothelial cells freshly isolated from different regions of porcine aorta of known susceptibilities to atherosclerosis . The inner curvature of the aortic arch is constantly exposed to DF , and prone to atherosclerosis while the nearby descending thoracic aorta is subjected to UF , and resistant to atherosclerosis ( Figure 7A ) . En face images showed that endothelia located in aortic arch exhibited cobblestone morphology and greater amounts of ROS ( measured by dihydroethidium bromide ) ( Figure 7B ) compared to elongated endothelia in the nearby descending aorta ( Figure 7C ) . Endothelial cells ( 95% purity ) were scraped immediately after the swine were sacrificed for Western Blot analyses ( Wu et al . , 2015; Fang et al . , 2010 ) . As shown in Figure 7D and E , we detected significantly increased protein expression of NOX4 , HIF-1α and glycolytic enzymes SLC2A1 , HK2 , and PDK1 , as well as phosho-p65 , in endothelia isolated form the inner curvature of swine aortic arch as compared to those isolated from descending thoracic aorta in the same animal . Similarly , using immunofluorescence , we confirmed that NOX4 ( Figure 7—figure supplement 1A , B ) and HIF-1α were increased in parts of murine aortas , after live perfusion with fixative followed by immunofluorescence ( Figure 7—figure supplement 1C , D ) . These in vivo results are consistent with the in vitro studies demonstrating that disturbed flow promotes endothelial glycolytic metabolism and reduced mitochondrial function by increasing NOX4-mediated ROS production and consequent HIF-1α stabilization . 10 . 7554/eLife . 25217 . 036Figure 7 . Athero-susceptible regions of aorta express higher levels of ROS and HIF-1α . ( A ) Aortas were harvested from pigs due for slaughter in less than 10 min after sacrifice . The inner curvature ( B ) of the aortic arch ( AA ) or descending thoracic ( DT ) aorta ( C ) was cut out and immediately stained with dihydroethidium bromide ( DHE ) before fixation in cold 4% paraformaldehyde . The sections were then permeabilized and stained with lectin . Other aortas were dissected out and immediately washed with PBS before a #10 scalped was passed along the inner curve of the aorta or along the descending thoracic aorta . Endothelial cells were immediately stored in cold lysis buffer . ( D ) Western blotting for NOX4 , HIF-1α , phospho-p65 , HK2 , SLC2A1 , PDK1 , and β-actin of AA and DT samples . ( E ) The Western blots are quantified for NOX4 , HIF-1α , phospho-p65 , HK2 , SLC2A1 , and PDK1 ( n = 4 ) . The AA region of the pig aortas have significantly more expression of all these enzymes . p<0 . 05; p<0 . 005 as determined by Student’s t-test . Data represent mean ± SEM . Bar is 10 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 03610 . 7554/eLife . 25217 . 037Figure 7—figure supplement 1 . NOX4 and HIF-1α immunofluorescence in descending thoracic aorta ( DT ) and aortic arch ( AA ) . Mice were anesthetized and perfused with fixative prior to immunofluorescence staining . ( A ) NOX4 is has higher fluorescence in the AA than DT . ( B ) Quantification of fluorescence 26–50 cells per region per mouse . ( C ) HIF-1α has higher nuclear fluorescence in AA than DT . ( D ) Quantification of HIF-1α nuclear fluorescence 37–64 cells per region per mouse . Scale bar: 10 μm . p<0 . 005 , ***p<0 . 0005 by non-parametric Kolmogorov-Smirnov test . DOI: http://dx . doi . org/10 . 7554/eLife . 25217 . 037 We discovered a previously uncharacterized role for HIF-1α in metabolic reprogramming and activation of vascular endothelium under disturbed flow , which simulates hemodynamics associated with atherosclerosis . Although flow-dependent metabolic plasticity has been implicated in endothelial pathophysiology ( Doddaballapur et al . , 2015; Cucullo et al . , 2011 ) , the underlying molecular mechanisms are poorly understood . Here , we evaluated the DF-induced regulation of endothelial metabolism by a combination of whole-genome RNA sequencing , pathway predictions , bioenergetics measurements , and molecular/biochemical analyses . We found that DF increases endothelial glycolysis and reduces mitochondrial respiratory capacity by promoting NOX4-dependent ROS-mediated HIF-1α stabilization . In agreement with the in vitro investigations , NOX4 , ROS , HIF-1α , and key glycolytic activators were also increased in vivo in endothelial cells isolated from the inner curvatures of porcine aortic arch , which represents an area where cells are exposed to DF when compared to nearby descending thoracic aorta where cells are subjected to UF . These findings demonstrate that a metabolic switch comprising HIF-1α-dependent reduction in respiratory capacity and increased glycolysis as important molecular signatures that characterize the athero-susceptible endothelium associated with DF . Our RNA-seq analyses predicted that KLF2 and HIF-1α are the major transcriptional regulators that control the endothelial phenotypes related to athero-relevant flows . Subsequent mechanistic investigations further demonstrate that DF-induced expression of HIF-1α promotes endothelial glycolytic metabolism and reduces mitochondrial oxidative phosphorylation . Our data are consistent with reports on glycolysis playing a central role in the promotion of angiogenesis ( De Bock et al . , 2013b; Potente et al . , 2011 ) . However , these previous studies did not apply differential flow , which has long been known to result in different EC phenotypes . Our data are also consistent with reports of mitochondria acting as signaling organelles ( Quintero et al . , 2006 ) ; the acute reduction in mitochondrial oxidation under DF may act as a shunt to oxygenate tissues further away from the vasculature . Hypoxia-inducible factors ( HIFs ) function as master regulators of oxygen homeostasis , and HIF-1α and HIF-2α are the most widely studied members of the HIF family ( Semenza , 2012; Prabhakar and Semenza , 2012 ) . While HIF-1α is expressed in all cells of all metazoan species , HIF-2α is only expressed in certain cell types of vertebrate species including endothelial cells . Unlike HIF-2α , HIF-1α controls the expression of multiple gene products that mediate glycolysis ( Prabhakar and Semenza , 2012 ) . For instance , HIF-1α activates transcription of PDK1 , which phosphorylates and inactivates the catalytic subunit of PDH , the enzyme that converts glucose-derived pyruvate into acetyl-CoA for entry into the mitochondrial tricarboxylic acid ( TCA ) cycle ( Kim et al . , 2006 ) . Inhibition of PDK1 only partially restored the mitochondrial reserve capacity in HAECs under DF compared to HIF-1α knockdown , which almost completely rescued mitochondrial reserve capacity to the level detected under UF . These findings indicate that besides PDK1 , HIF-1α exerts its effects on mitochondrial oxidative phosphorylation via additional signaling pathways . One possibility is the flow-mediated regulation of Lactate Dehydrogenase A ( LDHA ) , a key enzyme that catalyzes the conversion of pyruvate to lactate in the final step of anaerobic glycolysis . Indeed , our RNA-seq analyses demonstrated up-regulation of LDHA expression by DF; this effect was suppressed by HIF-1α inhibition . Increased LDHA is predicted to reduce the availability of pyruvate to enter the TCA cycle and hence limit the substrate availability for oxidative phosphorylation . Our results identified NOX4-dependent ROS as a major contributor to promote DF-induced HIF-1α stabilization . The data are consistent with a previous study showing that oscillatory flow increases NOX4 expression and ROS production in cultured bovine aortic endothelial cells ( Hwang et al . , 2003 ) . In agreement with these in vitro investigations , we detected increased NOX4 expression and ROS generation , accompanied by elevated HIF-1α , in vivo in endothelia isolated from the athero-susceptible swine aortic arch . ROS are known to stabilize HIF-1α even in normoxic conditions ( Bonello et al . , 2007 ) . HIF-1α stabilization has also been shown to occur in stretch-based assays by stretch-induced inhibition of succinate dehydrogenase , and is thought to mediate epithelial protection during ventilator-induced lung injury ( Eckle et al . , 2013 , 2014 ) . While it is possible that accumulation of mitochondrial TCA substrates contributes to HIF-1α stabilization under DF , flow by itself is unlikely to cause mechanical stretch in either the in vitro assays or in the aorta in vivo , as UF is more likely than DF to cause stretch ( UF has 10-fold higher shear stress compared to DF ) . HIF-1α mRNA is also thought to increase after prolonged stretch in rat skeletal muscle capillary beds ( Milkiewicz et al . , 2007; Milkiewicz and Haas , 2005 ) and the inferior vena cava ( Lim et al . , 2011 ) . However , we found no transcriptional differences between UF and DF ( Figure 2C ) . Another major finding of the present study is the demonstration that DF-induced endothelial metabolic reprogramming contributes to increased inflammatory gene expression . DF-induced expression of pro-inflammatory genes was markedly attenuated with the knockdown of SLC2A1 , PDK1 , or HIF-1α ( loss-of-function ) , and increased by HIF-1α overexpression ( gain-of-function ) . Corroborating our in vitro results , we detected increased HIF-1α , SLC2A1 , and PDK1 expression in vivo in swine aortic arch ECs . These results are in agreement with the reported increase in the expression of HIF-1α in carotid artery plaques ( Vink et al . , 2007; Sluimer et al . , 2008 ) . Aortic arch ECs have exposed to DF in vivo also been shown to exhibit NF-κB activation ( Hajra et al . , 2000; Fang et al . , 2010 ) . Although the underlying molecular mechanisms remain to be elucidated , one possibility is that increased glycolysis promotes endothelial inflammation via lactate-induced activation of NF-κB signaling in a manner described in immune cells such as macrophages ( Végran et al . , 2011; Covarrubias et al . , 2015 ) . However , it is important to point out that we cannot rule out the possibility that HIF-1α promotes vascular inflammation by additional pathways that are independent of metabolic changes . For example , our RNA-seq analyses demonstrated that inhibition of HIF-1α in DF-treated ECs rescued the expression of anti-inflammatory molecules KLF2 , KLF4 , and PPAP2B ( a coronary arterial disease candidate gene identified by genome-wide association studies ) ( SenBanerjee et al . , 2004; Wu et al . , 2015 ) . While we did not study the role of KLF2 , Doddaballapur et al . showed that KLF2 is required for UF-induced reduction in the expression of glycolytic genes in human umbilical vein endothelial cells ( Doddaballapur et al . , 2015 ) . Importantly , Inflammatory stimuli such as TNFα can by itself reduce the UF-responsiveness of KLF2 ( Huang et al . , 2017 ) ; it remains to be seen if metabolic changes might also play a role under such stimuli . Thus , differential flow waveforms may amplify EC metabolic phenotypes ( and subsequent pro- or anti-inflammatory states ) by providing transcriptional mechanisms to increase glycolysis and reduce respiration under DF and to reduce glycolysis and increase respiration under UF . Lesion-targeted therapies may fill the current treatment gap related to systemic risk factor management ( Kuo et al . , 2014 ) . Our in vitro and in vivo results suggest spatial inhibition of glycolytic metabolism and HIF-1α signaling may be a suitable approach for future arterial wall-based therapies that target disturbed flow-activated endothelium associated with focal atherosclerosis . Consistent with the results demonstrating HIF-1α inhibition reduces the disturbed flow-induced inflammation , endothelium-deletion of HIF-1α was recently shown to significantly ameliorate atherosclerotic burden in apoe−/− mice ( Akhtar et al . , 2015 ) . In summary , while studies suggest inhibition of glycolysis as a potential therapy to reduce angiogenesis ( Schoors et al . , 2014 ) , we propose that there may be a role in promoting the athero-protective flow phenotype in order to reverse EC activation such as increased vascular permeability and inflammation . Further investigations in metabolic effects of athero-protective flow might provide useful insight into the pathways that may be targeted to promote mitochondrial health and reduce athero-susceptibility . Here , we described that metabolic plasticity of endothelium is dynamically and reversibly regulated by hemodynamics suggesting a critical role for mechanical forces in actively mediating metabolic changes that drive endothelial functions in health and disease . Human aortic endothelial cells ( HAECs ) were purchased from Lonza ( Allendale , NJ ) ( CC-2535 , lot number 0000305906 or 0000297640 or 0000336393 ) . The primary cells were isolated from donated human tissue ( male , age 27 , female , 54 , and male , age 22 , respectively ) . The cells were tested for mycoplasma . The cells were verified to be endothelial cells through cell morphology ( alignment with flow ) and presence of CD31 . There was no alpha actin expression . For experiments , cells were grown in EGM-2 supplemented with SingleQuots from Lonza ( CC-3156 and CC-4176 ) and Antibiotic-Antimycotic from Gibco ( Grand Island , NY ) ( 15240062 ) . For flow experiments , cells were plated in 6-well plates at 4 × 105 cells/well , and after 24 hr , dextran ( Sigma-Aldrich , St . Louis , MO , 31392 ) was added to media to final concentration of 4% . All primary cultures were used from passage 6 to 10 . THP-1 ( human monocyte ) cells were a gift from the Sperling lab at the University of Chicago . Cells were grown in RPMI media ( ThermoFisher , Waltham , MA ) and 10% FBS ( Biowest USA , Riverside , MO ) . A flow device consisting of a computerized stepper motor UMD-17 ( Arcus Technology , Livermore , CA ) and a 1° tapered stainless steel cone was used to generate the physiologically-relevant shear stress pattern ( Wu et al . , 2015 ) . The flow device was placed in a 37°C incubator with 5% CO2 . HAECs at 100% confluence , maintained in EGM2- medium containing 4% dextran in 6-well plates , were subjected to unidirectional steady flow ( UF ) or disturbed flow ( DF ) for 24–72 hr before cells being harvested ( Dai et al . , 2004 ) . Static cells used the same dextran-containing media above and did not utilize the flow devices . Cells were lysed with buffer containing 8M deionized urea , 1% SDS , 10% glycerol , 60 mM Tris-HCl , 5% betamercaptoethanol ( all chemicals from Sigma-Aldrich ) . The lysates are passed through an insulin syringe three times and resolved on 4–12% Bis-Tris gels ( Invitrogen ) and transferred to a 0 . 2 μm PVDF membrane before blocking in either 5% non-fat milk in Tris-buffered saline and 0 . 1% Tween-20 ( TBST ) or 5% bovine serum albumin in TBST . Blots were then incubated in primary antibodies as described below . RNA as isolated form cells using Trizol and the Zymo Direct-zol RNA MiniPrep kit and reverse transcribed using High-Capacity cDNA Reverse Transcription Kit ( ThermoFisher ) . Quantitative mRNA expression was determined by real-time RT-PCR using SYBR Green MasterMix ( Roche , Indianapolis , IN ) . The following primer sequences were used ( IDT , Coralville , IA ) : SLC2A1- 5’-GAACTCTTCAGCCAGGGTCC-3’ , 5’-ACCACACAGTTGCTCCACAT-3’ HK2- 5’-GCTTGGAGCCACCACTCACCC-3’ , 5’-AGCCAGGAACTCTCCGTGTTCTGT-3’ IL8 , 5’-TGTGCCTTGGTTTCTCCTTT-3’ 5’-GCTTCCACATGTCCTCACAA-3’ CCL2 , 5’-AGCAGCAAGTGTCCCAAAGA-3’ 5’-TTGGGTTTGCTTGTCCAGGT-3’ VCAM1 , 5’-GATACAACCGTCTTGGTCAG-3’ 5’-TAATTCCTTCACATAAATAAACCC-3’ ND1 5’-AGCCCTACTCCACTCAAGCA-3’ 5’-GCTGCGAACAGAGTGGTGAT-3’ ND5 5’-GCCCTACTCCACTCAAGCAC-3’ 5’-TGAAGAAGGCGTGGGTACAG-3’ HIF-1α 5'-GGCGCGAACGACAAGAAAAA-3' 5'-GTGGCAACTGATGAGCAAGC-3' PDK1 5’- CCAGGACAGACAATACAAGTGGT-3’ 5’ GAATCGGGGGATAAACGCCT-3’ EPAS1 5’-GCGCACCTCGGACCTTCA-3’ 5’- TCTCCGAGCTACTCCTTTTCTTC-3’ EGLN3 5’-CACAGCGAGGGAATGAACCT-3’ 5’-TCCTGCTGTTAAGGCTTCCG-3’ VEGFA 5’-CTCTACCTCCACCATGCCAA-3’ 5’-GCATGGTGATGTTGGACTCC-3’ KLF2 5’-GAACCCATCCTGCCGTCCTT-3’ 5’-CACGCTGTTGAGGTCGTCG-3’ KLF4 5’-ATCTCAAGGCACACCTGCG-3’ 5’- CCTGGTCAGTTCATCTGAGCG-3’ GAPDH 5’- TGCACCACCAACTGCTTAGC-3’ 5’- GGCATGGACTGTGGTCATGAG-3’ ACTB 5’- TCCCTGGAGAAGAGCTACGA-3’ 5’- AGGAAGGAAGGCTGGAAGAG-3’ UBB 5’- ATTTAGGGGCGGTTGGCTTT-3’ 5’- TGCATTTTGACCTGTTAGCGG-3’ Glycolytic rates were measured using the XFe24 Extracellular Flux Analyzer ( Seahorse Bioscience , North Billerica , MA ) . After siRNA , chemical inhibitors , or flow experiments , HAECs were collected and seeded at a density of 4 × 104/well on Seahorse plates and allowed to adhere for 4 hr in a standard incubator . Cells were next equilibrated with XF Base media ( Seahorse ) at 37°C for one hour in an incubator lacking CO2 . Glycolysis stress test was performed by sequential treatments with glucose ( 10 mM ) , oligomycin ( 1 . 0 μM ) and 2-DG ( 100 mM ) ) . Mitochondrial stress test was performed by sequential treatments with oligomycin ( 1 . 0 μM ) , carbonyl cyanide-4- ( trifluoromethoxy ) phenylhydrazone ( FCCP ) ( 1 . 0 μM ) , and rotenone/antimycin A ( 1 . 0 μM each ) . For determination of oxidative phosphorylation substrate utilization , cells were equilibrated in XF Base media supplemented with glucose ( 5 mM ) and glutamine ( 2 mM ) . For glucose utilization determination , UK5099 ( 2 μM ) and then BPTES ( 3 μM ) /Etoxomir ( 4 μM ) ( 4 M ) or BPTES ( 3 μM ) /Etoxomir ( 4 μM ) and then UK5099 ( 2 μM ) were added sequentially . Dichloroacetate ( DCA ) at 4 mM was maintained in the media at all times for DCA experiment . All chemicals form Sigma-Aldrich . HAECs ( 4 × 105 cells ) were transfected with an siRNA ( 50 nmol ) using RNAiMAX ( Life Technologies , Carlsbad , CA ) as described by the manufacturer . Media was exchanged the following day for flow media . siRNA products are obtained from Qiagen ( Venlo , The Netherlands ) and the siRNAs used are as follows: Non-targeting siRNA sense: 5’-UUCUCCGAACGUGUCACGUdTdT 3’ 1027310 Non-targeting siRNA antisense: 5’-ACGUGACACGUUCGGAGAAdTdT 3’ 1027310 HIF-1α siRNA sense: 5’-GAAGAACUAUGAACAUAAATT-3’ SI02664053 HIF-1α siRNA antisense: 5’-UUUAUGUUCAUAGUUCUUCCT-3’ SI02664053 NOX4 siRNA sense: 5’-CCAGGAGAUUGUUGGAUAATT-3’ SI02642500 NOX4 siRNA antisense: 5’-UUAUCCAACAAUCUCCUGGTT-3’ SI02642500 PDK1 siRNA sense: 5’- CGAACUAGAACUUGAAGAATT-3’ SI00605752 PDK1 siRNA antisense: 5’-UUCUUCAAGUUCUAGUUCGGG-3’ SI00605752 SLC2A1 siRNA sense: 5’-CCACGAGCAUCUUCGAGAATT-3’ SI03089401 SLC2A1 siRNA antisense: 5’-UUCUCGAAGAUGCUCGUGGAG-3’ SI03089401 In vitro transcription for wild type ( T7-HIF-1α ) or mutated HIF-1α ( T7-mHIF-1α ) mRNA transcripts were performed using mMESSAGE mMACHINE T7 Ultra Kit ( Life Technologies ) following manual instructions . HAECs were transfect with 40 ng of T7-KLF2 or T7-ctrl with Lipofectamine MessengerMAX ( Life Technologies ) for 6 hr according to manufacturer’s instructions . Human HIF-1α ( hHIF-1α ) expression plasmid was purchased from GeneCopoeia ( Rockville , MD ) . Mutagenesis of P402 , P564 and N803 in hHIF-1α open reading frame was performed using QuikChange Multi Site-Directed Mutagenesis Kit ( Agilent ( Newport , DE ) ) according to the manufacturer’s protocol . Mutations were confirmed by DNA sequencing . Primer sequences for mutagenesis: hHIF-1α_P402A_F: 5'-TAACTTTGCTGGCCGCTGCCGCTGGAGACAC-3' hHIF-1α_P564G_F: 5'-GATTTAGACTTGGAGATGTTAGCTGGATATATCCCAATGGATGATGACTTC-3' hHIF-1α_N803A_F: 5'-CTGACCAGTTATGATTGTGAAGTTGCTGCTCCTATACAAGGCAG-3' Wild type HIF-1α and mHIF-1α plasmids were used to generate the DNA templates for in vitro transcripts with the primers listed below . In vitro transcription for wild type ( T7-HIF ) or mHIF-1α ( T7-mHIF-1α ) mRNA transcripts was performed using mMESSAGE mMACHINE T7 Ultra Kit ( Life Technologies ) following manual instructions . HAECs were transfected with T7-HIF-1α or T7-mHIF-1α mRNA transcripts using Lipofectamine MessengerMAX ( Life Technologies ) according to manufacturer’s instructions . hT7_F: 5'-CCA CTG CTT ACT GGC TTA TCG-3’ hHIF-1α _stop: 5'-TCA GTT AAC TTG ATC CAA AGC TCT G-3’ A 6-well plate had its bottom sawed off and hand-filed down . A #1 microscope coverslip ( Ted-Pella , Redding , CA ) sufficient to cover the 35 mm2 area was adhered with aquarium seal and allowed to dry over-night . The plates were re-sterilized with ethanol and UV for 48 hr prior to use . 1 hr before cell plating , 0 . 1% gelatin in PBS was pipetted onto the cover glass and incubated at 37°C . HAECs were then plated onto the coverglass overnight before flow experiments . After flow experiments , the media was changed to EGM-2 with either 5 mM CellRox Orange ( ThermoFisher ) or 500 μM MitoTracker Deep Red ( ThermoFisher ) and placed in the incubator for 30 min . The media was changed again for EGM-2 and the cells imaged immediately . A standard mercury lamp and TRITC ( CellRox ) or Cy5 ( MitoTracker ) filter set was used on an inverted microscope ( Zeiss Axio 200 ) . HAECs were plated at 4 × 105 cells/well overnight before either UF or DF flow for 48 hr . Immediately after flow , cells were washed with warm PBS three times before incubating with Krebs-Ringer-Phosphate-HEPES ( KPRH ) buffer ( 20 mM HEPES , 5 mM KH2PO4 , 1 mM MgSO4 , 1 mM CaCl2 , 136 mM NaCl , 4 . 7 mM KCl , pH 7 . 4 ) with 2% BSA for 1 hr . Cells were again washed with PBS three times and incubated in 1 mM 2-NBDG and PBS for 30 min . The cells were again washed with warm PBS three times and KPRH/BSA was added to the cells . The cells were immediately imaged on a Nikon Ti-Eclipse ( Tokyo , Japan ) microscope at 20x using a FITC filter set . 20 random locations were imaged in both epi-fluorescence and brightfield . Fluorescence brightness was quantified using a custom MATLAB script , deposited at the journal website . Reagents for the assay are from Abcam #136955 ( Cambridge , MA ) . HAECs were plated at 4 × 105 cells/well overnight before either UF or DF flow for 48 hr . Cells were washed with warm PBS three times before trypsinized and re-plated at 20 , 000 cells/well ( four replicates per condition ) in 90 μL of EGM2 media in a flat white-bottom plate and allowed to settle for 4 hr at 37 C . Next , either 10 μL PBS , 10 μL of rotenone/antimycin ( final concentration 1 μM each ) or 10 μL 2-deoxyglucose ( final concentration 50 mM ) were added to the cells , and incubated at 37 C for 30 min . The cells were then assayed according to the manufacturer’s protocol . Porcine aortas were obtained from a local slaughterhouse ( Ruwaldt Packing Co . ( Hobart , IN ) ) . Three-year old male pigs destined to enter the food supply were sacrificed and submerged in boiled in water for 10 s per industry protocol . They were rapidly dissected and the aortas removed . The aortas were cut open and washed quickly with cold PBS before gently passing a #10 scalpel across the lumen . 100 μL of lysis buffer was quickly added to the cell scraping and the samples frozen on dry ice . The entire process took less than 10 min . Swine protein lysates were immunoblotted for smooth muscle cell marker SM22-alpha to verify the endothelial purity . Endothelial purity was detected at greater than 95% . All animal care and treatment procedures were approved by the University of Chicago Institutional Animal Care and Use Committee . C57BL/6J mice ( Jackson Laboratory , Bar Harbor , ME ) were anesthetized with trimobromoethanol ( 2 . 5% , 17 . 5 μL/kg , intraperitoneal injection ) ( Sigma ) . As soon as their reflexes were blunted , as assessed by hindleg pinching , the mice were supinated and their limbs taped to a dissection board . Mice were sprayed with 70% ethanol prior to dissection: the thoracic cage was opened and the heart visualized . 5 mL of warm PBS was flushed through the left ventricle with a 25 gauge butterfly needle with the IVC open . Next , 4% paraformaldehyde ( PFA ) ( Sigma ) in PBS was perfused through the left ventricle at a rate of 2 mL/min for 15 min . At the end of perfusion , 5 mL of PBS was again used to wash out any remaining PFA . The aorta was then removed from the thorax and cleaned . The heart and aorta were removed en bloc and the descending thoracic aorta separated from the arch , below the left subclavian artery . For immunofluorescence , the aortas were opened along the dorsal aspect before pinning to a rubber board . The aortas were permeablized with 0 . 1% Triton-X 100/PBS for 10 min and washed with 5% BSA and tris-buffered saline with TBST three times , prior to incubation with 1:100 HIF-1α antibody ( rabbit Ig ) or 1:50 NOX4 antibody ( rabbit Ig ) in 5% BSA and TBST overnight at 4C , together with 1:100 CD31 antibody ( rat Ig ) . After primary antibody incubation , the samples were again washed with 5% BSA and TBST three times and incubated with 1:1000 goat anti-rabbit IgG Alexa 488 ( ThermoFisher , A11034 ) , 1:1000 chicken anti-rat IgG Alexa 594 ( ThermoFisher , A21471 ) and 1:2000 Hoechst ( ThermoFisher , 62249 ) for 1 hr at room temperature . The samples were again washed with 5% BSA and TBST three times prior to mounting on a coverslip for imaging . Imaging was performed on a spinning disk confocal microscope . Dichloroacetate was purchased from Sigma-Aldrich ( 347795 ) . Used at 4 mM final concentration . EUK134 was purchased from Sigma-Aldrich ( SML0743 ) and used at 1 mM final concentration . NEMO-binding domain peptide ( 480025 ) from Millipore . Rotenone and antimycin were purchased from Sigma-Aldrich . Primary antibodies used were HIF-1α ( 1:500 in 5% non-fat milk and TBST for Western , 1:100 in 5% BSA and TBST for immunofluorescence , Cayman , Ann Arbor , MI , 10006421 , RRID:AB_10099184 ) , SLC2A1 ( 1:1000 in 5% BSA and TBST , ProteinTech , Rosemont , IL , 21829–1-AP , RRID:AB_10837075 ) , HK2 ( 1:1000 in 5% BSA and TBST , Cell Signaling , Danvers , MA , C64G5 , RRID:AB_2232946 ) , β-actin ( 1:5000 in 5% BSA and TBST , Abcam , ab6276 , RRID:AB_2223210 ) , PDK1 ( 1:1000 in 5% non-fat milk and TBST , Cell Signaling , C47H1 , RRID:AB_1904078 ) , NOX4 ( 1:500 in 5% milk and TBST for Western , 1:50 in 5% BSA and TBST for immunofluorescence , Santa Cruz , Dallas , TX , sc-30141 , RRID:AB_2151703 ) , CD31 ( BD , Franklin Lakes , NJ 550274 , RRID:AB_393571 ) , EPAS1 ( 1:1000 in 5% BSA and TBST , Novus Biologicals , Littleton , CO , NB100-122 , RRID:AB_10002593 ) . Secondary antibodies for Western blotting: Goat anti-mouse IgG , HRP conjugate ( 1:5000 in 5% non-fat milk and TBST , Calbiochem , 401253 , RRID:AB_437779 ) . Anti-rabbit IgG , HRP conjugate ( 1:3000 in 5% non-fat milk and TBST , Cell Signaling , 7074S , RRID:AB_2099233 ) . HAECs were plated at 4 × 105 cells/well in a 6-well plate overnight before either UF or DF flow for 48 hr before proceeding with the protocol according to the manufacturer . Briefly , the cells were trypsinized and resuspended in sample buffer . 1/10 vol of detergent was added and the sample mixed and stored on ice for 10 min . The sample was then centrifuged and the supernatant collected , and the protein concentration measured with a detergent-compatible BCA protein assay ( Pierce ) , according to standard protocol . Sample volumes were adjusted so that the total amount of protein loaded onto dipsticks was the same . The PDH dipsticks were first blocked and before wicking sample . The dipsticks were then washed with sample buffer before the addition of activity buffer . The reaction was allowed to proceed for 1 hr before washing with deionized water and drying . The dipsticks were the imaged by a flatbead scanner and quantified with ImageJ . The final intensity was normalized to total protein . HAECs were plated at 4 × 105 cells/well in a 6-well plate overnight before either UF or DF flow for 48 hr . The cells were removed from the flow devices and washed with PBS three times . The cells were then starved in KPRH/BSA buffer for 40 min before washing again in PBS three times . Samples were then incubated in either 10 mM 2-deoxyglucose ( 2-DG ) or PBS ( control ) for 20 min . The HAECs were then washed with PBS three times before trypsinization and cell counting . The final cell number was normalized across all experiment and control samples . The samples were pelleted and resuspended in PBS with addition of 80 μL Extraction Buffer and vigorously mixed . After on freeze/thaw cycle on dry ice , the the samples were incubated at 85°C for 40 min . The lysates were cooled on ice for 5 min and then 10 μL of Neutralization Buffer was added . The samples were mixed and centrifuged . The supernatants were then assayed according to the manufacturer protocol , and final 412 nm absorbance reading normalized to non-2DG treated controls . The 412 nm absorbance was compared against a standard curve according to manufacturer’s protocol . HAECs were subjected to 50 nM siRNA targeted against HIF-1α or control for 24 hr , prior to DF for 48 hr . 1 hr prior to removal from flow , THP-1 cells at 5-fold number of HAECs were pelleted and resuspended in 600 μL serum-free RPMI ( 900 rpm , 5 min , room temperature ) . The THP-1 cells were then incubated with 5 μM Calcein AM dye ( ThermoFisher ) for 30 min at 37 C . The cells were then pelleted and resuspended in 600 μL ( 900 rpm , 5 min , room temperature ) serum free RPMI . After removal of flow devices , the HAEC flow medium was replaced with regular EGM2 ( 1 mL ) . 100 μL of labeled THP-1 cells was then added to the HAEC cells and let incubate at 37 C for 1 hr , with gentle rocking every 30 min . The HAEC+THP-1 cells were then washed with warm PBS five times , 2 mL per well . Fluorescence of the Calcien AM dye was then measured on a Cytation 3 ( Biotek , Winooski , VT ) device in area scanning mode , with gain of 80 , and excitation 492 nm , and emission 550 nm . Briefly , quality of reads was assessed using fastQC . Reads were aligned to GENCODE hg38 . p2 reference genome using Tophat2 version 2 . 1 . 1 . Transcripts were assembled using the bam files from Tophat2 using Cufflinks version 2 . 1 . 1 . The transcript files from cufflinks were merged using cuffmerge . Cuffquant was used to estimate abundances , prior to analysis by cuffdiff to estimate differential gene expression . The cut off for differentially expressed genes was an FDR of 0 . 05 ( Trapnell et al . , 2012 ) . The complete unedited RNA-seq datasets are available at DOI: https://doi . org/10 . 5281/zenodo . 260122 ( Wu et al . , 2017b ) and DOI: https://doi . org/10 . 5281/zenodo . 260120 ( Wu et al . , 2017a ) . The list of differentially expressed genes after processing has been uploaded to the publisher’s website . Genes used for GSEA ( Subramanian et al . , 2005; Mootha et al . , 2003 ) were all genes that were tested by cuffdiff , regardless of whether or not they were differentially expressed in either condition . All other genes that were not tested were deemed too lowly expressed or detected to be relevant . The gene set collection used was the h . all . v5 . 1 . symbols . gmt [Hallmarks] gene sets database . GSEA performed 1000 permutations . Phenotype labels corresponded to the conditions of the experiment . Did not collapse dataset to gene symbols . Permutation type was gene_set . Chip platform was ftp://gseaftp . broadinstitute . org/pub/gsea/annotations/GENE_SYMBOL . chip . Enrichment statistic was weighted . Metric for ranking genes was Signal2Noise . The gene list sorting mode was real and done in descending ordering mode . The max size was 500 and the minimum was 15 for gene sets . Cutoff for significant gene sets was an FWER less than 1 . 0 . In this study , we selected to report the ten most significant gene sets . Ingenuity Pathway Analysis ( IPA , Redwood City , CA ) , www . ingenuity . com ) is a system that transforms a list of genes into a set of relevant networks based on extensive records maintained in the Ingenuity Pathways Knowledge Base ( Calvano et al . , 2005 ) . Highly-interconnected networks are predicted to represent significant biological function ( Ravasz et al . , 2002 ) . IPA Upstream Regulator Analytics was used to computationally predict the putative upstream transcriptional regulators that contribute to the gene expression changes in HAECs under athero-relevant hemodynamics . 3757 differentially expressed genes ( DEGs ) identified by the RNA-seq at a false discovery rate cut off value of q < 0 . 05 were uploaded to the IPA software . The analysis was conducted based on prior knowledge of expected effects between transcriptional regulators and their target genes stored in the Ingenuity Knowledge Base . The putative upstream regulators were ranked by an overlap p-value which calls likely upstream regulators based on significant overlap between dataset genes and known targets regulated by a transcriptional regulator . The activation z-score statistic is a weighted sum of activating and inhibiting interactions on a given gene set ( Krämer et al . , 2014 ) . We used DAVID ( Huang et al . , 2009a , Huang et al . , 2009b ) version 6 . 7 . Only differentially expressed genes were included and were separated if they had a positive or negative fold change from the RNA sequencing analysis . Summary results used were from GOTERM_BP_FAT . The top ten most significant gene ontology ( GO ) terms were used and the p-values were reported . Gene list of differentially expressed genes was used at www . metascape . org . Express analysis was used . The data were analyzed in Prism 7 ( GraphPad Software Inc , La Jolla , CA ) or Excel . All data are shown as mean ± standard error of the mean ( SEM ) . All data are in two groups . We explored individual differences with two-tailed Student’s t test . Statistical significance was defined as p<0 . 05 . A P value of less than 0 . 05 was considered significant . Due to experimental constraints , only eight biological replicates ( four control , four experiment ) are able to be run simultaneously with flow-devices . All plotted experiments are biological replicates except for Seahorse metabolic experiments , which are technical replicates . Seahorse experiments were run at least twice . All p-values are available at the publisher’s website . All procedures were in accordance with National Institutes of Health guidelines , and the use of vertebral animal related tissues obtained from outside the University of Chicago was approved by the Institutional Animal Care and Use Committee . RNA-seq differentially expressed genes for either UF or DF , and for DF with siRNA directed towards HIF-1α can be found online at the journal website . The complete unedited RNA-seq datasets are available at Flow transcriptome of human aortic endothelial cells , DOI: https://doi . org/10 . 5281/zenodo . 260122 ( Wu et al . , 2017b ) and HIF-1α knockdown under disturbed flow in human aortic endothelial cells , DOI: https://doi . org/10 . 5281/zenodo . 260120 ( Wu et al . , 2017a ) . Source code for image analysis is available as Source code 1 , writen in MATLAB code .	Atherosclerosis is the build-up of fatty material inside the blood vessels , and is one of the leading causes of heart disease and stroke . The blood vessels affected are typically inflamed for many years before the condition develops , and the condition often occurs at sites where blood vessels branch or turn . The cells that line the inside of the blood vessels are known as endothelial cells . Flowing blood exerts a force upon the endothelial cells , named “shear force” , which is similar to how wind bends plants . When the blood flows in one direction , the shear forces are high , the endothelial cells are tightly held together , and the vessels are less likely to become inflamed . However , the flow of blood is disturbed around turns or branch points . This means thatthe shear forces are lower and that the gaps between the endothelial cells are bigger . Low shear forces also mean that the endothelial cells release chemical signals that promote the inflammation and ultimately leads to atherosclerosis . Though low shear forces play an important role in “activating” endothelial cells to promote inflammation , it was not clear how this happens . Wu et al . now show that when shear forces inside blood vessels are low , endothelial cells promote inflammation by modifying their own metabolism . The experiments involved applying either high or low shear forces to endothelial cells that had originally been collected from a major blood vessel of human donors , and then grown in the laboratory . Wu et al . then analyzed the gene activity of these endothelial cells and discovered that low shear forces activate a selected pool of genes . The activated genes are mainly responsible for two cellular processes: glycolysis and the response to hypoxia . Glycolysis is a process that releases energy by breaking down the sugar glucose , while hypoxia refers to the situation when cells do not receive enough oxygen . Further molecular analyses revealed that low shear forces stabilize a particular protein involved in the response to hypoxia , named HIF-1α , and that this protein is responsible for stimulating glycolysis . Finally , Wu et al . showed that increasing glycolysis in endothelial cells was enough to cause the blood vessels to become inflamed . Going forward , a better understanding of how low shear forces modify the metabolism of endothelial cells in blood vessels and consequently promote inflammation will help scientists to tackle new questions about how atherosclerosis begins and develops . In the longer-term , these findings might also lead to the development of new treatments to atherosclerosis and similar diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2017	HIF-1α is required for disturbed flow-induced metabolic reprogramming in human and porcine vascular endothelium
The ability to recognize small organic molecules and chemical modifications of host molecules is an essential capability of the adaptive immune system , which until now was thought to be mediated mainly by B cell antigen receptors . Here we report that small molecules , such as cyanine 3 ( Cy3 ) , a synthetic fluorescent molecule , and 4-hydroxy-3-nitrophenylacetyl ( NP ) , one of the most noted haptens , are γδ T cell antigens , recognized directly by specific γδ TCRs . Immunization with Cy3 conjugates induces a rapid Cy3-specific γδ T cell IL-17 response . These results expand the role of small molecules and chemical modifications in immunity and underscore the role of γδ T cells as unique adaptive immune cells that couple B cell-like antigen recognition capability with T cell effector function . The adaptive immune system consists of B cells , αβ T cells and γδ T cells . While αβ T cells perform all well-defined functions attributed to T cells , γδ T cells and αβ T cells are present together in all but the most primitive vertebrates . This suggests that each cell type performs unique functions and that both are necessary for host immune competence . Indeed , although γδ T cells and αβ T cells have similar effector functions , γδ T cells and αβ T cells are distinct in their antigen recognition and activation requirements and in their antigen-specific repertoire and effector function development . These differences underlie γδ T cells' unique contribution to host immune defense ( Chien et al . , 2014 ) . Diversity in antigen receptor specificity is the hallmark of the adaptive immune system . Serological analysis of small chemical compound immune recognition was one of the earliest experimental demonstrations that B cells can mount responses to diverse antigens with specificity ( Landsteiner and van der Scheer , 1931; Landsteiner and Chase , 1937 ) . Haptens were characterized as small organic molecules which , when conjugated to a protein , induce a strong hapten-specific B cell response . Since then , antibody responses to haptens have been used extensively to investigate antibody affinity maturation , germinal center formation , and the development of memory B cell responses ( Jack et al . , 1977; Jacob et al . , 1991; McHeyzer-Williams and McHeyzer-Williams , 2005 ) . Antibodies specific for pathogen-produced small compounds and chemical modifications of host molecules have also served as a means of pathogen surveillance ( Daneshvar et al . , 1989; Temmerman et al . , 2004 ) and to monitor injury or altered physiological states ( Vossenaar et al . , 2004; Kim et al . , 2006; Yang and Sauve , 2006 ) . Thus , small molecule recognition is an important capability of the adaptive immune system . Although hapten-specific αβ T cells have been reported and studied in the context of suppressor T cell function , as exemplified by the work of Dorf et al . ( Sherr and Dorf , 1981 ) , interaction between the T cell receptor ( TCR ) and the hapten ligand has not been demonstrated . Moreover , it is well established that the antigen-specific repertoires of peripheral αβ T cells are largely limited to peptides that are processed from protein antigens in complex with the host major histocompatability complex ( MHC ) molecules ( Huseby et al . , 2005; Van Laethem et al . , 2007 ) . This restriction on antigen specificity is a consequence of the thymic development process ( Van Laethem et al . , 2012 ) . Thus , adaptive immune recognition of small molecules seems to be mainly mediated by B cells rather than T cells . While γδ T cells , like αβ T cells , require thymic maturation before entering the periphery ( Ohno et al . , 1993 ) , this process does little to constrain the γδ T cell antigen-specific repertoire ( Jensen et al . , 2008 ) . In addition , although fetal-derived γδ T cells in murine skin and the reproductive tract express non-variant TCRs , adult human and murine γδ T cells in other lymphoid compartments ( blood , lymph node , spleen , and intestine ) express diverse TCRs ( Chien et al . , 2014 ) . Analysis of γδ TCR CDR3 sequence diversity and length distribution suggest that these T cells have extensive antigen recognition capability and that as a group , γδ TCRs are more similar to immunoglobulins ( Igs ) than to αβ TCRs ( Rock et al . , 1994 ) . Since the requirements of γδ T cell antigen recognition are similar to those of B cells , we investigated whether γδ T cells , like B cells , can recognize haptens . Here , we report that Cyanine 3 ( Cy3 ) , a synthetic fluorescent molecule , is a γδ T cell antigen , recognized directly by specific γδ TCRs . Immunization with Cy3 induces γδ T cells to mount a Cy3-specific IL-17 response . IL-17 is a T cell cytokine , which is essential in the initiation of the inflammatory response . We also identified γδ TCRs specific for 4-hydroxy-3-nitrophenyl acetyl ( NP ) , one of the most commonly studied haptens in investigation of antibody response . These results enlarge the scope of the γδ T cell antigen-specific repertoire and suggest a way for this category of antigens to induce a T cell response . To test whether γδ T cells can recognize small molecules , we chose Cyanine 3 ( Cy3 ) for analysis . Cy3 is a synthetic dye with two modified indole groups joined by a polymethine chain ( Figure 1A ) . It is highly fluorescent and can be used for FACS analysis directly . 10 . 7554/eLife . 03609 . 003Figure 1 . Cy3 is a γδ T cell antigen . ( A ) Chemical structure of Cyanine 3 ( Cy3 ) . FACS analysis of ( B ) Cy3 tetramer ( Cy34-SAv ) staining of splenic γδ T cells in the presence of 10-fold molar excess of moth cytochrome c peptide coupled SAv ( MCC4-SAv ) ; ( C ) NX6/58α-β- cells stained with Cy3-MCC-SAv or PE-MCC-SAv; ( D ) NX6/58α-β- cells stained with Cy3-MCC-SAv in the absence ( left ) , or presence of anti-Cy3 Fab ( right ) . ( E ) IL-2 production by NX6/58α-β- cells activated by the indicated amount of plate-bound Cy3-OVA , OVA , PE , anti-CD3 for 16 hr . ( F ) The saturating binding curves of Cy34-SAv and un-conjugated SAv to a soluble form of NX6 as determined by surface plasmon resonance . No detectable binding was observed for 1 mM applications of PE or BSA ( not shown ) . ( G ) Kinetics of Cy34SAv binding to NX6/58α-β- cells . t1/2 was determined using real time flow cytometry in the presence of anti-Cy3 antibody Fab fragments ( left ) . KD was determined from Scatchard analysis ( right ) . All results are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03609 . 00310 . 7554/eLife . 03609 . 004Figure 1—figure supplement 1 . NX6/58α-β- cells stained with different fluorescently labeled ovalbumin preparations . Flow cytometry analysis of NX6/58α-β− cells stained with Cy3- OVA , FITC-OVA and APC-OVA . DOI: http://dx . doi . org/10 . 7554/eLife . 03609 . 00410 . 7554/eLife . 03609 . 005Figure 1—figure supplement 2 . Correlation between the mean fluorescence intensities of PE-SAv and Cy34SAv on red blood cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03609 . 005 Cy3 tetramer ( Cy34SAv ) ( a recombinant mutant of streptavidin ( Ramachandiran et al . , 2007 ) labeled with four Cy3 molecules at the C-terminal cysteine in each of the four identical subunits ) stained ∼0 . 05–0 . 2% of normal splenic γδ T cells , but not G8/Rag2−/− γδ TCR transgenic cells ( specific for the nonclassical MHC class I T10 and T22 ) ( Bluestone et al . , 1988; Schild et al . , 1994; Figure 1B ) . We then identified Cy3-specific γδ TCRs on a single cell level by sorting these cells and sequencing their TCR genes . 58α-β- cells expressing Cy3-specific γδ TCRs bound Cy3-ovalbumin ( Cy3-OVA ) , Cy3-bovine serum albumin ( Cy3-BSA ) , Cy3-MCC-streptavidin ( moth cytochrome C ( MCC ) -derived peptide , Cy3-labeled at the N-terminus , biotinylated at the C-terminus , and tetramerized with streptavidin ) , but not FITC or APC labeled OVA , nor PE-MCC peptide/streptavidin ( Figure 1C , Figure 1—figure supplement 1; Table 1 ) . Moreover , Cy3-MCC-streptavidin staining of a Cy3-specific γδ TCR NX6/58α-β- was inhibited by the inclusion of Fab fragments of an anti-Cy3 antibody ( clone A-6; Santa Cruz Biotechnology ) ( Figure 1D ) . In addition , NX6/58α-β- cells were activated by plate-bound Cy3-OVA , but not unmodified OVA ( Figure 1E ) . Binding of the soluble form of a Cy3-specific γδ TCR ( NX6 ) to Cy34SAv can be demonstrated by surface plasmon resonance ( Biacore ) with an apparent KD of 78 . 2 nM ( Figure 1F ) . We also examined the affinity of Cy34SAv binding to NX6 expressed on 58α-β- cells . Scatchard analysis showed an apparent nanomolar KD ( 1 . 8 nM ) with a half-life of ∼26 min ( Figure 1G ) . Taken together , these results indicate that Cy3 is an antigen of γδ T cells , recognized directly by specific γδ TCRs . 10 . 7554/eLife . 03609 . 006Table 1 . TCR sequences of Cy3 and NP-specific γδ TCRsDOI: http://dx . doi . org/10 . 7554/eLife . 03609 . 006VδND1ND2NJδVγNJγCy3NX6Vδ8C A A SAT D KVγ1C A V WS RS G T S W V KC5Vδ6AC A L W E LGG G I RA SD KVγ1C A V WT RG T S W V KNP1G9Vδ4C A L M E RRG YR R D TR AD KVγ4C S Y G SYS S G F H K1E3Vδ6BC A L S E LG GG GS AT D KVγ1C A V WK K TG T S W V K1B2Vδ4C A L M E RVGL YR R D TS L AT D KVγ1C A VFS G T S W V KEach pair of γ and δ chain sequences were identified from a single Cy3 or NP-specific γδ T cell derived from mouse splenocytes and verified by their ability to confer NP- or Cy3-specific binding to 58α-β- cells expressing the TCR . To determine whether γδ T cells can mount a hapten-specific response , we immunized mice subcutaneously with Cy3–chicken gamma globulin ( Cy3-CGG ) in aluminum hydroxide ( alum ) and analyzed Cy3-specific γδ T cells in the draining lymph nodes with a Cy3-OVA staining reagent . For comparison , we also analyzed Cy3-specific γδ T cells in mice immunized with CGG/alum . Alum was used because it is a non-antigenic adjuvant ( Eisenbarth et al . , 2008 ) , and we chose subcutaneous immunization because it focuses the immune response to the draining lymph nodes . We found that prior to immunization , ∼80% of Cy3-specific γδ T cells in the lymph nodes were CD44lo , a phenotype typical of naïve T cells . Within 24 hr after immunization , Cy3-specific γδ T cells up-regulated CD44 in Cy3-CGG-immunized mice , but not in CGG-immunized mice ( Figure 2A ) . BioMark analysis showed that Cy3-specific γδ T cells express the mRNA coding for RORγt , IL-17A and IL-17F 60 hr after immunization ( Figure 2B ) . Consistent with this observation , analysis of Cy3-specific γδ T cell responses in IL-17F reporter mice ( Il-17fThy1 . 1/Thy1 . 1 ) ( Lee et al . , 2009 ) and staining showed that 60 hr after Cy3-CGG immunization , activated Cy3-specific γδ T cells expressed the Thy1 . 1 reporter or IL-17 protein ( Figure 2C ) . In addition , we found that activated Cy3-specific γδ T cells expressed the receptors for IL-1 and IL-23 ( Figure 2B ) , a characteristic similar to our analysis of activated PE-specific γδ T cells in an immune response ( Zeng et al . , 2012 ) . The expression of inflammatory cytokine receptors allows antigen-activated γδ T cells to integrate signals from antigen receptors and cytokine receptors to mount an enhanced and sustained response ( Zeng et al . , 2012 ) . 10 . 7554/eLife . 03609 . 007Figure 2 . Cy3-specific γδ T cell response after immunization . ( A ) CD44 expression on Cy3-OVA+ ( red ) and Cy3-OVA− γδ T cells in the draining lymph nodes of mice immunized with Cy3-CGG-alum or CGG-alum 24 hr prior . ( B ) BioMark analysis of CD62LloCD44hi Cy3+ and CD62LhiCD44lo Cy3− γδ T cells isolated from the draining lymph nodes of C57BL/6 mice immunized with Cy3-CGG 60 hr prior ( 5 cells/sample ) . The heatmap , where rows are individual genes and columns are individual samples , indicates the expression or non-expression of a gene/sample pair ( relative to the β2m expression ) . Upper panel shows genes expressing higher ( p < 0 . 001 ) in Cy3+ cells than that in Cy3− cells . Middle panel shows non-varying genes . Bottom panel shows genes expressing lower ( p < 0 . 001 ) in Cy3+ cells than that in Cy3− cells . ( C ) Thy1 . 1 expression on γδ T cells from IL-17fThy1 . 1/Thy1 . 1 mice immunized with Cy3-CGG-alum 60 hr prior , representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03609 . 007 Taken together , the observations that Cy3-specific γδ T cells can be activated and produce IL-17 upon Cy3-CGG , but not CGG , immunization indicates that γδ T cells , like B cells , are capable of mounting specific responses to small molecules . To test the generality of the observation that γδ T cells can recognize small molecules , we chose 4-hydroxy-3-nitrophenyl acetyl ( NP ) for analysis . NP is one of the most commonly studied hapten molecules in investigations of antibody responses ( Jack et al . , 1977; Jacob et al . , 1991; McHeyzer-Williams and McHeyzer-Williams , 2005 ) , and NP is structurally unrelated to Cy3 ( Figure 3A ) . 10 . 7554/eLife . 03609 . 008Figure 3 . NP is a γδ T cell antigen . ( A ) Chemical structure of 4-hydroxy-3-nitrophenyl acetyl ( NP ) . Flow cytometry analysis of ( B ) NP67-PE staining of γδ T cells from C57BL/6 or G8/Rag2−/− mouse splenocytes and PE staining of γδ T cells from B6 splenocytes; ( C ) staining of 58α-β- cells expressing an NP-specific γδ TCR , 1G9 , with NP43-CGG-Cy5 or CGG-Cy5 , showing staining in relation to γδ TCR expression ( left ) or as a histogram ( right ) ; ( D ) staining of 58α-β- cells expressing an NP-specific γδ TCR , 1E3 , with NP43-CGG-Cy5 , NP26-BSA-Cy5 , or BSA-Cy5 ( left ) and NP67-PE alone , NP67-PE with a 20-fold molar excess of anti-NP Fab , or PE ( right ) . ( E ) IL-2 production by 1E3/58α-β- cells activated by the indicated amount of plate-bound NP25-KLH , KLH ( light gray bars ) , or 0 . 1 μg/ml anti-CD3 . ( F ) Sensorgram and steady state analysis of NP43-CGG ( 0–7 μM ) binding to soluble 1G9 TCR measured by surface plasmon resonance . Apparent KD was determined by steady state analysis of SPR measurements ( circles ) . Equal concentrations of un-modified CGG were tested ( squares ) , as well as NP43-CGG with a PE-specific γδ TCR , MA2 ( triangles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03609 . 008 NP conjugated to a fluorescent protein , phycoerythrin ( PE ) , is routinely used to identify NP-specific B cells in FACS analysis . We found that NP-PE stained ∼0 . 14% of splenic γδ T cells of normal mice ( left panel ) , but not G8/Rag2−/− γδ TCR transgenic cells ( middle panel ) . Consistent with the observation that PE is a γδ T cell antigen ( Zeng et al . , 2012 ) , we found ∼0 . 03% of splenic γδ T cells stained with PE under the same staining conditions ( right panel ) . After accounting for background staining and for PE staining , we estimated that ∼0 . 1% of total γδ T cells could be NP-specific ( Figure 3B ) . We further identified NP-specific γδ TCRs on a single cell level . Expressing NP-specific γδ TCRs in 58α-β- cells enables these cells to be stained with NP-CGG-Cy5 , but not CGG-Cy5 ( Figure 3C; Table 1 ) . Further investigation showed 58α-β- cells expressing NP-specific γδ TCRs could also be stained with NP-BSA-Cy5 and NP-PE , but not with BSA-Cy5 or PE ( Figure 3D , left panel ) . In addition , NP-PE staining was inhibited by the inclusion of Fab fragments of an anti-NP antibody ( clone H33Lγ; G . Kelsoe ) ( Figure 3D , right panel ) . Furthermore , 58α-β- cells expressing NP-specific γδ TCRs produced IL-2 in response to plate-bound NP-keyhole limpet hemocyanin ( NP-KLH ) , but not plate-bound KLH in a dose-dependent manner ( Figure 3E ) . The observations that only molecules containing the NP conjugation stain NP-specific γδ TCR-expressing cells , that NP-conjugate staining is blocked by an anti-NP Fab , and that an immobilized NP-conjugate can activate NP-specific γδ T cells indicate that NP is recognized directly by specific γδ TCRs . Indeed , direct binding between soluble NP-specific γδ TCRs ( 1G9 ) and NP-conjugates was also demonstrated using surface plasmon resonance ( Figure 3F ) . The measured apparent KD for the interaction between NP43-CGG and the 1G9 TCR was 0 . 66 μM . NP43-CGG exhibited no binding to the PE-specific γδ TCR , MA2 ( Zeng et al . , 2012 ) , and CGG did not bind 1G9 ( Figure 3F ) . Taken together , these results show that NP is a γδ T cell antigen and is recognized directly by specific γδ TCRs . At the turn of the last century , Landsteiner pioneered the use of small synthetic molecules , known as haptens , to induce an antibody response . When coupled with carrier proteins , haptens induce a strong ( hapten ) specific , ( αβ ) T cell-dependent B cell response . Since the hapten modification provides a defined epitope for analysis of the antibody response , haptenated proteins have been used extensively to characterize the development of B cell responses , and NP is one of the most commonly studied haptens . Although Cy3 has not been used in this context previously , high affinity , isotype-switched Cy3-specific antibodies are widely available commercially , indicating that Cy3 is also highly immunogenic , similar to other well-studied hapten molecules . Our demonstration that γδ T cells can directly recognize and respond to these molecules represents a significant expansion in the scope of the γδ T cell antigen-specific repertoire . This is the first demonstration that γδ TCRs can interact directly with small molecules . In this context , prior work has shown that a collection of small pyrophosphate-containing organic molecules can stimulate human Vγ9Vδ2-expressing γδ T cells ( also referred to as Vγ2Vδ2 by the Seidman et al . nomenclature ) in vitro in a TCR-dependent manner ( Chien et al . , 2014 ) . These molecules , collectively known as phosphoantigens ( pAgs ) include isopentenyl pyrophosphate ( IPP ) , an intermediate of the human mevalonate pathway , and ( E ) -4-hydroxy-3-methyl-but-2-enyl-pyrophosphate ( HMBPP ) , a microbial isoprenoid intermediate . However , recent reports indicate that pAgs do not interact directly with the γδ TCR . Instead , Vγ9Vδ2 T cell activation by pAgs is through the recognition of an allosteric change in the extracellular domain of a cell surface molecule , butyrophilin 3A1 , which is induced in response to intracellular accumulation of pAg ( Wang et al . , 2013; Sandstrom et al . , 2014 ) . Our past studies indicate that γδ T cells need not encounter cognate antigen in the thymus to signal through the TCR , mature , and exit to the periphery . Peripheral γδ T cells derived from γδ thymocytes that have not previously encountered thymic ligands produce IL-17 upon TCR triggering ( Jensen et al . , 2008 ) . Indeed , we have identified multiple foreign molecules which are γδ T cell antigens: phycoerytherin ( PE ) , a member of the phycobiliprotein family , which is located on the tip of photosynthetic antenna of red algae and cyanobacteria ( Zeng et al . , 2012 ) , and here , the haptens Cy3 and NP as γδ T cell antigens . Moreover , both Cy3 and PE-specific γδ T cells differentiate toward an IL-17-producing phenotype with similar activation kinetics upon antigen encounter ( Zeng et al . , 2012 ) ( and this manuscript ) : within 24 hr after immunization , PE- or Cy3 specific γδ T cells in the draining lymph node showed activated phenotypes , such as becoming CD44hi and CD62Llo . Activated antigen-specific γδ T cells express RORγt 48 hr after immunization and , after another 12 hr , IL-17A and IL-17F . Significantly , the expression of inflammatory cytokine receptors such as IL-1R and IL-23R are induced on antigen activated γδ T cells . The cytokine-receptor signaling provides a ‘second signal’ in addition to TCR engagement to perpetuate the response in inflammation ( Zeng et al . , 2012 ) . The inflammatory response is an essential mechanism in the host response to infection and injury . In vertebrates , it requires IL-17 , a cytokine primarily made by T cells . IL-17 induces the maturation and release of neutrophils from the bone marrow ( Stark et al . , 2005 ) . Neutrophil infiltration focuses the immune response at the site of infection or injury , where antigen-specific αβ T cells subsequently proliferate and gain effector functions after stimulation by professional antigen-presenting cells and a particular cytokine environment . In acute infection , the host must make IL-17 rapidly without prior antigen exposure . The results presented here , together with our previous studies , suggest that γδ T cells are uniquely suited for the initial IL-17 response and provide a way for haptens to elicit this important T cell response . In this context , it has been noted that haptenation enhances the immunogenicity of the carrier protein , but this effect is independent of innate immune recognition and signaling ( Palm and Medzhitov , 2009 ) . While serological responses to haptens were first demonstrated to illustrate the capability of the immune system to recognize diverse antigens , it appears that adaptive immune recognition of hapten-like pathogen-derived organic compounds and chemical modifications of host molecules can serve as a means of pathogen surveillance and monitoring of injury or altered physiological states . The synthetic hapten molecule NP is structurally similar to nitrated tyrosine ( 3-NTyr ) . 3-NTyr-containing proteins are formed in the presence of peroxynitrite , one of the side products of reactive oxygen and nitrogen species produced during the early stages of inflammation ( Beckman et al . , 1992; Ischiropoulos et al . , 1992a; Ischiropoulos et al . , 1992b; Beckmann et al . , 1994 ) . Tyrosine nitration has been demonstrated in a variety of infectious and inflammatory contexts , such as Trypanosoma cruzi infection ( Naviliat et al . , 2005; Dhiman et al . , 2008 ) and atherosclerosis ( Beckmann et al . , 1994 ) . There have been reports indicating that these pathological processes are driven in part by IL-17 and γδ T cells ( Stemme et al . , 1991; Kleindienst et al . , 1993; Lima and Titus , 1996; Hashmi and Zeng , 2006; Sardinha et al . , 2006; Cheng et al . , 2008; van Es et al . , 2009 ) . Furthermore , the synthetic hapten molecule Cy3 contains two modified indole groups joined by polymethine bonds . The indole molecule is a noted bacterial product and signaling molecule , which accumulates at the site of bacterial infection and affects antibiotic resistance and other virulence factors ( Martino et al . , 2003; Lee et al . , 2007; Hirakawa et al . , 2009; Lee et al . , 2010; Kim et al . , 2011 ) . An indole group also forms the side chain of tryptophan . Altered tryptophan metabolism along the kynurenine pathway and an unrestrained γδ T cell IL-17 response were identified as the causes of lethal pulmonary aspergillosis in a mouse model of chronic granulomatous disease ( Romani et al . , 2008 ) . Whether hapten-specific γδ T cells also recognize structurally similar natural products , such as 3-NTyr and indole groups , is unclear . Regardless , our observations that small molecules and chemical modifications on proteins are γδ T cell targets suggest a new category of antigen specificity in addition to cell surface molecules such as the non-classical MHC class I molecules T10 and T22 , MHC class I-related chain A/B ( MICA/B ) , and endothelial protein C receptor ( EPCR ) ( Willcox et al . , 2012 ) that can activate γδ T cells in infection and inflammation . The role of γδ T cells in hapten-driven pathological situations is currently unclear , and with these new findings worthy of future study . Allergic contact dermatitis ( ACD ) represents a specific example of a delayed-type-hypersensitivity response with a hapten-driven mechanism . γδ T cells have been implicated in mouse models of ACD . Some studies suggest that γδ T cells assist αβ T cells in adoptive transfer of contact sensitivity ( Ptak and Askenase , 1992 ) , while others suggest that γδ T cells regulate effector αβ T cell responses ( Guan et al . , 2002 ) . Given our findings that γδ T cells can recognize haptens and mount a hapten-specific immune response , studies of the role of hapten-specific γδ T cells in processes like ACD could yield interesting results . Although diversity in antigen receptor specificities is the hallmark of the adaptive immune system , effective adaptive immune responses are focused in antigen specificity . This is best illustrated in αβ T cell-dependent antibody responses , wherein only αβ T cells that can recognize proteins that are internalized and presented by B cells and displayed as peptide/MHC complexes on cell surface can provide B cell help . Thus , only haptens coupled to proteins , which can be processed and presented for αβ T cell recognition , can induce a hapten-specific antibody response . While αβ T cells are responsible for the development of high affinity , isotype-switched antibodies , we found that γδ T cells recognize and respond to noted B cell antigens such as PE , NP and Cy3 . In addition , in a case of human autoimmune myositis , where clonally expanded γδ T cells destroy muscle fiber , the targets of γδ T cells were also the targets of autoantibodies known as anti-Jo-1 ( Bruder et al . , 2012 ) . These observations indicate that an overlap between the γδ T cell and B cell antigen-specific repertoires . If the frequencies of other antigen-specific γδ T cells were also in a similar range as that of PE , Cy3 and NP , then the numbers of distinct γδ T cell antigens would be ∼103–104 . The size of the B cell antigen-specific repertoire was estimated as roughly 105 , based largely on antigen-specific B cell frequencies of 0 . 004–0 . 007% for nitrophenyl ( NP ) , dinitrophenyl ( DNP ) , and trinitrophenyl ( TNP ) . These values were obtained using antigen-specific spleen foci formation , ( Press and Klinman , 1974; Stashenko and Klinman , 1980 ) and are likely to be underestimates , as this assay requires extensive proliferation of individual clones . In fact , FACS analysis showed that in naïve mice , 0 . 1% of the B cells are PE-specific and 0 . 02% allophycocyanin ( APC ) -specific ( Pape et al . , 2011 ) . Using these values , the size of the antigen-specific B cell repertoire would be ∼1000–5000 , in the same range as that estimated for γδ T cells . Regardless of the extent of overlap between B cell and γδ T cell antigen-specific repertoires , our results here support previous observations ( Bruder et al . , 2012; Zeng et al . , 2012 ) that γδ T cells and β cells can recognise the same antigen . In particular , NP- and PE have been used extensively as model antigens to elucidate principles of antibody affinity maturation , germinal center formation and the development of memory B cell responses . These studies should provide a context to study the roles of γδ T cells in the development of an integrated adaptive immune response . Cy3 labeling of biotinylated moth cytochrome c ( MCC ) peptide ( residues 88–103 ) , ovalbumin ( OVA ) ( Sigma , St . Louis , MO ) , BSA ( Sigma ) , CGG ( EMD Millipore , Billerica , MA ) , and streptavidin ( SAv ) was carried out with Cy3 maleimide and amine-reactive labeling kits ( GE Healthcare , Little Chalfont , UK ) . NP ( 4-hydroxy 3-nitrophenylacetyl ) -phycoerythrin ( PE ) was prepared using NP-O succinymidyl ester ( NP-OSu ) ( Biosearch Technologies , Petaluma , CA ) . NP-chicken gamma globulin ( NP-CGG ) and NP-bovine serum albumin ( NP-BSA ) ( Biosearch Technologies ) were fluorescently labeled with Cyanine 5 ( Cy5 ) on amine groups ( Cy5 Mono-Reactive Dye , GE Healthcare ) . C57BL/6 mice were purchased from Jackson Laboratories and housed in the Stanford Animal Facility for at least one week before use . IL-17fThy1 . 1/Thy1 . 1 mice and G8/Rag2−/− TCR transgenic mice were bred and housed in the pathogen-free Stanford Animal Facility . All experiments were performed in accordance with the Institutional Biosafety Committee and the Institutional Animal Care and Use Committee . 200 μg each of Cy3-CGG and CGG in aluminum hydroxide ( Imject Alum; Thermo Scientific , Waltham , MA ) per mouse and subcutaneous immunization were used in all studies . Antibodies were purchased from either eBioscience or BD Biosciences unless otherwise stated . All analyses were performed on a BD LSR II flow cytometer . γδ T cells were enriched from mouse splenocytes by positive selection as described ( Jensen et al . , 2008 ) . For NP experiments , staining of enriched γδ T cells was performed using 15 μg/ml NP43-CGG-Cy5 or 0 . 02 μg/ml NP67-PE or PE , along with PE or APC conjugated anti-TCRδ ( GL-3 ) , APC-Cy7 and Pacific Blue-labeled antibodies to αβ TCR ( H57-597 ) , B220 ( RA2-6B2 ) , F4/80 ( BM8 ) , Gr-1 ( RB6-8C5 ) , and CD11b ( M1/70 ) , and Aqua Amine live/dead stain ( Invitrogen Molecular Probes , Eugene , OR ) . APC-Cy7 , Pacific Blue , and Aqua positive cells were excluded from analysis . Anti-NP antibody ( clone H33L γ; G . Kelsoe ) Fab fragments were prepared using the Pierce Fab Preparation kit . For Cy3 experiments , enriched γδ T cells were stained with Cy3-conjugated protein ( 0 . 5 μM ) on ice for 1 hr , along with APC conjugated GL-3 , Aqua Amine , FITC conjugated antibodies to αβ TCR , B220 , CD11b , CD11c ( N418 ) , Gr-1 , and F4/80 . FITC and Aqua-positive cells were excluded from the analysis . For the analysis of CD44 expression , enriched γδ T cells were stained with FITC-conjugated antibody to CD44 ( IM7 ) , APC conjugated GL-3 , and Cy3-OVA . For the analysis of Thy1 . 1 expression on cells isolated from IL-17fThy1 . 1/Thy1 . 1 reporter mice , enriched γδ T cells were stained with FITC conjugated antibody to Thy1 . 1 ( OX-7; Biolegend ) , Pacific Blue conjugated antibody to CD62L ( MEL-14 ) , APC conjugated GL-3 , and Cy3-OVA . Both analyses included the addition of Aqua Amine and APC-Cy7 labeled antibodies to αβ TCR , B220 , CD11b , CD11c , Gr-1 , and F4/80 , with Aqua and APC-Cy7-positive cells excluded from analysis . TCRs from Cy3- or NP-specific γδ T cells were identified at a single cell level and full length γ and δ TCR chain sequences were cloned and expressed in the 58α-β- cell line as described ( Shin et al . , 2005; Zeng et al . , 2012 ) . 58α-β- cells expressing γδ TCRs were stimulated with plate-bound NP- or Cy3-conjugates , the corresponding unmodified protein , or anti-CD3 . The supernatant was collected and assayed for IL-2 production as described ( Zeng et al . , 2012 ) . Measurement of the kinetics of antigen binding to cell surface-expressed γδ TCR by real-time FACS analysis was carried out as described ( Zeng et al . , 2012 ) . Briefly , 58α-β- cells expressing a Cy3-specific γδ TCR NX6 were stained with 40 nM Cy34SAv for 1 hr at 4°C . Cells were spun down and resuspended in FACS buffer with 1600 nM anti-Cy3 ( clone A-6 , Santa Cruz Biotechnology , Dallas , TX ) Fab , prepared using the Pierce Fab Preparation Kit . Cy34SAv binding was recorded by flow cytometry over 1 . 5 to 3 hr at 5 or 10 min intervals ( <10 s for each measurement ) . The data were fit using a first-order decay kinetic model to obtain the off-rate ( koff ) and half-life ( t1/2 ) . Scatchard analysis of Cy3 binding to NX6/58α-β- cells was carried out as described in Zeng et al . ( 2012 ) with some modifications . 1 × 105 cells were incubated with 27 . 34–1 . 71 nM Cy34SAv . To quantify cell surface bound Cy34SAv , we biotinylated red blood cells ( RBCs ) to generate cells with different surface biotin densities ( Huang et al . , 2010 ) . The same batch of biotinylated RBCs was stained with either PE-SAv or Cy34SAv . A linear correlation ( Figure 1—figure supplement 2 ) between the mean fluorescence intensities of PE-SAv and Cy34SAv was constructed , so that the Cy34SAv staining intensities could be converted to PE-SAv intensities , which were used to calculate the number of bound ligands by comparing them with the standard PE calibration curve . Soluble γδ TCRs were produced as described ( Zeng et al . , 2012 ) . Briefly , the extracellular domains of the γ and δ chains ( residues 1–273 and 1–242 , respectively ) were cloned in frame with a gene encoding a rhinovirus protease site , followed by acidic ( TCR-δ ) or basic ( TCR-γ ) leucine zippers and a ( histidine ) 6 tag in the pMSCV-P2 and Z4 retroviral expression vectors . These vectors contain an internal ribosome entry site followed by puromycin resistance gene for γ chain or zeocin resistance gene for δ chain and expressed in BHK-21 cells . Surface plasmon resonance using the Biacore system was used for quantitative measurements of TCR-ligand interactions . All Biacore measurements were performed on a Biacore 3000 instrument using a CM5 chip . 10 , 000 RU of anti-TCRδ was immobilized using amine linkages; anti-TCRβ was immobilized as a reference surface . Roughly 300 RU of γδ TCR was injected into the system , allowed to stabilize for 1 min , then a range of concentrations of analytes were injected , followed by a 2 min dissociation time . For NP-specific TCRs , NP43-CGG and CGG , were tested; for the Cy3-specific TCR , Cy34-SAv and streptavidin were tested . 10 mM glycine pH 2 . 5 was used at the end of each cycle to remove bound TCR and ligand . Specific binding was assessed by subtracting a blank buffer injection for each cycle . The dose response curves for NP43-CGG , CGG , Cy34-SAv , and unconjugated SAv for specific binding were measured by averaging signal between 10 and 20 s at the end of each analyte injection , as very slow unbinding was observed . Quantitative analysis of transcript expression of Cy3-specific γδ T cells was carried out with the BioMark system as follows: 60 hr after Cy3-CGG immunization , γδ T cells were enriched from the draining lymph nodes of immunized mice , then incubated with Cy3-KLH ( 0 . 5 μM ) for 6 hr in vitro . CD62LloCD44hi Cy3+ and CD62LhiCD44lo Cy3− γδ T cells were then FACS sorted into a PCR plate with five cells per well for the analysis . The primers for BioMark qPCR were purchased from Applied Biosystems . The sequences are described in Supplementary file 1 . Analyses of the expression data were performed with the R statistical package v . 3 . 0 . 2 . To compare the transcriptional profiles of Cy3+ and Cy3− cells , we performed differential expression analysis using a two-sample Mann–Whitney test . Prior to hypothesis testing , we removed any gene that did not vary across the entire sample of Cy3+ and Cy3− cells . Genes were considered significantly different at a Bonferroni-corrected p-value < 0 . 0019 . Gene expression differences were displayed in a two-dimensional heatmap false colored based on transcript expression levels .	Our immune system responds to invading microbes—such as viruses and bacteria—and tries to eliminate the threat via two distinct but connected systems: the innate and the adaptive immune systems . Cells of the innate immune system patrol our organs and tissues in an effort to identify and eliminate threats with a quick but general response , which is similar for many different pathogens . This first line of defense also escalates the immune response by activating the adaptive immune system . Unlike the innate immune response , the adaptive immune response targets unique molecules of different sizes , shapes and chemical compositions—ranging from small organic molecules to large pathogens . The adaptive immune system consists of three types of immune cells: B cells , alpha beta ( αβ ) T cells and gamma delta ( γδ ) T cells . These cells have proteins on their surfaces that function as receptors; when the receptors recognize and bind to a foreign molecule ( called antigen ) , the cell becomes activated . This then triggers a cascade of events that help to clear the infection and help immune cells to rapidly respond to any future infection by the same pathogen . αβ T cells and γδ T cells respond to different triggers , but perform similar tasks—while B cells perform tasks that are different from those of T cells . An effective immune response often involves both B cells and T cells . One important way that the adaptive immune system can identify an invading microbe or monitor for damaged or abnormal cells is by recognizing chemicals produced by pathogen and chemical modifications of host molecules . And while B cells are able to do this , αβ T cells are not . Zeng et al . now show that γδ T cells can also recognize and mount response against this type of antigen . γδ T cells were shown to detect both a small synthetic fluorescent dye , and a chemical modification that has been extensively studied for B cell responses over the last 80 years . Following on from these findings , the next challenge is to identify γδ T cells that recognize molecules or chemical compounds produced during infection or disease , and to define these cells' role in immunity .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2014	Gamma delta T cells recognize haptens and mount a hapten-specific response
Phagocytic immune cells kill pathogens in the phagolysosomal compartment with a cocktail of antimicrobial agents . Chief among them are reactive species produced in the so-called oxidative burst . Here , we show that bacteria exposed to a neutrophil-like cell line experience a rapid and massive oxidation of cytosolic thiols . Using roGFP2-based fusion probes , we could show that this massive breakdown of the thiol redox homeostasis was dependent on phagocytosis , presence of NADPH oxidase and ultimately myeloperoxidase . Interestingly , the redox-mediated fluorescence change in bacteria expressing a glutathione-specific Grx1-roGFP2 fusion protein or an unfused roGFP2 showed highly similar reaction kinetics to the ones observed with roGFP2-Orp1 , under all conditions tested . We recently observed such an indiscriminate oxidation of roGFP2-based fusion probes by HOCl with fast kinetics in vitro . In line with these observations , abating HOCl production in immune cells with a myeloperoxidase inhibitor significantly attenuated the oxidation of all three probes in bacteria . When bacteria encounter professional phagocytic immune cells , such as neutrophils or macrophages , they are engulfed and phagocytized . Within the phagosome , an intracellular compartment formed during phagocytosis , bacteria are exposed to a complex mixture of toxins , chiefly among them oxidative and nitrosative species ( Hurst , 2012; Segal , 2005; Urban et al . , 2006; Winterbourn and Kettle , 2013 ) . This mechanism , termed respiratory burst , is initiated by the reduction of oxygen to superoxide radicals through NADPH oxidase NOX2 , an enzyme that is assembled at the phagosomal membrane ( Chanock et al . , 1994; Segal and Abo , 1993 ) . From superoxide , other reactive oxygen species ( ROS ) , such as hydrogen peroxide ( H2O2 ) , are formed and released into the phagosomal space ( Hampton et al . , 1998 ) . The respiratory burst is potentiated through lysosomal degranulation that releases myeloperoxidase ( MPO ) and other microbicidal proteins into the phagosome ( Hurst , 2012 ) . MPO catalyzes the formation of hypochlorous acid ( HOCl ) , a strong oxidant , from H2O2 and chloride ions ( Furtmüller et al . , 2003 ) . These ROS can oxidize and damage virtually any cellular molecule , and together with other subsequent mechanisms , ultimately lead to microbial death . In contrast , individuals with chronic granulomatous disease ( CGD ) , a genetic disease with impaired NADPH oxidase activity , as well as mice that lack components of the NADPH oxidase are strongly susceptible to microbial infection ( Hampton et al . , 1998; Holmes et al . , 1967; Mandell , 1974; Segal , 2005; Segal et al . , 2000; Vethanayagam et al . , 2011; Winkelstein et al . , 2000 ) . The mechanisms that kill bacteria in phagocytic immune cells are still not fully understood . However , investigation of the phagosomal environment is quite challenging due to its transient nature and the complexity of the mixture of oxidative species ( Nüsse , 2011 ) . Several methods have been employed for the detection of phagosomal ROS , including fluorescent redox-sensitive dyes . However , most of these methods have a number of limitations such as non-specificity , irreversibility , non-quantitative information , or a lack of subcellular localization ( Nauseef , 2014 ) . The most widespread compound to detect H2O2 in intact cells is 2’ , 7’-dihydrodichlorofluorescein ( H2DCF ) , which can be oxidized to fluorescent 2’ , 7’-dichlorofluorescein ( DCF ) ( Chen et al . , 2010; Maghzal et al . , 2012 ) . DCF oxidation , however , is considered mainly qualitative , as it is observable in the absence of H2O2 and is stimulated by metals , peroxidases , and cytochrome c . Therefore , it does not provide detailed quantitative and compartment-specific information ( Meyer and Dick , 2010; Rota et al . , 1999; Tarpey et al . , 2004 ) . Recently , roGFP2 ( reduction-oxidation-sensitive green fluorescent protein 2 ) has been used to study oxidative and nitrosative stress dynamics in Salmonella inside macrophages ( van der Heijden et al . , 2015 ) . roGFP2 has several advantages when compared to commercially available fluorescent redox-sensitive dyes . As a GFP variant , it can be genetically introduced into virtually any biological system and can be even targeted to specific cellular compartments ( Dooley et al . , 2004; Hanson et al . , 2004 ) . Its redox state , which depends on the redox state of the biological system , can then be measured with the help of an engineered pair of cysteine residues close to the fluorophore . The reversible disulfide bond formation between these cysteines triggers a slight conformational change , which results in a reversible change of the protonation status of the fluorophore . The reduced and oxidized form of roGFP2 therefore have distinct fluorescence excitation maxima at 395 and 490 nm , respectively ( Dooley et al . , 2004 ) . Either the 405/488 nm ratio with laser-based excitation or 390/480 nm ratio on filter-based recording devices can thus be used to directly determine the probe’s redox state ( Meyer and Dick , 2010 ) . This ratiometric approach compensates for variations due to differences in absolute roGFP2 concentrations , allowing for quantitative monitoring . These probes thus allow compartment-specific real-time ratiometric quantification of the intracellular redox status in prokaryotic as well as eukaryotic cells ( Arias-Barreiro et al . , 2010; Bhaskar et al . , 2014; Meyer and Dick , 2010; van der Heijden et al . , 2015 ) . Here , we report the use of three different roGFP2-based fluorescent redox probes to quantitatively track the redox state of bacteria during the phagocytic process . Using the H2O2-sensitive roGFP2-Orp1 probe expressed in the cytoplasm of Escherichia coli , we could show with fluorescence spectroscopy and quantitative fluorescence microscopy that phagocytosis by a neutrophil-like cell line leads to probe oxidation within seconds . Comparison of roGFP2-Orp1’s oxidation kinetics to the oxidation kinetics of the glutathione-specific Grx1-roGFP2 probe and the oxidation kinetics of unfused roGFP2 suggested that the presence of a strong oxidant in the phagosome is over-riding the specificity of the fusion probes . Based on previous in vitro studies and chemical inhibition of myeloperoxidase , we conclude that HOCl is the major reactive species during the onset of the respiratory burst in neutrophils and initiates non-specific thiol oxidation in phagocytized bacteria . When bacteria are phagocytized by professional phagocytic immune cells , they are exposed to a toxic cocktail of reactive oxygen and nitrogen species , including hydrogen peroxide ( H2O2 ) generated by the action of NADPH-oxidase and superoxide dismutase . We wanted to monitor the increase in H2O2 and potentially other reactive species inside the bacterial cell during those host-pathogen interactions in situ . Thus , we expressed the roGFP2 fusion probe roGFP2-Orp1 in Escherichia coli MG1655 . This probe is specifically designed to measure H2O2 in biological systems . We could express roGFP2-Orp1 stably in E . coli from a plasmid ( Figure 1A ) . Using fluorescence spectroscopy , we could determine the oxidation state of the probe inside the E . coli cytoplasm using the ratio between the excitation wavelengths of 405 and 488 nm ( Dooley et al . , 2004; Gutscher et al . , 2008; Hanson et al . , 2004 ) . Addition of the strong oxidant Aldrithiol-2 ( AT-2 , 2 , 2′-Dipyridyl disulfide ) to the bacterial cells led to full oxidation of the probe , while addition of DTT resulted in full reduction ( Figure 1D and G ) . The exposure to reactive species in the phagolysosome could also interfere with the glutathione redox potential ( EGSH ) within the cell . Thus , we introduced an expression plasmid encoding Grx1-roGFP2 into E . coli . Grx1-roGFP2 was specifically designed to measure the glutathione redox potential ( EGSH ) . We could stably express Grx1-roGFP2 in E . coli ( Figure 1B ) , fully reduce it with DTT , and fully oxidize it with AT-2 ( Figure 1E and H ) . Additionally , we expressed unfused roGFP2 in E . coli and performed the same experiments ( Figure 1C , F , I ) . Having established that the probe could be fully reduced and fully oxidized in E . coli , we then tested the response of the roGFP2-Orp1 probe towards exogenous H2O2 . As expected , roGFP2-Orp1 changed its redox state when E . coli was exposed to a bolus of exogenous hydrogen peroxide at concentrations as low as 10 µM ( Figure 1J ) . Within a short period of time , the probe’s redox state returned to the pre-H2O2 steady state , indicating that the bacterial cell can recover from the oxidative insult . Higher concentrations of exogenous hydrogen peroxide led to longer recovery times of the probe , until at concentrations of 10 mM , the probe did no longer recover ( Figure 1J ) . Neutrophil granulocytes are the first line of defense against bacteria in the blood stream . They can phagocytize microorganisms and kill them . One of the main killing factors employed by neutrophils is the production of reactive oxygen and nitrogen species during the so-called oxidative burst ( Segal , 2005; Verchier et al . , 2007 ) . For our experiments , we selected the myeloid PLB-985 cell line that can be differentiated to neutrophil-like cells ( Pivot-Pajot et al . , 2010 ) . To differentiate PLB-985 cells we exposed them to DMSO ( Pivot-Pajot et al . , 2010 ) and added interferon γ ( IFNγ ) . After 5 days of incubation , differentiated cells showed significantly higher expression of CD11b and CD64 than undifferentiated cells , while CD16 and CD66b expression was unaltered ( Figure 2A–D ) . Differentiated cells lost their spherical shape , and the cells showed a morphology typical of neutrophils when stained with May-Grünwald-Giemsa stain ( Figure 2E–H ) . Differentiated cells internalized IgG-opsonized E . coli , demonstrating their phagocytic capacity ( Figure 3A ) . In contrast , the usage of undifferentiated PLB-985 cells or non-opsonized E . coli led to significantly decreased internalization of bacteria , despite their presence in the surrounding medium ( Figure 3B and D ) . When IFNγ was not added during the differentiation process , phagocytosis was less pronounced ( Figure 3C ) . Our microscopic data was corroborated and quantified by flow cytometry ( Figure 3G–I ) . For subsequent experiments , we used DMSO+ IFNγ-differentiated cells and opsonized E . coli , because under those conditions the neutrophil-like cells had the highest phagocytic capacity . We then monitored the oxidation state of roGFP2-Orp1 in E . coli during phagocytosis . Thus , roGFP2-Orp1-expressing E . coli were co-incubated with differentiated neutrophil-like PLB-985 cells and the ratio of fluorescent intensity at 405/488 nm excitation was measured as readout of the probe’s oxidation state in a fluorescence plate reader . The oxidation state of the probe increased , when opsonized E . coli expressing roGFP2-Orp1 were incubated with neutrophil-like cells . The oxidation kinetic showed a gradual increase in the 405/488 nm ratio , as soon as E . coli was incubated with neutrophil-like PLB-985 , reaching a plateau after 60 min of incubation and remaining stably oxidized until the end of the measurement ( 2 hr ) . In contrast , the probe was not oxidized when bacteria were incubated with medium or undifferentiated cells ( Figure 4A ) . When E . coli expressing Grx1-roGFP2 was cultured with neutrophil-like PLB-985 cells , the redox response of the probe was highly similar to the response observed with roGFP2-Orp1 ( Figure 4B ) , and we essentially observed the same oxidation kinetics with the unfused roGFP2 probe ( Figure 4C ) . The oxidation kinetics of the roGFP2-Orp1 probe in E . coli cultured with differentiated PLB-985 cells was not as fast as in an E . coli bacteria suspension that was directly exposed to high concentrations of H2O2 ( compare Figure 1J and Figure 4A ) . We argued that this could mean that bacteria experience redox stress from the host cells only upon phagocytosis . To examine if phagocytosis indeed plays a role in probe oxidation in the E . coli cytoplasm , we blocked phagocytosis by treating the neutrophil-like cells with Cytochalasin D ( CD ) . Cytochalasin D-pre-treated , differentiated PLB-985 cells did phagocytize opsonized E . coli less effectively than untreated , differentiated PLB-985 cells or cells treated with DMSO ( vehicle control ) ( Figure 5A ) . To verify the effective blocking of phagocytosis , we used opsonized FITC-labeled Zymosan and tested the phagocytic capacity of Cytochalasin D-treated and untreated PLB-985 cells . In this case , we quenched the fluorescence of attached Zymosan particles , giving us a more exact readout of the phagocytic capacity . Zymosan phagocytosis was almost completely blocked in Cytochalasin D-treated neutrophil-like cells when compared to the control or cells pre-incubated with DMSO ( vehicle control ) ( Figure 5B ) . Having established the inhibition of phagocytosis , we tested the redox state of roGFP2-Orp1 in E . coli in the presence of differentiated , Cytochalasin D-treated PLB-985 cells . The 405/488 nm ratio of roGFP2-Orp1 did not change significantly over time ( Figure 5C ) . Similarly , oxidation of the Grx1-roGFP2 fusion and unfused roGFP2 probe was phagocytosis-dependent ( Figure 5D & E ) indicating that bacteria are exposed to reactive oxygen species only when internalized . Our data with exogenous hydrogen peroxide demonstrated that the roGFP2-Orp1 probe reacts with very fast kinetics toward hydrogen peroxide . In contrast , the probe’s 405/488 nm ratio in E . coli cells exposed to neutrophil-like cells changed more gradually . In combination with the significantly lower probe oxidation in E . coli exposed to phagocytosis-impaired neutrophil-like cells , we hypothesized that phagocytosis is the rate-limiting step for probe oxidation . To test if the probes are indeed oxidized only upon phagocytosis , we monitored the dynamics of probe oxidation using quantitative fluorescence microscopy . Our experiments show that the probe remains in a reduced state in bacteria not yet phagocytized , but this changes promptly within seconds upon phagocytosis ( Figure 6 , Video 1 ) . The gradual increase in probe oxidation in E . coli exposed to neutrophil-like cells thus reflects the gradual phagocytosis of individual bacteria over time . The initial superoxide generated during the oxidative burst is produced by NADPH oxidase . Superoxide itself is highly reactive , but also the originator of further reactive species that are produced subsequently during the oxidative burst , chiefly among them H2O2 , through superoxide dismutase action ( Dupré-Crochet et al . , 2013; Klebanoff et al . , 2013; Segal et al . , 1980; Segal , 2005; Winterbourn , 2014 ) . Thus , we tested the influence of NOX2 on probe oxidation in E . coli . The generation of reactive species by PLB-985 cells was confirmed using the oxidant-sensitive dye 2' , 7'-Dichlordihydrofluorescein-diacetate ( H2DCFDA ) , which is oxidized intracellularly to 2' , 7'-Dichlorfluorescein ( DCF ) by a number of reactive oxygen species ( Chen et al . , 2010; Yazdani , 2015 ) . Neutrophil-like PLB-985 cells , when stimulated with PMA , showed higher DCF fluorescence than non-stimulated cells ( Figure 7A ) , confirming previous reports that these cells do indeed produce ROS ( Segal et al . , 1980 ) . ROS production was also stimulated by opsonized E . coli ( Figure 7B ) . Based on the previous findings and our own data , we argued that cells lacking NOX2 activity should not be able to induce roGFP2-Orp1 probe oxidation in E . coli cells . PLB-985 cells lacking gp91phox , the catalytic subunit of NOX2 , were indeed unable to induce significant oxidation of roGFP2-Orp1 in E . coli ( Figure 8 ) . Although PLB-985 cells lacking NOX2 activity were incapable to induce significant oxidation of roGFP2-Orp1 in E . coli , the virtually identical behavior of all three roGFP2-based probes in our phagocytosis experiments is inconsistent with superoxide-derived H2O2 as the main factor in probe oxidation . All three probes have different specificities: Unfused roGFP2 , for example , is known to react slowly with H2O2 and glutathione in vitro and thus is not well-suited for the detection of transient and weak oxidative stress ( Meyer and Dick , 2010 ) . Additionally , the Grx1-roGFP2 probe rapidly equilibrates with the cell’s glutathione redox potential , but is not directly affected by hydrogen peroxide ( Gutscher et al . , 2008 ) . However , we showed recently , that , in vitro , the specificity of these three probes breaks down in the presence of strong oxidants . All roGFP2-based probes react similar and with fast kinetics with hypochlorous acid ( HOCl ) and polysulfides in vitro ( Müller et al . , 2017b; 2017a ) . Addition of 10 µM or more HOCl to E . coli expressing any of the three probes led to instant probe oxidation ( Figure 9A–C ) . As activated neutrophils are thought to produce high concentrations of HOCl through the action of myeloperoxidase ( Klebanoff , 2005 ) , we investigated , if this enzyme is involved in the unspecific probe oxidation in phagocytized E . coli . Thus , we used the myeloperoxidase inhibitor 4-aminobenzoic acid hydrazide ( ABAH ) . Pre-treatment of neutrophil-like cells with ABAH resulted in a significant attenuation of the probes’ response ( Figure 10A–C ) and release of reactive species as measured with an 2' , 7'-Dichlordihydrofluorescein-diacetate assay ( Figure 7 ) . Phagocytosis was not affected by ABAH treatment ( Figure 10E–F ) . We thus conclude that hypochlorous acid is the major reactive species that leads to intracellular thiol oxidation as observed in all three roGFP2-based probes . Phagocytosis accompanied by the respiratory burst is an important mechanism by the innate immune system to protect against invading bacteria . However , our knowledge on the early events within the phagosome as well as inside engulfed bacteria is quite limited . This is mainly due to a lack of methods that would allow specific , spatio-temporal , and quantitative measurements of ROS and changes in the cell's redox status ( Balce and Yates , 2013; Kalyanaraman et al . , 2012; Winterbourn , 2014 ) . To close this gap , we set up a phagocytosis assay using neutrophil-like PLB-985 cells and E . coli bacteria expressing roGFP2-based probes in their cytoplasm . Our initial hypothesis was that H2O2 plays a major role in microbial oxidation during early phagocytosis , as NADPH oxidase is the first enzyme in the respiratory burst cascade and the generated superoxide is thought to be quickly converted to hydrogen peroxide ( Segal et al . , 1980; Segal , 2005 ) . We therefore used roGFP2-Orp1 in E . coli to specifically measure H2O2 accumulation in the bacterial cell . The probe was oxidized rapidly inside E . coli in the presence of 1 mM H2O2 , but recovery was observable within approximately 30 min . Permanent probe oxidation was achieved upon addition of 10 mM H2O2 , suggesting that the bacteria are incapable of detoxifying these high concentrations . We then tested the response of roGFP2-Orp1 , but also Grx1-roGFP2 and unfused roGFP2 to co-incubation and phagocytosis by PLB-985 . All probes were effectively oxidized under these conditions and showed virtually the same kinetics . This was somewhat unexpected , since the roGFP2-Orp1 probe typically shows high specificity toward hydrogen peroxide , whereas the Grx1-roGFP2 probe was designed to rapidly and specifically equilibrate with the glutathione redox couple ( Gutscher et al . , 2008 ) : in vitro , H2O2 treatment of purified Grx1-roGFP2 did not lead to significant probe oxidation even at concentrations as high as 100 µM ( Müller et al . , 2017a; 2017b ) . In the same vein , unfused roGFP2 only slowly equilibrates with the glutathione redox couple and did not show significant oxidation by hydrogen peroxide and oxidized glutathione in vivo and in vitro ( Meyer and Dick , 2010; Müller et al . , 2017a; 2017b ) . However , we previously showed that all three probes are effectively oxidized by HOCl in vitro ( Müller et al . , 2017a; 2017b ) , and HOCl is well known to be highly reactive with most thiols ( Peskin and Winterbourn , 2001; Storkey et al . , 2014 ) . To test if this holds true in an in vivo setting in bacteria as well , we treated E . coli expressing any of the three probes with increasing concentrations of HOCl . The response was instant at all concentrations and comparable for all three probes . It is therefore likely that a complex mixture of reactive species , with HOCl as the main component , leads to the oxidation of the probes in E . coli . In a parallel approach , ROS formation in differentiated PLB-985 cells stimulated with PMA was found strongly increased as well . Others have compared the activity of PLB-985 ( and cell line HL-60 , which is essentially the same cell line [Drexler et al . , 2003] ) to that of neutrophils . It was shown that this cell line , when differentiated to neutrophils , expresses slightly less myeloperoxidase than blood neutrophils ( Pivot-Pajot et al . , 2010 ) and had a generally weaker functional response . However , in other studies , superoxide production and MPO activity have been described to be comparable or even exceeding that of human neutrophils ( Hua et al . , 2000; Thompson et al . , 1988 ) . Since HOCl and other reactive species produced during the oxidative burst are ultimately derived from superoxide produced by NADPH-oxidase , roGFP2-based probe oxidation was largely diminished when PLB-985 cells lacking NOX2 activity were used in the co-cultivation assays . Our data also indicates that E . coli experiences the largest amount of oxidative stress within the phagolysosome , since the probe expressed in non-phagocytized bacteria was significantly less oxidized in the same co-cultivation assay ( Figure 6; Video 1 ) . This is somewhat surprising , as especially H2O2 is thought to easily diffuse through membranes ( Dupré-Crochet et al . , 2013 ) . Our hypothesis that HOCl could be responsible for the unspecific roGFP2 probe oxidation in phagocytized E . coli was further substantiated using the myeloperoxidase inhibitor ABAH . Pre-treatment of neutrophil-like cells with ABAH resulted in a significant decrease of the probes’ oxidation and release of reactive species as measured with a DCF assay . Taken together , these results suggest HOCl is most likely the major reactive species responsible for roGFP2 oxidation in E . coli ingested by PLB-985 cells . Our results provide some insights into the nature of ROS released within the phagosome . While production of different ROS at the onset of phagocytosis was known and described before and has been observed using roGFP2 expressed in bacteria ( van der Heijden et al . , 2015 ) , the exact composition and concentrations of individual reactive species as well as their time-resolved release remain largely elusive ( Dupré-Crochet et al . , 2013; Hurst , 2012; Klebanoff et al . , 2013; Winterbourn and Kettle , 2013 ) . As soon as neutrophils ingest opsonized bacteria and the vacuole is formed , degranulation begins within seconds and NADPH oxidase becomes activated and produces superoxide , which dismutates to H2O2 ( Segal et al . , 1980; Segal , 2005 ) . When the granules are fused to the phagosome , MPO levels increase , leading to the formation of HOCl and derived chloramines ( Hurst , 2012; Winterbourn and Kettle , 2013 ) . This HOCl generation in the phagolysosome could be directly observed in porcine neutrophils phagocytizing zymosan particles with the help of a chemical probe selective for HOCl ( Koide et al . , 2011 ) . Due to its high abundance , MPO thus can be regarded as a sink for hydrogen peroxide , which would favor the view that it does not accumulate to concentrations high enough to harm bacteria ( Dupré-Crochet et al . , 2013; Klebanoff et al . , 2013; Winterbourn and Kettle , 2013 ) . Instead , HOCl and potentially HOCl-derived chloramines seem to play the major role in the oxidation of bacterial macromolecules ( Chapman et al . , 2002; Clark and Borregaard , 1985; Klebanoff et al . , 2013; Vissers and Winterbourn , 1987 ) . These results are in agreement with previous studies in which bacterial killing following phagocytosis was analyzed ( Palazzolo et al . , 2005; Schwartz et al . , 2009 ) . GFP bleaching in phagocytized bacteria suggested that cytoplasmic HOCl concentrations were significant ( Hurst , 2012; Winterbourn and Kettle , 2013 ) . Our approach allowed real-time tracking of the roGFP2 oxidation state during phagocytosis and thus provided strong evidence of the presence of HOCl or derived chloramines within the cytoplasm of bacteria within seconds after phagocytosis . Interestingly , most individuals with MPO deficiency do not particularly suffer from microbial infections ( Kutter et al . , 2000; Nauseef , 1988; Parry et al . , 1981 ) . Inhibition of MPO , the enzyme producing HOCl in neutrophils , led to lower probe oxidation but did not fully inhibit oxidation , indicating that ROS produced prior to HOCl can still affect the thiol redox state of proteins in E . coli’s cytoplasm , or alternatively , that ABAH did not fully inhibit MPO , as has been described ( Björnsdottir et al . , 2015; Parker et al . , 2011 ) . In contrast , in the absence of NOX2 , roGFP2 remained fully reduced , indicating that the absence of superoxide , which is needed for the formation of peroxynitrite , H2O2 , and further derived ROS and RNS including HOCl , prevents the breakdown of the bacterial thiol redox state . The human myeloid leukemia cell line PLB-985 ( obtained from DSMZ , German collection of microorganisms and cell culture ) was cultured in RPMI-1640 medium supplemented with 10% FBS and 1% GlutaMAX ( Life Technologies , Carlsbad , CA ) at 37°C in a humidified incubator at 5% CO2 . Cells were authenticated based on their ability to differentiate to neutrophils and the associated expression of the respective surface markers ( see below ) . Their mycoplasma status was not tested by us , however , all cell lines distributed by DSMZ are certified mycoplasma negative . Cell cultures were passaged twice a week to maintain a cell density between 2 × 105 and 1 × 106 · mL−1 . For neutrophil-like phenotype differentiation , exponentially growing cells at a starting density of 2 × 105 · mL−1 were cultured in RPMI 1640 medium supplemented with 10% FBS , 1% GlutaMax and 1 . 25% DMSO for 5 days ( Pivot-Pajot et al . , 2010 ) . The phagocytic function of the PLB-985 cells was stimulated with 2000 U · mL−1 interferon-γ ( ImmunoTools , Friesoythe , Germany ) , added to the culture on day 4 during the differentiation period ( Tlili et al . , 2011 ) . Cell viability was monitored by trypan blue exclusion and was typically >90% . The efficiency of differentiation was estimated by morphological analysis with May-Grünwald-Giemsa stain and flow cytometric analysis of the expression of surface markers CD11b and CD64 using specific phycoerythrin ( PE ) -conjugated antibodies ( eBioscience , San Diego , CA ) . 105 PE stained cells were monitored by flow cytometer BD FACSCanto II ( Becton , Dickinson and Company , Franklin Lakes , NJ ) equipped with three lasers , with blue ( 488 nm , air-cooled , 20 mW solid state ) , red ( 633 nm , 17 mW HeNe ) and violet wavelengths ( 405 nm , 30 mW solid state ) . The red fluorescence ( PE emission ) was collected after passing through a 585/42 nm band pass ( BP ) filter . Data was analyzed using Flow Jo software Version 10 . 2 ( Tree Star Inc . , Ashland , OR ) . Construction of plasmid pCC_roGFP2 for expression of unfused roGFP2 ( Table 1 ) in E . coli was described previously ( Müller et al . , 2017a ) . For the expression of roGFP2-Orp1 and Grx1-roGFP2 in E . coli , the respective gene regions were amplified using the primer pairs listed in Table 1 from pQE-based vectors that served as template ( Table 2 ) . Subsequently , the PCR products were cloned into empty pCC using the restriction enzymes NdeI and EcoRI . pCC expresses proteins in E . coli with an IPTG-inducible Tac promoter ( Masuch et al . , 2015 ) . E . coli XL1 blue was used as a cloning host . The pCC vectors containing the three redox-sensitive fluorescent probes ( roGFP2 , Grx1-roGFP2 , and roGFP2-Orp1 ) were subsequently transformed into E . coli MG1655 and 100 µg/mL ampicillin was added to the growth medium for maintenance of the plasmid ( Table 2 ) . E . coli strains harboring pCC vectors containing the roGFP2-based probes ( Table 2 ) were cultured in LB liquid medium with 200 µg/mL of ampicillin at 37°C overnight . The optical density at 600 nm ( OD600 ) was measured and the bacterial suspension was diluted to an OD600 of 0 . 1 with fresh medium and cultured at 37°C for ~2 hr until an OD600 of 0 . 5–0 . 8 was reached . The expression of roGFP2-based probes was then induced with 100 µM IPTG and the culture was incubated at 20°C overnight . These bacterial cells , now containing roGFP2-based probes , were then washed twice in 40 mM HEPES buffer ( pH 7 . 4 ) and re-suspended in 1 mL HEPES buffer to a final OD600 of 0 . 3 . The fluorescence intensity was measured in an FP-8500 spectrofluorometer ( Jasco , Tokyo , Japan ) . The emission wavelength was fixed at 510 nm and excitation wavelength was scanned from 350 to 500 nm . Bandwidths of excitation and emission were set to 5 nm . The cell suspension in the cuvette was continuously stirred with a magnetic stir bar and the temperature of the temperature controller EHC-813 ( Jasco ) was set to 25°C . Fluorescence excitation ratios ( 405/488 nm ) were used as measurement of probe oxidation ( Arias-Barreiro et al . , 2010; Meyer and Dick , 2010 ) . Oxidation with 100 µM aldrithiol-2 ( AT-2 , Sigma-Aldrich , St . Louis , MO ) and reduction with 50 mM dithiothreitol ( DTT , Sigma-Aldrich ) were used to fully oxidize and fully reduce the probes , respectively ( Figure 1D–F ) . E . coli cells expressing roGFP2-based probes as described above were washed twice in PBS pH 7 . 4 and resuspended in PBS pH 7 . 4 with 0 . 5% FBS to a final OD600 of 0 . 1–0 . 3 . FBS was omitted in experiments where no PLB-985 cells were present ( Figure 1J and Figure 8 ) . Fifty microliters of this E . coli suspension were placed in the wells of a black , clear-bottom 96-well plate ( Nunc , Rochester , NY ) . Fluorescence intensity was recorded during every minute for 10 min using the Synergy H1 multi-detection microplate reader ( Biotek , Bad Friedrichshall , Germany ) at the excitation wavelengths 405 and 488 nm and emission wavelength at 530 nm . Then 50 µL of the selected chemical solution or the respective PLB-985 cell suspension was added to the wells and the fluorescence intensity was monitored for another 2 hr under the same conditions . The signals of fully oxidized and fully reduced probes were obtained by adding 100 µM AT-2 or 50 mM DTT to the bacteria culture at the start of the 2 hr measurement . The fluorescence excitation ratios ( 405/488 nm ) were used as measurement of probe oxidation and all values were normalized to the values obtained for fully oxidized ( AT-2-treated ) and for fully reduced ( DTT-treated ) bacteria cultures . E . coli cells expressing roGFP-based probes ( as described above ) were opsonized by incubation for 30 min at 37°C with 1 mg . mL−1 human immunoglobulin G ( Sigma-Aldrich ) in PBS pH 7 . 4 , then washed twice in PBS and re-suspended in PBS with 0 . 5% FBS to an OD600 of 0 . 1 . 50 µL of opsonized E . coli was then added to 50 µL of differentiated PLB-985 cell suspension ( 107 mL−1 ) in PBS with 0 . 5% FBS . Thus a ratio of 10 E . coli bacteria per PLB-985 cell was used to initiate phagocytosis . Cells and bacteria were then co-incubated at 37°C , and phagocytosis was stopped by adding 100 µL of ice-cold PBS at certain time points . Cells were fixed with 4% paraformaldehyde and samples were kept on ice until subjected to flow cytometry . Samples were analyzed using a BD FACSCanto II flow cytometer ( Becton , Dickinson and Company ) with an argon laser operating at 488 nm using the 530/30 emission filter to detect the fluorescence of the roGFP2-based probes expressed in phagocytized bacteria . For each sample , a total of 10 , 000 viable cells were counted . The mean fluorescence intensity ( MFI ) multiplied by the percentage of viable cells that had ingested fluorescent particles was used to evaluate the phagocytic capacity of PLB-985 cells . In some experiments , phagocytosis was inhibited by treatment of differentiated PLB-985 cells with 250 µM cytochalasin D ( Sigma-Aldrich ) for 30 min at 37°C before co-incubation with opsonized bacteria . Paraformaldehyde-fixed cells from the phagocytosis assay were centrifuged at 200 g for 5 min and resuspended in PBS containing 0 . 1% Tween 20 and 2 . 5 µg . mL−1 TRITC-conjugated phalloidin ( Sigma-Aldrich ) . Cells were incubated with phalloidin for 40 min at room temperature , protected from light . Excess phalloidin was removed by centrifugation and cells were resuspended by pipetting up and down in buffered glycerol containing 0 . 5 µg/mL DAPI ( Sigma-Aldrich ) . Cells were then visualized on a slide in a BX51 fluorescence microscope with a U‐UCD8 condenser , a U‐LH100HGAPO burner , a U‐RFL‐T power supply , a 63X/1 . 4 NA oil objective and a 450–490 nm excitation/500–550 emission bandpass filter ( Olympus , Tokyo , Japan ) . Images were collected with a CC12 digital color camera and the Cell Imaging Software ( Olympus ) and composite figures were prepared using ImageJ ( Schneider et al . , 2012 ) and Photoshop CS5 ( Adobe Systems , San Jose , CA ) software . Differentiated PLB-958 cells were stained with 0 . 25 µM Celltracker Deep Red ( Thermo Fisher Scientific , USA ) in RPMI 1680 at 37°C and 5% CO2 for 30 min , washed once with PBS and diluted in PBS with 0 . 5% FBS to a final concentration of 107 cells . mL−1 . 1 mL of the cell suspension was poured onto an imaging dish ( µ-Dish 35 mm , high , Ibidi , DE ) . Opsonized roGFP2-Orp1 E . coli cells were added with a ratio of five E . coli cells to one PLB-958 cell . Fluorescence images were acquired with an LSM 880 ELYRA PS . 1 microscope ( Carl Zeiss Microscopy GmbH , Jena , Germany ) . Images were acquired in three different channels: channel I: Ex405nm/Em513nm , channel II: Ex488nm/Em513nm , channel III: Ex561nm/Em674nm , bandwidth settings channels I and II: 13 nm , channel III: 59 nm . Individual single channel images were exported using ZEN 2 . 1 ( Zeiss , DE ) . Ratiometric images were generated with ImageJ 1 . 51e ( National Institutes of Health , USA ) as described ( Collins , 2007 ) . The image background was corrected using a rolling ball algorithm and images were transferred to 32-bit format . Images were thresholded and converted to binary mask . Images from channel I and channel II were aligned using the ImageJ plugin „MultiStackReg“ and ratiometric images were calculated using „Ratio Plus“ . Ratiometric images were colored using the „Lookup table“ feature from the plugin „NucMed“ . The display range of all ratiometric images were adjusted to the same range before converting them into RGB format . For the ratiometric time series assembly , images were smoothed in order to reduce noise . After background subtraction , normalized 405/488 nm ratio image series were calculated and assembled to a movie using Software kindly provided by Fricker ( 2016 ) . Intracellular oxidation of 2' , 7'-dichlorodihydrofluorescein ( DCFH ) to 2' , 7'-dichlorofluorescein ( DCF ) by PLB-985 cells was analyzed using the cell permeable derivative 2' , 7'-dichlorodihydrofluorescein diacetate ( DCFH-DA ) . PLB-985 cells ( 107 . mL−1 ) were pre-incubated with 1 . 25 µM DCFH-DA ( Thermo Fisher Scientific , Waltham , MA ) in PBS with 0 . 5% FBS during 15 min at 37°C . Then 50 µL were placed in the wells of a black , clear-bottom 96-well plate ( Nunc ) . Fluorescence intensity was recorded every 1 min for 10 min using the Synergy H1 multi-detection microplate reader ( Biotek ) at an excitation wavelength of 488 nm and an emission wavelength of 525 nm . Cells were then activated by addition of 50 ng · mL−1 phorbol 12-myristate 13-acetate ( PMA; Sigma-Aldrich ) or incubation with 10-fold excess of E . coli bacteria . Fluorescence intensity was then recorded every 1 min for 1 hr . Myeloperoxidase activity in neutrophil-like PLB-985 cells was inhibited by pre-incubation of cells with 500 µM 4-aminobenzoic acid hydrazide ( ABAH; Sigma-Aldrich ) for 30 min at 37°C prior to subsequent experiments . All 96-well-plate-based fluorescence measurement experiments were repeated at least three times with biologically independent replicates and the results are expressed as the mean ±standard deviation as represented by error bars . Representative data is shown for fluorescence spectroscopy , flow cytometry and microscopy image data .	A group of cells of the immune system defends the body against infections by wrapping themselves around bacteria , and effectively ‘eating’ them . During this process , called phagocytosis , the cell also douses the bacterium with a deadly cocktail of chemicals , including an antiseptic – hydrogen peroxide – and bleach . This mixture chemically burns , and then kills , the invader . The immune cells create hydrogen peroxide and bleach through chemical reactions that require two enzymes , NOX2 and MPO . The NOX2 enzyme is activated first , and produces a compound which is then transformed into hydrogen peroxide . In turn , hydrogen peroxide is used by MPO to make bleach . Phagocytosis is still poorly understood , and difficult to study: for example , it is not clear when the toxic mix is released , and which of its components are the most important . Here , Degrossoli et al . peer into this process: to do so , they genetically engineer bacteria and give them a built-in chemical burn tracker . The bacteria are made to carry fluorescent proteins which normally glow under blue light , but start to also react to violet light if they are exposed to a chemical burn . Under the microscope , when these bacteria encounter immune cells , they start glowing under violet light only a few seconds after they have been phagocytized . This shows that , during phagocytosis , the chemical mix is used almost immediately . The new technique also reveals that cells without a working NOX2 enzyme – which cannot produce hydrogen peroxide – could not burn the bacteria . However , hydrogen peroxide is also used by MPO to create bleach . If just MPO is deactivated , the cells can burn the bacteria , but much less efficiently . This , and the speed with which these fluorescent proteins were burnt , shows that the bleach is the main component of the toxic mix used during phagocytosis . Chronic granulomatous disease is a condition where patients can have a faulty version of NOX2 , which makes it harder for them to fight infection . Understanding the mechanisms and the enzymes associated with phagocytosis could lead to improved treatment in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "microbiology", "and", "infectious", "disease" ]	2018	Neutrophil-generated HOCl leads to non-specific thiol oxidation in phagocytized bacteria
Misfolded proteins in the lumen of the endoplasmic reticulum ( ER ) are retrotranslocated into the cytosol and polyubiquitinated before being degraded by the proteasome . The multi-spanning ubiquitin ligase Hrd1 forms the retrotranslocation channel and associates with three other membrane proteins ( Hrd3 , Usa1 , Der1 ) of poorly defined function . The Hrd1 channel is gated by autoubiquitination , but how Hrd1 escapes degradation by the proteasome and returns to its inactive ground state is unknown . Here , we show that autoubiquitination of Hrd1 is counteracted by Ubp1 , a deubiquitinating enzyme that requires its N-terminal transmembrane segment for activity towards Hrd1 . The Hrd1 partner Hrd3 serves as a brake for autoubiquitination , while Usa1 attenuates Ubp1’s deubiquitination activity through an inhibitory effect of its UBL domain . These results lead to a model in which the Hrd1 channel is regulated by cycles of autoubiquitination and deubiquitination , reactions that are modulated by the other components of the Hrd1 complex . Proteins translocated into the endoplasmic reticulum ( ER ) undergo quality control , such that only folded proteins are moved through the secretory pathway . If a protein does not reach its native folded state , it is eventually retrotranslocated across the ER membrane , polyubiquitinated , extracted from the membrane , and degraded by the proteasome , a pathway called ER-associated protein degradation ( ERAD ) ( for recent reviews , see Berner et al . , 2018; Preston and Brodsky , 2017; Wu and Rapoport , 2018 ) . Work in S . cerevisiae showed that substrates use four distinct ERAD pathways , depending on the localization of their misfolded domains . ERAD-L substrates contain misfolded domains in the ER lumen , ERAD-M substrates are misfolded within the membrane , ERAD-C substrates are membrane proteins with misfolded cytosolic domains , and ERAD-INM handles misfolded proteins in the inner nuclear membrane . These pathways use different ubiquitin ligases: ERAD-L and -M use the Hrd1 ligase , ERAD-C the Doa10 ligase , and ERAD-INM the Asi ligase complex ( Carvalho et al . , 2006; Foresti et al . , 2014; Huyer et al . , 2004; Khmelinskii et al . , 2014; Vashist and Ng , 2004 ) . These ligases are multi-spanning membrane proteins with cytosolic RING finger domains . Following polyubiquitination , all pathways converge at the Cdc48 ATPase ( p97 or VCP in mammals ) ( Bays et al . , 2001; Jarosch et al . , 2002; Rabinovich et al . , 2002; Ye et al . , 2001 ) . This ATPase cooperates with a cofactor ( Ufd1/Npl4 ) to extract polyubiquitinated substrates from the membrane ( Stein et al . , 2014 ) . Among the ubiquitin ligases , the function of Hrd1 is best understood . Hrd1 forms a complex with three other membrane proteins ( Hrd3 , Usa1 , Der1 ) ( Carvalho et al . , 2006; Denic et al . , 2006; Gardner et al . , 2000 ) . Hrd3 is a single-spanning membrane protein with a large lumenal domain that interacts with substrates and Hrd1 ( Gauss et al . , 2006a; Gauss et al . , 2006b ) . In the absence of Hrd3 , Hrd1 is strongly autoubiquitinated and rapidly degraded ( Gardner et al . , 2000 ) . Usa1 is a double-spanning membrane protein that serves as a linker between Hrd1 and Der1 and facilitates the oligomerization of Hrd1 ( Carvalho et al . , 2010; Horn et al . , 2009 ) . It also has a ubiquitin-like ( UBL ) domain of poorly defined function; the UBL domain is dispensable for the degradation of ERAD substrates , but is required for the efficient degradation of Hrd1 in a hrd3Δ strain ( Carroll and Hampton , 2010; Vashistha et al . , 2016 ) . Der1 is a multi-spanning protein required for ERAD-L , but not ERAD-M; it probably recognizes misfolded substrates in the ER lumen and facilitates their insertion into Hrd1 ( Knop et al . , 1996; Mehnert et al . , 2014 ) . Recent results suggest that the Hrd1 ligase forms a protein-conducting channel ( Baldridge and Rapoport , 2016 ) . Overexpression of Hrd1 in S . cerevisiae cells bypasses the requirement for the other components of the complex , while all downstream components , such as the ubiquitination machinery and Cdc48 ATPase complex , are still needed ( Carvalho et al . , 2010 ) . These results suggest that Hrd1 is the only essential membrane protein for a basic ERAD-L process . A cryogenic electron microscopy ( cryo-EM ) structure shows that the membrane-spanning segments of Hrd1 surround a deep aqueous cavity , supporting the idea that Hrd1 can form a channel ( Schoebel et al . , 2017 ) . In vitro experiments further demonstrate that Hrd1 reconstituted into proteoliposomes allows a misfolded substrate domain to retrotranslocate across the lipid bilayer ( Baldridge and Rapoport , 2016 ) . This process requires autoubiquitination of Hrd1 , leading to the suggestion that Hrd1 forms a ubiquitin-gated channel . The important autoubiquitination event occurs in the RING finger domain , as mutation of crucial lysines in this domain blocks retrotranslocation in vitro and ERAD-L in vivo ( Baldridge and Rapoport , 2016 ) . If the Hrd1 channel is activated by autoubiquitination , how is Hrd1 spared from degradation and returned to its inactive ground state ? Here , we identify Ubp1 as a membrane-bound deubiquitinating enzyme ( DUB ) that reverses the polyubiquitin modification of Hrd1 and allows Hrd1 to escape uncontrolled degradation . The Hrd1 partner Hrd3 serves as a brake for autoubiquitination , while the UBL domain of Usa1 attenuates Ubp1’s activity , allowing Hrd1 autoubiquitination and activation . This delicate balance allows Hrd1 to undergo cycles of autoubiquitination and deubiquitination during ERAD . Our previous experiments indicated that Hrd1 is autoubiquitinated in wild-type yeast cells ( Baldridge and Rapoport , 2016 ) . The protein is moderately stable , with a half-life of about 100 min . We therefore reasoned that overexpression of a DUB that reverses the modification of Hrd1 would increase the steady state levels of the ligase . We overexpressed 23 different DUBs in yeast cells that also express Flag-tagged Hrd1 . The levels of Hrd1 were determined by immunoblotting with anti-Flag antibodies ( Figure 1A ) . The strongest increase of Hrd1 levels was seen with Ubp1 , even though several other DUBs were expressed at a higher level than Ubp1 ( Figure 1—figure supplement 1A ) . Ubp1 overexpression had no effect on the levels of Hrd3 or Usa1 ( Figure 1—figure supplement 1B , C ) . A more sensitive analysis of Ubp1 function can be performed with hrd3Δ cells , in which Hrd1 autoubiquitination and degradation are greatly accelerated ( half-life of about 30 min ) . Indeed , while Hrd1 levels are very low in hrd3Δ cells ( Figure 1B; lane 3 ) , Ubp1 overexpression increased Hrd1 to about the same level as in wild-type cells ( Figure 1B; lane 8 versus 1 ) . This increase was not seen when an enzymatically inactive mutant of Ubp1 ( C110S ) was overexpressed ( Figure 1B , lane 13 ) , indicating that the deubiquitination activity of Ubp1 is required for Hrd1 stabilization . An increase of Hrd1 was also seen when Ubp1 was overexpressed in cells lacking Usa1 or Der1 ( Figure 1B; lanes 9 versus 4 and lane 10 versus 5 ) . Cycloheximide-chase experiments confirmed that Hrd1 is rapidly degraded in hrd3Δ cells and becomes more stable in cells overexpressing Ubp1 ( Figure 1C ) . Overexpression of Ubp1 also inhibited Hrd1 degradation in cells lacking Der1 or Usa1 ( Figure 1—figure supplement 2A , B ) . Enzymatically inactive Ubp1 ( C110S ) had no effect on Hrd1 degradation ( Figure 1C; Figure 1—figure supplement 2A , B ) . Furthermore , wild-type Ubp1 had no effect when co-expressed with an enzymatically inactive Hrd1 mutant ( Figure 1D ) , indicating that Ubp1 reverses autoubiquitination of Hrd1 . Consistent with this conclusion , the deletion of Ubp1 , which should increase ubiquitination of Hrd1 , decreased the levels of Hrd1 ( Figure 1E; lanes 9–12 versus 1–4 ) . As expected , the additional deletion of the stabilizing Hrd1-partner Hrd3 further reduced Hrd1 levels ( Figure 1E; lanes 13–16 ) . To further test whether Ubp1 reverses autoubiquitination of Hrd1 , we purified His-tagged Hrd1 under denaturing conditions and subjected the material to SDS-PAGE and immunoblotting with anti-ubiquitin antibodies ( Figure 1F ) . Indeed , the level of polyubiquitinated Hrd1 was reduced in wild-type or hrd3Δ cells overexpressing Ubp1 , even though the levels of unmodified Hrd1 were increased ( Figure 1F ) and the total level of ubiquitinated proteins in the cell remained unchanged ( Figure 1—figure supplement 2C ) . It is important to note that these samples were not treated with proteasomal inhibitors as in previous experiments ( Carroll and Hampton , 2010 ) to prevent the accumulation of ubiquitinated Hrd1 in hrd3Δ cells . To determine whether Hrd1 and Ubp1 interact in vivo , we performed co-immunoprecipitation experiments from yeast cells expressing Hrd1-Flag and Ubp1-V5 . Indeed , a small , but reproducible amount of Hrd1-Flag was precipitated with antibodies to the V5 epitope ( Figure 1G ) . The low level of co-precipitation may be explained by the fact that Ubp1 is 20-fold more abundant than Hrd1 ( Figure 1—figure supplement 2D ) . The specificity of the interaction is supported by control experiments with Orm2 , another integral membrane of the endoplasmic reticulum that is approximately 20 times more abundant than Hrd1: Ubp1-V5 did not precipitate Orm2-Flag ( Figure 1G ) . The interaction of Hrd1 with Ubp1 was maintained in the absence of Hrd3 or Usa1 ( Figure 1—figure supplement 2E ) . To test whether Ubp1 affects the function of Hrd1 in ERAD , we tested the degradation of the ERAD-L substrate carboxypeptidase Y* ( CPY* ) ( Finger et al . , 1993 ) in cycloheximide-chase experiments . In the absence of Ubp1 ( Figure 1H , solid square with dashed black line ) , CPY* degradation was slower than in cells expressing Ubp1 ( Figure 1H , circles with solid black line ) . Importantly , overexpression of Ubp1 did not prevent degradation of CPY* ( Figure 1H , triangles with red line ) or the steady-state levels of CPY* ( Figure 1—figure supplement 1D ) . The degradation of the overexpressed ERAD-L substrate , GFP-CPY* was also promoted by Ubp1 expression ( in Figure 1—figure supplement 3A ) . Surprisingly , Ubp1 was not required for ERAD-M and overexpression of Ubp1 only slightly slowed degradation of ERAD-M substrates ( Figure 1—figure supplement 3B and C ) . These data support the idea the Ubp1 is promoting optimal ERAD-L , but not ERAD-M . Taken together , these results demonstrate that Hrd1 and Ubp1 interact with one another , and that Ubp1 counteracts the autoubiquitination activity of Hrd1 thereby inhibiting Hrd1 degradation and regulating Hrd1 function . Ubp1 contains an N-terminal transmembrane segment followed by a cytosolic linker and catalytic domain ( Figure 2A ) . Ubp1 has been reported to be expressed as two distinct polypeptides; one with the transmembrane segment and the other lacking it ( called ΔTM in the scheme in Figure 2A ) , generated from an internal translation start site ( Met67 ) ( Schmitz et al . , 2005 ) . When Ubp1 lacking the transmembrane segment was overexpressed in hrd3Δ cells , no Hrd1 stabilization was observed ( Figure 2B ) . However , the transmembrane segment of full-length Ubp1 could be replaced by the unrelated transmembrane segment of the Cue4 protein ( Figure 2C ) . Even when the entire N-terminal segment preceding the catalytic domain was replaced by a segment derived from Cue4 , Ubp1 retained activity in stabilizing Hrd1 ( Figure 2C ) . These results demonstrate that the role of the Ubp1 transmembrane segment is primarily to anchor Ubp1 at the ER membrane , which likely facilitates its interaction with the multi-spanning membrane protein Hrd1 . In fact , the membrane anchor of Ubp1 could be replaced by the Hrd1-interacting H-domain of Usa1; this soluble fusion protein was able to completely stabilize Hrd1 ( Figure 2C ) . The enzymatically inactive version of this construct ( H-domain Ubp1 ( C110S ) ) had a small effect on Hrd1 degradation ( Figure 2—figure supplement 1A ) , perhaps due to its overexpression . Together , these results suggest that Hrd1 stabilization is primarily achieved by H domain-dependent recruitment of the catalytic domain of Ubp1 . Control experiments showed that all Ubp1-constructs were expressed at about the same level ( Figure 2—figure supplement 1B ) . To further test the specificity of Ubp1 , we used the cytosolic mammalian DUB Usp2 ( Baker et al . , 2005 ) . Both Ubp1 and Usp2 are members of the ubiquitin-specific protease ( USP ) family that generally has low ubiquitin-chain linkage specificity ( Mevissen and Komander , 2017 ) . Overexpression of the catalytic core of Usp2 had only a minor effect on Hrd1 stability ( Figure 2D ) . Similar results were obtained with a fusion protein containing the transmembrane domain of Ubp1 followed by Usp2 ( TM Usp2; Figure 2D ) , indicating that mere membrane targeting of the DUB is insufficient to stabilize Hrd1 . However , stabilization was observed when Usp2 was targeted directly to Hrd1 by the Hrd1-interacting H domain of Usa1 ( Figure 2D ) , even though this construct was expressed at a lower level than transmembrane Usp2 ( Figure 2—figure supplement 1B ) . Next , we tested if the Ubp1 chimeras co-immunoprecipitated with Hrd1 . Both wild-type Ubp1 and TM ( Cue4 ) Ubp1 interacted weakly with Hrd1 ( Figure 2E ) . In contrast , Ubp1 missing its transmembrane segment ( Ubp1 ΔTM ) did not interact with Hrd1 but a soluble version of Ubp1 carrying the H-domain of Usa1 showed strong interaction with Hrd1 ( Figure 2E ) . Thus , there is a general correlation between Ubp1’s Hrd1 interaction and its effect on Hrd1 stabilization . Together , these results suggest that wild-type Ubp1 is targeted to Hrd1 by both its transmembrane segment and the USP domain . They also provide further support for the idea that Ubp1 acts on polyubiquitinated Hrd1 . Next , we analyzed the effect of Ubp1 overexpression on substrate degradation in yeast strains lacking various ERAD components . As reported previously ( Bordallo et al . , 1998; Buschhorn et al . , 2004; Carvalho et al . , 2006; Denic et al . , 2006; Knop et al . , 1996 ) , the degradation of CPY* was drastically reduced when either Hrd1 , Usa1 , Der1 , or Hrd3 were absent ( compare dashed lines in Figure 3A with those in Figure 3B–E ) , confirming that these components are all required for ERAD-L . Ubp1 overexpression only had an effect in hrd3Δ cells , where it rescued CPY* degradation ( solid line in Figure 3E versus those in Figure 3A–D ) . The degradation rate approached that in wild-type cells ( Figure 3A ) . These results suggest that Hrd3’s major function is to inhibit the ubiquitination activity of Hrd1 . Importantly , bypassing Hrd3 by Ubp1 overexpression does not obviate the need for the other ERAD-L components in CPY* degradation ( Figure 3A–D ) . To further test the function of Ubp1 , we overexpressed Ubp1 in hrd1Δhrd3Δ cells together with wild-type Hrd1 . As expected from the results presented before , Ubp1 overexpression stabilized Hrd1 ( Figure 3F ) and allowed degradation of CPY* ( Figure 3G ) , even in the absence of Hrd3 . A Hrd1 mutant ( Hrd1 KRK; Hrd1 with lysine to arginine substitutions at amino acids 319 , 325 , 366 , 368 , 370 , 371 , 373 , 387 , 407 ) that cannot be autoubiquitinated in its RING finger domain and is therefore inactive in retrotranslocation in vitro and in vivo ( Baldridge and Rapoport , 2016 ) , was also stabilized by Ubp1 overexpression ( Figure 3H ) . This is consistent with the fact that the protein can still be polyubiquitinated in other domains ( Baldridge and Rapoport , 2016; Stein et al . , 2014 ) . However , as expected , the higher Hrd1 KRK levels did not result in CPY* degradation ( Figure 3I ) . Ubp1 overexpression had little effect on the stability of catalytically inactive Hrd1 ( C399S ) ( Figure 3J ) or on the degradation of CPY* in Hrd1 ( C399S ) expressing cells ( Figure 3K ) , consistent with the notion that Ubp1 acts on polyubiquitinated Hrd1 . Because Ubp1 does not accelerate the degradation of CPY* in the presence of retrotranslocation-deficient Hrd1 KRK or inactive Hrd1 ( C399S ) ( Figure 3I and K ) , we conclude that Ubp1 overexpression does not bypass normal Hrd1 function . Stabilization of wild-type Hrd1 by Ubp1 overexpression also resulted in the accelerated degradation of the ERAD-M substrate Erg3 ( Figure 3L ) ( Christiano et al . , 2014; Jaenicke et al . , 2011 ) . Surprisingly however , the Hrd1 KRK variant was also able to mediate the degradation of Erg3 ( Figure 3M ) , in contrast to the ERAD-L substrate CPY* ( Figure 3I ) . Overexpression of Ubp1 did not bypass Hrd1 function because inactive Hrd1 ( C399S ) was unable to support the degradation of Erg3 ( Figure 3N ) . It therefore appears that autoubiquitination of Hrd1 is not required for the degradation of this ERAD-M substrate . Previous reports demonstrated that Usa1 is required for Hrd1 autoubiquitination and degradation in a hrd3Δ strain ( Carroll and Hampton , 2010; Vashistha et al . , 2016 ) . This activity could be assigned to the UBL domain of Usa1 , but it remained unclear how this domain affects Hrd1 stability ( Carroll and Hampton , 2010 ) . We wondered if the UBL domain might inhibit Ubp1’s DUB activity; in its absence ( or absence of the entire Usa1 protein ) , Ubp1 would effectively reduce ubiquitination and thus degradation of Hrd1 . To test this idea , we used a hrd3Δusa1Δ strain , in which Hrd1 is much more stable than in a hrd3Δ single deletion strain ( Carroll and Hampton , 2010; see also Figure 4—figure supplement 1A ) . Nevertheless , there was still some Hrd1 autoubiquitination and degradation in this strain , as overexpression of wild-type Ubp1 , but not of enzymatically inactive Ubp1 ( C110S ) , resulted in increased Hrd1 levels ( Figure 4A , lanes 5–8 versus 9–12 ) . Overexpression of Usa1 lacking its UBL domain ( Usa1 ΔUBL ) drastically stabilized Hrd1 ( lanes 21–24 ) , in contrast to wild-type Usa1 , which only had a small effect on Hrd1 stability ( lanes 17–20 ) . Thus , the UBL domain of Usa1 indeed seems to attenuate Ubp1 activity . The co-overexpression of Usa1 and Ubp1 in a hrd3Δ strain had little effect on Hrd1 levels and Hrd1 degradation compared to overexpression of Ubp1 alone ( Figure 4B , lanes 5–8 versus 1–4 ) . However , when Usa1 ΔUBL was co-expressed with Ubp1 , Hrd1 was strongly stabilized ( lanes 9–12 ) , indicating that the UBL domain of Usa1 indeed inhibits Ubp1 activity . In a triple-deletion mutant ( hrd3Δubp1Δusa1Δ ) , Hrd1 was rapidly degraded ( Figure 4C , lanes 1–4 ) , but could be stabilized by its partner protein Hrd3 ( lanes 5–8 ) or by increasing deubiquitination through Ubp1 overexpression ( lanes 9–12 ) . However , as expected from our model , Usa1 or Usa1 ΔUBL had no effect in the absence of Ubp1 ( lanes 17–20 and 21–24 ) . When Ubp1 was expressed from its endogenous promoter in the triple-deletion mutant ( hrd3Δubp1Δusa1Δ ) , the additional expression of Usa1 ΔUBL caused stabilization of Hrd1 ( Figure 4D , lanes 9–12 ) , whereas expression of wild-type Usa1 only had a small effect ( Figure 4D , lanes 5–8 ) . Collectively , these results support the idea that the UBL domain of Usa1 inhibits the activity of Ubp1 . To test whether Hrd1 activity was affected by Usa1 or Usa1 ΔUBL , we followed the degradation of both ERAD-L and ERAD-M substrates . Overexpression of Usa1 partially inhibited CPY* turnover ( Figure 4—figure supplement 1B , lanes 9–12 and Carvalho et al . , 2010 ) and overexpression of Usa1 ΔUBL had a slightly stronger effect ( Figure 4—figure supplement 1B , lanes 13–16 ) . To test whether this effect was limited to ERAD-L substrates , we followed the degradation of ERAD-M substrates . We tested the constitutive degradation of the Hmg2 non-responder mutant ( Hmg2-NR1-GFP; substitution of amino acids 348–352 TFYSA to ILQAS ) and also the regulated degradation of Hmg2-GFP by flow cytometry ( Shearer and Hampton , 2005 ) . Degradation of these ERAD-M substrate still required Hrd1 but was unaffected by the overexpressing Usa1 or Usa1 ΔUBL ( Figure 4—figure supplement 1C and D ) . The fact that overexpression of Usa1 and Usa1 ΔUBL disrupted ERAD-L , but not ERAD-M , suggests that ERAD-L was likely compromised by overexpressed Usa1 sequestering Der1 ( Carvalho et al . , 2010; Horn et al . , 2009 ) . Taken together , these results support the idea that the UBL domain does not affect substrate degradation directly ( Carroll and Hampton , 2010; Vashistha et al . , 2016 ) . Our results reveal a novel mechanism by which the central ERAD component Hrd1 is regulated ( Figure 4E ) . Previous experiments had shown that Hrd1 is activated for ERAD by autoubiquitination; polyubiquitinated Hrd1 allowed a polypeptide segment of the substrate to move across the membrane ( Baldridge and Rapoport , 2016; Stein et al . , 2014 ) . This mechanism , however , raised the question of how polyubiquitinated Hrd1 escapes degradation and returns to its inactive ground state . Our results now indicate that Ubp1 , a membrane-bound DUB , is responsible for resetting Hrd1 ( Figure 4E ) . Ubp1 counteracts Hrd1’s autoubiquitination and therefore reduces its degradation ( Figure 1B–F ) , while also regulating Hrd1’s activity in ERAD-L ( Figure 1H and Figure 1—figure supplement 3A–C ) . Ubp1 is an unusual DUB because it has an N-terminal transmembrane segment that is required for its activity towards Hrd1 . However , there seem to be additional , so far unidentified , determinants of specificity within the catalytic domain of Ubp1 ( Figure 2 ) . The activity of Ubp1 must be tightly regulated , as excessive deubiquitination would prevent Hrd1 autoubiquitination and activation , and inadequate deubiquitination would result in Hrd1 degradation . Our results show that Usa1 , a component associated with Hrd1 , regulates Ubp1 activity; Usa1’s UBL domain attenuates the activity of Ubp1 and thereby allows autoubiquitination of Hrd1 ( Figure 4E ) . Autoubiquitination is also regulated by Hrd3 , a lumenal binding partner of Hrd1 ( Gardner et al . , 2000 ) . Hrd3 serves as a brake of Hrd1’s ubiquitination activity ( Figure 4E ) . We propose that substrate binding to Hrd3 releases this brake ( Figure 4E ) , allowing autoubiquitination of Hrd1 , which in turn opens the channel for retrotranslocation of the substrate . Hrd3 is not absolutely required for a basic ERAD process , as substrates are still degraded in its absence when Hrd1 is stabilized by Ubp1 overexpression . Taken together , our results indicate that Hrd1 undergoes cycles of autoubiquitination and deubiquitination , which are regulated by Hrd1-interacting components . Surprisingly , we found that the ERAD-M substrate Erg3 does not require Hrd1 autoubiquitination or deubiquitination for its degradation . It is therefore possible that the model depicted in Figure 4E is only applicable to ERAD-L substrates . ERAD-M substrates might enter Hrd1 laterally with their transmembrane segments and perhaps this does not require autoubiquitination of Hrd1 . Alternatively , and more likely , ERAD-M ( and ERAD-C ) substrates may use the Der1 homolog Dfm1 for their extraction into the cytosol ( Neal et al . , 2018 ) . Dfm1 seems to mediate the cytosolic extraction of these ERAD substrates even in the absence of Hrd1 ( or Doa10 ) , which would explain why autoubiquitination and deubiquitination of Hrd1 is not required . The presence of Ubp1 is required for optimal performance of ERAD-L ( Figure 1H ) . The decreased degradation rate in the absence of Ubp1 can be explained by decreased Hrd1 activity . Or alternately , decreased ERAD-L degradation could be explained by enhanced Hrd1 degradation caused by its insufficient deubiquitination . The effect of Ubp1 deletion is relatively small , perhaps because a combination of other DUBs can replace Ubp1 . The overexpression of Ubp1 has no effect on the degradation of ERAD-L substrates and slightly slows the degradation of ERAD-M substrates ( Figure 1—figure supplement 3B , C and Figure 4—figure supplement 1C , D ) . Perhaps , ERAD-M substrates are affected because they are accessible to Ubp1 on the cytosolic side of the membrane prior to their extraction from the membrane by Dfm1 ( Neal et al . , 2018 ) . An alternate possibility is that deubiquitination of Hrd1 reduces its activity towards ERAD-M substrates . While Ubp1 does slow down the degradation rate of ERAD-M substrates , this balance could be advantageous for substrate specificity ( Zhang et al . , 2013 ) . Our results explain several puzzling observations in the literature . It has long been known that the deletion of Hrd3 , which interacts with Hrd1 on the ER lumenal side , results in excessive Hrd1 autoubiquitination and degradation ( Gardner et al . , 2000 ) . These reactions were attenuated when Usa1 or Usa1’s UBL domain were also deleted ( Carroll and Hampton , 2010 ) . Our results now explain these results by the demonstration that the UBL domain inhibits Ubp1 . In the absence of this domain , Ubp1 is hyperactive and reduces polyubiquitination of Hrd1 and thus its subsequent degradation ( Figure 4E ) . This model also explains why the UBL domain of Usa1 has no direct effect on the degradation of ERAD substrates ( Figure 4—figure supplement 1B–D ) ( Carroll and Hampton , 2010 ) . Whether the UBL domain of Usa1 inhibits Ubp1 by a physical interaction or indirectly remains to be determined . However , functional interactions between UBL domains and DUBs have been reported , with DUBs being allosterically regulated or competitively inhibited by specific UBL domains ( Faesen et al . , 2012 ) . Although the absence of the UBL domain of Usa1 restored wild-type levels of Hrd1 in hrd3Δ cells , substrate degradation was still defective ( Vashistha et al . , 2016 ) . Our experiments show that both Hrd1 levels and ERAD can be restored in a hrd3Δ strain when Ubp1 is overexpressed . Thus , while Hrd3 is normally involved in ERAD , we suggest it primarily performs a supporting role for Hrd1 through regulating Hrd1’s autoubiquitination activity and helping to recruit ERAD substrates ( Carvalho et al . , 2010; Gardner et al . , 2001 ) . Clearly , Hrd1 regulation is complex and one explanation for the previously published results is that Hrd1 autoubiquitination was reduced to a level at which channel opening was severely inhibited . In contrast , Ubp1 overexpression ( in the presence of Usa1 ) might still allow a low level of autoubiquitination that is sufficient for retrotranslocation , but not for efficient Hrd1 degradation . This interpretation implies that Ubp1 overexpression does not completely overcome the inhibitory function of the UBL domain , perhaps because the interaction between Hrd1 and Ubp1 is weak ( Figure 1G ) while that between Hrd1 and Usa1 is strong , or because Usa1 sterically prevents the access of Ubp1 to Hrd1 ( Carvalho et al . , 2006; Horn et al . , 2009 ) . At this point , it is unclear exactly how Ubp1 is recruited to Hrd1 . Deletion of either Hrd3 or Usa1 doesn’t have a strong effect on the Hrd1/Ubp1 interaction ( Figure 1—figure supplement 2E ) . This could mean Hrd1 and Ubp1 interact directly , or that Ubp1 is recruited through the Hrd1-attached ubiquitin chain . Another possibility is that the interaction is mediated by another , unidentified protein , which might explain the low level of co-precipitation of Ubp1 and Hrd1 ( Figure 1G and Figure 1—figure supplement 2E ) . While Hrd1 polyubiquitinates both itself and ERAD substrates , deubiquitination of these components occurs by separate mechanisms . It is somewhat surprising that CPY* degradation is not accelerated by the increased Hrd1 levels caused by Ubp1 overexpression ( Figures 1H and 3A ) . However , Hrd1 functions in a complex with other components , including Hrd3 , Usa1 , and Der1 , the levels of which remain unchanged ( Figure 1—figure supplement 1 ) . Importantly , overexpression of Ubp1 does not prevent the degradation of CPY* ( Figures 1H and 3A ) , indicating that Ubp1 does not act on polyubiquitinated ERAD-L substrate . In fact , an ERAD substrate needs to retain its polyubiquitin chain until it has been extracted from the membrane by the Cdc48/p97 ATPase complex , because the polyubiquitin chain serves as a recognition signal for the Ufd1/Npl4 cofactor of Cdc48 . Substrate deubiquitination occurs by cytosolic DUBs , one of which is Otu1 ( Yod1 in mammals ) , an enzyme that binds to the Cdc48 ATPase through its UBX-like domain ( Stein et al . , 2014 ) . As long as the ATPase complex is bound to the membrane , Cdc48 interacts with Ubx2 , which prevents access of Otu1 . Once the ATPase has pulled the substrate out of the membrane and has moved into the cytosol , Otu1 can trim the ubiquitin chain , the substrate is released from Cdc48/p97 , and then transferred to the proteasome ( Bodnar and Rapoport , 2017; Ernst et al . , 2009; Neal et al . , 2017 ) . Thus , ERAD involves two distinct deubiquitination reactions , one catalyzed by Ubp1 at the ER membrane to reset Hrd1 , and another catalyzed by Otu1 in the cytosol to release substrate from Cdc48/p97 . Integral membrane DUBs are relatively rare . In both yeast and mammals , there are only two DUBs that are integral membrane proteins ( out of 24 and 90 enzymes , respectively ) ( Komander et al . , 2009 ) . In S . cerevisiae , Ubp1 is the only known DUB in the ER membrane ( Schmitz et al . , 2005 ) . Our results show that the N-terminal membrane anchor is required for reversing the ubiquitin-modification of Hrd1 but that it can be replaced by a transmembrane segment from a different protein . The apparent 20-fold excess of Ubp1 over Hrd1 makes it is likely that Ubp1 has additional substrates which could contribute indirectly to Hrd1 stability . A shorter version of Ubp1 , which lacks the transmembrane segment , has been reported to have a role in endocytosis ( Schmitz et al . , 2005 ) . Although one would expect the function of Ubp1 to be evolutionarily conserved , there is no obvious homolog in higher organisms . However , Usp19 has been reported to serve as a DUB for polyubiquitinated Hrd1 ( Harada et al . , 2016 ) . Usp19 is present at the ER membrane ( Hassink et al . , 2009 ) , is proposed to play a role in ERAD ( Lee et al . , 2014 ) , and is one of two mammalian DUBs with transmembrane segments ( Komander et al . , 2009 ) . Mammalian Hrd1 is associated with HERP ( Schulze et al . , 2005 ) , a putative Usa1 homolog with a UBL domain ( Kokame et al . , 2000 ) , but whether Usp19 functions in a similar way as Ubp1 in yeast remains to be investigated . Deletion strains used in this study were purchased from GE Dharmacon and are derivatives of BY4741 ( MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 ) or BY4742 ( MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 ) ( see Supplementary file 1 ) . The ubp1Δ strain was generated by a transforming a PCR-amplified targeting cassette with the LiAc/PEG methods ( Gietz and Schiestl , 2007 ) . Double and triple deletion strains were generated by crossing and sporulation . Genotypes were verified by PCR . Plasmids were constructed using standard restriction cloning or Gibson assembly . All plasmids used in this study were centromeric plasmids ( Sikorski and Hieter , 1989 ) and where indicated , overexpression was driven by the GPD/TDH3 promoter . For a list of plasmids used in this study , see Supplementary file 2 . Cycloheximide-chase degradation assays were performed as described previously ( Gardner et al . , 1998 ) with the following modifications . Cells were grown to mid-log phase ( 0 . 4–0 . 7 OD600/mL ) in synthetic dropout media . The cells were pelleted at 2000 x g for 5 min and resuspended to 2 . 5 OD600/mL in fresh media . At time ‘0 min’ the culture was supplemented with 50 μg/mL cycloheximide and an aliquot was taken and centrifuged as above . The cell pellet was either flash frozen in liquid N2 or resuspended in lysis buffer . The remaining culture was incubated at 30°C with samples taken as indicated . Cell pellets were resuspended in lysis buffer ( 10 mM MOPS , pH 6 . 8 , 1% SDS , 8M urea , 10 mM EDTA , fresh protease inhibitors ) at 25 OD600/mL with an equivalent volume of acid-washed glass beads ( 0 . 1 mm , Bio-Spec ) . After vortexing for 2 min , an equal volume of urea sample buffer ( 125 mM Tris pH 6 . 8 , 4% SDS , 8M urea , 10% β-mercaptoethanol ) was added and mixed . The samples were incubated at 65°C for 5 min before SDS-PAGE , transfer to PVDF membrane and immunoblotting ( THE Anti-DYKDDDK Antibody , Genscript; HA clone 3F10 , Roche ) and detection by chemiluminescence with Western Lightning Plus-ECL ( Perkin-Elmer ) on a ChemiDoc MP ( Bio-Rad ) . For quantification of the immunoblot band intensities , band intensities were normalized based upon total protein in the sample quantified within the same gels using Bio-Rad Stain-Free Dye Imaging Technology or by normalization to the intensity of PGK1 ( clone 22C5D8 , Thermo Scientific ) . Cells expressing GPD-driven Hmg2-GFP or Hmg2-NR1-GFP with the indicated plasmids were grown to mid-log phase in synthetic dropout media . Cells were pelleted and resuspended in synthetic medium and transferred into a 96-well round bottom plate . To follow degradation of Hmg2-NR1-GFP , the medium was supplemented with 50 μg/mL cycloheximide . To induce degradation of Hmg2-GFP , the media was supplemented with 10 μg/mL zaragozic acid ( Cayman Chemical ) . The cells were grown for 4 hr , then pelleted and washed with ice-cold PBS ( 137 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 , 1 . 8 mM KH2PO4 , pH 7 . 4 ) . The washed cells were resuspended at 1 OD600/mL in PBS containing the viability dye Sytox Blue ( Invitrogen ) at 1 μM . Cells were kept at 4°C prior to flow cytometry using a Bio-Rad ZE5 Cell Analyzer with Everest software . At least ten thousand events were analyzed using forward/side scatter to identify single cells and Sytox Blue fluorescence was used to exclude dead cells . GFP fluorescence was measured from the 488 nm laser with a 509 nm/24 nm bandpass filter set , while Sytox Blue fluorescence was measured from the 405 nm laser with a 460 nm/22 nm bandpass filter set . Data were analyzed and figures were generated using FlowJo V10 . 6 ( FlowJo LLC . ) . The low-expressing GFP population is present in all traces and is indicated under the dashed lines . Denaturing pulldowns of Hrd1 were performed as described previously ( Baldridge and Rapoport , 2016 ) with the following modifications . Cells with a centromeric plasmid bearing an endogenous Hrd1 promoter and a Hrd1-His10 were grown to mid-log phase and lysed in 50 mM HEPES pH 7 . 4 , 300 mM KCl , protease inhibitors , 1 mM PMSF , 1 . 5 µM pepstatin , 8M urea ( to prevent additional Hrd1 autoubiquitination ) , and 5 mM NEM ( to inhibit deubiquitinating enzymes ) . The lysates were centrifuged at 2000 x g for 10 min , and the supernatant was collected and re-centrifuged for 30 min in a Ti45 rotor at 42 , 000 rpm ( RCFavg 138 , 001 ) . The membranes were solubilized in 50 mM HEPES pH 7 . 4 , 300 mM KCl , protease inhibitors , 1 mM PMSF , 6M urea , 1 . 5% Triton X-100 final and 25 mM imidazole ) for 1 hr at 4°C . His-tag Dynabeads ( Life Technologies ) were added ( 0 . 25 mL per 1 , 500 OD cells ) and incubated for an additional 1 hr . The beads were washed three times with a 30-fold excess buffer . Hrd1-His10 was eluted with buffer including 400 mM imidazole . The samples were analyzed by SDS-PAGE and immunoblotting with anti-Hrd1 and anti-ubiquitin antibodies ( clone P4D1 , Santa Cruz ) . Unbound IMAC flow-through was used for a loading control to demonstrate equal material input . Cells with a centromeric plasmids bearing a combination of Hrd1-3xFlag , Orm2-3xFlag , and endogenous or GPD driven Ubp1-3xHA , Ubp1-3xV5 , or Ubp1-3xV5 variants were grown to mid-log phase , pelleted , washed once with water and flash frozen in liquid nitrogen . Cells were thawed on ice and treated for 10 min with spheroplasting buffer ( 50 mM HEPES , pH 7 . 4 , 150 mM NaCl , 1M sorbitol ) supplemented with 10 mM DTT . Cells were pelleted and washed once with spheroplasting buffer before resuspending in spheroplasting buffer supplemented with 20 µg zymolyase 100T ( per 5 OD ) . Cells were incubated at 30°C for 60 min , then pelleted at 3200 x g for 5 min . The spheroplasts were resuspended in IP buffer ( 50 mM HEPES , pH 7 . 4 , 150 mM NaCl with freshly added 1 mM PMSF and 3 µM pepstatin A ) and lysed with 7–10 strokes of a tight-fitting Dounce homogenizer . Lysed cells were pelleted at 20 , 000 x g for 20 min . The supernatant was discarded and the pellet ( microsome fraction ) was resuspended in IP buffer supplemented with 1% Decyl Maltose Neopentyl Glycol ( DMNG ) . Microsomes were solubilized for 1 hr at 4°C , before input samples were taken and 7 . 5 ODs of solubilized proteins were mixed with 20 µL ( 40 µL slurry ) Anti-V5 Agarose Affinity Gel ( Millipore , A7345 ) and rolled for 2 hr at 4°C . The bound proteins were washed seven times with a 35-fold excess of IP buffer containing 0 . 1% DMNG and eluted with 2x SDS-PAGE sample buffer . The samples were analyzed by SDS-PAGE and immunoblotting with anti-Flag ( THE DYKDDDDK Tag antibody , Genscript ) and anti-V5 antibodies ( THE V5 Tag Antibody , GenScript ) with the inputs loaded at 2% . All experiments in this submission were repeated at least three times with biological replicates from independent yeast transformations .	Just like factories make mistakes when producing products , cells make mistakes when producing proteins . In cells , a compartment called the endoplasmic reticulum is where about one third of all proteins are produced , and where new proteins undergo quality control . Damaged or misfolded proteins are removed by a process called endoplasmic reticulum-associated degradation ( ERAD for short ) , because if damaged proteins accumulate , cells become stressed . One type of ERAD is driven by a protein called Hrd1 . Together with other components , Hrd1 labels damaged proteins with a ubiquitin tag that acts as a flag for degradation . Hrd1 has a paradoxical feature , however . To be active , Hrd1 tags itself with ubiquitin but this also makes it more prone to becoming degraded . How does Hrd1 remain active while avoiding its own degradation ? To address this question , Peterson et al . forced budding yeast cells to produce high levels of 23 different enzymes that remove ubiquitin tags . One of these enzymes , called Ubp1 , was able remove the ubiquitin tag from Hrd1 , though it had not been seen in the ERAD pathway before . Further experiments also showed that Ubp1 was able to regulate Hrd1 activity , making Ubp1 a regulator of Hrd1 dependent protein quality control . Without protein quality control , damaged proteins can contribute to various diseases . ERAD is a common quality control system for proteins , present in many different species , ranging from yeast to animals . Therefore , understanding how ERAD works in budding yeast may also increase understanding of how human cells deal with damaged proteins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2019	Cycles of autoubiquitination and deubiquitination regulate the ERAD ubiquitin ligase Hrd1
Tumor suppressor p53 is the most frequently mutated gene in tumors . Many mutant p53 ( mutp53 ) proteins promote tumorigenesis through the gain-of-function ( GOF ) mechanism . Mutp53 proteins often accumulate to high levels in tumors , which is critical for mutp53 GOF . Its underlying mechanism is poorly understood . Here , we found that BAG2 , a protein of Bcl-2 associated athanogene ( BAG ) family , promotes mutp53 accumulation and GOF in tumors . Mechanistically , BAG2 binds to mutp53 and translocates to the nucleus to inhibit the MDM2-mutp53 interaction , and MDM2-mediated ubiquitination and degradation of mutp53 . Thus , BAG2 promotes mutp53 accumulation and GOF in tumor growth , metastasis and chemoresistance . BAG2 is frequently overexpressed in tumors . BAG2 overexpression is associated with poor prognosis in patients and mutp53 accumulation in tumors . These findings revealed a novel and important mechanism for mutp53 accumulation and GOF in tumors , and also uncovered an important role of BAG2 in tumorigenesis through promoting mutp53 accumulation and GOF . Tumor suppressor p53 plays a central role in tumor prevention ( Levine et al . , 2006; Levine and Oren , 2009; Vousden and Prives , 2009 ) . Trp53 is the most frequently mutated gene in human tumors; it is mutated in over 50% of all tumors . Majority of Trp53 mutations are missense mutations that are localized in the p53 DNA binding domain ( DBD ) , including several mutational hotspots in tumors ( e . g . , R175 , R248 , and R273 ) ( Harris and Hollstein , 1993; Freed-Pastor and Prives , 2012; Muller and Vousden , 2014 ) . Many tumor-associated mutant p53 ( mutp53 ) proteins not only lose the tumor suppressive function of wild-type p53 ( wtp53 ) , but also gain new oncogenic activities independently of wtp53 , which is defined as mutp53 gain-of-function ( GOF ) ( Freed-Pastor and Prives , 2012; Muller and Vousden , 2014 ) . So far , many mutp53 GOFs have been identified , including promoting tumor growth , metastasis , chemoresistance and metabolic changes ( Lang et al . , 2004; Olive et al . , 2004; Muller et al . , 2009; Blandino et al . , 2012; Freed-Pastor et al . , 2012; Cooks et al . , 2013; Zhang et al . , 2013 ) . Under the non-stressed condition , wtp53 protein levels are kept low in normal cells and tissues mainly through the proteasomal degradation mediated by E3 ubiquitin ligase MDM2 , the most critical negative regulator for wtp53 ( Brooks and Gu , 2006; Hu et al . , 2012 ) . At the same time , as a direct transcriptional target of p53 , MDM2 is up-regulated by p53 under both non-stressed and stressed conditions . Thus , p53 and MDM2 forms a negative feedback loop to tightly regulate p53 protein levels in cells . However , mutp53 proteins often become stable and accumulate to high levels in tumors , which is critical for mutp53 GOF in tumorigenesis and contributes greatly to tumor progression ( Oren and Rotter , 2010; Li et al . , 2011a; Freed-Pastor and Prives , 2012; Muller and Vousden , 2013 ) . It had long been thought that the inability of MDM2 to degrade mutp53 was the main cause for mutp53 protein accumulation in tumors . However , recent studies from mice with knock-in of R172H or R270H mutp53 ( equivalent to human R175H and R273H mutp53 , respectively ) challenged this concept . Mutp53 protein is kept at low levels in normal tissues but accumulates to very high levels in tumors ( Lang et al . , 2004; Olive et al . , 2004 ) . Furthermore , loss of MDM2 in mutp53 knock-in mice leads to mutp53 protein accumulation in normal tissues , which in turn promotes tumor development ( Terzian et al . , 2008 ) . Recent studies including ours also showed that MDM2 retains the ability to degrade mutp53 in in vitro cultured cells ( Lukashchuk and Vousden , 2007; Zheng et al . , 2013 ) . These results strongly suggest that while MDM2 maintains mutp53 protein levels low in normal tissues , the disruption of MDM2-mediated mutp53 degradation in tumors could be a main cause for the frequently observed mutp53 protein accumulation in tumors . Currently , the mechanism underlying the disruption of MDM2-mediated mutp53 degradation in tumors is poorly understood . Destabilizing mutp53 to inhibit mutp53 GOF is being actively tested as a novel and promising strategy for cancer therapy . Understanding the underlying mechanism for mutp53 accumulation is critical for the development of novel targets and strategies for cancer therapy . In this study , to investigate the mechanism underlying mutp53 accumulation in tumors , we screened for proteins interacting with mutp53 using liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) assays in tumors from R172H mutp53 knock-in mice , and identified BAG2 as a novel mutp53 binding protein that plays a critical role in promoting mutp53 accumulation in tumors . BAG2 belongs to the Bcl-2 associated athanogene ( BAG ) family , which is characterized by the BAG domain . As a group of multifunctional proteins , BAG proteins interact with a variety of proteins and take part in diverse cellular processes , including cell division , cell death and differentiation ( Takayama and Reed , 2001; Kabbage and Dickman , 2008 ) . Currently , the role of BAG2 in tumorigenesis and its underlying mechanism are poorly understood . We found that mutp53 binds to BAG2 and promotes the nuclear translocation of BAG2 . The BAG2-mutp53 interaction in the nucleus inhibits the ubiquitination and degradation of mutp53 mediated by MDM2 , and thereby promotes mutp53 accumulation and mutp53 GOF in tumorigenesis . Knockdown of BAG2 greatly decreases mutp53 protein levels in tumors and compromises mutp53 GOF in tumorigenesis . BAG2 is frequently overexpressed in various types of human tumors . BAG2 overexpression is associated with poor prognosis in cancer patients and mutp53 accumulation in tumors . These results revealed a novel and critical mechanism for mutp53 protein accumulation in tumors , and strongly suggest that BAG2 is a potential target for therapy in tumors carrying mutp53 . Our results also uncovered an important role of BAG2 in tumorigenesis and revealed that promoting mutp53 accumulation and GOF is a novel mechanism for BAG2 in tumorigenesis . R172H mutp53 knock-in ( Trp53R172H/R172H ) mice mainly develop lymphomas in the spleen and thymus ( Lang et al . , 2004; Olive et al . , 2004 ) . Mutp53 protein levels are drastically increased in majority of tumors from Trp53R172H/R172H mice but are very low in normal tissues . To investigate the mechanism underlying mutp53 accumulation in tumors , we screened for proteins interacting with mutp53 in thymic lymphomas of Trp53R172H/R172H mice with drastic mutp53 accumulation ( n = 3 ) using immunoprecipitation ( IP ) assays with an anti-p53 antibody followed by LC-MS/MS assays ( Figure 1A ) . Normal tissues of Trp53R172H/R172H mice with low mutp53 levels were used as controls . LC-MS/MS assays identified a list of potential proteins binding to mutp53 in the thymic lymphomas of Trp53R172H/R172H mice ( Figure 1B ) . Several known mutp53-binding proteins , including HSP90 , Myosin , Cct8 and Pontin ( Muller et al . , 2005; Trinidad et al . , 2013; Arjonen et al . , 2014; Zhao et al . , 2015 ) , were among the list of proteins identified in tumors of Trp53R172H/R172H mice , which validated our approach . The complete list of proteins that bound to mutp53 in Trp53R172H/R172H tumors was listed in Table 1 . 10 . 7554/eLife . 08401 . 003Figure 1 . Identification of proteins interacting with mutant p53 ( mutp53 ) protein in tumors from Trp53R172H/R172H mice . ( A ) Work flow for identification of proteins interacting with mutp53 protein . Lysate of thymic lymphomas and normal thymus from Trp53R172H/R172H mice were subjected to co-immunoprecipitation ( co-IP ) using anti-p53 ( FL393 ) beads . Eluted proteins were separated in a 4–15% SDS PAGE gel and analyzed by LC-MS/MS . ( B ) The table of a list of protein candidates that interacted with mutp53 protein . ( C ) The interaction of mutp53 with BAG2 in thymic lymphomas of Trp53R172H/R172H mice was confirmed by co-IP assays followed by Western blot assays . Thymic lymphomas from Trp53R172H/R172H mice and p53−/− mice as well as normal thymic tissue from Trp53R172H/R172H mice were subjected to co-IP assays using an anti-p53 antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 00310 . 7554/eLife . 08401 . 004Figure 1—figure supplement 1 . The interaction of mutp53 with BAG2 in normal mouse tissues of Trp53R172H/R172H mice . Normal thymus , spleen and kidney tissues from Trp53R172H/R172H mice were subjected to co-IP assays using an anti-p53 antibody ( FL393 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 00410 . 7554/eLife . 08401 . 005Table 1 . The list of identified mutp53-interaction protein candidatesDOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 005Gene namesAverage countsNormal thymusThymic lymphomaCtsb019Tfrc015 . 5Bag2015Stip1014 . 5Hyou1014 . 5Cad014 . 5Rps19014 . 5Pfn1014Cand1011 . 5Hspa2011Lcp1011Sar1a010 . 5Fam49b010 . 5Khsrp010Ifi4709 . 5Cse1l09 . 5Ipo509 . 5Hsp90b109Hspbp108 . 5Rfc508 . 5Tkt08 . 5myosin08Hadhb08Hsp70166Phgdh120Myh9581 . 5Hspd1232 . 5Rpl9-ps4116Ubr5113 . 5Dars113 . 5Iqgap1112Slc25a3111 . 5Rars111 . 5Ruvbl2111Ddb1110Hsph1438 . 5Dnajb419Aldoa18 . 5Pcna216 . 5Eprs18Hsp90215 . 5Gm9755212 . 5Dnaja1318Atp5b318Cltc741Gm5506526Dnaja2315Bag5731 . 5Rps7522 . 5Ywhae29Eef21038 . 5Adsl27 . 5Hsp90ab12074 . 5Gnb2l1622Copg27Rpl2327Psmc627Pcbp227Pcbp1310Pabpc4515 . 5Hspa877237 . 5Fcgr41338 . 5Mcm7411 . 5Hadha38 . 5Kpnb1514Atp5a1925Pontin411Bat338Pdia637 . 5Dnajc73886Rps15a613 . 5Aldh249Trim28511Eef1a11635St13613Cct83654Psmd1169 Interestingly , BAG2 was identified as a potential mutp53 binding protein ( Figure 1B ) . The BAG2-mutp53 interaction in Trp53R172H/R172H tumors was confirmed by co-IP followed by Western blot assays ( Figure 1C ) . The weak interaction between BAG2 and mutp53 was also observed in normal tissues from Trp53R172H/R172H mice , including thymus , spleen and kidney ( Figure 1C , Figure 1—figure supplement 1 ) . To investigate whether BAG2 specifically interacts with mutp53 in human cells , human p53-null lung cancer H1299 cells were transfected with human BAG2-HA expression vectors together with human wtp53 or mutp53 ( R175H ) expression vectors . Co-IP assays employing either anti-p53 or anti-HA antibodies showed that BAG2 preferentially bound to mutp53 compared with wtp53 ( Figure 2A ) . In addition to R175H , the strong BAG2-mutp53 interaction was observed in H1299 cells with ectopic expression of different mutp53 proteins , including R248W and R273H , respectively ( Figure 2B ) . The interaction between the endogenous BAG2 and mutp53 proteins was also observed in several human cancer cell lines , including human colorectal cancer HCT116 p53R248W/− , HT-29 and SW480 cell lines which contain a single copy of Trp53 gene with R248W and R273H mutation , respectively , human breast cancer SK-BR-3 , MDA-MB-468 cell lines which contain a single copy of Trp53 gene with R175H and R273H mutation , respectively , and human hepatocellular carcinoma Huh-7 cell lines which contain a single copy of Trp53 gene with Y220C mutation ( Figure 2C , Figure 2—figure supplement 1 ) . Together , these results demonstrate that BAG2 is a novel mutp53-specific binding partner , and this interaction is conserved in both mouse tumors and human cancer cells . 10 . 7554/eLife . 08401 . 006Figure 2 . BAG2 is a mutp53-specific binding partner as determined by reciprocal co-IP assays in human cell lines . ( A ) Ectopically expressed BAG2 preferentially interacted with mutp53 ( R175H ) protein compared with wild-type p53 ( wtp53 ) protein in H1299 cells . H1299 cells were transiently transfected with vectors expressing mutp53 ( R175H ) or wtp53 together with HA-tagged BAG2 ( BAG2-HA ) expression vectors . Antibodies used for IP assays: HA for BAG2-HA and DO-1 for p53 . ( B ) BAG2 interacted with several hotspot mutp53 proteins ( R175H , R248W and R273H ) in H1299 cells . H1299 cells were transiently transfected with vectors expressing mutp53 ( R175H , R248W or R273H ) together with BAG2-HA expression vectors . ( C ) The interaction of endogenous BAG2 with mutp53 ( R248W ) was observed in human colorectal cancer HCT116 p53R248W/− cells containing one allele of mutant p53 gene ( R248W ) . ( D ) BAG2 interacted with mutp53 DNA binding domain ( DBD ) . Upper panel: Schematic diagram showing the domain structure of mutp53 ( R175H ) . Lower Panel: H1299 cells were transiently transfected with expression vectors of HA-tagged mutp53 ( R175H ) fragments together with BAG2-Flag expression vectors . Antibodies used for IP: Flag for BAG2-Flag proteins . ( E ) BAG2 preferentially interacted with the DBD of mutp53 ( R175H ) but not wtp53 DBD . H1299 cells were transiently transfected with expression vectors of HA-tagged mutp53 ( R175H ) DBD or wtp53 DBD together with BAG2-Flag expression vectors . ( F ) Mutp53 interacted with the Bcl-2 associated athanogene ( BAG ) domain of BAG2 . Left panel: Schematic diagram showing the domain structure of BAG2 . Right panel: H1299 cells were transiently transfected with expression vectors of mutp53 ( R175H ) together with Flag-tagged BAG2 fragments . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 00610 . 7554/eLife . 08401 . 007Figure 2—figure supplement 1 . The interaction of endogenous BAG2 with mutp53 in several human tumor cell lines containing endogenous mutp53 . The interaction of endogenous BAG2 with mutp53 was observed in human colorectal cancer HT-29 and SW480 cell lines which contain one allele of mutant Trp53 gene ( R273H ) , human breast cancer SK-BR-3 and MDA-MB-468 cell lines which contain one allele of mutant Trp53 gene ( R175H and R273H , respectively ) , and human hepatocellular carcinoma Huh-7 cells which contain one allele of mutant Trp53 gene ( Y220C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 00710 . 7554/eLife . 08401 . 008Figure 2—figure supplement 2 . The interaction of BAG2 with mutp53 ( R248W and R273H ) DBD in H1299 cells . BAG2 interacted with DBD of mutp53 ( R248W and R273H ) but not the mutp53 fragments lacking DBD . H1299 cells were transiently transfected with expression vectors of the HA-tagged DBD of mutp53 ( R248W and R273H ) or mutp53 fragment lacking DBD ( P6 as indicated in Figure 2D ) together with BAG2-Flag expression vectors . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 008 p53 protein contains two transcriptional activation domains ( AD1 and AD2 ) , a sequence-specific DBD , a tetramerization domain and a C-terminal domain ( C-ter ) . To define the regions of mutp53 required for the BAG2-mutp53 interaction , expression vectors of fragments containing different mutp53 domains with HA-tag ( Figure 2D , upper panel ) and BAG2-Flag expression vectors were co-transfected into p53-null H1299 cells . Results of co-IP assays using an anti-Flag antibody showed that BAG2 interacted with all mutp53 ( R175H ) fragments containing the mutp53 DBD ( P1-P5 in Figure 2D ) , but not the fragment lacking the mutp53 DBD ( P6 in Figure 2D ) . Furthermore , BAG2 preferentially bound to DBDs of different mutp53 ( R175H , R248W and R273H ) but not wtp53 DBD ( Figure 2E , Figure 2—figure supplement 2 ) . The regions of BAG2 required for the BAG2-mutp53 interaction was examined by co-transfecting cells with vectors expressing different Flag-tagged BAG2 deletion mutants ( Figure 2F , left panel ) and mutp53 ( R175H ) expression vectors followed by co-IP assays . BAG2 contains a BAG domain ( amino acids 91–211 ) at the C-terminus ( Dai et al . , 2005 ) . The fragments containing the BAG domain interacted with mutp53 while the N-terminus of BAG2 protein lacking the BAG domain did not interact with mutp53 ( Figure 2F ) . Interestingly , the binding of mutp53 to the BAG2 fragment which lacks the N-terminus is much weaker compared with the full length ( FL ) BAG2 protein . It is possible that the N-terminus of BAG2 has an additional role for efficient BAG2-mutp53 complex formation although itself does not directly interact with mutp53 . These results demonstrate that mutp53 DBD and BAG domain of BAG2 are essential for the BAG2-mutp53 interaction . It was reported that BAG2 stabilizes some of its binding proteins , such as PINK1 and ataxin3-80Q ( Che et al . , 2013 , 2014 ) . To investigate whether BAG2 regulates mutp53 protein levels , endogenous BAG2 was knocked down by 2 different siRNA oligos and its impact upon mutp53 protein levels was evaluated in HCT116 p53R248W/− cells and p53-null Saos2 cells with stable ectopic expression of different mutp53 ( Saos2-R175H , Saos2-R248W and Saos2-R273H ) . The knockdown of BAG2 was confirmed at both mRNA and protein levels by real-time PCR and Western blot assays , respectively ( Figure 3A , B ) . While BAG2 knockdown showed no apparent effect on mutp53 mRNA levels ( Figure 3—figure supplement 1 ) , BAG2 knockdown greatly decreased the mutp53 protein levels in cells ( Figure 3A ) . The effect of BAG2 overexpression on mutp53 protein levels was also determined in these cells . Ectopic BAG2 expression by vectors clearly increased mutp53 protein levels ( Figure 3C ) , while had no clear effect on mutp53 mRNA levels in cells ( Figure 3—figure supplement 2 ) . These results demonstrate that BAG2 increases mutp53 protein levels in cells . 10 . 7554/eLife . 08401 . 009Figure 3 . BAG2 promotes mutp53 protein accumulation in human cancer cells through the inhibition of the ubiquitination and degradation of mutp53 mediated by MDM2 . ( A ) Knockdown of endogenous BAG2 by 2 different siRNA oligos decreased the mutp53 protein levels in HCT116 p53R248W/− and Saos2 cells with stable ectopic expression of mutp53 ( Saos2-R175H , Saos2-R248W and Saos2-R273H ) . The knockdown of BAG2 by siRNA at the protein level was examined by Western blot assays . ( B ) The efficient knockdown of BAG2 by siRNA was confirmed at the mRNA level by real-time PCR . Data are present as mean ±SD ( n = 3 ) . ( C ) Ectopic expression of BAG2 by transfection of BAG2-HA expression vectors increased the mutp53 protein levels in cells . ( D ) Knockdown of endogenous BAG2 by siRNA decreased the mutp53 protein levels in HCT116p53R248W/− , Saos2-R175H , Saos2-R248W and Saos2-R273H cells but not in these cells treated with the proteasome inhibitor MG132 ( 40 µM for 6 hr ) . ( E ) BAG2 inhibited the degradation of mutp53 ( R175H ) mediated by MDM2 in H1299 cells . Indicated combination of expression vectors of BAG2-HA , mutp53 ( R175H ) , MDM2 were transfected into the cells . ( F ) Knockdown of MDM2 abolished the effect of BAG2 knockdown on mutp53 protein level . Knockdown of endogenous BAG2 decreased mutp53 protein levels in Saos2-R175H cells but not in cells with knockdown of endogenous MDM2 . ( G ) BAG2 reduced the interaction of mutp53 with MDM2 in H1299 cells as determined by IP assays . Indicated combination of expression vectors of BAG2-HA , mutp53 ( R175H ) and MDM2 were transfected into the cells . Antibodies used for IP: DO-1 for p53 . ( H ) Ectopic BAG2 expression decreased the ubiquitination levels of mutp53 in H1299 cells . Cells were transfected with indicated combination of expression vectors of BAG2-HA , mutp53 ( R175H ) , His-ubiquitin ( His-Ub ) , followed by MG132 treatment . Mutp53 ubiquitination was determined by IP using DO-1 antibody ( for mutp53 ) followed by Western blot assays using an anti-Ub antibody . ( I ) Knockdown of endogenous BAG2 increased the ubiquitination levels of mutp53 in Saos2-R175H cells . Cells were transfected with indicated combination of BAG2 siRNAs and expression vectors of His-Ub followed by MG132 treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 00910 . 7554/eLife . 08401 . 010Figure 3—figure supplement 1 . Knockdown of BAG2 has no apparent effect on mutp53 mRNA expression levels in human cancer cells . Knockdown of endogenous BAG2 had no apparent effect on the mRNA levels of mutp53 in HCT116 p53R248W/− and Saos2 cells with stable ectopic expression of mutp53 ( Saos2-R175H , Saos2-R248W and Saos2-R273H ) . Data are present as mean ±SD ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 01010 . 7554/eLife . 08401 . 011Figure 3—figure supplement 2 . Ectopic expression of BAG2 has no apparent effect on mutp53 mRNA expression levels in human cancer cells . Ectopic expression of BAG2 had no apparent effect on the mRNA levels of mutp53 in HCT116 p53R248W/− and Saos2 cells with stable ectopic expression of mutp53 ( Saos2-R175H , Saos2-R248W and Saos2-R273H ) . Data are present as mean ±SD ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 01110 . 7554/eLife . 08401 . 012Figure 3—figure supplement 3 . The expression levels of BAG2 in H1299 , Saos2 and HCT116 p53−/− cells . ( A ) The mRNA expression levels of BAG2 were examined in cells by using Taqman real-time PCR and normalized with actin . Data are presented as mean ±SD ( n = 6 ) . ( B ) The protein levels of BAG2 were examined in cells by Western blot assays . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 012 BAG2 is a co-chaperone protein , which can regulate the ubiquitination and degradation of some proteins ( Dai et al . , 2005; Che et al . , 2013 ) . Here , we investigated whether BAG2 promotes mutp53 protein accumulation through the inhibition of mutp53 protein ubiquitination and degradation . Since endogenous BAG2 expression levels are relatively higher in Saos2 and HCT116 p53R248W/− cells compared with H1299 cells as determined at the RNA and protein levels ( Figure 3—figure supplement 3 ) , experiments with knockdown of endogenous BAG2 were performed by using Saos2 and HCT116 p53R248W/− cells , and experiments with ectopic BAG2 expression were performed by using H1299 cells . We found that blocking proteasomal degradation by the proteasome inhibitor MG132 largely abolished the effect of BAG2 knockdown on mutp53 protein levels in HCT116 p53R248W/− , Saos2-R175H , Saos2-R248W and Saos2-R273H cells ( Figure 3D ) . Ectopic expression of MDM2 clearly down-regulated mutp53 R175H in H1299 cells co-transfected with vectors expressing mutp53 R175H and MDM2 ( Figure 3E ) , which is consistent with previous reports ( Lukashchuk and Vousden , 2007; Zheng et al . , 2013 ) . Notably , co-expression of BAG2 largely reduced the degradation of mutp53 protein mediated by MDM2 ( Figure 3E ) . Consistently , knockdown of endogenous MDM2 clearly increased mutp53 protein levels in Saoa2-R175H cells ( Figure 3F ) . Notably , the effect of BAG2 knockdown on mutp53 protein levels was greatly reduced in cells with MDM2 knockdown , indicating that the effect of BAG2 knockdown on mutp53 protein levels is largely mediated by MDM2 ( Figure 3F ) . MDM2 directly binds to mutp53 to negatively regulate mutp53 . Co-expression of BAG2 clearly decreased the interaction of MDM2 with mutp53 in H1299 cells , which could be an important mechanism by which BAG2 inhibits MDM2-mediated mutp53 degradation ( Figure 3G ) . To investigate whether BAG2 regulates mutp53 protein through inhibiting mutp53 ubiquitination , in vivo ubiquitination assays were employed . Ectopic BAG2 expression reduced ubiquitination of mutp53 in H1299 cells ( Figure 3H ) . Knockdown of endogenous BAG2 by siRNA increased ubiquitination of mutp53 in Saos2-R175H cells ( Figure 3I ) . These results demonstrate that BAG2 interacts with mutp53 , and inhibits MDM2 binding to and degradation of mutp53 , which leads to the mutp53 accumulation in cells . It has been reported that BAG2 proteins were mainly localized in the cytoplasm ( Dai et al . , 2005 ) . Indeed , in H1299 cells with ectopic expression of BAG2 alone , BAG2 proteins were predominantly localized in the cytoplasm as determined by immunofluorescence ( IF ) staining ( Figure 4A ) . Interestingly , we found that mutp53 promoted BAG2 nuclear translocation; ectopic expression of mutp53 ( R175H , R248W and R273H ) , which is mainly localized in the nucleus , clearly increased the translocation of BAG2 from the cytoplasm to the nucleus in cells transfected with vectors expressing BAG2 together with mutp53 . Furthermore , BAG2 was largely co-localized with mutp53 in the nucleus ( Figure 4A ) . In contrast , ectopic expression of wtp53 , which is also mainly localized in the nucleus , did not have an obvious effect on BAG2 nuclear translocation in cells ( Figure 4A ) . The effect of mutp53 on BAG2 nuclear translocation was also confirmed by Western blot assays using whole cell lysates and nuclear extracts isolated from H1299 cells transfected with BAG2 vectors alone or together with mutp53 vectors ( Figure 4B ) . Both mutp53 and MDM2 proteins contain a nuclear localization signal ( NLS ) and are mainly localized in the nucleus , where MDM2 binds to and ubiquitinates mutp53 protein . The translocation of BAG2 to the nucleus where it interacts with mutp53 may play an important role in blocking MDM2 to bind to and degrade mutp53 . To test this possibility , we constructed the vector expressing the NLS mutant of mutp53 R175H ( mutp53NLS ) by mutating Lys305 , Arg306 , Lys319 , Lys320 and Lys321 to Ala as reported ( O'Keefe et al . , 2003 ) . Unlike mutp53 proteins which were mainly localized in the nucleus , mutp53NLS proteins were mainly localized in the cytoplasm as shown by IF staining ( Figure 4A ) . While mutp53NLS readily interacted with BAG2 as determined by co-IP assays ( Figure 4C ) , mutp53NLS could not promote the nuclear translocation of BAG2 . BAG2 was mainly localized in the cytoplasm in H1299 co-transfected with vectors expressing BAG2 and mutp53NLS ( Figure 4A ) . Notably , ectopic expression of MDM2 showed a limited effect on degradation of mutp53NLS protein compared with mutp53 ( R175H ) ( Figure 4D ) . Furthermore , co-expression of BAG2 had no obvious effect on mutp53NLS protein levels in cells co-transfected with expression vectors of BAG2 , mutp53NLS and MDM2 ( Figure 4D ) . These results strongly suggest that mutp53 promotes BAG2 nuclear localization and the BAG2-mutp53 interaction in the nucleus inhibits MDM2-mediated mutp53 protein degradation . 10 . 7554/eLife . 08401 . 013Figure 4 . Mutp53 promotes the nuclear translocation of BAG2 . ( A ) H1299 cells were transiently transfected with vectors expressing BAG2-HA together with or without expression vectors of mutp53 ( R175H , R248W , R273H , or R175HNLS ) and wtp53 . The protein localization of BAG2 and p53 in cells was determined by immunofluorescence ( IF ) staining . Antibody used for IF: Flag for BAG2-Flag and FL393 for p53 . Nuclei were stained with DAPI . Left panels: representative IF images . Scale bar: 10 µm . Right panels: quantification of the subcellular distribution of BAG2 in 200 cells for each independent experiment . Numerical data are presented in Figure 4—source data 1 . Data are present as mean ±SD ( n = 4 ) . p < 0 . 05; *p < 0 . 001 . ( B ) Mutp53 promotes the nuclear translocation of BAG2 in H1299 cells as determined by Western blot assays . The protein levels of BAG2 were determined in whole cell lysates and nuclear extracts prepared from H1299 cells transfected with vectors expressing BAG2-HA together with or without mutp53 ( R175H , R248W or R273H ) . ( C ) BAG2 interacted with mutp53NLS ( R175HNLS ) as determined by co-IP assays . H1299 cells were transfected with vectors expressing BAG2-Flag and mutp53 R175H or mutp53 R175HNLS . ( D ) MDM2 had a much reduced effect on degradation of mutp53NLS compared with mutp53 ( R175H ) . While BAG2 inhibited the degradation of mutp53 ( R175H ) mediated by MDM2 , it had no obvious effect on mutp53NLS protein levels in H1299 cells transfected with vectors expressing BAG2-HA , MDM2 and mutp53NLS . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 01310 . 7554/eLife . 08401 . 014Figure 4—source data 1 . % of cells with different BAG2 localization in H1299 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 014 The accumulation of mutp53 proteins is critical for mutp53 GOF in tumorigenesis ( Blandino et al . , 2012; Muller and Vousden , 2014 ) . Chemoresistance is one of the most important mutp53 GOFs ( Napoli et al . , 2012; Masciarelli et al . , 2014 ) . 5-flurorouracil ( 5-FU ) , which can induce apoptosis in cells , is one of the most commonly used chemotherapeutic agents for a wide variety of human cancers . 5-FU induced less apoptosis in Saos2-R175H , Saos2-R248W and Saos2-R273H cells compared with Saos2-Con cells as determined by Annexin V staining and the levels of cleaved Caspase 3 protein , demonstrating that mutp53 promotes chemoresistance , which is consistent with previous reports ( Napoli et al . , 2012; Masciarelli et al . , 2014 ) ( Figure 5A , B ) . Notably , knockdown of BAG2 increased 5-FU-induced apoptosis in Saos2-R175H , Saos2-R248W and Saos2-R273H cells but showed a very limited effect in Saos2-Con cells ( Figure 5A , B ) . Consistently , 5-FU induced less apoptosis in HCT116 p53R248W/− cells compared with HCT116 p53−/− cells . Knockdown of BAG2 increased 5-FU-induced apoptosis in HCT116 p53R248W/− but not HCT116 p53−/− cells ( Figure 5C , D ) . These results demonstrate that BAG2 , which promotes mutp53 protein accumulation , promotes mutp53 GOF in chemoresistance . 10 . 7554/eLife . 08401 . 015Figure 5 . BAG2 promotes mutp53 gain-of-function ( GOF ) in chemoresistance . ( A , B ) BAG2 knockdown increased 5-FU-induced apoptosis in Saos2 cells in a largely mutp53-dependent manner . The endogenous BAG2 was knocked down by siRNA in Saos2-Con , Saos2-R175H , Saos2-R248W and Saos2-R273H cells followed by 5-FU treatment ( 4 mM ) for 48 hr . In A , Annexin V assays were used to determine the percentage of apoptotic cells . Data are present as mean ±SD , n = 4 . p < 0 . 05; p < 0 . 01; p < 0 . 001 . In B , the levels of cleaved Caspase 3 , which reflect the degree of apoptosis of cells , were determined by Western blot assays . ( C , D ) BAG2 knockdown increased 5-FU-induced apoptosis in HCT116 p53R248W/− cells but had a limited effect in HCT116 p53−/− cells as determined by Annexin V assays ( C ) and Western blot assays for the cleaved Caspase 3 protein levels ( D ) . Numerical data for A and C are presented in Figure 5—source data 1 , 2 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 01510 . 7554/eLife . 08401 . 016Figure 5—source data 1 . % of apoptosis induced by 5-FU in Saos2 cells with and without ectopic expression of mutp53 . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 01610 . 7554/eLife . 08401 . 017Figure 5—source data 2 . % of apoptosis induced by 5-FU in HCT116 cells with and without mutp53 . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 017 A critical GOF of mutp53 is to promote metastasis ( Lang et al . , 2004; Olive et al . , 2004 ) . We found that BAG2 promotes mutp53 GOF in metastasis . Migration is a critical step of metastasis . Compared with p53-null cells ( Saos2-Con and HCT116 p53−/− cells ) , mutp53 ( R175H , R248W and R273H in Saos2 cells and R248W in HCT116 p53R248W/− cells ) promoted migration of cells as determined by transwell assays ( Figure 6A , B ) . Notably , knockdown of BAG2 by either siRNA oligos or shRNA vectors largely abolished the promoting effect of mutp53 on migration in these cells ( Figure 6A , B , Figure 6—figure supplement 1 ) . The effect of BAG2 on mutp53 GOF in metastasis was further examined in vivo . HCT116 p53R248W/− and HCT116 p53−/− cells stably transduced with shRNA vectors against BAG2 and control cells transduced with control shRNA vectors were injected into the tail vein of nude mice to evaluate the formation of lung metastatic tumors . Mutp53 ( R248W ) greatly promoted lung metastatic tumor formation in nude mice; HCT116 p53R248W/− cells formed significantly higher number and larger size of tumors compared with HCT116 p53−/− cells ( Figure 6C ) . Notably , this effect was greatly abolished by knockdown of BAG2 ( Figure 6C ) . These results demonstrate that BAG2 promotes mutp53 GOF in metastasis . 10 . 7554/eLife . 08401 . 018Figure 6 . BAG2 promotes mutp53 GOF in promoting metastasis and tumor cell growth . ( A ) Knockdown of endogenous BAG2 by 2 siRNA oligos preferentially inhibited the migration ability of Saos2-R175H , Saos2-R248W and Saos2-R273H cells compared with Saos2-Con cells as determined by transwell assays . Left panel: representative images form a portion of the field . Right panel: quantification of average number and area of migrated cells/field . ( B ) Knockdown of endogenous BAG2 preferentially inhibited the migration ability of HCT116 p53R248W/− cells compared with HCT116 p53−/− cells . For A , B , date are presented as mean ±SD , n = 4 . p < 0 . 001 . ( C ) BAG2 knockdown greatly inhibited lung metastasis of HCT116 p53R248W/− cells but had a limited effect on HCT116 p53−/− cells in vivo . HCT116 p53R248W/− and HCT116 p53−/− cells stably infected with shRNA against BAG2 and their control cells were injected into the nude mice via the tail vein . The number and size of lung metastatic tumors were determined at 6 weeks after inoculation . Left panel: representative H&E images of lung sections . Scale bar: 200 µm . Middle and Right panels: quantification of average number ( middle panel ) and area ( right panel ) of lung metastatic tumors , respectively . Date are presented as mean ±SD , n = 8/group . p < 0 . 01; *p < 0 . 001 . ( D ) Knockdown of BAG2 by shRNA preferentially inhibited the anchorage-independent growth in HCT116p53R248W/− cells but not HCT116 p53−/− cells . Upper panel: representative images of cell colonies in soft agar . Lower panel: quantification of average number of colonies/field . Date are presented as mean ±SD , n = 4 . p < 0 . 001 . ( E ) BAG2 knockdown inhibited the growth of HCT116 xenograft tumors in a largely mutp53-dependent manner . HCT116 p53R248W/− and HCT116 p53−/− cells stably infected with shRNA against BAG2 and their control cells were employed for xenograft tumor formation in nude mice . Upper panel: A representative image of xenograft tumors . Lower panel: growth curves of xenograft tumors . Tumor volumes are presented as mean ±SD , n = 6/group . p < 0 . 001 . ( F ) BAG2 knockdown decreased mutp53 protein levels in HCT116 p53R248w/− xenograft tumors as determined by Western blot assays . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 01810 . 7554/eLife . 08401 . 019Figure 6—figure supplement 1 . Knockdown of endogenous BAG2 by shRNA vectors inhibited mutp53 GOF in promoting migration in cells . ( A ) Knockdown of endogenous BAG2 by shRNA vectors was confirmed at the protein level in HCT116 p53R248W/− and HCT116 p53−/− cells by Western blot assays . ( B , C ) Knockdown of endogenous BAG2 by shRNA vectors preferentially inhibited the migration ability of HCT116 p53R248W/− cells compared with HCT116 p53−/− cells as determined by the transwell assays . ( B ) Representative images from a portion of the field . ( C ) Quantification of average number ( left panel ) and area ( right panel ) of migrated cells/field . Date are presented as mean ±SD , n = 6 . *p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 01910 . 7554/eLife . 08401 . 020Figure 6—figure supplement 2 . BAG2 knockdown by shRNA inhibited the proliferation rate in HCT116 p53R248W/− cells but not HCT116 p53−/− cells . Numbers of viable cells of HCT116 p53R248W/− but not HCT116 p53−/− cells with and without knockdown of BAG2 were counted daily by the Vi-CELL cell counter ( Beckman Coulter ) . Date are presented as mean ±SD , n = 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 020 The mutp53 GOFs also include the abilities to promote proliferation of tumor cells and anchorage-independent cell growth ( Zhang et al . , 2013 ) . As shown in Figure 6D and Figure 6—figure supplement 2 , mutp53 ( R248W ) promoted proliferation and anchorage-independent growth of HCT116 cells . Notably , knockdown of BAG2 clearly inhibited the rates of cell proliferation and anchorage-independent growth in HCT116 p53R248W/− but not HCT116 p53−/− cells . The xenograft tumorigenesis assays were further performed to investigate whether BAG2 knockdown reduced mutp53 GOF in promoting tumor growth in vivo . As shown in Figure 6E , knockdown of BAG2 in HCT116 p53R248W/− cells significantly inhibited the growth of xenograft tumors , whereas knockdown of BAG2 in HCT116 p53−/− had much less effect on the growth of xenograft tumors . Furthermore , knockdown of endogenous BAG2 clearly decreased mutp53 protein levels in HCT116 p53R248W/− tumors as determined by Western blot assays ( Figure 6F ) , which is consistent with the results obtained from in vitro cultured cells . These results demonstrate that BAG2 promotes mutp53 GOFs in tumor cell growth . Results from our study have demonstrated that BAG2 interacts with mutp53 and inhibits mutp53 degradation , which in turn promotes mutp53 protein accumulation and enhances mutp53 GOF in tumorigenesis . BAG2 expression was found elevated in many types of human tumors , including colorectal cancers , lung cancers , breast cancers and sarcomas , compared with normal tissues as analyzed in 4 databases from Oncomine ( GSE20842 , Gaedcke et al . , 2011; GSE10072 , Landi et al . , 2008; GSE3744 , Richardson , 2006; GSE21122 , Taylor et al . , 2010 ) ( Figure 7A ) . The amplification of BAG2 was observed in many types of human tumors as analyzed by employing the cBioportal for Cancer Genomics ( Figure 7—figure supplement 1 ) , suggesting that gene amplification is an important mechanism for BAG2 overexpression in tumors . We further investigated whether BAG2 overexpression is associated with poor prognosis in cancer patients by using the PrognoScan database . PrognoScan , which has a large collection of publicly available database with microarray data and clinical information , can assess the prognostic power of gene expression levels ( Mizuno et al . , 2009 ) . As shown in Figure 7B–E , BAG2 overexpression is associated with poor disease free survival in colorectal cancer patients ( HR = 1 . 40 , p = 0 . 022 ) , poor disease specific survival in lung cancer patients ( HR = 2 . 4 , p = 0 . 00001 ) , poor relapse free survival in breast cancer patients ( HR = 1 . 3 , p = 0 . 00014 ) and poor distant recurrence free survival in soft tissue cancer patients ( HR = 1 . 67 , p = 0 . 00001 ) . These results suggest the significant prognostic value of BAG2 expression levels for patients with various types of cancer . 10 . 7554/eLife . 08401 . 021Figure 7 . BAG2 is overexpressed in many human tumors and high levels of BAG2 are associated with mutp53 protein accumulation in human tumors . ( A ) BAG2 mRNA levels are elevated in human cancers , including colorectal cancers , lung cancers , breast cancers and sarcomas . BAG2 mRNA levels in normal and cancer tissues are presented as box plots based on data in four different datasets obtained from the Oncomine database . The expression levels of BAG2 are expressed in terms of a log2 median-centered intensity which is calculated by normalizing the intensity of BAG2 probe to the median of the probe intensities across the entire array . ( B–E ) High levels of BAG2 are associated with poor prognosis in cancer patients . Kaplan–Meier curves indicating the disease free survival of 226 colorectal cancer patients ( B ) , the disease specific survival of 90 lung cancer patients ( C ) , the relapse free survival of 204 breast cancer patients ( D ) and the distant recurrence free survival of 140 soft tissue cancer patients ( E ) . The survival information and expression levels of BAG2 were obtained from the public available databases ( GSE14333 for B , GSE14814 for C , GSE12276 for D , and GSE30929 for E ) and analyzed by PrognoScan , a web based platform evaluating the prognostic power of gene expression levels . ( F ) BAG2 overexpression correlates with mutp53 protein accumulation ( p = 0 . 036 , χ2 test ) but not wtp53 protein accumulation in human colorectal cancers . BAG2 mRNA levels were determined in human colorectal cancers and normalized with β-actin . ( G ) Schematic model depicting that mutp53 interacts with BAG2 and promotes BAG2 nuclear translocation to inhibit MDM2-mediated mutp53 protein degradation , which in turn promotes mutp53 protein accumulation and GOF in tumorigenesis . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 02110 . 7554/eLife . 08401 . 022Figure 7—figure supplement 1 . Amplification of the BAG2 gene was observed in many human tumors . Percentage of tumors showing genetic alterations , including amplification ( red ) , mutation ( green ) and deletion ( blue ) , in the BAG2 gene in different human tumors . Data were obtained from the cBioportal for Cancer Genomics ( http://www . cbioportal . org ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 022 The correlation between BAG2 overexpression and mutp53 accumulation was further investigated in a cohort of human colorectal cancer samples with known p53 mutation status and p53 protein levels ( n = 100 ) ( Zheng et al . , 2013 ) . p53 mutation status was determined by direct sequencing of exons 2–11 of p53 and the p53 protein levels were determined by IHC staining as previously described ( Zheng et al . , 2013 ) . All tumors carrying mutp53 and a small percentage of tumors with wtp53 showed positive staining for p53 ( >10% cells are stained ) . Tumors were divided into 2 groups according to median BAG2 expression levels as determined by Taqman real-time PCR assays . There is a clear correlation between high BAG2 expression and mutp53 accumulation ( Figure 7F ) . In tumors with mutp53 , 66 . 7% of tumors ( 18 out of 27 ) with high BAG2 expression displayed high p53 staining ( >30% cells are stained ) while only 40 . 9% tumors ( 9 out of 22 ) with low BAG2 expression had high p53 staining ( p = 0 . 035 ) . In contrast , in tumors with wtp53 , there is no correlation between BAG2 expression and p53 accumulation . Among these tumors , 21 . 7% of tumors ( 5 out of 23 ) with high BAG2 expression and 17 . 9% of tumors ( 5 out of 28 ) with low BAG2 expression displayed low p53 staining ( 10–30% cells are stained ) , respectively ( p = 0 . 36 ) ( Figure 7F ) . These results demonstrate that BAG2 overexpression is significantly correlated with accumulation of mutp53 protein in colorectal cancers . Many tumor-associated mutp53 proteins gain new oncogenic activities independently of wtp53 , which is critical for mutp53 to promote tumorigenesis . While wtp53 proteins are kept at low levels in normal tissues under normal conditions , mutp53 proteins often accumulate to high levels in tumors , which is critical for mutp53 GOF in tumorigenesis ( Terzian et al . , 2008; Oren and Rotter , 2010; Muller and Vousden , 2013; Liu et al . , 2015 ) . Currently , the mechanism for mutp53 accumulation in tumors is poorly understood . Results from mouse genetic experiments with knockout of MDM2 in mutp53 mice and cell-based in vitro experiments suggest that MDM2 maintains mutp53 protein levels low in normal tissues , whereas some changes occurred in tumors disrupt MDM2-mediated mutp53 degradation , thereby leading to mutp53 accumulation . Recent studies have reported that HSP90 can bind to mutp53 and reduce mutp53 degradation mediated by MDM2 . Knockdown of HSP90 by siRNA or blocking HDAC6-HSP90 axis by SAHA induced destabilization of mutp53 and inhibited its GOF in tumorigenesis ( Li et al . , 2011a , 2011b ) . Our recent study showed that tumor-derived MDM2 short isoforms inhibited full-length MDM2-mediated mutp53 degradation , which promoted mutp53 accumulation and enhanced GOF in tumorigenesis ( Zheng et al . , 2013 ) . In this study , we searched for the changes that occurred in tumors to disrupt MDM2-meidated mutp53 degradation by screening for mutp53 binding protein in tumors from mutp53 knock-in mice . Results from this study identified that BAG2 is a novel mutp53 binding protein that promotes mutp53 protein accumulation , which revealed a novel mechanism for mutp53 accumulation in tumor cells . BAG2 belongs to the BAG family , which is characterized by the BAG domain . The BAG domain is a conserved region located at the C-terminus of the BAG-family proteins that binds to the ATPase domain of Hsc70 and has general nucleotide exchange activities towards Hsc70 ( Takayama and Reed , 2001; Kabbage and Dickman , 2008 ) . Therefore , proteins containing the BAG domain often functions as a co-chaperone protein . Aside from the formation of the BAG-Hsc70 interaction , BAG proteins functionally interact with many proteins to regulate cellular functions . The expression of BAG2 has been detected in many tissues . Recently , it was reported that BAG2 interacts with and stabilizes PINK1 and ataxin3-80Q , proteins involved in neurological diseases , through inhibiting their ubiquitination and degradation ( Che et al . , 2013 , 2014 ) . BAG2 can also deliver Tau to the proteasome for protein degradation independently of ubiquitination ( Carrettiero et al . , 2009 ) . In this study , we found that BAG2 preferentially binds to mutp53 at the DBD domain . BAG2 can interact with many different mutp53 , including several tumor-associated mutational hotspots . This BAG2-mutp53 interaction is conserved in both human tumor cells and mouse tissues . It is unclear how BAG2 can discriminate between the DBD of a diverse range of mutp53 proteins and wtp53 . BAG2 is a co-chaperone protein . It is possible that conformational changes of mutp53 proteins lead to its association with chaperone and co-chaperone proteins . It remains unclear whether BAG2 interacts with mutp53 directly or interacts with mutp53 through other protein , such as Hsc70 , which will be of interest to investigate in future studies . The role of BAG2 in tumor is poorly understood . In this study , we found that mutp53 interacted with BAG2 and promoted the translocation of BAG2 from the cytoplasm to the nucleus , where BAG2 inhibited the binding of MDM2 to mutp53 and the ubiquitination and degradation of mutp53 protein mediated by MDM2 . It is unclear how the BAG2-mutp53 interaction interferes with the MDM2-mutp53 interaction since MDM2 binds to the N-terminus of mutp53 whereas BAG2 binds to the DBD of mutp53 . Future studies are needed to further understand its mechanism . Results from this study showed that BAG2 promoted mutp53 protein accumulation in tumor cells , which in turn promoted mutp53 GOF in tumorigenesis ( Figure 7G ) . Knockdown of endogenous BAG2 significantly inhibited cell proliferation , migration , metastasis and chemoresistance of tumor cells in a largely mutp53-dependent manner in cultured cells and/or in mice . These results strongly suggest that targeting BAG2 could be developed as a novel strategy to destabilize mutp53 and inhibit its GOF in tumorigenesis . Importantly , analysis of several database from Oncomine showed that BAG2 is frequently overexpressed in many types of cancer ( Figure 7A ) . Overexpression of BAG2 is significantly associated with poor prognosis in different types of cancer ( Figure 7B–E ) . Furthermore , our results showed that BAG2 overexpression in colorectal tumors is significantly associated with mutp53 protein accumulation ( Figure 7F ) . These results strongly suggest that BAG2 plays an important role in tumorigenesis and promoting mutp53 accumulation and GOF is a novel mechanism for BAG2 in tumorigenesis . It is unclear why not all tumors with BAG2 overexpression showed the accumulation of mutp53 protein . It is possible that additional mechanisms are involved in the regulation of mutp53 protein levels and/or the BAG2-mutp53 interaction . It will be interesting to examine whether BAG2 displays weak or no interaction with mutp53 protein in this subgroup of tumor samples in future studies . It is also worth noting that while normal tissues from Trp53R172H/R172H mice express a lot of BAG2 , there is a limited amount of interacted BAG2-mutp53 protein complex and no clear accumulation of mutp53 proteins in the normal tissues ( Figure 1C and Figure 1—figure supplement 1 ) . These results suggest that some tumor-specific events might contribute to the effect of BAG2 on mutp53 accumulation . Data from cBioportal showed amplification of the BAG2 gene in many types of human tumors , suggesting that gene amplification is an important mechanism for BAG2 overexpression in human tumors . Considering that less than 10% of tumors had the amplification of BAG2 in majority of tumor types , it is possible that additional mechanisms contribute to BAG2 overexpression in tumors , which needs further investigation in future studies . Taken together , results from this study demonstrate that BAG2 interacts with mutp53 to prevent its degradation by MDM2 , leading to mutp53 accumulation in tumor cells and enhanced mutp53 GOF in tumorigenesis . Knockdown of BAG2 greatly reduces mutp53 protein levels in tumor cells and greatly compromises mutp53 GOF in tumorigenesis , including tumor growth , metastasis and chemoresistance . Considering that BAG2 is frequently overexpressed in cancer cells , our findings revealed a new and important mechanism for mutp53 protein accumulation in tumors . Trp53 is the most frequently-mutated gene in tumors . Mutp53 protein is frequently accumulated in tumors , which is critical for mutp53 GOF in tumor development . Therefore , mutp53 has become an extremely attractive target for tumor therapy . Our findings that BAG2 promotes mutp53 protein accumulation and mutp53 GOF in tumorigenesis strongly suggest that BAG2 could be a potential target for cancer therapy in tumors containing mutp53 . Human lung cancer H1299 , osteosarcoma Saos2 , breast cancer SK-BR-3 , MDA-MB-468 , colorectal cancer HT29 , SW480 and hepatocellular carcinoma Huh-7 cell lines were obtained from ATCC ( Manassas , VA ) . Human HCT116 p53R248W/− cells were gifts from Dr Bert Vogelstein at Johns Hopkins University . Stable cell lines expressing mutp53 R175H , R248Q and R273H were established as previously described ( Zheng et al . , 2013 ) . p53−/− mice were obtained from Jackson Laboratory ( Bar Harbor , ME ) and Trp53R172H/R172H mice were gifts from Dr Gigi Lozano at MD Anderson Cancer Center . Expression vectors of BAG2-HA ( pcDNA-HA-BAG2 ) were gifts from Dr Cam Patterson at University of North Carolina . Expression vectors of mutp53 fragments containing different domains were obtained by using site-directed mutagenesis to introduce R175H mutation into expression vectors of wtp53 fragments containing different domains , which were generous gifts from Dr Xinbin Chen at University of California , Davis . R175H mutp53NLS expression vectors were obtained by using site-directed mutagenesis . Primers used for site mutagenesis and cloning for mutp53 fragments , Flag-tagged FL BAG2 and BAG2 fragments are listed in Table 2 . Retroviral shRNA vectors against human BAG2 were purchased from Open Biosystems ( Thermo Scientific , Waltham , MA , Cat#V2LHS-27769 ) . Two different siRNA oligos against MDM2 were purchased from Qiagen ( Germantown , MD , Cat#SI00300846 ) and Dharmacon ( Lafayette , CO , Cat#M-003279-01 ) . Two different siRNA oligos against BAG2 were purchased from IDT ( Coralville , Iowa ) . siRNA targeting BAG2: siRNA-1: 5′-GUU GGC UUU AGC GUU GAU CUU CGC CUG-3′; siRNA-2: 5′-GUG UCA GUA GAA ACA AUU AGA AAC C-3′ . 5-FU and MG132 were purchased from Sigma ( St . Louis , MO ) . 10 . 7554/eLife . 08401 . 023Table 2 . Sequences of the primer sets used for site-directed mutagenesis and amplifying p53 and BAG2 fragmentsDOI: http://dx . doi . org/10 . 7554/eLife . 08401 . 023Name of fragmentsPrimer sequencesFor site-directed mutagenesis Mutp53 R175H-HA P1 ( aa 1–363 ) , P2 ( aa 43–393 ) , P3 ( aa 43–363 ) Forward5′-GAG GTT GTG AGG CAC TGC CCC CAC CAT-3′Reverse5′-ATG GTG GGG GCA GTG CCT CAC AAC CTC-3′ R175H mutp53NLSForward 15′-GTT GGG CAG TGC TGC CGC AGT GCT CCC TGG GGG CAG-3′Reverse 15′-CTG CCC CCA GGG AGC ACT GCG GCA GCA CTG CCC AAC-3′Forward 25′-TGA AAT ATT CTC CAT CCA GTG GTG CCG CCG CTG GCT GGG GAG AGG AGC TGG TGT TGT TG-3′Reverse25′-CAA CAA CAC CAG CTC CTC TCC CCA GCC AGC GGC GGC ACC ACT GGA TGG AGA ATA TTT CA-3′For amplifying p53 and BAG2 fragments Mutp53 R175H-HA , P4 ( aa 93–393 ) Forward5′-GCG AAT TCA CCA TGG GCT ACC CAT ACG ATG TTC CAG ATT ACG CTC TGT CAT CTT CTG TCC CTT-3′Reverse5′-GAT CGA ATT CTC AGT CTG AGT CAG GCC CTT-3′ Mutp53 R175H-HA , P5 ( aa 93–325 ) , wtp53-DBD , Mutp53 R248W-DBD Mutp53 R273H-DBDForward5′-GCG AAT TCA CCA TGG GCT ACC CAT ACG ATG TTC CAG ATT ACG CTC TGT CAT CTT CTG TCC CTT-3′Reverse5′-GCG AAT TCT CAT CCA TCC AGT GGT TTC TT-3′ Mutp53 R175H-HA , P6 ( Δaa 101–300 ) Forward 15′-GCG AAT TCA CCA TGG GCT ACC CAT ACG ATG TTC CAG ATT ACG CTG AGG AGC CGC AGT CAG ATC C-3′Reverse 15′-CTT AGT GCT CCC TGG CTG GGA AGG GAC AGA-3′Forward 25′-TCT GTC CCT TCC CAG CCA GGG AGC ACT AAG-3′Reverse 25′-GAT CGA ATT CTC AGT CTG AGT CAG GCC CTT-3′ BAG2-FlagForward5′-CGG AAT TCA TGG CTC AGG CGA AGA-3′Reverse5′-CGG GAT CCA TTG AAT CTG CTT TCA GCA T-3′ BAG2 B1-FlagForward5′-CGG AAT TCA TGG CTC AGG CGA AGA-3′Reverse5′-CGG GAT CCT CTT CCC ATC AAA CGG TT-3′ BAG2 B2-FlagForward5′-CGG AAT TCA CCA TGG GAA GAA CTC TCA CCG TT-3′Reverse5′-CGG GAT CCA TTG AAT CTG CTT TCA GCA T-3′ To determine mutp53 binding partners in mouse tissues , mouse mutp53 protein complexes were purified from lysates from tumor and normal tissues of mutp53R172H/R172H mice by IP using anti-p53 ( FL393 ) beads and eluted with 0 . 1 M Glycine solution . Eluted materials were separated in a 4–15% Tris SDS gel and visualized by silver staining using the silver staining kit ( Invitrogen , Grand Island , NY ) and coomassie blue staining . Coomassie blue-stained protein bands were excised from the gel and subsequently analyzed by LC-MS/MS at the Biological MS facility of Rutgers University . IP assays were performed as previously described ( Zheng et al . , 2013 ) . In brief , 1 mg cell or tissue lysates in NP-40 buffer were used for IP using anti-p53 ( DO-1 for human cells and FL393 for mouse tissues , Santa Cruz , Dallas , Texas ) , anti-HA and anti-Flag antibodies to pull down mutp53 , BAG2-HA and BAG2-Flag protein , respectively . Standard Western blot assays were used to analyze the levels of protein . Nuclear extracts were prepared by using Qproteome Nuclear Protein Kit ( Qiagen ) . Antibodies against p53 ( FL393; 1:2000 dilution; Santa Cruz ) , MDM2 ( 2A10; 1:1000 dilution ) , Flag ( 1:10 , 000 dilution; Sigma ) , BAG2 ( 1:1000 dilution; Aviva Systems Biology ) , HA ( 1:4000 dilution; Roche ) , α-Tubulin ( C-5286; 1:1000 dilution; Santa Cruz ) , Lamin A/C ( SC-7293; 1:1000 dilution; Santa Cruz ) , cleaved-caspase 3 ( D175; 1:1000 dilution; Cell Signaling ) , and β-actin ( 1:20 , 000 dilution; Sigma ) were used in this study . IF staining was performed as previously described ( Zheng et al . , 2013 ) . Antibodies against p53 ( FL393 ) and Flag were used to detect p53 and BAG2-Flag , respectively . Slides were then incubated with Alexa Fluor 555 Goat Anti-Rabbit IgG ( H + L ) and Alexa Fluor 488 Goat Anti-mouse IgG ( H + L ) ( Invitrogen ) . Nuclei were stained with 4′ , 6-diamidino-2-phenylindole ( DAPI; Vector , Burlingame , CA ) . RNA from cells was prepared with RNeasy kit ( Qiagen ) . RNA from formalin fixed and paraffin-embedded colorectal tumor sections was prepared with High Pure miRNA Isolation kit ( Roche , Indianapolis , IN ) . The cDNA was prepared by using High Capacity cDNA Reverse Transcription Kit ( Applied Biosystems , Grand Island , NY ) . Primers for Taqman real-time PCR assays were purchased from Applied Biosystems . The expression of genes was normalized with the β-actin gene . In vivo ubiquitination assays were performed as previously described ( Liu et al . , 2014 ) . In brief , cells were transfected with different expression vectors , including mutp53 R175H , BAG2-HA and His-ubiquitin , or transfected with siRNA against BAG2 together with His-ubiquitin expression vectors . At 24 hr after transfection , cells were treated with MG132 for 6 hr . The levels of mutp53 ubiquitination were determined by IP using DO-1 antibody followed by Western blot assays with an anti-ubiquitin antibody ( P4D1; 1:1000; Santa Cruz ) . Annexin V staining was used to determine apoptosis as previously described ( Yu et al . , 2014 ) . In brief , cells were stained by using Muse Annexin V and Dead Cell Assay Kit ( Millipore ) and analyzed in a bench flow cytometry , the Muse Cell Analyzer ( Millipore , Billerica , MA ) . The transwell system ( BD Biosciences , San Jose , CA ) was employed for cell migration assays as previously described ( Zheng et al . , 2013 ) . In brief , cells in FBS-free medium were seeded into upper chambers . The lower chamber was filled with medium supplemented with 10% FBS . Cells on the lower surface of upper chambers were counted after culturing at 37°C for 24 hr . Anchorage-independent growth assays were performed as previously described ( Li et al . , 2014 ) . In brief , cells were seeded in 6-well plates coated with media containing 0 . 6% agarose , and cultured in media containing 0 . 3% agarose . Colonies were stained and counted after 2–3 weeks . Cells ( 5 × 106 in 0 . 2 ml PBS ) were injected subcutaneously ( s . c . ) into 8-week-old BALB/c athymic nude mice ( Taconic ) . Tumor volumes were measured every 2 days for 3 weeks . Tumor volume = 1/2 ( length × width2 ) ( n = 6 mice/group ) . Tumor samples were processed for routine histopathological examination . In vivo lung metastasis assays were performed as previously described ( Zheng et al . , 2013 ) . In brief , HCT116 p53R248W/− and HCT116 p53−/− cells with or without knockdown of BAG2 by shRNA vectors ( 1 × 106 in 0 . 1 ml PBS ) were injected into 8-week-old nude mice via the tail vein ( n = 8 mice/group ) . The mice were sacrificed at 6 weeks after the inoculation . The numbers of lung tumors were counted under a dissecting microscope and confirmed by histopathological analysis . The areas of tumor nodules were quantified in 8 representative images taken at 10 × magnification by using the imageJ software . Animal protocols were approved by the IACUC committee of Rutgers University . PrognoScan ( http://www . prognoscan . org/ ) , which has a large collection of publicly available database with microarray data and clinical information ( Mizuno et al . , 2009 ) , was used to analyze the prognostic power of BAG2 expression levels in colorectal cancer patients ( GSE14333 , Sieber et al . , 2010 ) , lung cancer patients ( GSE14814 , Tsao et al . , 2010 ) , breast cancer patients ( GSE12276 , Bos et al . , 2009 ) , and soft tissue cancer patients ( GSE30929 , Gobble et al . , 2011 ) . A cohort of the de-identified colorectal cancer tissues with known p53 mutation status and p53 protein levels was obtained from the database of the First Affiliated Hospital of Harbin Medical University ( Harbin , China ) with an IRB approval ( Zheng et al . , 2013 ) . None of these patients received pre-surgical chemotherapy . The differences in xenograft tumor growth among groups were analyzed for statistical significance by ANOVA , followed by Student's t-tests using a GraphPad Prism software . Kaplan–Meier statistics were performed to analyze the significance of differences in survival of patients among different groups using software in PrognoScan website . All other p values were obtained using Student's t-test or χ2 test . Values of p < 0 . 05 were considered to be significant .	Cancer can develop if cells in the body acquire mutations that enable them to grow rapidly to form a mass called a tumor . The gene that encodes a protein called p53 is the most commonly mutated gene in human tumors . Most of these mutations result in the production of mutant p53 proteins that are similar in size to the normal protein , but do not work in the same way . The normal p53 protein is known as a ‘tumor suppressor’ because it promotes the repair of damaged genetic material and stops the cell from dividing while this repair is underway . Also , it can instruct a cell to die if the damage is too great to repair . However , many of the mutant p53 proteins stop performing these roles and gain new activities that promote tumor growth instead . These activities often rely on the mutant p53 proteins accumulating to very high levels , but it is not clear why this happens . Here , Yue , Zhao et al . used biochemical techniques to search for other proteins that can bind to mutant p53 proteins . The experiments reveal that a protein called BAG2 binds to mutant p53 and promotes its accumulation in cancer cells , which increases the activity of mutant p53 in driving tumor growth . Loss of BAG2 leads to a reduction in the level of mutant p53 in cells and inhibits the activity of mutant p53 . Using a public database of genetic data from human tumors , Yue , Zhao et al . found that human tumor cells often contain higher levels of BAG2 than normal cells . Furthermore , patients with tumors that had high levels of BAG2—and therefore accumulated mutant p53 proteins—were less likely to have positive outcomes after medical treatment . Yue , Zhao et al . 's findings suggest that increased production of BAG2 in many tumors may be responsible for the accumulation of mutant p53 proteins that drive tumor growth . A future goal is to develop a new treatment strategy that targets BAG2 in tumors to prevent the accumulation of mutant p53 proteins and therefore block the growth of tumors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2015	BAG2 promotes tumorigenesis through enhancing mutant p53 protein levels and function
Formation of the three embryonic germ layers is a fundamental developmental process that initiates differentiation . How the zebrafish pluripotency factor Pou5f3 ( homologous to mammalian Oct4 ) drives lineage commitment is unclear . Here , we introduce fluorescence lifetime imaging microscopy and fluorescence correlation spectroscopy to assess the formation of Pou5f3 complexes with other transcription factors in real-time in gastrulating zebrafish embryos . We show , at single-cell resolution in vivo , that Pou5f3 complexes with Nanog to pattern mesendoderm differentiation at the blastula stage . Later , during gastrulation , Sox32 restricts Pou5f3–Nanog complexes to the ventrolateral mesendoderm by binding Pou5f3 or Nanog in prospective dorsal endoderm . In the ventrolateral endoderm , the Elabela / Aplnr pathway limits Sox32 levels , allowing the formation of Pou5f3–Nanog complexes and the activation of downstream BMP signaling . This quantitative model shows that a balance in the spatiotemporal distribution of Pou5f3–Nanog complexes , modulated by Sox32 , regulates mesendoderm specification along the dorsoventral axis . Oct4 is a key transcription factor ( TF ) in the pluripotency regulatory network ( Boyer et al . , 2005; Esch et al . , 2013; Loh et al . , 2006; Pan et al . , 2002 ) . Previous studies have suggested that Oct4 drives cell lineage commitment in a dose-dependent manner ( Niwa et al . , 2000; Thomson et al . , 2011 ) , switching gene regulatory regions ( Aksoy et al . , 2013 ) by diffusive behaviors ( Kaur et al . , 2013; Plachta et al . , 2011 ) and mediating changes in chromatin structure ( Abboud et al . , 2015; Carey et al . , 2014; Hogan et al . , 2015 ) . The mechanism by which Oct4 is able to specify naive cells in a pluripotent state into one of the three germ layers is not completely understood . The function of TFs can be modulated by their interactions with other TFs ( Phair et al . , 2004 ) . In vitro assays of Oct4 protein interactions have been performed ( Liang et al . , 2008; van den Berg et al . , 2010; Wang et al . , 2006 ) ; however , they do not reflect the complexity of a live embryo and do not provide information about the spatiotemporal complex formation that occurs during embryonic development . Therefore , to determine the mechanism by which Oct4 regulates cell fate decisions and patterning , it is important to investigate Oct4 complex formation with other TFs in the context of a developing embryo . Mammalian Pou5f1/Oct4 can functionally replace its paralogue Pou5f3 ( Frankenberg et al . , 2014 ) in zebrafish embryos , with evidence showing that overexpressed Oct4 can rescue the phenotype of maternal-zygotic ( MZ ) spg embryos that lack Pou5f3 function ( Onichtchouk et al . , 2010 ) . Maternal Pou5f3 regulates dorsoventral ( DV ) patterning ( Belting et al . , 2011; Reim and Brand , 2006 ) and endoderm formation ( Lunde et al . , 2004; Reim et al . , 2004 ) , whereas the establishment of the mid–hindbrain boundary requires zygotic expression of Pou5f3 ( Belting et al . , 2001; Reim and Brand , 2002 ) . Pou5f3 induces mesendoderm ventralization through activation of the BMP pathway and the expression of the Vent ( Vent , Ved and Vox ) family of TFs ( Reim and Brand , 2006 ) . In addition , although Pou5f3 is required in mesendoderm progenitors for sox17 activation to specify endoderm formation ( Lunde et al . , 2004; Reim et al . , 2004 ) , it is not required for upstream regulators of endoderm , which are properly induced in MZspg ( Lunde et al 2004; Reim et al . , 2004 ) . In contrast , Nanog is critical for endoderm induction through the Mxtx-Nodal pathway where it induces sox32 in mesendodermal cells and other early , endoderm regulators , such as gata5 , mixer , nrd1 and nrd2 ( Xu et al . , 2012 ) . Uniquely , Sox32 , in the presence of Pou5f3 , activates sox17 expression in endodermal cells ( Alexander et al . , 1999; Kikuchi et al . , 2001; Lunde et al . , 2004; Reim et al . , 2004 ) . Loss- and gain-of-function genetics experiments , as well as investigations at the mRNA level , have sought to identify various roles for Pou5f3 ( Belting et al . , 2001; Burgess et al . , 2002; Lunde et al . , 2004; Onichtchouk et al . , 2010; Reim and Brand , 2006 ) , Nanog ( Schuff et al . , 2012; Xu et al . , 2012 ) and Sox32 ( Kikuchi et al . , 2001; Reim et al . , 2004 ) during zebrafish development . Here , we exploit fluorescence lifetime imaging microscopy ( FLIM ) and fluorescence correlation spectroscopy ( FCS ) to study , at the protein level , the TF complexes and dynamics that underlie cell fate commitment in vivo . We present a quantitative model to describe how Pou5f3–Nanog complexes , modulated by Sox32 , can specify mesendoderm cell lineage differentiation in a spatiotemporal manner along the DV axis . To investigate how Pou5f3 controls early cell lineage differentiation in vivo , we used a phenotype complementation assay to rescue MZ spg mutant embryos with a GFP-Oct4 fusion protein . The GFP-Oct4 fusion protein was able to complement the spg phenotype in 30% of injected embryos . Because rescued embryos could only be identified from 75% epiboly onwards , we could not analyze earlier developmental events . Alternatively , morpholino ( MO ) -mediated knockdown of maternal Pou5f3 specifically blocks Pou5f3 activity in 100% of injected embryos , which arrest at the blastula stage ( Burgess et al . , 2002 ) . This depletion approach allowed us to discriminate embryos that are rescued by the GFP-Oct4 mRNA from those not rescued , which remained arrested at the blastula stage . Figure 1—figure supplement 1 shows the phenotypes of pou5f3 morphants and the details of the rescue . FCS ( Figure 1a ) has been previously used to study dynamic processes , such as blood flow ( Pan et al . , 2007 ) or morphogen gradients ( Yu et al . , 2009 ) in living zebrafish embryos . Recent studies have described the use of FCS to analyze TF protein activity in iPScells ( Lam et al . , 2012 ) and pre-gastrula mouse embryos ( Kaur et al . , 2013 ) . In cells , TFs can be found free ( free fraction , F1 ) or as complexes poised to interact with DNA and regulate gene expression ( bound fraction , F2 ) . Using FCS ( Figure 1a ) , we hence sought to ascertain fluctuations in the fluorescence intensity of GFP-Oct4 over a timeframe of milliseconds and calculate the autocorrelation functions ( ACFs ) at different developmental stages . To obtain GFP-Oct4 protein concentrations and diffusion kinetics of single cells in rescued embryos , the ACFs were fit using the two-component anomalous diffusion model ( Material and methods and Figure 1—figure supplement 2 ) . At the blastula stage ( oblong stage; 3 . 5 hpf ) , the GFP-Oct4 concentration was 44 . 39 ± 1 . 54 nM ( Figure 1c–e and Figure 1—source data 1 ) . Non-rescued embryos arrested at the oblong stage ( Figure 1b ) had similar Oct4 concentrations ( 43 . 90 ± 18 . 30 nM ) to those of rescued embryos ( Figure 1c–e and Figure 1—source data 1 ) . However , the DNA-bound fraction ( F2 ) was significantly lower in the non-rescued embryos ( 0 . 19 ± 0 . 08 ) as compared with the rescued ones ( 0 . 27 ± 0 . 01 ) ( p<0 . 0001; Figure 1c–e and Figure 1—source data 1 ) . When a construct of Oct4 lacking its homeodomain ( GFP-Oct4ΔHD ) was used instead , the DNA-bound fraction further decreased , and all embryos arrested at the oblong stage ( 0 . 12 ± 0 . 01 ) ( p<0 . 0001; Figure 1c–e and Figure 1—source data 1 ) . These results suggest that specific levels of the Oct4 DNA-bound active fraction—and , by extension , Pou5f3—are crucial for proper embryo gastrulation . 10 . 7554/eLife . 11475 . 003Figure 1 . Oct4 DNA-bound active fraction controls zebrafish gastrulation . ( a ) Schematic diagram of fluorescence correlation spectroscopy ( FCS ) . GFP-tagged nuclear protein is localized in the nucleus of embryonic cells . Fluorescence molecules diffuse through a confocal volume ( <1 μm3 ) within a single-cell nucleus and generate fluctuating fluorescence intensity . The autocorrelation function ( ACF ) of the fluctuation is fit to obtain the absolute protein concentration ( C ) and the diffusion coefficient ( D ) , where N is the number of molecules . Scale bar: 10 μm . ( b ) Lateral view of pou5f3 morphant embryos expressing GFP-Oct4 rescued by GFP-Oct4 mRNA at the blastula [3 . 5 hr post-fertilization ( hpf ) ] and gastrula ( 7 hpf ) stages and non-rescued embryos ( arrested ) at the blastula stage . The non-rescued embryos also express GFP-Oct4 but remain at the blastula stage and do not develop further . Scale bar: 200 μm . ( c ) ACF of the intensity traces of GFP-Oct4 and GFP-Oct4ΔHD in rescued and non-rescued embryos at the blastula stage . The ACF were fit by two-component anomalous diffusion model . Curves are normalized to compare differences in protein activity ( indicated by arrows ) . ( d ) Raw data of residuals from fit curves shown in c . ( e ) Concentration and DNA-bound fraction levels derived from the FCS measurements in c . Values represent the mean ± SEM of data from three to five independent experiments ( n = 39–125 cell nuclei from 10 to 15 embryos **p<0 . 0001; p<0 . 01 ) . n . s . over bars indicates non-significant differences . See also Figure 1—figure supplements 1–3 , Figure 1—source data 1 and Materials and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 00310 . 7554/eLife . 11475 . 004Figure 1—source data 1 . Quantification of GFP-Oct4 concentration and activity in zebrafish rescued and non-rescued embryos . Values for concentration and diffusion parameters were derived from the analysis of FCS data with the ACFs fit by two-component anomalous diffusion model . D1 , D2: Diffusion coefficients of the fast and slow diffusion components , respectively . F2: Slow component fraction . α1 , α2: anomalous parameters of the fast and slow diffusion components , respectively . Values represent mean ± SEM of data from three to five independent experiments ( n represents the number of cell nuclei from 10 to 15 embryos; **p<0 . 0001; p<0 . 01; p<0 . 05 ) . Details of the rescue and the FCS analyses are shown in Figure 1—figure supplements 1–3 and Material and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 00410 . 7554/eLife . 11475 . 005Figure 1—figure supplement 1 . GFP-Oct4 rescues zebrafish Pou5f3 function . ( a ) Phenotypes of pou5f3 knockdown using morpholino ( MO ) antisense . Arrows show a constriction in the interface between the yolk and the blastoderm , prohibiting gastrulation in embryos arrested at the blastula stage after MO injection . Embryos not arrested show zygotic phenotype: they do not develop the mid–hindbrain boundary ( MHB ) at the 24-hpf stage . ( b ) Relative percentages of different phenotypes according to the dose of pou5f3 MO injected . Phenotypes are expressed as a percent of whole ( n = 150–400 embryos per condition ) . As a control , a mismatch MO sequence was injected resulting in 100% of embryos with wt phenotype . 100% of embryos were arrested at the blastula stage using 8 ng of MO . Lower-dosed 4-ng and 2-ng embryos continued through gastrulation but did not develop the MHB , according to the percentages shown . ( c ) Phenotypes of rescued embryos . Rescue of maternal pou5f3 function , as shown by restoring normal gastrulation . Rescue of maternal-zygotic pou5f3 function , as shown by MHB formation at 24 hpf . ( d and e ) Relative percentages of ( d ) maternal and ( e ) maternal-zygotic rescue according to different amounts of injected mRNA . Phenotypes of the rescued embryos are expressed as a percent of whole ( n = 170–400 embryos per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 00510 . 7554/eLife . 11475 . 006Figure 1—figure supplement 2 . One- and two-component anomalous diffusion model for GFP and GFP-Oct4 . ( a ) Autocorrelation function ( ACF ) of free GFP fit by two- and one-component anomalous diffusion models . Two-component anomalous diffusion model converged to effective one-component anomalous diffusion for a real one-component system , such as free GFP . ( b ) Raw data of residuals from fit curves shown in a . ( c ) Quantitative parameters derived from the ACFs . F1: Fast component fraction . D1 , D2: Diffusion coefficients of the fast and slow diffusion components , respectively . α1 , α2: anomalous parameters of the fast and slow diffusion components , respectively . I/N: Ratio of mean intensity extracted directly from the time traces and the mean number of molecules in the focus extracted from the fit , referred to as molecular brightness . n: number of cells . ( d ) The GFP-Oct4-free diffusion coefficient ( D1 ) determined by global fitting and averaging individual fits yields provides similar diffusion time , tauD1 . Values represent mean ± SEM of data from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 00610 . 7554/eLife . 11475 . 007Figure 1—figure supplement 3 . Oct4 concentration and DNA-bound active fraction in embryos rescued with different amount of GFP-Oct4 mRNA . ( a ) Autocorrelation function ( ACF ) of the intensity traces of GFP-Oct4 by the two-component fit anomalous diffusion model in rescued embryos with different amounts of GFP-Oct4 mRNA at blastula ( oblong; 3 . 5 hpf ) and gastrula ( 60% epiboly; 7 hpf ) stages . Double-headed arrow in the graph indicates the difference in the amplitude of the curves , which corresponds to the difference in protein concentrations . ( b ) Raw data of residuals from fit curves shown in a . ( c , d ) Concentration ( c ) and DNA-bound fraction ( d ) derived from the FCS measurements in a . Changes in the GFP-Oct4 concentration at oblong stage in rescued embryos were concordant with the proportion of mRNA injected . However , the Oct4 DNA-bound fraction did not significantly change with varying concentrations . Concentration and Oct4 DNA-bound fraction remained similar in rescued embryos at 60% epiboly . Values represent mean ± SEM of data from three independent experiments ( n > 35 cell nuclei from 10 to 15 embryos; p<0 . 01 ) . n . s . over bars indicates non-significant differences . See also Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 007 We next set out to determine the tissue-specific distribution of the DNA-bound active fraction ( F2 ) of Pou5f3 within a developing embryo at the blastula stage ( oblong; 3 . 5 hpf ) using GFP-Oct4 . In zebrafish , unlike in mammals , it is possible to follow mesendoderm and ectoderm formation during early development ( Fong et al . , 2005 ) ( Figure 2a and Figure 2—figure supplement 1 ) . For that purpose , zygotes were injected with GFP-Oct4 mRNA followed by dextran red as a lineage tracer into two central cells at the 16-cell stage , such that the dextran red-positive cells are entirely ectodermal , whereas 83 . 7% of the negative cells are mesendodermal ( Figure 2—figure supplement 1 ) . FCS measurements were performed for both labeled and unlabeled cells . The ACFs of the intensity traces were fit using a two-component anomalous diffusion model . Mesendodermal cells showed a significantly higher proportion of the DNA-bound active fraction of GFP-Oct4 protein as compared with ectodermal cells ( 0 . 27 ± 0 . 01 vs 0 . 19 ± 0 . 01 , respectively; p<0 . 0001; Figure 2—figure supplement 1 and Figure 2—source data 1 ) . Dextran red-labeled mesendodermal cells were used as a control for the assay ( Figure 2—figure supplement 2 ) . Thus , at the oblong stage , the DNA-bound active fraction of Oct4 was higher in the mesendoderm than in the ectoderm lineage . 10 . 7554/eLife . 11475 . 008Figure 2 . Oct4 and Nanog bind in mesendoderm of zebrafish blastula embryos . ( a ) Schematic location of the presumptive mesendoderm ( ME ) and ectoderm ( EC ) in an embryo at oblong stage [3 . 5 hr post-fertilization ( hpf ) ] . Scale bar: 200 µm . ( b ) Schematic diagram of FLIM-FRET ( Fluorescence Lifetime Microscopy–Forster Resonance Energy Transfer ) . GFP lifetime ( t1 ) of the donor is reduced if an acceptor ( mCherry ) is in close proximity ( 1–10 nm ) ; this is the reduced lifetime ( t2 ) . If the acceptor is not in close proximity ( >10 nm ) to the donor , donor lifetime remains unchanged ( t2 similar to t1 ) . Lifetimes are measured with time-correlated single-photon counting . ( c ) Lifetime values and FLIM images of GFP-Oct4 alone and in the presence of a linked mCherry protein in single cells . Scale bar: 10 µm . In the same graph , lifetime values and FLIM images of GFP-Oct4ΔHD alone and co-expressing mCherry-Nanog in single nuclei . Scale bar: 5 µm . ( d ) Lifetime values and FLIM images of GFP-Oct4 lifetime alone and in the presence of mCherry-Nanog in the nucleus of individual cells from mesendoderm or ectoderm . Scale bar: 5 µm . The percentage of binding is indicated at the top right corner of the FLIM images . Values represent the median and quartile ranges of data from three to five independent experiments ( n = 20–40 cell nuclei from 10 to 15 embryos; p<0 . 0001 ) . n . s . over bars indicates non-significant differences . See also Figure 2—figure supplements 1–4 , Figure 2—source data 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 00810 . 7554/eLife . 11475 . 009Figure 2—source data 1 . Quantification of GFP-Oct4 and GFP-Nanog activity in mesendoderm and ectoderm of wt and morphant zebrafish embryos . Diffusion parameters values were derived from analysis of FCS data with the ACFs fit by two-component anomalous diffusion model . D1 , D2: Diffusion coefficients of the fast and slow diffusion components , respectively . F2: Slow component fraction . α1 , α2: anomalous parameters of the fast and slow diffusion components , respectively . ME: mesendoderm . EC: ectoderm . Values represent mean ± SEM from three to five independent experiments ( n represents the number of cell nuclei from 10 to 15 embryos; p<0 . 0001; p<0 . 001 ) . Details of the MOs and FCS analyses are shown in Figure 2—figure supplements 1 , 3 and Material and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 00910 . 7554/eLife . 11475 . 010Figure 2—source data 2 . FCCS parameters of GFP-Nanog and mCherry-Oct4 in mesendoderm and ectoderm of blastula embryos ( oblong stage; 3 . 5 hpf ) . Diffusion parameters values were derived from analysis of FCCS data with the ACFs and CCF fit by two-component anomalous diffusion model . D1 , D2: Diffusion coefficient of the fast and slow diffusion component , respectively . α1 , α2: anomalous parameter of the fast and slow diffusion component , respectively . Kd: dissociation constant at equilibrium; values were obtained from the slopes of the fitted linear line when plotting the concentration of GFP-Nanog ( CN ) concentration of mCherry-Oct4 ( CO ) versus the concentration of the proteins association ( CNO ) . If the proteins are associated , there will be a linear line; in cases where no association exists , there is no linear relationship ( N . L ) . Association: fraction of proteins diffusing together in the same complex . Details of the FCCS analysis are shown in Figure 2—figure supplement 4 and Materials and methods . ME: mesendoderm . EC: ectoderm . Values represent mean ± SEM from three to five independent experiments with n > 15 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 01010 . 7554/eLife . 11475 . 011Figure 2—figure supplement 1 . GFP-Oct4 dynamics in blastula embryos . ( a ) GFP-Oct4 expressed in the blastoderm at the blastula stage ( oblong; 3 . 5 hpf ) . Ectoderm ( EC ) cells are traced by dextran red; non-labeled cells are mesendoderm ( ME ) . Scale bar: 200 μm . ( b ) Cells from the blastoderm expressing GFP-Oct4 . Staining as in a . Scale bar: 20 μm . ( c ) ACF of the intensity traces of GFP-Oct4 in EC and ME in wild-type embryos . The ACF were fit by two-component anomalous diffusion model . Curves are normalized to compare differences in protein activity . ( d ) Raw data of residuals from fit curves shown in c . ( e ) DNA-bound fraction derived from the previous ACFs . Values represent the mean ± SEM of data from three to five independent experiments ( n > 60 cell nuclei from 10 to 15 embryos; **p<0 . 0001 ) . ( f ) ACF of the intensity traces of GFP-Oct4 in nanog morphant embryos . The ACF were fit by two-component anomalous diffusion model . Curves are normalized to compare differences in protein activity . ( g ) Raw data of residuals from fit curves shown in f . ( h ) DNA-bound fraction derived from the previous ACFs . Values represent the mean ± SEM of data from three to five independent experiments ( n > 60 cell nuclei from 10 to 15 embryos; p<0 . 001 ) . See also Figure 2—source data 1DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 01110 . 7554/eLife . 11475 . 012Figure 2—figure supplement 2 . Dextran red does not interfere in the FCS measurements . ( a ) Embryo ( oblong stage; 3 . 5 hpf ) showing mesendoderm ( ME ) labeled by dextran ( red ) leaving the ectoderm ( EC ) unlabeled . Scale bar: 200 µm . ( b ) ACF of the intensity traces of GFP-Oct4 in ME cells labeled or unlabeled . FCS data were fit by two-component anomalous diffusion model . ( c ) Raw data of residuals from fit curves shown in b . ( d ) DNA-bound fraction and diffusion coefficient derived from the FCS measurements in b . Values represent mean ± SEM of data from three independent experiments ( n = 40–100 cell nuclei from 10 to 15 embryos ) . n . s . over bars indicates non-significant differences . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 01210 . 7554/eLife . 11475 . 013Figure 2—figure supplement 3 . GFP-Nanog dynamics in blastula embryos . ( a ) Cells from the blastoderm ( oblong stage; 3 . 5 hpf ) expressing GFP-Nanog . Ectoderm ( EC ) cells are positive for dextran red; mesendoderm ( ME ) cells are negative . Scale bar: 20 μm . ( b ) ACF of the intensity traces of GFP-Nanog in EC and ME in wild-type embryos . The ACF were fit by two-component anomalous diffusion model . Curves are normalized to compare differences in protein activity . ( c ) Raw data of residuals from fit curves shown in b . ( d ) DNA-bound fraction derived from the previous ACFs . Values represent the mean ± SEM of data from three to five independent experiments ( n > 60 cell nuclei from 10 to 15 embryos; *p<0 . 0001 ) . ( e ) ACF of the intensity traces of GFP-Nanog in EC and ME in pou5f3 morphant embryos . The ACF were fit by two-component anomalous diffusion model . Curves are normalized to compare differences in protein activity . ( f ) Raw data of residuals from fit curves shown in e . ( g ) DNA-bound fraction derived from the previous ACFs . Values represent the mean ± SEM of data from three to five independent experiments ( n > 60 cell nuclei from 10 to 15 embryos; p<0 . 001 ) . See also Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 01310 . 7554/eLife . 11475 . 014Figure 2—figure supplement 4 . Oct4 and Nanog cross-correlate in mesendoderm of blastula embryos . ( a ) Schematic diagram of a setup for fluorescence cross-correlation spectroscopy ( FCCS ) . GFP- and mCherry-tagged nuclear proteins diffuse through a confocal volume ( <1 μm3 ) within a single-cell nucleus and generate fluctuating fluorescence intensities . The autocorrelation functions ( ACFs ) and the cross-correlation function ( CCF ) of the fluctuations are fit to obtain the absolute protein concentration ( C ) and the diffusion coefficient ( D ) , where N is the number of molecules . CCF is shown in black . ( b ) ACFs and CCFs of the intensity traces of GFP-Nanog co-expressed with mCherry-Oct4 in mesendoderm and ectodermal cells of blastula embryos ( oblong stage; 3 . 5 hpf ) fit by two-component anomalous diffusion model . ( c ) Raw data of residuals from fit curves shown in b . ( d ) Kd plots for GFP-Nanog and mCherry-Oct4 associations . Kd values were obtained from the slopes of the fitted linear line . If the proteins are associated , there will be a linear line when plotting the concentration of GFP-Nanog ( CN ) concentration of mCherry-Oct4 ( CO ) versus the concentration of the protein association ( CNO ) , since Kd represents the constant of the association at equilibrium . There is no linear relationship in cases where there is no association . ( e ) Kd plots for GFP co-expressed with mCherry protein ( negative control ) and GFP-mCherry linked protein ( tandem; positive control ) . Values for GFP and mCherry proteins could not be linearly fitted . Association: fraction of proteins diffusing together in the same complex . See also Figure 2—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 014 Next , the GFP-Oct4 DNA-bound active fraction was studied in mesendoderm of nanog morphants at the same stage ( oblong; 3 . 5 hpf ) . A low dose of nanog MO , which did not cause severe developmental defects , was co-injected with GFP-Oct4 mRNA . This co-injection led to a reduction in the bound fraction of GFP-Oct4 from 0 . 27 ± 0 . 01 to 0 . 22 ± 0 . 01 ( p<0 . 001; Figure 2—figure supplement 1 and Figure 2—source data 1 ) . As a control , we carried out the reverse experiment to assess the GFP-Nanog DNA-bound active fraction in pou5f3 morphants , measuring a reduction from 0 . 21 ± 0 . 01 to 0 . 16 ± 0 . 01 ( p<0 . 001; Figure 2—figure supplement 3 and Figure 2—source data 1 ) . These FCS data suggest that the Pou5f3- and Nanog-binding fractions are dependent on each other in mesendoderm lineage . Since Oct4 and Nanog share many genomic binding sites ( Loh et al . , 2006; van den Berg et al . , 2010 ) , we next investigated whether they interact in vivo in a cell lineage-dependent manner . We used FCCS ( Fluorescence Cross-Correlation Spectroscopy; Bacia et al . , 2006; Krieger et al . , 2015 ) to study the diffusion of both proteins simultaneously in mesendoderm and ectoderm at the oblong stage ( 3 . 5 hpf; Figure 2—figure supplement 4 and Figure 2—source data 2 ) . The cross-correlation function indicates whether GFP-Oct4 and mCherry-Nanog diffuse together within the same complex , and by calculating the dissociation protein constants ( Kd ) , we can then determine the binding affinity of GFP-Oct4 and mCherry-Nanog . We measured a Kd of 15 . 34 ± 1 . 6 nM and 61 . 9 ± 7 . 5 nM in the mesendoderm and ectoderm , respectively , suggesting higher binding affinity in mesendoderm at the oblong stage ( Figure 2—figure supplement 4 and Figure 2—source data 2 ) . To confirm the interaction of Oct4 and Nanog in a cell-lineage dependent manner , we next used FLIM-FRET ( Fluorescence Lifetime Imaging Microscopy–Förster Resonance Energy Transfer ) in single cells of the developing embryo . FLIM follows the lifetime of the excited state of fluorescent molecules and can be used to monitor protein–protein interactions via FRET ( Sun et al . , 2011; Margnieanu et al . , 2016 ) ( Figure 2b ) . The proximity of a GFP fusion protein as a donor and an mCherry fusion protein as an acceptor reduces the lifetime of the donor , indicating a protein–protein interaction . Contrarily , no change in the lifetime indicates that the proteins are not interacting ( Figure 2b ) . We have previously shown that Oct4 interacts with Sox2 in ESCs and iPSCs using this method ( Lam et al . , 2012 ) . Here , we used a GFP-mCherry tandem protein as a positive control to document the decrease in GFP lifetime when it interacts with mCherry as compared to the lifetime of GFP alone in zebrafish embryos ( p<0 . 0001; Figure 2c ) . mCherry-Nanog mRNA , co-injected with GFP-Oct4ΔHD mRNA was used as negative control , because the DNA binding domain of Oct4 is needed for its interaction with Nanog and , therefore , the GFP-Oct4ΔHD lifetime should not change in the presence of mCherry-Nanog ( Figure 2c ) . In the nuclei of individual cells within the mesendoderm and ectoderm of oblong stage embryos ( 3 . 5 hpf; Figure 2d ) , we found a significant reduction in the GFP-Oct4 lifetime in the presence of mCherry-Nanog in the mesendoderm ( 1 . 79 ± 0 . 05 ns; p<0 . 0001 ) but not in the ectoderm ( 2 . 28 ± 0 . 01 ns; Figure 2d ) as compared with the GFP-Oct4 lifetime in the absence of mCherry-Nanog ( 2 . 30 ± 0 . 01 ns; Figure 2d ) ; the binding percentages ( Orthaus et al . , 2009 ) in the mesendoderm and ectoderm were 27% and 2 . 2% , respectively ( Figure 2d ) . These FLIM-FRET results demonstrate a higher proportion of Oct4–Nanog complexes in the mesendoderm as compared with that in ectoderm at the oblong stage and suggest that Pou5f3 and Nanog co-regulate mesendoderm targets during gastrulation . Given the proportion of complexes in the oblong stage , we next sought to investigate the spatial localization of Oct4–Nanog complexes in ventral , lateral and dorsal ectoderm and mesendoderm at the 50% epiboly stage ( germ ring; 5 . 7 hpf ) . This time we used FCCS to explore the diffusion of GFP-Nanog and mCherry-Oct4 and their binding affinities in the different ectodermal and mesendodermal areas . At 50% epiboly , the high Kd value indicated a low-binding affinity between Oct4 and Nanog in the ventrolateral ectodermal cells ( Kd: 57 . 09 ± 7 . 06 nM ) , and almost a complete absence of cross-correlation in the dorsal ectoderm ( Kd: 201 ± 49 . 4 nM ) ( Figure 3—figure supplement 1 and Figure 3—source data 1 ) . In contrast , the Kd values in the ventral and lateral mesendoderm ( 5 . 4 ± 0 . 4 nM and 11 . 1 ± 0 . 73 nM , respectively ) indicated a high-binding affinity in these regions as compared with that in the dorsal mesendoderm ( 36 . 8 ± 3 . 5 nM ) ( Figure 3—figure supplement 2 ) and Figure 3—source data 1 ) . We further confirmed these mesendodermal interactions using FLIM-FRET . We found that the GFP-Oct4 lifetime was significantly reduced in the presence of mCherry-Nanog in the ventral and lateral mesendoderm ( 2 . 29 ± 0 . 01 ns in absence of mCherry-Nanog to 2 . 02 ± 0 . 01 ns and 2 . 03 ± 0 . 02 ns , respectively; p<0 . 0001; Figure 3b ) but not in dorsal mesendoderm ( 2 . 31 ± 0 . 01 ns; Figure 3b ) . The percentage of binding was greater than 20% in the ventral and lateral mesendoderm but only 4 . 7% in the dorsal area ( Figure 3b ) . 10 . 7554/eLife . 11475 . 015Figure 3 . Oct4 and Nanog complexes in the ventrolateral mesendoderm . ( a ) wt embryo showing the ventral ( V ) –lateral ( L ) - and dorsal ( D ) -mesendoderm ( ME ) at the 50% epiboly [5 . 7 hr post-fertilization ( hpf ) ] before commencement of involution . LiCl-dorsalized embryo at 5 . 7 hpf . Arrows show the dorsal organizer of the wt embryo and radialized dorsal structures along the germ ring of dorsalized embryos . Scale bar: 200 µm ( b ) Lifetime values and FLIM images of GFP-Oct4 lifetime alone and in the presence of mCherry-Nanog in the nuclei of individual cells measured at different locations within the mesendoderm . Values represent the median and quartile ranges of data from three to five independent experiments ( n = 30–70 cell nuclei from 10 to 15 embryos; ***p<0 . 0001 ) . The percentage of binding is indicated at the top right corner of the FLIM images . Scale bar: 5 µm . n . s . over bars indicates non-significant differences . See also Figure 3—figure supplements 1 , 2 and Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 01510 . 7554/eLife . 11475 . 016Figure 3—source data 1 . FCCS parameters of GFP-Nanog and mCherry-Oct4 in mesendoderm and ectoderm of gastrula embryos ( 50% epiboly; 5 . 7 hpf ) . Diffusion parameters were derived from analysis of FCCS data with the ACFs and CCF fit by two-component anomalous diffusion model . D1 , D2: Diffusion coefficients of the fast and slow diffusion components , respectively . α1 , α2: anomalous parameters of the fast and slow diffusion components , respectively . Kd: dissociation constant at equilibrium; values were obtained from the slopes of the fitted linear line when plotting the concentration of GFP-Nanog ( CN ) concentration of mCherry-Oct4 ( CO ) versus the concentration of the proteins association ( CNO ) . If the proteins are associated , there will be a linear line; in cases where no association exists , there is no linear relationship . Association: fraction of proteins diffusing together in the same complex . Details of the FCCS analysis are explained in the Materials and methods . V-ME: ventral mesendoderm; ME: lateral mesendoderm; D-ME: dorsal mesendoderm; V-EC: ventral ectoderm; D-EC: dorsal ectoderm . Values represent mean ± SEM from three to five independent experiments with n > 15 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 01610 . 7554/eLife . 11475 . 017Figure 3—figure supplement 1 . Nanog and Oct4 cross-correlation in the ectoderm of gastrula embryos . ( a ) ACFs and CCFs of the intensity traces of GFP-Nanog co-expressed with mCherry-Oct4 in ectoderm ( EC ) cells of gastrula ( 50% epiboly; 5 . 7 hpf ) embryos fit by two-component anomalous diffusion model . ( b ) Raw data of residuals from fit curves shown in a . ( c ) Kd plots for GFP-Nanog and mCherry-Oct4 associations . Kd values were obtained from the slopes of the fitted linear line . If the proteins are associated , there will be a linear line when plotting the concentration of GFP-Nanog ( CN ) * concentration of mCherry-Oct4 ( CO ) versus the concentration of the protein association ( CNO ) , since Kd represents the constant of the association at equilibrium . There is no linear relationship in cases where there is no association . Association: fraction of proteins diffusing together in the same complex . See also Figure 3—source data 1DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 01710 . 7554/eLife . 11475 . 018Figure 3—figure supplement 2 . Nanog and Oct4 cross-correlation in ventrolateral mesendoderm of gastrula embryos . ( a ) ACFs and CCFs of the intensity traces of GFP-Nanog co-expressed with mCherry-Oct4 in ventral , lateral and dorsal mesendoderm ( ME ) cells of gastrula embryos ( 50% epiboly; 5 . 7 hpf ) fit by two-component anomalous diffusion model . ( b ) Raw data of residuals from fit curves shown in a . ( c ) Kd plots for GFP-Nanog and mCherry-Oct4 associations . Kd values were obtained from the slopes of the fitted linear line . If the proteins are associated , there will be a linear line when plotting the concentration of GFP-Nanog ( CN ) * concentration of mCherry-Oct4 ( CO ) versus the concentration of the protein association ( CNO ) , since Kd represents the constant of the association at equilibrium . There is no linear relationship in cases where there is no association . Association: fraction of proteins diffusing together in the same complex . See also Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 018 Finally , we used LiCl treatment , which activates Wnt/β-catenin signaling ( Shao et al . , 2012 ) , to assess the distribution of Oct4–Nanog complexes in dorsalized embryos . Embryos at the 32-cell stage were treated with LiCl and monitored for changes in Oct4–Nanog complexes using FLIM-FRET . As expected , LiCl-treated embryos exhibited radialized dorsal structures along the germ ring ( Figure 3b ) . The cells from these radial dorsal structures showed a similar percentage of Oct4–Nanog binding to that of cells from the dorsal mesendoderm of non-treated embryos ( 2% and 4 . 7% , respectively; Figure 3b ) . These results suggest that uneven yet specific levels of the Pou5f3–Nanog complex may drive DV mesendodermal patterning . Pou5f3 is known to act in DV patterning ( Reim and Brand , 2006 ) . To address whether Nanog is involved with Pou5f3 in driving DV patterning , we performed a series of Nanog loss-of-function experiments using varying amounts of nanog MO ( Figure 4—figure supplement 1 ) . With 1 . 6 ng of nanog MO , we observed that 90% of morphants were Class ( C ) IV ( poorly developed axial structures , and no apparent eyes , trunk or tail ) ( Figure 4a , b ) ; with 0 . 4 ng of nanog MO , 70% of morphants showed C III ( truncated bodies and hypoplastic eyes ) and C II phenotypes ( normal head structures , a short body axis , a bent tail and no fins ) ( Figure 4a , b ) . Near complete rescue ( C I ) was observed when 0 . 1 ng of nanog mRNA ( lacking the MO-target site ) was co-injected with 0 . 4 ng of MO ( Figure 4a , b ) . Embryos injected with nanog mismatch-MO ( 5 mismatch nucleotides; nanog* ) developed normally ( data not shown ) . 10 . 7554/eLife . 11475 . 019Figure 4 . Pou5f3 and Nanog promote ventral fate . ( a ) nanog MO-injected larvae show severely affected ( Class ( C ) IV ) , less-severely affected ( C III ) , mildly affected ( C II ) and least affected ( C I ) phenotypes . ( b ) Relative percentages of C I , C II , C III and C IV larvae according to dose of nanog MO injected ( 1 . 6 ng of nanog MO injection , n = 167; 0 . 8 ng of nanog MO injection , n = 186; 0 . 4 ng of nanog MO injection , n = 143 ) . Co-injection of nanog MO ( 0 . 4 ng ) with nanog* mRNA ( 0 . 1 ng ) leads to over 80% wt-like larvae as opposed to 20% wt-like larvae in its absence ( n = 126 ) . ( c–e ) Embryos are at 50%-epiboly except where indicated . Embryos are in top views except lateral views for bmp2b- , oct4- and sox17-stained embryos . Dorsal is to the right-hand side . Markers were analysed following injection of 0 . 8 ng of nanog MO at the 1-cell stage . ( c ) chd expression in the dorsal margin is expanded ventrally in 30%-epiboly nanog morphants relative to wt embryos ( 86% , n = 40 ) , and is uniformly expressed in the blastoderm of nanog morphants at 50%-epiboly relative to wt embryos ( 94% , n = 66 ) . gsc expression in the prospective shield is expanded ventrally within the germ ring in nanog morphants relative to wt embryos at the early gastrula stage ( 71% , n = 47 ) . bmp2b expression in the ventral ectoderm and organizer is markedly reduced in nanog morphants relative to wt embryos at mid-gastrulation ( 96% , n = 45 ) . Expression of vox , a BMP target , is greatly diminished in nanog morphants relative to wt embryos ( 86% , n = 68 ) . Expression of vent in the ventral margin is nearly absent in nanog morphants relative to wt embryos ( 95% , n = 56 ) . At the early-gastrula stage , pou5f3 expression in the blastoderm is reduced in nanog morphants compared to wt embryos ( 97% , n = 44 ) . ( d ) Effect of MZspg , nanog MO and MZspg/nanog MO on the expression of chd and vox . chd expression in the organizer of wt embryos ( 100% , n = 92 ) is ventrally expanded in MZspg embryos ( 98% , n = 75 ) and nanog morphants ( 90% , n = 97 ) . In nanog MO-injected MZspg embryos , chd expression is further expanded in the entire blastoderm ( 92% , n = 62 ) . vox expression in the ventral margin of wt embryos ( 100% , n = 96 ) is markedly reduced in MZspg embryos ( 97% , n = 60 ) and nanog morphants ( 88% , n = 84 ) . In nanog MO-injected MZspg embryos , vox expression is completely lost ( 94% , n = 58 ) . ( e ) Effect of oct4 mRNA in nanog MO , and nanog mRNA in MZspg mutant on chd expression . chd expression is ventrally expanded in MZspg embryos ( 98% , n = 75 ) relative to wt embryos ( 100% , n = 92 ) . nanog mRNA cannot cause chd expansion when injected into wt embryos ( 93% , n = 65 ) and cannot rescue chd ventral expansion when injected into MZspg embryos ( 94% , n = 52 ) . chd expression is ventrally expanded in nanog morphants ( 90% , n = 97 ) relative to wt embryos ( 100% , n = 92 ) . Pou5f3 mRNA cannot cause chd expansion when injected into wt embryos ( 92% , n = 67 ) and cannot rescue chd ventral expansion caused by nanog depletion when co-injected with nanog MO ( 88% , n = 64 ) . Data are from three to five independent experiments ( n = 40–150 ) . See also Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 01910 . 7554/eLife . 11475 . 020Figure 4—figure supplement 1 . Nanog controls dorsoventral ( DV ) patterning . ( a ) Maternal nanog mRNA is ubiquitously expressed in each blastomere of the eight-cell stage embryo . After mid-blastula transition , nanog expression is uniform in the blastoderm at the dome stage . At 50% epiboly , nanog mRNA is detected in all cells of the blastoderm . nanog mRNA accumulation is minimal by 100% epiboly . At 1 day , nanog transcripts are no longer detected . ( b ) Position of the nanog ATG-blocking morpholino relative to the translation initiation start of the zebrafish nanog allele . ( c ) RT-PCR analysis shows that , relative to actin , nanog expression in the developing embryo lasts until the onset of neurulation at early gastrulation , with high maternal contribution in oocytes . ( d ) qRT-PCR analysis reveals that nanog depletion or overexpression differentially affects the transcription of selected marker genes . Values represent mean ± SE of data from three independent experiments; p<0 . 01 except for cyclinB1 . ( e ) nanog morphants , relative to wt embryos , lack non-neural ectoderm , as documented by the absence of gata2 expression at the end of gastrulation ( 88% , n = 52 ) . Similarly , expression of dlx3 is lost in nanog morphants relative to wt embryos ( 85% , n = 48 ) . Expression of the forebrain–midbrain marker otx2 is radialized in nanog morphants at 100% epiboly relative to wt embryos ( 95% , n = 41 ) . ntl expression in the dorsal midline and myoD in the somites are mis-localized and absent , respectively , in nanog morphants relative to wt embryos at the 8–10 somite stage ( 97% , n = 52 ) . The endoderm marker sox17 is dramatically reduced in nanog morphants relative to wt embryos at the mid-gastrula stage ( 92% , n = 65 ) . sqt expression in the blastoderm margin at 30% epiboly is markedly reduced in nanog morphants relative to wt embryos ( 94% , n = 39 ) . Presumptive segmental plate expression of her1 at the 1–4 somite stage is lost in nanog morphants relative to wt embryos ( 95% , n = 45 ) . The ventrolateral mesoderm markers eve1 and tbx6 are reduced and absent , respectively , in nanog morphants relative to wt at 50% epiboly ( 97% , n = 25; 100% , n = 25 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 020 We further analyzed the specification of the DV axis , which is driven by a gradient of BMP activity , by measuring the expression of specific markers . In particular , the expression of chd at the dorsal pole was up-regulated dramatically: at the 30% epiboly stage , chd was expressed circumferentially , whereas at 50% epiboly , chd transcripts were ubiquitous ( Figure 4c ) . gsc , a dorsally expressed gene primarily controlled by the Nodal pathway , was also up-regulated but to a lesser extent , suggesting that Nanog largely controls the levels of BMP signaling ( Figure 4c ) . Whereas bmp2b expression was localized at the embryonic shield and broadly expressed ventrally at the 70% epiboly stage in controls , nanog morphants showed significantly down-regulated bmp4 ( data not shown ) and bmp2b expression ( Figure 4c ) . Consistently , the expression of direct targets of BMP signaling , such as vox and vent , was greatly reduced in nanog morphants at the 50% epiboly stage ( Figure 4c ) . The spatial distribution of pou5f3 transcripts in nanog morphants during 50% epiboly was not affected , but the level of expression was reduced ( Figure 4c ) . Together , these results indicate that nanog knockdown leads to the dorsalization of embryos and that maternal nanog activity is necessary for ventral cell specification by BMPs . The dorsalization and loss of endoderm in nanog morphants is reminiscent of the phenotype of MZspg mutants , which are completely devoid of Pou5f3 ( Lunde et al . , 2004; Reim and Brand , 2006 ) . This supports and reinforces the idea that Pou5f3 and Nanog may participate in the same developmental program . To test this , MZspg mutants and nanog morphants were compared with embryos deficient in both MZspg and nanog using markers of DV patterning . Expansion of the dorsal marker , chd , was more pronounced in MZspg/nanog-deficient embryos as compared with nanog morphants or MZspg mutants at the 50% epiboly stage ( Figure 4d ) . Conversely , the ventral marker , vox , was absent in MZspg/nanog-deficient embryos but only reduced in both nanog morphants and MZspg mutants ( Figure 4d ) . Lastly , we tested if nanog overexpression could compensate for the loss of pou5f3 or if pou5f3 could compensate for the depletion of nanog . Injection of nanog mRNA into MZspg mutants failed to rescue the expression of chd ( Figure 4e ) . Similarly , injection of pou5f3 mRNA into nanog morphants could not rescue DV patterning ( Figure 4e ) . These results suggest that pou5f3 and nanog display overlapping functions during DV patterning but cannot compensate for one another . This is consistent with the notion that Nanog must cooperate with Pou5f3 to promote ventral fate . The paucity of Oct4–Nanog complexes in the dorsal mesendoderm at 50% epiboly suggested that some other TFs could partner with Oct4 to replace Nanog . Lineage tracing experiments have shown that dorsal endodermal precursors arise from cells in the dorsal mesendoderm near the margin before the cells involute to form the hipoblast , and that those situated above are restricted to the mesoderm ( Warga and Nüsslein-Volhard , 1999 ) . The dorsal side is easily detected by the presence of the ‘shield’ at 50% epiboly , but the ventrolateral endodermal cells are derived from dispersed precursors located close to the margin , which makes them impossible to distinguish morphologically at this stage . Previously , we have shown that Oct4 and Nanog form complexes in the whole mesendoderm at 3 . 5 hpf and , by 4 hpf , Sox32 expression starts in the dorsal endodermal cells and extends to ventrolateral endoderm ( Thisse et al . , 2001 ) . Thus , we next tested whether Sox32 could compete with Nanog for Oct4 binding in the prospective dorsal endoderm . In sox32 morphants , we saw a significant reduction in the lifetime of GFP-Oct4 in the presence of mCherry-Nanog ( 2 . 31 ± 0 . 01 ns to 2 . 16 ± 0 . 03 ns , p<0 . 0001; Figure 5a-c ) , and an increase in the binding percentage from 1% to 15% , indicating that the Oct4–Nanog complex forms in the absence of Sox32 in the dorsal endoderm precursors at 50% epiboly ( Figure 5a-c ) . To verify the existence of Oct4–Sox32 complexes in these cells in vivo , we followed the GFP-Sox32 lifetime in the presence and absence of mCherry-Oct4 . We found that the GFP-Sox32 lifetime was significantly reduced from 2 . 54 ± 0 . 01 ns in the absence of mCherry-Oct4 to 2 . 21 ± 0 . 04 ns in presence of mCherry-Oct4 , with a binding percentage of 18% ( p<0 . 0001; Figure 5a-c ) . Figure 5—figure supplement 2 and Figure 5—source data 1 show that Nanog also interacts with Sox32 in the dorsal endoderm at 50% epiboly . The co-existence of these Oct4–Sox32 and Nanog–Sox32 complexes was further confirmed in ventrolateral endodermal cells at 60% epiboly ( 7 hpf ) using the Tg ( sox17:GFP-UTRN ) line , which reports sox17 promoter activity by driving actin-GFP ( Woo et al . , 2012 ) ( Figure 5—figure supplement 2 and Figure 5—figure supplement 1 ) . Together , these results suggest that Sox32 prevents the formation of Pou5f3–Nanog complexes at the 50% epiboly stage in dorsal aspects and restricts them to the ventrolateral mesendoderm . 10 . 7554/eLife . 11475 . 021Figure 5 . Sox32 competes with Nanog for Oct4 binding at dorsal endoderm of gastrula embryos . ( a ) The schematic shows the main germ layers of the embryo at 50% epiboly ( 5 . 7 hpf ) with ectoderm in blue and mesendoderm in green . The dorsal-restricted endoderm precursors are shown in yellow . ( b , c ) Lifetime values ( b ) and FLIM images ( c ) of GFP-Sox32 and GFP-Oct4 alone and in the presence of mCherry-Oct4 and mCherry-Nanog , respectively , in the nuclei of individual cells of dorsal endoderm precursors . Scale bar: 5 µm values represent the median and quartile ranges of data from three to five independent experiments ( n = 20–30 cell nuclei from 10 embryos; ***p<0 . 0001 ) . See also Figure 5—figure supplement 2 , Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 02110 . 7554/eLife . 11475 . 022Figure 5—source data 1 . FCCS parameters of GFP-Sox32 and mCherry-Nanog in endoderm of gastrula embryos ( 50% epiboly; 5 . 7 hpf ) . Diffusion parameters were derived from analysis of FCCS data with the ACFs and CCF fit by two-component anomalous diffusion model . D1 , D2: Diffusion coefficients of the fast and slow diffusion components , respectively . α1 , α2: anomalous parameters of the fast and slow diffusion components , respectively . Kd: dissociation constant at equilibrium; values were obtained from the slopes of the fitted linear line when plotting the concentration of GFP-Sox32 ( CS ) concentration of mCherry-Nanog ( CN ) versus the concentration of the proteins association ( CSN ) . If the proteins are associated , there will be a linear line; in cases where no association exists , there is no linear relationship . Association: fraction of proteins diffusing together in the same complex . Details of the FCCS analysis are explained in the Materials and methods . D-E: dorsal endoderm; V-E: ventral endoderm; Values represent mean ± SEM from three to five independent experiments with n > 15 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 02210 . 7554/eLife . 11475 . 023Figure 5—figure supplement 1 . Sox32 binds Oct4 in ventrolateral endoderm . Lifetime values of GFP-Sox32 lifetime alone and co-expressed with mCherry-Oct4 in the nuclei of individual cells from the row next to the margin in ventrolateral mesendoderm at 50% epiboly ( 5 . 7 hpf ) before involution starts . Those showing a decrease in the lifetime are endoderm precursors and those with similar lifetime are mesoderm precursors . ( b ) GFP-Sox32 and mCherry-Oct4 co-expressed in Tg ( sox17:GFP-UTRN ) embryos at 60% epiboly ( 7 hpf ) . Scale bar: 20 μm . ( c ) Lifetime values and FLIM images of GFP-Sox32 lifetime alone and co-expressed with mCherry-Oct4 in the nuclei of individual cells in ventrolateral endoderm at 60% epiboly ( 7 hpf ) . The percentage of binding is indicated at the top right corner of the FLIM image . Values represent the median and quartile ranges of data from three independent experiments ( n = 20–30 cell nuclei from 10 embryos; ***p<0 . 0001 ) . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 02310 . 7554/eLife . 11475 . 024Figure 5—figure supplement 2 . Nanog and Sox32 interact in endoderm . ( a , b ) ACFs and CCFs of the intensity traces of GFP-Sox32 co-expressed with mCherry-Nanog in ( a ) the dorsal endoderm precursors at 50% epiboly ( 5 . 7 hpf ) and ( b ) in ventrolateral endoderm cells in Tg ( sox17:GFP-UTRN ) embryos at 60% epiboly ( 7 hpf ) fit by two-component anomalous diffusion model . ( c , d ) Raw data of residuals from fit curves shown in a , b , respectively . ( e ) Kd plots for GFP-Sox32 and mCherry-Nanog associations . Kd values were obtained from the slopes of the fitted linear line . If the proteins are associated , there will be a linear line when plotting the concentration of GFP-Sox32 ( CS ) concentration of mCherry-Nanog ( CN ) versus the concentration of the protein association ( CSN ) , since Kd represents the constant of the association at equilibrium . There is no linear relationship in cases where there is no association . Association: fraction of proteins diffusing together in the same complex . ( f ) Lifetime values of GFP-Sox32 lifetime alone and co-expressed with mCherry-Nanog in the nuclei of individual cells of dorsal and ventrolateral endoderm at the stages previously described . Values represent the median and quartile ranges of data from three independent experiments ( n = 20–30 cell nuclei from 10 embryos; p<0 . 01; p<0 . 001 ) . See also Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 024 To ascertain whether Oct4 and Nanog still bind in ventrolateral mesoderm and endoderm lineages at 60% epiboly ( 7 hpf; ( Figure 6a–c ) , we measured the GFP-Oct4 lifetime in the absence and presence of mCherry-Nanog . We observed a significant reduction in GFP-Oct4 lifetime from 2 . 31 ± 0 . 01 ns to 2 . 01 ± 0 . 02 ns and 1 . 98 ± 0 . 02 ns in the ventrolateral mesoderm and endoderm , respectively ( p<0 . 0001; ( Figure 6a–c ) , suggesting that Pou5f3–Nanog complexes may be involved in ventrolateral patterning at later stages . 10 . 7554/eLife . 11475 . 025Figure 6 . Sox32 modulates Oct4 and Nanog complexes in ventrolateral endoderm of gastrula embryos . ( a ) The schematic shows the Ventrolateral ( VL ) - and dorsal ( D ) -mesendoderm layers at 60% epiboly ( 7 hpf ) , with mesoderm in orange and endoderm in yellow . ( b , c ) Lifetime values ( b ) and FLIM images ( c ) of GFP-Oct4 alone and in the presence of mCherry-Nanog in the nuclei of individual cells within VL-Mesoderm and VL-Endoderm cells of wt and elabr13 mutants . Values of FLIM data represent the median and quartile ranges of data from three to five independent experiments ( n = 20–40 cell nuclei from 10 embryos; p<0 . 0001 ) . Scale bar: 5 µm . ( d ) qRT-PCR analysis of sox32 relative to actin in wt and elabr13 mutant embryos . Values represent mean ± SEM of data from three independent experiments ( p<0 . 01 ) . ( e ) Graphs show percentage of binding of GFP-Oct4 and mCherry-Nanog in VL-Mesoderm and VL-Endoderm of wt and elabr13 mutant embryos . Values represent the median and quartile ranges from data of three to five independent experiments ( n = 20–40 cell nuclei from 7 to 10 embryos; p<0 . 01 ) . Values represent mean ± SEM of data from three independent experiments ( p<0 . 01 ) . n . s . over bars indicates non-significant differences . ( f ) qRT-PCR analysis relative to actin reveals different transcription levels of bmp2b , bmp4 , bmp2a and her5 in elabr13 mutants at 60% epiboly ( 7 hpf ) . Values represent mean ± SEM of data from three independent experiments ( *p<0 . 01 ) . ( g ) bmp4 expression ( top view , dorsal is to the right-hand side ) is ventrally reduced in elabr13 mutants compared with wt embryos . her5 expression ( dorsal view ) is dorsally upregulated in elabr13 mutants related to wt embryos . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 02510 . 7554/eLife . 11475 . 026Figure 6—figure supplement 1 . Ela/Aplnr pathway refines ventrolateral Bmp signaling . Expression of the ventrolateral mesoderm markers , tbx6 and eve1 , was slightly down-regulated in elabr13 relative to wt embryos at 50% epiboly . Chd expression in the dorsal margin and gsc in the prospective shield does not change in elabr13 ascompared with wt embryos at 50% epiboly . Ectoderm markers , gata2 ( non-neural ectoderm ) and otx2 ( forebrain-midbrain ) , remained similar in elabr13 relative to wt embryos at 75% epiboly . DOI: http://dx . doi . org/10 . 7554/eLife . 11475 . 026 Elabela mutant ( elabr13 ) embryos have a significantly lower number of endodermal progenitors that express sox17; yet , the total levels of sox17 are still higher in elabr13 embryos than in control siblings ( Chng et al . , 2013 ) . Likewise , sox32 levels are higher in elabr13 embryos than in control siblings ( p<0 . 01; Figure 6d ) . To gain insight into this conundrum , we next followed Oct4–Nanog complexes in ventrolateral mesodermal and endodermal cells of Tg ( sox17:GFP-UTRN; elabr13 ) embryos . We observed a significant decrease in the GFP-Oct4 lifetime , suggesting that Pou5f3 and Nanog also interact in the ventrolateral mesoderm and endoderm of elabr13 embryos ( Figure 6b ) . Interestingly , in contrast to mesodermal cells , the percentage of Oct4–Nanog complexes in endodermal cells was significantly reduced from 21% to 15% ( p<0 . 01; Figure 6e ) . These results support the idea that Sox32 competes with Nanog for Pou5f3 binding to regulate cell fate . Next , we tested whether the decrease in the percentage of Oct4–Nanog complexes in ventrolateral endoderm of elabr13 embryos would affect BMP signaling . We found that bmp4 expression was more reduced and restricted to the ventral area of elabr13 embryos . bmp2b and bmp2a expression was also lower in elabr13 embryos as compared to wt embryos ( Figure 6f ) . The expression of ventrolateral mesoderm bmp targets , such as tbx6 and eve1 , were slightly reduced in elabr13 as compared with wt embryos , whereas her5 , which is expressed in a dorsal subpopulation of endodermal precursors ( Müller et al . , 1996b ) , was significantly increased ( Figure 6f–g ) . Dorsal mesoderm markers , such as gsc and chd , were not significantly upregulated in elabr13 embryos relative to wt embryos . Ventral and dorsal ectoderm , marked by gata2 and otx2 , respectively , were not affected in elabr13 mutants ( Figure 6—figure supplement 1 ) . These results suggest that elevated levels of Sox32 , as observed in Elabela mutant embryos , may decrease the proportion of Pou5f3–Nanog complexes in ventrolateral endodermal cells , which , in turn , may lower BMP signaling in the mesendoderm , but not ectoderm , lineage . Protein dynamics critically regulate tissue differentiation and morphogenesis . Although the dynamics of receptor–ligand interactions ( Ries et al . , 2009; Shi et al . , 2009 ) and the activity gradients of morphogens ( Dubrulle et al . , 2015; Müller et al . , 2012; Yu et al . , 2009 ) have been studied in zebrafish and Xenopus embryos , very little is known about TF dynamics and the interactions that drive early cell commitment in vivo . In this work , we provide quantitative data for Oct4—and , by extension , Pou5f3—DNA binding , and the spatiotemporal complexes that form to achieve proper germ layer formation during gastrulation . Our FCS experiments show that the rescue of maternal spg function depends on Oct4 DNA-binding . Others have proposed that TFs output can be influenced by the dynamic changes following cellular stress ( Cai et al . , 2008; Ashall et al . , 2009; Purvis et al . , 2012: Petrenko et al . , 2013 ) . In particular , chromatin modifications can influence long- and short-lived TF DNA binding in early mouse embryos ( White et al , 2016 ) . Hence , it is possible that the cellular stress inherent to our rescue assay may , in part , account for the decrease in the Oct4 DNA-bound fraction in our non-rescued embryos . Because we are able to document differences in the distribution of the Oct4 DNA-bound fraction between distinct cell lineages , with higher fractions measured in mesendoderm versus ectoderm , our results cannot be explained by a non-specific stochastic rescue . DNA accessibility is thought to impact the degree to which Oct4 and Sox2 are able to find their target binding sites and interact ( Lam et al . , 2012 ) . Our results , showing a lineage dependency on such an Oct4–DNA interaction , support the view that a switch in the partners of Pou5f3 must play a role in cell fate determination . Using FLIM , we provide the first evidence for complex formation between Oct4 and other TFs in developing embryos . Nanog , which activates zygotic transcription with Pou5f3 and SoxB1 ( Lee et al . , 2013; Leichsenring et al . , 2013 ) , formed a significantly higher percentage of complexes with Oct4 in the mesendoderm as compared with the ectoderm at the blastula stage . Given that Sox proteins in mesendoderm do not co-regulatePou5f3 targets ( Onichtchouk et al . , 2010 ) , we speculate that Pou5f3 and Nanog may act together to control zygotic genes in mesendoderm and thereby limit lineage commitment in pluripotent cells . At the gastrula stage , our results show that Nanog cooperates with Pou5f3 in ventrolateral mesendoderm to promote ventral fate by acting upstream of BMP signaling . BMP signaling is initially uniform at the blastula stage and is shaped by inhibitors from the dorsal side at the beginning of gastrulation ( Blader et al . , 1997; Kishimoto et al . , 1997; Reversade and De Robertis , 2005 ) . The presence of Oct4–Nanog complexes at the ventrolateral but not dorsal mesendoderm suggests that , during the transition from blastula to gastrula , dorsal factors may compete with Nanog to partner with Pou5f3 , which in turn , may restrict the expression of BMP ligands to the ventral aspects of the embryo . In dorsal endoderm precursors , Oct4–Nanog complexes are detectable following the knock down of sox32 , indicating that Sox32 , which interacts in vivo with Oct4 in these cells , competes with Nanog for Oct4 binding . We also found that Nanog complexes with Sox32 in the same group of cells , suggesting that Sox32 prevents the formation of Pou5f3–Nanog complexes in dorsal endoderm precursors by binding either with Pou5f3 or Nanog . The presence of the Sox32–Nanog complex suggests that Nanog , in addition to its role with Pou5f3 upstream of BMP signaling and in regulating ventrolateral endoderm through the Nodal pathway ( Xu et al . , 2012 ) , may have other roles in patterning the DV axis in zebrafish . We also provide evidence that Oct4–Nanog , Sox32–Oct4 and Sox32–Nanog complexes coexist in ventrolateral endodermal cells during gastrulation . The Elabela / Aplnr pathway is essential for proper endoderm differentiation , as it regulates the proliferation and migration of the endoderm precursors ( www . elabela . com; Chng et al . , 2013; Pauli et al . , 2014 ) . Interestingly , when the Ela pathway is inhibited , the levels of sox32 increase , accompanied by a decrease in the percentage of Oct4–Nanog complexes in endodermal cells . Thus , these results support the idea that Sox32 regulates Pou5f3–Nanog complexes by protein competition and , thus , modulates endoderm formation in a spatiotemporal manner along the DV axis . The modification of these Pou5f3–Nanog complexes in ventrolateral endodermal cells of Ela-null embryos alters the expression of BMP targets , such as her5 , eve1 , tbx6 and her5 , in particular , is known to control the anteroposterior migration of endodermal progenitors ( Tiso et al . , 2002 ) , a process that is acutely affected in Ela-null embryos . We did not detect significant changes in the expression of the dorsal markers , gsc and chd , suggesting that the observed decrease in BMP signaling is not sufficient to overtly dorsalize the embryo . Thus , our data show that the Elabela / Aplnr pathway , by controlling sox32 levels , quantitatively modulates Oct4-Nanog complex formation to refine BMP signaling during DV patterning of the endoderm . Overall , our in vivo measurements of TF complex formation and our understanding of the protein competition among Pou5f3 , Nanog and Sox32 provide a quantitative overview of how key TFs control the upstream transcription of important morphogens , including the BMP pathway , that initiate histotopic differentiation along the three embryonic axis . Adult zebrafish of the wt ( AB ) strain were kept and bred under standard conditions at 28 . 5°C ( Westerfield , 1993 ) . The spgm793-/+ and Tg ( sox17:GFP-UTRN ) strains were obtained from the laboratory of Drs . Wolfgang Driever ( University of Freiburg , Germany ) and Didier Stainier ( Max Planck Institute , Germany ) , respectively . To generate embryos that were both maternal and zygotic mutants for pou5f3 , mutant adult carriers ( MZspg ) were generated using zebrafish pou5f3 mRNA-mediated rescue of spgm793-/- embryos , as described previously ( Reim and Brand , 2006 ) . elabr13 mutants have been previously described ( Chng et al . , 2013 ) . Embryos were collected by natural spawning and staged as described elsewhere ( Kimmel et al . , 1995; Westerfield , 1993 ) . Zebrafish nanog and pou5f3 were cloned into pCS2+ with the following primers: nanog forward: 5’-GTTTATCTAACGGCGAAATGGCG-3’ , nanog reverse: 5’GCAACCCATGACATCACTGCCT-3’ , pou5f3 forward: 5’-ATGACGGAGAGAGCGCAGA-3’ , and pou5f3 reverse: 5’-TTAGCTGGTGAGATGACCCAC-3’ . To generate a zebrafish nanog construct that would be immune to MO translation inhibition ( nanog ) , the following forward primer was used: 5’-GGCACCATGGCAGATTGGAAAATGCCGGTG-3’ . GFP , GFP-mCherry , GFP-Oct4 , GFP-Oct4ΔHD and mCherry-Oct4 were synthetized as previously described ( Lam et al . , 2012; Plachta et al . , 2011 ) . To generate GFP-Nanog and mCherry-Nanog fusion proteins , cDNA was amplified using sequence-specific primers and cloned directionally into BamHI and NotI sites of the pXJ40:GFP and pXJ40:mCherry expression vectors , respectively . Sox32 and Vox were provided by Dr . Frederic M . Rosa ( Institute National de la Santé et de la Recherche Médicale , France ) , amplified with specific primers , and cloned into BamHI and Xhol sites of pXJ40:GFP . Capped mRNAs were synthesized with the SP6 or T7 mMessage mMachine Kit ( Ambion Inc . , Thermo Fisher Scientific , Austin , TX ) . The antisense morpholino oligonucleotides ( MO ) were manufactured by Gene Tools: nanog MO: 5’- CTGGCATCTTCCAGTCCGCCATTTC-3’ and nanog 5-mismatch MO: 5’- CTGcCATgTTgCAcTCCcCCATTTC -3’ . pou5f3 and sox32 MOs have been previously described ( Burgess et al . , 2002; Sakaguchi et al . , 2001 ) . MO and mRNAs were injected at the one-cell stage in doses as indicated in the Results . For rescue experiments , mRNAs devoid of the MO binding site were co-injected with the MO . Dextran red ( 200 pl ) was injected into two of the four central cells of 16-cell stage embryos . Chorionated embryos at the 32-cell stage were treated with 0 . 3 M LiCl solution for 8 min after mRNA injection at one-cell stage ( Shao et al . , 2012 ) . Total RNA was extracted from a group of 10 embryos at the designated stage using Trizol reagent ( Sigma-Aldrich , St Louis , MO ) . cDNAs were synthesized from 1 µg of total RNA using random hexamers ( Promega , Madison , WI ) and Superscript III reverse transcriptase ( Invitrogen , Life Technologies , Carlsbad , CA ) . RT-PCR was performed for 20 cycles and zebrafish actin was used as a loading control . qRT-PCR , normalized to actin expression levels , was performed with SYBR Green Master Mix ( Applied Biosystems , Foster City , CA ) . The reactions were carried out in triplicate for each experiment , and data are expressed as the mean ± SEM . Significant differences were considered to be those with a p value of <0 . 05 and *<0 . 01 . All qRT-PCR primers are listed in the Supplementary file 1 . The following clones were used to prepare antisense probes for in situ hybridization: gata2 ( Detrich et al . , 1995 ) , dlx3 ( Akimenko et al . , 1994 ) , otx2 ( Mori et al . , 1994 ) , myoD ( Weinberg et al . , 1996 ) , ntl ( Schulte-Merker et al . , 1994 ) , sox17 ( Alexander and Stainier , 1999 ) , sqt ( Feldman et al . , 1998 ) , her1 ( Müller et al . , 1996a ) , pou5f3 ( Takeda et al . , 1994 ) , chd ( Miller-Bertoglio et al . , 1997 ) , gsc ( Thisse et al . , 1994 ) , bmp2b ( Martinez-Barbera et al . , 1997 ) , bmp4 ( Martinez-Barbera et al . , 1997 ) , vox ( Melby et al . , 2000 ) , vent ( Melby et al . , 2000 ) , her5 ( Tiso et al . , 2002 ) and tbx6 ( Hug et al . , 1997 ) . The nanog probe was generated by digestion of the nanog open reading frame in pCS2+ with ClaI and transcription using T7 polymerase . WISH was carried out as previously described ( Tian et al . , 2010 ) ( refer to http://www . reversade . com-a . googlepages . com/protocols for detailed protocols ) . Images were viewed using the Zeiss Axioplan microscope and captured with the Zeiss AxioCam HRc camera ( Zeiss , Oberkochen , Germany ) . Embryos were mounted at specified stages and orientations in 35-mm glass-bottomed petri dishes using 0 . 8% low-melting agarose and covered with egg water . The fluorescence images were obtained using an Olympus FV-1000 confocal microscope ( Olympus , Tokyo , Japan ) . The excitation light source was a 488- and 559-nm cw laser with a dichroic mirror of 488/559 and respective GFP and mRFP emission filters along with DIC channel images . FCS experiments were performed on a PicoHarp 300 TCSPC module ( PicoQuant , Berlin , Germany ) attached to an Olympus FV-1000 confocal microscope ( Olympus ) with a 60 × 1 . 2 W objective . The excitation light source was a 488-nm cw Ar laser with a 488/559-nm dichroic mirror and a 520/35-nm emission filter . The fluctuating photons in the confocal volume were detected using a SPAD detector . The pinhole was set to 80 µm , corresponding to 0 . 2 µm back-projected into the focal plane , and the data were acquired using the SymPhoTime 200 software ( PicoQuant ) . Calibration and measurement of the confocal volume is crucial to extract concentrations and diffusion coefficients . This was performed on a day-to-day basis before measurements were taken under identical settings using a solution of 1 nM Atto 488 with a known diffusion coefficient of 400 µm2s–1 at room temperature ( Wachsmuth et al . , 2015 ) . At three measurement points , intensity time traces were recorded for 30 s in the nuclei of embryos expressing either free GFP or Oct4-GFP . A range of 10 to 18 µW laser power was used to minimize photobleaching . The FCCS acquisition was performed using a setup similar to that used for the FCS measurement , with the exception that an additional laser line 559 cw was used and the mCherry channel was detected using the 615/45 nm filter along with a 560 nm Dichroic mirror to split the green and red channel emission . GFP alone and mCherry alone samples were measured before measuring the experimental samples to correct for cross-talk . The confocal volume for the red channel was measured using Rhodamine 6G . FCS data analysis followed the previously established workflow ( Wachsmuth et al . , 2015 ) : After calculation of the ACFs and after correction for slow fluctuations , such as photobleaching , the data were fit with a two-component anomalous diffusion model . This very general model converged to effective one-component anomalous diffusion for a real one-component system , such as free GFP . Using the radius and volume of the focus , the diffusion coefficients and concentrations were calculated as previously described ( Wachsmuth et al . , 2015 ) . To ensure that the confocal volume was not affected by aberrations induced by the refractive index mismatch among water , the medium , and the interior of the embryo , we determined the ratio of the mean intensity extracted directly from the time traces and the mean number of molecules in the focus extracted from the fit , referred to as molecular brightness . Whereas the first is very robust , the second is very sensitive to aberrations and decreases in response ( Yu et al . , 2009 ) . Virtually , all FCS measurements revealed a molecular brightness inside a window of ± 10% around the mean value of GFP measured in the most peripheral nuclei; this allowed us to ensure that the resulting concentrations and diffusion coefficients were not biased due to aberrations ( Figure 1—figure supplement 2 ) . The free diffusion coefficient ( D1 ) was determined by global fitting: the ACFs were fit with the two-component anomalous diffusion model and the fast component average was taken to fix the free diffusion time value . With the free diffusion time determined , the rest of the parameters were recalculated . Using GFP-Oct4 as an example , global fitting provides a tauD1 of 3030 ± 250 μs; averaging individual fits yields a similar tauD1 of 3000 ± 390 μs ( Figure 1—figure supplement 2 ) . Kd values and percentage of association were calculated as previously described ( Shi et al . , 2009 and Sudhaharan et al . , 2009 ) . Time domain FLIM experiments were performed on a Time Correlated Single Photon Counting ( TCSPC ) system ( PicoQuant ) attached to an Olympus FV-1000 confocal microscope ( Olympus ) with a 60 × 1 . 2 W objective . The excitation light source was a 485-nm pulsed diode laser controlled by a Sepia II ( PicoQuant ) driver with a dichroic mirror of 488/559 and a 520/30 emission filter . Individual photon arrivals were detected using a SPAD detector , and events were recorded by a PicoHarp 300 TCSPC module . Lifetime analysis was carried out using SymPhoTime 200 software . Mono- and bi-exponential fittings were applied . The percentage of binding was calculated from the amplitudes derived from the bi-exponential fitting , as previously shown ( Orthaus et al . , 2009 ) . Statistical analyses and graphs were generated using GraphPad Prism , version 6 . 0 . Unpaired , two-tailed Student’s t-tests with Welch correction were performed if data passed the normality assumptions; if data did not pass the normality test , it was analysed by the Mann–Whitney method . Median and interquartile ranges are graphed as box and whiskers of non-normal data , and bar graphs show the mean and standard error of mean ( SEM ) for normal data .	As an animal embryo develops , cells divide and establish three distinct layers called the ectoderm , mesoderm and endoderm . Proteins called transcription factors control this process by regulating the activity of particular genes . Two or more transcription factors may interact to modulate each other’s activity . Zebrafish embryos provide an ideal model system for monitoring how these embryonic layers form and the interactions between transcription factors in real-time because they are transparent and develop outside their parents . Pou5f3 and Nanog are two key transcription factors involved in this process in zebrafish . However , it is not clear how Pou5f3 and Nanog instruct cells to become ectoderm , mesoderm or endoderm . Perez Camps et al . used imaging techniques to study Pou5f3 and Nanog . The experiments show that Pou5f3 and Nanog bind together to form complexes that instruct cells to form the temporary layer that later gives rise to both the mesoderm and endoderm . The cells in which there are less Pou5f3 and Nanog complexes form the ectoderm layer . To develop the body shape of adult zebrafish , the embryos need to give individual cells information about their location in the body . For example , a signal protein called bone morphogenetic protein ( BMP ) accumulates on the side of the embryo that will become the underside of the fish . Perez Camps et al . show that once the endoderm , mesoderm and ectoderm have formed , Pou5f3–Nanog complexes regulate BMP signalling to specify the underside of the fish . Meanwhile , in the endoderm on the opposite side , another transcription factor called Sox32 binds to individual Pou5f3 and Nanog proteins . This prevents Pou5f3 and Nanog from forming complexes and determines which side of the embryo will make the topside of the fish . A future challenge is to explore other transcription factors that may prevent Pou5f1 and Nanog from binding in the mesoderm and ectoderm of the topside of the fish .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "developmental", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Quantitative imaging reveals real-time Pou5f3–Nanog complexes driving dorsoventral mesendoderm patterning in zebrafish
A fundamental question of biology is what determines organ size . Despite demonstrations that factors within organs determine their sizes , intrinsic size control mechanisms remain elusive . Here we show that Drosophila wing size is regulated by JNK signaling during development . JNK is active in a stripe along the center of developing wings , and modulating JNK signaling within this stripe changes organ size . This JNK stripe influences proliferation in a non-canonical , Jun-independent manner by inhibiting the Hippo pathway . Localized JNK activity is established by Hedgehog signaling , where Ci elevates dTRAF1 expression . As the dTRAF1 homolog , TRAF4 , is amplified in numerous cancers , these findings provide a new mechanism for how the Hedgehog pathway could contribute to tumorigenesis , and , more importantly , provides a new strategy for cancer therapies . Finally , modulation of JNK signaling centers in developing antennae and legs changes their sizes , suggesting a more generalizable role for JNK signaling in developmental organ size control . Within a species , organ size is remarkably reproducible . While extrinsic factors like hormones are required for growth , classic transplantation experiments indicate that intrinsic factors within organs determine size ( Bryant and Simpson , 1984 ) . For example , embryonic limb buds transplanted from a large species of salamander onto a small species grow to the size characteristic of the donor ( Twitty and Schwind , 1931 ) . Similar findings have been made in quail and chick limbs ( Iten and Murphy , 1980; Wolpert , 1978 ) , rat hearts and kidneys ( Dittmer et al . , 1974; Silber , 1976 ) , and mouse thymuses ( Metcalf , 1963 ) . Consistently , developing Drosophila wings transplanted into adult abdomens grow to the proper size , indicating that the information determining size is located within the developing organ ( García-Bellido , 1965 ) . Indeed , the Drosophila wing is a classic model system for studying organ size , as its size is highly replicable ( García-Bellido and Merriam , 1971; García-Bellido , 1965 ) , and all adult precursor cells are located within the pouch region of the developing larval imaginal disc ( García-Bellido et al . , 1973 ) ( Figure 1A , grey ) . Despite extensive work , the molecular mechanisms underlying intrinsic organ size control remain unclear ( Vogel , 2013 ) . While morphogens direct both patterning and growth of developing organs ( Tabata and Takei , 2004 ) , a link between patterning molecules and growth control pathways has not been established ( Schwank et al . , 2011 ) . 10 . 7554/eLife . 11491 . 003Figure 1 . Localized JNK activity exists in the developing wing . ( A ) Schematic of wing precursor cells ( grey ) in the developing disc ( A , anterior; P , posterior ) . ( B-F ) Antibody staining against active , phosphorylated JNK ( pJNK , green; DAPI , blue ) labels a stripe in wildtype ( B-C ) but not JNKK mutant ( D-E , hepr75/Y ) third instar discs . Boxed region in ( B ) and ( D ) is magnified in ( C ) and ( E ) , respectively . Weak pJNK signal is also detected along the dorsal/ventral boundary . pJNK stripe staining is lost in JNKK mutant clones ( F , hepr75 , clone is negatively marked in F’ ) . ( G-I ) pJNK localizes to the same cells in which ptc is expressed ( G , ptc>RFP , red ) along the A/P boundary , and is lost following JNK phosphatase expression ( H , ptc>puc , RFP , red ) or RNAi-mediated knockdown of bsk within the ptc domain ( I , ptc>bskRNAi , RFP , red ) . Bar: 50 um ( B-F , H-I ) and 25 um ( G ) . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 00310 . 7554/eLife . 11491 . 004Figure 1—figure supplement 1 . pJNK recognizes endogenous JNK activity in developing wing discs . Related to Figure 1 . ( A-C ) Wildtype Canton-S wing discs stained for DAPI ( blue ) and pJNK ( green ) during ( A ) early third instar ( L3 ) , ( B ) mid-third instar , and ( C ) late third instar . ( D ) Wing disc stained for DAPI ( blue ) , pJNK ( green ) , and puc-lacZ ( red ) . Boxes indicate areas enlarged in E and F . ( E ) Notum cells are positive for pJNK and puc-lacZ . ( F ) Blade cells show a stripe of pJNK staining but no detectable puc-lacZ . ( G-H ) A second , independently generated antibody against pJNK from Promega shows a similar pattern in third instar discs . ( G ) Whole wing disc and ( H ) wing blade . ( I ) Inhibition of JNK signaling in the dorsal compartment reduces pJNK staining ( green ) ( ap>puc ) . ( J ) ptc-Gal4 expresses in a stripe in early L3 stage . ( K ) Inhibition of JNK in all wing blade cells ( rn>bskRNAi#1 , RFP ) or ( L ) in ptc cells ( ptc>bskRNAi#2 , RFP ) eliminates pJNK ( green ) signal . ( M ) Western blot analysis of larval extracts from Canton-S ( Lane 1 ) and hepr75/Y ( Lane 2 ) animals . pJNK is predicted to be ~43kD . Loading control ( bottom ) is alpha-tubulin . Bar: 50 um . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 004 The Jun N-terminal Kinase ( JNK ) pathway promotes proliferation during regeneration and tumor growth ( Bosch et al . , 2005; Igaki et al . , 2006; Ryoo et al . , 2004; Srivastava et al . , 2007; Wu et al . , 2010 ) . In fact , JNK-induced proliferation is often non-autonomous ( Enomoto and Igaki , 2012; Pastor-Pareja et al . , 2008; Ryoo et al . , 2004; Sun and Irvine , 2011; Wu et al . , 2010 ) . Basket ( Bsk ) is the singular Drosophila JNK and is activated by phosphorylation by the JNKK Hemipterous ( Hep ) ( Glise et al . , 1995; Stronach , 2005 ) . Canonical JNK signaling acts through the transcription co-factor Jun , which regulates migration and apoptosis ( Stronach , 2005 ) . Although the role of JNK in activating Yorkie signaling and growth during regeneration and tumorigenesis is clear ( Enomoto and Igaki , 2012; Sun and Irvine , 2011; Sun and Irvine 2013 ) , it is not known to regulate proliferation and growth during developmental size control . Here we show that localized JNK activity in the developing wing is established by Hedgehog ( Hh ) signaling and controls wing size through a non-canonical , Jun-independent signaling mechanism that inhibits the Hippo pathway . Two independently generated antibodies that recognize the phosphorylated , active form of JNK ( pJNK ) specifically label a stripe in the pouch of developing wildtype third instar wing discs ( Figure 1B–C and Figure 1—figure supplement 1G–H ) . Importantly , localized pJNK staining is not detected in hemizygous JNKK mutant discs ( Figure 1D–E; hepr75/Y ) , in clones of JNKK mutant cells within the stripe ( Figure 1F; hepr75 , FRT10/Ubi-GFP , FRT10;; MKRS , hs-FLP/+ ) , following over-expression of the JNK phosphatase puckered ( puc ) ( Figure 1—figure supplement 1I; ap-Gal4 , UAS-puc ) , or following RNAi-mediated knockdown of bsk using two independent , functionally validated RNAi lines ( Figure 1—figure supplement 1K–L; rn-Gal4 , UAS-bskRNAi#1 or ptc-Gal4 , UAS-bskRNAi#2; see Experimental Genotypes for full genotypes and conditions ) ( Glise et al . , 1995; MacDonald et al . , 2013; Martín-Blanco et al . , 1998; Pérez-Garijo et al . , 2013; Weber et al . , 2000; Xu and Rubin , 1993 ) . The stripe of localized pJNK staining appeared to be adjacent to the anterior-posterior ( A/P ) compartment boundary , a location known to play a key role in organizing wing growth , and a site of active Hedgehog ( Hh ) signaling ( Basler and Struhl , 1994; Tabata and Kornberg , 1994; Zecca et al . , 1995 ) . Indeed , pJNK co-localizes with the Hh target gene patched ( ptc ) ( Figure 1G; ptc-Gal4 , UAS-RFP ) . Expression of the JNK phosphatase puc in these cells specifically abrogated pJNK staining ( Figure 1H; ptc-Gal4 , UAS-puc ) , as did RNAi-mediated knockdown of bsk ( Figure 1I and Figure 1—figure supplement 1L; ptc-Gal4 , UAS-bskRNA#i1or2 ) . Together , these data indicate that the detected pJNK signal reflects endogenous JNK signaling activity in the ptc domain , a region of great importance to growth control . Indeed , while at earlier developmental stages pJNK staining is detected in all wing pouch cells ( Figure 1—figure supplement 1A ) , the presence of a localized stripe of pJNK correlates with the time when the majority of wing disc growth occurs ( 1000 cells/disc at mid-L3 stage to 50 , 000 cells/disc at 20 hr after pupation , ( Garcia-Bellido , 2009 ) , so we hypothesize that localized pJNK plays a role in regulating growth . Inhibition of JNK signaling in the posterior compartment previously led to the conclusion that JNK does not play a role in wing development ( McEwen and Peifer , 2005 ) . The discovery of an anterior stripe of JNK activity spurred us to re-examine the issue . Since bsk null mutant animals are embryonic lethal , we thus conditionally inhibited JNK signaling in three independent ways in the developing wing disc . JNK inhibition was achieved by RNAi-mediated knockdown of bsk ( bskRNAi#1or2 ) , by expression of JNK phosphatase ( puc ) , or by expression of a dominant negative bsk ( bskDN ) . These lines have been independently validated as JNK inhibitors ( MacDonald et al . , 2013; Martín-Blanco et al . , 1998; Perez-Garijo et al . , 2013; Weber et al . , 2000 ) . Inhibition of JNK in all wing blade cells ( rotund-Gal4 , rn-Gal4 ) or specifically in ptc-expressing cells ( ptc-Gal4 ) resulted in smaller adult wings in all cases , up to 40% reduced in the strongest cases ( Figures 2A–F , 2J–K , and Figure 2—figure supplement 1D ) . Importantly , expression of a control transgene ( UAS-GFP ) did not affect wing size ( Figure 2—figure supplement 1B–C; ptc-Gal4 , UAS-GFP ) . This contribution of JNK signaling to size control is likely an underestimate , as the embryonic lethality of bsk mutations necessitates conditional , hypomorphic analysis . Nevertheless , hypomorphic hepr75/Y animals , while pupal lethal , also have smaller wing discs ( Figure 2—figure supplement 1G ) , as do animals with reduced JNK signaling due to bskDN expression ( Figure 2—figure supplement 1H–I; ap-Gal4 , UAS-bskDN ) . Importantly , total body size is not affected by inhibiting JNK in the wing . Even for the smallest wings generated ( rn-Gal4 , UAS-bskDN ) , total animal body size is not altered ( Figure 2—figure supplement 1A , E ) . 10 . 7554/eLife . 11491 . 005Figure 2 . Modulation of localized JNK signaling changes wing size . Inhibition of JNK in all wing blade cells ( B-E , J ) or within the ptc domain ( F , K ) decreases adult wing size compared to controls ( A , C-E , J , rn> ) or ( F , K , ptc> ) . Note that autonomous reduction between longitudinal veins 3 and 4 accounts for a small portion of the global reduction . Apoptosis inhibition does not rescue the small wing phenotype ( red , G , rn>p35 , bskDN ) . ( H-I , L ) Increased JNK signaling within the ptc domain following eiger expression causes an increase in disc size ( I , ptc>egr , RFP , red; DAPI , blue ) compared to controls ( H , ptc>RFP , red ) . ( L ) This is increase is dependent on bsk ( ptc>egr , bskDN ) but not affected by diap1 or p35 expression ( ptc>egr , diap1 or ptc>egr , p35 ) . Due to high pupal lethality , disc size was analyzed when animals reached the wandering third instar stage . ( M-O ) JNK inhibition does not affect cell size ( N-O , rn>bskDN ) . ( P-Q ) Increased JNK signaling within the ptc domain causes an increase in proliferation ( Q , ptc>egr , RFP , red; EdU , green ) compared to controls ( P , ptc>RFP , red; EdU , green ) . EdU of boxed region in ( P ) and ( Q ) is shown in ( R ) and ( S ) , respectively . ( T ) Quantification of mean EdU signal in wing pouch regions between ptc>RFP and ptc>egr animals . Whiskers are SD . For box plots of area quantifications , whiskers represent maximum and minimum values ( J-L , O ) . -**=p<0 . 05–0 . 0001 . n . s . = not significant . Bar: 50 um . See also Figure 2—figure supplements 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 00510 . 7554/eLife . 11491 . 006Figure 2—figure supplement 1 . JNK inhibition does not affect body size or cell death , but rather cell proliferation . Related to Figure 2 . ( A ) Control rotund-Gal4 ( rn> ) alone female fly ( left ) . Inhibiting JNK in the entire wing ( rn>bskDN ) leads to a female fly with smaller , well-patterned wings ( right ) . Black bars highlight difference in wing size . ( B-C ) Expression of a control transgene ( UAS-GFP ) does not affect wing size ( ptc>GFP ) . ( D ) Quantification of relative wing size for knockdown of bsk with a second RNAi line ( bskRNAi#2 ) . ( E ) Adult body length is not affected by inhibiting JNK by rn-Gal4 . ( F ) Inhibition of JNK with rn-GAL4 delays development . ( G ) hepr75/Y animals have smaller wing discs than controls ( Canton-S or hepr75/+ ) , even when adjusted for delayed developmental time ( 7d AEL ) . ( H-I ) JNK inhibition ( red , dorsal half ) causes a reduction in wing pouch size compared to its matched control ( blue , ventral half ) ( ap>bskDN , red ) . ( J ) JNK inhibition ( dorsal half ) reduces cell proliferation by phosphorylated histone 3 ( PH3 ) staining ( green ) compared to its matched control ( ventral half ) ( ap>bskDN ) . ( K ) Control discs ( ap>RFP , blue ) do not show a difference in PH3 staining between dorsal and ventral halves ( ratio = 1 . 04 ) , while JNK inhibited ones do ( Ratio = 0 . 86 , red ) . ( L ) Control wing pouch ( rn>RFP , red ) stained for cleaved Caspase 3 ( CCP3 , green ) . ( M ) Inhibition of JNK in all pouch cells ( rn>bskDN , RFP , red ) does not induce apoptosis as assayed by CCP3 staining ( green ) . ( N ) Positive control expression of wildtype JNK ( bskAY ) causes apoptosis and CCP3 staining ( green ) . Two-sided student’s t-test: -**p<0 . 05–0 . 001 . Bar: 50 um . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 00610 . 7554/eLife . 11491 . 007Figure 2—figure supplement 2 . Activating JNK signaling increases wing disc size independent of cell death or developmental timing . Related to Figure 2 . ( A-C ) Age-matched wing discs expressing RFP by ptc-GAL4 ( control , A ) or RFP and egr by ptc-GAL4 ( B ) . ( C ) Wing disc area quantification for A-B . ( D-F ) Induction of apoptosis in the ptc domain reduces wing disc size . ( D ) Control ptc>RFP wing . ( E ) Expression of UAS-hid ( ptc>hid ) decreases wing size . ( F ) Quantification of D-E . ( G ) Size increase due to egr expression depends on bsk activity ( ptc>egr , bskDN ) , but is not affected by expression of diap1 ( G , ptc>egr , diap1 ) or p35 ( I , ptc>egr , p35 ) . Quantification of G-I is presented in Figure 2L . Two-sided student’s t-test: -p<0 . 05–0 . 01 . Bar: 100 um . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 00710 . 7554/eLife . 11491 . 008Figure 2—figure supplement 3 . JNK inhibition does not affect Dpp or EGFR signaling . Related to Figure 2 . ( A-C ) Wing discs stained for the EGFR reporter pERK ( green ) . ( A ) Control wing disc ( ap>RFP , red ) . ( B ) Inhibition of EGFR signaling in the dorsal half of the disc ( ap>EGFRRNAi , RFP , red ) decreases dorsal pERK ( green ) staining , while ( C ) inhibition ofJNK signaling ( ap>bskDN , RFP , red ) does not . ( D-F ) Wing discs stained for the Dpp reporter pSMAD ( green ) . ( D ) Control ( ap>RFP , red ) . ( E ) Inhibition of Dpp signaling in the dorsal half of the disc ( ap>dppRNAi , RFP , red ) abolishes dorsal pSMAD ( green ) staining , while ( F ) inhibition of JNK signaling ( ap>bskDN , RFP , red ) does not . ( G ) Quantification of pSMAD fluorescence , as a ratio of dorsal to ventral staining . ap>dppRNAi causes a dramatic decrease in the ratio , while JNK inhibition ( ap>bskDN ) does not produce a statistically significant change ( p=0 . 17 ) . ( H ) pSMAD gradient fluorescence plot by distance along the A-P axis . Ventral ( blue ) is control , while dorsal ( red ) is knockdown of dpp . ( I ) pSMAD gradient fluorescence plot by distance along the A-P axis . Inhibiting JNK signaling ( dorsal , red ) does not affect pSMAD gradient formation ( compare blue to red ) . ( J ) Control rn-Gal4 alone control . ( K ) RNAi-mediated knockdown of dpp causes a reduction in wing veins and a more pronounced effect on AP than PD length . ( L ) Inhibition of JNK does not cause wing vein loss , but does cause a global reduction in size . AFU . : arbitrary fluorescence units . Bar: 50 um . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 00810 . 7554/eLife . 11491 . 009Figure 2—figure supplement 4 . Inhibiting EGFR or Dpp signaling does not affect pJNK establishment . Related to Figure 2 . Inhibition of EGFR ( A ) or Dpp ( B ) by RNAi does not have an effect on pJNK ( green ) . Bar: 50 um . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 009 To test whether elevation of this signal can increase organ size , we expressed eiger ( egr ) , a potent JNK activator ( Igaki et al . , 2002 ) , within the ptc domain ( ptc-Gal4 , UAS-egr ) . Despite induction of cell death as previously reported ( Igaki et al . , 2002 ) and late larval lethality , we observed a dramatic increase in wing disc size without apparent duplications or changes in the shape of the disc ( Figures 2H–I and 2L; ptc-Gal4 , UAS-egr ) . While changes in organ size could be due to changing developmental time , wing discs with elevated JNK signaling were already larger than controls assayed at the same time point ( Figure 2—figure supplement 2A–C; ptc-Gal4 and ptc-Gal4 , UAS-egr ) . Similarly , inhibition of JNK did not shorten developmental time ( Figure 2—figure supplement 1F; rn-Gal4 , UAS-bskDN ) . Thus , changes in organ size by modulating JNK activity do not directly result from altering developmental time . Finally , the observed increase in organ size is not due to induction of apoptosis , as expression of the pro-apoptotic gene hid does not increase organ size ( Figure 2—figure supplement 2D–F ) . In contrast , it causes a decrease in wing size ( Figure 2—figure supplement 2D–F ) . Furthermore , co-expression of diap1 or p35 did not significantly affect the growth effect of egr expression ( p>0 . 05; Figure 2L and Figure 2—figure supplement 2H–I; ptc-Gal4 , UAS-egr , UAS-diap1 and ptc-Gal4 , UAS-egr , UAS-p35 ) , while the effect was dependent on Bsk activity ( p<0 . 05; Figure 2L and Figure 2—figure supplement 2G; ptc-Gal4 , UAS-egr , UAS-bskDN ) . In stark contrast to known developmental morphogens , we did not observe any obvious defects in wing venation pattern following JNK inhibition ( Figure 2A–B ) , suggesting that localized pJNK may control growth in a pattern formation-independent manner . Indeed , even a slight reduction in Dpp signaling results in dramatic wing vein patterning defects ( Figure 2—figure supplement 3K ) . Second , inhibiting Dpp signaling causes a reduction in wing size along the A-P axis , while JNK inhibition causes a global reduction ( Figure 2—figure supplement 3J–L ) . Furthermore , ectopic Dpp expression increases growth in the form of duplicated structures ( Zecca et al . , 1995 ) , while increased JNK signaling results in a global increase in size ( Figure 2H–I ) . Molecularly , we confirm that reducing Dpp signaling abolishes pSMAD staining , while quantitative data shows that inhibiting JNK signaling does not ( Figure 2—figure supplement 3D–I ) . Furthermore , we also find that Dpp is not upstream of pJNK , as reduction in Dpp signaling does not affect pJNK ( Figure 2—figure supplement 4B ) . Together , the molecular data are consistent with the phenotypic results indicating that pJNK and Dpp are separate programs in regulating growth . Consistent with our findings , during the revision of this manuscript , it has been suggested that Dpp does not play a primary role in later larval wing growth control ( Akiyama and Gibson , 2015 ) . Finally , we found that inhibition of JNK does not affect EGFR signaling ( pERK ) and that inhibition of EGFR does not affect the establishment of pJNK ( Figure 2—figure supplement 3A–C and 4A ) . A difference in size could be due to changes in cell size and/or number . We examined wings with reduced size due to JNK inhibition and did not detect changes in cell size or apoptosis ( Figure 2M–O and Figure 2—figure supplement 1L–N; rn-Gal4 , UAS-bskDN ) , suggesting that pJNK controls organ size by regulating cell number . Consistently , the cell death inhibitor p35 was unable to rescue the decreased size following JNK inhibition ( Figure 2G; rn-Gal4 , UAS-p35 , UAS-bskDN ) . Indeed , inhibition of JNK signaling resulted in a decrease in proliferation ( Figure 2—figure supplement 1J–K; ap-Gal4 , UAS-bskDN ) , while elevation of JNK signaling in the ptc domain resulted in an increase in cell proliferation in the enlarged wing disc ( Figure 2P–T; ptc-Gal4 , UAS-egr ) . Importantly , this increased proliferation is not restricted to the ptc domain , consistent with previous reports that JNK can promote proliferation non-autonomously ( Enomoto and Igaki , 2012; Pastor-Pareja et al . , 2008; Ryoo et al . , 2004; Sun and Irvine , 2011; Wu et al . , 2010 ) . To determine the mechanism by which pJNK controls organ size , we first considered canonical JNK signaling through its target Jun ( Ip and Davis , 1998 ) . Interestingly , RNAi-mediated knockdown of jun in ptc cells does not change wing size ( Figure 3A–B and Figure 3—figure supplement 1C–F; ptc-Gal4 , UAS-junRNAi#1or2; Both RNAi lines can effectively inhibit jun activity , Figure 3—figure supplement 1A–B ) , which is consistent with previous analysis of jun mutant clones in the wing disc ( Kockel et al . , 1997 ) . Furthermore , in agreement with this , a reporter of canonical JNK signaling downstream of jun ( puc-lacZ [Ring and Martinez Arias , 1993] ) is not expressed in the pJNK stripe ( Figure 1—figure supplement 1F ) . Finally , knockdown of fos ( kayak , kay ) alone or with junRNAi did not affect wing size ( Figure 3—figure supplement 1G–H; rn-Gal4 , UAS-kayRNiA#1or2 and rn-Gal4 , UAS-junRNAi#1 , UAS-kayRNiA#1or2 ) . Together , these data indicate that canonical JNK signaling through jun does not function in the pJNK stripe to regulate wing size . 10 . 7554/eLife . 11491 . 010Figure 3 . Non-canonical JNK signaling regulates wing size . RNAi-mediated knockdown of Jun within the ptc stripe does not change adult wing size ( A-B , red , ptc>junRNAi compared to blue , ptc> ) . RNAi-mediated knockdown of jub does change global wing size ( C-D , red , ptc>jubRNAi compared to blue , ptc> ) . Expression of yki in all wing cells ( E-F , red , rn>yki , bskDN compared to blue , rn> ) or within the ptc stripe ( G-H , red , ptc>bskDN , yki compared to blue , ptc> ) rescues wing size following JNK inhibition . RNAi-mediated knockdown or overexpression of yki in ptc cells decreases or enlarges wing size , respectively ( I-J , red , ptc>ykiRNAi , blue , ptc> , and K-L , red , ptc>yki , blue , ptc> ) . ( M-N ) Inhibition of JNK signaling does not enhance the phenotype of Yki inhibition alone ( M , red , ptc>bskDN , ykiRNAi; blue , ptc>ykiRNAi ) . ( O-P ) RNAi-mediated knockdown of fj modifies the Yki growth phenotype ( O , red , ptc>yki , fjRNAi; blue , ptc>yki ) . For box plots , whiskers represent maximum and minimum values . *=p<0 . 0001 . See also Figure 3—figure supplements 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 01010 . 7554/eLife . 11491 . 011Figure 3—figure supplement 1 . Jun RNAi line validation and loss of kayak phenotypes . Related to Figure 3 . RNAi-mediated knockdown of Jun in ap domain cells decreases puc expression ( puc-lacZ , green ) ( B ) compared to controls ( A ) . Dotted line indicates puc+ cells that co-localize with ap-Gal4 . Note decreased puc-lacZ staining in this domain following Jun inhibition . However , ( C-D ) inhibition of Jun in all wing cells by RNAi-mediated knockdown does not show a phenotype . ( E-F ) A second Jun RNAi line does not show a phenotype when expressed in ptc-expressing cells . ( G-H ) Inhibition of kayak/fos ( red , rn>kayRNAi ) does not affect wing size , nor does inhibiting jun and kay together ( green , rn>kayRNAi , junRNAi ) . Individually , kayRNAi lines induced a thorax closure defect when driven by ap-Gal4 . For box plots , whiskers represent maximum and minimum values . Bar: 5DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 01110 . 7554/eLife . 11491 . 012Figure 3—figure supplement 2 . JNK interacts with Yki to cause global changes in wing size . Related to Figure 3 . ( A ) Schematic for measuring the ratio of anterior to posterior wing area . ( B ) Local ( ptc-driven ) inhibition of JNK or increased Yki expression affects the anterior and posterior compartments equally . ( C-D ) The effect of inhibiting JNK signaling can be partially suppressed in a lats heterozygous mutant background ( C , red , rn>bskDN; latse26-1/+ ) . ( G-H ) Inhibition of fj alone does not change wing size ( G , red , ptc>fjRNAi , blue , ptc> ) , albeit it slightly changes wing shape , likely due to its effect on polarity . ( I-J ) Over-expression of fj causes a decrease in wing size ( I , red , ptc>fj ) . For box plots , whiskers are maximum and minimum values . Two-sided student’s t-test: -**p<0 . 05–0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 012 In search of such a non-canonical mechanism of JNK-mediated size control , we considered the Hippo pathway . JNK signaling regulates the Hippo pathway to induce autonomous and non-autonomous proliferation during tumorigenesis and regeneration via activation of the transcriptional regulator Yorkie ( Yki ) ( Bakal et al . , 2008; Enomoto and Igaki , 2012; Sun and Irvine , 2011 ) . Recently it has been shown that JNK activates Yki via direct phosphorylation of Jub ( Sun and Irvine , 2013 ) . To test whether this link between JNK and Jub could account for the role of localized pJNK in organ size control during development , we performed RNAi-mediated knockdown of jub in the ptc stripe , and observed adults with smaller wings ( Figure 3C–D; ptc-Gal4 , UAS-jubRNAi#1or2 ) . Indeed , the effect of JNK loss on wing size can be partially suppressed in a heterozygous lats mutant background ( Figure 3—figure supplement 2C–D; rn-Gal4 , UAS-bskDN , latse26-1/+ ) and increasing downstream yki expression in all wing cells ( Figure 3E–F; rn-Gal4 , UAS-yki , UAS-bskDN ) or just within the ptc domain ( Figure 3G–H; ptc-Gal4 , UAS-yki , UAS-bskDN ) can rescue wing size following JNK inhibition . These results suggest that pJNK controls Yki activity autonomously within the ptc stripe , leading to a global change in cell proliferation . This hypothesis predicts that the Yki activity level within the ptc stripe influences overall wing size . Consistently , inhibition of JNK in the ptc stripe translates to homogeneous changes in anterior and posterior wing growth ( Figure 3—figure supplement 2A–B ) . Similarly , overexpression or inhibition of Yki signaling in the ptc stripe also results in a global change in wing size ( Figure 3I–L and Figure 3—figure supplement 2A–B; ptc-Gal4 , UAS-yki; ptc-Gal4 , UAS-ykiRNAi ) . It is important to note that the yki expression line used is wild-type Yki , which is still affected by JNK signaling . For this reason , the epistasis experiment was also performed with activated Yki , which is independent of JNK signaling ( UAS-ykiS111A , S168A , S250A . V5; ( Oh and Irvine , 2009 ) . Expression of this activated Yki in the ptc stripe caused very large tumors and lethality ( data not shown ) . Importantly , inhibiting JNK in this context did not affect the formation of these tumors or the lethality ( data not shown; ptc-Gal4 , UAS-ykiS111A , S168A , S250A . V5 , UAS-bskDN ) . Furthermore , inhibiting both JNK and Yki together does not enhance the phenotype of Yki inhibition alone ( Figure 3M–N and Figure 3—figure supplement 2E–F; ptc-Gal4 , UAS-ykiRNAi , UAS-bskDN and ptc-Gal4 , UAS-ykiRNAi , UAS-puc ) , further supporting the idea that Yki is epistatic to JNK , instead of acting in parallel processes . Mutants of the Yki downstream target four-jointed ( fj ) have small wings with normal patterning , and fj is known to propagate Hippo signaling and affect proliferation non-autonomously ( Ambegaonkar et al . , 2012; Harvey and Tapon , 2007; Strutt et al . , 2004; Villano and Katz , 1995; Willecke et al . , 2008 ) . Although RNAi-mediated knockdown of fj in ptc cells does not cause an obvious change in wing size , it is sufficient to block the Yki-induced effect on increasing wing size ( Figure 3O–P and Figure 3—figure supplement 2G–H; ptc-Gal4 , UAS-yki , UAS-fjRNAiand ptc-Gal4 , UAS-fjRNAi ) . However , overexpression of fj also reduces wing size , which makes it not possible to test for a simple epistatic relationship ( ptc-Gal4 , UAS-fj; Figure 3—figure supplement 2I–J ) . Overall , these data are consistent with the notion that localized pJNK regulates wing size not by Jun-dependent canonical JNK signaling , but rather by Jun-independent non-canonical JNK signaling involving the Hippo pathway . While morphogens direct both patterning and growth of developing organs ( Tabata and Takei , 2004 ) , a link between patterning molecules and growth control pathways has not been established ( Schwank et al . , 2011 ) . pJNK staining is coincident with ptc expression ( Figure 1G ) , suggesting it could be established by Hh signaling . During development , posterior Hh protein travels across the A/P boundary , leading to activation of the transcription factor Cubitus interruptus ( Ci ) in the stripe of anterior cells ( Domínguez et al . , 1996; Schwartz et al . , 1995 ) . To test whether localized activation of JNK is a consequence of Hh signaling through Ci , we performed RNAi-mediated knockdown of ci and found that the pJNK stripe is eliminated ( Figure 4A–B; ptc-Gal4 , UAS-ciRNAi#1or2 ) . Consistently , adult wing size is globally reduced ( Figures 4D and 4G ) . In contrast , we do not observe a change in pJNK stripe staining following RNAi-mediated knockdown of dpp or EGFR ( Figure 2—figure supplement 4A–B ) . Expression of non-processable Ci leads to increased Hh signaling ( Price and Kalderon , 1999 ) . Expression of this active Ci in ptc cells leads to an increase in pJNK signal and larger , well-patterned adult wings ( Figures 4C , E , and 4G; ptc-Gal4 , UAS-CiACT ) . The modest size increase shown for ptc>CiACT is likely due to the fact that higher expression of this transgene ( at 25°C ) leads to such large wings that pupae cannot emerge from their cases . For measuring wing size , this experiment was performed at a lower temperature ( 20°C , see Experimental Genotypes ) so that the animals were still viable . Furthermore , inhibition of JNK in wings expressing active Ci blocks Ci’s effects , and resulting wings are similar in size to JNK inhibition alone ( Figure 4F–G; ptc-Gal4 , UAS-CiACT , UAS-bskDN ) . Together , these data indicate that Hh signaling through Ci is responsible for establishing the pJNK stripe . 10 . 7554/eLife . 11491 . 013Figure 4 . Hh signaling through Ci establishes localized pJNK . RNAi-mediated knockdown of Ci in ptc cells abrogates pJNK ( green ) staining ( A-B , ptc>CiRNAi , RFP compared to ptc>RFP ) and results in smaller adult wings ( D , red , ptc>CiRNAi compared to blue , ptc> ) . Expression of activated Ci in the ptc domain leads to increased pJNK staining ( green ) ( C , ptc>CiACT , RFP ) and a larger wing ( E , red , ptc>CiACT compared to blue , ptc> ) . Inhibition of JNK signaling in these cells blocks the effect of activated Ci ( red , F , ptc>CiACT , bskDN ) . For the box plot ( G ) , whiskers represent maximum and minimum values . -**=p<0 . 001–0 . 0001 . Bar: 50 um . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 013 To determine the mechanism by which Ci activates the JNK pathway , we compared transcriptional profiles of posterior ( red , hh+ ) and ptc domain cells ( green , ptc+ ) isolated by FACS from third instar wing discs ( Figure 5A; Materials and methods ) . Of the total 12 , 676 unique genes represented on the microarray , 50 . 4% ( 6 , 397 ) are expressed in ptc domain cells , posterior cells , or both ( log2 normalized expression ≥6 . 5; Figure 5—figure supplement 1A–D; Supplementary file 1; Materials and methods ) . We thresholded on a false discovery rate <0 . 01 and fold change ≥1 . 5 and found that 5 . 7% ( 363 ) of expressed genes were upregulated in ptc cells and 3 . 8% ( 242 ) were downregulated ( Figure 5—figure supplement 1D; Supplementary file 2; Materials and methods ) . Hh pathway genes known to be differentially expressed are identified ( Figure 5B ) . We next asked whether any JNK pathway genes are differentially expressed and found that dTRAF1 expression is more than five-fold increased in ptc cells ( Figure 5C ) , while other JNK pathway members are not differentially expressed ( Figure 5C; Supplementary file 1; Supplementary file 2 ) . 10 . 7554/eLife . 11491 . 014Figure 5 . Hedgehog signaling establishes pJNK by elevating dTRAF1 expression . ( A ) ptc cells ( green , ptc+ ) and posterior cells ( red , hh+ ) from third instar wing discs were dissociated and sorted by FACS . RNA was isolated and hybridized to microarrays . Differentially expressed genes were identified . ( B ) Hedgehog pathway genes known to be differentially expressed are identified . Genes upregulated in ptc cells ( ptc+ ) compared to posterior ( hh+ ) cells are highlighted in green and downregulated in red . Genes with log2 normalized expression ≥6 . 5 are considered expressed . ( C ) JNK pathway gene dTRAF1 is >5-fold upregulated in ptc cells . ( D-I ) RNAi-mediated knockdown of dTRAF1 eliminates pJNK ( green ) staining ( E , ptc>dTRAFRNAi#1 , RFP , red ) and leads to smaller adult wings ( F-I , rn>dTRAFRNAi#1 or ptc>dTRAFRNAi#1 ) . ( J ) Ci inhibition causes a ~30% decrease in dTRAF1 expression in 3rd instar wing discs , relative to endogenous control Rp49 . Whiskers are SD . For box plots , whiskers are maximum and minimum values ( H-I ) . -**=p<0 . 05–0 . 0001 . Bar: 50 um . See also Figure 5—figure supplement 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 01410 . 7554/eLife . 11491 . 015Figure 5—figure supplement 1 . Transcriptional profiling quality control and additional dTRAF1 validation . Related to Figure 5 . Quality assessment analyses were conducted with post-normalized data . ( A ) Microarrays cluster by condition , indicating that biological effects are driving variability . ( B ) Principle components analysis similarly groups biological replicates . Outliers were not detected in ( A ) or ( B ) . ( C ) Density plots of the log2 normalized expression in ptc domain ( ptc+ ) or posterior ( hh+ ) samples are very similar in shape and range , further suggesting comparable signal quality between the two arrays . Probe sets with a median log2 normalized expression ≥6 . 5 in ptc+ samples , hh+ cells , or both , were considered expressed ( Supplementary file 1; Materials and methods ) . This corresponds to 6854 genic probe sets ( 47 . 3% ) . ( D ) Quantile-quantile plot for the differential expression analysis . Based on a criteria of minimum fold change ≥1 . 5 and false discovery rate ( FDR ) ≤0 . 01 , 624 of 6 , 854 genic probe sets ( 9 . 1% ) are differentially expressed , with 376 ( 5 . 5% ) upregulated in ptc+ samples ( green ) and 248 ( 3 . 6% ) downregulated in ptc+ samples ( red , Supplementary file 2; Materials and methods ) . The dashed line indicates the tuning parameter , delta , which is chosen according to the specified FDR ( ≤0 . 01 ) . Inhibition of dTRAF1 expression by a second RNAi line also abolishes pJNK staining ( E , ptc>dTRAFRNAi#2 , and ( F ) leads to a smaller adult wing ( red ) compared to control ( blue ) . ( G ) Quantification of adult wing size . ( H ) Multiple Ci binding sites ( red lines ) are identified within the dTRAF1 gene region ( green ) . Height of red line indicates significance of the binding site . Boxes indicate exons , and arrowed lines indicate introns in the direction of transcription . For box plot , whiskers represent maximum and minimum values . =p<0 . 0001 . Bar: 50 um . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 01510 . 7554/eLife . 11491 . 016Figure 5—figure supplement 2 . Inhibiting dTRAF1 can modify an activated Ci phenotype . Related to Figure 5 . ( A ) Compared to control wings ( blue , ptc> ) , inhibiting dTRAF1 while activating Ci still leads to a dTRAF1 phenotype of a smaller wing ( red , ptc>CiACT , dTRAF1RNAi ) . Compare to Figure 4E , G . For box plot , whiskers represent maximum and minimum values . =p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 016 dTRAF1 is expressed along the A/P boundary ( Preiss et al . , 2001 ) and ectopic expression of dTRAF1 activates JNK signaling ( Cha et al . , 2003 ) . Thus , positive regulation of dTRAF1 expression by Ci could establish a stripe of pJNK that regulates wing size . Indeed , we identified Ci binding motifs in the dTRAF1 gene ( Figure 5—figure supplement 1H ) , and a previous large-scale ChIP study confirms a Ci binding site within the dTRAF1 gene ( Chr2L: 4367100- 4371393; [Biehs et al . , 2010] ) . Consistently , a reduction in Ci led to a 29% reduction in dTRAF1 expression in wing discs ( Figure 5J; ptc-Gal4 , UAS-CiRNAi ) . Given that the reduction of dTRAF1 expression in the ptc stripe is buffered by Hh-independent dTRAF1 expression elsewhere in the disc ( Preiss et al . , 2001 ) , this 29% reduction is significant . Furthermore , inhibition of dTRAF1 by RNAi knockdown abolished pJNK staining ( Figure 5D–E and Figure 5—figure supplement 1E; ptc-Gal4 , UAS-dTRAF1RNAi#1or2 ) . Finally , these animals have smaller wings without obvious pattern defects ( Figure 5F–I and Figure 5—figure supplement 1F–G ) . Conversely , overexpression of dTRAF1 causes embryonic lethality ( ptc-Gal4 , UAS-dTRAF1 ) , making it not possible to attempt to rescue a dTRAF1 overexpression wing phenotype by knockdown of bsk . Nevertheless , it has been shown that dTRAF1 function in the eye is Bsk-dependent ( Cha et al . , 2003 ) . Finally , inhibition of dTRAF1 modulates the phenotype of activated Ci signaling ( ptc-Gal4 , UAS-dTRAF1RNAi , UAS-CiACT; Figure 5—figure supplement 2 ) . Together , these data reveal that the pJNK stripe in the developing wing is established by Hh signaling through Ci-mediated induction of dTRAF1 expression . Finally , we detected localized centers of pJNK activity during the development of other imaginal discs including the eye/antenna and leg ( Figures 6A and 6G ) . Inhibition of localized JNK signaling during development caused a decrease in adult antenna size ( Figures 6B–C and 6F; dll-Gal4 , UAS-bskDN ) and leg size ( Figures 6H–I and 6L; dll-Gal4 , UAS-bskDN ) . Conversely , increasing JNK signaling during development resulted in pupal lethality; nevertheless , overall sizes of antenna and leg discs were increased ( Figures 6D–E and 6J–K; dll-Gal4 , UAS-egr ) . Together , these data indicate that localized JNK signaling regulates size in other organs in addition to the wing , suggesting a more universal effect of JNK on size control . 10 . 7554/eLife . 11491 . 017Figure 6 . Modulation of localized JNK signaling within the developing antenna or leg changes organ size . pJNK ( green ) staining of wildtype antenna/eye ( A ) and leg ( G ) third instar discs . Inhibition of JNK in the developing antenna ( B-C , F , dll>bskDN ) or leg ( H-I , L , dll>bskDN ) leads to a smaller adult organ . Increased JNK activation within the antenna ( D-E , dll>egr , RFP , red ) or leg disc ( J-K , dll>egr , RFP , red ) causes an increase in disc size . ( M ) Model of how localized JNK signaling regulates wing size during development . Engrailed ( En ) controls Hh signaling , leading to a stripe of active Ci along the A/P boundary . Ci increases transcription of dTRAF1 , activating JNK ( pJNK , green ) . JNK acts in a non-canonical , Jun-independent manner to regulate Yki or Yki-dependent signaling . As the human dTRAF1 homolog , TRAF4 , and Hippo components are amplified in numerous cancers , these findings provide a new mechanism for how the Hh pathway could contribute to tumorigenesis ( Camilleri-Broët et al . , 2006; Harvey et al . , 2013 ) . For box plots , whiskers represent maximum and minimum values ( F , L ) . ***=p<0 . 0001 . Bar: 100 umDOI: http://dx . doi . org/10 . 7554/eLife . 11491 . 017 Intrinsic mechanisms of organ size control have long been proposed and sought after ( Bryant and Simpson , 1984; Vogel , 2013 ) . Our study reveals that in developing Drosophila tissues , localized , organ-specific centers of JNK signaling contribute to organ size in an activity level-dependent manner . Such a size control mechanism is qualitatively distinct from developmental morphogen mechanisms , which affect both patterning and growth ( Zecca et al . , 1995 ) . Aptly , this mechanism is still integrated in the overall framework of developmental regulation , as it is established in the wing by the Hh pathway ( Figure 6M ) . Our data indicate that localized JNK signaling is activated by Ci-mediated induction of dTRAF1 expression . Furthermore , we discovered that it is not canonical Jun-dependent JNK signaling , but rather non-canonical JNK signaling that regulates size , possibly through Jub-dependent regulation of Yki signaling , as described for regeneration ( Sun & Irvine , 2013 ) ( Figure 6M ) . As the human dTRAF1 homolog , TRAF4 , and Hippo components are amplified in numerous cancers ( Camilleri-Broët et al . , 2006; Harvey et al . , 2013 ) , these findings provide a new mechanism for how the Hh pathway could contribute to tumorigenesis . More importantly , these findings offer a new strategy for potential cancer therapies , as reactivating Jun in Hh-driven tumors could lead tumor cells towards an apoptotic fate . Fly crosses were maintained at 25°C on standard cornmeal-molasses media unless otherwise indicated ( see Experimental Genotypes ) . When possible , crosses were established so that every experimental animal had an in-vial Gal4 alone control . For experiments that necessitated precise developmental staging , 2 hr egg lays were conducted on apple juice agar plates with yeast paste . For all other experiments , females were allowed to lay eggs on standard media for 24 hr , after which they were removed and progeny were considered as 12 +/- 12 hr after egg lay . The following stocks were utilized: ( 1 ) Canton-S ( 02 ) y , hepr75 , FRT10 . 1/FM7iGFP ( Glise et al . , 1995 ) ( 2 ) Ubi-GFP , FRT10 . 1;; hs-FLP , MKRS/TM6B ( 3 ) UAS-puc ( III ) ( Martín-Blanco et al . , 1998 ) ( 4 ) w; ap-GAL4 , UAS-src-RFP; Sb/TM6B ( 5 ) w; ptc-GAL4 , UAS-src-RFP; Sb/TM6B ( 6 ) UAS-bskRNAi ( II and III ) VDRC 34138 ( Perez-Garijo et al . , 2013 ) and BDSC 32977 ( 7 ) w , UAS-bskDN ( X ) ( 8 ) w;; UAS-bskDN/TM6B ( 9 ) w;; rn-GAL4/TM6B ( 10 ) y , UAS-p35; Adv/CyO; Sb/TM6B ( 11 ) w; Sp/CyO; UAS-egr/MKRS ( 12 ) UAS-diap1 ( III ) BDSC 6657 ( 13 ) UAS-bskAY ( II ) BDSC: 6407 ( 14 ) UAS-CiRNAi ( II and III ) BDSC 31236 and 31236 ( 15 ) UAS-Ci5m/TK-GFP ( “UAS-CiACT” ) ( Price and Kalderon , 1999 ) ( 16 ) pucE69/TM6B ( “puc-lacZ” ) ( Ring and Martinez Arias , 1993 ) ( 17 ) UAS-dTRAF1RNAi ( X and III ) VDRC 21213 and 21214 ( 18 ) UAS-junRNAi ( III ) BDSC 31595 and VDRC 10835 ( 19 ) UAS-kayRNAi#1 ( III ) BDSC 33379 and 31322 ( 20 ) UAS-jubRNAi ( III and II ) BDSC 32923 and 41938 ( 21 ) y , w;; latse26-1/TM6B ( 22 ) yw; UAS-yki . GFP; Sb/TM6B BDSC 28815 ( Oh and Irvine , 2008 ) ( 23 ) UAS-ykiRNAi/TM3 BDSC 31965 ( 24 ) UAS-fjRNAi/TM3 BDSC 28009 ( 25 ) UAS-fj . V5 ( III ) BDSC 44252 ( 26 ) w; dll-Gal4 , UAS-src . RFP/CyO ( 27 ) UAS-dppRNAi ( III ) BDSC 25782 ( 28 ) UAS-EGFRRNAi ( III ) BDSC 25781 ( 29 ) UAS-ykiS111A . S168A . S250A . V5 ( III ) BDSC 28817 Antibody staining was performed according to standard procedures for imaginal discs . The following antibodies were used: rabbit PhosphoDetectTM anti-SAPK/JNK ( pThr183 , pTyr185 ) ( 1:100 , Calbiochem , immunogenic sequence is 100% identical to D . melanogaster bsk/JNK ) , rabbit anti-ACTIVE® JNK ( 1:100 , Promega , immunogenic sequence is 100% identical to D . melanogaster bsk/JNK ) , rabbit anti-cleaved-caspase 3 ( 1:250 , Cell Signaling ) , mouse anti-betagalactosidase ( 1:500 , Sigma ) , rabbit anti-pERK ( 1:75 , Cell Signaling ) , rabbit anti-pSMAD ( 1:75 , Cell Signaling ) , rabbit anti-phosphorylated histone 3 ( 1:250 , Cell Signaling ) , goat Alexa-488-conjugated anti-rabbit IgG ( 1:250 , Invitrogen ) , goat Alexa-488-conjugated anti-mouse IgG ( 1:250 , Invitrogen ) , goat Alexa-555-conjugated anti-rabbit IgG ( 1:250 , Invitrogen ) . EdU staining was performed according to established protocol ( Gouge and Christensen , 2010 ) using the Click-iT EdU cell proliferation assay kit ( Invitrogen ) , Grace’s Media ( Invitrogen ) and a 10 min EdU incubation . Imaginal discs to be imaged by confocal microscopy were mounted in Vectashield mounting media with DAPI ( Vector Labs ) . Confocal images were taken with a Zeiss LSM510 Meta confocal microscope or a Leica TCS SP8 STEAD 3X confocal microscope with 405nm , 488nm , 561nm , and 633nm lasers . Both microscopes gave similar results . Measurements of disc size were performed from images of at least fifteen discs using NIH Image-J software . Whole Canton-S and hepr75/Y larvae were lysed in standard RIPA buffer with protease and phosphatase inhibitors . Proteins were separated by SDS-PAGE using a 4–15% acrylamide gel ( BioRad ) , transferred for 1 hr at 4°C , and probed with primary antibodies: rabbit anti-pJNK ( Calbiochem , 1:1000 ) and mouse anti-alpha tubulin ( Sigma , 1:4000 ) . HRP-conjugated secondary antibodies ( anti-rabbit and anti-mouse ) were used at 1:5000 . ECL ( Pierce ) was used for detection with film . Adult wings , legs , or antenna were dissected in 70% ethanol , mounted in Permount mounting media ( Fisher Scientific ) , and imaged with a Leica DFC300FX camera on a Leica MZ FLIII stereomicroscope . Measurements of wing size were performed from images of twenty to sixty female flies using NIH Image-J software . Wing images were false-colored and overlayed to scale using Adobe Photoshop CS3 software . Cell size was measured by dividing the number of hairs ( 1 hair/cell ) by a set area using Adobe Photoshop CS3 software . Mean EdU signal was measured in Adobe Photoshop CS3 . Measurements of antenna or leg size were performed from images of at least twenty male flies for each genotype using NIH Image-J software . To determine whether differences in area were statistically significant , two-sided student’s t-tests were performed using raw data values , matched for temperature and sex . Box plots were generated where whiskers represent maximum and minimum , a plus sign indicates the mean , a horizontal line within the box indicates the median , and the box represents the 25–75% quartile range . Both parametric and non-parametric analyses were performed , and p-values less than 0 . 05 were considered significant . Data are presented as relative to the mean of the matched Gal4-alone control . For each of three biological replicates , 200 pairs of wing imaginal discs were dissected from third instar larvae of the genotypes hh-Gal4; UAS-mCD8GFP or ptc-Gal4; UAS-mCD8GFP . Discs were stored in Schneider's Drosophila Media ( 21720 , Invitrogen ) plus 10% FBS ( 10438 , Invitrogen ) on ice for less than two hours prior to cell dissociation . Discs were washed twice with 1 ml cell dissociation buffer ( Sigma , C-1544 ) . Elastase ( Sigma , E-0258 ) was diluted to 0 . 4 mg/ml in fresh cell dissociation buffer once discs were ready . Discs were incubated for 20 min at room temperature in 0 . 4 mg/ml elastase with stirring by a magnetic micro stirring bar . Undissociated tissue was spun out , cell viability was measured using the Beckman Vi-CELL Cell Viability Analyzer ( >80% ) , and cells were immediately isolated using the BD FACSAria II system within the Stanford FACS facility . Dead cells labeled with propidium iodide ( P3566 , Invitrogen ) were excluded during FACS , and purity of sorted cells was greater than 99% by post-sorting FACS analysis . Total RNA was extracted from sorted cells ( RNeasy , Qiagen ) , quality was assessed with the Agilent Bioanalyzer 2100 ( RIN > 7 . 0 ) , and microarray analysis was performed in the Stanford Protein and Nucleic Acid Facility ( Affymetrix D . mel GeneChip Genome 2 . 0 microarrays ) . All analyses were conducted in R version 3 . 1 . 1 ( 2014-07-10 ) . Expression values were determined using the affy package ( Gautier et al . , 2004 ) , available from BioConductor ( http://bioconductor . org ) . The automatically downloaded Drosophila 2 . 0 CDF environment was utilized . Probe level data from the CEL files were imported using the function ReadAffy and converted to expression values using the function rma with default settings . This method implements robust multi-array average ( RMA ) for background correction followed by quantile normalization . PM correction was not performed . Probe level expression values were combined into probe set expression measures using medianpolish , the standard summary method employed in RMA ( Irizarry et al . , 2003 ) . Expression values are log2 transformed . Post-normalization microarray quality assessment was conducted using the arrayQualityMetrics package ( Kauffmann et al . , 2009 ) , available from BioConductor . Default settings were used , with ptc domain ( ptc+ ) versus posterior ( hh+ ) as the covariate in intgroup . Biological replicates cluster together in a dendrogram of inter-array difference , estimated as the mean absolute difference between the data of the arrays ( Figure 5—figure supplement 1A ) , indicating that biological effects are stronger than any batch effects . Similarly , principle components analysis also separates biological replicates into two clusters ( Figure 5—figure supplement 1B ) . Outliers were not detected by either of these methods . Probe sets were mapped to genes using the drosophila2 . db annotation package ( version 3 . 0 . 0 ) , available from BioConductor . 14 , 481 of 18 , 952 ( 76 . 4% ) probe sets map to gene isoforms—12 , 676 ( 87 . 5% ) of which correspond to unique genes ( some genes are mapped by ≥1 probe set ) . In order to minimize technical artifacts , probe sets mapping to the same gene were not combined . Based on the distribution observed in the density plot of normalized probe set expression values , probe sets ( genes ) with median log2 expression value ≥6 . 5 in at least one condition ( ptc+ and/or hh+ ) were considered to be expressed ( Figure 5—figure supplement 1C ) . According to these criteria , 7 , 228 of 18 , 952 probe sets ( 38 . 1% ) are expressed . This corresponds to 6 , 854 of 14 , 481 gene isoforms ( 47 . 3% ) , which corresponds to 6 , 397 of 12 , 676 unique genes ( 50 . 4% , Figure 5—figure supplement 1D , Supplementary file 1 ) . To identify probe sets ( genes ) differentially expressed between ptc+ and posterior ( hh+ ) samples , we used the samr package , an R implementation of significance analysis of microarrays ( Tusher et al . , 2001 ) . This package is available from CRAN ( http://cran . r-project . org/ ) . Only expressed probe sets mapping to genes ( 6 , 854 ) were considered in this analysis . Differentially expressed probe sets were identified with the function SAM , using a two class unpaired response type , the t-statistic as the test statistic , and a false discovery rate ( FDR ) threshold of 0 . 01 . The maximum number of possible permutations ( 720 ) was used . To ensure these results are biologically meaningful , we further trimmed this list to probe sets with a minimum 1 . 5 fold change between ptc+ and hh+ cells . Based on these criteria , 624 of 6 , 854 probe sets ( 9 . 1% ) are differentially expressed , with 376 ( 5 . 5% ) upregulated in ptc+ samples and 248 ( 3 . 6% ) downregulated in ptc+ samples ( Figure 5—figure supplement 1D , Supplementary file 2 ) . A gene was considered differentially expressed if any mapped probe set was differentially expressed . Therefore , of the 6 , 397 unique expressed genes , 604 ( 9 . 4% ) are differentially expressed , 363 ( 5 . 7% ) upregulated and 242 ( 3 . 8% ) downregulated . One gene , Tie , was mapped by probe sets both up- and down-regulated . The quantile-quantile plot in Figure 5—figure supplement 1D was prepared using the samr . plot function . Total RNA was extracted from third instar wing discs from ptc-Gal4 or ptc-Gal4 , UAS-CiRNAi animals using a standard TriZol extraction . RNA was reverse transcribed using the iScript cDNA Synthesis Kit ( Bio-Rad ) according to manufacturer’s instructions . dTRAF1 expression was quantified relative to Rp49 ( RpL32- FlyBase , endogenous control ) by real-time PCR performed in triplicate using the SYBR Green fast kit ( Applied Biosystems ) and an Applied Biosystems machine according to the manufacturer’s instructions . The following primers were used: dTRAF1 , 5’-GCACTCCATCACCTTCACAC-3’ and 5’-TAGCTGATCTGGTTCGTTGG-3’; Rp49 , 5′-GGCCCAAGATCGTGAAGAAG-3′ and 5′-ATTTGTGCGACAGCTTAGCATATC-3′ . The Drosophila Ci positional weight matrix from the BioBase TRANSFAC database was queried against the Drosophila melanogaster genome with a p-value <0 . 0001 ( chosen based on known Ci binding sites within ptc ) using FIMO ( MEME ) and aligned back to the UCSC genome browser .	A key challenge in biology is to understand what determines size . As an animal grows , signals are produced that control the size of its organs . Many of the signaling pathways that regulate size during normal animal development also contribute to the formation of tumors . Therefore , it is important to find out exactly how the signaling molecules that regulate size are linked to those that regulate tumor growth . A protein called JNK activates a signaling pathway that triggers tumor growth . JNK signaling also stimulates cells to multiply in tissues that need repair , but it is not known whether it also regulates the size of organs during animal development . Here , Willsey et al . investigate whether JNK is active in the developing wings of fruit flies , which are commonly used as models of animal development . The experiments show that JNK is active in a stripe across the developing wing and is required for the wing to grow to its proper size . A master signal protein called Hedgehog is responsible for establishing this stripe of JNK activity . Unexpectedly , rather than acting through its usual signaling pathway , JNK activates the Hippo pathway in the wing to control organ size during development . Willsey et al . ’s findings highlight potential new targets for cancer therapies . A future challenge will be to find out whether small patches of JNK signaling are found in the developing organs of other animals , and whether they can help explain how size changes between species .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2016	Localized JNK signaling regulates organ size during development
Pain is the most prominent symptom of osteoarthritis ( OA ) progression . However , the relationship between pain and OA progression remains largely unknown . Here we report osteoblast secret prostaglandin E2 ( PGE2 ) during aberrant subchondral bone remodeling induces pain and OA progression in mice . Specific deletion of the major PGE2 producing enzyme cyclooxygenase 2 ( COX2 ) in osteoblasts or PGE2 receptor EP4 in peripheral nerve markedly ameliorates OA symptoms . Mechanistically , PGE2 sensitizes dorsal root ganglia ( DRG ) neurons by modifying the voltage-gated sodium channel NaV1 . 8 , evidenced by that genetically or pharmacologically inhibiting NaV1 . 8 in DRG neurons can substantially attenuate OA . Moreover , drugs targeting aberrant subchondral bone remodeling also attenuates OA through rebalancing PGE2 production and NaV1 . 8 modification . Thus , aberrant subchondral remodeling induced NaV1 . 8 neuronal modification is an important player in OA and is a potential therapeutic target in multiple skeletal degenerative diseases . Subchondral bone is an integral component of the joint , absorbing compressive forces during movement ( Martel-Pelletier et al . , 2016 ) . Physiological subchondral bone remodeling maintains its structural integrity and supports the overlying articular cartilage . Age or trauma ( Hayami et al . , 2004 ) related alteration of subchondral bone is a principal risk factor of osteoarthritis ( OA ) , the most common joint disease ( Berenbaum et al . , 2018 ) characterized by cartilage destruction ( Glasson et al . , 2005; Kim et al . , 2014a ) , subchondral sclerosis ( Suri et al . , 2007 ) and synovitis ( Benito et al . , 2005; Sellam and Berenbaum , 2010 ) . Pain , as the major symptom for OA patients ( Lane et al . , 2010 ) , often leads to physical disability and mortality in senior patients ( Chen et al . , 2017 ) . Although the major local source is not clearly defined ( Malfait and Schnitzer , 2013 ) , there is evidence showing that synovitis ( Sellam and Berenbaum , 2010 ) and subchondral bone marrow lesion ( BML ) are highly relevant to OA pain . BML is a fluid enriched area under magnetic resonance imaging ( MRI ) and is characterized as bone marrow edema , fibrosis , microfractures or trabecular pattern alterations in pathological examinations , with high relevance to abnormal bone remodeling . However , how subchondral BML induces OA pain , still remains largely unknown . We previously showed that aberrant subchondral bone remodeling in response to altered mechanical loading patterns in OA was initiated by over-activation of transforming growth factor β1 ( TGF-β1 ) . High levels of subchondral TGF-β1 signaling induces mesenchymal stem cells ( MSC ) clustering and leads to the uncoupling of osteoblastic bone formation , osteoclastic bone resorption and angiogenesis . Moreover , we found overactivated osteoclasts in BML aggravated OA pain by secreting the axon guidance molecule Netrin-1 to induce subchondral sensory innervation ( Zhu et al . , 2019 ) . The increased sensory innervation provides a structural base for the transmission of nociceptive signals from the subchondral bone to the central nervous system . However , how these nerve fibers are activated and sensitized during OA progression remains to be elucidated . The persistent low grade of local inflammation is another hallmark of OA ( Midwood et al . , 2009; Tu et al . , 2019 ) . During the uncoupled bone remodeling in OA progression , a series of pro-inflammatory factors ( Kapoor et al . , 2011; Sellam and Berenbaum , 2010 ) like prostaglandin E2 ( PGE2 ) ( Chen et al . , 2019 ) , interleukin-1β ( IL-1β ) , interleukin-6 ( IL-6 ) are released into the subchondral bone area ( Massicotte et al . , 2002 ) . We previously showed that PGE2 was a crucial factor in both pain sensation and bone metabolism ( Tu et al . , 2019 ) . However , the main source remains largely unknown . We newly discovered a feedback mechanism on sensory nerve regulation of bone mass . PGE2 concentration is inversely related to bone mass and sensory nerves monitors bone density by responding to the concentration of PGE2 in bone ( Chen et al . , 2019 ) . We found that when bone mass is decreased , the enzymatic activity of cyclooxygenase 2 ( Cox2 ) in osteoblastic lineage cells was significantly increased and thus catalyzes more arachidonic acid into PGE2 . At the early stage of OA , the bone density is temporarily decreased , which resembles the low bone density as seen in osteoporosis . Currently , there is still lack of information whether the increased PGE2 in OA subchondral bone is also attributed to the increased Cox2 activity in osteoblastic cells in response to low bone mass . PGE2 functions as an inflammatory mediator and a neuromodulator that alters neuronal excitability ( Samad et al . , 2001 ) . In the four types of G-protein-coupled EP receptors ( EP1-EP4 ) that mediate the functions of PGE2 , EP4 receptor is considered as the primary mediator of PGE2- evoked inflammatory pain hypersensitivity and sensitization of sensory neurons ( Boyd et al . , 2011; Chen et al . , 2008; Lin et al . , 2006; McCoy et al . , 2002; Nakao et al . , 2007; Southall and Vasko , 2001; Taylor-Clark et al . , 2008 ) , as evidenced by that the specific EP4 receptor antagonists could reduce acute and chronic pain ( Nakao et al . , 2007 ) , including OA pain ( Abdel-Magid , 2014 ) . Increased neuronal excitability contributes to the generation of hypersensitivity in various types of chronic pain ( Kuner , 2010; Malfait and Schnitzer , 2013 ) . PGE2 has been shown to potentiate several ion channels in neurons to enhance neuronal excitability ( Funk , 2001 ) . Voltage-gated sodium channel ( NaV ) , a member of the tetrodotoxin-resistant sodium channel ( TTX-R ) , is mainly expressed in small- and medium-sized Dorsal root ganglion ( DRG ) neurons and their fibers . The NaV is responsible for initiating and propagating electrical signal transmission by inducing Na+ influx to start action potential firing . PGE2 has been shown to modulate the sodium current of the TTX-R in DRG neurons and promote Nav1 . 8 trafficking to the cell surface ( England et al . , 1996; Liu et al . , 2010 ) . Therefore , the PGE2 induced neuronal hypersensitivity is likely to be mediated by the Nav 1 . 8 during OA progression . Among the 9 subtypes of NaVs ( NaV1 . 1–1 . 9 ) , NaV1 . 8 ( Akopian et al . , 1996 ) is the main drug target due to its highly relevant to pain signal transmission , and restricted distribution in primary nociceptive neurons ( Akopian et al . , 1999; Julius and Basbaum , 2001 ) . Gain of function mutations in human in the promoter region of NaV1 . 8 directly induces pain hypersensitivity ( Duan et al . , 2018 ) . Interestingly , animals lacking NaV1 . 8 display significant lower mechanical pain sensitivity with modest changes in heat or innocuous touch sensitivities ( Akopian et al . , 1999; Basbaum et al . , 2009 ) . This specificity of NaV1 . 8 in transducing mechanical pain signals makes it highly possible in the participation of mechanical allodynia in OA . Moreover , post-transcriptional modifications of NaV1 . 8 including phosphorylation ( Gold et al . , 1998; Hudmon et al . , 2008; Wu et al . , 2012 ) and methylglyoxalation ( Bierhaus et al . , 2012 ) further regulate its activity . A recent study demonstrated that inhibition of the expression of NaVs in nociceptive neurons was effective in OA pain alleviation ( Miller et al . , 2017 ) , with the detailed molecular mechanism remained to be clarified ( Strickland et al . , 2008 ) . In this study , we take the initiative to show that aberrant subchondral bone remodeling contributes to neuronal hypersensitivity during OA progression . Excessive PGE2 is synthesized by osteoblastic lineage cells in response to the low bone density at the early stage of OA . Excessive PGE2 sensitize sensory fibers innervates subchondral bone by upregulating the expression of sodium channel NaV1 . 8 in both subchondral bone nerve fibers and DRG neuron body , which contributes to peripheral mechanical allodynia during OA progression . Therefore , we developed a small molecule conjugate by linking the TGFβ type receptor 1 ( Tβ1R ) inhibitor ( LY-2109761 ) and alendronate ( Aln ) ( Hayami et al . , 2004 ) to achieve bone-targeted delivery . We used this conjugate ( Aln-Ly ) as a proof of concept drug to test whether reversing the aberrant bone remodeling by synergistically inhibiting osteoclast bone resorption and the excessive TGF-β activity can substantially reduce the PGE2 production and subsequent mechanical hypersensitivity that generated in OA subchondral bone . To identify the primary voltage-gated sodium channel in subchondral sensory fibers that responsible for mechanical hypersensitivity during OA progression , we tested the expression levels of different sensory related sodium channel NaVs in OA mice post anterior cruciate ligament transection ( ACLT ) . The transcription levels of mRNAs that encode NaVs in DRG including NaV1 . 1 , NaV1 . 2 , NaV1 . 3 , NaV1 . 6 , NaV1 . 7 , NaV1 . 8 , NaV1 . 9 were measured by qPCR using mRNA isolated from mouse ipsilateral L3-5 DRGs one month post-ACLT or sham surgery . Compares to the sham-operated group , the expression of mRNA encoding NaV1 . 8 increased 2 . 5 folds in the ACLT group as the highest upregulation among all the NaVs . The mRNAs encoding NaV1 . 7 and NaV1 . 9 showed moderate upregulation in the ACLT group relative to that of the Sham group while the changes of NaV1 . 1 NaV1 . 2 NaV1 . 3 or NaV1 . 6 were not detected ( Figure 1a ) . Therefore , we further investigated NaV1 . 8 protein expression in the immune-histological analysis of OA subchondral bone . The intensity of NaV1 . 8 immunofluorescence in subchondral bone was also elevated about 2 to 3 folds in OA mice compared to sham- operated mice one- or two-months post-surgery ( Figure 1b and c ) . To examine whether the increase of NaV1 . 8 expression is limited to in a certain subtype ( s ) of the DRG neuron that innervates subchondral bone in OA mice , NaV1 . 8 was co-stained with different markers for sensory nerve subtypes based on the current classification of DRG neurons ( Usoskin et al . , 2015 ) . The expression rate of NaV1 . 8 in total nerve fibers ( labeled by pan neuron marker PGP9 . 5 ) innervated in subchondral bone significantly increased post-ACLT ( Figure 1—figure supplement 1a , f ) . Moreover , the elevated NaV1 . 8 expression was highly co-localized with the peptidergic nociceptor marked by calcitonin gene-related peptide ( CGRP ) ( Brain et al . , 1985; Figure 1—figure supplement 1b , f ) and mechanosensitive low-threshold mechanoceptors ( labeled by PIEZO2 ) ( Eijkelkamp et al . , 2013 ) . The expression of NaV1 . 8 was also slightly elevated in the synovium ( Figure 1—figure supplement 1f 1 hr , j ) . Both western blot analysis ( Figure 1f ) and immunostaining ( Figure 1—figure supplement 1f 1 g , i ) of ipsilateral lumbar 3–5 DRG confirmed the upregulation of NaV1 . 8 expression at the DRG level . We then further validated whether DRG neurons with upregulated NaV1 . 8 expression directly innervates fibers in the subchondral bone . We injected a neurophilic fluorescent dye ( DiI ) into the subchondral bone to label the distal nerve fibers ( Ferreira-Gomes et al . , 2010 ) . We found that the NaV1 . 8+ neurons labeled by DiI significantly increased in the DRG neurons of OA rats relative to sham-operated rats ( Figure 1g , h ) , indicating that DiI was transported into DRG neurons by the sensory fibers innervated subchondral bone in a retrograde manner . These results suggest that the expression of pain-related sodium channel NaV1 . 8 is upregulated in DRG neurons and their axons that innervate subchondral bone during progression . We then examined the association between NaV1 . 8 neuronal activity and OA pain . Von Frey test showed that the hind paw withdrawal threshold ( HPWT ) dropped nearly 60% and maintained at this level throughout the two month-period post ACLT relative to the sham-operated group , suggesting that development of mechanical allodynia in OA mice ( Figure 1g; Chen et al . , 2017 ) . To assess the potential role of NaV1 . 8 in DRG neuronal excitability , we used an in vivo DRG imaging in PirtGCaMP3fl/-mice that we recently developed . In this genetically targeted mice , genetic-encoded Ca2+ indicator GCaMP3 is specifically expressed in >95% of all DRG neurons under the control of the Pirt promoter . In the PirtGCaMP3fl/-mice , the excitability of the nociceptive neurons in DRG can be visualized by fluorescence signals of calcium influx . The number of excited DRG neurons ipsilateral to the surgery significantly increased in OA mice compared to sham-operated mice , and importantly , administration of NaV1 . 8 inhibitor ( A-803467 ) ( Jarvis et al . , 2007 ) blunted the signal in DRG ( Figure 1h and i ) . To validate the excitation of DRG neurons related to NaV1 . 8 , we performed patch-clamp in the DRG neurons that were isolated from mice that underwent ACLT or sham surgery . The action potential number and NaV1 . 8 current density significantly elevated in ACLT mice relative to sham-operated mice , and the elevations were blocked by A-803467 ( Figure 1j–l , Figure 1—figure supplement 1f 1 k and l ) . Thus , the activation of NaV1 . 8 mediates OA pain related DRG neuron hypersensitivity . We then examined the mechanism of upregulation of NaV1 . 8 expression during OA progression . To examine whether excessive PGE2 contributes to the upregulation of NaV1 . 8 , we firstly performed immunostaining of cyclooxygenase 2 ( Cox2 ) in subchondral bone sections . Cox2 expression was significantly increased in subchondral bone and primarily in osteocalcin positive osteoblastic cells post ACLT mice compared with sham-operated mice ( Figure 2a and b ) . Consistently , PGE2 concentration in subchondral bone increased about three times in OA mice relative to sham-operated mice ( Figure 2c , Figure 2—figure supplement 1a–d ) . To examine if elevated PGE2 upregulates the expression of NaV1 . 8 for the mechanical allodynia in OA , we generated osteoblast specific Cox2 deficient mice ( Cox2Oc-/- mice ) by crossbreeding Cox2fl/fl mice with Bglap-Cre mice . PGE2 concentration in subchondral bone was significantly lower in Bglap-Cre::Cox2fl/fl mice compared with Cox2fl/fl mice post ACLT ( Figure 2h ) . Notably , the NaV1 . 8 immunofluorescence intensity in subchondral bone immunostaining was also significantly reduced in Bglap-Cre::Cox2fl/fl mice compared with Cox2fl/fl mice ( Figure 2d and i ) . In addition , the number of NaV1 . 8+ neurons in DRG was also significantly decreased ( Figure 2e and j ) in the Bglap-Cre::Cox2fl/fl mice relative to Cox2fl/fl mice . We then investigated whether a decrease of PGE2 alleviates OA pain . We crossed Bglap-Cre::Cox2fl/fl mice with Pirt GCaMP3fl/- mice to measure the DRG neuronal excitability in Bglap-Cre::Cox2fl/fl::Pirt GCaMP3fl/- mice post ACLT . Pirt-GCaMP3 DRG imaging showed that the number of excited neurons was significantly reduced in Bglap-Cre::Cox2fl/fl::PirtGCaMP3fl/- mice compared with Cox2fl/fl::Pirt GCaMP3fl/- mice post ACLT ( Figure 2f and k ) . Moreover , the single neuron excitability in ipsilateral L4 DRG was functionally tested by whole cell patch clamp electrophysiology . The whole cell current clamp revealed that the action potential firing number was significantly decreased in Bglap-Cre::Cox2fl/fl mice compared with Cox2fl/fl mice after ACLT ( Figure 2g and l ) . Concurrently , the NaV1 . 8 current density was reduced for about 40% recorded by the whole-cell voltage-clamp ( Figure 2g and m ) . The mechanical allodynia was simultaneously attenuated in Bglap-Cre::Cox2fl/flmice relative to Cox2fl/fl mice as revealed by the Von Frey behavior test ( Figure 2n ) . Catwalk analysis also showed that the Maximal Contact At and Maximal Intensity of ipsilateral hind paw was significantly higher in Bglap-Cre::Cox2fl/fl mice compared with Cox2fl/fl mice post ACLT ( Figure 2o ) . Thus , PGE2 derived from osteoblastic cells stimulates the pain hypersensitivity in OA mice likely by upregulating of NaV1 . 8 in subchondral nociceptive neurons . To examine whether EP4 at sensory neurons is the primary receptor that propagates PGE2 signals in upregulating NaV1 . 8and OA pain , we specifically knocked out EP4 , the skeletal pain related receptor for PGE2 ( Yoshida et al . , 2002 ) , in peripheral sensory nerves by crossbreeding Advillin-Cre ( Avil-Cre ) ( Zurborg et al . , 2011 ) mice with Ptger4 fl/fl mice ( Ptger4 is the gene that encodes EP4 receptor ) . Consistently with Bglap-Cre::Cox2fl/fl mice , the intensity of NaV1 . 8 immunofluorescence was significantly reduced in Avil-Cre::Ptger4 fl/fl mice compared with Ptger4 fl/fl mice post-ACLT ( Figure 3a , d , Figure 2—figure supplement 1e–h ) . We then further confirm this finding in DRG neurons that cultured in PGE2 . We found that NaV1 . 8 protein expression was significantly reduced by siRNA against EP4 in western blot analysis of levels from the cell lysates . Knocking-down the expression of EP1-EP3 did not have a significant effect on Nav1 . 8 expression ( Figure 2—figure supplement 1i ) . To determine the effect of conditional deletion of EP4 on DRG neuronal excitability , we generated Avil-Cre::Ptger4 fl/fl::Pirt - GCaMP3fl/- and Ptger4 fl/fl::Pirt GCaMP3fl/- ACLT mice . In vivo ipsilateral L4 DRG Pirt GCaMP3 imaging demonstrated significantly dampened excitability in the Ptger4 fl/fl:: Pirt - GCaMP3fl/- mice relative to the control group ( Figure 3b and e ) . The patch-clamp analysis further revealed that the DRG neuronal hypersensitivity and NaV1 . 8 currents were significantly reduced in Avil-Cre::Ptger4 fl/fl mice relative to Ptger4 fl/fl mice post-ACLT ( Figure 3c , f , g ) . Moreover , both Von Frey test and catwalk analysis showed attenuation of OA pain when EP4 is conditionally deleted in sensory neurons ( Figure 3h and i ) . Thus , the EP4 receptor expressed in DRG neurons is responsible for the propagation of subchondral PGE2-induced upregulation of NaV1 . 8 and neuronal excitability in OA mice . To investigate the mechanism of PGE2 stimulated upregulation of NaV1 . 8 expression , RT-qPCR was performed with mRNA isolated from primary DRG neurons treated with PGE2 . The result showed that NaV1 . 8 transcription levels were significantly elevated at 6 and 12 hr after incubation with PGE2 ( Figure 4a ) . Moreover , PGE2 stimulated phosphorylation of protein kinase A ( PKA ) ( Gold et al . , 1998 ) and cAMP response element-binding protein ( Creb1 ) ( Lonze and Ginty , 2002; Figure 4b ) . Notably , the effect of PGE2 and Forskolin , a cAMP stimulant in the upregulation of NaV1 . 8 protein expression was dampened by Creb1 inhibitor 666–15 ( Figure 4c ) , indicating PGE2 stimulates Nav1 . 8 expression through the PKA-Creb1 signaling pathway . Consistently , the neuronal excitability and NaV1 . 8 current density stimulated by PGE2 was abolished by the application of PKA inhibitor ( PKI ) or CREB inhibitor 666–15 ( Figure 4d–f ) . Co-immunofluorescence staining further demonstrated that PKA levels significantly increased in NaV1 . 8 positive DRG neurons in ACLT mice compared with sham-operated mice ( Figure 4g and h ) . To examine the mechanism of PGE2-induced NaV1 . 8 transcription , we performed chromatin immunoprecipitation ( ChIP ) assay with three potential pCreb1-binding elements in the NaV1 . 8 promoter . ChIP assay revealed that pCreb1 binds to the NaV1 . 8 promoter at binding site two to stimulate transcription of NaV1 . 8 gene ( Figure 4i–k ) . Taken together , our findings reveal that PGE2 induces transcription of NaV1 . 8 by stimulation phosphorylation PKA and pCreb1 , which directly binds to NaV1 . 8 promoter . We next tested whether the deletion of NaV1 . 8+ neurons could attenuate OA pain . Scn10a -Cre mice were crossed with Rosa26iDTRfl/fl mice to generate Scn10a -Cre:: Rosa26iDTRfl/fl mice . In these mice , the NaV1 . 8+ neurons underwent apoptosis upon receiving the injection of diphtheria toxin ( Buch et al . , 2005 ) . The ablation of NaV1 . 8+ neurons had no effect on the articular cartilage deterioration , as shown by similar OARSI scores between Scn10a -Cre:: Rosa26iDTRfl/fl mice and Rosa26iDTRfl/fl mice post-ACLT ( Figure 5a , e ) . The efficacy of specific neuron ablation was evidenced by a significant reduction of NaV1 . 8+ signals at both subchondral bone and DRG level ( Figure 5b , c , f , g ) . Consistently , the DRG hypersensitivity was reduced as indicated by a decreased AP firing ( Figure 5d and h ) . We further investigated whether the ablation of NaV1 . 8+ sensory neurons reduces the pain in OA mice by catwalk gait analysis ( Lakes and Allen , 2016 ) . The results showed that max intensity , which reflected mechanical pain sensitivity ( Kameda et al . , 2017 ) , was increased in Scn10a -Cre:: Rosa26iDTRfl/fl mice ( Figure 5i ) . Similarly , the Von Frey test displayed a significant reduction of HPWT in Scn10a -Cre:: Rosa26iDTRfl/fl mice compared with Rosa26iDTRfl/fl mice after ACLT ( Figure 5j ) . Taken together , the Scn10a -Cre:: Rosa26iDTRfl/fl mice indicates that specific ablation NaV1 . 8 can alleviate OA pain in OA mice . We previously showed that inhibition of excessive TGF-β activity attenuated OA progression by restringing the coupling of subchondral bone remodeling ( Figure 6—figure supplement 1a–n , Figure 6—figure supplement 2 4a-e ) . We have developed a small molecule drug by conjugating TGF-β type I receptor kinase inhibitor ( TβR1I ) covalently with alendronate through a metabolically cleavable carbamate linkage ( Qin et al . , 2018 ) . The conjugate is effectively delivered to the bone surface where TβR1I is released by cleavage of the carbamate linkage in vivo . ( Figure 6—figure supplement 3a ) . Administration of the conjugate in human MSCs effectively inhibited TGF-β signaling evidenced by a significant reduction of pSMAD2/3 ( Figure 6—figure supplement 3b and c ) . As excessive PGE2 production and subsequent upregulation of Nav1 . 8 are triggered by abnormal bone remodeling , we investigated whether conjugate treatment can alleviate OA pain by downregulates the activity of Nav1 . 8 . As expected , the articular cartilage degeneration was attenuated with a weekly injection of conjugate 100 ug/kg in ACLT mice compared with the vehicle group , with a significant improvement of the OARSI score ( Figure 6a and g ) . In concurrent with the cartilage protection , the subchondral bone microarchitecture was improved in μCT analysis of ACLT mice treated with the conjugate treatment compared with the vehicle group ( Figure 6b and h , Figure 6—figure supplement 3f ) . As shown in Figure 6c and i , phosphorylation of Smad2/3 was effectively inhibited by the conjugate in the subchondral bone . The number of TRAP+ osteoclastic cells and Osterix+ pre-osteoblast were reduced ( Figure 6—figure supplement 3e and g , Figure 6d and j ) . As a result , the BML in tibial subchondral bone was significantly reduced in the conjugate treated group ( Figure 6f ) further indicating that coupling of the osteoclast bone resorption and osteoblastic bone formation were restored . Finally , we examined whether the conjugate effect on the improvement of subchondral bone structure and articular cartilage degeneration could also relieve OA pain . Interestingly , subchondral PGE2 concentration was significantly reduced in ACLT mice with the conjugate treatment relative to the vehicle group ( Figure 6l ) . Importantly , the expression of NaV1 . 8 was also reduced in both subchondral bone ( Figure 6e and k ) , and ipsilateral lumbar DRG ( Figure 6m ) . Moreover , the electrophysiological tests demonstrate that data showed that conjugate treatment blunted the upregulation of DRG neuron activity ( Figure 6n and o ) and NaV1 . 8 currents ( Figure 6n and p ) in ACLT mice . The effect of the conjugate on joint pain related behaviors were examined in the Catwalk test . HPWT , maximal contact AT and swing phase were significantly ameliorated in ACLT mice with the administration of conjugate relative to the vehicle group ( Figure 6q and r ) . These data suggest that alendronate-TβR1I conjugate relieves OA pain by modifying the disease . This was likely achieved by the decrease of PGE2 in the improvement of subchondral bone structure . Pain is the major symptom of OA , the most prevalent skeletal degenerative disease with no effective disease-modifying drugs . To date , the major local source and pathophysiological mechanisms of OA pain remain poorly understood , impeding the development of mechanism based strategies for OA pain attenuation . Based on clinical observations , in this study , we hypothesized that aberrant subchondral bone remodeling could be highly responsible for OA pain . Aberrant bone remodeling significantly stimulates the PGE2 production in subchondral bone , with the osteoblastic cell being the major source of production . Accordingly , OA pain alleviation can be achieved by specifically knocking out the PGE2 producing enzyme Cox2 in osteoblast or its receptor EP4 in peripheral sensory nerve , likely through reducing the expression of NaV1 . 8 . In particular , the direct ablation of NaV1 . 8+ DRG neurons demonstrates that NaV1 . 8 overexpression at least partially mediates neuronal hypersensitivity in OA progression . Importantly , we demonstrated that pharmacologically inhibition of aberrant bone remodeling has superior treatment effect than purely blocking pain transduction pathway as evidenced by that conjugate improved subchondral bone structure , attenuated cartilage degeneration and ameliorated OA pain simultaneously while deleting NaV1 . 8+ neurons only alleviated OA pain without disease-modifying effect in OA pathologies . Generally , central ( Kuner , 2010 ) and peripheral sensitization ( Miller et al . , 2017 ) are two principal components for OA pain . Since surgical removal of a part of arthritic knee joint in total knee replacement can immediately relieve OA pain ( Skou et al . , 2015 ) , it is believed that peripheral input is indispensable in OA pain sensitization . Several joint structures are plausible sources of OA pain ( e . g . , the synovium , tendons ) , but clinical tests do not reliably attribute the pain to those structures . Synovium , because of its dense innervation of sensory nerves , is thought to be one of the important sources of OA pain ( Kc et al . , 2016 ) . Low grade of synovitis in OA could also be able to stimulate sensory nerve endings . However , human studies showed that synovial sensory nerve declined in some of the degenerative OA patients ( Dominique Muschter et al . , 2017; Sellam and Berenbaum , 2010 ) , making this hypothesis inconclusive . Similarly , the increase of vascular and nerve growth in meniscus ( Ashraf et al . , 2011 ) and fat pad ( Bohnsack et al . , 2005 ) suggests that they might also be a source of pain . Several lines of clinical evidence point to the potential role of subchondral bone in the mechanical allodynia during OA progression ( Kwoh , 2013; Zhu et al . , 2019 ) . This is clinically supported by the immediate pain relief in OA patient after removal of degraded cartilage and underlying subchondral bone in joint surgery ( Mittag et al . , 2016 ) . Since articular cartilage is not innervated by sensory nerve , therefore , the densely innervated subchondral bone could be an essential local source for clinical OA pain . Identifying the main source of pain and related mechanisms is essential for the treatment of OA pain . The subchondral bone transmits mechanical loads produced by body weight and muscle activity . It is highly adaptable , with the ability to model and remodel in response to loading stresses . During OA development , the subchondral bone undergoes aberrant remodeling , leading to pathologic lesions . MRI studies have shown lower bone mineral density , also known as bone marrow lesions ( BML ) , and more severe disruption of subchondral bone architecture in patients with OA ( Dore et al . , 2009; Majumdar et al . , 2004 ) . Subchondral BML is the first sign of OA in animal models ( Libicher et al . , 2005 ) and strongly correlate with knee pain in humans ( Davies-Tuck et al . , 2009 ) . We previously have demonstrated aberrant bone remodeling of subchondral bone at the onset and pathological development of OA ( Zhen et al . , 2013 ) . Studies have showed perivascular sensory and sympathetic nerve fibers breach the subchondral bone in OA compared to normal joint ( Mapp and Walsh , 2012; Walsh et al . , 2010 ) . Recently , we found that excessive Netrin-1 secreted by osteoclasts in subchondral bone induces sensory nerve axonal growth in OA ( Zhu et al . , 2019 ) . We also found that during bone remodeling , PGE2 , produced from arachidonic acid by the enzymatic activity of Cox2 , activates EP4 in sensory nerves . In the present study , we found that the abnormal bone remodeling and temporary decrease of bone density in subchondral bone at the early stage of OA resembles the pathological changes as seen in osteoporosis . This explains why Cox2 activity and subsequent PGE2 production increased in response to the structural changes in OA subchondral bone . The increased nociceptive innervation to OA subchondral bone secondary to excessive netrin-1 secretion by the osteoclasts therefore favors the PGE2 induced neuronal excitations . Taken together , we believe that the development of OA pain is a synergistic result involving central sensitization in spinal cord in conjunction with peripheral input from subchondral bone , synovium , meniscus and fat pad etc . The molecular mechanism of neuronal sensitization remains a poorly understood facet of OA pathophysiology . It is widely accepted that neuronal plasticity including activation , transcriptional modification and post-transcriptional of ion channels related to electrical excitability could contribute to the generation of chronic pain ( Woolf and Salter , 2000 ) . Giving the essential role of action potential firing in peripheral nerve input , the possible involvement of ion channels was investigated in recent studies . Our findings suggest that NaV1 . 8 is the most upregulated NaV channel with a restricted localization in DRG . We therefore focused on NaV1 . 8 to the possible molecular events based on the extensive over-activation of subchondral bone remodeling in OA progression . We found the expression rate of NaV1 . 8 was significantly elevated in subchondral bone marrow , sciatic nerve and ipsilateral lumbar DRG levels . And the further screening analysis showed the expression rate of NaV1 . 8 mainly elevated in CGRP+ nociceptive fiber and piezo2+ low threshold mechanoceptive fibers . This pattern of modification of NaV1 . 8 expression in sensory neurons could explain the high sensitivity in polymodal nociception and mechanoception after ACLT . Functionally , this upregulation of expression led to larger NaV1 . 8 currents and higher excitability of DRG neurons after ACLT . The NaV1 . 8 currents are thought to be essential for action potential firing at the initial state . Being activated by PGE2 , the lager NaV1 . 8 currents could make the DRGs easier for action potential firing , thus transmitting the pain signals into higher centers for mechanical allodynia . In the short term , phosphorylation of NaV1 . 8 by PGE2 may increase the inward currents by opening the NaV1 . 8 ion gating mechanism ( Hudmon et al . , 2008 ) . Also , PGE2 increases the expression of NaV1 . 8 in a relatively long term of stimulation by PKA signaling . However , the role of other modalities of modulations like phosphorylation , methylglyoxalation need to be investigated in future studies . Nevertheless , future studies should be conducted to explain how this elevation of NaV1 . 8 could be integrated and translated into the central nervous system as pain signals . Pain sensation and OA progression are often dissociated . Late stage of radiographic OA patients with extensive subchondral bone sclerosis may result in less pain sensation and early stage patients with significant subchondral BML can be very painful . Moreover , the anti-nerve growth factor ( NGF ) tanezumab administration to OA patients relief pain with no significant attenuation on OA cartilage protection or subchondral bone sclerosis ( Lane and Corr , 2017 ) . Consistently , although pain relief can be achieved to some extent by targeting NaV1 . 8 , the cartilage or subchondral bone was not significantly protected in OA progression after ACLT in our study . These results indicate a comprehensive therapy for OA should target upstream events that cause OA progression and pain . Here , we provide a proof of principle for the potential of small molecule drug balancing aberrant bone remodeling to attenuate mechanical allodynia and OA progression in general . We achieved the bone-targeted TGF-β inhibition by conjugating the TGF-β type I receptor inhibitor ( LY-2109761 with an osteoclast targeting drug alendronate . Consistent with our previous findings , the conjugate rebalanced the uncoupled subchondral bone remodeling and reduced over-activated osteoblastic bone formation through targeting aberrant TGF-β signaling in ACLT mice model . This rebalance was effective in pain alleviation through a reduction of excessive PGE2 released into subchondral bone marrow and down-regulation of NaV1 . 8 expression and electric property to reduce mechanical allodynia in OA . Meanwhile , the reconstruction of subchondral bone architecture protected the overlying cartilage destruction and delayed the progression of the overall OA process ( Figure 7 ) . We believe this mechanism based management of OA pain may shed light on the strategy of various musculoskeletal disorders with chronic pain symptom . Nevertheless , because the pharmacological or toxicological aspect of the conjugate in vivo are largely unknown , further enhancement of therapeutic effect may be achieved with optimization of dosing and delivery strategies . We purchased C57BL/6J ( WT ) 3 months old male mice from Jackson Laboratories . We purchased Sprague Dawley ( SD ) 3 months old male rats from Charles River company . To develop the mechanical instability related OA model , we used ACLT surgery ( Malfait and Little , 2015 ) . Briefly , after ketamine and xylazine anesthesia , the left ACL was surgically transected and sham operations were performed on other groups of mice . For the time-course experiments , mice were euthanized at 4 , 8 or 12 weeks after surgery ( n = 6 per group ) . The Rosa26iDTRfl/fl mice were purchased from Jackson Laboratory . The Advillin-Cre ( Avil-Cre ) and PirtGCaMP3 mouse strain were kindly provided by Xingzhong Dong ( The Johns Hopkins University ) . The Bglap-Cre mice were provided by Thomas J . Clemens ( The Johns Hopkins University ) . The Cox2fl/fl mice were kindly provided by Harvey Herschman ( University of California , Los Angeles ) . The Ptger4 fl/fl mice were provided by Brian L . Kelsall ( the National Institutes of Health ) . The Scn10a-Cre mice were kindly provided by Yun Guan ( The Johns Hopkins University ) . Heterozygous Bglap-Cre mice were crossed with a Cox2fl/fl mouse; the offspring were intercrossed to generate the following genotypes: WT , Bglap-Cre , Cox2fl/fl , Bglap-Cre::Cox2fl/fl . Cox2fl/fl or Bglap-Cre::Cox2fl/fl mice were further crossed with PirtGCaMP3fl/- mice to generate Cox2fl/fl::PirtGCaMP3fl/- mice or Bglap-Cre::Cox2::PirtGCaMP3 fl/- mice for in vivo GCaMP3 DRG imaging . Heterozygous Avil-Cre mice were crossed with Ptger4fl/fl mice . The offspring were intercrossed to generate the following genotypes: wild type ( referred as ‘WT’ in the text ) , Avil-Cre ( Cre recombinase expressed driven by Advillin promoter ) , Ptger4fl/fl , Avil-Cre:: Ptger4fl/fl ( conditional deletion of EP4 receptor in Advillin lineage cells ) . Ptger4fl/fl or Avil-Cre:: Ptger4fl/fl mice were further crossed with Pirt-GCaMP3 fl/- mice to generate Ptger4fl/fl::Pirt-GCaMP3fl/- mice or Avil-Cre:: Ptger4fl/fl::PirtGCaMP3 fl/ mice for in vivo GCaMP3 DRG imaging . Heterozygous Scn10a-Cre mice were crossed with the Rosa26iDTRfl/f mouse; the offspring were intercrossed to generate the following genotypes: WT , Scn10a-Cre::Rosa26iDTRfl/fl , Scn10a-Cre::Rosa26iDTRfl/fl mice . We injected 12-week-old Scn10a-Cre:: Rosa26iDTRfl/fl or Rosa26iDTRfl/fl mice with 1 μg/kg DTX intraperitoneally three times per week after ACLT for 4 weeks . We obtained femurs , tibiae and DRG from the mice after euthanasia . For conjugate injections , we used intraperitoneal injection method and 1 mg/kg per week dosage according to previous toxicological experiments . All animals were maintained at the animal facility of The Johns Hopkins University School of Medicine . All the experimental protocols were approved by the Animal Care and Use Committee of The Johns Hopkins University ( Protocol number: Mo18M308 ) . After approval by the Institutional Review Board of The Johns Hopkins Hospital , , we collected tibial plateau specimens from eight individuals with osteoarthritis that underwent total knee arthroplasty . The knee joints from three healthy young adults underwent lower limb amputations after trauma serves as healthy controls . The demographic data of patients were collected . The samples were used to perform histology and immunohistochemistry after decalcification . Immediately after euthanasia , we resected and fixed the animals knee joints or DRG in 10% buffered formalin for 24 hr , decalcified them in 0 . 5 M ethylenediaminetetraacetic acid ( EDTA , pH 7 . 4 ) for 14 d and embedded them in paraffin or gelatin solution ( 20% D-sucrose , 2% Polyvinylpyrrolidone ( PVP ) and 8% gelatin in PBS ) . Four-micrometer sagittal oriented sections of the medial compartment of left knees were processed for hematoxylin and eosin , safranin orange and fast green and Tartrate-resistant acid phosphatase ( TRAP ) staining ( Sigma ) . For immunohistology and immunofluorescence , slides ( 4 μm for immunohistology , 20 μm for DRG , 60 μm for knee immunofluorescence ) were incubated with antigen retrieval buffer ( Dako , S169984-2 ) at 96°C for 15 min , gradually cooled to room temperature and washed with tris-buffered saline with Tween ( TBST ) . After blocking , the slides were incubated with primary antibodies overnight at 4°C . Secondary antibody ( 1:200 ) was used to incubate the samples for 1 hr at room temperature . For immunohistochemical staining , a horseradish peroxidase–streptavidin detection system ( Dako ) was used to detect immunoactivity , followed by counterstaining with hematoxylin ( Sigma-Aldrich ) . For immunofluorescence , the fluorescent conjugated secondary antibody ( 1:200 ) was applied . The photographs of the immunohistology sections were recorded by light microscopy ( DP71 microscope camera , Olympus ) and analyzed by OsteoMeasure XP software ( OsteoMetrics ) . We calculated OARSI scores as previously described ( Glasson et al . , 2010 ) . The OARSI scores were evaluated by two independent graders and the averages were taken . For the immunofluorescence , the photographs were shot under laser confocal microscopy ( Zeiss , LSM 780 ) and Zen 2 . 2 software . The mice knees were scanned using high-resolution μCT ( SkyScan 1275 , Bruker microCT ) as previously described ( Zhen et al . , 2013 ) . The scanner was set at a voltage of 65 kVp , a current of 153 μA and a resolution of 5 . 7 μm per pixel . We reconstructed and analyzed outcomes using NRecon v1 . 6 , and CTAn v1 . 9 , respectively . Three-dimensional reconstructions were done by CTVol v2 . 0 ( Bruker microCT ) . We defined the region of interest to cover the trabecular part of the medial compartment of tibial subchondral bone , and five consecutive images from the medial tibial plateau were used for 3-dimensional reconstruction . We analyzed 3D parameters as following: TV ( total tissue volume; containing both trabecular and cortical bone ) , BV/TV ( trabecular bone volume per tissue volume ) and Tb . Pf ( trabecular pattern factor ) . We performed in vivo μMRI studies on a horizontal 9 . 4T Bruker Biospec preclinical scanner according to our previous protocol ( Zhen et al . , 2013 ) . Briefly , we showed subchondral BML by T2-weighted scanning with 2D RARE ( rapid acquisition with relaxation enhancement ) sequence , a TE/TR ( echo time/repetition time ) of 15 . 17 ms/3 , 000 ms , 30 slices at 0 . 35 mm thickness , 1 . 75 cm ×1 . 75 cm field of view ( FOV ) with a matrix size of 256 × 128 . The fat suppression was done in T2-weighted imaging with a chemical shift selective fat saturation pulse tuned to the fat resonant frequency . Bilateral lumbar DRGs were harvested from 4 week male WT mice . For DRG neuron culture medium , MEM was supplemented with 5% fetal bovine serum ( Gibco ) , 2X penicillin and streptomycin solution ( Gibco ) , 1X GlutaMAX ( Thermo Fisher ) , 20 μM 5-fluoro-2-deoxyuridine ( Sigma-Aldrich ) and 20 μM uridine ( Sigma-Aldrich ) . DRG neurons were digested and dissociated with 1 mg/ml collagenase D ( Roche ) for 90 min and then 1X TrypLE Express solution ( Thermo Fisher ) for 15 min . The dissociated DRG neurons were placed on a precoated dish with 100 μg / ml poly-D-lysine ( thermal fisher ) and 10 μg / ml laminin ( thermal fisher ) . 100 ng / ml Nerve growth factor ( R and D ) was applied to maintain the neuronal activity . After 24 hr incubation , PGE2 ( 1 μM ) or PBS were applied to stimulate the DRG neurons . In vitro RNA interference was performed using commercially available RNAi products from Thermal Scientific ( s72365 , s72370 , s72373 , and s72375 ) and the protocol was followed by the manufacture’s instruction . Briefly , after neuron seeding for 24 hr , media were replaced for the cells to be prepared for transfection . Lipofectamine RNAi MAX ( 13778100 , Invitrogen ) was diluted in OptiMEM ( 31985062 , Thermal Fisher ) and incubated for 5 min , then mixed with siRNAs or scramble control RNAs for five mins . Diluted DNA and Lipofectamine RNAi MAX were mixed and incubated at room temperature for 20 min and then used to transfect the DRG neurons . The medium was replaced 10 hr following transfection and neurons were harvested 24 hr after transfection . Similarly , the human GFP labeled MSC was purchased from Cyagen and is cultured in MEM with 10% fetal bovine serum ( Gibco ) , 1X penicillin and streptomycin solution ( Gibco ) . We used Pirt-GCaMP3f/- mice in DRG imaging . In order to monitor the activity of large populations of DRG neurons in intact live animals , Dr Xinzhong Dong’s Lab developed an in vivo imaging technique by using PirtGCaMP3 genetically engineered mice , in which the genetic-encoded Ca2+indicator GCaMP3 is specifically expressed in >95% of all DRG neurons by Pirt promoter ( Kim et al . , 2008; Kim et al . , 2014b ) . After surgical exposure of ipsilateral L4 DRG , in vivo imaging was immediately performed . Similarly as previously described ( Miller et al . , 2018 ) , the animals were maintained under inhalation anesthesia with assisted ventilation through endotracheal incubation . A laser scanning confocal microscope ( Leica LSI microscope system ) with a water immersed lens was used to capture the fluorescent signals . Live images were acquired at 10 frames with 600 Hz in frame-scan mode per 6–7 s , at depths below the dura ranging from 0 to 70 µm . 25 g of direct compression was applied to the ipsilateral knee after ACLT or sham surgery using a rodent pincher ( IITC Life Science ) to stimulate DRG neuronal firing . The duration of the mechanical force application maintained 15–30 s after 40–50 s of baseline imaging and the activated neuron number was counted and analyzed . Whole-cell current-clamp recordings were performed to perform the action potential of DR neurons according to the previous study ( Bierhaus et al . , 2012 ) . Only small and medium-sized DRG neurons with a resting membrane potential more negative than −40 mV were recorded . The extracellular solution contained ( in mM ) : NaCl 140 , KCl 4 , CaCl2 2 , MgCl2 1; HEPES 10 , NaOH 4 . 55 , glucose 5 ( pH 7 . 4 , 300–310 mOsm/kg H2O ) . The pipette solution contained ( in mM ) : KCl 135 , MgCl2 0 . 1 , Mg-ATP 1 . 6 , HEPES 10 , EGTA 2 , ( pH 7 . 3 at 25°C , adjusted with NaOH ) . The voltage was firstly clamped at −60 mV . For action potential stimulation , the frequency is by 2 × and 3 × rheobase and ramp current stimulation ( 0 . 1 , 0 . 3 , 0 . 5 , and 1 . 0 nA/sec ramp current ) . To measure the TTX resistant NaV1 . 8 currents in DRG neurons , the voltage-clamp technique was used . For recordings on DRG neurons , the extracellular solution contained ( mM ) : NaCl 60 , KCl 3 , Choline-Cl 80 , CaCl2 0 . 1 , MgCl2 0 . 1 , HEPES 10 , tetraethylammonium chloride 10 , glucose 10 and CdCl2 0 . 1 ( pH adjusted to 7 . 4 , 300–310 mOsm/kg H2O ) TTX ( 1 uM ) and TC-N 1572 ( 1 . 6uM ) were applied to the solution to block TTX sensitive sodium current and NaV1 . 9 currents . The pipette solution contained ( mM ) CsF 140 , EGTA 5 , MgCl 1 , and HEPES 10 , glucose 10 ( pH 7 . 4 , osmolarity 285–295 mOsm/kg H2O ) . Only cells with an initial seal >1 GΩ were recorded . The NaV1 . 8 currents were recorded responding to potential from –70 to +50 mV in 10 mV increments . The maximal current densities ( pA/pF ) were calculated and analyzed . Electronic Von Frey hair algesiometer ( IITC Life Science ) was used to measure the hind paw withdrawal threshold . Before starting the test , mice were separately placed in elevated Plexiglas chambers on metal mesh flooring for 30 mins . A von Frey hair with bending force ( 0 . 6 g , 1 g , 1 . 4 g , 2 g , 4 g ) was exerted perpendicular to the plantar surface of the hind paw until it just bent and the hind paw of mice of elevated . The force displayed on the electronic device were recorded . The threshold force required to elicit withdrawal of the paw was determined three times on each hind paw and averaged . Gait analysis was performed on mice 4 weeks after ACLT by the CatWalk system ( Noldus ) according to our previous protocol ( Zhen et al . , 2013 ) . Briefly , each mouse was placed walkway and allowed to allow the free movement from one side to the other side for at least three times . Mice were trained previously in the formal experiment . After the recording of mouse gait , several parameters were generated , and 5 of the most relevant parameters to OA pain were analyzed . ( 1 ) stands , ( 2 ) maximal contact at . ( 3 ) maximal ( 4 ) swing and ( 5 ) single stance . Western blotting was performed on the lysates of DRG neuron culture and tibial subchondral bone marrow . The samples were separated by SDS-PAGE gel and transferred onto a nitrocellulose membrane ( Bio-Rad Laboratories ) . After incubation with specific primary and secondary antibodies , signals were detected by an enhanced chemiluminescence kit ( Amersham Biosciences ) . The primary antibodies used are as follow rabbit anti-NaV1 . 8 ( 1:500 , ASC-016 , Alomone ) , rabbit anti-CREB ( 1:2000 , #9179 , Cell Signaling Technology ) , rabbit anti-pCREB ( 1:1000 , 9198 , Cell Signaling Technology ) , rabbit anti-PKA c- c-α ( 1:1000 , 4782 , Cell Signaling Technology ) and rabbit anti-GAPDH ( 1:1000 , 5174 , Cell Signaling Technology ) . The experiments were repeated three times and a representative film was selected . To measure the concentration of PGE2 in the subchondral bone marrow of mice tibiae , a PGE2 ELISA kit ( 514010 , Cayman ) was used according to the manufacturer’s manual . Briefly , we harvest the subchondral bone and then homogenized by ultrasound . The supernatant was aspired after high-speed centrifugation ( 13 , 200 g ) for 10 mins . The concentration of PGE2 was normalized by total protein concentration using the BCA assay . The ChIP assay was carried out using the epiquik ChIP Kit ( Epigentek catalog number: P-2002–1 ) . Briefly , the cultured lumbar DRG cells were crosslinked with 1% formaldehyde at for 10 min . After the collection of the cell , the sonication was performed until the DNA was broken into fragments with a mean length of 200 bps – 500 bps . The samples were subjected to immunoprecipitation with 2 mg of rabbit antibodies against pCreb1 ( CST , 1:50 ) for 90 min at room temperature and 10% of the sample for immunoprecipitation was used as an input ( a positive control ) . After purification , the DNA fragments were amplified using qRT-PCR with the primers for NaV1 . 8 promoter listed in Supplementary Table 2 . Retrograde tracing was performed at 3-month-old male SD rats ( Charles River Laboratories ) ( 300–400 g , n = 6 per group ) 2 months after ACLT . According to the previous study ( Ferreira-Gomes et al . , 2010 ) , a 20 mm parapatellar incision was made over the medial side of the left knee . Ipsilateral femoral and tibial subchondral bone were subject to retrograde labeling . We injected 2 μl DiI ( Molecular Probes; with 5 mg/ml in N , N dimethylformamide ) into the femoral and tibial subchondral bone areas using a Hamilton syringe with a 27-gauge needle . Immediately after injection , bone wax was used to seal the drilling holes to prevent tracer leakage . Animals were euthanized 2 weeks after retrograde injection and the left lumbar DRGs ( L3-5 ) were isolated for fluorescence detection . Twenty sections from each DRG were used for statistical analysis . Data are presented as means ± standard deviations . Error bars represent standard deviations . We used unpaired or paired two-tailed Student’s t-tests for comparisons between two groups , and one-way ANOVA with Bonferroni post hoc test for multiple comparisons , in comparison between three or more groups , two-way ANOVA with Bonferroni post hoc test were used . All data demonstrated a normal distribution and similar variation between groups . For all experiments , p<0 . 05 was considered to be significant .	Many people will suffer from joint pain as they age , particularly in their knees . The most common cause of this pain is osteoarthritis , a disease that affects a tissue inside joints called cartilage . In a healthy knee , cartilage acts as a shock absorber . It cushions the ends of bones and enables them to move smoothly against one another . But in osteoarthritis , cartilage gradually wears away . As a result , the bones within a joint rub against each other whenever a person moves . This makes activities such as running or climbing stairs painful . But how does this pain arise ? Previous work has implicated cells called osteoblasts . Osteoblasts are found in the area of the bone just below the cartilage . They produce new bone tissue throughout our lives , enabling our bones to regenerate and repair . Each time we move , forces acting on the knee joint activate osteoblasts . The cells respond by releasing a key molecule called PGE2 , which is a factor in pain pathways . The joints of people with osteoarthritis produce too much PGE2 . But exactly how this leads to increased pain sensation has been unclear . Zhu et al . now complete this story by working out how PGE2 triggers pain . Experiments in mice reveal that PGE2 irritates the nerve fibers that carry pain signals from the knee joint to the brain . It does this by activating a channel protein called Nav1 . 8 , which allows sodium ions through the membranes of those nerve fibers . Zhu et al . show that , in a mouse model of osteoarthritis , Nav1 . 8 opens too widely in response to binding of PGE2 , so the nerve cells become overactive and transmit a stronger pain sensation . This means that even small movements cause intense pain signals to travel from the joints to the brain . Building on their findings , Zhu et al . developed a drug that acts directly on bone to reduce PGE2 production , and show that this drug reduces pain in mice with osteoarthritis . At present , there are no treatments that reverse the damage that occurs during osteoarthritis , but further testing will determine whether this new drug could one day relieve joint pain in patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "neuroscience" ]	2020	Aberrant subchondral osteoblastic metabolism modifies NaV1.8 for osteoarthritis
PRDM9 binding localizes almost all meiotic recombination sites in humans and mice . However , most PRDM9-bound loci do not become recombination hotspots . To explore factors that affect binding and subsequent recombination outcomes , we mapped human PRDM9 binding sites in a transfected human cell line and measured PRDM9-induced histone modifications . These data reveal varied DNA-binding modalities of PRDM9 . We also find that human PRDM9 frequently binds promoters , despite their low recombination rates , and it can activate expression of a small number of genes including CTCFL and VCX . Furthermore , we identify specific sequence motifs that predict consistent , localized meiotic recombination suppression around a subset of PRDM9 binding sites . These motifs strongly associate with KRAB-ZNF protein binding , TRIM28 recruitment , and specific histone modifications . Finally , we demonstrate that , in addition to binding DNA , PRDM9's zinc fingers also mediate its multimerization , and we show that a pair of highly diverged alleles preferentially form homo-multimers . In humans and mice , PRDM9 determines the locations of meiotic recombination hotspots ( Baudat et al . , 2010; Myers et al . , 2010; Parvanov et al . , 2010 ) . PRDM9 is expressed early in meiotic prophase ( Sun et al . , 2015 ) , during which its C2H2 Zinc-Finger ( ZF ) domain binds DNA at particular motifs and its PR/SET domain trimethylates surrounding histone H3 proteins at lysine 4 ( H3K4me3; Hayashi et al . , 2005 ) and at lysine 36 ( H3K36me3; Wu et al . , 2013; Eram et al . , 2014; Powers et al . , 2016; Davies et al . , 2016; Grey et al . , 2017; Yamada et al . , 2017 ) . At a subset of PRDM9 binding sites , SPO11 is recruited to form Double Strand Breaks ( DSBs ) ( Neale and Keeney , 2006; Smagulova et al . , 2011 ) . These DSBs undergo end resection and the resulting single-stranded DNA ends are decorated with the meiosis-specific protein DMC1 ( Neale and Keeney , 2006 ) . In vivo experiments to date have mapped the locations of intermediate events in recombination by performing Chromatin ImmunoPrecipitation with high-throughput sequencing ( ChIP-seq ) against the H3K4me3 mark and the DMC1 mark in testis tissue from mice and humans ( Baker et al . , 2014; Smagulova et al . , 2011; Brick et al . , 2012; Pratto et al . , 2014; Davies et al . , 2016 ) , or by sequencing DNA fragments that remain attached to SPO11 after DSB formation in mice ( Lange et al . , 2016 ) . Recent studies have also published direct PRDM9 ChIP-seq results using a custom antibody in mouse testes ( Baker et al . , 2015a; Walker et al . , 2015; Grey et al . , 2017 ) . To study the DNA-binding properties of mouse PRDM9 , one study sequenced genomic DNA fragments bound in vitro by recombinant proteins containing only the PRDM9 ZF array ( Walker et al . , 2015 ) . In humans , recombination hotspots identified by DMC1 mapping and by Linkage Disequilibrium ( LD ) mapping have enabled the discovery of human PRDM9 binding motifs ( Myers et al . , 2008 , 2010; Hinch et al . , 2011; Pratto et al . , 2014; Davies et al . , 2016 ) . However , these published motifs are neither sufficient nor necessary to predict genome-wide PRDM9 binding , DSB formation , or recombination events ( Myers et al . , 2010; Pratto et al . , 2014 ) , and it has been suggested that binding might be influenced by chromatin features in cis ( Walker et al . , 2015 ) . Moreover , not all PRDM9 binding sites become hotspots ( Baker et al . , 2014; Grey et al . , 2017 ) , and the reasons for this remain unclear . In particular , apart from PRDM9 motifs themselves , there are no specific DNA sequence features that have been shown to modulate recombination rate in cis in mammals . The H3K4me3 mark has been associated with meiotic recombination initiation in budding yeast ( Borde et al . , 2009 ) , which lack PRDM9 , as well as in PRDM9 knockout mice ( Brick et al . , 2012 ) . Recent work has suggested that this histone mark is bound by CXXC1 , a protein that also binds to PRDM9’s KRAB domain and to the axis-associated protein IHO1 ( Imai et al . , 2017 ) . Because the H3K4me3 mark is also found at active gene promoters ( Santos-Rosa et al . , 2002 ) , PRDM9 has been hypothesized to play a role in meiotic gene regulation , in addition to its role in initiating recombination ( Hayashi et al . , 2005; Mihola et al . , 2009 ) . In fact , PRDM9 was shown to activate transcription in a reporter gene assay ( Hayashi et al . , 2005 ) , and its SET domain has been shown to de-repress a subset of genes when tethered to their promoters ( Cano-Rodriguez et al . , 2016 ) . However , recent experiments demonstrate full fertility in transgenic mice with completely remodeled PRDM9 binding landscapes ( Baker et al . , 2014; Davies et al . , 2016 ) , suggesting that PRDM9 has no essential role in gene activation . This does not preclude the possibility that PRDM9 may play a secondary gene regulatory role in meiosis . PRDM9 has also been shown to bind to itself and form multimers in transfected cells , while maintaining its ability to bind DNA and trimethylate histones ( Baker et al . , 2015b ) . However , it is not known which domains of PRDM9 mediate this multimer formation activity nor whether PRDM9 allelic variation impacts multimerization . To investigate the properties of PRDM9’s zinc-fingers in humans as they relate to the questions posed above , we expressed several engineered versions of PRDM9 in a mitotic human cell line ( HEK293T ) , then performed various high-throughput sequencing experiments . While this approach cannot reproduce cell-type-specific phenomena found only in spermatocytes and oocytes , it nevertheless enables us to infer some of the fundamental rules governing the behavior of PRDM9 in the nucleus . Indeed , as we describe below , this system replicates many of the key properties of PRDM9 binding in vivo . In these cells , we performed ChIP-seq against human PRDM9 , H3K4me3 , H3K36me3 , and chimp PRDM9 , as well as ATAC-seq ( Assay for Transposase-Accessible Chromatin with high-throughput sequencing ) to examine nucleosome positioning and DNA accessibility , and RNA-seq to examine gene expression . Importantly , by comparing data from transfected and untransfected cells ( in which there is weak endogenous PRDM9 expression ) , we can observe the same genomic sites with and without the effects of PRDM9 overexpression . This approach also allows us to rapidly engineer and test various different alleles and truncations of PRDM9 to explore the properties of its individual domains . Further , our results are complemented by previously published data on LD-based recombination hotspots ( Frazer et al . , 2007 ) , DSB hotspots decorated by DMC1 ( Pratto et al . , 2014 ) , H3K4me3 in human testes ( Pratto et al . , 2014 ) , and histone modifications across human cell types ( Kundaje et al . , 2015 ) , which we jointly analyze to understand the regulation of recombination outcomes downstream of PRDM9 binding . As described below , our results implicate a widespread role for other zinc-finger genes in suppressing , rather than activating , meiotic recombination in humans . We performed ChIP-seq in HEK293T cells transfected with the human PRDM9 reference allele ( the ‘B’ allele ) containing an N-terminal YFP tag that was targeted for immunoprecipitation . To identify regions bound by PRDM9 , we modeled binding enrichment relative to a measure of local background coverage at each position in the genome ( detailed in Appendix 1 ) , which accounts for local differences in sequencing coverage , including differences attributable to the known aneuploidy of this cell line ( Graham et al . , 1977; Bylund et al . , 2004; Lin et al . , 2014 ) . This yielded 170 , 198 PRDM9 binding peaks across the genome ( p<10−6 ) , demonstrating that PRDM9 can bind with some affinity to many sites outside of recombination hotspots , which number in the tens of thousands ( Myers et al . , 2005; Pratto et al . , 2014 ) . This large number of peaks likely results from the high expression level of PRDM9 in this system , providing sensitivity to detect even weak binding interactions , although it may be attributable in part to the chromatin organization of this cell type . We compared our ChIP-seq data with a set of 18 , 343 published in vivo human DSB hotspot peaks from DMC1 ChIP-seq experiments in testis samples ( Pratto et al . , 2014 ) . We found evidence for binding at 74% of DSB hotspots ( at p<10−3 ) after correcting for chance overlaps ( see Materials and methods ) . The proportion bound in our system is greater ( up to 82% ) at DSB hotspots >15 Mb from telomeres , which show elevated recombination rates in human males ( Dib et al . , 1996; Pratto et al . , 2014; Figure 1—figure supplement 1a ) . Overlap probabilities increase with both PRDM9 binding strength and DMC1 heat ( Figure 1b; Figure 1—figure supplement 1b ) . Furthermore , at PRDM9 binding sites , we observed peaks in LD-based recombination rates ( HapMap CEU map , Frazer et al . , 2007 ) , which increase with PRDM9 binding strength ( Figure 1c–d ) , as does DMC1 enrichment ( Figure 1—figure supplement 2c ) . Therefore , despite cell-type differences between our HEK293T expression system and the chromatin environment of early meiotic cells , our binding peaks capture the majority of biologically relevant recombination hotspots and reveal many additional non-hotspot sites bound by PRDM9 in HEK293T cells . Next , we leveraged the large number and high resolution of our ChIP-seq peaks to search for sequence motifs at PRDM9 binding sites using a Bayesian de novo motif-finding algorithm ( described in Davies et al . , 2016 and in Materials and methods ) . Rather than yielding a single motif described by a position weight matrix ( PWM ) , this algorithm allows binding sites to be described by a mixture of multiple motifs enriched in peak centers . The algorithm identified seven non-degenerate motifs , representing distinct PRDM9 binding modes . These explain 75% of the strongest 1000 binding peaks , falling to 53% of all peaks ( Figure 1a ) . The remaining peaks contain mostly degenerate , GC-rich sequences ( Figure 1—figure supplement 3 ) , similar to DMC1 hotspots in transgenic mice containing this same human PRDM9 allele ( Davies et al . , 2016 ) and interpretable as binding to clusters of individually weaker motif matches in mostly GC-rich regions . While each of the seven motifs has a close internal match to the published 13-mer found in human recombination hotspots ( Myers et al . , 2008 ) , allowing for multiple binding modalities revealed that the zinc fingers predicted to bind upstream of this 13-mer ( ZFs 1–6 ) can show comparably high sequence specificity ( Figure 1a ) . We aligned our seven motifs to each other and to an in-silico motif prediction ( based on the zinc-finger domain’s amino acid sequence alone; Myers et al . , 2010; Persikov et al . , 2009; Persikov and Singh , 2014 ) , revealing differences across motifs driven mainly by variable internal spacings ( Figure 1a ) alongside smaller differences in base-pair preferences ( e . g . Motif 5 ) . The region corresponding to ZF5 and ZF6 is predicted to span 6 bp , but in Motifs 4–7 this region spans only 2 bp , and in Motif 1 it spans only 5 bp . Interestingly , we only observed these three particular spacings , and the expected 6 bp binding footprint is observed only for Motifs 2 and 3 , which explain a relatively small proportion of peaks ( 6% ) . This alternative spacing cannot be captured in a single motif , possibly explaining why ZFs 1–6 have shown weak sequence specificity in previously published hotspot motifs ( Myers et al . , 2008 , 2010; Hinch et al . , 2011; Pratto et al . , 2014 ) . Alternative spacing within motifs could explain how long zinc-finger arrays like PRDM9’s are able to consecutively bind DNA despite theoretical physical constraints ( Persikov and Singh , 2011 ) , similar to multivalent CTCF binding ( Nakahashi et al . , 2013 ) . Our results are also consistent with recent findings that truncated mouse PRDM9 alleles can stably bind discontinuous submotifs , though at reduced specificities , with subsets of zinc fingers ( Striedner et al . , 2017 ) . ZF5 and ZF6 , which overlap the variably spaced region , have large , aromatic tryptophan residues at the DNA-contacting ‘−1’ position ( Figure 1a ) . They also lack the positively charged DNA-contacting residues found in the most sequence-specific zinc fingers in the array ( consistent with an electrostatic attraction to the negatively charged DNA ) . We speculate that these bulky , uncharged middle zinc fingers might fail to bind DNA strongly and may act more like a linker between the more strongly binding zinc fingers found upstream and downstream . Interestingly , we observed a lower mean LD-based recombination rate ( Frazer et al . , 2007 ) around Motif 7 peaks , not explained by differences in PRDM9 binding enrichment , promoter overlap , repeat overlap , or H3K4me3 enrichment ( Figure 1d , Figure 1—figure supplement 4 ) . We hypothesized that Motif 7 might be favorably bound by the B allele and thus underrepresented in LD-based recombination maps , which are dominated by historical recombination events initiated by the more common A allele of PRDM9 , which differs at a single DNA-contacting amino acid in ZF5 ( Baudat et al . , 2010 ) . To test this hypothesis , we searched for our seven motifs in DSB hotspots unique to an individual with an A/B PRDM9 genotype , then compared these to DSB hotspots found in homozygous A/A individuals ( Pratto et al . , 2014 ) . We found that Motif 7 is two-fold enriched in A/B-only hotspots relative to A/A hotspots , while all other motifs are found in more similar proportions between the two sets ( Figure 1d ) . Motif 7 also resembles , but extends , a motif previously identified in A/B-only hotspots ( Pratto et al . , 2014 ) . We conclude that the B allele must bind Motif 7 with greater affinity than does the A allele , demonstrating distinguishable binding preferences between these highly similar PRDM9 alleles . We investigated the histone methylation activity of PRDM9 by performing ChIP-seq against the H3K4me3 mark in transfected and untransfected cells . After subtracting sites overlapping ‘pre-existing’ H3K4me3 peaks ( those present in untransfected cells ) , we found that 95% of PRDM9 binding peaks show H3K4me3 following transfection ( p<0 . 01 ) , and this proportion increases to 100% with increasing PRDM9 binding enrichment ( see Figure 1b ) . That is , PRDM9 makes the H3K4me3 mark essentially everywhere it binds , regardless of the pre-existing chromatin substrate , with H3K4me3 signal strength increasing with PRDM9 binding strength ( r=0 . 48 , Figure 1—figure supplement 1c , Figure 1—figure supplement 2 ) . As observed in mice ( Davies et al . , 2016; Powers et al . , 2016; Grey et al . , 2017 ) , we also observe localized H3K36me3 deposition at bound sites ( see Figure 1—figure supplement 1d ) . Apart from depositing H3K4me3/H3K36me3 locally around its binding sites , PRDM9 has been shown to phase surrounding nucleosomes in vivo in mice ( Baker et al . , 2014 ) . To investigate this behavior in transfected HEK293T cells , we performed ATAC-seq and found that full-length PRDM9 appears to phase surrounding nucleosomes even in this completely different cell type and expression system ( see Figure 2—figure supplement 3a ) . However , when we transfected a truncated version of PRDM9 including only the zinc-finger domain , we saw no evidence of nucleosome phasing around PRDM9 binding sites ( see Figure 2—figure supplement 3b ) . Instead , its ATAC-seq coverage pattern appears similar to that of unstransfected cells or of cells transfected with a truncated version of PRDM9 excluding the zinc-finger domain ( Figure 2—figure supplement 3c , d ) . We confirmed that this ‘ZF only’ truncated protein localizes to the nucleus ( see Figure 2—figure supplement 4 ) , and previous studies have shown that PRDM9’s ZF array is sufficient to bind DNA ( Walker et al . , 2015; Striedner et al . , 2017 ) . This suggests that PRDM9’s nucleosome phasing behavior stems not only from the binding of its ZF array to DNA , but may involve steric effects of the non-ZF region or require histone methylation . A study in mice has shown that , in the absence of PRDM9 , DSBs localize to active promoters marked with H3K4me3 , suggesting that PRDM9 may serve to provide alternative H3K4me3 sites to compete with and direct recombination away from promoters ( Brick et al . , 2012 ) . However , our ChIP-seq data revealed that , surprisingly , of the 12 , 982 protein-coding genes with H3K4me3 surrounding their Transcription Start Site ( TSS ) in our untransfected cells ( p<10−5 ) , 81% have a PRDM9 binding peak center within 500 bp of the TSS , compared to only 6% expected by chance overlap ( yielding a corrected overlap fraction of 79% ) . At promoters with little or no prior H3K4me3 , the proportion bound by PRDM9 decreases to 15% ( corrected for chance overlaps , Figure 2a ) , though this difference could potentially be explained by increasing power to detect weak binding events at more active genes . If we concentrate only on the strongest quartile of PRDM9 binding enrichment at promoters , we see that roughly 10% of promoters are strongly bound , regardless of H3K4me3 enrichment ( Figure 2a ) . Previous datasets in humans have been unable to detect this affinity for promoters because they relied on H3K4me3 , DMC1 , or LD mapping as proxies for inferring PRDM9 binding sites ( Baker et al . , 2015b; Pratto et al . , 2014; Myers et al . , 2010 ) . Since active promoters contain PRDM9-independent H3K4me3 peaks , they are filtered out from H3K4me3 analyses , and since DSBs are suppressed at promoters ( at least in the presence of PRDM9 , as shown by Brick et al . , 2012 ) , promoters are underrepresented in DMC1 and LD-based recombination hotspots . One recent study mapped binding of the human PRDM9 B allele in HEK293T cells by ChIP-exo , yielding a conservative set of 839 peaks after stringent filtering ( Imbeault et al . , 2017 ) . Of these 839 peaks , 87% overlap our 170 , 198 peaks , and they are similarly enriched in promoters ( 18% occur within 500 bp of a TSS , versus 6% when shifted 5 kb , compared to 15% and 7% with our peaks , respectively ) . To exclude the possibility that PRDM9 binding peaks observed at promoters were false positives ( Jain et al . , 2015 ) , we performed two ChIP-seq replicates on cells transfected with a PRDM9 construct in which we replaced the human ZF domain with the ZF domain from the chimpanzee w11a allele , which is not predicted to bind the GC-rich DNA commonly found at promoters ( Auton et al . , 2012; Schwartz et al . , 2014 ) . We found that the chimp allele binds a T-rich motif ( Figure 2—figure supplement 1c ) , and only 5% of chimp PRDM9 peaks occur within 500 bp of a human peak center , below the 8% expected by chance ( Figure 2—figure supplement 1b ) . In contrast to results for human PRDM9 , only 3% of promoters fall within 500 bp of a chimp PRDM9 peak , versus 9% expected by chance overlap , confirming that the promoter peaks we observe for the human allele are unlikely to be ChIP-seq artifacts . Furthermore , motif identification at human PRDM9’s promoter binding sites identified the expected binding motifs at similar frequencies to non-promoter peaks , except for a twofold enrichment of Motif 7 ( Figure 2b ) . Interestingly , Motif 7 is also the B-allele-enriched motif , so PRDM9’s promoter affinity might also differ between common human alleles . We suggest that these GC-rich motifs , together with accessible chromatin , enable human PRDM9 to consistently bind to promoter regions in HEK293T cells ( Figure 2a ) . Notably , however , PRDM9 peaks in promoters tend to have lower mean enrichment estimates across a range of motif FIMO scores ( Figure 2—figure supplement 2d ) . It is also worth noting that in vivo mapping of PRDM9 binding will be required to confirm that promoter binding occurs in meiotic cells , although it is difficult to understand how this sequence-dependent binding could be cell-type-specific across all promoters . Although there is widespread binding of human PRDM9 to promoters in HEK293T cells , we observe little to no elevation in local recombination rate or testis DMC1 enrichment at these binding sites ( Figure 2c , d , Figure 2—figure supplement 2e , f ) . In the absence of PRDM9 , DSBs localize to promoters in mice ( Brick et al . , 2012 ) , but in light of our results , it remains difficult to explain how recombination might be suppressed at promoters despite direct PRDM9 binding . A second mark , H3K36me3 , is also deposited by PRDM9 at many of its binding sites in vivo ( Powers et al . , 2016 ) , and it shows a similar pattern to H3K4me3 around DSB sites in mice ( Yamada et al . , 2017 ) . At both non-promoter and promoter PRDM9 peaks , we observed a similar enrichment of H3K36me3 in transfected relative to untransfected cells ( Figure 2e , f ) , confirming that PRDM9 indeed binds these sites . However , a very strong depletion of H3K36me3 around promoters in untransfected cells means that absolute levels of H3K36me3 remain low in promoters , relative to non-promoter binding sites ( Figure 2g ) . Interestingly , the amount of H3K36me3 deposited by PRDM9 at promoters negatively correlates with the amount of H3K4me3 enrichment at those promoters in untransfected cells , and this cannot be explained by differential PRDM9 binding ( Figure 2h ) . This suggests that at highly active promoters , PRDM9 is less able to deposit H3K36me3 , or this mark is actively removed . This difference between promoter and non-promoter binding sites could in principle explain the lack of recombination at promoters , if the simultaneous presence of both H3K36me3 and H3K4me3 influences recombination initiation , as has been suggested by Powers et al . ( 2016 ) and shown to be consistent with DSB data by Yamada et al . ( 2017 ) . In humanized mice , in vivo DSB hotspot sites favor motif positions with lower PRDM9-independent H3K4me3 levels than genomic background ( Davies et al . , 2016 ) , and this seems highly concordant with our human results . We have shown that human PRDM9 binds promoters and deposits the H3K4me3 mark wherever it binds in HEK293T cells , which raises the possibility that PRDM9 may affect gene expression , given that H3K4me3 is highly enriched at active promoters ( Santos-Rosa et al . , 2002 ) . Tethering PRDM9’s SET domain to other promoter-binding proteins has been shown to de-repress gene expression in a context-dependent manner ( Cano-Rodriguez et al . , 2016 ) , leading us to hypothesize that full-length human PRDM9 might also be able to activate gene expression . We therefore performed RNA-seq in cells transfected with human PRDM9 , along with control samples that were either untransfected , transfected with the chimp allele , or transfected with a construct containing only the human zinc-finger domain ( and incapable of H3K4me3 deposition; referred to as ‘ZF only’; all constructs illustrated in Figure 5a ) . Seven transcripts showed overwhelming evidence of being differentially expressed in cells transfected with the human allele versus all other samples , with all seven being upregulated by PRDM9 presence . Five overlap known genes: MEG3 , ONECUT3 , LGALS1 , VCX , and CTCFL . Interestingly , the latter two genes are normally expressed only in spermatogenesis ( Lahn and Page , 2000; Sleutels et al . , 2012 ) . We validated expression induction at these two genes using qPCR ( Figure 3 ) . CTCFL is a variant of chromatin regulator CTCF , and in mice it has been shown to be expressed exclusively in pre-leptotene spermatocytes ( Sleutels et al . , 2012 ) . Male knockout mice show greatly reduced fertility due to meiotic arrest ( Sleutels et al . , 2012 ) , and variants at CTCFL influence genome-wide recombination rates in human males ( Kong et al . , 2014 ) . CTCFL may be involved in organizing the meiotic chromatin landscape and regulating the transcription of meiotic genes ( Sleutels et al . , 2012 ) . We found that CTCFL RNA levels increase 28-fold after transfection with the human allele , from a nearly undetectable baseline transcription level ( Figure 3; we note this may underestimate the true relative expression level given that transfection efficiency is not 100% ) . PRDM9 binds strongly to a GC-rich repeat near the CTCFL TSS and deposits H3K4me3 , which is absent in untransfected cells ( Figure 3 ) . The chimp PRDM9 allele , in contrast , does not bind near the TSS and does not show elevated transcript levels after transfection ( Figure 3 ) . VCX encodes a small , highly charged protein of unknown function and has been previously studied for its involvement in PRDM9-related non-homologous recombination events and X-linked ichthyosis ( Myers et al . , 2008; Van Esch et al . , 2005 ) . We found that PRDM9 does not in fact bind near the annotated VCX TSS , but instead in the middle of the gene and very strongly at a minisatellite array of PRDM9 binding motifs ( Myers et al . , 2008 ) near the terminus of the gene ( Figure 3—figure supplement 1 ) . PRDM9 adds the H3K4me3 mark throughout the gene’s coding regions in a pattern similar to that seen in testes ( Figure 3—figure supplement 1 ) . RNA-seq coverage suggests normal splicing , but use of an alternative promoter that excludes the first , untranslated exon ( Figure 3—figure supplement 1 ) . We note that this result does not establish whether human PRDM9 is necessary or sufficient for CTCFL and VCX expression in vivo , but still PRDM9 is demonstrably able to trigger the transcription of these genes in a way that depends on the binding of its zinc fingers . Previous work has shown that Prdm9 expression begins in pre-leptotene cells in mice ( Sun et al . , 2015 ) , concurrent with Ctcfl expression ( Sleutels et al . , 2012 ) and thus supports the possibility that PRDM9 may promote CTCFL transcription in vivo . The failure of the chimp allele to bind to or activate the expression of human CTCFL further suggests that this behavior may not be essential across organisms , although the chimp allele might in principle still bind the CTCFL promoter in the chimp genome . Similarly , there is no evidence that human PRDM9 alleles with very different binding preferences , such as the C allele , would bind the same promoter . Also notably , the motif bound at the CTCFL promoter is Motif 7 , so the A and B alleles may bind this locus with different affinities . 43 additional genes showed weaker evidence of being activated by human PRDM9 binding near their annotated transcription start sites , with 41 showing increases , as opposed to decreases , in expression ( Figure 3—source data 2 ) . We lack power to detect small changes in gene expression , especially decreases in expression ( Trapnell et al . , 2012 ) . Nonetheless , it is likely that effects of similar magnitude to CTCFL and VCX are quite rare . Our data do make it clear that PRDM9 binding and histone trimethylation near a promoter can trigger or enhance gene expression in some cases . Furthermore , this effect on gene expression is not likely to result from PRDM9 binding alone but from its trimethylation activity , given that transfection with the zinc fingers alone does not trigger expression . Further work will need to establish if promoter-binding PRDM9 alleles are able to regulate gene expression in vivo , whether as an accidental side effect of binding or specifically functional , though this work may remain challenging in humans . Although our seven motifs ( Figure 1a ) improve our understanding of PRDM9 binding , even the top-scoring 0 . 1% of motif matches genome-wide have only a 50% chance of overlapping an actual PRDM9 binding peak ( see Figure 2—figure supplement 2a ) . Moreover , at best we only observe a 55% correlation between H3K4me3 and DMC1 enrichment values from testis data surrounding our PRDM9 binding sites ( Figure 1—figure supplement 2f ) . Therefore , other influences such as wider sequence and chromatin contexts must impact both binding and downstream recombination outcomes . The only specific known mammalian sequence feature so far identified as influencing either PRDM9 binding , or downstream recombination events , is the PRDM9 binding motif itself . Thus , it is uncertain which factors prevent or promote hotspot occurrence , whether these act in cis or trans , and what these might be . A powerful approach to identify factors that might influence PRDM9 binding and subsequent hotspot formation is to search for sequence motifs predicting these outcomes . Identified motifs are likely to have a causal influence , so they can help address whether particular histone modifications associated with those motifs have a genuinely causal role themselves . We hypothesized that sequence motifs unrelated to PRDM9 binding might have strong local effects on recombination outcomes , but these motifs might evade detection if they operate only at a minority of recombination hotspots . To attempt to overcome this and control for the effects of local genetic context , we focused on hotspots centering within one family of retrotransposon elements , called THE1B repeats , which are the most strongly hotspot-enriched among all human repeats ( Myers et al . , 2008 ) . PRDM9 binds directly to a subset of THE1B repeat copies containing matches to its target motif ( Figure 4a ) , in a known region of the repeat ( Myers et al . , 2008 , see Appendix 2 ) , and THE1B-centered hotspots contribute a substantial fraction of all human A- and B-allele controlled recombination ( 4 . 6% measured by DMC1 mapping; Pratto et al . , 2014 ) . We analyzed over 20 , 000 THE1B repeats throughout the human genome , which share highly similar sequences perturbed by random mutations . These mutations allowed us to precisely dissect the impact of particular sequence motifs on PRDM9 binding , and on downstream DSB formation ( as measured by DMC1 mapping , from Pratto et al . , 2014 ) and crossover activity ( as measured by LD mapping , from Frazer et al . , 2007 ) . We used conditional association testing to identify collections of motifs that independently correlate with PRDM9 binding or recombination ( see Appendix 2 ) . Seventeen distinct motifs ( Figure 4a ) were found to influence PRDM9 binding to THE1B copies in HEK293T cells ( Figure 4—source data 1 ) . All map within the predicted PRDM9 binding region and span the entire region , confirming that all of PRDM9’s zinc fingers are involved in binding . Motifs promoting PRDM9 binding associated with higher H3K4me3 enrichment in testes ( data from Pratto et al . , 2014 ) and with increasing LD/DMC1 hotspot probability , so the same motifs must operate in vivo ( Figure 4a; detailed in Appendix 2 ) . Importantly for the results described below , binding of PRDM9 does not associate strongly with any sequence motifs outside the directly bound region , so it might act as a local ‘pioneer’ protein at least on this background , despite results in mice ( Grey et al . , 2017 ) . We then independently tested for the presence of motifs influencing recombination hotspot formation conditional on presence of a PRDM9 binding site in HEK293T cells . We identified an initial seven such motifs ( Figure 4a; detailed in Appendix 2; Figure 4—source data 1 ) . Only three of these map within the PRDM9 binding region and correspond to stronger/weaker PRDM9 enrichment . The remaining four motifs show no association whatsoever with PRDM9 binding in HEK293T cells , and map well outside the PRDM9 binding motif ( Figure 4a ) . We refer to these as ‘non-PRDM9 recombination-influencing motifs’ . The strongest signal is for the motif ATCCATG ( joint p=2 . 8×10-9 for LD-hotspots , OR = 0 . 32 ) , whose presence within a THE1B repeat produces a 2 . 5-fold reduction in the surrounding recombination rate at PRDM9-bound THE1B repeats ( Figure 4b ) . ATCCATG presence also reduces the local recombination rate around THE1B repeats not bound by PRDM9 , implying a more general , PRDM9-independent mechanism of recombination suppression ( Figure 4b ) . Notably , this suppression extends beyond the boundaries of the THE1B repeat itself . We observed strong testis H3K4me3 enrichment at THE1B repeats containing PRDM9 binding motifs regardless of whether ‘ATCCATG’ was present , and after conditioning on the strength of the PRDM9 motif match ( Figure 4b ) . Therefore , this motif must suppress recombination downstream of PRDM9 binding in vivo . In fact , presence of the modifier motif ATCCATG actually modestly increased the testis H3K4me3 signal , even at THE1B copies not containing a PRDM9 motif and not bound by PRDM9 in HEK293T cells ( Figure 4b ) , which we return to below . Similar results were observed for the other three non-PRDM9 recombination-influencing motifs . We hypothesized that the recombination-influencing motifs described above might be bound by chromatin-modifying proteins . To examine this possibility , we independently searched for motifs that could predict chromatin states within THE1B elements . Specifically , we searched de novo for motifs associated with 15 previously identified chromatin states , and individual histone modifications , across each of 125 somatic cell types ( Kundaje et al . , 2015 ) . Strikingly , we observed that the motif ATCCATG ( independently identified above as the strongest non-PRDM9 recombination-influencing motif ) is also the strongest single predictor of the ‘heterochromatin’ state , marked by enriched H3K9me3 . THE1B repeats containing ATCCATG are heterochromatin-enriched in over half of cell types , especially in embryonic stem cells , and exhibit a strong localized increase in H3K9me3 ( Figure 4c ) . More surprisingly , we also observed a weak , but significant , localized increase in H3K4me3 signal ( p=7 . 5×10−13; Figure 4c ) . We also saw the same weak H3K4me3 peak in testes , after restricting analysis to THE1B repeats not bound by PRDM9 ( Figure 4b , c ) , indicating this modification operates fully independently of PRDM9 . This weak increase might reflect genuine partial co-occurrence of H3K9me3 and H3K4me3 at the same locus ( but possibly on different alleles , or in different cells ) , or in theory it could be explained by non-specificity of experimental antibodies for these two histone modifications . We reasoned that we might more generally exploit the subtle H3K4me3 signal elevation ( whatever its underlying cause ) as a potential marker also of H3K9me3 elevation in germline tissues by examining H3K4me3 in testes ( Pratto et al . , 2014 ) . We performed de novo motif finding to identify PRDM9-independent 7-mers associated with testis H3K4me3 in THE1B repeats definitively not bound by PRDM9 ( detailed in Appendix 2 ) . This identified eighteen motifs significantly associated with non-PRDM9 H3K4me3 ( after Bonferroni correction , Figure 4a ) . The motif ATCCATG remained the most strongly associated ( p<10−25 ) , with eight other motifs clustered around it ( Figure 4a ) . Confirming that these motifs also predict H3K9me3 levels , we observed almost perfect positive correlation ( r = 0 . 93 ) between H3K4me3 signal strength in testes and H3K9me3 ( as well as H3K4me3 ) in particular ROADMAP ESC lines ( Figure 4—figure supplement 1c ) . Therefore , these 18 motifs predict both H3K9me3 and H3K4me3 , broadly observable across somatic cells and ( at least for the latter mark ) testes also , and so we refer to this set as ‘non-PRDM9 H3K9me3/H3K4me3 motifs . ’ In addition to the top-scoring motif , ATCCATG , many or all of the remaining 17 non-PRDM9 H3K9me3/H3K4me3 motifs evidently impact meiotic recombination ( Figure 4—source data 1; p<0 . 00036 for effect size correlation ) . All four of the non-PRDM9 recombination-influencing motifs we found overlap at least one of these 18 independently derived non-PRDM9 H3K9me3/H3K4me3 motifs ( Figure 4a; note that power differences account for the smaller size of the former motif set ) . Summing these 18 motif influences to produce a score for each THE1B repeat using only its DNA sequence , we see more than a threefold difference in the probability of observing a recombination hotspot across PRDM9-bound THE1B copies between the top and bottom 10% quantiles of the score ( Figure 4d ) . Given that we are only able to examine the region within each 1–2 kb recombination hotspot corresponding to the 354 bases of the THE1B element , this likely underestimates the true impact of local sequence on whether hotspots occur or not . Notably , our testing for association with other histone-defined chromatin states ( e . g . states enriched for H3K27me3 ) in ROADMAP-studied cell types identified many more sequence motifs . These included the known binding targets of two proteins , DUX4 and ZBTB33 , that were previously shown to bind to THE1B elements , with DUX4 showing strong expression in testes ( Young et al . , 2013; Wang et al . , 2012 ) . However , only those motifs associated with heterochromatin and H3K9me3/H3K4me3 overlapped our non-PRDM9 recombination-influencing motifs . Thus , only a particular subset of chromatin modifications correspond to suppressed recombination , in THE1B repeats at least . Overall , this analysis of thousands of human hotspots reveals that in cis , it is not simply PRDM9 binding that influences whether hotspots occur . Multiple sequence motifs exist that do not prevent PRDM9 binding , but instead modify the average amount of recombination that occurs downstream of binding , over two-fold for a single motif ( ATCCATG ) . Given this diversity even within THE1B-centered hotspots , completely different motifs might operate to modulate recombination activity in other hotspots , either centered in different repeats or in non-repeat DNA . In contrast to this complexity , examination of histone modifications reveals a common signature across recombination-influencing motifs , with strong alterations in the specific histone mark H3K9me3 and weaker signals for H3K4me3 . This suggests that the mechanism of action across motifs might share fundamental similarities . Both H3K4me3 and H3K9me3 marks correlate negatively with recombination across all human hotspots ( Figure 4d; Figure 4—figure supplement 1b ) , and reduced levels of non-PRDM9 H3K4me3 within hotspots has been observed in mice ( Brick et al . , 2012; Davies et al . , 2016 ) . The large class of human KRAB-ZNF genes represent an obvious set of motif-binding candidates that might explain H3K9me3 deposition within THE1B repeats and more broadly . In many such genes , the KRAB domain recruits TRIM28 , which in turn recruits histone-modifying proteins including SETDB1 , which lead to H3K9me3 deposition on nearby nucleosomes ( Schultz et al . , 2002; Imbeault et al . , 2017 ) . We therefore examined recent data measuring genome-wide binding of 222 KRAB-ZNF proteins in humans , and sites where TRIM28 is present in embryonic stem cells , for overlap with THE1B repeats ( Imbeault et al . , 2017; Appendix 2 ) . Notably , although PRDM9 is a KRAB-ZNF protein , its KRAB domain does not interact with TRIM28 ( Imai et al . , 2017 ) . We identified three KRAB-ZNF proteins ( ZNF100 , ZNF430 and ZNF766 ) , as well as TRIM28 , that are enriched for binding in THE1B repeats and also associate genome-wide with H3K9me3 deposition . We identified binding motifs for each of these four proteins within THE1B repeats . Strikingly , ATCCATG overlapped the second most significant motif for TRIM28 recruitment , and additional motif analysis for TRIM28 revealed a large ( 51 bp ) motif , fully spanning a cluster of eight motifs associated with H3K9me3/H3K4me3 and recombination rate ( Figure 4a ) , and presumably representing the binding target of one or more KRAB-ZNF protein ( s ) whose binding targets have not yet been experimentally characterized . The three ZNF proteins also all bind sites overlapping those implicated in impacting H3K9me3/H3K4me3 and meiotic recombination , two in the same region as the TRIM28 motif , but with differing sequence specificity ( Figure 4a ) . Thus , while binding maps are not yet available for every human KRAB-ZNF protein , those that bind THE1B repeats consistently operate to reduce recombination , and TRIM28 recruitment can explain the strongest signals we see . Across all our PRDM9 binding peaks ( not only those in THE1B elements ) , 36 . 5% fall within 500 bp of a binding site of at least one of the KRAB-ZNF proteins with available data ( Imbeault et al . , 2017 ) , suggesting that such repression might be important in regulating recombination more generally . To test this , we individually analyzed the KRAB-ZNF proteins with at least 30 instances of a KRAB-ZNF binding peak occurring near a PRDM9 binding peak ( after excluding DNase HS regions and promoters , which are often bound by multiple different proteins ) , for their effect on whether a hotspot occurs at these PRDM9 binding peaks ( Appendix 2 ) . This revealed a universal negative trend ( Figure 4e ) typified by a twofold reduction in recombination locally at TRIM28-marked sites genome-wide , with every gene except one ( ZNF282 , which was non-significant ) inferred to reduce hotspot odds . Binding of almost all KRAB-ZNF genes tested correlated positively with H3K9me3 , and those genes with strongest H3K9me3 enrichment showed the strongest suppression of recombination locally ( Figure 4e ) . Together , our results indicate a mechanism of cis recombination repression affecting thousands of human PRDM9 binding sites . Binding of KRAB-ZNF proteins to specific sequence motifs within or nearby the PRDM9 binding site , followed by TRIM28 recruitment and H3K9me3 deposition , universally acts to strongly repress local recombination . Perhaps surprisingly , this can occur without preventing PRDM9 binding or H3K4me3 deposition . We suggest that this is the mechanism at play for the recombination-suppressing , H3K9me3-promoting ATCCATG motif , which we suspect is bound by a KRAB-ZNF protein whose binding sites have not yet been mapped . Many KRAB-ZNF genes bind to specific sets of retrotransposon repeats ( THE1B repeats represent one example ) , so this repressive mechanism is likely to act to reduce recombination around many particular repeats . Finally , we used our THE1B dataset to examine the relationship between PRDM9 binding and broad-scale recombination rates genome-wide while controlling for local genetic context . To do so , we partitioned THE1B repeats into quintiles of increasing recombination rate in the surrounding 1 Mb in males ( independently measured by Kong et al . , 2002 ) . We observed that DMC1 enrichment increases >10-fold with surrounding recombination rate across both telomeric and non-telomeric regions , but H3K4me3 enrichment in testes , a proxy for meiotic PRDM9 binding , shows no association whatsoever ( Figure 4—figure supplement 2 ) . Therefore , in broad ‘hotter’ regions , double-strand breaks and crossovers occur at much higher frequencies , completely independently of the local sequence ( which is similar in THE1B repeats genome-wide ) or the local level of PRDM9 binding . This proves that , at least in human males , megabase-scale recombination rates throughout the genome are not associated with PRDM9’s ability to bind and deposit H3K4me3 , consistent with previous observations in the specific case of elevated human male recombination in telomeres ( Pratto et al . , 2014 ) . Our results thus far have added to the already complex array of evolutionary forces buffeting PRDM9 , relating to its ability to influence gene expression or to the co-binding of other zinc-finger proteins near its binding sites . Another dimension of evolutionary constraint may arise from PRDM9’s ability to bind to itself and form functional multimers . Previous work has shown that PRDM9 as a whole can multimerize and that hetero-multimers of the human A and C alleles can bind the sequence targets of either allele and trimethylate surrounding histones ( Baker et al . , 2015b ) . However , it remains unknown which PRDM9 domain is responsible for this observed multimerization behavior . We sought to determine whether multimerization might involve PRDM9’s ZF domain in any way , given other examples of ZF domains mediating protein-protein interactions ( McCarty et al . , 2003; Lee et al . , 2007 ) . To do so , we co-expressed PRDM9 constructs with different ZF domain properties and performed co-ImmunoPrecipitation ( co-IP ) experiments , thus extending our study from PRDM9’s DNA-binding properties to its protein binding properties . First , to confirm the ability of the PRDM9 alleles we study here to form multimers ( Baker et al . , 2015b ) , we performed co-IP experiments with full-length human B-allele PRDM9 constructs differentially tagged with HA and V5 epitopes and co-transfected into HEK293T cells . Following IP against the HA-tagged construct , we detected the V5-tagged construct very robustly; and conversely ( Figure 5—figure supplement 1 ) . This is consistent with human PRDM9 binding strongly to itself , as demonstrated previously in HEK293 cells ( Baker et al . , 2015b ) . To narrow the PRDM9 domain ( s ) responsible for this self-binding behavior , we split the full-length human B-allele PRDM9 cDNA into two pieces: one containing only the C-terminal Zinc-Finger domain ( the ‘ZFonly’ construct ) , and one containing everything else ( the ‘noZF’ construct; illustrated in Figure 5a ) . We co-transfected these constructs and full-length PRDM9 into HEK293T cells in various combinations . The full-length human construct and the ZFonly construct localized to the nucleus , but the noZF construct localized throughout the cell , confirming a dominant role for the ZF domain in nuclear localization ( Figure 2—figure supplement 4 , Collin et al . , 2013; Wang et al . , 2014 ) . Interestingly , the ZF domain alone appears to be responsible for most of PRDM9’s self-binding activity ( Figure 5b ) . Following co-transfection of noZF-HA and noZF-V5 , and despite very high expression levels visible in the input , only a very faint co-IP band is visible in the absence of the ZF array . Because the mock control lane is clean ( Figure 5—figure supplement 2a ) , this band likely reflects a real but weak self-binding capability mediated by the non-ZF portion of PRDM9 ( though we cannot rule out a role for the ‘early zinc finger’ ) . In complete contrast , we saw an intense co-IP band when co-transfecting ZFonly-HA with ZFonly-V5 . Therefore , the zinc-finger domain of one PRDM9 protein can bind strongly to the zinc-finger domain of another , while the rest of the protein interacts more weakly . We confirmed this result by co-transfecting full-length , V5-tagged human PRDM9 with either noZF-HA or ZFonly-HA , revealing that the ZFonly construct is sufficient to bind and pull down the full-length construct . This finding replicated in a repeat experiment , and when reversing the direction of the IP-western experiment ( Figure 5—figure supplement 2b ) . No co-IP band is seen in a negative control experiment in which we co-transfected the noZF construct with the ZFonly construct ( Figure 5b ) , ruling out an interaction between the ZF domain and the rest of PRDM9 or any interaction between the epitope tags used . Our results remained unchanged following complete DNA digestion by benzonase in the ZFonly-ZFonly co-IP experiment ( Figure 5—figure supplement 3a ) , implying that DNA is not required for the observed interaction between ZF domains . Finally , to examine the specificity of ZF array binding , we replaced the final exon containing the human ZF array with a synthesized cDNA matching the final exon of the chimpanzee reference PRDM9 allele ( w11a ) containing 18 zinc fingers ( compared to 12 in the human allele , allowing us to resolve them as two distinct bands ) , and with different DNA-binding preferences . We refer to the resulting tagged constructs as Chimp-HA and Chimp-V5 ( Figure 5a ) . To test the relative efficiency of homo- versus hetero-multimerization , we performed direct competition experiments . We transfected cells with three constructs: for example , Chimp-V5 plus Chimp-HA plus Human-HA . In this case Chimp-V5 would be the ‘bait’ pulled down by IP with anti-V5 , and Chimp-HA and Human-HA would be the co-IP ‘prey’ detected by western blotting with anti-HA ( we replicated by reversing the tags ) . The results show that Chimp PRDM9 pulls down Chimp PRDM9 more than twofold more efficiently than it pulls down Human PRDM9 . Similarly , Human PRDM9 pulls down Human PRDM9 more than twofold more efficiently than it pulls down Chimp PRDM9 ( Figure 5c ) . Thus , PRDM9 preferentially forms homo-multimers rather than hetero-multimers , at least for ZF arrays as highly diverged as Human and Chimp . These findings replicated after completely digesting DNA with benzonase ( Figure 5—figure supplement 3 ) . Because chimp and human PRDM9 ChIP-seq peaks almost never overlap ( Figure 2—figure supplement 1b ) , we can rule out the possibility that heteromultimer formation between these two alleles results from co-binding to short DNA fragments that may be protected from benzonase digestion by PRDM9 . That is , these results also confirm that PRDM9 multimer formation must be mediated by protein-protein interactions , not by protein-DNA interactions , though we still cannot formally rule out a role for DNA in enhancing this protein-protein interaction . The extremely rapid evolution of PRDM9’s zinc fingers , both within and between species , is one of the most striking features of this remarkable protein . Our results imply that over and above their role in positioning recombination sites and a role in chromosome synapsis ( Davies et al . , 2016 ) , several other factors might influence this evolution . We showed here that PRDM9’s zinc-finger domain can impact its ability to form multimers , its ability to activate gene expression , and its ability to initiate recombination , in particular if it binds near promoters or near targets of other zinc-finger proteins . PRDM9’s zinc-finger array has been regarded primarily as a DNA-binding domain with no other demonstrated functions , although studies of other zinc-finger proteins have shown that ZF domains can participate in highly specific protein-protein interactions , including with each other ( McCarty et al . , 2003; Lee et al . , 2007 ) . The mammalian gene with the most similar ZF-array to PRDM9 is ZNF133 , whose zinc fingers have an almost identical consensus sequence , apart from at DNA-contacting bases , to PRDM9 . ZNF133 has been shown to interact with PIAS1 ( which interestingly is recruited to DNA damage sites; Galanty et al . , 2009 ) via its zinc fingers , which can simultaneously bind its protein and DNA targets ( Lee et al . , 2007 ) . Thus , it seems credible that multimerization interactions involving PRDM9 might involve its zinc fingers , and it further seems plausible that PRDM9’s zinc-finger domain might be able to mediate interactions with other proteins . Currently , we can only speculate about what function PRDM9 multimerization might serve if it occurs in meiosis . If biased multimerization occurs in vivo between different PRDM9 alleles ( mediated by their variable zinc-finger domains ) , it could have important meiotic impacts in PRDM9 heterozygotes , although further study is needed , for example to determine if hetero-multimers form less efficiently between the human A , B and C alleles . Together with binding affinity differences , variable hetero-multimerization might impact PRDM9 dominance patterns , and dominance over less advantageous existing alleles could further increase the evolutionary advantage enjoyed by some newly arising alleles ( Baker et al . , 2015b ) or potentially play a role in the dosage sensitivity of PRDM9 in causing hybrid infertility in mice ( Flachs et al . , 2012; Ségurel et al . , 2011 ) . One intriguing hypothesis is that multimer formation may play some role in PRDM9-mediated homologue pairing , which we previously identified as a potential mechanism to explain the role of PRDM9 in fertility and speciation in mice ( Davies et al . , 2016 ) . In this case , a preference for homo-multimer formation would have obvious advantages . Our results also highlight the key impact of zinc-finger variation on PRDM9 binding at both fine and broad scales . We observed no fewer than seven different modes of human PRDM9 binding with different internal spacings between several DNA-contacting zinc fingers ( Figure 1a ) , a pattern not detected in previous studies . Binding is strongly impacted by all zinc fingers—as we observed in THE1B repeats and has been previously shown for mouse alleles ( Billings et al . , 2013 ) —and involves extensive sequence specificity not captured by a single shared motif . However , the chimpanzee w11a PRDM9 allele binds differently not only at fine scales but also broad scales ( Figure 2—figure supplement 1 ) and avoids promoters . Similarly , a recent study in mice ( Grey et al . , 2017 ) found that two mouse PRDM9 alleles do not directly bind at promoters . When Spo11 was present to form DSBs , additional PRDM9 peaks appeared at a small number of promoters—hypothesized as due to indirect recruitment ( Grey et al . , 2017 ) . An earlier study in mice with AT-rich PRDM9 binding motifs suggested that PRDM9 may direct recombination away from promoters by depositing competitive H3K4me3 marks ( Brick et al . , 2012 ) . In contrast to these alleles in chimp and mouse , we observed human PRDM9 directly binding to many promoter regions , previously unobserved due to filtering of PRDM9-independent H3K4me3 peaks and the evident suppression of DSB formation at these sites ( Pratto et al . , 2014; Baker et al . , 2015b ) . Given the similarity of promoter composition and organization across cell types , the human A/B alleles likely bind to promoters in vivo as well , although we cannot exclude the possibility that such binding is prevented somehow , and further study will need to determine the promoter affinities of other human PRDM9 alleles . Our results imply that the suppression of recombination at promoters ( including those that we show are bound by PRDM9 ) cannot simply be due to PRDM9 binding away from promoters . Interestingly , PRDM9 deposits less H3K36me3 at promoters compared to non-promoters , particularly at promoters with higher levels of PRDM9-independent H3K4me3 ( Figure 2 ) . We speculate that , if the co-occurrence of the H3K4me3 and H3K36me3 marks is essential for recombination initiation ( as suggested by Powers et al . , 2016; Yamada et al . , 2017 ) , then the relative lack of H3K36me3 at PRDM9-bound promoters could explain why these binding sites fail to initiate recombination . Of course , this does not explain why recombination tends toward promoters in the absence of PRDM9 , be it in knockout mice ( Brick et al . , 2012 ) or lineages that have lost PRDM9 ( Baker et al . , 2017 ) , such as dogs ( Auton et al . , 2013 ) . Together with the discovery of a fertile woman with two nonfunctional copies of PRDM9 ( Narasimhan et al . , 2016 ) , these results highlight the unresolved complexity surrounding PRDM9’s role in meiosis . Adding to this complexity is our finding that PRDM9 can influence the transcriptional activity of a subset of bound genes , such as the spermatogenesis-specific CTCFL and VCX genes , in transfected HEK293T cells . Speculatively , this pleiotropic effect may even help to explain why a single PRDM9 allele predominates in many human populations . That is , while a multitude of alleles may function equally well in specifying sites of meiotic recombination initiation , perhaps a subset can positively affect fertility by binding to and enhancing the expression of meiotic genes such as CTCFL , and these alleles are consequently driven to high frequency by positive selection . We also observed that a predicted submotif shared by many western chimp PRDM9 alleles ( Schwartz et al . , 2014 ) corresponds precisely to a group of chimp zinc fingers with the strongest influence on binding targets ( Figure 2—figure supplement 1c ) , similar to the prior observation of a group of ‘C-type’ human PRDM9 alleles that are diverse overall , but again overlap in the region identified to most strongly influence binding ( Hinch et al . , 2011; Berg et al . , 2011; Pratto et al . , 2014 ) . This apparent sharing of binding specificities between alleles could potentially be driven by PRDM9’s effects on transcription , its propensity to form multimers , and/or its ability to bind symmetrically to homologous chromosomes in heterozygotes ( Davies et al . , 2016 ) . Further work will need to explore the extent to which these behaviors are functionally important in vivo . Aside from recombination suppression at promoters , our results shed light on an additional level of recombination regulation occurring downstream of PRDM9 binding . Sequence-specific binding by the large collection of KRAB-ZNF genes is associated with localized recombination suppression at scales >1 kb , without suppressing nearby PRDM9 binding , or H3K4me3 deposition , either in transfected cells ( this study ) or in testes ( Pratto et al . , 2014 , Figure 4e ) . This implies that hundreds of motifs exist that mark sites of local recombination suppression . In contrast , we observe no impact of the presence/absence of binding sites for proteins such as DUX4 ( Young et al . , 2013 ) on recombination , despite our observing clear effects of the DUX4 binding motif on local chromatin marks ( Figure 4—source data 1 ) . Instead , perhaps only certain chromatin modifications suppress recombination . At their binding sites , many KRAB-ZNF proteins recruit TRIM28 which in turn recruits histone remodeling proteins including SETDB1 and HP1 , depositing the H3K9me3 modification ( Schultz et al . , 2002; Imbeault et al . , 2017 ) , which has been associated with suppression of meiotic recombination in mice ( Buard et al . , 2009; Walker et al . , 2015; Yamada et al . , 2017 ) . It has been suggested that KRAB-ZNF-induced heterochromatin may serve to stabilize repetitive sequences by preventing non-allelic homologous recombination ( NAHR ) ( Vogel et al . , 2006; Iyengar et al . , 2011 ) . Furthermore , PRDM9 has been shown to interact with both readers and writers of H3K9me3 ( Parvanov et al . , 2017 ) . Interestingly , we also saw a weak increase in H3K4me3 signal whenever H3K9me3 increased , and this signal is also observed in testes , implying the motifs we find can impact chromatin modifications in this tissue , and—unlike PRDM9—in many somatic cell types also . Most KRAB-ZNF proteins bind repeats , and they constitute the largest family of transcription factors in mammals , with rapid evolution ( Imbeault et al . , 2017 ) . Evidence suggests that the KRAB domain may have first evolved in an ancient ancestor of PRDM9 and then spread ( Birtle and Ponting , 2006 ) , so it is interesting that these partial descendants of PRDM9 appear to disrupt meiotic recombination . In general , KRAB-ZNF genes appear to emerge concomitantly with the spread of particular transposon families , and they play a role in repressing transposon activity ( Imbeault et al . , 2017; Jacobs et al . , 2014; Wolf et al . , 2015; Rowe et al . , 2013 ) . Paradoxically though , they often remain active long after their targets lose transpositional activity ( Imbeault et al . , 2017 ) . Our results suggest that one possible reason might be an adaptive role for KRAB-ZNF genes in specifically suppressing meiotic recombination in and around repeats , which otherwise could be prone to mediating deleterious genomic rearrangements ( as proposed by Zamudio et al . , 2015 regarding DNA methylation at transposons ) . If so , evolution of PRDM9 to bind new repeats might , in turn , lead to co-evolution of ZNF genes to suppress meiotic recombination at a subset of those repeats . We note that the meiotic effects of KRAB-ZNF proteins might be apparent even if they are not expressed in meiotic cells , as their chromatin marks might be transmitted epigenetically from precursor cells ( Rowe et al . , 2013 ) . However , previous work has shown that KRAB-ZNF co-repressors are essential for normal gametogenesis in mice . Namely , the H3K9me3 methyltransferase SETDB1 is required to silence endogenous retroviruses in mouse primordial germ cells ( Liu et al . , 2014 ) , and germline knockout of TRIM28 leads to sterility ( Weber et al . , 2002 ) . Further study will need to determine which , if any , KRAB-ZNF proteins are active in human meiotic cells . Another consequence of KRAB-ZNF-mediated meiotic recombination suppression is that not only PRDM9 binding sites , but potentially many other sites within hotspots , are predicted to cause DSB initiation asymmetry , and thus are likely to be subject to biased transmission—as seen previously for PRDM9 motifs and GC-biased gene conversion in hotspots ( Boulton et al . , 1997; Coop and Myers , 2007; Myers et al . , 2010; Baker et al . , 2015a; Smagulova et al . , 2016; Davies et al . , 2016 ) . Unlike self-destructive drive at PRDM9 motifs , such drive would bias the evolution of features with broad impacts across cell types , towards increased KRAB-ZNF binding and hence constitutive silencing of hotspot regions , even if this silencing is selectively disadvantageous . Recent work by ( Yamada et al . , 2017 ) has demonstrated that as many as a third of meiotic DSBs occur within repetitive sequences in B6 mice , although DSB frequencies vary substantially among different classes of repeats , with most classes being depleted for DSBs . The authors hypothesize that PRDM9 may evolve to target transposons for meiotic recombination so that the effects of hotspot death will rapidly inactivate them by driving mutations or deletions of the PRDM9 binding site to fixation ( and this advantage might compensate for the risk of NAHR at those repeats; Yamada et al . , 2017 ) . Our work suggests that PRDM9 binding to transposable elements might also inactivate them in a second way: by accelerating their evolution towards constitutive silencing by KRAB-ZNF proteins . In this model , hotspot self-destructive drive would be mirrored by the rapid accumulation of new KRAB-ZNF binding sites within PRDM9-bound transposable elements—a prediction that should be examined empirically by future studies . On the other hand , given strong DSB suppression at promoters , nearby PRDM9 binding sites might be immune from the effects of hotspot death , which would otherwise act to abolish its binding and drive potentially deleterious mutations—including any which might weaken the promoter—to fixation in these regions . Indeed , the potentially destructive or repressive effects of hotspot death could explain why meiotic recombination is directed away from functional elements like promoters , and towards deleterious elements like transposons , at least in humans and mice . A cDNA was custom synthesized to contain the full-length ( 2 , 685 bp ) PRDM9 transcript from the human reference genome ( GRCh37 ) , which is the B allele of PRDM9 . 218 synonymous base changes were engineered into the exon containing the zinc-finger domain in order to distinguish the synthetic copy of PRDM9 from the endogenous copy and to facilitate proper synthesis of this highly repetitive region . We cloned this cDNA into the pLEXm transient expression vector ( Aricescu et al . , 2006 ) by ligation with a Venus ( YFP ) tag at its N-terminus , fused using an AgeI restriction site . A similar synthesized construct was designed to match exon 10 of the chimp PRDM9 reference allele ( the ‘w11a’ allele , 2 , 022 bp , codon optimized for human expression and non-repetitiveness ) . Exons 1–9 were amplified from the human construct , and the chimp allele was fused at the N-terminus with an XbaI site . The ZFonly and noZF alleles were amplified using internal primers designed inside the full-length human construct . For the C-terminally tagged constructs , a 198 bp HA and 213 bp V5 linker were synthesized ( having the sequence linker-TwinStrep-linker-HA/V5-linker-P2A ) and cloned between each respective PRDM9 allele and a YFP tag using KpnI and AgeI sites , respectively . C-terminally tagged constructs were cloned into the pLENTI CMV/TO Puro DEST vector ( Addgene plasmid # 17293; Campeau et al . , 2009 ) , owing to its higher transient expression efficiency and to test the possibility of stable lentiviral transduction . Cloning into this vector was performed using the Gateway recombinase-based cloning system ( Thermo Fisher Scientific , Waltham , MA ) . Constructs were cloned , amplified , and isolated using an Qiagen ( Germany ) EndoFree Plasmid Giga Kit to yield transfection-quality DNA , which was verified by restriction digestion and Sanger sequencing . HEK293T cells were chosen owing to their high transfection efficiency , rapid growth rate , and low-cost media requirements . Cells were purchased directly from the ATCC ( ATCC CRL-3216; RRID:CVCL_0063 ) , with a certificate of analysis confirming cell line identity by Short Tandem Repeat profiling and confirming lack of mycoplasma contamination . All experiments were carried out on cells cultured for less than five passages from the purchased stock reference strain . Large-scale transfections of the N-terminal GFP-tagged Human PRDM9 construct were performed as described ( Aricescu et al . , 2006 ) . Cells were grown in DMEM media ( 10% FCS , 1X NEAA , 2 mM L-Glut , Sigma D6546; Millipore Sigma , Burlington , MA ) in 200 ml roller bottles at 37∘C/5% CO2 . A transfection cocktail was prepared for each bottle by adding 0 . 5 mg of chloroform-purified construct DNA to 50 ml of serum-free DMEM ( 1X NEAA , 2 mM L-glut ) and 1 mg polyethylenimine , followed by a 10 min incubation , and then addition of 375 μg of kifunensine . After the cells reached 75% confluence , the growth medium was removed from each roller bottle and replaced with 200 ml low-serum DMEM ( 2% FCS , 1X NEAA , 2 mM L-Glut ) and 50 ml transfection cocktail . Cells were then incubated for 72 hr to enable expression of the transfected construct . Expression was verified by fluorescence microscopy , and we consistently observed visible fluorescence in at least 50% of cells for all samples prior to harvesting . We performed all subsequent smaller-scale transfections of the C-terminally tagged constructs in the pLENTI vector using the FuGENE-HD transfection reagent according to manufacturer instructions ( Promega , Madison , WI ) . HEK293T cells ( ATCC CRL-3216; RRID:CVCL_0063 ) were thawed and incubated at 37°C with 5% CO2 in DMEM ( Sigma D6546 ) supplemented with 10% fetal bovine serum ( Sigma F7524 ) , 1X L-Glutamine ( Sigma G7513 ) , and 1X penicillin/streptomycin ( Sigma P0781 ) . The night before transfection , confluent cells were trypsinized ( Sigma T3924 ) , diluted in growth medium , and counted on an automatic hemocytometer ( Bio-Rad TC20 , Hercules , CA ) . For each replicate , 15 million cells were seeded in 30 ml growth medium in a T175 cell culture flask . The following morning , cells were transfected by mixing 30 μg total construct DNA into 800 μl OPTI-MEM ( Thermo Fisher Scientific 31985062 ) , then carefully adding 90 μl FuGENE-HD Transfection Reagent and flicking to mix , incubating at room temperature for 15 min , and then adding the mixture dropwise to each dish while swirling gently to mix . After 48 hr , cells were imaged briefly with a fluorescent microscope to confirm expression ( and transfection efficiency >50% ) , and were subsequently harvested . As negative controls , additional cells were seeded at the same time but were not transfected . ChIP-seq was performed according to an online protocol produced by Rick Myers’s laboratory ( Johnson et al . , 2007 ) , which was used to produce much of the ENCODE Project’s ChIP-seq data ( ENCODE Project Consortium , 2012 ) , with several optimizing modifications . Slight modifications were made for the smaller-scale transfection experiments with C-terminally tagged constructs . Crosslinking was performed in 1% formaldehyde for 5 min . Input chromatin was ‘pre-cleared’ to remove chromatin bound non-specifically by the beads . For each sample , 50 μl of equilibrated magnetic beads were resuspended in 100 μl PBS/BSA and added to the chromatin samples for pre-clearing for two hours at 4°C with rotation . Beads were removed , and 100 μl of pre-cleared chromatin was set aside for the input control . 5 μl ChIP-grade rabbit polyclonal antibody ( Abcam anti-HA ab9110 RRID:AB_307019 , anti-V5 ab9116 RRID:AB_307024 , anti-H3K4me3 ab8580 RRID:AB_306649 , or anti-H3K36me3 ab9050 RRID:AB_306966 ) was added to the remaining pre-cleared chromatin and incubated overnight at 4°C with rotation . 50 μl beads were washed and resuspended as before , then incubated with the chromatin samples for 2 hr at 4°C with rotation . After washing and decrosslinking , samples were further incubated with 80 μg RNAse A at 37°C for 60 min and then with 80 μg Proteinase K at 55°C for 90 min . DNA was submitted to the Oxford Genomics Centre for library preparation , sequencing , and mapping . For the N-terminal YFP-Human experiments , ChIP and input chromatin DNA samples from transfected and untransfected cells were sequenced in multiplexed paired-end Illumina ( San Diego , CA ) HiSeq1000 libraries , yielding 51 bp reads . Samples from transfected cells were multiplexed across 3 lanes , yielding roughly 77–101 million properly mapped read pairs ( i . e . fragments ) per replicate . Samples from untransfected cells ( processed independently ) were multiplexed across 2 lanes , yielding roughly 60–99 million properly mapped fragments per sample . For the C-terminal tag experiments , ChIP and input chromatin DNA samples from transfected and untransfected cells were sequenced all together in 6 lanes of paired-end Illumina HiSeq2500 libraries ( rapid mode ) , yielding 51 bp reads with 37 to 64 million reads per replicate . Coverage was chosen in each experiment to exceed recommendations for doing ChIP-seq with sufficient power to detect the majority of true binding events ( Landt et al . , 2012 ) . Sequencing reads were aligned to hg19 using BWA ( v0 . 7 . 0-r313 , option -q 10 , Li and Durbin , 2009 , RRID:SCR_010910 ) followed by Stampy ( v1 . 0 . 23-r2059 , option -bamkeepgoodreads , Lunter and Goodson , 2011 , RRID:SCR_005504 ) , and reads not mapped in a proper pair or with an insert size larger than 10 kb were removed . Read pairs representing likely PCR duplicates were also removed by samtools rmdup ( v0 . 1 . 19–44428 cd , Li et al . , 2009 , RRID:SCR_002105 ) . Pairs for which neither read had a mapping quality score greater than 0 were removed . For samples with only one replicate , fragments were split at random into two equally-sized pseudo-replicates . Fragment coverage from each replicate was then computed at each position in the genome using in-house code and the samtools ( v0 . 1 . 19–44428 cd , RRID:SCR_002105 ) and bedtools ( v2 . 23 . 0 , genomecov -d , RRID:SCR_006646 ) packages ( Li et al . , 2009; Quinlan and Hall , 2010 ) . Visualization ( producing browser screengrabs ) was done using the WashU Epigenome Browser ( Zhou et al . , 2011 , RRID:SCR_006208 ) . Details of the ChIP-seq samples are listed in Figure 1—source data 1 . Our peak calling algorithm is fully described in Appendix 1 . We compared the C-terminal Human-HA/V5 data with the N-terminal YFP-Human data and found strong overlap between the peak sets ( 60% ) but a poor correlation in raw coverage values or in our computed enrichment values ( r = 0 . 3 ) . We explored this further and noticed that the newer sequencing run had a strong increase in coverage of GC-rich regions ( nearly two-fold higher input coverage in regions with >60% GC ) , perhaps owing to differences in the ChIP protocol or to downstream differences in the library prep and sequencing steps ( Illumina HiSeq 1000 versus Illumina HiSeq 2500 ) . We also cannot exclude any effects due to the different placement of the tags . Due to this strong GC bias , we utilized the N-terminal YFP-Human dataset exclusively for most analyses of the human allele , except when directly comparing to data obtained using the C-terminal Human-HA/V5 constructs ( ATAC-seq , RNA-seq , H3K36me3 ChIP-seq , Chimp ChIP-seq ) . When comparing peak sets to determine overlap proportions , one must account for chance overlaps owing to the width and number of peaks being compared . For comparisons between single-base peak centers and DSB hotspot intervals , for example , we computed the expected number of chance overlaps c between the n peak centers and the t hotspot intervals , each with width wi , in a genome of size g as ( 1 ) c=∑i∈t ( 1− ( g−wig ) n ) . For more complicated comparisons , for example between two sets of intervals , we computed chance overlaps by randomly shifting the positions of one set of intervals uniformly in the interval [−60000 , 60000] , then counted the resulting overlaps to estimate c . Given f observed overlaps between the sets of n and t peaks , we can compute the corrected overlap fraction , o/t as follows . Let o/t be the proportion of systematic overlaps , c/t be the fraction of chance overlaps , and f/t be the proportion of total overlaps . The probability of no overlap is simply the product of the complements of chance and systematic overlaps , as follows: ( 1−f/t ) = ( 1−o/t ) ( 1−c/t ) . Solving for o/t then yields: ( 2 ) o/t=1−1−f/t1−c/t . Note that this method is only suitable when the number of chance overlaps is smaller than the number of total overlaps . For each peak , a 300 bp sequence ( centered on the called peak center ) was extracted from the reference sequence ( hg19 ) . Ab initio motif calling was performed on sequences from the top 5 , 000 peaks ( ranked by enrichment ) that passed a set of stringent filters ( p<10-10 , enrichment >2 , C . I . width ≤50 , no bases overlapping annotated repeats , number of input reads between 10%ile and 90%ile , and ≥30 reads from ChIP rep1 + ChIP rep2 ) . Motif calling proceeded in two stages: seeding motif identification , and joint motif refinement . Each seeding motif was obtained by first counting all 10-mers present in all input sequences , and from the top 50 most frequently occurring 10-mers , the one with the greatest over-representation in the central 100 bp of each peak sequence was chosen . This seeding 10-mer was then refined for 100 iterations as described in ( Davies et al . , 2016 ) , and all peak sequences containing matches to this refined motif were removed . From the remaining sequences , a new 10-mer was found and refined into a seeding motif , and this process was iterated up to 20 times . The 20 resulting seeding motifs were then refined jointly for 200 iterations as described ( Davies et al . , 2016 ) . Three separate runs were performed for each sample to verify consensus . For the YFP-Human peaks , a run producing 17 final motifs was chosen , and of these the 7 motifs with ≥85% of matches occurring in the central 100 bp of each peak sequence were chosen as the final set in order to remove degenerate motifs ( i . e . those with little base specificity at any position ) as well as likely false positives ( such as a match to the motif for the AP1 transcription factor ) . For the Chimp-HA/V5 peaks , only two motifs were produced , one of which was a degenerate CT-rich motif found in only 10% of peaks ( but not centrally enriched ) , so it was filtered out ( not shown ) . These final motifs were then force-called on the full set of peaks ( without any peak filtering ) by rerunning the refinement algorithm ( Davies et al . , 2016 ) with the option to not update the motifs with each iteration . The motif with the greatest posterior probability ( of at least 0 . 75 ) of a match was reported for each peak , along with position and strand . For identifying motif matches genome wide , we used FIMO ( version 4 . 10 . 0; Bailey et al . , 2015 ) . ATAC libraries were prepared as described ( Buenrostro et al . , 2013 ) . Briefly , 50 , 000 cells were lysed in 10 mM Tris-HCl pH 7 . 4 , 10 mM NaCl , 3 mM MgCl2 , 0 . 1% IGEPAL CA-630 and the nuclei were pelleted at 500g for 10 min . The transposition reaction was carried out for 30 min at 37°C using the Nextera DNA Sample Preparation Kit ( Illumina ) according to the manufacturer’s instructions . The libraries were purified using the MinElute PCR Purification Kit ( Qiagen ) , PCR amplified , multiplexed , and sequenced by the Oxford Genomics Centre on an Illumina HiSeq2500 ( rapid mode ) to produce 60–77 million sequenced fragments ( 51 bp , paired-end reads ) per sample . Reads were mapped to the hs37d5 reference ( Abecasis et al . , 2012 ) using BWA ( v0 . 7 . 0-r313 , Li and Durbin , 2009 ) followed by Stampy ( v1 . 0 . 23-r2059 , with option –bamkeepgoodreads , Lunter and Goodson , 2011 ) . PCR duplicates , mtDNA-mapped reads , reads not mapped in a proper pair , reads with mapping quality equal to 0 , and pairs with an insert size larger than 2 kb were removed using samtools ( v0 . 1 . 19–44428 cd , Li et al . , 2009 ) , leaving ∼11 million fragments per sample . Using in-house code , fragments were split by size into inter-nucleosome ( 51–100 bp ) and mono-nucleosome fragments ( 180–247 bp ) , and the position of the central base in each fragment was reported , as described ( Buenrostro et al . , 2013 ) . This yielded ∼1 million inter-nucleosome and ∼3 million mono-nucleosome fragments per sample . Fragment center coverage was computed genome-wide using bedtools ( Quinlan and Hall , 2010 ) . Total RNA was extracted using the RNeasy kit ( Qiagen ) from three biological replicates ( independently transfected in separate wells in parallel ) per sample . For quantitative PCR analysis , RNA was reverse-transcribed using Expand Reverse Transcriptase ( Roche ) , according to the manufacturer’s instructions . qPCR reactions were carried out in duplicate for each sample using Fast SYBR Green Master Mix ( Applied Biosystems , Foster City , CA ) on a CFX real-time C1000 thermal cycler ( Bio-Rad ) , following the manufacturer’s guidelines . Data were analyzed using the CFX 2 . 1 Manager software ( Bio-Rad ) and normalized to the Tata binding protein ( TBP ) gene . Relative gene expression levels were calculated using the Δ⁢ΔCt method , after averaging the two technical replicates for each sample . Statistical analysis was carried using a one-tailed t test . Primer sequences ( from Hines et al . , 2010 and Lahn and Page , 2000 ) and Ct values are given in Figure 3—source data 1 . Total RNA was submitted to the Oxford Genomics Centre for mRNA enrichment , library preparation , and sequencing . Samples were multiplexed and sequenced on an Illumina Hi-Seq2500 ( rapid mode ) , yielding 71–98 million 51 bp read pairs per sample . We created a custom reference sequence by merging the hs37d5 reference ( used by the 1000 Genomes Project to improve mapping quality Abecasis et al . , 2012 ) with the construct and vector sequences transfected into our cells . Data were analyzed using the Tuxedo software package ( Trapnell et al . , 2012 ) . Reads were mapped and processed using TopHat ( version 2 . 0 . 13 , options –mate-inner-dist=250 –mate-std-dev 80 –transcriptome-index = Ensembl . GRCh37 . genes . gtf , RRID:SCR_013035 ) ; followed by Cufflinks , CuffQuant , and CuffDiff ( version 2 . 2 . 1 , RRID:SCR_014597 , RRID:SCR_001647 ) ; then analyzed using CummeRbund ( RRID:SCR_014568 ) . We searched for all genes with evidence of H3K4me3 within 500 bp of a TSS in the human-transfected sample ( p<0 . 05 , force-calling , requiring >5 input reads ) and with defined FPKM values in the untransfected sample . Of the 14 , 667 genes passing these filters , 10 , 652 ( 73% ) have a human PRDM9 binding peak within 500 bp of the TSS . Of these , 873 showed at least some evidence of differential expression between the human-transfected and untransfected samples ( p<0 . 05 ) , and of these , 76 are significant after correction for multiple testing , with 43 significant only in the human-transfected sample ( p<0 . 05 after Benjamini-Hochberg correction ) . For each experiment , 10 million HEK293T cells ( ATCC CRL-3216; RRID:CVCL_0063 ) were seeded in 20 ml growth medium in a 15 cm round cell culture dish . The following morning , cells were transfected by mixing 30 μg total DNA into 800 μl OPTI-MEM ( Thermo Fisher Scientific 31985062 ) , then carefully adding 90 μl FuGENE-HD Transfection Reagent and flicking to mix , incubating at room temperature for 15 min , and then adding the mixture dropwise to each dish while swirling gently to mix . After 48 hr , cells were imaged briefly with a fluorescence microscope to confirm expression and were subsequently harvested . As negative controls , additional cells were seeded at the same time but were not transfected . Dishes were aspirated to remove media and cells were washed with cold PBS . 2 ml of cold lysis buffer ( 1% Triton X-100 , 150 mM NaCl , 50 mM Tris pH 8 . 0 plus 2X final concentration of Roche cOmplete Protease Inhibitor Cocktail Tablets ) were added and cells were collected into 2 ml Eppendorf tubes using a cell scraper . Tubes were incubated on ice for 30 min and lysates were dounced 20 times in a 2 ml dounce homogenizer with a tight pestle to help shear nuclear membranes . Cells were spun at 2000g for 5 min to remove chromatin and cell debris . 100 μl of lysate was set aside as an input control , and the remainder was split evenly among experimental and mock IP conditions . 2 μg of primary antibody ( Abcam ChIP-grade rabbit polyclonal anti-HA ab9110 RRID:AB_307019 or anti-V5 ab9116 RRID:AB_307024 , or rabbit polyclonal IgG isotype control ab171870 RRID:AB_2687657 ) was added and lysates were incubated for 1 hr at 4°C with rotation . For each sample , 25 μl of magnetic beads ( Invitrogen M-280 Sheep Anti-Rabbit Dynabeads ) was equilibrated by washing 3 times in 1 ml cold PBS/BSA ( 1X PBS , 5 mg/ml BSA , filtered with 0 . 45-micron filter ) , then resuspending in 25 μl PBS/BSA . Beads were added to the lysates and incubated for an additional hour at 4∘C . Tubes were spun down and placed on a magnetic rack for 1 min . Beads were pipetted up and down in 1 ml cold lysis buffer and rotated for 3 min at 4∘C . Washing steps were repeated 4 more times , with all steps taking place in a cold room at 4∘C . Beads were resuspended in 20 μl 2X Laemmli western loading buffer and boiled for 5 min at 100∘C . Beads were removed on a magnetic stand and supernatants were diluted two-fold . The total protein concentrations of input lysates were estimated using a Pierce BCA Protein Assay Kit ( Thermo Fisher Scientific 23227 ) and a NanoDrop spectrophotometer ( Thermo Fisher Scientific ) . 4X Laemmli buffer was added to 50 μg of input protein to a final concentration of 1X then boiled for 5 min at 100∘C . Samples were run on 10-well 7 . 5% Bio-Rad mini-Protean TGX pre-cast gels at 150 Volts in standard TGX running buffer for approximately 1 hr , using 5 μl of Full-Range Rainbow Ladder ( VWR 95040–114 , Radnor , PA ) in one well . Gels were then assembled onto a Bio-Rad mini Trans-Blot transfer pack ( with PVDF membrane ) according to manufacturer instructions and run on a Trans-Blot Turbo machine on the Mixed MW setting ( 2 . 5A , up to 25V , 7 min ) . Membranes were quickly removed and transferred to 50 ml conical tubes , then blocked for 5 min with rotation in 10 ml Blocking Buffer ( 5% milk in PBS with 0 . 1% Tween-20 ) , which was then poured off . Primary antibodies were diluted 1:5 , 000 in 5 ml blocking buffer and added to the membranes and incubated for 1 hr at room temperature with rotation . Membranes were washed 3 times for 5 min each in PBST ( PBS with 0 . 1% Tween ) . Secondary antibody ( Amersham ECL Donkey anti-Rabbit IgG , HRP-linked , NA934 RRID:AB_772206; GE Healthcare Life Sciences , Pittsburgh , PA ) was diluted 1:30 , 000 in blocking buffer , then 5 ml was added to each membrane and they were incubated for 1 hr at room temperature with rotation . Membranes were washed an additional three times in PBST and one final time in PBS . Blots were imaged using a Bio-Rad Clarity ECL kit according to manufacturer instructions and placed between sheets of transparency film to prevent drying during imaging . Imaging was performed using a Bio-Rad ChemiDoc MP Instrument using chemiluminescence hi-sensitivity settings and signal accumulation mode for various exposure times . Image processing was performed in the Bio-Rad ImageLab software ( RRID:SCR_014210 ) , in which relative bands intensities were quantified by densitometry . HEK293T cells ( ATCC CRL-3216; RRID:CVCL_0063 ) were co-transfected for 48 hr with equimolar mixtures of pLenti constructs encoding V5-or HA-tagged full-length ( FL ) human ( h ) or chimp ( c ) PRDM9 , or the zinc-Finger ( ZF ) domain only , using Fugene HD transfection reagent according to the manufacturer’s guidelines ( Promega ) . Cells were lysed for 30 min on ice in buffer containing 50 mM Tris-HCl pH 8 . 0 , 150 mM NaCl , 1% Triton X-100 and a cocktail of protease inhibitors ( Roche ) . Cell debris were pelleted by centrifugation at 4°C for 20 min at 20 , 000g . Protein extracts were incubated in the presence or absence of 125 U/ml benzonase ( Sigma ) and 2 mM MgCl_2 for 1 hr at 4°C with gentle rotation , and clarified again by centrifugation for 15 min at 16 , 000g . Note a pellet is visible after treatment with benzonase . Extracts were incubated for 1 hr at 4°C with 2 μg of anti-V5 antibody ( Abcam ab9116 RRID:AB_307024 ) and a further 1 hr with 25 µl Dynabeads M-280 ( Thermo Fisher Scientific ) . After 5 washes in lysis buffer , the immunocomplexes were eluted from the beads for 5 min at 100°C in 2x Laemmli sample buffer ( Bio-Rad ) and resolved on a 4–15% ( ZF ) or 7 . 5% ( FL ) Mini-PROTEAN TGX precast gel ( Bio-Rad ) alongside 50 μg of input extracts ( measured by BCA assay , Thermo Fisher Scientific 23227 ) . Proteins were transferred onto PVDF membranes and PRDM9 was detected by western blot following standard procedures . Blots were blocked overnight in PBS containing 0 . 1% Tween-20% and 5% milk , and incubated for 1 hr at room temperature with anti-HA ( Abcam ab9110 RRID:AB_307019 ) or anti-V5 ( Abcam ab9116 RRID:AB_307024 ) antibodies ( 1:5 , 000 dilution ) , and appropriate ECL HRP-conjugated IgG secondary antibodies ( Amersham ECL Donkey anti-Rabbit IgG , HRP-linked , NA934 RRID:AB_772206 ) with 3 washes in PBS-Tween buffer in between . Protein signals were revealed using the ECL Prime western blotting detection reagent according to the manufacturer’s recommendations ( GE Healthcare ) . To assess benzonase digestion efficiency , input protein extracts were diluted 1:20 in 0 . 1% SDS , and DNA concentration was measured on a nanodrop . 2 μg of DNA from each sample was analyzed on a 2% agarose gel in the presence of 0 . 1% SDS . HEK293T cells ( ATCC CRL-3216; RRID:CVCL_0063 ) were seeded onto glass coverslips pre-treated with Poly-L-Lysine ( Millipore Sigma ) . Transfections with FL , ZF only and no ZF V5-tagged PRDM9 constructs were carried out for 24 hr , as described above . Cells were fixed for 20 min in chilled methanol , washed three times in PBS , permeabilized for 10 min in PBS containing 0 . 1% Triton X-100 , washed again , and blocked for 1 hr at RT in PBS supplemented with 0 . 1% Tween 20% and 1% BSA . Cells were immunostained with an anti-V5 antibody ( Abcam ab9116 RRID:AB_307024 ) overnight at 4∘C , washed , and incubated for 1 hr at RT with an appropriate secondary antibody conjugated to the Alexa Fluor 594 dye ( Thermo Fisher Scientific A21207 RRID:AB_141637 ) . Coverslips were mounted in medium containing DAPI ( Vectashield , Vector Laboratories , United Kingdom ) and the cells were observed on a Olympus ( Japan ) BX60 microscope for epifluorescence equipped with a Sensys CCD camera ( Photometrics , Tucson , AZ ) . Images were captured using the Genus Cytovision software ( Leica Microsystems , Germany ) . Sequencing reads , genome-wide fragment coverage depth , peak calls , and differential gene expression files are available with GEO accession https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE99407 . Source code is available in the Github repository https://github . com/altemose/PRDM9-map ( Altemose , 2017; copy archived at https://github . com/elifesciences-publications/PRDM9-map ) .	Human cells have two copies of each chromosome: one from the mother , and one from the father . When cells divide to form sex cells , such as sperm or egg cells , the maternal and paternal chromosomes line up next to each other and swap some of their DNA . This process , known as genetic recombination , creates different versions of genes and ensures that we are all unique – or genetically diverse . Recombination is a complex process that is largely controlled by a protein called PRDM9 . This protein binds DNA at particular spots on the chromosome and directs other proteins to carry out recombination nearby . However , not all of PRDM9’s binding sites are known , and not all regions that PRDM9 binds to undergo recombination . Until now , it was not understood why this is the case at fine scales . To investigate this further , Altemose et al . activated the human version of PRDM9 in human kidney cells grown in the laboratory . The results showed that PRDM9 often bound near the start sites of genes , although these regions rarely undergo recombination in humans . When PRDM9 bound near these sites , it sometimes turned the gene on , which suggests that it may also help to regulate the activity of genes . Moreover , a specific group of DNA-binding proteins , called KRAB-ZNF proteins , appear to suppress recombination wherever they bind , which explains why some PRDM9 binding sites do not recombine . Lastly , Altemose et al . discovered that the part of PRDM9 that binds to DNA can also bind to other copies of PRDM9 proteins . This self-binding ability might play a role in bringing together the maternal and paternal chromosomes at the correct spots during recombination . Together , these results shed new light on the recombination process , which is a driving force in the formation of new species and essential for fertility . A next step will be to study these results further in tissues of the reproductive organs . This will provide a better understanding of the forces that shape human evolution .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2017	A map of human PRDM9 binding provides evidence for novel behaviors of PRDM9 and other zinc-finger proteins in meiosis
Vulnerability to noise-induced tinnitus is associated with increased spontaneous firing rate in dorsal cochlear nucleus principal neurons , fusiform cells . This hyperactivity is caused , at least in part , by decreased Kv7 . 2/3 ( KCNQ2/3 ) potassium currents . However , the biophysical mechanisms underlying resilience to tinnitus , which is observed in noise-exposed mice that do not develop tinnitus ( non-tinnitus mice ) , remain unknown . Our results show that noise exposure induces , on average , a reduction in KCNQ2/3 channel activity in fusiform cells in noise-exposed mice by 4 days after exposure . Tinnitus is developed in mice that do not compensate for this reduction within the next 3 days . Resilience to tinnitus is developed in mice that show a re-emergence of KCNQ2/3 channel activity and a reduction in HCN channel activity . Our results highlight KCNQ2/3 and HCN channels as potential targets for designing novel therapeutics that may promote resilience to tinnitus . Tinnitus , the perception of sound in the absence of acoustic stimulus , is frequently caused by exposure to loud sounds ( Shargorodsky et al . , 2010 ) and can be detrimental to the quality of life for millions of tinnitus sufferers ( Roberts et al . , 2010; Shargorodsky et al . , 2010 ) . Although the development of tinnitus is strongly correlated with acoustic trauma and noise-induced hearing loss , the absence of tinnitus after exposure to loud sounds—resilience to tinnitus—has been observed in a significant percentage of the population both in humans and in animal models ( Yankaskas , 2012; Zeng et al . , 2012; Li et al . , 2013 ) . Many studies have investigated the plasticity mechanisms underlying susceptibility to tinnitus; however , much less is known about the resilience mechanisms . Elucidating the resilience mechanisms could be vital for both developing pharmacological interventions to prevent and treat tinnitus , and understanding the mechanisms that differentiate pathogenic plasticity that leads to tinnitus from homeostatic plasticity that prevents the induction of tinnitus . Here , we studied the resilience mechanisms in the dorsal cochlear nucleus ( DCN ) , an auditory brainstem nucleus . Ablation of the DCN 3–5 months after the acoustic trauma did not affect the psychophysical evidence of tinnitus ( Brozoski and Bauer , 2005 ) . However , when bilateral DCN lesion occurred prior to noise exposure , it prevented the induction of tinnitus ( Brozoski et al . , 2012 ) . These results suggest that although DCN is not essential for the maintenance of tinnitus , it is indispensable for the induction of tinnitus . One of the major neural correlates for tinnitus induction is increased spontaneous firing rates in DCN fusiform cell , termed hyperactivity ( Zhang and Kaltenbach , 1998; Brozoski et al . , 2002; Li et al . , 2013 ) . Whereas a persistent change in neuronal firing properties in other auditory centers such as the ventral cochlear nucleus , inferior colliculus , auditory thalamus or auditory cortex may be important for tinnitus maintenance ( Eggermont and Roberts , 2004; Vogler et al . , 2011; Kalappa et al . , 2014; Ropp et al . , 2014 ) , transient increases in DCN neuronal spontaneous firing rates appear crucial for tinnitus induction ( Li et al . , 2013; Ropp et al . , 2014 ) . After exposure to tinnitus-inducing sounds , key physiological properties of DCN fusiform cells are different between noise-exposed animals that develop tinnitus ( tinnitus animals ) and animals that do not develop tinnitus ( non-tinnitus animals ) . Namely , although all noise-exposed animals show similar shifts in their hearing thresholds , only a portion of them display DCN hyperactivity and develop tinnitus ( Longenecker and Galazyuk , 2011; Koehler and Shore , 2013; Li et al . , 2013; Longenecker et al . , 2014 ) . Importantly , fusiform cell hyperactivity in tinnitus mice is associated with reduced KCNQ2/3 currents at hyperpolarized membrane potentials , due to a depolarizing shift in the voltage dependence of KCNQ2/3 channel opening ( Li et al . , 2013 ) . This reduction , along with decreased synaptic inhibition and increased synaptic excitation , leads to DCN hyperactivity ( Wang et al . , 2009; Middleton et al . , 2011; Pilati et al . , 2012; Zeng et al . , 2012 ) . On the other hand , non-tinnitus mice do not show fusiform cell hyperactivity after sound exposure and express normal levels of KCNQ2/3 currents ( Li et al . , 2013 ) . Nonetheless , knowledge of the biophysical mechanisms associated with resilience to tinnitus is still in its infancy . Because in vitro studies have shown that the intrinsic properties of fusiform cells generate spontaneous firing in these neurons ( Leao et al . , 2012 ) , we blocked excitatory and inhibitory synaptic transmission to investigate the intrinsic mechanisms underlying resilience to tinnitus in a mouse model of noise-induced tinnitus . We evaluated fusiform cell intrinsic excitability in control , tinnitus and non-tinnitus mice . We then extended this analysis by pharmacologically and biophysically isolating the ion channels whose noise-induced plasticity was associated with fusiform cell spontaneous firing rates and tinnitus or non-tinnitus behavior . Finally , we manipulated channel activity in vivo and tested the effects of these manipulations on the induction of tinnitus . Our results highlight the importance of KCNQ2/3 and HCN channels in the resilience to tinnitus and illuminate therapeutic paths that may enhance resilience for tinnitus prevention . To study the neural mechanisms underlying resilience to noise-induced tinnitus , we employed an animal model of tinnitus that permits behavioral separation of tinnitus from non-tinnitus mice . According to this model , behavioral evidence of tinnitus is evaluated based on the inability of tinnitus mice to detect a silent sound gap in a continuous background sound , because their tinnitus ‘fills in the gap’ ( Turner et al . , 2006; Longenecker and Galazyuk , 2011; Li et al . , 2013 , but see Hickox and Liberman , 2014 and ‘Discussion’ ) . When a silent gap is introduced in a constant background sound before a startle stimulus ( gap trial ) , normal mice show reduced startle amplitude compared to their response to a startle stimulus preceded by the same background sound but without any gap ( no-gap trial; Figure 1A top left , diagram showing gap and no-gap trials; Figure 1A top middle , gray and black , representative startle responses from a sham-exposed , control mouse ) . Noise-exposed mice that do not detect the silent gap show similar startle amplitudes in gap and no-gap trials and are considered tinnitus mice ( Figure 1A top middle , gray and green , representative startle responses from a tinnitus mouse ) . Noise-exposed mice that detect the silent gap show reduced startle amplitudes in gap trials and are considered non-tinnitus mice ( Figure 1A top right , gray and blue , representative startle responses from a non-tinnitus mouse ) . To quantify the gap detection ability of control and noise-exposed mice , we calculated the gap startle ratio before and after exposure ( Figure 1B ) . Gap startle ratio is the maximum startle amplitude in gap trials divided by the maximum startle amplitude in no-gap trials . Moreover , to investigate the frequency specificity of noise-induced tinnitus , we quantified the gap startle ratio for background sounds with different frequencies ( ‘Materials and methods’ ) . Our results showed that 7 days after noise exposure , 52 . 4% ( 11/21 ) of the noise-exposed mice exhibited a significant gap detection deficit at high-frequency ( 20–32 kHz ) but not at low-frequency ( 10–16 kHz ) background sounds . This deficit is indicated by an increase in gap startle ratios and is consistent with behavioral evidence of high-frequency tinnitus ( Figure 1B middle; Figure 1—figure supplement 1A–D; ‘Materials and methods’ ) . Increases in gap startle ratios in tinnitus mice are not consistent with noise-induced impairments in temporal processing or noise-induced inability to hear the background sound , because prepulse inhibition , PPI , was similar among control , tinnitus , and non-tinnitus mice ( Figure 1A bottom left , diagram showing startle-only and prepulse trials; Figure 1A bottom middle and right , representative startle responses from control , tinnitus and non-tinnitus mice; Figure 1C , summary graphs of PPI ratio; Figure 1—figure supplement 1E–G , ‘Materials and methods’ ) . PPI ratio reflects the inhibition of startle response by a preceding non-startling sound of similar intensity as the background sound used in gap detection ( Figure 1C ) . Moreover , tinnitus and non-tinnitus mice displayed similar hearing thresholds before and after noise-exposure , as evidenced by their similar auditory brainstem response ( ABR ) thresholds ( Figure 2A , B; ‘Materials and methods’ ) . In response to acoustic stimuli , ABRs reflect the synchronous activity of auditory brainstem nuclei arising from the auditory nerve ( wave I ) to the inferior colliculus , IC ( wave V , Figure 2A ) . Similar ABR thresholds may be accompanied with hidden differences in suprathreshold amplitudes of wave I . Because these differences could reflect differential degeneration of the auditory nerve ( Kujawa and Liberman , 2009; Furman et al . , 2013 ) , we compared the wave I amplitude of ABRs in response to suprathreshold sounds between tinnitus and non-tinnitus mice , before and after noise exposure . Whereas wave I amplitudes were reduced after noise-exposure for all noise-exposed mice , tinnitus and non-tinnitus mice showed no differences in wave I amplitude , suggesting no difference in the damage of afferent nerve terminals between these two groups ( Figure 2C , D ) . Taken together , our results suggest that neither differential noise-induced hearing threshold shifts nor differential auditory nerve damage can explain the behavioral differences between tinnitus and non-tinnitus mice . 10 . 7554/eLife . 07242 . 003Figure 1 . Mouse model of tinnitus allows for behavioral separation of noise-exposed mice with either vulnerability or resilience to tinnitus . A . Top left: diagram illustrating gap and no-gap trials in the gap detection behavioral assay . Top right: representative startle responses in no-gap ( always in gray ) and gap trials from control ( black ) , tinnitus ( green ) , and non-tinnitus ( blue ) mice . AU: Arbitrary unit . Bottom left: diagram illustrating startle-only and prepulse trials in the prepulse inhibition ( PPI ) behavioral assay . Bottom right: representative startle responses in startle-only ( always in gray ) and prepulse trials from control ( black ) , tinnitus ( green ) , and non-tinnitus ( blue ) mice . B . Summary graphs of gap startle ratio ( response to gap trial/response to no-gap trial ) before ( open bar ) and 1 week after noise exposure ( filled bar ) for high- and low-frequency background sounds ( high-frequency background , 20–32 kHz , control: n = 21 , p = 0 . 6; tinnitus: n = 11 , p < 0 . 001; non-tinnitus: n = 10 , p = 0 . 6; low frequency background , 10–16 kHz , control: n = 21 , p = 0 . 42; tinnitus: n = 11 , p = 0 . 06; non-tinnitus: n = 10 , p = 0 . 07 ) . C . Summary graphs of prepulse inhibition ratio ( PPI ratio , response to prepulse trial/response to startle-only trial ) before ( open bar ) and 1 week after noise exposure ( filled bar ) for high- and low-frequency prepulse ( high-frequency background , 20–32 kHz , control: n = 21 , p = 0 . 25; tinnitus: n = 11 , p = 0 . 18; non-tinnitus: n = 10 , p = 0 . 36; low frequency background , 10–16 kHz , control: n = 22 , p = 0 . 56; tinnitus: n = 11 , p = 0 . 17; non-tinnitus: n = 10 , p = 0 . 56 ) . Asterisk , p < 0 . 05 . Error bars indicate SEM . See end of the manuscript for detailed values in B and C . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 00310 . 7554/eLife . 07242 . 004Figure 1—figure supplement 1 . Tinnitus behavior ( gap detection deficit ) is detected with high-frequency background sounds . A . Probability distribution of changes in gap startle ratio ( Δ gap startle ratio ) in sham-exposed ( control ) mice . Δ gap startle ratio represents changes in control mice between postnatal day P17- P20 and P24 - P27 . Data were fitted by a normal distribution ( red curve , μ = −0 . 02 , δ = 0 . 15 , n = 21 ) . Δ gap startle ratios smaller than 0 . 28 ( dotted line ) , which is the point that is 2 standard deviations above the mean and is used as the threshold for evaluating tinnitus , represent 98 . 6% of the experimental Δ gap startle ratios and 98 . 5% of the fitted distribution . B . Summary graph of gap startle ratio for different frequencies of background for control mice: 10 kHz , before: 0 . 69 ± 0 . 03 , n = 13 , after: 0 . 79 ± 0 . 04 , n = 17 , p = 0 . 08; 12 kHz , before: 0 . 65 ± 0 . 04 , n = 8 , after: 0 . 70 ± 0 . 04 , n = 18 , p = 0 . 47; 16 kHz , before: 0 . 67 ± 0 . 03 , n = 15 , after: 0 . 66 ± 0 . 03 , n = 18 , p = 0 . 67; 20 kHz , before: 0 . 64 ± 0 . 02 , n = 14 , after: 0 . 67 ± 0 . 03 , n = 18 , p = 0 . 47; 24 kHz , before: 0 . 74 ± 0 . 02 , n = 17 , after: 0 . 69 ± 0 . 05 , n = 14 , p = 0 . 34; 32 kHz , before: 0 . 71 ± 0 . 03 , n = 19 , after: 0 . 69 ± 0 . 05 , n = 16 , p = 0 . 77 . C . Summary graph of gap startle ratio for different frequencies of background for tinnitus mice: 10 kHz , before: 0 . 65 ± 0 . 04 , n = 8 , after: 0 . 70 ± 0 . 04 , n = 10 , p = 0 . 36; 12 kHz , before: 0 . 61 ± 0 . 06 , n = 5 , after: 0 . 71 ± 0 . 03 , n = 10 , p = 0 . 10; 16 kHz , before: 0 . 66 ± 0 . 05 , n = 8 , after: 0 . 74 ± 0 . 06 , n = 10 , p = 0 . 30; 20 kHz , before: 0 . 62 ± 0 . 06 , n = 7 , after: 0 . 80 ± 0 . 05 , n = 11 , p = 0 . 04; 24 kHz , before: 0 . 68 ± 0 . 02 , n = 7 , after: 0 . 86 ± 0 . 05 , n = 9 , p = 0 . 006; 32 kHz , before: 0 . 59 ± 0 . 03 , n = 8 , after: 0 . 86 ± 0 . 03 , n = 12 , p < 0 . 0001 . D . Summary graph of gap startle ratio for different frequencies of background for non-tinnitus mice: 10 kHz , before: 0 . 69 ± 0 . 05 , n = 7 , after: 0 . 61 ± 0 . 07 , n = 10 , p = 0 . 40; 12 kHz , before: 0 . 78 ± 0 . 03 , n = 6 , after: 0 . 58 ± 0 . 06 , n = 8 , p = 0 . 02; 16 kHz , before: 0 . 71 ± 0 . 04 , n = 6 , after: 0 . 72 ± 0 . 04 , n = 9 , p = 0 . 83; 20 kHz , before: 0 . 73 ± 0 . 05 , n = 6 , after: 0 . 64 ± 0 . 06 , n = 9 , p = 0 . 32; 24 kHz , before: 0 . 66 ± 0 . 04 , n = 7 , after: 0 . 70 ± 0 . 07 , n = 9 , p = 0 . 67; 32 kHz , before: 0 . 71 ± 0 . 03 , n = 7 , after: 0 . 78 ± 0 . 05 , n = 6 , p = 0 . 26 . E . Summary graph of PPI ratio for different frequencies of background sound for control mice: 10 kHz , before: 0 . 61 ± 0 . 02 , n = 21 , after: 0 . 58 ± 0 . 02 , n = 21 , p = 0 . 24; 12 kHz , before: 0 . 66 ± 0 . 03 , n = 21 , after: 0 . 62 ± 0 . 02 , n = 21 , p = 0 . 34; 16 kHz , before: 0 . 61 ± 0 . 04 , n = 21 , after: 0 . 68 ± 0 . 03 , n = 19 , p = 0 . 21; 20 kHz , before: 0 . 69 ± 0 . 02 , n = 20 , after: 0 . 69 ± 0 . 02 , n = 21 , p = 0 . 90; 24 kHz , before: 0 . 63 ± 0 . 05 , n = 21 , after: 0 . 64 ± 0 . 04 , n = 19 , p = 0 . 87; 32 kHz , before: 0 . 55 ± 0 . 04 , n = 21 , after: 0 . 56 ± 0 . 04 , n = 21 , p = 0 . 87 . F . Summary graph of PPI ratio for different frequencies of background sound for tinnitus mice: 10 kHz , before: 0 . 52 ± 0 . 06 , n = 11 , after: 0 . 55 ± 0 . 05 , n = 11 , p = 0 . 79; 12 kHz , before: 0 . 52 ± 0 . 06 , n = 11 , after: 0 . 61 ± 0 . 05 , n = 11 , p = 0 . 32; 16 kHz , before: 0 . 51 ± 0 . 07 , n = 11 , after: 0 . 60 ± 0 . 03 , n = 10 , p = 0 . 24; 20 kHz , before: 0 . 53 ± 0 . 06 , n = 11 , after: 0 . 58 ± 0 . 04 , n = 11 , p = 0 . 51; 24 kHz , before: 0 . 50 ± 0 . 05 , n = 11 , after: 0 . 54 ± 0 . 04 , n = 11 , p = 0 . 56; 32 kHz , before: 0 . 50 ± 0 . 04 , n = 11 , after: 0 . 56 ± 0 . 04 , n = 11 , p = 0 . 28 . G . Summary graph of PPI ratio for different frequencies of background sound for non-tinnitus mice: 10 kHz , before: 0 . 62 ± 0 . 08 , n = 10 , after: 0 . 54 ± 0 . 05 , n = 10 , p = 0 . 36; 12 kHz , before: 0 . 59 ± 0 . 05 , n = 9 , after: 0 . 52 ± 0 . 04 , n = 9 , p = 0 . 3; 16 kHz , before: 0 . 50 ± 0 . 09 , n = 8 , after: 0 . 55 ± 0 . 05 , n = 9 , p = 0 . 57; 20 kHz , before: 0 . 52 ± 0 . 06 , n = 9 , after: 0 . 61 ± 0 . 06 , n = 10 , p = 0 . 31; 24 kHz , before: 0 . 42 ± 0 . 08 , n = 9 , after: 0 . 54 ± 0 . 09 , n = 10 , p = 0 . 36; 32 kHz , before: 0 . 58 ± 0 . 04 , n = 9 , after: 0 . 71 ± 0 . 04 , n = 10 , p = 0 . 04 . Asterisk , p < 0 . 05 . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 00410 . 7554/eLife . 07242 . 005Figure 2 . Tinnitus and non-tinnitus mice show similar ABR thresholds and suprathreshold ABR wave I amplitudes . A . Representative raw traces of auditory brainstem response ( ABR ) in response to click tone presented at different intensities ( dB ) . I–V represent the different waves of the ABR . B . Summary graph of ABR thresholds for tinnitus ( green ) and non-tinnitus ( blue ) mice before ( solid line ) and 7 days ( dashed line ) after noise exposure ( n = 5–11 , no statistical difference was observed between tinnitus and non-tinnitus mice ) . C . Summary graph of suprathreshold wave I amplitudes for tinnitus ( green ) and non-tinnitus ( blue ) mice before noise exposure for high frequency ( 20–32 kHz ) sound stimulation ( n = 12–25 , no statistical difference was observed between tinnitus and non-tinnitus mice ) . D . Summary graph of suprathreshold wave I amplitudes for tinnitus ( green ) and non-tinnitus ( blue ) mice 7 days after noise exposure for high frequency ( 20–32 kHz ) sound stimulation ( n = 4–10 , no statistical difference was observed between tinnitus and non-tinnitus mice ) . See end of the manuscript for detailed values for B–D . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 005 Next we explored whether behavioral differences between tinnitus and non-tinnitus mice are due to differential intrinsic plasticity in DCN fusiform cells , which are the first targets of the auditory nerve in the central nervous system . Tinnitus mice present reduced KCNQ2/3 channel activity and increased spontaneous firing rates in DCN fusiform cells 7 days after noise exposure ( Li et al . , 2013 ) . In contrast , non-tinnitus mice are associated with normal KCNQ2/3 channel activity and normal spontaneous firing rates of fusiform cell ( Li et al . , 2013 ) . These findings suggest that in resilient , non-tinnitus mice either there is no transient reduction in KCNQ2/3 channel activity post noise exposure or that KCNQ2/3 channel activity is transiently reduced but recovers by 7 days post noise exposure . To distinguish between these two possibilities , we measured KCNQ2/3 current amplitudes in fusiform cells from noise-exposed mice 4 days post noise exposure . Because fusiform cells from DCN regions that represent high , but not low , frequency sounds are involved in tinnitus-related biophysical changes ( Li et al . , 2013 ) , all in vitro electrophysiological recordings were conducted on fusiform cells from high-frequency DCN regions ( ≥20 kHz , dorsal part ) . To quantify KCNQ2/3 currents , we held fusiform cells at −30 mV for 5 s and then stepped the voltage to −50 mV for 1 s to unmask the slow deactivation of KCNQ2/3 channels ( Figure 3A ) . Consistent with previous measurements of KCNQ2/3 currents in fusiform cells ( Li et al . , 2013 ) , this protocol revealed a slowly deactivating current , which was significantly reduced by XE991 application ( 10 μM , a specific KCNQ channel blocker; Figure 3A ) . Because KCNQ2/3 heteromeric channels mediate the KCNQ currents in fusiform cells ( Li et al . , 2013 ) , the XE991-sensitive component of these recordings represents the KCNQ2/3 current amplitude . 4 days noise-exposed mice showed a significant reduction in KCNQ2/3 current amplitude ( Figure 3B; ‘Materials and methods’ ) . A Boltzmann fit of the KCNQ2/3 conductance–voltage ( G–V ) function showed that the Gmax of KCNQ2/3 currents was not different between 4 days sham- and 4 days noise-exposed mice ( Figure 3C , E ) , but the V1/2 was shifted to more depolarized potentials in the 4 days noise-exposed mice ( Figure 3C , D; ‘Materials and methods’ ) . These results suggest that the reduction of KCNQ2/3 currents in 4 days noise-exposed mice is due , at least in part , to a depolarizing shift in the V1/2 of KCNQ2/3 channels , which is mechanistically similar to the reduction of KCNQ2/3 currents in tinnitus mice when assessed 7 days after noise exposure ( Li et al . , 2013 ) . 10 . 7554/eLife . 07242 . 006Figure 3 . Noise-exposed mice show reduced KCNQ2/3 channel activity 4 days after noise exposure; this reduction is due to a depolarizing shift of V1/2 . A . Representative traces illustrating XE991 ( 10 μM ) -sensitive KCNQ2/3 currents ( Top ) in fusiform cells from a 4 days sham-exposed ( black ) and a 4 days noise-exposed ( yellow ) mouse in response to a voltage step to −50 mV from a holding potential of −30 mV ( Bottom; for clearer representation , currents were truncated along time axis ) . B . Summary graph showing KCNQ2/3 current amplitude in fusiform cells from 4 days sham-exposed and 4 days noise-exposed mice ( 4 days sham-exposed mice , 73 . 8 ± 9 . 05 pA , n = 6; 4 days noise-exposed mice , 39 . 16 ± 6 . 7 pA , n = 15 , p = 0 . 01 ) . C . Representative conductance–voltage relationship of XE991-sensitive current in 4 days sham-exposed ( dark gray ) and 4 days noise-exposed mice ( light gray ) . Gray and red lines represent Boltzmann fits . D . Summary graph for Boltzmann fit parameter V1/2 ( 4 days sham-exposed: −28 . 3 ± 1 . 9 mV , n = 5 , 4 days noise-exposed: −20 . 3 ± 2 . 0 mV , n = 4 , p = 0 . 03 ) . E . Summary graph for Boltzmann fit parameter Gmax ( 4 days sham-exposed: 39 . 1 ± 10 . 2 nS , n = 5 , 4 days noise-exposed: 37 . 7 ± 20 . 2 nS , n = 4 , p = 0 . 90 ) . F . Effect of retigabine injection 4–6 days after noise exposure on the percentage of mice that develop tinnitus . ( Noise-exposed mice + saline at day 4–6: 57 . 1% , n = 14 , noise-exposed mice + retigabine at day 4–6: 31 . 3% , n = 16 , p = 0 . 03 ) . Asterisk , p < 0 . 05 . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 006 Given that fusiform cells from non-tinnitus mice display control-level KCNQ2/3 currents 7 days post noise exposure ( Li et al . , 2013 ) , our results suggest that it is the recovery in KCNQ2/3 channel activity , not the lack of reduction in KCNQ2/3 currents , which is linked with the resilience to tinnitus . Moreover , previous findings showed that in vivo pharmacological activation of KCNQ currents with either KCNQ2-5 or KCNQ2/3 channel activators , retigabine or SF0034 , respectively ( Tatulian et al . , 2001; Kalappa et al . , 2015 ) , prevented the development of tinnitus ( Li et al . , 2013; Kalappa et al . , 2015 ) . The ability of KCNQ channel activators to prevent the development of tinnitus is probably occurring through the reversal of the pathogenic reduction of KCNQ2/3 channel activity and the promotion of the natural resilience to tinnitus . However , in those experiments KCNQ channel activators were administered 30 min after noise exposure and then twice a day for an additional 5 days . Here , we tested whether retigabine application initiated at day 4 , instead of 30 min after noise exposure , is sufficient to reduce the incidence of tinnitus at day 7 . Application of retigabine at day 4 significantly reduced the percentage of mice that developed tinnitus ( Figure 3F; ‘Materials and methods’ ) , suggesting that the recovery of KCNQ2/3 currents between day 4 and 7 is crucial for tinnitus resilience . Taken together , our results suggest that experience-dependent bidirectional plasticity of KCNQ2/3 channel activity is associated with the vulnerability and resilience to noise-induced tinnitus: experience-dependent depression in KCNQ2/3 channel activity is important for vulnerability , while experience-dependent recovery in KCNQ2/3 channel activity is important for resilience to tinnitus . Our results highlight that pathogenic ( reduction ) and homeostatic plasticity ( recovery ) in KCNQ2/3 channel activity are associated with tinnitus and non-tinnitus behavior , respectively . However , it is not known whether fusiform cells from non-tinnitus mice are biophysically similar to the fusiform cells from control mice , or they have undergone homeostatic plasticity in other channels , which could further contribute to tinnitus resilience . To answer this question , we compared the biophysical properties between fusiform cells from control and non-tinnitus mice 7 days after sham or noise exposure . We found that although control mice and non-tinnitus mice show similar non-tinnitus behavior ( Figure 1B ) , similar fusiform cell spontaneous firing rates ( Li et al . , 2013 ) , similar levels of KCNQ2/3 channel activity ( Li et al . , 2013 ) , and similar spike parameters ( Supplementary file 1; ‘Materials and methods’ ) , non-tinnitus mice are biophysically distinct from control mice . When we blocked fusiform cell spontaneous firing activity with tetrodotoxin ( 0 . 5 μM , a specific sodium channel blocker ) , we found that fusiform cells from non-tinnitus mice showed more hyperpolarized resting membrane potential ( RMP; Figure 4A ) . Because fusiform cell RMP is similar between control and tinnitus mice ( Li et al . , 2013 ) , this finding suggests specific changes in subthreshold ionic conductance ( s ) occurring only in non-tinnitus mice . Moreover , when we used small current steps to evaluate the onset and steady state input resistance ( Rin; Figure 4D; ‘Materials and methods’ ) , we found that onset Rin was not different among control , tinnitus , and non-tinnitus mice ( Figure 4B ) . However , non-tinnitus mice displayed a significantly increased steady-state Rin , which was revealed later in the voltage response ( Figure 4C; ‘Materials and methods’ ) . Together , our results revealed that fusiform cells from non-tinnitus mice exhibit a decrease in a slowly activating and deactivating depolarizing conductance that is open at subthreshold potentials . These results indicate that although control and non-tinnitus mice show similar non-tinnitus behavior , non-tinnitus mice are biophysically distinct from control mice . Importantly , these results highlight the involvement of homeostatic plasticity of additional , non-KCNQ , ionic conductances in underlying resilience to tinnitus . 10 . 7554/eLife . 07242 . 007Figure 4 . Non-tinnitus mice are biophysically distinct from control mice . A . Summary graph of resting membrane potential ( RMP ) of fusiform cells from control and non-tinnitus mice after blocking spontaneous firing with 0 . 5 μM TTX ( control: −64 . 2 ± 1 . 3 mV , n = 14; non-tinnitus: −67 . 5 ± 0 . 7 mV , n = 15 , p = 0 . 04 ) . B . Summary graph of onset input resistance ( Rin ) of fusiform cells as measured in D from control , tinnitus and non-tinnitus mice ( control: 97 . 9 ± 8 . 4 ΜΩ , n = 8; tinnitus: 90 . 1 ± 4 . 8 ΜΩ , n = 7; non-tinnitus: 105 . 7 ± 18 . 5 ΜΩ , n = 7 , p = 0 . 73 ) . C . Summary graph of steady-state Rin of fusiform cells as measured in D from control , tinnitus , and non-tinnitus mice ( control: 58 . 2 ± 3 . 6 ΜΩ , n = 8; tinnitus: 54 . 8 ± 4 . 2 ΜΩ , n = 7; non-tinnitus: 108 . 18 ± 15 . 9 ΜΩ , n = 7 , p = 0 . 04 ) . D . Representative voltage traces of fusiform cells from control , tinnitus , and non-tinnitus mice ( Top ) in response to small depolarizing and hyperpolarizing current steps ( Bottom , −60 pA–60 pA , 20 pA step ) for measuring input resistance . Shaded areas indicate the region for evaluating onset ( starting from the peak of the voltage response , 100 ms width ) and steady-state Rin ( starting at the voltage response 1 . 75 s after the current injection , 250 ms width ) . Asterisk , p < 0 . 05 . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 007 Hyperpolarization-activated cyclic nucleotide-gated channels ( HCN channels ) exhibit slowly activating and deactivating kinetics and are important regulators of subthreshold dynamics in fusiform cells ( Leao et al . , 2012 ) . We therefore hypothesized that a decrease in HCN channel activity mediates the hyperpolarized RMP and the increased steady-state Rin in non-tinnitus mice . To test this hypothesis , we injected hyperpolarizing current to activate HCN channels . HCN channel activation led to a characteristic rebound in the membrane voltage ( voltage sag ) that was sensitive to the specific HCN channel blocker ZD7288 ( 10 μM; Figure 5A , red: before ZD7288; black: after ZD7288 ) . We quantified HCN channel activity by measuring the ZD7288-sensitive sag ratio , which is the difference between the peak voltage response ( Vpeak ) and steady-state voltage response ( Vss ) , normalized to the Vpeak ( Sag ratio = ( Vpeak−Vss ) /Vpeak100% ) . Consistent with our hypothesis , fusiform cells from non-tinnitus mice showed significantly reduced HCN channel activity ( Figure 5B ) . To confirm whether the reduction in HCN channel activity was responsible for the non-tinnitus-specific intrinsic properties , we evaluated the effect of ZD7288 on steady-state Rin and RMP . Indeed , ZD7288 application abolished the differences in steady-state input resistance and in RMP among control , tinnitus , and non-tinnitus mice ( Figure 5C–E ) . Together , our results are consistent with the notion that reduced HCN channel activity is a major contributor for the more hyperpolarized RMP and increased steady-state Rin in non-tinnitus mice . Moreover , our results show that noise-induced reduction in HCN channel activity is another critical biophysical change associated with resilience to tinnitus . 10 . 7554/eLife . 07242 . 008Figure 5 . Reduced HCN channel activity in non-tinnitus mice underlies the biophysical differences between control and non-tinnitus mice . A . Representative voltage traces ( Top ) of fusiform cells from control , tinnitus , and non-tinnitus mice in response to a hyperpolarizing current step ( Bottom ) for measuring HCN channel activity before ( red ) and after ( black ) 10 μΜ ZD7288 . HCN channel activity is measured by calculating the sag ratio ( Vpeak−Vss ) /Vpeak100 ( % ) that is sensitive to ZD7288 . B . Summary graph showing HCN channel activity as measured by the protocol described in A ( control: 15 . 5 ± 2 . 2% , n = 8; tinnitus: 14 . 1 ± 1 . 1% , n = 6; non-tinnitus: 8 . 3 ± 1 . 1% , n = 8 , p = 0 . 02 ) . C . Representative voltage traces of fusiform cells from control and non-tinnitus mice in response to current steps for measuring Rin as in Figure 4D but now in the presence of 10 μΜ ZD7288 . D . Summary graph showing steady-state Rin in control , tinnitus , and non-tinnitus mice in 10 μΜ ZD7288 ( control , 72 . 0 ± 15 . 1 MΩ , n = 4; tinnitus , 64 . 4 ± 8 . 0 MΩ , n = 6; non-tinnitus , 81 . 9 ± 16 . 4 MΩ , n = 6 , p = 0 . 9 ) . E . Summary graph showing resting membrane potential ( RMP ) in control , tinnitus , and non-tinnitus mice in 10 μΜ ZD7288 ( control , −78 . 4 ± 2 . 6 mV , n = 6; tinnitus , −78 . 4 ± 0 . 9 mV , n = 6; non-tinnitus , −76 . 8 ± 2 . 1 mV , n = 8 , p = 0 . 4 ) . Asterisk , p < 0 . 05 . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 008 Because both recovery in KCNQ currents and reduction in HCN channel activity are associated with resilience to tinnitus , we hypothesized that either KCNQ plasticity occurs before HCN channel plasticity and drives the resilience pathway or that plasticity in HCN channels occurs before KCNQ channel plasticity and plays the determinant role for tinnitus resilience . To distinguish between these two possibilities , we measured HCN channel activity 4 days after noise exposure . Our results revealed that although noise-exposed mice showed significantly reduced KCNQ2/3 currents in fusiform cells 4 days after noise exposure ( Figure 3B ) , HCN channel activity was not different between sham-exposed and noise-exposed mice ( Figure 6A , B ) . These results indicate that the decrease in KCNQ2/3 activity happens before the reduction of HCN channel activity . 10 . 7554/eLife . 07242 . 009Figure 6 . Noise-induced HCN plasticity occurs after KCNQ2/3 plasticity; KCNQ2/3 enhancement promotes HCN channel activity reduction . A . Representative voltage traces ( Bottom ) from fusiform cells in response to hyperpolarizing current step ( Top ) for measuring HCN channel activity in 4 days sham-exposed ( black ) and 4 days noise-exposed mice ( yellow ) . B . Summary graph showing HCN channel activity in fusiform cells from 4 days sham-exposed and 4 days noise-exposed mice , measured as in Figure 5A , B ( 4 days sham-exposed mice , 27 . 2 ± 2 . 5% , n = 4; 4 days noise-exposed mice , 24 . 4 ± 3 . 5% , n = 7 , p = 0 . 7 ) . C . Representative voltage traces ( Bottom ) from fusiform cells in response to hyperpolarizing current step ( top ) for measuring HCN channel activity in noise-exposed with saline injection ( black ) and noise-exposed mice with retigabine injection ( red ) . D . Summary graph showing HCN channel activity , as measured as in Figure 5A , B . ( Noise-exposed and saline-injected mice , 26 . 1 ± 0 . 6% , n = 13; noise-exposed retigabine-injected , 17 . 9 ± 0 . 5% , n = 14 , p = 0 . 008 ) . Asterisk , p < 0 . 05 . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 009 Given the initial appearance of KCNQ current reduction 4 days after noise exposure and because pharmacological activation of KCNQ currents with retigabine 4 days after noise exposure is sufficient to prevent the development of tinnitus ( Figure 3F ) , we hypothesized that the recovery of KCNQ channels not only drives resilience to tinnitus but also promotes the reduction in HCN channel activity in non-tinnitus mice . To test this hypothesis , we measured the effect of retigabine on HCN current amplitude ( intraperitoneal [IP] injections , 10 mg/kg 30 min post exposure then twice daily for 5 days; ‘Materials and methods’ ) . Consistent with our hypothesis , we found that fusiform cells from retigabine-injected mice , which showed reduced incidence of tinnitus and reduced spontaneous firing rate of fusiform cells ( Figure 7A , Figure 7—figure supplement 1A , C , D ) , displayed reduced HCN channel activity ( Figure 6C , D ) . For some mice , we used flupirtine , another KCNQ channel activator ( Mackie and Byron , 2008; ‘Materials and methods’ ) . Moreover , application of KCNQ channel activators did not affect ABR thresholds , wave I amplitude , or PPI ratios ( Figure 7B–D , Figure 7—figure supplement 1B ) , suggesting that these activators do not affect either noise-induced hearing threshold shifts or auditory nerve damage . These results are consistent with recent studies showing that application of retigabine does not have any effect on hearing thresholds ( Sheppard et al . , 2015 ) . However , application of retigabine reduced hearing loss in an animal model of sodium salicylate-induced sensorineural hearing loss and tinnitus ( Sheppard et al . , 2015 ) . This is probably due to the fact that different mechanisms mediate sodium salicylate- and noise-induced hearing loss . Taken together , our results suggest that increases in KCNQ2/3 channel activity promote a decrease in fusiform cell HCN channel activity and resilience to tinnitus . 10 . 7554/eLife . 07242 . 010Figure 7 . Injection of KCNQ activators after noise exposure reduces the incidence of tinnitus development without affecting threshold and suprathreshold ABRs . A . Percentage of mice that develop tinnitus ( noise-exposed mice with intraperitoneal ( IP ) injection of vehicle , 50% , n = 18 , noise-exposed mice with IP injection KCNQ channel activators , 25% , n = 20 , p = 0 . 02 ) . For noise-exposed mice with IP injection of vehicle ( 11 for retigabine vehicle and 7 flupirtine vehicle at 10 mg/kg ) ; for noise-exposed mice with IP injection of KCNQ channel activators ( 10 for retigabine and 10 for flupirtine at 10 mg/kg ) . B . Summary graph of ABR thresholds from saline- ( gray ) and retigabine-injected ( red ) mice before ( solid line ) and 7 days after ( dashed line ) noise exposure and injection ( n = 4–9 , no statistical difference was observed between retigabine- and saline-injected mice ) . C . Summary graph of suprathreshold wave I amplitudes for noise-exposed mice + saline ( gray ) and noise-exposed mice + retigabine ( red ) before noise exposure for high frequency ( 20–32 kHz ) sound stimulation ( n = 5–15 , no statistical difference was observed between retigabine- and saline-injected mice ) . D . Summary graph of suprathreshold wave I amplitudes for noise-exposed mice + saline ( gray ) and noise-exposed mice + retigabine ( red ) after noise exposure and injection for high-frequency ( 20–32 kHz ) sound stimulation ( n = 5–20 , no statistical difference was observed between retigabine- and saline-injected mice ) . See end of the manuscript for detailed values for B–D . Asterisk , p < 0 . 05 . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 01010 . 7554/eLife . 07242 . 011Figure 7—figure supplement 1 . In vivo administration of KCNQ channel activators prevents the development of tinnitus and reduces the spontaneous firing rate of fusiform cells . A . Gap startle ratio of noise-exposed mice with KCNQ channel activator injection ( high frequency background sound: before , 0 . 62 ± 0 . 02 , after , 0 . 68 ± 0 . 02 , n = 20 , p = 0 . 06; low frequency background sound: before , 0 . 70 ± 0 . 02 , after , 0 . 63 ± 0 . 03 , n = 20 , p = 0 . 06 ) and with vehicle injection ( high frequency background sound: before , 0 . 61 ± 0 . 02 , after , 0 . 70 ± 0 . 03 , n = 18 , p = 0 . 01; low frequency background sound: before , 0 . 66 ± 0 . 02 , after , 0 . 72 ± 0 . 03 , n = 18 , p = 0 . 08 ) . B . PPI ratio of noise-exposed mice with KCNQ channel activator injection ( high frequency background sound: before , 0 . 56 ± 0 . 03 , after , 0 . 49 ± 0 . 03 , n = 20 , p = 0 . 12; low frequency background sound: before , 0 . 54 ± 0 . 03 , after , 0 . 48 ± 0 . 02 , n = 20 , p = 0 . 16 ) and with vehicle injection ( high frequency background sound: before , 0 . 47 ± 0 . 03 , after , 0 . 51 ± 0 . 03 , n = 18 , p = 0 . 11; low frequency background sound: before , 0 . 44 ± 0 . 04 , after , 0 . 52 ± 0 . 03 , n = 18 , p = 0 . 36 ) . C . Representative spontaneous action potentials of fusiform cells from noise-exposed mice injected with either saline ( Upper trace , gray ) or retigabine ( Lower trace , red ) twice a day for 6 days . D . Summary graph showing spontaneous firing rate of fusiform cells from noise-exposed mice injected with either saline or retigabine twice a day for 6 days ( noise-exposed mice with saline: 15 . 0 ± 1 . 5 Hz , n = 5; noise-exposed mice with activator: 7 . 5 ± 2 . 2 Hz , n = 7 , p = 0 . 02 ) . Whole-cell voltage-follower mode recordings ( current clamp , at I = 0 ) were performed 7 days after noise exposure and in the presence of excitatory and inhibitory receptor antagonists ( 20 μM DNQX , 20 μM SR95531 , and 0 . 5 μM strychnine ) . Asterisk , p < 0 . 05 . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 011 4 days after noise exposure , fusiform cells from noise-exposed mice displayed reduced KCNQ2/3 currents and control level HCN channel activity . This biophysical profile is similar to the fusiform cell profile from tinnitus mice 7 days post exposure . Therefore , we hypothesized that 4 days noise-exposed mice would show behavioral evidence of tinnitus and fusiform cell hyperactivity . Surprisingly , 4 days noise-exposed mice showed normal fusiform cell spontaneous firing activity and no behavioral evidence of tinnitus ( Figure 8A , B , Figure 8—figure supplement 1 ) . The lack of association between reduced KCNQ2/3 channel activity and fusiform cell hyperactivity may be explained by the hyperpolarized RMP in fusiform cells from 4 days noise-exposed mice ( Figure 8C ) , which is probably due to plasticity in some other ionic conductance ( s ) ( ‘Discussion’ ) . These results suggest that biophysical changes that are associated with increases in fusiform cell spontaneous firing rates lead to tinnitus , while biophysical changes that maintain normal levels of spontaneous firing rates are associated with tinnitus resilience . 10 . 7554/eLife . 07242 . 012Figure 8 . 4 days after noise exposure , mice have reduced KCNQ2/3 current amplitude but do not show either hyperactivity or tinnitus . A . Percentage of mice that develop tinnitus in 4 days sham-exposed and 4 days noise-exposed mice ( 4 days sham-exposed: n = 19; 4 days noise-exposed: n = 20 , p = 0 . 28 ) . B . Summary graph showing spontaneous firing rate in fusiform cells , assessed with whole-cell , voltage-follower mode recordings ( current clamp , at I = 0 ) , from 4 days sham-exposed and 4 days noise-exposed mice ( 4 days sham-exposed: 14 . 3 ± 4 . 7 Hz , n = 6; 4 days noise-exposed: 5 . 6 ± 2 . 3 Hz , n = 6 , p = 0 . 13 ) . C . Summary graph showing resting membrane potential ( RMP ) in 4 days sham-exposed mice and 4 days noise-exposed mice ( 4 days sham-exposed: −64 . 4 ± 1 . 2 mV , n = 6; 4 days noise-exposed: −67 . 7 ± 0 . 8 mV , n = 14 , p = 0 . 04 ) . Asterisk , p < 0 . 05 . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 01210 . 7554/eLife . 07242 . 013Figure 8—figure supplement 1 . 4 days noise-exposed mice exhibit similar gap detection and PPI compared to 4 days sham-exposed mice . A . Probability distribution of changes in gap startle ratio ( Δ gap startle ratio ) before and 4 days after sham exposure . The distribution of Δ gap startle ratio represents the changes in gap startle ratios in control ( sham-exposed ) mice between postnatal P20 - P23 and P24 - P27 . Data were fitted by a normal distribution ( gray curve , μ = 0 . 09 , δ = 0 . 18 , n = 15 ) . Δ gap startle ratios smaller than 0 . 48 ( dotted line , which is the point that is 2 standard deviations above the mean and used as the threshold for evaluating tinnitus ) represent 98 . 5% of the experimental population and 98 . 5% of the fitted distribution . B . Probability distribution of Δ gap startle ratio before and 4 days after noise exposure . Data were fitted by a normal distribution ( gray curve , μ = 0 . 07 , δ = 0 . 12 , n = 15 ) . C . Comparison of Δ gap startle ratio distribution between 4 days sham-exposed and 4 days noise-exposed mice shown in A and B for different thresholds . With the threshold ranging from 1 to −1 in 0 . 02 increments , the fraction of Δ gap startle ratios above threshold in 4 days noise-exposed mice was plotted against the fraction of Δ gap startle ratios above threshold in 4 days sham-exposed mice . Because this relationship was not different from the diagonal line ( dotted line ) , we concluded that the fractions of Δ gap startle ratios above threshold in 4 days noise-exposed mice and 4 days sham-exposed mice were not different . D . Summary graph of gap startle ratio for different frequencies of background sound for 4 days sham-exposed mice ( 10 kHz , before: 0 . 67 ± 0 . 03 , n = 16 , after: 0 . 78 ± 0 . 03 , n = 20 , p = 0 . 05; 12 kHz , before: 0 . 67 ± 0 . 03 , n = 14 , after: 0 . 73 ± 0 . 03 , n = 18 , p = 0 . 14; 16 kHz , before: 0 . 62 ± 0 . 02 , n = 16 , after: 0 . 68 ± 0 . 04 , n = 18 , p = 0 . 27; 20 kHz , before: 0 . 63 ± 0 . 04 , n = 12 , after: 0 . 73 ± 0 . 03 , n = 19 , p = 0 . 051; 24 kHz , before: 0 . 64 ± 0 . 02 , n = 19 , after: 0 . 82 ± 0 . 04 , n = 16 , p < 0 . 001; 32 kHz , before: 0 . 68 ± 0 . 03 , n = 17 , after: 0 . 79 ± 0 . 04 , n = 18 , p = 0 . 01 ) and 4 days noise-exposed mice ( 10 kHz , before: 0 . 74 ± 0 . 02 , n = 15 , after: 0 . 70 ± 0 . 03 , n = 20 , p = 0 . 42; 12 kHz , before: 0 . 72 ± 0 . 03 , n = 17 , after: 0 . 70 ± 0 . 04 , n = 21 , p = 0 . 68; 16 kHz , before: 0 . 68 ± 0 . 04 , n = 16 , after: 0 . 79 ± 0 . 03 , n = 19 , p = 0 . 04; 20 kHz , before: 0 . 69 ± 0 . 02 , n = 18 , after: 0 . 77 ± 0 . 04 , n = 21 , p = 0 . 12; 24 kHz , before: 0 . 61 ± 0 . 04 , n = 12 , after: 0 . 76 ± 0 . 03 , n = 19 , p = 0 . 005; 32 kHz , before: 0 . 69 ± 0 . 03 , n = 15 , after: 0 . 85 ± 0 . 03 , n = 18 , p < 0 . 001 ) . E . Summary graph of PPI ratio for different frequencies of background sound for 4 days sham-exposed mice ( 10 kHz , before: 0 . 62 ± 0 . 03 , n = 21 , after: 0 . 57 ± 0 . 04 , n = 19 , p = 0 . 13; 12 kHz , before: 0 . 63 ± 0 . 03 , n = 19 , after: 0 . 59 ± 0 . 03 , n = 21 , p = 0 . 28; 16 kHz , before: 0 . 60 ± 0 . 06 , n = 20 , after: 0 . 64 ± 0 . 03 , n = 19 , p = 0 . 51; 20 kHz , before: 0 . 60 ± 0 . 03 , n = 19 , after: 0 . 64 ± 0 . 03 , n = 21 , p = 0 . 29; 24 kHz , before: 0 . 67 ± 0 . 04 , n = 21 , after: 0 . 66 ± 0 . 04 , n = 20 , p = 0 . 87; 32 kHz , before: 0 . 762 ± 0 . 04 , n = 20 , after: 0 . 68 ± 0 . 03 , n = 18 , p = 0 . 21 ) and 4 days noise-exposed mice ( 10 kHz , before: 0 . 54 ± 0 . 05 , n = 21 , after: 0 . 47 ± 0 . 04 , n = 21 , p = 0 . 33; 12 kHz , before: 0 . 50 ± 0 . 06 , n = 21 , after: 0 . 49 ± 0 . 05 , n = 21 , p = 0 . 91; 16 kHz , before: 0 . 50 ± 0 . 05 , n = 19 , after: 0 . 53 ± 0 . 06 , n = 17 , p = 0 . 65; 20 kHz , before: 0 . 58 ± 0 . 04 , n = 19 , after: 0 . 60 ± 0 . 04 , n = 21 , p = 0 . 63; 24 kHz , before: 0 . 53 ± 0 . 07 , n = 18 , after: 0 . 58 ± 0 . 05 , n = 18 , p = 0 . 55; 32 kHz , before: 0 . 68 ± 0 . 05 , n = 17 , after: 0 . 63 ± 0 . 06 , n = 17 , p = 0 . 46 ) . Asterisk , p < 0 . 05 . Error bars indicate SEM . 35 dB , noise-exposed + saline , 815 . 2 ± 68 . 3 nV , n = 15 , noise-exposed + retigabine , 826 . 1 ± 76 . 8 nV , n = 14 , p = 0 . 92; 40 dB , noise-exposed mice + saline , 947 . 8 ± 81 . 0 nV , n = 16 , noise-exposed + retigabine , 1054 . 5 ± 104 . 8 nV , n = 14 , p = 0 . 42; 45 dB , noise-exposed + saline , 963 . 2 ± 69 . 9 nV , n = 14 , noise-exposed + retigabine , 1163 . 5 ± 129 . 1 nV , n = 13 , p = 0 . 16; 50 dB , noise-exposed + saline , 1022 . 6 ± 64 . 3 nV , n = 16 , noise-exposed + retigabine , 1207 . 6 ± 99 . 9 nV , n = 14 , p = 0 . 12; 55 dB , noise-exposed + saline , 1247 . 1 ± 76 . 7 nV , n = 13 , noise-exposed + retigabine , 1266 . 6 ± 134 . 5 nV , n = 12 , p = 0 . 90; 60 dB , 1168 . 4 ± 74 . 8 nV , n = 9 , noise-exposed + retigabine , 1392 . 2 ± 170 . 8 nV , n = 10 , p = 0 . 26; 65 dB , noise-exposed + saline , 1097 . 1 ± 117 . 0 dB , n = 5 , noise-exposed + retigabine , 1183 . 3 ± 156 . 0 nV , n = 5 , p = 0 . 67 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 013 Taken together , we show that noise exposure leads to down regulation of KCNQ2/3 channel activity by 4 days after noise exposure . At this time , no tinnitus has developed yet , probably due to the absence of fusiform cell hyperactivity . Mice that show a natural recovery of KCNQ2/3 channel activity and a reduction in HCN channel activity display normal level of spontaneous firing rates and are resilient to tinnitus . Mice that show preservation of reduced KCNQ2/3 channel activity until 7 days post noise exposure show fusiform cell hyperactivity and develop tinnitus ( Figure 9 ) . 10 . 7554/eLife . 07242 . 014Figure 9 . Biophysical mechanisms underlying the development of vulnerability and resilience to noise-induced tinnitus . Diagram illustrating the noise-induced plasticity of KCNQ2/3 and HCN channel activity , the emergence of DCN hyperactivity , and the development of vulnerability and resilience to tinnitus . DOI: http://dx . doi . org/10 . 7554/eLife . 07242 . 014 Given the subjective nature of tinnitus perception , there is not currently any behavioral paradigm in either animals or humans that provides for objective evidence of tinnitus . Despite this inherent limitation for studying mechanisms underlying tinnitus perception , deficits in the gap-prepulse inhibition of the acoustic startle reflex ( gap detection ) is the most widely used behavioral test for assessing behavioral evidence of tinnitus in animals ( Galazyuk and Hebert , 2015 ) . Research based on this behavioral assay has revealed important findings for the mechanisms underlying tinnitus , including neuronal hyperactivity , increased neuronal bursting activity , increased neuronal synchrony , degradation of frequency tonotopy , reduced GABAergic/glycinergic inhibition , enhanced GABAergic tonic inhibition , enhanced glutamatergic excitation , and changes in the biophysical properties of KCNQ channels ( Engineer et al . , 2011; Middleton et al . , 2011; Wang et al . , 2011; Zeng et al . , 2012; Li et al . , 2013; Kalappa et al . , 2014 , 2015; Sametsky et al . , 2015 ) . In several cases , these findings have led to ongoing clinical trials . For example , a recent study on the role of vagus nerve stimulation paired with tones for treating tinnitus was based on the gap detection paradigm in rats ( Engineer et al . , 2011 ) and is currently under clinical trial . Our studies , which also utilized the gap detection paradigm , revealed a novel KCNQ2/3 channel activator as a clinical candidate for preventing tinnitus ( Kalappa et al . , 2015 ) . Because the gap detection paradigm is based on a reflex , unlike other animal models of tinnitus requiring extensive behavioral training ( Bauer and Brozoski , 2001; Lobarinas et al . , 2004 ) , it allows for faster tinnitus screening of a larger number of animals and permits the separation of tinnitus from non-tinnitus animals . However , the interpretation that tinnitus ‘fills in’ the gap in the background sound that precedes the startle stimulus is not compatible with a recent study , which showed that gap detection deficits in noise-exposed mice were evident only when the gap was placed immediately , but not 80 ms , before the startle stimulus ( Hickox and Liberman , 2014 ) . If tinnitus just ‘fills in’ the gap , then gap detection deficits in tinnitus animals are expected to be independent of where the silent gap is placed . Moreover , although human studies revealed gap detection deficits in tinnitus sufferers , which are consistent with the use of the gap detection in animal studies , the frequency of the tinnitus percept did not fully match the frequency of the background sound where gap detection deficits were observed ( Fournier and Hebert , 2013 ) . This lack of correspondence in frequency is also inconsistent with the ‘fill in’ theory . Other human studies , which used higher-level cognitive tasks but not the acoustic reflex for assessing gap detection , did not reveal gap detection deficits in tinnitus sufferers ( Campolo et al . , 2013; Boyen et al . , 2015 ) . Taken together , we propose that although caution should be taken when interpreting gap detection deficits , the gap-prepulse inhibition of the acoustic startle reflex is a useful behavioral assay for elucidating tinnitus mechanisms . KCNQ2/3 channels are slowly activating , non-inactivating voltage-dependent potassium channels that are open at hyperpolarized ( subthreshold ) voltages . As a result , they control subthreshold membrane potential and serve as a powerful brake of neuronal firing activity ( Maljevic et al . , 2008; Brown and Passmore , 2009 ) . Disorders that are characterized by neuronal hyperexcitability , such as epilepsy , neuropathic pain , and tinnitus are linked to genetic or experience-dependent reductions in KCNQ2/3 channel activity ( Biervert et al . , 1998; Dedek et al . , 2001; Kullmann , 2002; Wuttke et al . , 2007; Li et al . , 2013 ) . Pharmacological activation of KCNQ channels by retigabine , an FDA approved anti-epileptic drug that activates KCNQ2-5 , or by SF0034 , a potent KCNQ2/3 activator , prevent seizures , neuropathic pain and the development of tinnitus ( Gunthorpe et al . , 2012; Li et al . , 2013; Kalappa et al . , 2015 ) . Because in vitro studies have shown that KCNQ2/3 channel activity is plastic and can be enhanced in response to increased neuronal activity ( Wu et al . , 2008; Brown and Randall , 2009; Misonou , 2010 ) , it is possible that endogenous , non-pharmacologically driven recovery in KCNQ2/3 channel activity may provide resilience to KCNQ2/3-related pathology . However , such KCNQ2/3-mediated resilience mechanisms have not been observed in hyperexcitability-related disorders . Here , we report that natural , non-pharmacologically driven recovery in KCNQ2/3 channel activity is linked with the resilience to noise-induced tinnitus . This novel finding suggests a temporal window during which endogenous , intrinsic mechanisms can restore KCNQ2/3 channel activity and lead to tinnitus resilience . Elucidation of these mechanisms will unmask previously unknown plasticity mechanisms of KCNQ2/3 channel activity , and lead to new targets for drug development towards enhancing resilience to noise-induced tinnitus in humans . KCNQ2/3 channels mediate the native neuronal M-type current ( Wang et al . , 1998 ) , a slowly activating voltage-gated potassium current blocked by muscarinic acetylcholine ( M ) receptors ( Brown and Adams , 1980 ) . KCNQ2/3 currents are strongly inhibited not only by activation of M receptors , but also by activation of other G protein-coupled receptors that reduce membrane phosphatidylinositol- ( 4 , 5 ) -bisphosphate ( PIP2 ) levels , which is the major determinant in KCNQ channel gating ( Marrion , 1997; Zhang et al . , 2003 ) . Our previous studies have revealed the important role of cholinergic activity and M receptors , namely M1 and M3 , in fusiform cell synaptic plasticity ( Zhao and Tzounopoulos , 2011 ) . Importantly , because noise exposure increases basal cholinergic activity in the DCN ( Jin et al . , 2006; Kaltenbach and Zhang , 2007; Manzoor et al . , 2013 ) , we propose that noise-induced increases and decreases ( or recovery to baseline levels ) in basal ( tonic ) M receptor signaling may underlie the bidirectional plasticity of KCNQ2/3 channel activity . HCN channels are non-inactivating cation channels that open at hyperpolarized voltages . Because of their voltage-dependent properties and their reversal potential at −30 mV , HCN channels not only depolarize the RMP , but also reduce membrane resistance and therefore stabilize the membrane voltage by opposing its alterations in response to synaptic inputs ( Pape , 1996; Robinson and Siegelbaum , 2003; Biel et al . , 2009 ) . Therefore , changes in HCN channel activity affect intrinsic and synaptic excitability in opposing directions . This differential influence of HCN channels on synaptic and intrinsic neuronal excitability has implicated HCN plasticity in the vulnerability and resilience to several neurological disorders characterized by abnormal synaptic or intrinsic excitability , such as epilepsy , neuropathic pain , depression , and Parkinson’s disease ( Chaplan et al . , 2003; Biel et al . , 2009; Chan et al . , 2011; Emery et al . , 2011; Friedman et al . , 2014 ) . Although future studies are needed to test the potential causal relationship between reduction of HCN channel activity in fusiform cell and the resilience to tinnitus , our results show that reduction of HCN channel activity is associated with resilience to tinnitus and with normal levels of fusiform cell spontaneous firing activity ( Figures 4 and 5 ) . We suggest that reduced HCN channel activity in non-tinnitus mice prevents fusiform cell hyperactivity and contributes to tinnitus resilience by hyperpolarizing the RMP of fusiform cell ( Figures 4 and 5 ) . Moreover , decreased HCN channel activity increased the steady-state Rin of fusiform cells ( Figures 4C and 5D ) , which is expected to enhance responsiveness of fusiform cells to synaptic inputs . Therefore , reduction of HCN channel activity in fusiform cells may contribute to enhanced evoked activity and potentially to noise-induced hyperacusis—the perception of moderate-level sounds as intolerably loud . Previous studies have shown that large increases in postsynaptic calcium lead to enhancement in HCN current amplitude via NMDA receptor ( NMDAR ) activation , CaMKII activation and increased HCN channel expression ( Fan et al . , 2005 ) , while smaller increases in calcium through L-type calcium channels and/or activation of mGluRs and PKC activation lead to activity-dependent decreases in HCN expression ( Brager and Johnston , 2007; Chan et al . , 2011 ) . Pharmacological enhancement of KCNQ channels with retigabine reduces spontaneous firing rate in fusiform cells ( Figure 7—figure supplement 1C , D ) and promotes a decrease in HCN current amplitude ( Figure 6D ) . Therefore , we propose that decreases in spontaneous firing that may be caused by the enhancement in KCNQ2/3 channel activity lead to changes in intracellular calcium , which , in turn , may trigger a homeostatic mechanism that decreases HCN currents in an effort to normalize spontaneous spike rates . Immunohistochemical studies have shown that the HCN2 subunit is expressed in fusiform cells that lack HCN1 subunit expression ( Koch et al . , 2004 ) , suggesting that HCN2 isoforms may mediate the noise-induced plasticity in fusiform cells . Therefore , we propose that manipulations that reduce HCN2 channel activity may serve as potential therapeutic path for preventing the development of tinnitus . Reduction of KCNQ2/3 currents is associated with increased spontaneous firing rate and tinnitus behavior 7 days after noise exposure ( Li et al . , 2013 ) . However , at 4 days after noise exposure , fusiform cells show reduced KCNQ2/3 currents ( Figure 3B ) but without hyperactivity ( Figure 8B ) . This lack of association between reduced KCNQ2/3 channel activity and neuronal hyperactivity can be explained by the hyperpolarized RMP of fusiform cells 4 days after noise exposure ( Figure 8C ) . Because various levels of inwardly rectifying potassium channels ( Kir ) set the diverse RMP of fusiform cells ( Leao et al . , 2012 ) , we suggest that noise-induced increases in Kirs may mediate this hyperpolarization of RMP and hence resilience to tinnitus . We conclude that reduction in KCNQ2/3 channel activity is a robust mechanism for triggering tinnitus , but its pathogenic effect is gated by the RMP of fusiform cells . Therefore , besides targeting KCNQ2/3 channel activators for preventing tinnitus , manipulations that hyperpolarize the fusiform cell RMP may also provide additional therapeutic approaches . Fusiform cells from control , 4 days noise-exposed and 7 days non-tinnitus mice show similar spontaneous firing rates but with different combinations of ionic conductances ( Figures 3 , 5 and 8 and ( Li et al . , 2013 ) ) . These findings complement and extend previous experimental and theoretical work showing that similar neuronal output can result from multiple combinations of intrinsic and synaptic properties ( Edelman and Gally , 2001; Prinz et al . , 2004; Marder and Goaillard , 2006; Goaillard and Dufour , 2014; Ratte et al . , 2014 ) . In accordance with this view , activity-dependent changes in conductances that affect neuronal excitability frequently trigger homeostatic , compensatory changes in different conductances , which result in constant neuronal output ( Desai et al . , 1999; LeMasson et al . , 1993; O'Leary et al . , 2010; O'Leary et al . , 2013; O'Leary et al . , 2014 ) . Our results are consistent with such homeostatic mechanisms and highlight that recovery of KCNQ channel activity is associated with a reduction in HCN channel activity and the maintenance of normal spontaneous firing rates in non-tinnitus mice ( Figures 3 , 5 and ( Li et al . , 2013 ) . Because homeostatic and coordinated regulation of potassium and HCN currents occurs in different species and neuronal circuits , such as dopaminergic neurons in rat susbstantia nigra pars compact , neurons of the lobster stomatogastric ganglion and octopus and MSO auditory brainstem neurons ( MacLean et al . , 2003; Oertel et al . , 2008; Khurana et al . , 2011; Amendola et al . , 2012 ) , we propose that coordinated plasticity of potassium and HCN channels may represent a general biophysical strategy for achieving neuronal homeostasis . The fact that multiple molecular pathologies underlie hyperexcitability-related disorders , such as neuropathic pain and epilepsy , has led to the suggestion that drugs that simultaneously target more than one type of ion channels could treat these disorders more effectively ( Goaillard and Dufour , 2014; Klassen et al . , 2011; Ratte et al . , 2014 ) . Similarly , because plasticity of multiple conductances is involved in tinnitus , we propose that a combination of drugs that enhance KCNQ2/3 and reduce HCN channel activity represents a potent therapeutic approach that will enhance resilience and reduce vulnerability to tinnitus . Moreover , simultaneous pharmacological enhancement of KCNQ2/3 and reduction of HCN channel activity is expected to act synergistically in stabilizing spontaneous firing rates in fusiform cells . This synergistic effect may in turn reduce the required concentration for each individual drug to exert its effect and therefore may lead to increased potency and reduced toxicity . Animals were handled and sacrificed according to methods approved by the Institutional Animal Care and Use Committee of the University of Pittsburgh . ICR ( CD-1 ) male and female mice were noise exposed at postnatal day P17–P20 . After anesthetizing the mouse with 1–1 . 5% isoflurane , a pipette tip that was connected with the speaker was inserted into left ear canal of the mouse ( unilateral noise exposure ) . Narrow bandpass noise with a 1 kHz bandwidth centered at 16 kHz was presented at 116 dB SPL ( dB ) for 45 min . For sham-exposed ( control ) mice , the procedures were identical with the noise-exposed mice but without noise presentation . For 4 days sham- and noise-exposure experiments , P20—P23 ( instead of P17—P20 ) mice were used for sham or noise exposure , so that electrophysiological recordings and behavioral assessments 4 days after noise exposure were performed on mice that had similar age with the mice that were used for experiments 7 days after sham or noise exposure . The gap detection paradigm ( Turner et al . , 2006; Li et al . , 2013 ) was used for assessing behavioral evidence of tinnitus . Gap detection of sham- or noise-exposed mice was assessed before exposure , 4 days or 1 week ( 6–7 days ) after sham or noise exposure . Detailed testing has been described in Li et al . ( 2013 ) . Briefly , the gap detection testing consists of two types of trials , gap trials and no-gap trials ( Figure 1A top left ) . In gap trials , a sound gap was embedded in a narrow bandpass background sound ( 1 kHz bandwidth centered at 10 , 12 , 16 , 20 , 24 , and 32 kHz presented at 70 dB ) , which was followed by the startle stimulus ( 20 ms white noise burst at 104 dB ) ; a 50-ms sound gap was introduced 130 ms before the startle stimulus . No-gap trials were the same as gap trials , but that no gap was introduced in the background sound . Gap trials and no-gap trials were presented as paired stimuli for the same background sound frequency , and were delivered in an alternating manner . The startle response represents the waveform of the downward pressing force that the mouse applies onto the platform in response to the startle stimulus . The ability of a mouse to detect sound gap was quantified by the gap startle ratio , which is the ratio of the peak-to-peak value of the startle waveform ( amplitude in Arbitrary Unit , AU ) in gap trials over the peak-to-peak value of the startle waveform of the paired no-gap trials . Prepulse inhibition ( PPI ) is the inverse of gap detection and was tested together with gap detection before , 4 days , or 1 week after sham or noise exposure . PPI testing consists of prepulse trials and startle-only trials , which were delivered in an alternating manner ( Figure 1A , bottom left ) . In prepulse trials , a brief non-startling sound ( prepulse ) of similar intensity as the background sound used in the gap detection test ( 50 ms , 70 dB bandpass sound with 1 kHz bandwidth centered at 10 , 12 , 16 , 20 , 24 , and 32 kHz ) . The prepulse was presented 130 ms before the startle stimulus . Startle-only trials were similar to the prepulse trials , but no prepulse was delivered . PPI was quantified by the PPI ratio , which is the ratio of the peak-to-peak value of the startle waveform in prepulse trials over the peak-to-peak value of the startle waveform in startle-only trials . For each sham- or noise-exposed mouse , gap detection and PPI ratios were averaged from three rounds of testing before and after sham or noise exposure . Each round of testing included 72 pairs of gap and no-gap trials for gap detection ( 6 background sound frequencies , 12 pairs for each frequency ) and 30 pairs of prepulse and startle-only trials for PPI ( 6 prepulse sound frequencies , 5 pairs for each frequency ) . Gap startle and PPI ratios were analyzed separately . Maximum absolute amplitude of the startle response and root mean square of baseline movement preceding the startle response ( RMS baseline ) were measured with a Labview-based recording system . Mean and standard deviation of RMS baseline of gap , no-gap , prepulse , and startle-only trials were measured to assess the variability of baseline movement . Trials with RMS baseline amplitude above or below the mean ± 2 . 5 standard deviations were eliminated . When a trial was eliminated , its paired trial was also eliminated . For gap detection trials of the same background frequency , gap startle ratios were sorted in an ascending manner . To control for variability of gap startle ratios tested with the same frequency , ratios that showed an increase of more than 0 . 5 from the preceding value were excluded; the following values were also excluded . If more than 5 ratios were excluded within a frequency , the gap startle ratio for this frequency was not used . For background and prepulse sound of the same testing frequency , individual gap startle and PPI ratios were averaged and generated the average ratio for each round of testing . Average gap startle and PPI ratios of each frequency were then averaged across the three rounds of pre-exposure and post-exposure testing to generate the final ratio value . According to previous established criteria , gap startle ratios that were bigger than 0 . 9 before sham or noise exposure , or bigger than 1 . 1 after exposure were excluded ( Li et al . , 2013 ) . Similarly , PPI ratios that were bigger than 1 before or after exposure were excluded ( Li et al . , 2013 ) . Changes in gap startle ratio before and after exposure ( Δ gap startle ratio ) were calculated by subtracting the post-exposure ratio from the pre-exposure ratio for each testing frequency . The probability distribution of Δ gap startle ratios from all testing frequencies of sham-exposed mice was fitted with a Gaussian distribution , which permitted the calculation of the mean ( μ ) and the standard deviation ( δ ) of the probability distribution ( Figure 1—figure supplement 1A , Figure 8—figure supplement 1A , B ) , as described previously ( Li et al . , 2013 ) . For evaluating the behavioral evidence of tinnitus , we calculated the point that is 2 standard deviations above the mean and used this value as the threshold ( Li et al . , 2013 ) . Mice that presented Δ gap startle ratio higher than threshold value in at least one tested frequency were considered tinnitus mice ( Li et al . , 2013 ) . To determine whether Δ gap startle ratios from 4 days noise-exposed mice were different from Δ gap startle ratios from 4 days sham-exposed mice , we calculated the fraction of Δ gap startle ratios that are above thresholds in 4 days sham- and noise-exposed mice , with thresholds ranging from 1 to −1 with 0 . 02 increments ( Figure 8—figure supplement 1C ) . Auditory brainstem response ( ABR ) thresholds were measured before , 4 days and 7 days after noise exposure following gap detection and PPI test . Measurements were conducted in a sound-attenuating chamber ( ENV-022SD; Med Associates , St . Albans , VT , United States ) . Mice were anesthetized initially with 3% isoflurane in oxygen , and then maintained with 1–1 . 5% . To present the sound stimuli , a pipette tip was fixed to the end of a plastic tube ( 2 . 5 cm in length ) , which was attached to the speaker ( CF-1; Tucker Davis Technologies , Alachua , FL , United States ) and was inserted into the left ear canal . ABR thresholds were obtained for 1-ms clicks and 3-ms tone bursts of 10 , 12 , 16 , 20 , 24 , and 32 kHz presented at a rate of 18 . 56/s in response to stimuli with different intensities ( 80 dB–15 dB , −5 dB step ) . Stimuli were produced using the System 3 software package from Tucker Davis Technologies . Evoked potentials were averaged 1024 times and filtered using a 300- to 3000-Hz bandpass filter . Wave I amplitude of each ABR response was measured as the peak-to-peak amplitude of the first wave that could be identified by eye from the baseline variation . Wave I amplitude for different stimulation intensities at 20 kHz , 24 kHz , and 32 kHz were averaged as response to high-frequency sounds ( Figure 2C , D; Figure 7C , D ) . For these experiments , ICR ( CD-1 ) mice ( P17—P20 ) , both male and female mice were used . For experiments where administration of vehicle and KCNQ activators start 30 min after noise exposure , 20 noise-exposed mice were injected with KCNQ channel activators: 10 mice with retigabine as its dihydrochloride salt and 10 mice with flupirtine , another KCNQ channel activator ( Mackie and Byron , 2008 ) . 18 noise-exposed mice were treated with vehicle: 11 mice with 0 . 9% saline paired with the retigabine group and 7 mice with 30% propylene glycol , 5% Tween 80 , 65% D5W paired with the flupirtine group . Prior to noise exposure , all mice were assessed for gap detection and PPI . For the KCNQ channel activator group , retigabine or flupirtine were administered 30 min after noise exposure via IP injection at a dose of 10 mg/kg . In the vehicle group , the same volume of vehicle solution was administered 30 min after noise exposure via IP injection . All mice were further administered with KCNQ channel activator or vehicle twice a day every 12 hr for 5 days . Gap detection and PPI were retested 24 hr after the final injection . Only mice from the retigabine and saline vehicle group were used for electrophysiological recordings ( Figure 6C , D and Figure 7—figure supplement 1C , D ) . For experiments where injection of retigabine or saline started at the end of day 3 after noise exposure , injections continued 3 times a day every 8 hr for 3 days at a dose of 10 mg/kg . Gap detection and PPI were tested before noise exposure and 24 hr after the final injection . Retigabine dihydrochloride was obtained from Santa Cruz Biotechnology ( LKT Laboratories , St . Paul , MN , United States ) and Alomone ( Jerusalem , Israel ) . Flupirtine maleate was obtained from Selleck Chemicals ( Houston , TX , United States ) . Coronal slices of the left DCN ( 210 μm ) were prepared from control and noise-exposed mice ( P24—P27 ) . Immediately , after brain slices were prepared , they were incubated in normal artificial cerebral spinal fluid ( ACSF ) at 36°C for 1 hr and then at room temperature . Fusiform cells were visualized using an Olympus upright microscope under oblique illumination condenser equipped with a XC-ST30 CCD camera and analog monitor . Cells were identified based on their morphological and electrophysiological characteristics . The preparation of slices and the identification for fusiform cells have been described in detail previously ( Tzounopoulos et al . , 2004 ) . The incubation as well as external recording solution contained ( in mM ) : 130 NaCl , 3 KCl , 1 . 2 KH2PO4 , 2 . 4 CaCl2 . 2H2O , 1 . 3 MgSO4 , 20 NaHCO3 , 3 NaHEPES , and 10 D-glucose , saturated with 95% O2/5% CO2 . DNQX ( 20 μM , AMPA and Kainate receptor antagonist , Abcam , Cambridge , MA , United States ) , strychnine ( 0 . 5 μM , glycine receptor antagonist , Sigma–Aldrich ) , SR95531 ( 20 μM , GABAa receptor antagonist , Abcam ) were used to block glutamatergic , glycinergic as well as GABAergic synaptic transmission , respectively . XE991 ( 10 μM , KCNQ channel blocker , Abcam ) was applied for blocking KCNQ currents . Tetrodotoxin ( TTX , selective inhibitor for sodium channel , 0 . 5 μM , Abcam ) was used to block spiking activity . ZD7288 ( 10 μM , blocker for HCN channel , Abcam ) was used to block HCN channel activity . Recordings were performed at 34–37°C using an inline heating system ( Warner Instruments , Hamden , CT , United States ) with perfusion speed maintained ( 4–6 ml min−1 ) . For whole-cell voltage and current clamp experiments , pipettes ( 3–5 ΜΩ ) were filled with a K+-based internal solution containing ( in mM ) : 113 K-gluconate , 4 . 5 MgCl2 , 2 . 6 H20 , 14 Tris-phosphocreatine , 9 HEPES , 0 . 1 EGTA , 4 Na2ATP , 0 . 3 Tris-GTP , 10 Sucrose , pH 7 . 3 , and 300 mOsmol . Liquid junction potential of −11 mV was corrected . Access resistance was monitored throughout the experiment from the size and shape of the capacitive transient in response to a 5-mV depolarization step . Recordings with access resistance larger than 15 MΩ were eliminated . Recordings were performed with Clampex 10 . 2 and Multiclamp 700B amplifier interfaced with Digidata 1440A data acquisition system ( Axon Instruments ) . For whole-cell voltage clamp experiments , fast , slow capacitive currents as well as series resistance ( Rs ) were compensated ( 70% , bandwidth 15 kHz ) . In voltage clamp ramp experiments , KH2PO4 was removed from ACSF . External CsCl2 ( 1 mM ) and CdCl2 ( 200 μM ) were used to block hyperpolarization-activated cyclic nucleotide-gated channels ( HCN , Ih channel ) and calcium channels , respectively . All recording protocols , except for the gap-free recording in current clamp , were applied below 0 . 1 Hz to eliminate potential short-term plasticity effects . Spike parameters were analyzed from spontaneous spikes in whole-cell , voltage-follower mode recordings ( current clamp , at I = 0; synaptic transmission was pharmacologically blocked ) . To assess spike properties , 20 consecutive spontaneous spikes were aligned at the negative peak and averaged . Spike threshold is the membrane potential at which the depolarization slope exceeds 10 V/s . Spike amplitude is the voltage difference between the spike threshold and the peak voltage of the spike . Depolarization and hyperpolarization slope indicate the maximum positive slope during the depolarization and minimum negative slope during hyperpolarization phase of the spike . Half height width is the width of the spike when voltage equals to ( spike threshold + half of spike amplitude ) . Fast afterhyperpolarization ( fAHP ) is the voltage difference between spike threshold and negative peak of the spike . Resting membrane potential ( RMP ) was measured with whole-cell , voltage-follower mode recordings ( current clamp , at I = 0 ) 5 min after TTX ( 0 . 5 μM ) application . ZD7288 was applied after TTX for evaluating its effect on RMP . Input resistance ( Rin ) was measured in current clamp mode through current injection ( −60 pA–60 pA , 20 pA step size , 2 s ) . Onset Rin is the slope of the current–voltage ( I–V ) relationship of the average membrane voltage response of the initial 100 ms starting from the peak of the voltage response . Steady-state Rin was calculated similarly to the onset Rin , but the last 250 ms of the voltage response were used . To quantify KCNQ2/3 currents , we measured the XE991-sensitive tail current amplitude in response to a voltage step to −50 mV from a holding potential of −30 mV , described in detail previously ( Li et al . , 2013 ) . To measure HCN channel activity , we injected bias current to maintain the membrane potential at −75 mV . A hyperpolarizing current step ( −550 pA , 2 s ) was then used to activate HCN channels . The biggest membrane potential change from −75 mV ( Vpeak ) , and the membrane potential change at the end of hyperpolarizing current ( Vss ) was used to calculate the sag ratio: Sag ratio = ( Vpeak−Vss ) /Vpeak × 100% ( Figure 5A ) . Sag ratio before and after ZD7288 ( 15 min application ) was subtracted and generated the ZD7288-sensitive sag ratio . In voltage-clamp ramp experiments , XE991-sensitive KCNQ currents elicited by slow voltage ramp ( 10 mV/s ) were converted to conductance ( G , nS ) according to Ohm's law: G = I/ ( V−Vr ) . I ( pA ) is the current amplitude at the membrane potential V ( mV ) , and Vr is the reversal potential of potassium [−85 . 5 mV; ( Leao et al . , 2012 ) ] . Conductance–voltage curves were then fitted with Boltzmann function to describe the voltage dependence of KCNQ activation ( Li et al . , 2013 ) ( Figure 3C , D , E ) . For data that were normally distributed ( based on Liliefors test ) , we conducted Student's t-test or One-Way Analysis of Variance ( ANOVA ) . Post-hoc analysis for one-way ANOVA was performed with Tukey's least significant test . For non-normally distributed data , we performed non-parametric Wilcoxon rank sum test or Kruskal–Wallis test . Binomial test was used for comparing percentages of tinnitus mice in response to different experimental manipulations .	Tinnitus is often described as ‘ringing in the ears’ . Though the phantom sounds , which are heard in the absence of any genuine external noise , can take a variety of forms including buzzing , whistling , or humming . While training the brain to pay less attention to these internally generated sounds can sometimes reduce the impact of tinnitus , many people find that the disorder reduces their quality of life significantly . One of the main causes of tinnitus is prolonged or repeated exposure to excessive noise . However , not everyone with such exposure develops tinnitus . Certain individuals appear to show a natural resilience . Identifying the basis of this resilience could make it possible to develop drugs that enhance these mechanisms , and thereby extend this protection to those who would otherwise be at risk of tinnitus . Li et al . have brought this a step closer by studying tinnitus resilience mechanisms in mice . Previous work by the same group revealed that in mice with tinnitus a group of neurons in the brainstem called ‘fusiform cells’ are overly active . These cells receive direct input from the ear , and their hyperactivity is largely due to ion channels called KCNQ2/3 channels being less active . These channels allow for potassium ions to flow across the membrane and thereby control the activity of fusiform cells . Li et al . now show that exposure to excessive noise causes a reduction in KCNQ2/3 activity in the exposed mice . However , in animals that successfully avoid developing tinnitus , KCNQ2/3 activity spontaneously recovers over the course of a few days . This recovery triggers a reduction in the activity of another type of ion channel , known as the HCN channel . The combined flexibility of KCNQ and HCN channels prevents tinnitus-associated hyperactivity in the fusiform cells . Drugs that increase activity of KCNQ2/3 channels , and/or reduce activity of HCN channels , could thus boost resilience to tinnitus . In the future , targeting both channel types at the same time could provide an effective treatment with minimal side effects .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Noise-induced plasticity of KCNQ2/3 and HCN channels underlies vulnerability and resilience to tinnitus
Evolutionary adaptation is a major source of antibiotic resistance in bacterial pathogens . Evolution-informed therapy aims to constrain resistance by accounting for bacterial evolvability . Sequential treatments with antibiotics that target different bacterial processes were previously shown to limit adaptation through genetic resistance trade-offs and negative hysteresis . Treatment with homogeneous sets of antibiotics is generally viewed to be disadvantageous as it should rapidly lead to cross-resistance . We here challenged this assumption by determining the evolutionary response of Pseudomonas aeruginosa to experimental sequential treatments involving both heterogenous and homogeneous antibiotic sets . To our surprise , we found that fast switching between only β-lactam antibiotics resulted in increased extinction of bacterial populations . We demonstrate that extinction is favored by low rates of spontaneous resistance emergence and low levels of spontaneous cross-resistance among the antibiotics in sequence . The uncovered principles may help to guide the optimized use of available antibiotics in highly potent , evolution-informed treatment designs . The efficacy of antibiotics for the treatment of infections is diminishing rapidly as bacteria evolve new mechanisms to resist antibiotics ( Laxminarayan et al . , 2013 ) . Resistance evolution is frequently observed during antibiotic therapy and can happen within days ( Bloemberg et al . , 2015; Hjort et al . , 2020; Tueffers et al . , 2019 ) . A failure to account for such rapid bacterial adaptation is likely a common reason for treatment failure ( Woods and Read , 2015; Zhou et al . , 2020 ) . For this reason , the field of evolutionary medicine specifically accounts for bacterial evolvability and seeks treatment solutions that are hard to overcome by genetic adaptation ( Andersson et al . , 2020; Merker et al . , 2020 ) . While an evolution-proof antibiotic remains to be found , the mechanisms that restrict evolutionary escape are starting to be revealed ( Bell and MacLean , 2018 ) . Such evolutionary insight may guide the design of effective and sustainable antibiotic therapy . An effective way of reducing the amount of evolutionary solutions is to administer several antibiotics either simultaneously ( i . e . , combination therapy ) or sequentially ( i . e . , sequential therapy ) . Tailored combination treatments make use of physiological and evolutionary constraints ( Baym et al . , 2016 ) . The emergence of resistance is delayed by combinations , when evolutionary escape requires multiple mutations and when drug interactions eliminate the intermediate genetic steps of single-drug resistance ( Chait et al . , 2007 ) , antibiotic tolerance ( Levin-Reisman et al . , 2017 ) , and heteroresistance ( Band et al . , 2019 ) . However , when genetic resistance to the combination is easily accessible , for example , through gene amplification of efflux pumps , then combination therapy can accelerate resistance emergence ( Pena-Miller et al . , 2013 ) . This undesired selective effect is potentially avoided by sequential drug application . Evolutionary escape from sequential treatments is constrained by negative hysteresis responses induced by specific antibiotics ( Roemhild et al . , 2018 ) and/or the emergence of genetic collateral sensitivity trade-offs ( Barbosa et al . , 2019; Yoshida et al . , 2017 ) . Negative hysteresis occurs when exposure to an antibiotic induces changes to bacterial physiology that transiently increase the killing efficacy of other antibiotics ( Roemhild et al . , 2018 ) . Collateral sensitivity is a genetic side effect of evolved resistance that too increases the efficacy of other antibiotics ( Szybalski and Bryson , 1952 ) . Collateral sensitivity is prevalent among pathogens and occurs especially between antibiotics with distinct mechanism of action ( i . e . , heterogeneous sets of antibiotics ) , while cross-resistance often emerges towards antibiotics with similar mode of action ( i . e . , homogeneous sets of antibiotics ) ( Barbosa et al . , 2017; Imamovic and Sommer , 2013; Lázár et al . , 2013; Maltas and Wood , 2019 ) . Thus , conventionally , multidrug treatments would avoid antibiotics from similar classes , with the rationale of limiting the overlap in the respective sets of resistance mutations , and thus the ensuing cross-resistance . The particular efficacy of sequential therapy has been confirmed with the help of evolution experiments under controlled laboratory conditions . Different types of sequential treatments have been tested . Some regimens involved a single switch between antibiotics , while others included multiple switches at short time intervals . One of the main findings was that the efficacy of sequential treatments depended both on the included antibiotics and the particular treatment sequence ( Fuentes-Hernandez et al . , 2015; Maltas and Wood , 2019; Roemhild et al . , 2015 ) . While fast sequential treatments did not exclude the eventual emergence of multidrug resistance , many significantly delayed bacterial adaptation compared to monotherapy ( Kim et al . , 2014; Roemhild et al . , 2015; Yoshida et al . , 2017 ) . A single antibiotic switch can also delay adaptation , dependent on the drug order , and it can additionally reverse previous resistance and resensitize bacterial populations to specific antibiotics ( Barbosa et al . , 2019; Hernando-Amado et al . , 2020; Imamovic and Sommer , 2013; Yen and Papin , 2017 ) . Moreover , our group previously demonstrated that fast sequential treatments with a heterogeneous set of three antibiotics – the fluoroquinolone ciprofloxacin ( CIP ) , the β-lactam carbenicillin ( CAR ) , and the aminoglycoside gentamicin ( GEN ) – delayed the emergence of multidrug resistance in the pathogen Pseudomonas aeruginosa ( Roemhild et al . , 2018 ) . The observed inhibition of evolutionary escape was manifested by the occurrence of population extinction , although antibiotic concentrations were below the minimal inhibitory concentration ( MIC ) . We further found that negative hysteresis at antibiotic switches reduced adaptation rates because it selected for distinct genetic changes . Several populations adapted to fast sequential treatment by independent mutations in the histidine kinase cpxS that only mildly increased resistance , thereby explaining the low rate of adaptation to the used antibiotics . Instead , the cpxS mutations suppressed negative hysteresis , demonstrating that adaptation was specific to the selective constraint imposed by the drug switches . Based on these findings , we assumed that the acting selective dynamics were ultimately a consequence of antibiotic heterogeneity . However , is this so ? Do selective dynamics differ for a homogenous set of drugs ? The primary aim of our current study was to assess the efficacy of sequential treatments with either heterogeneous or homogeneous sets of three antibiotics . We focused on P . aeruginosa strain PA14 as a tractable pathogen model system , for which comprehensive experimental reference data is available on resistance evolution ( e . g . , Barbosa et al . , 2019; Barbosa et al . , 2018; Hernando-Amado et al . , 2020; Roemhild et al . , 2018; Sanz-García et al . , 2018; Yen and Papin , 2017 ) . We performed similar evolution experiments as before , with three new sets of bactericidal antibiotics , two of which included only β-lactams , and one the three previously considered modes of action ( Figure 1—figure supplement 1A ) . The new heterogeneous drug set CIP , streptomycin ( STR ) , and doripenem ( DOR ) involved drug synergy and was expected to contribute to collateral sensitivity ( Barbosa et al . , 2018; Barbosa et al . , 2017 ) . The drug sets comprising three β-lactams , however , had all properties that would typically be avoided for the design of multidrug treatments . The three β-lactams CAR , cefsulodin ( CEF ) , and DOR have the same core structure and individually inhibit the DD-transpeptidase activity in cell-wall synthesis ( Walsh , 2003 ) . The collateral effects landscape between CAR-CEF-DOR was expected to be dominated by cross-resistance ( Barbosa et al . , 2017 ) and the three antibiotics showed neither synergy nor antagonism ( Barbosa et al . , 2018 ) . Resistance to these antibiotics may potentially be achieved through single mutations . The situation is replicated by the set of ticarcillin ( TIC ) , azlocillin ( AZL ) , and ceftazidime ( CEZ ) . In contrast to expectations , the triple β-lactam sequences showed high treatment potency . Therefore , the secondary aim of our study was to assess which characteristics constrained the ability of the bacteria to adapt to the β-lactam sequential treatments . We focused on one triple β-lactam set ( CAR-CEF-DOR ) and specifically tested the influence of antibiotic switching rate , switching regularity , negative hysteresis , the potential for spontaneous resistance evolution , and resulting cross-resistances on treatment efficacy . We challenged a total of 756 replicate P . aeruginosa populations with sequential treatments across three fully independent evolution experiments , each focused on a different set of three antibiotics ( Figure 1 , Figure 1—figure supplement 1 , Supplementary file 1A , Materials and methods ) . The antibiotic concentrations were calibrated to an inhibitory concentration of 75% ( IC75 ) , allowing bacteria to adapt to the imposed selection pressure . We used a serial dilution protocol for experimental evolution , with 2% culture transfer after 12 hr ( one transfer ) across a total of 96 transfers , equivalent to approximately 500 bacterial generations . Following the previous setup ( Roemhild et al . , 2018 ) , we recorded the evolutionary dynamics in response to 16 different treatments , belonging to four main treatment types: monotherapy , fast-regular , slow-regular , and random sequential therapy ( Figure 1 ) . Extinction of experimental populations differed considerably between the antibiotic sets . The two β-lactam sets produced a surprisingly high degree of extinction ( CAR-CEF-DOR and TIC-AZL-CEZ; extinct fraction 27 . 2 and 13 . 3% , respectively , Figure 1C ) . The observed extinction frequency was comparable to that observed in the previous experiment with CAR-CIP-GEN ( extinct fraction 15% , Figure 1C ) . CIP-DOR-STR caused no extinction , indicating that extinction was not explained by applying heterogeneous sets of antibiotics . Within the β-lactam sequential treatments , we observed that treatments that switched between antibiotics fast ( every transfer ) produced much higher extinction levels than those that switched slowly ( every four transfers ) or not at all ( Figure 1D ) . Most of the extinction events happened early in the experiment ( Figure 1E ) , indicating that the initial treatment steps are critical for adaptation of populations . We conclude that fast sequential β-lactam treatments showed a surprising ability to restrict bacterial adaptation . As this result was unexpected , we decided to research the mechanisms that constrain resistance emergence in β-lactam sequences . Given that the experiment involving CAR-CEF-DOR produced the highest fraction of extinct populations , we decided to focus further analyses on this set . The CAR-CEF-DOR triple β-lactam experiment was characterized in detail for changes in growth , evolved resistance , and whole-genome sequences in order to assess the selection dynamics involved . We calculated the relative growth yield ( see Materials and methods ) at the end of each transfer and found growth dynamics to be divided into three phases: an early phase of rapid adaptation ( transfers 1–12 ) , followed by a phase of gradual growth yield convergence ( transfers 13–48 ) , and a final plateau phase ( transfers 49–96 ) ( Figure 2A; the growth phases are separated by vertical dotted lines ) . We compared the main treatment types using general linear models ( GLM ) for each phase separately ( this fulfills the model assumption of response linearity ) . The early phase dynamics were characterized by significantly decelerated adaptation dynamics of the fast-regular group compared with monotherapy and slow-regular ( GLM , post hoc test , p<0 . 037 , Supplementary file 1B ) , but not random treatments . The slow-regular treatment did not differ significantly from monotherapy or random treatments ( GLM , post hoc test , p=0 . 469 , Supplementary file 1B ) . In the subsequent phase , growth yields of the groups converged to a plateau of roughly 90% relative yield , indicating similar final levels of adaptation ( the growth yields of main treatment groups showed no statistical differences in phases 2 and 3 , Supplementary file 1B ) . Alternating between the β-lactams fast and in a regular order therefore significantly constrained the growth of the bacterial populations . Intriguingly , in these fast sequential treatments , bacterial growth in the transfers with DOR was lower than in the transfers with the other two antibiotics ( Figure 2—figure supplement 1 ) , indicating an evolutionary constraint associated with the antibiotic DOR . We can rule out the alternative hypotheses that the reduced growth is explained by a stronger initial reduction in bacterial population size by DOR in comparison to the other two drugs or increased stochastic variation in dosage effects . All treatments were initiated using specifically standardized IC75 dosage ( see Materials and methods ) and at the IC75 , DOR showed very little variation ( Figure 1—figure supplement 1 ) . We thus hypothesize that the observed evolutionary constraint may be due to lower rate of DOR resistance emergence . To understand the dynamics of early adaptation in more detail , we measured the resistance profiles of 16 evolved populations after transfers 12 and 48 from the different antibiotic treatments ( representing the end of phases 1 and 2 , respectively; Figure 2B , C , Figure 2—figure supplements 2–5 , Supplementary file 1B–F; see Materials and methods ) . We randomly sampled 20 bacterial colonies from each population and characterized their resistance profile by broth microdilution . Resistance was measured for the three antibiotics of the evolution experiment and two additional clinically relevant antibiotics from different classes , ciprofloxacin and gentamicin . The resistance profiles in the early and the mid phases were found to be distinctly different . Resistance to the used β-lactams increased across the two time points only in some treatments , but not all ( Figure 2B , C , Figure 2—figure supplement 4 , Figure 2—figure supplement 5 , Supplementary file 1F ) , suggesting treatment-dependent evolutionary responses to the antibiotics . We assessed how the main treatment types varied in their β-lactam resistance using a GLM for each phase separately . Most treatment types varied significantly from each other in their multidrug β-lactam resistance in both phases ( Supplementary file 1C , D ) . The multidrug resistance in the early phase was in most cases constrained by the susceptibility to DOR ( e . g . , in the switching and monotherapy treatments ) . We additionally observed collateral responses of the treatment to the two non-β-lactams , which increased over time . We further used hierarchical clustering of the resistance profiles to assess the presence of subpopulations , followed by calculation of Shannon diversity for each population at both transfers . We found population diversity to be significantly higher at transfer 48 as compared to transfer 12 ( ANOVA , F = 6 . 2060 , p=0 . 01893 , Supplementary file 1E ) , indicating a diversification of the evolving lineages over time . Taken together , the population analysis of resistance profiles indicates that resistance evolution depends on the exact treatment protocol and that the dynamics of resistance emergence to DOR may be key for the observed deceleration of β-lactam adaptation in the fast-regular treatments . To identify the genomic changes underlying the first steps of β-lactam adaptation , we sequenced 33 whole genomes of the evolved and characterized isolates from the monotherapy , fast-regular , and slow-regular treatment types . Specifically , we sequenced three isolates from each population representing the distinct phenotypic subpopulations , assessed above . We found that all isolates , except those that received DOR monotherapy , had mutations in known resistance genes by the end of the early phase ( Table 1 ) . This agreed with the inferred resistance profiles where isolates from the DOR monotherapy did not show a noticeable amount of resistance at that stage ( Figure 2B ) . DOR resistance was , however , found at the end of the middle phase ( Figure 2C ) , and this was mirrored in the genomics with a non-synonymous mutation in the gene ftsI . This gene codes for the penicillin binding protein 3 ( PBP3 ) ( Liao and Hancock , 1995 ) , a common target of the three β-lactams ( Davies et al . , 2008; Fontana et al . , 2000; Rodriguez‐Tebár et al . , 1982; Rodríguez-Tebar et al . , 1982; Zimmermann , 1980 ) . ftsI was also found to be mutated in isolates from CAR monotherapy , although at a different site within the gene and associated with a different resistance profile than the DOR-associated ftsI variant ( Figure 2B ) . Isolates from CEF monotherapy contained mutations in pepA . This gene is responsible for the production of a protein required for cytotoxicity and virulence in P . aeruginosa ( Hauser et al . , 1998 ) . Although its role in antimicrobial resistance remains to be studied in detail , it was previously found to be mutated in P . aeruginosa strains resistant to certain β-lactams ( Cabot et al . , 2018; Sanz-García et al . , 2018 ) . The switching treatments selected for mutations in the above-listed and also in some additional genes . In particular , we identified mutations in nalD and phoQ , a negative regulator of the MexAB-OprM efflux pump and a two-component system , respectively . Mutations in these genes account for resistance to a variety of drugs in P . aeruginosa ( Barbosa et al . , 2021; Sobel et al . , 2005 ) . Further mutations were identified in some non-canonical β-lactam resistance genes such as rmcA , 23srRNA , 3-oxoacyl synthase , dnaX , and zipA ( Table 1 ) . Taken together , mutations in both canonical and non-canonical targets of β-lactam selection were identified in our experiment , and among these , DOR resistance mutations were found only later in the experiment , consistent with the obtained resistance profiles ( Figure 2B , C ) . Based on our detailed characterization of the CAR-CEF-DOR triple β-lactam experiment , we conclude that DOR has a key role in restricting evolutionary rescue as evidenced by the delayed acquisition of genetic resistance to it . As extinction was associated with antibiotic switches , we next focused on selective events that can occur at drug switches , such as hysteresis , an inducible physiological change . We characterized the complete hysteresis landscape between the three β-lactams: CAR , DOR , and CEF . We pretreated exponential phase cells with an antibiotic for only 15 min to ensure that cells are physiologically challenged but not subject to differential killing or replication . The pretreatment was followed by a change to fresh medium containing a second antibiotic as main treatment . We included controls of no pretreatment , or no main treatment ( Figure 3A ) . We found that negative hysteresis existed for several switches between the β-lactams ( Figure 3B , C , Figure 3—figure supplement 1 , Figure 3—figure supplement 2 ) . DOR and CAR displayed asymmetric bidirectional negative hysteresis with the switch from DOR to CAR , resulting in stronger negative hysteresis than the reverse . Negative hysteresis was also observed in the switch from CAR to CEF and CEF to CEF . To our surprise , only a single case of weak positive hysteresis was observed , although we generally anticipated it given that P . aeruginosa produces the AmpC β-lactamase ( Livermore , 1995 ) . We conclude that negative hysteresis is abundant between the studied β-lactams and is a potential predictor of treatment potency in the sequential β-lactam treatments . Since resistance to DOR was constrained in both the monotherapy and the switching treatments ( Figure 2B ) , we hypothesized that DOR resistance was difficult to achieve compared to the other two β-lactams . Resistance against a given drug can arise because of spontaneous direct resistance and/or because of collateral resistance from the preceding antibiotics in the sequence . As a first step , we thus measured the spontaneous direct resistance rate with the classic fluctuation assay using identical inhibitory concentrations of the three antibiotics ( Luria and Delbrück , 1943; Figure 4A , Supplementary file 1G ) . To determine the probability of indirect resistance in a second step , we isolated the obtained single-step mutants and quantified the fraction of cross-resistance towards the other two β-lactams with a patching assay ( Figure 4A ) . We used a comparatively large number of spontaneous mutants for this analysis ( n = 60 per antibiotic ) to capture the stochastic nature of evolution and , in this context , the potential importance of collateral effects for bacterial adaptation , as previously emphasized ( Nichol et al . , 2019 ) . We found that the spontaneous resistance rate was significantly lower for DOR than for CAR and CEF ( likelihood ratio test , p<0 . 0001 and p<0 . 01 , respectively; Supplementary file 1H , Figure 4B ) . Moreover , the resulting cross-resistance effects ( Figure 4C ) were particularly common towards CAR ( 93% of clones with spontaneous CEF resistance and 71% with DOR resistance ) and CEF ( 73% of originally CAR-resistant clones and 67% DOR-resistant clones ) . By contrast , the smallest levels of cross-resistance were expressed towards DOR ( 36% of originally CAR-resistant clones and 50% CEF-resistant clones ) . The overall fraction of cross-resistant clones was significantly smaller towards DOR than either CEF or CAR ( Fisher's exact test , p<0 . 0004; Supplementary file 1I ) . We conclude that of the three β-lactams DOR had the lowest probability for both direct and indirect resistance , thereby providing experimental support to the indication of constrained DOR resistance evolution obtained from the detailed phenotypic and genomic characterization of the evolved bacteria ( Figure 2 , Table 1 ) . We used the collected information to identify the critical determinant ( s ) of treatment efficacy in the CAR-CEF-DOR triple β-lactam experiment . We assessed the influence of either the two experimental predictors ( switching rate , temporal irregularity ) or the three biological predictors ( hysteresis , probability of spontaneous resistance , and resulting cross-resistance ) on each of the evolutionary responses extinction , rate of growth adaptation , and multidrug resistance , using separate GLM-based analyses ( see Materials and methods; Supplementary file 1J–O ) . For the biological predictors , we calculated the levels of cumulative hysteresis , cumulative probability of spontaneous resistance , and the cumulative levels of cross-resistance in each of the 16 individual treatments up to transfer 12 ( see Materials and methods ) . We focused our analysis on the early phase of evolution up to transfer 12 as it appeared most critical for treatment efficacy , especially for population extinctions that usually occurred early ( Figure 1E ) . Our analysis revealed that extinction was significantly associated with both the experimental predictors , switching rate ( GLM , F = 14 . 44 , p=0 . 0042 , Figure 5B , Supplementary file 1J–M ) and temporal irregularity ( GLM , F = 10 . 53 , p=0 . 0101 , Supplementary file 1M ) . Temporal irregularity further showed a statistical trend with multidrug resistance ( GLM , F = 4 . 19 , p=0 . 0711 , Supplementary file 1M ) . From our biological predictors , the cumulative cross-resistant fraction showed a significant association with extinction ( GLM , F = 10 . 42 , p=0 . 0121 , Supplementary file 1O ) , while cumulative probability of spontaneous resistance showed a statistical trend ( GLM , F = 4 . 14 , p=0 . 0763 , Supplementary file 1O ) . Indeed , the cumulative cross-resistant fraction and also the cumulative probability of spontaneous resistance are strongly correlated with extinction ( Figure 5B ) . The cumulative cross-resistant fraction is also strongly correlated with switching rate ( Figure 5—figure supplement 1 ) , most likely explaining the latter impact on extinction . By contrast , cumulative hysteresis levels did not have a significant influence on any of the evolutionary responses ( GLM , F = 0 . 16 , p=0 . 7015 , Supplementary file 1O ) . Taken together , our results suggest that in our sequential CAR-CEF-DOR treatments the switching rate , temporal irregularity of antibiotics , the probability of spontaneous resistance , and especially the resulting collateral effects ( maximized by switching rate ) determine treatment efficacy through their effect on bacterial extinction . The limiting factor appears to be constrained evolution of resistance and low levels of cross-resistance to DOR . Treatment with multiple β-lactam antibiotics is generally avoided due to the perceived fear of therapy failure from cross-resistance . Our work now challenges this widespread belief . We characterized the ability of replicate P . aeruginosa populations to evolve de novo resistance to sequential treatments with different drug sets . To our surprise , we found that sets of three β-lactams constrained bacterial adaptation by reducing bacterial survival . We demonstrate that treatment potency was determined by variation in the spontaneous rate of resistance to the β-lactams and the resulting collateral effects across sequential treatment protocols . Our initial screen of sequential protocols with different antibiotic triplets revealed that the triple β-lactam sequences are at least as effective at causing extinction as sequences of antibiotics with distinct modes of actions . This finding is at first sight counterintuitive , but at second sight not completely unexpected . The joint application of two β-lactam drugs was in fact tested and found effective in a few previous studies ( Rahme et al . , 2014 ) . For example , the β-lactam aztreonam was shown to interact synergistically with four other β-lactam drugs against multiple resistant isolates of Enterobacteriaceae and P . aeruginosa in vitro ( Buesing and Jorgensen , 1984 ) . A combination of ticarcillin with ceftazidime produced high efficacy in a rat peritonitis model ( Shyu et al . , 1987 ) . In a treatment of bacterial soft tissue infections , the combination of cefotaxime and mecillinam led to higher clinical response rates than the tested monotherapy ( File and Tan , 1983 ) . Further , the dual β-lactam combination of ceftazidime plus piperacillin was as effective as the combination of ceftazidime and tobramycin in granulocytopenic cancer patients ( Joshi et al . , 1993 ) . More recent studies demonstrated that a triple combination of meropenem , piperacillin , and tazobactam successfully constrained resistance evolution in Methicillin-resistant Staphylococcus aureus ( MRSA ) , both in vitro and in a mouse model ( Gonzales et al . , 2015 ) . In addition , the combination of cefotaxime and mecillinam was effective against Salmonella enterica harboring a mutant β-lactamase in a mouse model ( Rosenkilde et al . , 2019 ) . Our findings add to the high potency of treatments with multiple β-lactams . We conclude that the use of multiple β-lactams , either as a combination or sequentially , is a commonly underappreciated form of therapy and its use opens new avenues to better utilize our existing antibiotic armamentarium . Spontaneous rate of antibiotic resistance was found to play a critical role in the success of the CAR-CEF-DOR sequential treatment . The probability of spontaneous resistance on all three β-lactams was significantly different , with the rate of DOR resistance being the lowest . These rates determined the overall probability of acquiring direct resistance in treatment , which significantly correlated with the frequency of population extinction ( Figure 5B ) . Resistance rates were previously shown to vary towards different antibiotics , for example , in Escherichia coli ( El Meouche and Dunlop , 2018 ) and P . aeruginosa ( Oliver et al . , 2004 ) . This variation can arise from genetic factors such as mutational target space and physiological factors like activation of the bacterial SOS response ( Martinez and Baquero , 2000 ) . Such information on resistance rates has so far been used for predicting the occurrence of resistance against single drugs , and antibiotics that target multiple pathways in a cell are considered advantageous in this context ( Ross-Gillespie and Kümmerli , 2014 ) . One example of the latter are compounds against S . aureus that inhibit both DNA gyrase and topoisomerase IV ( Nyerges et al . , 2020 ) . The rate of resistance emergence may also be reduced by using adjuvants that target the SOS response ( Bell and MacLean , 2018 ) , as previously shown for compounds interfering with LexA activity leading to reduced resistance rates to ciprofloxacin and rifampicin in E . coli ( Cirz et al . , 2005 ) . Our study extends the role of resistance rates of antibiotics beyond this convention . We show that inclusion of an antibiotic with relatively low spontaneous resistance emergence can enhance the potency of a sequential treatment design . What could be the underlying reasons for the particular importance of DOR compared with the other β-lactams ? DOR belongs to the carbapenem subclass of the β-lactam antibiotics . Carbapenems possess broad activity against Gram-positive and Gram-negative bacteria ( Papp-Wallace et al . , 2011 ) and are active against many β-lactamase-producing microbes since their thiazolidinic ring makes them relatively resistant to β-lactamase-mediated hydrolysis ( Schafer et al . , 2009 ) . In contrast , the penicillin CAR is active mostly ( albeit not exclusively ) against Gram-negative bacteria ( Castle , 2007 ) while the activity of the cephalosporin CEF is restricted to P . aeruginosa ( Wright , 1986 ) . Within P . aeruginosa , all three antibiotics show high potency against a large variety of clinical isolates ( Castanheira et al . , 2009; Neu and Scully , 1984; Traub and Raymond , 1970 ) . Resistance rates for β-lactam antibiotics were assessed with different approaches across P . aeruginosa strains and clinical isolates , consistently showing that DOR has a particularly low propensity to select for resistance mutations , even when compared to other carbapenems ( Barbosa et al . , 2017; Barbosa et al . , 2021; Sakyo et al . , 2006; Tanimoto et al . , 2008; Mushtaq et al . , 2004; Fujimura et al . , 2009 ) . Therefore , the phenotype of reduced spontaneous resistance to DOR appears to be robustly expressed across different P . aeruginosa genotypes and does not extend to other carbapenems or β-lactams . One possible reason for this pattern may be variation in the range of β-lactam target proteins , in this case the penicillin binding proteins ( PBPs ) , and where DOR is known to bind more of these PBPs than do CAR or CEF ( Davies et al . , 2008; Fontana et al . , 2000; Rodriguez‐Tebár et al . , 1982; Rodríguez-Tebar et al . , 1982; Zapun et al . , 2008; Zimmermann , 1980 ) . Thus , target resistance to DOR would likely require a larger number of mutations than that to other β-lactams . Interestingly , another carbapenem , meropenem , targets the same PBPs as DOR ( Davies et al . , 2008 ) but has a higher resistance rate , suggesting that the underlying reasons for resistance rate variation are multifactorial . Taken together , effective resistance mutations against DOR seem to be less commonly available in P . aeruginosa in comparison to that against other drugs , including the here used CEF and CAR . A key determinant of treatment potency was the reduced level of spontaneous cross-resistance to the sequentially applied drugs ( Figure 5B ) . This effect was maximized by the switching rate ( Figure 5—figure supplement 1 ) . Our findings are consistent with the previously and repeatedly proposed importance of collateral sensitivity for the efficacy of sequential treatment protocols ( Barbosa et al . , 2019; Hernando-Amado et al . , 2020; Imamovic and Sommer , 2013; Kim et al . , 2014; Maltas and Wood , 2019; Yen and Papin , 2017 ) . Even though we did not measure collateral sensitivity directly , the lack of cross-resistance is related as it indicates that the mutant cells , which have become resistant to one drug , maintain at least ancestral levels of susceptibility against the second drug . Moreover , our study focused on spontaneous emergence of cross-resistance ( or lack thereof ) . By contrast , many previous studies established collateral effects after bacteria evolved resistance to the first drug over many generations , often followed by only a single antibiotic switch to assess the impact of collateral sensitivity on therapy success ( Barbosa et al . , 2019; Hernando-Amado et al . , 2020; Imamovic and Sommer , 2013; Yen and Papin , 2017 ) . Surprisingly , our study revealed potentially beneficial collateral effects between antibiotics of the same class . In fact , we chose these three β-lactams because our previous work demonstrated cross-resistance between most of them , although inferred upon multigenerational adaptation to the first drug ( Barbosa et al . , 2017 ) . Our current finding of a lack of cross-resistance among some of these drugs now suggests that spontaneous mutants may have different collateral profiles than the lines , which adapted over many generations . Our results further suggest that the collateral effects of spontaneous mutants are likely to be more pertinent for the design of sequential treatments with fast switches among antibiotics . This suggestion is supported by two previous studies , in which the efficacy of fast sequential treatments was optimized by considering collateral effects for either single-step mutants of S . aureus , obtained after 20 hr exposure to three distinct antibiotics for 20 hr ( Kim et al . , 2014 ) , or from Enterococcus faecalis populations adapted over 2 days to four distinct antibiotics ( Maltas and Wood , 2019 ) . As a side note , it is particularly interesting that our detailed resistance analysis consistently revealed almost all treatments to cause the evolution of collateral sensitivity towards the aminoglycoside gentamicin , but not the fluoroquinolone ciprofloxacin ( Figure 2B , C ) , possibly indicating yet another treatment option – in cases where the applied triple β-lactam sequential protocols fail . Temporal irregularity was additionally found to constrain bacterial adaptation . When bacteria experienced the antibiotics in an irregular pattern , this caused significantly increased extinction and to some degree reduced multidrug resistance . With CAR-DOR-CEF , the lowest multidrug resistance was observed in random sequential treatments ( Figure 2—figure supplement 5 ) , as also previously observed with CIP-GEN-CAR ( Roemhild et al . , 2018 ) . Environmental change anticipation has been documented in several microorganisms ( Mitchell et al . , 2009; Mitchell and Pilpel , 2011 ) , indicating their capability to specifically adapt to regular environmental change . Stochastic changes can make it harder to evolve anticipation ( Roemhild and Schulenburg , 2019 ) . Stochastic changes in environmental parameters were indeed found to constrain fitness in evolving bacteria ( Hughes et al . , 2007 ) and viruses ( Alto et al . , 2013 ) . We show that irregular antibiotic sequences have potential to inhibit bacterial resistance evolution . Unexpectedly , we further identified negative hysteresis for multiple combinations of the three β-lactams . However , cumulative hysteresis levels per treatment did not significantly associate with any of our measured evolutionary responses . In our previous study ( Roemhild et al . , 2018 ) , within the CAR-CIP-GEN combination , negative hysteresis was expressed for the switches from CAR to GEN and CIP to GEN . Yet , only the CAR-GEN hysteresis was significantly associated to the evolutionary responses . Thus , hysteresis interactions can exist between antibiotics from the same or different classes , but they need not impact the evolutionary outcome of a sequential treatment protocol each time . In the current study , it appears that spontaneous resistance effects and the resulting cross-resistance effects are dominant over the β-lactam hysteresis . One potential explanation could be that insensitivity to β-lactam hysteresis evolves quickly . Nevertheless , it clearly warrants further research to assess whether negative hysteresis between the β-lactam drugs is robustly shown across strains of P . aeruginosa or other bacterial species and can somehow be exploited in sequential therapy , in analogy to the previous results with antibiotics from different classes ( Roemhild et al . , 2018 ) . Taken together , our study highlights that the available antibiotics offer unexplored , highly potent treatment options that can be harnessed to counter the spread of drug resistance . It further underscores the importance of evolutionary trade-offs such as reduced cross-resistance in treatment design and introduces spontaneous resistance rates of component antibiotics as a guiding principle for sequential treatments . It is ironic that the differential cross-resistance landscape of the β-lactams was a key factor contributing to treatment potency , even though the risk of cross-resistance is usually used to reject β-lactam-exclusive treatments . The underlying reasons for differential spontaneous and long-term cross-resistance between these drugs ( including the underlying molecular mechanisms ) are as yet unknown and clearly deserve further attention in the future . We conclude that a detailed understanding of both spontaneous resistance rates and resulting cross-resistances against different antibiotics should be of particular value to further improve the potency of sequential protocols . All experiments were performed with P . aeruginosa UCBPP-PA14 ( Rahme et al . , 1995 ) . Bacteria were grown in M9 minimal medium supplemented with glucose ( 2 g/L ) , citrate ( 0 . 58 g/L ) , and casamino acids ( 1 g/L ) or on M9 minimal agar ( 1 . 5% ) or Lysogeny broth ( LB ) agar . Antibiotics were added as indicated . Cultures and plates were incubated at 37°C . Experiments included biological replicates ( initiated with independent clones of the bacteria , which were grown separately before the start of the experiment , or independent evolutionary lineages from the respective evolution treatments ) and technical replicates ( initiated from the same starting culture of the bacteria ) , as indicated below . For the experiments , treatment groups were run in parallel and randomized . Treatment names were masked in order to minimize observer bias . We used dose-response curves based on broth microdilution in order to determine antibiotic concentration causing inhibition level of 25% growth yield relative of untreated controls ( inhibitory concentration 75 [IC75] ) for the antibiotics azlocillin ( AZL ) , carbenicillin ( CAR ) , ciprofloxacin ( CIP ) , cefsulodin ( CEF ) , ceftazidime ( CTZ ) , doripenem ( DOR ) , gentamicin ( GEN ) , and ticarcillin ( TIC; see Supplementary file 1A for details on antibiotics ) . Briefly , bacteria were grown to exponential phase ( OD600 = 0 . 08 ) and inoculated into 96-well plates ( 100 µL per well , 5 × 106 CFU/mL ) containing linear concentration ranges close to MIC of the antibiotics in M9 medium . Antibiotic concentrations were randomized spatially . Bacteria were incubated for 12 hr after which optical density was measured in BioTek EON plate readers at 600 nm ( OD600 ) . We included six biological replicates and 1–2 technical replicates per concentration and antibiotic . Optical density was plotted against antibiotic concentration to obtain a dose-response curve . Model fitting was carried out using the package drc ( Ritz et al . , 2015 ) in the statistical environment R ( R Development Core Team , 2020 ) and the fitted curve was used to predict IC75 values ( Figure 1—figure supplement 1 ) . We carried out evolution experiments with the various combination of antibiotics according to the design described previously ( Roemhild et al . , 2018 ) . A total of 16 treatments were included ( Figure 1B ) . Treatments 1–4 were constant environments consisting of the monotherapy ( #1–3 ) and no drug control ( #4 ) . Treatments 5–10 were the regular switching treatments . They switched between the antibiotics in a regular predictable fashion , either every transfer ( fast; #5–7 ) or every fourth transfer ( slow; #8–10 ) . Treatments 11–16 consisted of the random treatments that switched fast in a temporally irregular fashion . The setup was designed to test the effect of switching rate and temporal irregularity . Every treatment consisted of 12 replicate populations ( initiated from six biological replicates × two technical replicates ) . All populations were started with an inoculum of 5 × 105 cells . Populations were propagated as 100 µL batch cultures in 96-well plates , with a transfer to fresh medium every 12 hr ( transfer size 2% v/v ) . Antibiotic selection was applied at IC75 throughout . We monitored growth by OD600 measurements taken every 15 min through the entire evolution experiment ( BioTek Instruments , USA; EON; 37°C , 180 rpm double-orbital shaking ) . Evolutionary growth dynamics were assessed by plotting the final OD achieved in every transfer ( relative to final OD of no drug control; relative yield ) . Adaptation rate was calculated with a sliding window approach , where adaptation rate was the inverse of the transfer at which the mean relative yield of a sliding window of 12 transfers reached 0 . 75 for the first time . Cases of extinction were determined at the end of the experiment by counting wells in which no growth was observed after an additional incubation in antibiotic-free medium . Samples of the populations were frozen in regular intervals in 10% ( v/v ) DMSO and stored at −80°C for later analysis . The evolution experiments were carried out for a total of 96 transfers . We characterized populations frozen at transfers 12 and 48 in detail because they represented the early and late phases of the evolution experiment . One population originating from a single biological replicate was chosen per treatment and plated onto LB agar . After incubation at 37°C , 20 colonies from each population were picked randomly and frozen in 10% ( v/v ) DMSO and stored at −80°C . These colonies , termed isolates , were considered to be representative biological replicates for each population . We constructed dose-response curves for the isolates using for each evolved population one technical replicate per isolate and four technical replicates of the ancestral PA14 strain , as described above , for the antibiotics CAR , CEF , DOR , GEN , and CIP . The integral of this curve for every isolate was calculated and the integral of the ancestral PA14 control subtracted . The resulting value was resistance of the isolate on the said antibiotic . We identified subpopulations in any given population by hierarchical clustering of the resistance profiles , as previously described ( Roemhild et al . , 2018 ) . Resistance of a population was calculated by averaging the resistance of the isolates . Resistance of the population on CAR , CEF , and DOR was added to obtain a single value for multidrug resistance . From the frozen isolates at transfer 12 , we chose three isolates per population ( i . e . , three biological replicates per population ) for whole-genome sequencing to determine possible targets of selection . Each resistance cluster in the population was represented in the sequenced isolates . For the DOR monotherapy , isolates from transfer 48 were also sequenced as no phenotypic resistance was observed at transfer 12 . Frozen isolates were thawed and grown in M9 medium at 37°C for 16–20 hr . We extracted DNA using a modified CTAB protocol ( von der Schulenburg et al . , 2001 ) and sequenced it at the Competence Centre for Genomic Analysis Kiel ( CCGA Kiel; Institute for Clinical Microbiology , University Hospital Kiel ) , using Illumina Nextera DNA Flex library preparation and the MiSeq paired-end technology ( 2 × 300 bp ) . Quality control on the resulting raw reads was performed with FastQC ( Andrews , 2010 ) and low-quality reads were trimmed using Trimmomatic ( Bolger et al . , 2014 ) . We then used MarkDuplicates from the Picard Toolkit ( http://broadinstitute . github . io/picard/ ) to remove duplicate reads and mapped the remaining reads to the P . aeruginosa UCBPP-PA14 genome ( available at http://pseudomonas . com/strain/download ) using Bowtie2 and samtools ( Langmead and Salzberg , 2012; Li et al . , 2009 ) . Variant calling was done using the GATK suite ( Poplin et al . , 2018 ) and the called variants were annotated using SnpEFF ( Cingolani et al . , 2012 ) and the Pseudomonas Genome Database ( https://www . pseudomonas . com/ ) . We removed all variants that were detected in the no drug control as they likely represent adaptation to the medium and not the antibiotic . The fasta files of all sequenced isolates are available from NCBI under the BioProject number: PRJNA704789 . The presence of cellular hysteresis was tested , following the previously developed protocol ( Roemhild et al . , 2018 ) . Bacterial cells were grown to exponential phase ( OD600 = 0 . 08 ) , diluted 10-fold , and treated with IC75 of the first antibiotic . In the treatments where the pretreatment did not require an antibiotic , none was added . These cells were allowed to incubate for 15 min at 37°C and 150 rpm ( pretreatment ) . After this , the first antibiotic was removed by centrifugation and fresh medium containing IC75 of a second antibiotic was added . In cases where the main treatment did not require an antibiotic , fresh medium without an antibiotic was added . Bacteria were now incubated for 8 hr at 37°C and 150 rpm ( main treatment ) . Bacterial count was monitored through the main treatment by spotting assays . We used three biological replicates per treatment and , for CFU counting , four technical replicates per biological replicate and treatment . Log10 CFU/mL were plotted against time to obtain time-kill curves ( Figure 3B ) . The level of hysteresis was calculated as the difference between the antibiotic switch and only main treatment curves . We determined the MIC on M9 agar for the antibiotics CAR , CEF , and DOR according to the EUCAST protocol ( https://doi . org/10 . 1046/j . 1469-0691 . 2000 . 00142 . x ) that was modified to account for inoculum effect in our fluctuation assay setup . UCBPP-PA14 was grown in M9 medium at 37°C for 20 hr . 5 × 105 cells were taken from the stationary phase cultures and spread on M9 agar plates containing doubling dilutions of the antibiotic . Plates were incubated at 37°C for 20–24 hr . MIC was read as the lowest concentration at which no growth of bacteria was seen . MIC determination for each antibiotic was done for three biological replicates ( no additional technical replication ) . We measured resistance rates on the three β-lactams using the classic fluctuation assay ( Luria and Delbrück , 1943 ) . Briefly , a single colony of UCBPP-PA14 was inoculated to 10 mL M9 and incubated at 37°C , 150 rpm for 20 hr . This primary culture was used to start 30 parallel cultures all having a starting concentration of 102 CFU/mL . The parallel cultures were considered biological replicates and incubated at 37°C , 150 rpm for 20 hr . Thereafter , 5 × 105 cells were plated onto MIC plates of CAR , CEF , and DOR . The plates were incubated for 40 hr at 37°C . The resulting mutant colonies were taken and patched on identical antibiotic MIC plates to ensure genetic resistance . Colonies that grew after patching were counted . We used counts from all 30 cultures to estimate resistance rate on each antibiotic using the package rSalvador ( Zheng , 2017 ) in R ( R Development Core Team , 2020 ) . We assessed the extent of cross-resistance associated with each β-lactam using the mutants obtained from the fluctuation assay . Sixty mutants with genetic resistance to a given β-lactam were considered biological replicates and patched onto MIC plates of the two other β-lactams . The patched plates were incubated for 16–20 hr at 37°C . If the mutant grew at MIC of the second β-lactam , it was counted as resistant . If it did not grow at the MIC of the second β-lactam , it was counted as susceptible . For each switch between two drugs , the fraction of cross-resistant mutants was calculated asNumberofmutantsthatgrewondrugBTotalmutantsisolatedondrugA To test whether the secondary antibiotic had an influence on the degree of cross-resistance of the mutants obtained from the fluctuation assay , we conducted a Fischer’s exact test followed by post hoc comparisons using the R package rcompanion ( Mangiafico , 2016 ) . The obtained p-values were then corrected for multiple testing using false discovery rate . To test whether main treatment types were associated with altered dynamics of adaptation in non-extinct populations , we analyzed the trajectories of relative growth yield ( as plotted in Figure 1E and Figure 2A ) of drug-treated populations using a GLM , including sequence ( ##1–16 ) and transfer as fixed factors and preculture and replicate population as nested random factors ( see Supplementary file 1B for details ) . Comparisons between main treatment groups were performed using pairwise post hoc tests and z statistics . All p-values were corrected for multiple testing using false discovery rate . The analysis was performed separately for the three time phases ‘early’ ( transfers 2–12 ) , ‘middle’ ( transfers 13–48 ) , and ‘late’ ( transfers 49–96 ) of the experiment , thus fulfilling the model assumption of response linearity . All statistical analyses were carried out in the statistical environment R ( R Development Core Team , 2020 ) . To test whether evolved populations displayed distinct multidrug β-lactam resistance depending on their main treatment type , we analyzed multidrug β-lactam resistance of evolved isolates – the sum of resistance values against CAR , CEF , and DOR ( as plotted in Figure 3B , C ) – using a GLM . The model included sequence ( ##1–16 ) as fixed factor and replicate population as nested random factor ( see Supplementary file 1C for detailed information ) . Comparisons between main treatment groups were performed using pairwise post hoc tests and z statistics . All p-values were corrected for multiple testing using false discovery rate . The analysis was performed separately for the ‘early’ ( after transfers 12 ) and ‘middle’ ( transfer 48 ) time points of the evolution experiment using the R statistical environment ( R Development Core Team , 2020 ) . To test whether our experimental ( switching rate and temporal irregularity ) and biological predictors ( hysteresis , probability of direct resistance , and cross effects ) were able to explain the variability in our evolutionary responses ( extinction , rate of growth adaptation , and multidrug resistance ) we carried out a GLM analysis . Values per treatment protocol for the biological predictors were calculated and the GLM analysis then carried out in R ( R Development Core Team , 2020 ) . We used the lm and anova commands and the main effects model: response ~ switching rate + irregularity for the experimental predictors and response ~ hysteresis + spontaneous resistance + mutant fraction cross-resistant for the biological predictors .	Overuse of antibiotic drugs is leading to the appearance of antibiotic-resistant bacteria; this is , bacteria with mutations that allow them to survive treatment with specific antibiotics . This has made some bacterial infections difficult or impossible to treat . Learning more about how bacteria evolve resistance to antibiotics could help scientists find ways to prevent it and develop more effective treatments . Changing antibiotics frequently may be one way to prevent bacteria from evolving resistance . That way if a bacterium acquires mutations that allow it to escape one antibiotic , another antibiotic will kill it , stopping it from dividing and preventing the appearance of descendants with resistance to several antibiotics . In order to use this approach , testing is needed to find the best sequences of antibiotics to apply and the optimal timings of treatment . To find out more , Batra , Roemhild et al . grew bacteria in the laboratory and exposed them to different sequences of antibiotics , switching antibiotics at different time intervals . This showed that sequential treatments with different antibiotics can limit bacterial evolution , especially when antibiotics are switched quickly . Unexpectedly , one of the most effective sequences used very similar antibiotics . This was surprising because using similar antibiotics should lead to the evolution of cross-resistance , which is when a drug causes changes that make the bacterium less sensitive to other treatments . However , in the tested case , cross-resistance did not evolve when antibiotics were switched quickly , thereby ensuring efficiency of treatment . Batra et al . show that alternating sequences of antibiotics may be an effective strategy to prevent drug resistance . Because the experiments were done in a laboratory setting it will be important to verify the results in studies in animals and humans before the approach can be used in medical or veterinary settings . If the results are confirmed , it could reduce the need to develop new antibiotics , which is expensive and time consuming .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2021	High potency of sequential therapy with only β-lactam antibiotics
Upon fertilization , the highly specialised sperm and oocyte genomes are remodelled to confer totipotency . The mechanisms of the dramatic reprogramming events that occur have remained unknown , and presumed roles of histone modifying enzymes are just starting to be elucidated . Here , we explore the function of the oocyte-inherited pool of a histone H3K4 and K9 demethylase , LSD1/KDM1A during early mouse development . KDM1A deficiency results in developmental arrest by the two-cell stage , accompanied by dramatic and stepwise alterations in H3K9 and H3K4 methylation patterns . At the transcriptional level , the switch of the maternal-to-zygotic transition fails to be induced properly and LINE-1 retrotransposons are not properly silenced . We propose that KDM1A plays critical roles in establishing the correct epigenetic landscape of the zygote upon fertilization , in preserving genome integrity and in initiating new patterns of genome expression that drive early mouse development . Gametes are highly differentiated cell types and fertilization of the oocyte by sperm requires major epigenetic remodelling to reconcile the two parental genomes and the formation of a totipotent zygote . In particular , the paternal genome arrives densely packed with protamines rather than histones , and the maternal epigenome is highly specialised . Maternal factors must unravel these specialised chromatin states to enable zygotic gene activation and development to proceed . Histone tail post-translational modifications ( PTMs ) , and more specifically lysine methylation , appear to be dynamically regulated during this first step of development ( Burton and Torres-Padilla , 2010 ) . Histone lysine methylation appears to have different biological read-outs , depending on the modified residue as well as the state of methylation ( mono- , di- or tri- ) . For example , methylation of histone H3 lysine 4 ( referred to as H3K4 methylation hereafter ) is mainly associated with transcriptionally active chromatin while methylation of histone H3 lysine 9 ( referred to as H3K9 methylation hereafter ) is usually linked to repressive chromatin . The incorrect setting of some of these histone marks in cloned animals have been correlated with their poor development potential , pointing to their importance during early stages of development ( Matoba et al . , 2014; Santos et al . , 2003 ) . However , the actors underlying these dynamic changes in histone modifications after fertilization and their impact on the appropriate regulation of zygotic genome function remain open questions . Lysine methylation is tightly regulated by distinct families of conserved enzymes , histone lysine methyltransferases ( KMTs ) , which add methyl groups and histone lysine demethylases ( KDMs ) which remove them ( Black et al . , 2012 ) . Importantly , KMTs and KDMs show different specificities for their target lysine substrates , as well as for the number of methyl group they can add or remove ( from unmethylated -me0- , to dimethylated –me2 , and trimethylated -me3; and vice versa ) . Several KMTs and KDMs have been disrupted genetically in model organisms , including mouse , and their loss often leads to lethality or to severe defects in embryogenesis , or else in tissue-specific phenotypes in adults . This has been linked to their important roles in cell fate maintenance and differentiation , as well as in genome stability ( Black et al . , 2012; Greer and Shi , 2012 ) . However , investigating their potential roles during the first steps of development , after fertilization is frequently hampered by their maternal mRNA and/or protein pool , which can persist during early embryogenesis and mask the potential impact that the absence of such factors might have ( Li et al . , 2010 ) . In mice , conditional knock-outs in the female germline that suppress the maternal store of mRNA and protein at the time of fertilization , can be used to examine protein function during the earliest steps of development ( de Vries et al . , 2000; Lewandoski et al . , 1997 ) . In this way , the roles of KMTs and KDMs during early embryogenesis are just starting to be explored . For example , it has been shown that depletion of maternal EZH2 affects the levels of H3K27 methylation in zygotes , although this did not lead to any growth defects during embryonic development ( Erhardt et al . , 2003; Puschendorf et al . , 2008 ) . Another study investigated maternal loss of Mll2 ( Mixed lineage leukemia 2 ) , encoding one of the main KMTs targeting H3K4 and revealed its essential role during oocyte maturation and for the embryos to develop beyond the two-cell stage , through gene expression regulation , ( Andreu-Vieyra et al . , 2010 ) . Importantly , in the presence of maternal EZH2 or MLL2 protein ( when wt/- breeders are used ) , both Ezh2 and Mll2 null embryos die much later in utero ( O'Carroll et al . , 2001; Glaser et al . , 2006 ) . The roles of these regulators of lysine methylation can thus be highly stage-specific , with very different effects at the zygote , early cleavage or later developmental stages . The LSD1/KDM1A protein ( encoded by the gene previously known as Lsd1 but subsequently renamed Kdm1a , which will be the used in this manuscript hereafter ) was the first histone KDM to be characterized to catalyse H3K4me1 and 2 demethylation and transcriptional repression ( Shi et al . , 2004 ) . KDM1A was later shown to demethylate H3K9me2 and to activate transcription ( Laurent et al . , 2015; Metzger et al . , 2005 ) . Genetic deletion of murine Kdm1a during embryogenesis obtained by mating of heterozygous animals showed early lethality prior to gastrulation ( Foster et al . , 2010; Macfarlan et al . , 2011 , Wang et al . , 2007; 2009 ) . In light of the above considerations , we set out to study the impact of eliminating or inhibiting the maternal pool of KDM1A during preimplantation development . We report for the first time the crucial role of Kdm1a following fertilization . The absence of KDM1A protein in zygotes derived from Kdm1a null oocytes led to a developmental arrest at the two-cell stage , with a severe and stepwise accumulation of H3K9me3 from the zygote stage , and of H3K4me1/2/3 at the two-cell stage . These chromatin alterations coincide with increased perturbations in the gene expression repertoire , based on single embryo transcriptomes , leading to an incomplete switch from the maternal to zygotic developmental programs . Furthermore , absence of KDM1A resulted in deficient suppression of LINE-1 retrotransposon expression , and increased genome damage , possibly as a result of increased LINE-1 activity . Altogether , our results point to an essential role for maternally-inherited KDM1A in maintaining appropriate temporal and spatial patterns of histone methylation while preserving genome expression and integrity to ensure embryonic development beyond the two-cell stage . To investigate whether Kdm1a might have a role during early mouse development we first assessed whether the protein was present in pre-implantation embryos using immunofluorescence ( IF ) and western blotting ( Figure 1A and B ) . A uniform nuclear localization of KDM1A within both parental pronuclei was observed by IF in the zygote , and at the two-cell stage . The protein was also readily detected by western blot analysis of total extracts of two-cell-stage embryos when compared to nuclear extracts of ESCs . Altogether , these data reveal the presence of a maternal pool of KDM1A . 10 . 7554/eLife . 08851 . 003Figure 1 . Kdm1a maternally deleted embryos arrest at two-cell stage . ( A ) Immunofluorescence using anti-KDM1A antibody ( red ) at the zygote and two-cell stage shows nuclear accumulation of KDM1A in control embryos ( top ) . Cre-mediated deletion of Kdm1a in maternal germline ( bottom ) leads to depletion of the protein after fertilization . Paternal pronucleus ( p ) , maternal pronucleus ( m ) and polar body ( pb ) are indicated . DNA is counterstained by DAPI ( blue ) . ( B ) western blot analysis ( left panel ) for ESC ( lane1 ) and two-cell stage embryo extracts ( lane 2 ) using anti-KDM1A antibody . Ponceau staining ( right panel ) is shown as loading control . Molecular weights ( kDa ) are indicated on the left . ( C ) Mating scheme and experimental outcomes for the different developmental stages used in this study: f/wt control embryos are obtained from superovulated Kdm1af/f females mated with wild-type males , while △m/wt mutant embryos are obtained from superovulated Kdm1af/f::Zp3cre females crossed with wild-type males . ( D ) Distribution of developmental stages found in f/wt and △m/wt embryos collected at embryonic day 2 ( E2 ) ( expected two-cell stage ) and after 24 hr of in vitro culture . Numbers of females used and numbers of oocytes/embryos analysedare shown under the graph . See also Figure 1—figure supplement 1 for oocyte analysisand Figure 1—figure supplement 2 for developmental stage distribution using natural matings without superovulation for females . ( E ) Bright field images representative for two consecutive days of in vitro culture for f/wt and △m/wt embryos collected at E2 . ( F ) Phenotypes and distribution of developmental stages obtained after 48 hr treatment in vitro culture with a catalytic inhibitor ( pargyline ) of KDM1A in wild-type zygotes recovered at 17 hr post hCG injection . Scale bars represent , 10 μm and 50 μm , in A and D , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 00310 . 7554/eLife . 08851 . 004Figure 1—figure supplement 1 . Kdm1a loss of function in female germline . ( A ) qPCR analysis of Kdm1a transcripts level from single ovulated oocytes cDNAs . Data are expressed as normalized mean expression levels ± sem for control ( in white , n = 7 ) versus mutant ( in black , n = 7 ) oocytes . Two asterisks indicate p<0 . 01 as calculated using a Student’s t-test . ( B ) IF analysis of control MII oocytes ( from Kdm1a f/f females; left ) or KDM1A depleted oocytes ( from Kdm1af/f::Zp3cre females; right ) with β-tubulin ( green ) , and metaphase chromosomes are counterstained with DAPI ( white ) . ( C ) Chromosome instability studied as a proportion of matured MII stages oocytes that exhibited normal alignment of chromosomes on the spindle versus matured MII stages oocytes with lagging chromosomes are shown in the graph . Oocyte genotypes are Kdm1af/f ( in white; n = 75 ) and Kdm1af/f::Zp3cre ( in black; n = 55 ) . ( D ) Proportions of two-cell stage embryos with micronuclei: f/wt control embryos are shown in white n = 60 , compared to Δm/wt mutant in black n = 40 ( C and D ) . Data are presented as the mean ± s . e . d of two or three independent experiments . Statistical difference was calculated using a chi-square test . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 00410 . 7554/eLife . 08851 . 005Figure 1—figure supplement 2 . Embryo recovery at day 2 post fertilization using natural matings . Distribution of developmental stages found in f/wt and Δm/wt embryos collected at embryonic day 2 ( E2; expected two-cell stage ) using natural breeding for Kdm1af/f or Kdm1af/f::Zp3cre females . Oocytes/embryos numbers , as well as the number of female per genotype are shown under the graph . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 005 To assess the function of KDM1A in early mouse embryo development , we deleted the Kdm1a gene in the female germline during oocyte growth . To this end Kdm1atm1Schüle Zp3cre females , carrying a new conditional allele for Kdm1a deletion engineered in the Schüle group ( Zhu et al . , 2014 ) , and a Zp3 promoter driven cre transgene exclusively expressed in oocytes ( Lewandoski et al . , 1997 ) were produced ( see also materials and methods ) . These animals are referred as Kdm1af/f::Zp3cre in this study ) . Kdm1af/f::Zp3cre females were then mated with wild-type males ( Figure 1C ) . We isolated one- and two-cell stage embryos derived from such crosses to obtain maternally depleted Kdm1a mutant embryos ( hereafter named △m/wt ) in parallel to control embryos ( hereafter named f/wt ) and we confirmed that the KDM1A maternal pool is absent by performing IF ( Figure 1A , bottom panels ) . In parallel , RT-qPCR analysis revealed the absence of Kdm1a mRNA in mutant oocytes ( Figure 1—figure supplement 1A ) . Numerous Kdm1af/f::Zp3cre females were housed with wild-type males for several months , however no progeny was ever obtained , in contrast to Kdm1af/f or f/wt females that produced the expected range of pup number ( 4 to 7; data not shown ) . This indicated that Kdm1af/f::Zp3cre females are sterile . To determine the possible causes of sterility , control Kdm1af/f and mutant Kdm1af/f::Zp3cre females were mated with wild-type males and embryos were recovered on embryonic day 2 ( E2 ) ( Figure 1D and E ) . The total number of oocytes or embryos scored per female was on average 17 for the mutant background ( 206 oocytes or embryos obtained for 12 females studied ) and 25 for the control ( 226 oocytes or embryos obtained for 9 females studied ) ( see Figure 1D ) . We found that the proportion of △m/wt two-cell stage embryos recovered ( 19% , n = 39/206 ) was far lower than that obtained with f/wt embryos ( 75% n = 170/226 ) ( Figure 1D ) . Using Kdm1af/f::Zp3cre females , we also noted a high percentage of fertilized and unfertilized oocytes blocked at meiosis II ( MII ) ( n = 95; 46% ) compared to those recovered from control females ( n = 34; 15% ) . Inspection of control ( n = 75 ) and mutant ( n = 55 ) MII oocytes revealed a high proportion of misaligned chromosomes on the metaphase spindle ( Figure 1—figure supplement 1B and C ) in mutants ( 41% ) compared to controls ( 17% ) , suggesting that a lack of maternal KDM1A can lead to chromosome segregation defects . Furthermore , upon fertilization , transmission of inherited chromosomal abnormalities was clearly evident , with the frequent presence of micronuclei in KDM1A maternally depleted two-cell embryos ( n = 40; 63% ) ( Figure 1—figure supplement 1D ) . Lastly , 19% ( n = 39 ) of mutant embryos were still at the zygote stage compared to 0% in controls ( Figure 1D ) . These results indicate that many MII oocytes lacking germline KDM1A are not competent at ovulation and that when fertilized their first cell cycle is delayed . Similar results were obtained when using females not subjected to superovulation for mating ( Fig1—figure supplement 2 ) . We next assessed the progress of surviving △m/wt two-cell embryos by culturing them in vitro . After 24 hr in culture , 96% of the △m/wt embryos were found to be arrested at the two-cell stage , unlike f/wt embryos where only 6% showed an arrest ( Figure 1D and E ) . The mutant embryos blocked at the two-cell stage did not progress further in development upon prolonged in vitro culture , and eventually fragmented , while the control embryos progressed towards the blastocyst stage ( Figure 1E ) . Taken together , these results suggest that the sterility of Kdm1a germline mutant females is in part caused by a severely compromised spindle organization in some oocytes in the second round of meiosis , as well as for the second round of cleavage after fertilization . This immediate loss of viability of the first generation embryos contrasts with the progressive effect seen across generations when spr-5 , the Kdm1a homologue in C . elegans is mutated in germline precursors for both gametes ( Katz et al . , 2009 ) . Also , targeted disruption of Lsd2/Kdm1b , the closest homologue of Kdm1a , in the mouse female germline , was reported to have no effect on oogenesis and early mouse development , but only later at mid-gestation , due to misregulation at some imprinted genes ( Ciccone et al . , 2009 ) . Our results show that KDM1B in the female germline is not sufficient to rescue the phenotype of KDM1A maternal depletion after fertilization . The developmental arrest observed at the two-cell stage of △m/wt embryos could be due to defects carried over by the mutant oocytes , particularly given the chromosome defects observed in a significant proportion of arrested oocytes , and/or to a requirement for KDM1A function after fertilization . To assess a requirement for KDM1A enzymatic activity in early embryos , we tested the impact of KDM1A catalytic inhibition specifically after fertilization . To this end , we treated wild-type zygotes with pargyline , a well-characterized potent chemical inhibitor of KDM1A enzymatic activity ( Fierz and Muir , 2012; Metzger et al . , 2005 ) and followed their development in vitro over 48 hr . As shown in Figure 1F , 76% embryos cultured with pargyline were found to be significantly blocked at the two-cell stage and 17% never progressed beyond the 3/4-cell stage . On the contrary , the majority ( 94% ) of mock treated embryos developed to the eight-cell stage within 48 hr , as expected . These data parallel the phenotype of genetic ablation of the KDM1A maternal pool , where 96% of △m/wt embryos are developmentally arrested at the two-cell stage and 4% at the 3/4-cell stage . Taken together , the genetic depletion and pargyline inhibition data strongly support a requirement for KDM1A enzymatic activity during the zygote and two-cell stage , for embryos to proceed beyond the two-cell stage . The above observations suggested that the histone demethylase KDM1A plays an important role in early development . At the zygote stage , H3K4 and H3K9 methylation levels appear to be tightly regulated and show highly parental specific patterns ( Arney et al . , 2002; Lepikhov and Walter , 2004; Santos et al . , 2005; Puschendorf et al . , 2008; Santenard et al . , 2010; Burton and Torres-Padilla , 2010 ) . Given that KDM1A has been implicated in the regulation of H3K4 and H3K9 mono and di methylation in previous studies ( Shi et al . , 2004; Metzger et al . , 2005; Di Stefano et al . , 2008; Katz et al . , 2009 ) , we investigated whether the methylation levels of these two histone H3 lysines were affected by KDM1A depletion in one-cell stage embryos . To this end , we collected f/wt and △m/wt embryos at embryonic day 1 and analysed them for both H3K4 and H3K9 methylation using specific antibodies against mono ( me1 ) , di ( me2 ) and tri ( me3 ) methylation ( Figure 2 ) . We used antibodies that show similar patterns in control zygotes to those previously published by others ( Arney et al . , 2002; Lepikhov and Walter , 2004; Puschendorf et al . , 2008; Santenard et al . , 2010; Santos et al . , 2005 ) ( see Material and Methods ) . We prioritised single-embryo analysis given the limited amount of material that can be recovered at these early developmental time points , particularly in the context of the depletion of KDM1A ( Figure 1D ) . We first analysed H3K4 methylation patterns ( Figure 2A and B ) . It was previously reported that the paternal pronucleus only gradually shows enrichment in H3K4me2 and me3 during the one-cell stage , while the female pronucleus is enriched with these marks from its oocyte origin ( Burton and Torres-Padilla , 2010; Lepikhov and Walter , 2004 ) . We compared maternal and paternal pronuclear patterns in control and mutant embryos and categorised them according to previously described nomenclature ( Adenot et al . , 1997 ) . In mid-stage zygotes , the absence of maternal KDM1A does not seem to affect overall H3K4me1 , me2 or me3 levels in either the maternal or paternal pronuclei ( Figure 2A and B ) . 10 . 7554/eLife . 08851 . 006Figure 2 . H3K9me3 heterochromatin levels are defined by maternally inherited KDM1A at the zygote stage . ( A and C ) IF using antibodies against me1 , me2 and me3 of ( A ) H3K4 ( in green ) and ( C ) H3K9 ( in red ) during zygotic development . Mid to late f/wt and △m/wt zygote are shown . Paternal pronucleus ( p ) , maternal pronucleus ( m ) and the polar body ( pb ) are indicated when present . DNA is counterstained with DAPI ( blue ) . In C , note that in △m/wt zygotes , H3K9me3 is increased in the maternal pronucleus ( grey arrowhead ) and is localized de novo in the paternal pronucleus ( yellow arrowhead ) . ( B and D ) Classification of embryos based on staining intensity scores for H3K4/K9me1/2/3 ) in the paternal versus maternal pronuclei in zygotes . Note that concerning H3K9me2 , 50% of △m/wt embryos have a strong staining versus 35% in controls ( which are also up to 20% with no IF signal ) . The most striking and only significant differences in proportions are seen for H3K9me3 both in maternal ( grey arrowheads ) and paternal ( yellow arrowheads ) pronuclei , with p<0 . 05 using a Chi square test . The scoring is as follows: light grey for no signal; medium green/red for moderate signal and dark green/red for strong signal . Number of embryos and their genotypes are indicated at the bottom of the graph . Scale bar in A and C represent 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 006 We also assessed whether H3K9 methylation levels were affected in zygotes lacking a maternal pool of KDM1A ( Figure 2C and D ) . H3K9me1 was reported to be equally enriched in both parental pronuclei , while H3K9me2 and me3 are exclusively present in the maternal pronucleus ( Arney et al . , 2002; Santos et al . , 2005; Lepikhov and Walter , 2004; Puschendorf et al . , 2008; Santenard et al . , 2010; Burton and Torres-Padilla , 2010 ) . We found that △m/wt embryos do not seem to differ from f/wt embryos in H3K9me1 levels ( Figure 2C and D ) . In the case of H3K9me2 , a complete absence of H3K9me2 staining in the paternal pronucleus was recorded for both control and mutant zygotes . However , we did note a small change in the proportion of embryos displaying H3K9me2 staining in the maternal pronucleus . This suggests that absence of KDM1A may slightly impact on oocyte-inherited H3K9me2 profiles . Although KDM1A was shown to specifically induce demethylation of H3K9me1/2 at target genes ( Laurent et al . , 2015; Metzger et al . , 2005 ) , we nevertheless assayed H3K9me3 patterns by IF , in case it could also accumulate in absence of KDM1A , due to the presence of specific H3K9 KMT ( Cho et al . , 2012; Puschendorf et al . , 2008 ) . H3K9me3 enrichment is a feature of constitutive heterochromatin , and has been shown to be zygotically enriched at the periphery of nucleolar like bodies ( NLBs ) within the maternal but not the paternal pronucleus ( Burton and Torres-Padilla , 2010; Puschendorf et al . , 2008; Santenard et al . , 2010 ) ( Figure 2C ) . In △m/wt zygotes , strikingly elevated levels H3K9me3 were found in the whole maternal pronucleus when compared to controls ( grey arrowhead , Figure 2C and D ) . Even more surprisingly , in △m/wt zygotes , H3K9me3 could be detected at the periphery of paternal NLBs ( yellow arrowhead ) , when compared to controls . Taken together , these observations show that the absence of maternal KDM1A protein results in specifically elevated levels of H3K9me3 in both parental genomes at the zygote stage , and suggest that KDM1A might be engaged with other chromatin modifiers to regulate H3K9me3 immediately after fertilization . In order to investigate whether KDM1A activity was important for the regulation of H3K4 and K9 methylation after the first cell cleavage , we examined two-cell stage f/wt and △m/wt embryos by IF to measure the relative fluorescence intensities ( Figure 3 ) . We found that the overall H3K4 methylation levels for mono , di and tri-methylation were significantly elevated in △m/wt two-cell embryos ( Figure 3A ) , with the most striking effect being seen for H3K4me3 where a six-fold increase was found in mutants compared to controls . Thus , a lack of KDM1A protein has a significant impact on H3K4 methylation levels at the two-cell stage . When H3K9me1 , me2 and me3 levels were also examined by IF , we found that all three marks were elevated , with the most significant effect being seen for H3K9me3 , which showed a 2 . 2 fold increase in fluorescence intensity particularly at DAPI dense regions of constitutive heterochromatin ( Figure 3B ) . 10 . 7554/eLife . 08851 . 007Figure 3 . Two-cell stage H3K4 and H3K9 methylation levels are altered upon absence of maternal KDM1A . Immunofluorescence stainings of two-cell stage embryos using antibodies against me1 , me2 and me3 of H3K4 ( A; in green ) and H3K9 ( B; in red ) were performed on f/wt ( left panels ) and △m/wt embryos ( right panels ) . Control and mutant samples were processed in parallel and acquired using similar settings at the confocal microscope . DNA is counterstained with DAPI ( blue ) . Projections of z-stacks are shown of representative embryos for each staining . Scale bars , 10 μm . Error bars represent S . E . M . By t-test; p<0 . 05 corresponds to * and p<0 . 001 to ** as performed on the number of embryos indicated below each picture . Below each image are shown the relative quantifications for IF intensity levels of me1 , me2 and me3 of △m/wt ( in black ) relative to f/wt ( in white ) in two-cell stage embryos . Note that no alteration for H3K27me3 or H4K20me3 could be detected for mutant two-cell stage embryos ( Figure 3—figure supplement 1A and B ) . Also , IF for pargyline-treated two-cell stage embryos revealed changes in both H3K4me3 and H3K9me3 patterns ( Figure 3—figure supplement 1C and D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 00710 . 7554/eLife . 08851 . 008Figure 3—figure supplement 1 . Immunofluorescence analysis of histone tail modifications upon maternal depletion or upon chemical inhibition of KDM1A for two-cell stage embryos . ( A and B ) Analysis of heterochromatin marks by IF with antibodies against me3 of H3K27 ( A ) and H4K20 ( B ) were performed on f/wt control embryos ( left panel ) in parallel to Δm/wt mutant embryos ( right panel ) . ( C and D ) Analysis by IF with antibodies against H3K4me3 ( C ) and H3K9me3 ( D ) of two-cell stage wild-type embryos cultured in mock conditions ( left panel ) or with pargyline ( right panel ) . Antibody staining is shown in green or red and DNA is counterstained with DAPI ( blue ) . The number of embryos processed is indicated under each picture . Shown are full projections of stack sections taken every 0 . 5 μm . Scale bar , 10 μ . Under each image is the graphical representation of the fluorescent mean intensity ± sem ( arbitrary unit , AU ) of Δm/wt ( in black ) relative to f/wt ( in white ) two-cell stage immunostained embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 008 To address the specificity of these effects of KDM1A on H3K4 and H3K9 methylation , we tested other histone marks , reported not to be targeted by KDM1A activity . Two such marks , H3K27me3 and H4K20me3 , both associated with heterochromatin , were analysed by IF in f/wt and △m/wt two-cell stage embryos . No significant changes in either of these marks could be detected in mutant compared to control embryos ( Figure 3—figure supplement 1A and B ) , underlining the specificity of the defects found in KDM1A maternally depleted embryos . As an additional control , we performed IF analysis of two-cell stage embryos generated from wild-type zygotes grown for 24 hr with pargyline . H3K4me3 and H3K9me3 patterns revealed changes in pargyline-treated when compared to mock-treated embryos ( Figure 3—figure supplement 1C and D ) . In both , a global increase in staining was detected when compared to controls , although to a slightly lesser extent than in Kdm1a mutant embryos . After fertilization , development initially proceeds by relying on the maternally inherited pool of RNA and protein , followed by massive induction of transcription of the zygotic genome in different waves as shown in Figure 4A . Newly produced transcripts corresponding to zygotic genome activation ( ZGA ) appear in two phases , first at the zygote stage ( corresponding to minor ZGA ) and subsequently at the two-cell stage ( major ZGA ) . Transition from the maternal pool to zygotic products is essential for successful developmental progression ( Flach et al . , 1982 ) . Previous work has shown that KDM1A affects transcription regulation during in vitro embryonic stem cell ( ESC ) differentiation or during peri- or post-implantation mouse development ( Foster et al . , 2010; Macfarlan et al . , 2011; Wang et al . , 2007; Zhu et al . , 2014 ) . However , its role has never been evaluated during the very first steps of embryogenesis , when appropriate transcriptional activity is crucial . 10 . 7554/eLife . 08851 . 009Figure 4 . Abnormal ZGA upon absence of KDM1A revealed by transcriptome analysis . ( A ) Schematic illustration of the sequential sources of RNA pool over embryonic development . ( B ) Histogram shows the percent of differentially expressed genes in the △m/wt versus f/wt embryos . Fold difference ( in log2 ) is annotated as upregulated ( with log2≥ 1; yellow ) , downregulated ( as log2≤-1; green ) and similar ( as 1<log2>-1; grey ) . Number of genes is indicated on the right of the graph . Details concerning the RNA seq analysis are described in Materials and Methods section and Supplementary file 1 ( C ) Hierarchical clustering analysis for gene expression pattern of 16 libraries shows dramatic expression changes between f/wt ( floxE1 to E8 ) and △m/wt ( △mE1 to E8 ) two-cell stage embryos . See also Figure 4—figure supplement 2 for analysis between two-cell stage and oocyte transcriptomes ( D ) RNA-seq data comparison with the different categories of the gene catalogue available at the Database of Transcriptome in Mouse Early Embryos ( DBTMEE ) generated an the ultralarge-scale transcriptome analysis ( Park et al . , 2013 ) . The total number of genes belongings to each class and found in our RNA seq is indicated on top of the graph ( see also Table 1 ) . ( E ) Graphical representation of the normalized mean expression levels ± sem for chromatin-encoding genes in f/wt ( in white ) or △m/wt ( in black ) MII oocytes ( Oo , n = 7 ) and two-cell stage embryos ( 2C , n = 10 ) . corresponds to p<0 . 05 and * to p<0 . 001 . ( F ) Top 6 representative GO terms ( biological functions ) enriched in △m/wt mutant embryos . Fold overrepresentation indicates the percentage of misregulated genes in a particular category over the percentage expected on the basis of all GO-annotated genes present within the sequencing . p-value indicates the significance of the enrichment . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 00910 . 7554/eLife . 08851 . 010Figure 4—figure supplement 1 . Immunostainings and RTqPCR analysis for assessing transcription of Kdm1a mutant two-cell stage embryos . ( A , B ) Example of IF using anti-PolIICTD antibodies ( A ) or anti-PolII Ser2P ( B ) for f/wt control and Δm/wt mutant two-cell stage embryos with below the mean ± s . e . m fluorescence graphical representation ( AU ) of Δm/wt mutant embryos ( in black ) relative to f/wt control embryos ( in white ) . Number of processed embryos is indicated under each image . Antibody signal in red , DAPI is blue . Scale bar , 10 μm . ( C ) Graphical representation of qPCR analysis for individual oocyte or two-cell stage embryos ( f/wt control in white , n = 10 or Δm/wt mutant in black , n = 11 ) plotting the mean expression levels ± sem of three housekeeping genes Gapdh , Hprt and Ppia ( according to GeNorm application ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 01010 . 7554/eLife . 08851 . 011Figure 4—figure supplement 2 . Transcriptome analysis of Kdm1a mutant versus control in oocytes or two-cell stage embryos . ( A ) Venn diagrams show the numbers of differentially expressed genes in absence of KDM1A , when considering a fold difference ( in log2 ) annotated as upregulated ( with log2≥ 1 ) or downregulated ( with log2≤-1 ) for two-cell stage embryos or ovulated oocytes ( B ) Principal component analysis for gene expression pattern of all libraries shows dramatic expression changes between f/wt and Δm/wt two- cell stage embryos , as well as with oocytes from both genotypes . ( C ) Venn diagram between oocyte stage or two-cell embryos for upregulated genes or down regulated genes ( when comparing mutants to controls ) . ( D ) Top 6 representative GO terms ( biological functions ) enriched in oocytes maternally deficient for KDM1A . Fold overrepresentation indicates the percentage of misregulated genes in a particular category over the percentage expected on the basis of all GO-annotated genes present within the sequencing . p-value indicates the significance of the enrichment . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 01110 . 7554/eLife . 08851 . 012Table 1 . Comparing the two-cell stage transcriptome of the Kdm1a mutant embryos to DBTMEE Numbers of genes found for the comparison of our two-cell stage RNA-seq data with the different categories for the gene catalogue found in DBTMEE . Total genes considered = 3811 and total genes changed = 1818 ( 48% ) . Our dataset cover the genes categorized on the public resource with a minimum of 96% of genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 012UpDownSimilarNot in our dataTotal in DBTMEEmaternal3606352131975minor ZGA540477504013771C transient34451190major ZGA10297297105842C transient815615612329MGA1328628613563 In the light of our results on chromatin changes described above , and to assess whether transcription might be affected by the lack of the KDM1A maternal pool , we performed IF analysis against PolII and its elongating form ( PolIISer2P ) , which did not reveal any obvious difference between f/wt and △m/wt two-cell stage embryos ( Figure 4—figure supplement 1A ) . For a direct comprehensive analysis of the transcriptome upon lack of KDM1A , we used RNA sequencing ( RNA-seq ) for single oocytes and single embryos at the two-cell stage on a cohort of control and mutant samples ( Figure 4; Figure 4—figure supplement 2; Supplementary file 1 ) . The method used is based on that of ( Tang et al . , 2010 ) which captures poly ( A ) tail mRNA and allows examination up to 3kb from the 3’ end . The quality of our single oocyte or embryo cDNAs was first checked by qPCR for three housekeeping genes ( Hprt , Gapdh , Ppia ) known to be stably expressed from oocytes to blastocysts ( Mamo et al . , 2007; Vandesompele et al . , 2002 ) . Control and mutant samples displayed similar relative expression for these three genes attesting to the quality of our samples ( Figure 4—figure supplement 1C ) . We next prepared cDNA libraries from single control and mutant oocytes ( n = 5 each ) , as well as individual f/wt and △m/wt embryos ( n = 8 each ) and performed Illumina-based deep RNA sequencing on these samples ( see experimental procedures and analysis for more details; Supplementary file 1 ) . We used DEseq as a normalization method across our samples to assess the relative gene expression between controls and mutants . At the two-cell stage , our analysis revealed two sets of genes that become either upregulated ( 21%; n = 2449; FDR = 5% ) or downregulated ( 24%; n = 2749 ) in the mutant when compared to control embryos ( Figure 4B; Figure 4—figure supplement 2A ) . Hierarchal clustering based on the transcription profiles showed that all Kdm1a mutant embryos clustered distinctly from the controls ( Figure 4C ) . Furthermore , the analysis of oocyte transcriptomes also revealed that there were fewer genes misregulated in Kdm1a mutant oocytes , than in Kdm1a mutant embryos ( Figure 4—figure supplement 2A ) . Moreover , Principal Component Analysis demonstrated that the gene expression patterns showed greater differences between controls and mutants at the two-cell stage , than in oocytes , and that the two different stages cluster away from each other ( Figure 4—figure supplement 2B ) . The stage comparison also showed that only a subset of genes were misregulated in common , in both oocytes and two-cell stage embryos , upon loss of maternal KDM1A ( Figure 4—figure supplement 2C ) . GO analysis of up or down regulated genes at the two-cell stage ( Figure 4F ) or oocytes ( Figure 4—figure supplement 2D ) revealed very little overlap in the specific biological functions affected by loss of function of KDM1A before and after fertilization , with the notable exception of cell cycle associated genes . This connects well with the observed phenotype for poor oocyte competence at fertilization and the total developmental arrest at the two-cell stage . These results reveal that absence of maternal KDM1A most likely leads to transcriptome changes during oocyte maturation , but to even more serious defects after zygotic gene activation , at the two-cell stage . The latter may be due in part to an aberrant maternal supply of transcripts/proteins , or else to aberrant transcriptional regulation of the zygotic genome in absence of maternal KDM1A . We assessed our two-cell stage RNA-seq data according to the recent Database of Transcriptome in Mouse Early Embryos ( DBTMEE ) ( Park et al . , 2013 ) . DBTMEE was built from an ultra-large-scale whole transcriptome profile analysis of preimplantation embryos , in which genes are classified depending on which transcription waves ( as in Figure 4A ) they are expressed . As shown in Figure 4D ( see also Table 1 ) , we assessed the percentage of genes of each of our classes ( up; down and not significantly changed ) that overlapped with the different DBTMEE categories of transcription switches , from oocyte to two-cell stage . Strikingly , the upregulated genes in △m/wt embryos fall essentially into the earliest stages and belong to genes annotated as maternal ( 37% of this category ) , as minor ZGA genes ( 39% ) and as zygotic-transient ( 38% ) . We checked whether the misregulation of these three categories of genes might originate from the oocyte stage changes . We found that only 56 out of 360 of maternal genes , 94 out of 540 of minor ZGA genes and 6 out of 34 of 1C transient ( Table 1 and data not shown ) were already upregulated in mutant oocytes . These results reinforce the conclusion that the maternal and zygotic pools of transcripts become more compromised as development proceeds toward the two-cell stage in mutant embryos , rather than being aberrant right from the Kdm1a mutant germline . In clear contrast , the majority of downregulated genes in the △m/wt were found to belong to the three categories of genes that are normally activated at the two-cell stage , with 50% in the major ZGA class , 37% in the two-cell transient and 50% in the MGA ( Mid zygotic gene activation ) . This suggests that absence of KDM1A compromises the activation of gene expression by the two-cell stage . In order to validate our RNA seq data and the analysis done , we selected four genes with characteristic expression profiles , Atrx ( maternal ) , H2Az ( major ZGA ) , Suv39h1 ( 2C-transient ) , Klf4 ( MGA ) , which all encode chromatin associated factors crucial for early mouse development . Validation was performed by RT-qPCR in control and mutant oocytes and two-cell embryos . As predicted from our RNA seq results ( Supplementary file 2 ) , H2AZ , Suv39h1 and Klf4 failed to be expressed at two-cell stage in Kdm1a mutant embryos ( Figure 4E ) . In contrast Atrx which is a known maternal factor , but which is zygotically expressed by the two cell stage , was correctly activated . No difference in expression of Suv39h1 , Klf4 and Atrx could be seen between controls and mutants at the oocyte stage , implying that the maternal pool of these mRNAs was not affected by the maternal KDM1A depletion . This single embryo transcriptome profiling data reveals an aberrant gene expression profile in Kdm1a mutant embryos , which is likely due to an absence or delay in the transcription switch from maternal-zygote to the two-cell stage pattern for a substantial set of genes ( 47%; 1818 out of 3811 considered; Figure 4—figure supplement 2 ) . Together with the changes in chromatin profiles that we observed at the two-cell stage , we conclude that part of the deficiency in developmental progression could be due to the inappropriate setting of a successful zygotic gene expression program upon KDM1A loss . A gene ontology ( GO ) analysis of the up-regulated genes classified as maternal to zygote-transient in Figure 4D , revealed a clear over-representation of genes involved in protein transport and localisation as well as contribution to cell cycle ( Figure 4F ) . GO analysis of the downregulated genes from major-to-mid-zygotic activation are implicated in ribosome biogenesis and translation processes ( Figure 4F ) . Collectively , these results suggest that KDM1A is necessary for the transcriptional regulation of specific genetic pathways implicated in fundamental biological functions such as protein production and localisation , and cell cycle regulation . These combined defects could be consistent with the inability of the mutant embryos to develop further than the two-cell stage Many transposable elements are known to be expressed in early mouse embryos , as early as zygotic stage , and some of these repeat elements might even be competent for new events of retrotransposition between fertilization and implantation ( Fadloun et al . , 2013; Kano et al . , 2009; Peaston et al . , 2004 ) . The repression of some of these transposable elements during preimplantation has been correlated with loss of active chromatin marks such as H3K4me3 , rather than acquisition of heterochromatic marks such as H3K9me3 ( Fadloun et al . , 2013 ) . Interestingly , a previous study using Kdm1a mutant mESCs and late preimplantation embryos found a significant impact on MERVL:LTR repeat expression ( for Murine endogenous retrovirus-like LTR ) , as well as a good correlation for the presence of remnant ERVs within 2kb of the transcription start site of KDM1A-repressed genes ( Macfarlan et al . , 2011 ) . The increased levels of H3K4me3 and H3K9me3 that we found in △m/wt two-cell stage embryos and the reported role of KDM1A in late preimplantation embryos prompted us to analyze the effects of maternal KDM1A depletion on repetitive element expression after fertilization . To this end , we investigated our RNA-seq data from control and Kdm1a mutant two-cell embryos for the relative expression of repetitive elements . As our single embryo RNA seq approach was based on oligo-dT priming this restricted our analysis to reads at the 3’ ends of transcripts , which somewhat limited our capacity to detect repeat variation . In particular we could not determine which specific LINE-1 families were expressed in the mutants , nor whether the LINE-1 reads we detected corresponded to full-length , and/or intact elements . Nevertheless , our results shows that by far the most abundant categories of expressed repeats at this stage of development were LTRs ( long terminal repeat ) and non-LTR retrotransposons in f/wt and △m/wt ( 95% and 92% , respectively ) ( Figure 5A ) . However , no significant impact on expression could be detected in the mutants , with the exception within the non-LTR elements , of quite a significant overrepresentation of LINEs , but not SINEs ( for Long/Short Interspersed Nuclear Elements element ) ( Figure 5A and B ) . We validated this result by RT-qPCR using individually prepared cDNAs of two-cell stage embryos for three transposable element classes . LINE-1 , SINE B1 and MuERV-L transcripts are all abundantly expressed in control and mutant embryos , but LINE-1 levels show a two-fold increase in the △m/wt embryos ( Figure 5C ) . No significant up-regulation was seen in ERV-promoter driven genes , that had previously reported to be affected by loss of KDM1A in ESCs ( Macfarlan et al . , 2011 ) . 10 . 7554/eLife . 08851 . 013Figure 5 . Increased LINE-1 protein levels and γH2AX foci in two-cell embryos depleted for KDM1A . ( A ) Pie chart representing the percent of each category of repeats analyzed in our 16 RNA-seq data of individual embryos . ( B ) Box-plot for percent of LINEs , SINEs and LTR element expression for f/wt ( white ) or △m/wt ( black ) embryos over the total of reads mapping repeats for each of our 16 samples of RNA-seq . Details of the analysis are in experimental analysis . ( C ) qPCR analysis for LINE-1 , SinesB1 and MuERV-L expression levels from individual two-cell stage cDNAs of f/wt ( white ) or △m/wt ( black ) . Each embryo is represented as a single bar . Data are expressed as normalized expression to three house-keeping genes . On the right of each graph is represented the mean ± sem . Two asterisks indicate p<0 . 01 as calculated using a Student’s t-test . ( D ) Nascent LINE-1 transcripts are detected by RNA FISH ( signal in red ) using a TCN7 probe on f/wt or △m/wt two-cell stage embryos . RNAse A treated control embryos processed in parallel display no signal for RNA transcription . Also see Figure 5—figure supplement 1 for Atrx expression control . ( E ) Quantification of LINE-1 RNA FISH . On the left , the graph represents the fluorescent quantification with the mean intensity of fluorescence plotted on the x axis against the respective maximum intensity on the y axis ) for each nucleus of the two–cell embryos for the two populations ( white squares = f/wt controls and black square = △m/wt mutants ) . On the right , box-plot representation of the entropy levels analysis of the RNA FISH images for the control versus mutant embryos as defined by Haralick parameters measuring the pattern of the image with each dot corresponding to a f/wt ( white ) or △m/wt ( black ) nucleus . P value is calculated with a student T-test and indicated that the two populations are significantly different . ( F ) IF of two-cell stage embryos using anti-ORF1 antibodies ( in red ) . A dotted line indicates the nucleus . Below is the graphical representation of the percentage of embryos displaying enriched fluorescent signal in either the cytoplasm ( cy ) or the nucleus ( nu ) for f/wt or △m/wt embryos . ( G ) IF of two-cell stage embryos using antibodies directed against phosphorylated histone H2A variant X ( γH2AX , in green ) for f/wt and △m/wt . Below is the corresponding quantification of embryo percentage according to the strength of γH2AX staining . DNA is counterstained by DAPI ( blue ) . Number of processed embryos is indicated . Scale bar , 2 μm ( D , G ) ) 10 μm ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 01310 . 7554/eLife . 08851 . 014Figure 5—figure supplement 1 . RNA FISH controls for LINE-1 ongoing transcription . RNA FISH using a probe recognizing the Atrx genomic locus ( signal in red; DAPI is blue ) on two-cell stage f/wt control embryo ( left ) and Δm/wt mutant embryo ( right ) . Numbers of embryo processed and percentage of nuclei showing pinpoints of nascent transcripts by RNA FISH assays are indicated under each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 01410 . 7554/eLife . 08851 . 015Figure 5—figure supplement 2 . EdU labeling and γH2AX immunofluorescence of Kdm1a mutant versus control two-cell stage embryos . ( A ) Two-cell stage embryos ( f/wt and Δm/wt ) were collected at 40–41 hr post hCG and culture for EdU pulse for 45 min . After fixation , they were analyzed by Click -It reaction to reveal the incorporated EdU ( shown in green ) , then subjected to immunostaining for γH2AX ( shown in red ) . DNA is counterstained with DAPI ( in blue ) . Shown are maximal projection of confocal z series of representative embryos . f/wt control embryo n = 17 and Δm/wt mutant embryo n = 11 . Scale bar , 10 μm . ( B ) quantification of embryo percentage according to the presence of EdU incorporation or γH2AX immunofluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 08851 . 015 To further assess the impact that KDM1A depletion has on active LINE-1 transcription , we used a single-cell method , RNA fluorescent in situ hybridization ( RNA FISH ) , which enables the detection of nascent transcripts . We first assessed the quality of our assay by checking the Atrx gene , known to be transcribed at the two-cell stage ( Patrat et al . , 2009 ) and expressed at similar levels in mutant and control ( Figure 5—figure supplement 1 ) . A comparable proportion of f/wt embryos and △m/wt two-cell embryos displayed detectable ongoing transcription , as registered by a pinpoint at this locus . Using a probe spanning the full-length LINE-1 element ( Chow et al . , 2010 ) , we detected LINE-1 RNA in control two-cell stage embryos as displayed by the punctate pattern in nuclei ( Fadloun et al . , 2013 ) , while RNase-A treated embryos showed no signal ( Figure 5D ) . In the maternally depleted embryos , the arrangement of fluorescent foci appeared extensively modified ( Figure 5D ) . This was confirmed upon analysis of the fluorescence intensity distributions ( Figure 5E left ) as well as the image composition for the foci ( Figure 5E right ) , which in both cases significantly separated the two types of samples . Our analysis revealed that the active LINE-1 transcription profiles were extensively modified upon the loss of maternal KDM1A . To investigate whether the increase in nascent LINE-1 transcription observed might correspond to full length LINE-1 elements , we assessed by IF for the presence of ORF1 , one of the two LINE-1 encoded proteins . At the two-cell stage , we found an approximately four-fold increase in the proportion of △m/wt embryos displaying a stronger IF signal , notably in the nucleus ( Figure 5F ) . These results suggest that the LINE-1 deregulation observed at the RNA level might indeed lead to the production and nuclear import of increased levels of LINE-1 ORF1 proteins . We next investigated whether expression of such proteins from transposable elements would have any consequences . We thus performed γH2AX IF staining to assess whether increased DNA damage signalling could be seen in △m/wt compared to f/wt embryos ( Figure 5G; Figure 5—figure supplement 2 ) . Half of the mutant embryos displayed a stronger staining for γH2AX , with a significant increase compared to controls ( Figure 5G ) . We also assessed whether this accumulation of γH2AX signals could also be related to replication delays , as reported previously in the case of maternal loss of two components of the polycomb complex PRC1 ( Posfai et al . , 2012 ) . We performed EdU pulse treatment ( a nucleoside analog of thymidine incorporated into DNA ) in two-cell embryos , at a stage when they have normally completed S phase ( 40–41 hr post hCG injection ) . This revealed that S phase is delayed in the △m/wt embryos given the incorporation of EdU in the mutants , while none of the control embryos used in parallel were stained ( Figure 5—figure supplement 2 ) . All the mutant embryos delayed in their replication displayed concomitantly intense γH2AX signals . However , 38% of the mutant embryos did not show any EdU incorporation , indicating that they exit S phase , yet , they still show high levels of γH2AX signals . Finally , although , no significant enrichment was directly found for DNA damage pathways when running our GO analysis ( Figure 4E ) , many genes related to DNA damage repair were upregulated ( Supplementary file 2 ) . Taken all together , these results suggest that the elevated DNA damage signalling observed could be independent from replication defaults in KDM1A maternally depleted embryos , but might be related either to changes in transcript levels for DNA damage genes or else to the observed increase in LINE-1 activity in Kdm1a mutant embryos at this stage . H3K4me1/2/3 levels have been shown to increase from the zygote to the two-cell stage , before decreasing again by the four-cell stage ( Shao et al . , 2014 ) . To date , only one H3K4 KMT , MLL2 , has been shown to be necessary at the two-cell stage ( Andreu-Vieyra et al . , 2010 ) . Here , we report that the maternal pool of the KDM , KDM1A , is also necessary at this stage , with its loss leading to global elevation of H3K4me1/2/3 . Noticeably , no changes in transcription levels of genes encoding the main H3K4me2/3 KMTs were recorded ( Supplementary file 2 ) . This suggests that KDM1A is a key regulator of H3K4 methylation post-fertilization . Moreover , in absence of KDM1A , the transcripts encoding for two main KMTs ( SUV39H2/KMT1B and SETDB1/KMT1E ) targeting H3K9me1/2 during preimplantation ( Cho et al . , 2012; Puschendorf et al . , 2008 ) are well detected in two-cell stage mutant embryos ( Supplementary file 2 ) . These KMTs could be able to generate H3K9me3 from the excess of H3K9me1/2 , produced because of the absence of KDM1A . In conclusion , KDM1A most likely acts in combination with other chromatin regulators in order to keep a tight balance of the global H3K4/K9 methylation levels during early embryonic development . One of the most striking consequences of lack of maternal KDM1A that we observed was the disruption of the wave-like gene expression patterns previously described at the onset of mouse development ( Hamatani et al . , 2004; Xue et al . , 2013 ) . At the two-cell stage , we saw a significant increase in mRNA levels of genes normally expressed maternally or at the zygote stage , and this increase relates more to post-fertilization disruption rather than inherited defects from Kdm1a mutant germline . The accompanying manuscript by Wasson et al reports similar findings concerning transcriptional regulation by maternal KDM1A in early stage post fertilization . Maternal and zygotic mRNA excess could reflect a reduced rate of mRNA degradation , maybe related to the developmental arrest , or else the severe impairment of the mutant embryos in the ribosome biogenesis pathways could preclude the translation machinery of their usage and clearance or else a change in the cytoplasmic polyadenylation of the maternal pool of mRNA could also be disturbing their utilization . Lastly , their abundance could also be due , maybe partly , to an increased transcription rate for these genes ( and more specifically the one corresponding to the minor ZGA ) . Given the accumulation of H3K4 methylation that we show in our study for the mutants at this stage and the proven link of this mark with enhanced transcription ( Black et al . , 2012 ) , we hypothesise that KDM1A might normally be involved in the transcriptional down-regulation of these genes via H3K4 demethylation . Chromatin-based repression is thought to be superimposed on zygotic genome activation and is necessary for the transition from the two-cell to the four-cell stage ( Ma , 2001; Ma and Schultz , 2008; Nothias et al . , 1995; Wiekowski et al . , 1997 ) . We propose that KDM1A might be part of such a mechanism , and required for a transition towards two-cell stage specific gene expression patterns ( ie in the major ZGA and MGA waves ) , and therefore for proper development beyond the two-cell stage . The significant absence of the major ZGA and MGA waves in the transcriptome of Kdm1a mutants supports this hypothesis . Whether misplaced or increased H3K9 methylation ( Figure 3B ) could be involved in failure of transcription activation is not known , but one can speculate that such repressive chromatin and/or absence of KDM1A itself might impair correct recruitment of transcription regulators . So far , a small subset of such factors ( TFs and co-regulators ) acting at ZGA-gene promoters has recently been suggested to orchestrate the appropriate gene expression patterns following fertilization ( Park et al . , 2013; Xue et al . , 2013 ) . Although , KDM1A was not reported in this study , our results suggest that maternal KDM1A is nonetheless crucial for shaping the transcriptome in early life . Its role in oocyte and embryogenesis may have long lasting effects , as reported in the accompanying paper by Wasson et al where a hypomorphic maternal KDM1A , associated with perinatal lethality , showed alterations in imprinted gene expression much later in life . The importance of the maternal pool of KDM1A opens up exciting prospects for the roles of this remarkable histone demethylase in early development . The control of repeat elements by epigenetic mechanisms , including histone KMTs and KDMs , may be critical in early development . Previous work has suggested that KDM1A may contribute to MERVL element repression in late pre-implantation embryos ( Macfarlan et al . , 2011 ) . We did not detect any impact on these elements in the Kdm1a mutants immediately post fertilization . However , we did see a small but significant increase in LINE-1 expression and LINE-1 ORF1 protein levels in the Kdm1a mutant embryos . This observation , together with the striking elevation in H3K4me3 levels , is of particular interest in the context of a recent study which proposed that loss of H3K4m3 at LINE-1 elements ( rather than a gain in H3K9 methylation ) might be critical for their repression during early pre-implantation development ( Fadloun et al . , 2013 ) . Whether this increase in LINE-1 expression actually leads to an increase in LINE-1 element retrotransposition ( ie new insertions ) remains to be seen , but the increase in LINE-1 proteins observed in Kdm1a mutant embryos is potentially consistent with such a possibility . In this context , we speculate that misregulation of LINE-1 elements in the absence of KDM1A might participate in the early developmental arrest that is observed , via an increased potential of genome instability and activation of some specific DNA damage checkpoints . The increase in γH2AX foci we detected in Kdm1a mutants , independently from replication stalling problems , could also be consistent with this hypothesis . Our results thus support the hypothesis that histone-based defence mechanisms act to safeguard the genome from LINE-1 retrotransposition during preimplantation development , when global DNA hypomethylation might compromises their usual silencing route ( Leung & Lorincz , 2012 ) . Finally , chromatin status and regulated expression of another family of repeats , located within pericentric heterochromatin , has been proposed to be involved in developmental progression after fertilization , ensuring correct chromosome segregation and heterochromatin propagation ( Probst et al . , 2010; Santenard et al . , 2010 ) . In the absence of KDM1A , we detected aberrant accumulation of H3K9me3 at presumptive pericentric heterochromatin ( NLBs ) post-fertilization , as well as lagging chromosomes in oocytes , and micronuclei accumulation following fertilisation . Collectively , this data points to maternal KDM1A protein having a potential role at pericentromere/centromere regions that merits future exploration . In conclusion , our findings demonstrate the instrumental role of KDM1A as a maternally provided protein at the beginning of life in shaping the histone methylation landscape and the transcriptional repertoire of the early embryo . Immunofluorescence was carried out as described previously ( Torres-Padilla et al . , 2006 ) , with some modifications . After removal of the zona pellucida with acid Tyrode’s solution ( Sigma ) , embryos were fixed in 4% paraformaldehyde , 0 , 2% sucrose , 0 . 04% Triton-X100 and 0 . 3% Tween20 in PBS for 15 min at 37°C . After permeabilisation with 0 . 5% Triton-X100 in PBS for 20 min at room temperature , embryos were washed in PBStp ( 0 . 05% Triton-X100; 1 mg/ml polyvinyl pyrrolidone ( PVP;Sigma ) ) then blocked and incubated with the primary antibodies in 1% BSA , 0 . 05% Triton-X100 for ∼16 hr at 4°C . Embryos were washed in PBStp twice and blocked 30 min in 1% BSA in PBStp and incubated for 2 hr with the corresponding secondary antibodies at room temperature . After washing , embryos were mounted in Vectashield ( Vector Laboratories , Burlingame , CA ) containing DAPI ( 4' , 6'-diamidino-2-phénylindole ) for visualizing the DNA . Full projections of images taken every 0 . 5 μm along the z axis are shown for all stainings , except for the ORF1 for which the middle section is shown only . Antibody staining for H3K4 methylation is in green , and in red for H3K9 methylation , DNA is counterstained with DAPI ( blue ) . For each antibody , embryos were processed identically and analyzed using the same settings for confocal acquisition Stainings were repeated independently at least twice . The following antibodies were used ( Antibody/Vendor/Catalog #/Concentration ) : anti-rabbit KDM1A/Abcam ( UK ) /ab17721/ 1:750 , anti mouse H3K4me1/Cosmobio ( Japan ) /MCA-MBAI0002/ 1:700 , anti mouse H3K4me2/Cosmobio /MCA-MBAI0003/ 1:700 , anti mouse H3K4me3/Cosmobio/MCA-MBAI0004/ 1:700 , anti-rabbit H3K9me1 kind gift from T . Jenuwein , anti mouse H3K9me2/Cosmobio /MCA-MBAI0007/ 1:500 , anti rabbit H3K9me2/ActiveMotif ( Carlsbad , CA ) /39239/ 1:800 , anti rabbit H3K9me3/ Millipore ( Billerica , MA/07–442/ 1:200 , anti-mouse H3K27me3/ Abcam/ab6002/ 1:400 , anti-rabbit H4K20me3/ Abcam/ab 9053/ 1:200 , anti-mouse γH2AX/ Millipore/05–623/ 1/200 , anti-mouse β-TUBULIN/ Invitrogen ( Carlsbad , CA ) /32–2600/ 1:1000 , anti-mouse POLII CTD4/ Millipore/05–623/1:200 , anti-rabbit POLII CTD4 S2P/Abcam/ab5095/1:200 , anti-rabbit ORF1 , kind gift from A . Bortvin/ 1:500 , Alexa488 goat anti-mouse IgG/ Invitrogen/A11029/ 1:500 , Alexa568 goat anti-rabbit IgG/ Invitrogen A11036/ 1:500 . 50 two-cell stage embryos were resuspended in 2-mercaptoethanol containing loading buffer and heated at 85°C for 15 m . SDS-PAGE , Ponceau staining , and immunoblots were performed following standard procedures using a Mini-PROTEAN Tetra Cell System ( Bio-Rad , Hercules , CA ) . Primary anti-KDM1A ( dilution 1:500 ) and secondary HRP-conjugated goat anti-rabbit ( DAKO , Santa Clara , CA , Cat . #K4002 ) were used . 2 μg of ESC nuclear extracts were used as control . RNA FISH was performed as described ( Patrat et al . , 2009 ) . Nick translation ( Vysis Abbott , Chicago , IL ) using Spectrum green or Spectrum red ( Vysis ) was used to label double stranded probes . The LINE-1 probe used consisted of a full- length Tf element cloned into a Bluescript plasmid as previously described ( Chow et al . , 2010 ) . The Atrx probe consisted of a BAC ( CHORI , Oakland , CA; reference RP23-260I15 ) . Briefly , embryos were taken at 42 hr post hCG and the zona pellucida was removed . Embryos were transferred onto coverslips previously coated in Denhardt’s solution , dried down for 30 min at room temperature , after all excess liquid was removed . Samples were fixed in 3% paraformaldehyde ( pH 7 . 2 ) for 10 min at RT and permeabilized in ice-cold PBS 0 . 5% triton for 1 min on ice and then directly stored in ETOH 70°C ethanol at -20°C until processed for RNA FISH . Hybridizations , without Cot1 competition for LINE-1 , were performed overnight at 37°C in a humid chamber . Excess of probes was eliminated through three washes in 2xSSC at 42°C for 5 min each . Slides were mounted in Vectashield containing DAPI . After zona pellucida removal and 3 consecutive washes in PBS-0 . 1% BSA , individual oocytes or whole two-cell stage embryos were transferred into a 0 . 2 ml eppendorf tube ( care was taken to add a minimum liquid volume of PBS BSA ) and directly frozen in -80°C until use . RNA was extracted and amplified as described previously ( Tang et al . 2010 ) . For quality control and gene expression analysis , quantitative real-time PCR was performed for gene expression on 1/10 dilution of cDNA preparation in 10 μl final volume with Power SYBR green PCR master mix ( Applied Biosystems , Foster City , CA ) on a ViiA7 apparatus ( Life Technologies ) . The level of gene expression was normalized to the geometric mean of the expression level ofFoster City Hprt , Gapdh and Ppia housekeeping genes as according to ( Vandesompele et al . , 2002 ) . For p<0 . 05 corresponds to * and p<0 . 001 to ** by t-test . The following primers used in this study are listed as name/ forward primer 5’ to 3’ / reverse primer 5’ to 3’ Hprt/ ctgtggccatctgcctagt / gggacgcagcaactgacatt , Gapdh/ ccccaacactgagcatctcc / attatgggggtctgggatgg , Ppia/ ttacccatcaaaccattccttctg / aacccaaagaacttcagtgagagc ( as in Duffie et al . , 2014Atrx/ tgcctgctaaattctccaca / aggcaagtcttcacagctgt , H2AZ/ acacatccacaaatcgctga / aagcctccaacttgctcaaa , Klf4/ agccattattgtgtcggagga/ agtatgcagcagttggagaac , Suv39h1/ ctgggtccacttgtctcagt/ ctgggaagtatgggcaggaa , SineB1/ gtggcgcacgcctttaatc / gacagggtttctctgtgtag ( Martens et al . , 2005 ) , MuERVL/ atctcctggcacctggtatg / agaagaaggcatttgccaga ( Macfarlan et al . , 2011 ) , Kdm1a/ tggagaacacacaatccgga / tgccgttggatctctctgtt , LINE-1 3’UTR/ atggaccatgtagagactgcca / caatggtgtcagcgtttgga For RNA deep sequencing , library construction was performed following Illumina ( San Diego , CA ) manufacturer suggestions . The 26 samples ( 5 f/f or wt/wt and 5 ∆m/∆m oocytes; 8 f/wt and 8 ∆m/wttwo-cell embryos ) were sequenced in single-end 49 bp reads on an Illumina HiSeq 2500 instrument . The depth of sequencing was ranged from 12 , 500 , 000 to 35 , 000 , 000 with an average around 18 , 000 , 000 reads per sample ( Supplementary file 1 ) . For the gene-based differential analysis , quality control was applied on raw data . Sequencing reads characterized by at least one of the following criteria were discarded from the analysis: ( more than 50% of low quality bases ( Phred score <5 ) ; more than 5% of N bases; more than 80% of AT rate At least 30% ( 15 bases ) of continuous A and/or T ) . Reads passing these filters were then aligned to the mouse mm10 genome using the TopHat software v2 . 0 . 6 ( Trapnell et al . , 2009 ) . Only unique best alignments with less than 2 mismatches were reported for downstream analyses . Count tables of gene expression were generated using the RefSeq annotation and the HTSeq v0 . 6 . 1 software ( Anders et al . , 2015 ) . The DESeq R package v1 . 16 . 0 ( Anders and Huber , 2010 ) was then used to normalize and identify the differentially expressed genes between control and mutant embryos . Genes with 0 counts in all samples were filtered out and only the 60% of the top expressed genes were used for the analysis , as described in the DESeq reference manual . Genes with an adjusted p-value lower than α = 0 . 05 were consider as differentially expressed . Hierarchical clustering analysis for gene expression pattern of 16 libraries was based on Spearman correlation distance and the Ward method , and performed using the hclust function implemented in the gplots v2 . 16 . 0 R package . In order to study the transposons expression , we performed the mapping of reads passing the quality control using the Bowtie v1 . 0 . 0 software ( Langmead et al . , 2009 ) . This mapping was performed in 2 steps: ( i ) reads aligned on ribosomal RNA ( unique best alignments with less than 3 mismatches in the seed ) ( GenBank identifiers:18S , NR_003278 . 3; 28S , NR_003279 . 1; 5S , D14832 . 1; and 5 . 8S , KO1367 . 1 ) were discarded ( ii ) remaining reads were aligned to the mouse mm10 genome , reporting a maximum of 10 , 000 genomic locations ( best alignments without mismatches ) . Aligned reads were then annotated and intersected with repeats annotation from the repeatMasker database . The transposon counts table was generated using the reads that fully overlap with an annotated repeat and for which all possible alignments are concordant , i . e associated with the same repeat family in more than 95% of cases . The resulting count table was normalized by the total number of reads aligned on repeats . Statistical analysis to identify repeat families with significant changes in expression between control and mutant embryos was performed using the limma R package v3 . 20 . 4 ( Ritchie et al . , 2015 ) . Repeats family with an adjusted p-value lower than α = 0 . 05 were consider as significant . The tool AmiGO 2 ( Carbon et al . , 2009 ) was used to perform the enrichment Gene Ontology items with the misregulated genes from the Kdm1a mutant two-cell stage embryos . The Gene Expression Omnibus ( GEO ) accession number for the data sets reported in this paper is GSE68139	During fertilization , an egg cell and a sperm cell combine to make a cell called a zygote that then divides many times to form an embryo . Many of the characteristics of the embryo are determined by the genes it inherits from its parents . However , not all of these genes should be “expressed” to produce their products all of the time . One way of controlling gene expression is to add a chemical group called a methyl tag to the DNA near the gene , or to one of the histone proteins that DNA wraps around . Soon after fertilization , a process called reprogramming occurs that begins with the rearrangement of most of the methyl tags a zygote inherited from the egg and sperm cells . This dynamic process is thought to help to activate a new pattern of gene expression . Reprogramming is assisted by “maternal factors” that are inherited from the egg cell . KDM1A is a histone demethylase enzyme that can remove specific methyl tags from certain histone proteins , but how this affects the zygote is not well understood . Now , Ancelin et al . ( and independently Wasson et al . ) have investigated the role that KDM1A plays in mouse development . Ancelin et al . genetically engineered mouse eggs to lack KDM1A and used them to create zygotes , which die before or shortly after they have divided for the first time . The zygotes display severe reprogramming faults ( because methyl tags accumulate at particular histones ) and improper gene expression patterns , preventing a correct maternal-to-zygotic transition . Further experiments then showed that KDM1A also regulates the expression level of specific mobile elements , which indicates its importance in maintaining the integrity of the genome . These findings provide important insights on the crucial role of KDM1A in establishing the proper expression patterns in zygotes that are required for early mouse development . These findings might help us to understand how KDM1A enzymes , and histone demethylases more generally , perform similar roles in human development and diseases such as cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "developmental", "biology" ]	2016	Maternal LSD1/KDM1A is an essential regulator of chromatin and transcription landscapes during zygotic genome activation
Mutations in the genes for PINK1 and parkin cause Parkinson’s disease . PINK1 and parkin cooperate in the selective autophagic degradation of damaged mitochondria ( mitophagy ) in cultured cells . However , evidence for their role in mitophagy in vivo is still scarce . Here , we generated a Drosophila model expressing the mitophagy probe mt-Keima . Using live mt-Keima imaging and correlative light and electron microscopy ( CLEM ) , we show that mitophagy occurs in muscle cells and dopaminergic neurons in vivo , even in the absence of exogenous mitochondrial toxins . Mitophagy increases with aging , and this age-dependent rise is abrogated by PINK1 or parkin deficiency . Knockdown of the Drosophila homologues of the deubiquitinases USP15 and , to a lesser extent , USP30 , rescues mitophagy in the parkin-deficient flies . These data demonstrate a crucial role for parkin and PINK1 in age-dependent mitophagy in Drosophila in vivo . Loss-of-function mutations in PARK2 and PINK1 , which encode the cytosolic E3 ubiquitin ligase parkin and the mitochondrial ubiquitin kinase PINK1 , respectively , are the most prevalent recessive causes of Parkinson’s disease ( PD ) , an age-dependent neurodegenerative disorder ( Corti et al . , 2011 ) . Parkin and PINK1 cooperate in the selective autophagic degradation of damaged mitochondria ( mitophagy ) ( Pickrell and Youle , 2015 ) . Upon mitochondrial damage , PINK1 is stabilized on the outer mitochondrial membrane ( OMM ) and phosphorylates ubiquitin and parkin at their respective S65 residues . This activates parkin to catalyze ubiquitination of many OMM proteins that are then either degraded by the proteasome or act as signals for autophagic clearance of the mitochondrion ( Pickrell and Youle , 2015 ) . Parkin-mediated mitochondrial ubiquitination and mitophagy are counteracted by specific deubiquitinases ( Bingol et al . , 2014; Cornelissen et al . , 2014 ) . PINK1/parkin-mediated mitophagy can be triggered in a variety of neuronal and non-neuronal cells in vitro , typically by exposure to mitochondrial depolarizing agents ( Narendra et al . , 2008; Geisler et al . , 2010; Cai et al . , 2012; Ashrafi et al . , 2014; Cornelissen et al . , 2014; Oh et al . , 2017 ) . Along with other mitochondrial quality control mechanisms PINK1/parkin-mediated mitophagy contributes to the maintenance of a healthy mitochondrial network ( Pickles et al . , 2018 ) . Despite the wealth of mechanistic information on PINK1/parkin-mediated mitophagy in cultured cells , questions still surround the existence and physiological relevance of this pathway in vivo ( Cummins and Götz , 2018; Whitworth and Pallanck , 2017 ) . Direct evidence for the occurrence of PINK1/parkin-mediated mitophagy in vivo is still scarce . Ubiquitin phosphorylated at S65 ( pS65-Ub ) , a biomarker of PINK1 activity , accumulates in brains from elderly human subjects ( Fiesel et al . , 2015 ) and from mice with a genetic defect in mitochondrial DNA proofreading ( Pickrell et al . , 2015 ) , suggesting that PINK1-mediated mitophagy is induced by aging and accumulation of mitochondrial damage . However , a recent study reported that mitophagy occurs independently of PINK1 in the mouse ( McWilliams et al . , 2018 ) . In Drosophila , deficiency of PINK1 or parkin causes reduced life span and severe flight muscle degeneration with accumulation of swollen mitochondria ( Greene et al . , 2003; Pesah et al . , 2004; Clark et al . , 2006; Park et al . , 2006 ) . In parkin mutant flies , half-lives of mitochondrial proteins are drastically prolonged , consistent with a role for parkin in mitophagy , but the effect of PINK1 on mitochondrial protein turnover is much more limited ( Vincow et al . , 2013 ) . Overall , it remains to be directly demonstrated whether PINK1 and parkin can target damaged mitochondria to lysosomes in vivo . Recently , the mt-Keima reporter was developed to quantitatively image mitophagy in vitro and in vivo ( Katayama et al . , 2011; Sun et al . , 2015 ) . Here , we used live mt-Keima imaging and correlative light and electron microscopy ( CLEM ) to assess mitophagy in Drosophila in vivo and to determine the role of parkin and PINK1 in this pathway . The phenotype of parkin- and PINK1-deficient flies is especially striking in flight muscle ( Greene et al . , 2003; Pesah et al . , 2004; Clark et al . , 2006; Park et al . , 2006 ) . We therefore focused on the role of mitophagy in this tissue . We generated flies that express mt-Keima specifically in muscle ( mef-2-Gal4 ) . The mt-Keima reporter is a mitochondrially targeted form of Keima , a fluorescent protein that resists degradation by lysosomal proteases ( Katayama et al . , 2011 ) . The peak of the mt-Keima excitation spectrum shifts when mitochondria are delivered to acidic lysosomes , which allows dual-excitation ratiometric quantification of mitophagy ( Katayama et al . , 2011 ) ( Figure 1A ) . Transmission electron microscopy ( TEM ) analysis showed that mt-Keima expression did not change the morphology of muscle mitochondria ( Figure 1B ) . The mt-Keima signal extensively colocalized with the mitochondrial protein ATP synthase β , confirming that mt-Keima was properly targeted to mitochondria ( Figure 1C ) . Interestingly , a small subset of mt-Keima structures had high 543 nm/458 nm ratio values , indicative of an acidic environment ( Figure 1D ) . These high 543/458 ratio mt-Keima puncta colocalized with the lysosomal dye LysoTracker ( Figure 1E ) . Thus , the high 543/458 ratio mt-Keima puncta probably represented mitochondria that had been delivered to lysosomes . The abundance of these puncta was very low in muscle cells of 1-week-old adult flies , but strongly rose with aging , showing an approximately tenfold increase by the age of 4 weeks ( Figure 1F , G ) . To assess the role of autophagy in the biogenesis of the high 543/458 ratio mt-Keima puncta , we overexpressed a kinase-dead version of Atg1 ( Atg1K38A ) , the homologue of mammalian ULK1 ( Toda et al . , 2008 ) . Atg1/ULK1 is needed in the early steps of autophagosome formation and is also involved in mitophagy ( Itakura and Mizushima , 2010; Itakura et al . , 2012 ) . When overexpressed , kinase-dead Atg1 exerts dominant-negative effects ( Scott et al . , 2007 ) . Overexpression of Atg1K38A suppressed the high levels of mitophagy observed in aged muscle cells ( Figure 1F , G ) . The low residual level of high 543/458 ratio mt-Keima puncta that persisted after Atg1K38A overexpression , may result from Atg1-independent mechanisms of autophagy induction ( Braden and Neufeld , 2016 ) . To demonstrate the occurrence of mitophagy in Drosophila indirect flight muscle more definitively , we resorted to CLEM ( Bishop et al . , 2011 ) . We first performed live mt-Keima imaging of muscle cells to identify regions of interest that contained both ‘acidic’ and ‘neutral pH’ mt-Keima structures ( Figure 2A ) . After fixation , we burned laser marks around the regions of interest using near-infrared branding ( NIRB ) to be able to re-identify these regions after processing for TEM ( Figure 2B–D ) . Interestingly , ‘neutral pH’ mt-Keima structures colocalized with typical mitochondria at the TEM level , while ‘acidic’ mt-Keima puncta showed remarkable overlay with smaller organelles with the characteristic electron-dense appearance of lysosomes ( Figure 2E–N ) . Many of the mt-Keima-positive lysosomes contained densely packed concentric membranes surrounding a dense core , giving them the appearance of multilamellar bodies ( Figure 2N ) . Multilamellar bodies are lysosomal organelles that are found in various cell types in physiological conditions but also accumulate in lysosomal storage diseases ( Blanchette-Mackie , 2000; Hariri et al . , 2000; Lajoie et al . , 2005 ) . We then crossed the mt-Keima-expressing flies with PINK1 loss-of-function mutant flies ( PINK1B9 ) and parkin RNAi flies ( Park et al . , 2006; Cornelissen et al . , 2014 ) . The low level of mitophagy in 1-week-old adult flies was not affected by deficiency of PINK1 or parkin ( Figure 3A , B ) . However , the age-dependent increase in mitophagy in 3- and 4-week-old control flies was completely abolished in the PINK1B9 and parkin RNAi flies ( Figure 3A , B ) . The DUBs USP15 and USP30 oppose parkin-mediated mitochondrial ubiquitination and mitophagy in cultured human cells . Interestingly , the yeast homologue of USP15 ( Ubp12 ) deubiquitinates the MOM protein Fzo1 , the yeast homologue of the mammalian parkin substrates mitofusin 1 and 2 , pointing to a conserved role for USP15 in deubiquitination at the MOM ( Anton et al . , 2013; SimoesSimões et al . , 2018 ) . Knockdown of the Drosophila homologs of USP15 ( CG8334 , hereafter called dUSP15 ) and USP30 ( CG3016 , hereafter dUSP30 ) largely rescues the mitochondrial defects of parkin-deficient fly muscle in vivo ( Bingol et al . , 2014; Cornelissen et al . , 2014 ) . To assess the effects of dUSP15 and dUSP30 on PINK1/parkin-mediated mitophagy in vivo , we knocked down dUSP15 and dUSP30 in mt-Keima-expressing flies using RNAi . Levels of dUSP15 mRNA in dUSP15 RNAi lines were 57 . 6 ± 2 . 5% of control levels ( n = 6 ) , and dUSP30 mRNA levels in dUSP30 RNAi lines were 16 . 4 ± 5 . 8% of controls ( n = 6 ) . Knockdown of dUSP15 and , to a lesser extent , dUSP30 rescued the mitophagy defect of two different parkin RNAi fly lines ( Figure 3C , D ) . The mitophagy defect of PINK1B9 flies was partially rescued by dUSP15 knockdown , but unaffected by dUSP30 knockdown ( Figure 3C , D ) . Dopaminergic neurons of parkin mutant Drosophila also accumulate abnormal mitochondria ( Burman et al . , 2012 ) . We therefore expressed mt-Keima in dopaminergic neurons ( TH-GAL4 ) . As in muscle cells , mitophagy increased with aging ( Figure 4A , B ) . Loss of parkin suppressed mitophagy in 4-week-old flies ( Figure 4A , B ) , and this deficit was rescued by dUSP15 knockdown ( Figure 4C , D ) . Our data show that mitophagy occurs in Drosophila flight muscle and dopaminergic neurons in vivo , even in the absence of exogenous mitochondrial toxins . Mitophagy in these cells rises with aging , and this age-dependent increase is abrogated by PINK1 or parkin deficiency . Thus , PINK1 and parkin appear to be particularly important for mitophagy during aging , which may explain why loss of these proteins in patients causes an age-dependent neurodegenerative disease rather than a congenital or developmental disorder . Consistent with this , pS65-Ub , a readout of PINK1 activity , is almost undetectable in postmortem brains from neurologically normal , young human subjects , but accumulates in brains from elderly individuals ( Fiesel et al . , 2015 ) . A recent study reported that PINK1 is dispensable for mitophagy in the mouse in vivo ( McWilliams et al . , 2018 ) . This conclusion was based on imaging using the mito-QC reporter , which differs from the mt-Keima probe used in our study . mito-QC is targeted to the OMM , whereas mt-Keima is localized to the mitochondrial matrix . Upon mitochondrial damage , activated parkin ubiquitinates a wide variety of OMM proteins , many of which are then extracted from the OMM and degraded by the proteasome before engulfment of the mitochondrion into an autophagosome ( Tanaka et al . , 2010; Chan et al . , 2011; Sarraf et al . , 2013 ) . The mito-QC probe consists of a tandem mCherry-GFP tag fused to a large portion of the OMM protein FIS1 ( McWilliams et al . , 2016 ) , which is itself a substrate of parkin ( Chan et al . , 2011; Sarraf et al . , 2013 ) . If mito-QC is ubiquitinated by parkin and extracted from the OMM , this would prevent mito-QC delivery to lysosomes and reduce its sensitivity as a reporter for PINK1/parkin-mediated mitophagy . An alternative explanation for the discrepant results could be that mice may not accumulate sufficient mitochondrial damage in their lifespan to activate PINK1/parkin-mediated mitophagy . The abundance of pS65-Ub in cerebral cortex is very low in wild-type mice , but is substantially higher in mice expressing a proof-reading-deficient version of mitochondrial DNA polymerase γ ( Pickrell et al . , 2015 ) . This suggests that a ‘second hit’ in addition to aging may be required to induce PINK1/parkin-mediated mitophagy in mice and may explain why PINK1 or parkin deficiency by itself does not cause a degenerative phenotype in this species . In a recent Drosophila study Lee et al . were unable to detect mitophagy in flight muscle and found no evidence for a major role for PINK1 or parkin in mitophagy in other fly tissues ( Lee et al . , 2018 ) . The discrepancy with our correlative mt-Keima and TEM imaging findings may be due to the fact that Lee et al . mostly relied on imaging with mito-QC , a probe that may be less sensitive for parkin-mediated forms of mitophagy , as discussed above . Also , Lee et al . restricted their analysis of flight muscle to 2-day-old or younger flies , and may thus have missed the PINK1/parkin-dependent rise in mitophagy that occurs at later ages . Recent publications proposed a distinction between basal and induced mitophagy ( McWilliams et al . , 2018; Lee et al . , 2018 ) . According to this terminology , the physiological mitophagy observed in our study could be labeled as basal . However , this may be confusing , because the term ‘basal’ suggests a relatively stationary background phenomenon , whereas our data show that ‘basal’ mitophagy in the fly is strongly induced by normal aging . We did not detect a significant reduction in mitophagy in 1-week-old PINK1B9 and parkin RNAi flies , at a time point when these flies already display mitochondrial abnormalities ( Greene et al . , 2003; Cornelissen et al . , 2014 ) . This suggests that the mitochondrial changes in 1-week-old PINK1B9 and parkin RNAi flies are caused by loss of aspects of PINK1 and parkin function that are unrelated to mitophagy . For example , parkin also regulates Ca2+ transfer from ER to mitochondria ( Gautier et al . , 2016 ) . PINK1 promotes mitochondrial complex I activity through phosphorylation of the complex I subunit NdufA10 ( Morais et al . , 2014 ) and is involved in crista junction remodeling via phosphorylation of dMIC60 ( Tsai et al . , 2018 ) . Alternatively , the early mitochondrial abnormalities in PINK1- and parkin-deficient flies may be due to more subtle reductions in mitophagy that are not detectable with mt-Keima imaging . Genetic manipulation in Drosophila is relatively straightforward . As illustrated by our dUSP15 and dUSP30 knockdown experiments , this novel mt-Keima fly model is a convenient tool to determine the impact of individual genes on PINK1/parkin-mediated mitophagy in vivo . This model will greatly facilitate the identification of targets for modulation of a pathway with growing relevance for neurodegenerative diseases . All Drosophila melanogaster stocks and experimental crosses were kept on standard corn meal and molasses food at room temperature . The mt-Keima construct ( mt/mKeima/pIND ( SP1 ) ) was a gift from Dr . A . Miyawaki ( RIKEN Brain Science Institute , Japan ) ( Katayama et al . , 2011 ) . The mt-Keima cDNA was cloned into the NotI and Xba1 sites of pUAS-attB and inserted in integration site VK20 after in-house injection . CG8334 transgenic UAS-RNAi ( 18981 ) and parkin KK UAS-RNAi ( 107919 , named parkin2 RNAi in this paper ) lines were obtained from the Vienna Stock Center ( VDRC ) and CG3016 transgenic UAS-RNAi line ( 3016 R-2 ) from NIG-Fly Stock Center . Parkin ( 37509 , named parkin1 RNAi in this paper ) and control ( luciferase , 31603 ) TRiP UAS-RNAi lines , mef-2-GAL4 , TH-GAL4 , PINK1B9 and Atg1K38A were obtained from Bloomington stock center ( Indiana , USA ) . To quantify parkin , CG8334 and CG3016 mRNA levels under control of mef-2-GAL4 , RNA was isolated from adult thoraces and real-time RT-PCR was performed as previously described ( Cornelissen et al . , 2014 ) using primers 5’-CCAGCAATGTCACCATCAAAG-3’ and 5’-GCGTGTCCACTCAGTCTG-3’ for parkin , 5’-GGAGTGACGCATCTTGAG-3’ and 5’-TTCTTTGGTATGGGTGGACTG-3’ for CG8334 and 5’ TACGCCATAGCAATCTGGGG-3’ and 5’- CTCGTGTATCTGCTGGCGTT-3’ for CG3016 . The data were normalized using RP-49 , a ribosomal gene . Real-time PCR showed that parkin mRNA levels in parkin1 RNAi flies were 49 , 7 ± 2 , 3% of control levels ( n = 5 ) . Parkin mRNA levels in parkin2 RNAi flies were determined previously ( Cornelissen et al . , 2014 ) . Flies were dissected in HL3 buffer . Indirect flight muscle fibers and complete brains were immobilized on a glass slide in low gelling temperature agarose and analyzed using a Leica TCS SP5 II confocal microscope equipped with a 63x objective lens ( HC PL APO 63x/1 . 4 CS2 ) , a multi-argon laser ( 458 , 476 , 488 nm ) and a He/Ne laser ( 543 nm ) . Mt-Keima was imaged in two channels via two sequential excitations ( 458 nm , green; 543 nm , red ) and using a 600 to 695 nm emission range . Images from random microscopic fields were captured and analyzed by an investigator blinded to genotype and age . For muscle analysis , at least 7 z-stacks with 0 . 2 µm slice thickness were taken per fly . For dopaminergic neuron analysis , at least 10 PPL1 neurons were analyzed per fly . Ratio ( 543/458 ) images were created using the Ratio Plus plugin in ImageJ . High ( 543/458 ) ratio areas were segmented and quantified with the Analyze Particles plugin in ImageJ . The total mitochondrial area was quantified with the Analyze Particles plugin by calculating the area of the total emission at 543 nm excitation . The parameter ( high [543/458] ratio area/total mitochondrial area ) was used as an index of mitophagy , as described ( Katayama et al . , 2011 ) . LysoTracker ( DND-26 , Thermo Fisher ) was imaged using a 488 nm excitation and a 495–550 nm emission filter . Indirect flight muscle was dissected in HL3 , live imaged for mt-Keima and fixed in 4% paraformaldehyde for 1 hr . Samples were washed three times for 10 min in PBS and placed in blocking solution ( 5% normal donkey serum in PBS , 0 . 1% Triton X-100 ) for 30 min , after which the samples were incubated overnight at 4°C with antibody against ATP synthase β ( 1:500; ab14730 , Abcam ) in PBS , 2% normal donkey serum , 0 . 1% Triton X-100 . After washing , samples were incubated for 2 hr with Alexa 488-conjugated secondary antibody in PBS , 2% normal donkey serum , 0 . 1% Triton X-100 . Samples were washed and mounted on a glass slide in Vectashield mounting medium ( Vector Laboratories ) . Alexa 488 fluorescence was imaged using a 488 nm excitation and a 495–550 nm emission filter . CLEM studies using near-infrared branding ( NIRB ) were performed as described previously ( Bishop et al . , 2011; Urwyler et al . , 2015 ) . Tissues were first imaged live in low gelling temperature agarose ( 3% in 0 . 1 M phosphate buffer , pH 7 . 4 [PB] ) , and fixed overnight in 0 . 5% glutaraldehyde , 4% PFA in 0 . 1 M PB , pH 7 . 4 at 4°C . Samples were rinsed in 0 . 1 M PB after which fixed samples were imaged using the Leica SP5 confocal microscope to check whether fixing affected overall tissue morphology . The bleaching function of the ZEN2010 software ( Zeiss ) was used to perform NIRB on a LSM 780 inverted confocal microscope . Branding marks were introduced to the tissue with a Mai Tai DeepSee two-photon laser ( Spectra-Physics ) at 800 nm and 40% maximal power output . Z-stacks of the region of interest were acquired before and after NIRB . Processed samples were post-fixed in 2% glutaraldehyde , 4% paraformaldehyde and 0 . 2% picric acid in 0 . 1 M PB at 4°C overnight or until further processing . Next , samples were osmicated for 1 hr in 2% OsO4 , 1 . 5% potassium ferrocyanide in PB and subsequently stained for 30 min in 0 . 2% tannic acid , followed by overnight incubation in 0 . 5% uranyl acetate in 25% methanol . Next day , samples were stained en bloc with lead aspartate and dehydrated in an ascending series of ethanol solutions followed by flat embedding in Agar 100 ( Laborimpex; Agar Scientific ) . Flat-embedded sections were mounted on aluminum pin stubs ( Gatan ) with conductive epoxy ( Circuit Works ) . A Zeiss Sigma Variable pressure SBF-SEM with 3View technology ( Gatan ) was used to approach the region of interest . Imaging was done at 1 . 3 kV with a pixel size of 20 nm and sections of 200 nm . Regions of interest could be located based on the branding marks and muscle fiber morphology . When the region of interest was reached , 70 nm thick , serial ultrathin sections were cut using a Reichardt Ultracut E ultramicrotome . All sections were collected as ribbons of 4–5 sections on triple slot grids ( Ted Pella ) . Images were taken on a JEOL JEM-1400 Transmission Electron Microscope operated at 80 kV . Values and error bars represent mean ± SEM , and n refers to the number of biological replicates . Significance of differences between conditions was analyzed with one-way ANOVA and post-hoc Holm-Sidak test ( SigmaStat 3 . 5 , Systat ) . No outliers were excluded .	Parkinson’s disease is a brain disorder where certain nerve cells slowly die , and the symptoms gradually worsen over time . While the risk of developing the condition increases with age , in certain patients the illness is caused by defects in two proteins , PINK1 and parkin . PINK1 and parkin help to manage mitochondria , the compartments in our cells that create molecules that serve as the energy currency for nearly all biological processes . When mitochondria get damaged , they release harmful substances that can kill their host cell . To prevent this , PINK1 and parkin can start a process known as mitophagy , which allows the cell to safely dispose of these dangerous mitochondria . Yet , mitophagy that is triggered by PINK1 and parkin has only been observed in cells grown in the laboratory; there is very little direct evidence that it also takes place in living organisms . If this mechanism does not happen in animals , then it is probably not relevant to Parkinson’s disease . Here , Cornelissen et al . genetically engineered fruit flies that carry a fluorescent marker which helps to track when and where damaged mitochondria are destroyed by a cell . The experiments revealed that mitophagy took place in muscles and in brain tissues . As the animals grew older , mitophagy became more frequent . However , this increase in mitophagy was not seen in insects that did not have PINK1 and parkin . These results showed that the role of PINK1 and parkin in mitophagy is not restricted to cells grown artificially . The fruit flies designed by Cornelissen et al . will be useful to investigate how PINK1 and parkin keep cells healthy by disposing of harmful mitochondria in living organisms . Ultimately , this may help to develop treatments that slow down the development of Parkinson’s disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "cell", "biology" ]	2018	Deficiency of parkin and PINK1 impairs age-dependent mitophagy in Drosophila
Gamma synchronization increases during movement and scales with kinematic parameters . Here , disease-specific characteristics of this synchronization and the dopamine-dependence of its scaling in Parkinson’s disease are investigated . In 16 patients undergoing deep brain stimulation surgery , movements of different velocities revealed that subthalamic gamma power peaked in the sensorimotor part of the subthalamic nucleus , correlated positively with maximal velocity and negatively with symptom severity . These effects relied on movement-related bursts of transient synchrony in the gamma band . The gamma burst rate highly correlated with averaged power , increased gradually with larger movements and correlated with symptom severity . In the dopamine-depleted state , gamma power and burst rate significantly decreased , particularly when peak velocity was slower than ON medication . Burst amplitude and duration were unaffected by the medication state . We propose that insufficient recruitment of fast gamma bursts during movement may underlie bradykinesia as one of the cardinal symptoms in Parkinson’s disease . Oscillatory brain rhythms are thought to contribute to the processing and long-range transmission of information as encoded in single-neuron activity through coherence of neuronal assemblies ( Fries , 2015 ) . Pathological oscillatory patterns could be related to disease-specific symptoms in Parkinson’s disease using invasive recordings in patients undergoing deep brain stimulation surgery for severe movement disorders ( Deuschl et al . , 2006; Dams et al . , 2016 ) . Most strikingly , an elevated synchronization in the subthalamic beta band activity is found in Parkinson’s disease that correlates with symptom severity in the dopamine-depleted state ( Williams et al . , 2003; Neumann et al . , 2016b; Kühn and Volkmann , 2017 ) and is attenuated by dopaminergic replacement therapy ( Kühn et al . , 2006 , 2009 ) as well as deep brain stimulation ( Eusebio et al . , 2011; de Hemptinne et al . , 2015; Oswal et al . , 2016 ) . In contrast to pathological activity at rest , less is known about the functional role of physiological movement-related changes of oscillatory patterns in the cortico-basal ganglia motor loop . In the human motor system , ongoing movement is accompanied by increases in cortical ( Crone et al . , 1998; Cheyne , 2013 ) and subcortical gamma ( 30–90 Hz ) oscillations ( Androulidakis et al . , 2007; Kempf et al . , 2009; Brücke et al . , 2012 , 2013 ) and strengthening of coupling in that frequency range ( Litvak et al . , 2012 ) . Cortical and subcortical movement-related gamma band synchronization seems a general and physiological phenomenon irrespective of the underlying disease . The power increase in high frequencies most likely relies on mechanistically different phenomena with a primarily oscillatory component in the frequency range 30–90 Hz and overlapping asynchronous spiking activity in frequencies >100 Hz ( Fries et al . , 2007; Manning et al . , 2009; Buzsáki and Wang , 2012 ) . Previous studies have shown that narrowband increases in gamma oscillations with a center frequency between 30–90 Hz and ~20 Hz width were of functional significance in encoding of movement parameters ( Jenkinson et al . , 2012; Joundi et al . , 2012; Tan et al . , 2015 ) . In line with this , human and animal studies have implicated that the basal ganglia play a major role in scaling of movement ( Desmurget et al . , 2003; Kravitz et al . , 2010; Grafton and Tunik , 2011; Brücke et al . , 2012; Joundi et al . , 2012; Oldenburg and Sabatini , 2015; Tan et al . , 2015 ) . The loss of their physiological activity may lead to hypo- or hyperkinetic movement disorders . Whether the encoding of movement scaling through subcortical oscillatory activity in the gamma band is dopamine-dependent , thus impaired in PD patients after withdrawal of dopaminergic medication , and correlates with movement slowing in the OFF dopaminergic state of the same patient , has to our knowledge never been investigated . Moreover , recent studies have highlighted that both task-specific and resting state oscillatory synchronization do not reflect a continuous activity but rather occur in bursts . While this has been primarily shown for beta activity in the basal ganglia ( Feingold et al . , 2015 ) , intracerebral recordings of cortical areas highlighted an encoding capacity through variations of gamma burst density in memory processing ( Kucewicz et al . , 2017; Lundqvist et al . , 2016 ) . In Parkinson’s disease , there is emerging evidence for a disadvantageous distribution toward long beta bursts ( Tinkhauser et al . , 2017 ) . In the gamma frequency range , neither the phenomenon of transient synchrony in motor processing nor how it may be influenced by dopamine signaling has so far been investigated . Here , we aimed at elucidating the patho-physiological role of velocity-related subthalamic gamma oscillations and their modulation by dopamine in patients with Parkinson’s disease . We hypothesized that movement-induced gamma oscillations that have been shown to drive the motor cortex ( Litvak et al . , 2012 ) are gradually scaled by kinematic velocity as seen in the upstream globus pallidus internus ( Brücke et al . , 2012 ) , the main projection-target of the subthalamic nucleus and major output nucleus of the basal ganglia . A coding deficiency in modulating subcortical gamma signaling through reduced burst rate in the hypodopaminergic state may contribute to the underlying pathomechanism of bradykinesia , one of the cardinal symptoms in Parkinson’s disease . The mean reaction time ( ±SEM ) of the 16 patients was 651 ± 176 ms across all trials , which was similar ( p=0 . 3 , permutation test ) to the reaction time in the subgroup of seven patients that additionally performed the task after withdrawal of the dopaminergic therapy ( reaction time-OFF: 622 ± 43 ms; small: 613 ± 43 ms , medium: 617 ± 39 ms , large: 637 ± 47 ms ) . Between movement conditions , reaction time was slightly but significantly shorter in the small compared to the medium and large condition ( small: 621 ± 23 ms , medium: 651 ± 23 ms , large: 648 ± 23 ms , small-medium: p=0 . 02 , small-large: p=0 . 002 ) in the patient cohort ON medication ( n = 16 ) and in the small compared to the large condition in the seven patients OFF medication ( small – large: p=0 . 03 ) . No difference between conditions was seen in those seven patients ON medication . In both medication states , there was a significant stepwise increase of both movement amplitude and movement velocity ( small <medium < large , p<0 . 001 ) between the three movement conditions in the time window from −183 ms to 670 ms around maximal velocity ( small-medium: −172 to 800 ms , medium-large: −150 to 670 ms , small-large: −190 to 800 ms around maximal velocity , p<0 . 05 , FDR-corrected ) , see Figure 1 . Patients were not significantly slower in the OFF compared to the ON state ( n = 7 , p=0 . 1 ) when averaged across all conditions but showed a significant decrease of movement velocity when performing the large condition ( p=0 . 05 ) as shown in Figure 1 . This indicates that patients in the OFF-dopaminergic state were primarily impaired in the execution of large and fast movements , while smaller and respective slower movements were less affected without dopaminergic replacement therapy . Time frequency representations averaged across all bipolar recordings and subsequently across patients revealed a specific movement-related pattern of synchronization and desynchronization that was significantly different from baseline activity ( p<0 . 05 , FDR-corrected ) in the subthalamic nucleus both contralateral and ipsilateral to the moved side ( Figure 2 , for the ipsilateral side , see Figure 2—figure supplement 2 ) . In all conditions , movement onset was preceded and accompanied by an ERS in the theta frequency range ( 2 to 8 Hz ) and an ERD in the beta band ( 13 to 30 Hz ) . The theta ERS started ~1 . 2 s before movement onset and persisted up to ~2 . 7 s after movement onset . The beta ERD started later ( ~0 . 4 s before movement onset ) and ended about ~1 . 7 s after movement onset . In addition , a pronounced synchronization in the gamma frequency range , spreading from 35 to 100 Hz , started shortly before movement onset at about ~0 . 3 s and persisted up to ~1 . 4 s for the smallest and ~3 s for the largest movement . This gamma synchronization during movement seems to segue into a post-movement beta ERS . However , when interpreting the results it should be kept in mind that the cycle number of the wavelet transform may affect the exact starting and ending time points of the desynchronization and synchronization phenomena . Gamma power around maximal velocity was first averaged across all contacts per electrode , second across hemispheres of one patient for statistical testing and then averaged across patients for visualization . Permutation testing of averaged time-frequency bands revealed a stepwise increase of gamma power with movement velocity in both hemispheres ( contralateral: small: 5 . 7% , medium: 9 . 4% , large: 11 . 6% , small vs large , p<0 . 001 , small vs medium , p<0 . 001 , medium vs large , p=0 . 02; ipsilateral: small: 4 . 1% , medium: 7 . 2% , large: 9 . 4% , small vs large , p=0 . 004 , small versus medium , p=0 . 004 , medium versus large , p=0 . 01; FDR-corrected ) , see Figure 2 . No such modulation was seen in the theta or beta band . Direct comparison of relative power changes in all time-frequency bins across movement conditions ( large , medium or small movement ) confirmed a parametric movement-related modulation of oscillatory activity in the contralateral hemisphere that was essentially restricted to the gamma band , ranging from 32 to 100 Hz in a period of up to 2 . 7 s after movement onset . The time-frequency clusters that displayed a significant change in mean oscillatory amplitude when compared to one of the other movement conditions are shown in Figure 2 , right column ( p<0 . 05 , permutation test , FDR-corrected ) . Movement-related changes in oscillatory activity were seen in both the ipsi- and the contralateral subthalamic nucleus to the moved side . To test for differences between ipsi- and contralateral hemispheres we conducted permutation tests separately for the previously defined time-frequency ranges across conditions and separately for each condition . Only the synchronization in the gamma band showed a trend toward a more pronounced gamma ERS in the subthalamic nucleus contralateral to the moved limb ( theta band: p=0 . 4 , beta band: p=0 . 8 , gamma band: p=0 . 08 ) , as seen in Figure 2 . Interestingly , the difference between gamma ERS of the ipsi- and contralateral subthalamic nucleus becomes significant when comparing the movement conditions separately . While large and medium movements entrain a significantly stronger gamma synchronization in the subthalamic nucleus contralateral rather than ipsilateral to the moved side , small movements go along with a similar oscillatory activity in both subthalamic nuclei ( small: p=0 . 9 , medium: p=0 . 01 , large: p=0 . 03 , permutation test ) as shown in Figure 2E . In contrast , mean relative power changes were similar in both hemispheres for all conditions in the theta band ( small: p=0 . 8 , medium: p=0 . 4 , large: p=0 . 5 , permutation test ) and the beta band ( small: p=1 , medium: p=1 , large: p=0 . 5 , permutation test ) . As seen in the behavioral data , increasing movement amplitude correlates with a significant increase in movement velocity . To further disentangle the influence of movement velocity and amplitude on gamma synchronization , gamma-band activity contralateral to the moved side was averaged across the three movement conditions and aligned to the time points movement onset , maximal velocity and maximal amplitude . Mean gamma power ( 0–0 . 5 s following the respective time point ) was compared between the three alignment points . This revealed a significantly stronger synchronization in the gamma band when aligned on maximal velocity ( 14 . 5% ) compared to maximal amplitude ( 6 . 9% , p<0 . 001 ) and movement onset ( 12 . 3% , p=0 . 02 ) , see Figure 3 . In all patients , there was a positive correlation between contralateral gamma band power ( 40–90 Hz , ±0 . 5 s around maximal velocity ) and the maximally reached velocity on a single-trial level that was significant in 8/16 patients . We further investigated the correlation for the entire spectrum of oscillatory activity and maximal velocity on a single-trial level separately for the hemispheres ipsi- and contralateral to the moved hand . When averaging the resulting correlation plots across all patients , a weak but significant positive correlation between relative power changes in the subthalamic nucleus contralateral to the moved hand and maximal velocity within a trial extended from 24 to 95 Hz and −400 to +600 ms around maximal velocity ( p<0 . 05 , FDR-corrected ) , as seen in Figure 3 . The significant cluster was prominent in the gamma band and reflects a finely tuned scaling of velocity-related power changes on a single-trial level . Importantly , the subthalamic nucleus ipsilateral to the movement side lacked correlation between velocity and relative power changes . The significant correlation between relative power changes and behavior on a single-trial level was specific for velocity . Other tested parameters such as reaction time , amplitude and accuracy did not significantly correlate across all patients on a single-trial level . To control for a potential fatigue effect leading to decreased peak velocity over trials or a subsequent compensatory mechanism in the gamma band , we investigated a potential systematic change of movement velocity or gamma band activity over the course of the experiment . There was no significant correlation between either variables over trials ( Spearman’s Rho = −0 . 09 , p=0 . 3 ) making a systematic fatigue effect unlikely . Correlation between motor impairment as assessed by the UPDRS-ON score and averaged relative power changes around maximal velocity showed a negative correlation between both mean theta ( Spearman’s ρ = −0 . 73 , p=0 . 006 ) and mean gamma power ( Spearman’s ρ = −0 . 55 , p=0 . 038 ) and the individual UPDRS-ON score as shown in Figure 4 . In contrast , averaged beta desynchronization during movement did not correlate with the symptom severity ON medication . Since we observed movement-related fine tuning of gamma-band synchronization mainly in the contralateral side , we focused our subgroup analysis of dopamine-related changes to the hemisphere contralateral to the moved hand . In both dopaminergic states , movement-related changes in oscillatory activity as seen when averaged across patients showed a similar general pattern with movement-related beta ERD and theta and gamma ERS ( Figure 5A , B ) . However , mean gamma synchronization during movement was significantly less pronounced in patients OFF medication ( relative power change ON = 15% , OFF = 9% , p<0 . 001 , permutation test , Figure 5C ) . When comparing movement conditions separately this lack in gamma power increase gradually enlarged toward the large movement condition ( difference ON - OFF: small = 7% , medium = 13% , large = 20% , Figure 5D ) . Similarly , on a behavioral level subjects OFF medication showed significant movement slowing during the large movement condition ( p=0 . 05; Figure 1 ) . Hence , parkinsonian motor impairment may be related to disturbed modulation capacity in gamma power scaling . Consequently , the parametric scaling of contralateral gamma synchronization with movement condition was lost OFF medication ( small vs medium , p=0 . 09; medium vs large , p=0 . 5 ) with only a significant difference between small and large movements ( p=0 . 018; Figure 5C ) . Theta ERS remained similar in both conditions . Relative changes in beta desynchronization differed between time points . While beta ERD was significantly more pronounced in the ON medication state when averaged around movement onset ( ON: −29 . 37 ± 19 . 27% , OFF: −16 . 36 ± 16 . 01% , p=0 . 037 ) , this difference did not reach significance during maximal velocity ( ON: −29 . 5 ± 19 . 5% , OFF: −22 . 3 ± 19 . 3% , p=0 . 062 ) , see Figure 2—figure supplement 3 . On average , 42 . 6 ± 3 . 5 gamma bursts with a duration of 27 . 1 ± 1 . 5 ms occurred in each trial . Gamma peak frequency was 69 . 4 ± 1 . 17 Hz ( range 62–81 Hz ) , suggesting that one gamma burst included in average 1 . 9 cycles of gamma oscillations . In the ON medication state , all gamma burst features increased significantly with movement when compared to baseline ( 26 hemispheres of 16 patients ) , see Figure 6B . Across patients , burst rate ( bursts per 0 . 5 s ) increased from 2 . 8 ± 0 . 25 to 3 . 4 ± 0 . 25 bursts during movement ( p<0 . 001 ) , burst amplitude by 51 ± 0 . 11% ( p<0 . 001 ) and burst duration from 26 . 9 ± 2 ms to 29 ± 1 . 9 ms ( p<0 . 001 ) , see Figure 6B . Relative changes in burst rate during movement correlated significantly higher with gamma power changes ( Spearman’s ρ = 0 . 8 ) than burst duration ( Spearman’s ρ = 0 . 39 , Burst rate vs amplitude: p=0 . 009 ) and burst amplitude ( Spearman’s ρ = 0 . 42 , Burst rate vs duration: p=0 . 05 ) , see Figure 6C . Additionally , relative changes in gamma burst rate during movement revealed a scaled increase toward the large movement condition ( large: 48 . 2 ± 6 . 5% , medium: 36 . 7 ± 5% , small: 29 . 3 ± 3 . 5% , L vs M: p=0 . 01 , L vs S: p=0 . 003 , M vs S: p=0 . 09 ) , see Figure 6D . Averaged relative changes in burst rate during movement correlated negatively with the UPDRS-ON score that was available in 14 patients ( normally distributed , Pearson’s r = −0 . 6 , p=0 . 008 ) , see Figure 6E . In the dopamine-depleted state , gamma burst rate during movement showed a significant decrease compared to the ON medication state ( ON: 55 ± 4% , OFF: 24 ± 4% , p<0 . 001 ) while increase in burst amplitude ( ON: 58 . 8 ± 16%; OFF: 39 . 6 ± 2 . 4% , p=0 . 4 ) and duration remained similar ( ON: 16 . 65 ± 0 . 04 ms; OFF: 10 . 9 ± 0 . 01 ms , p=0 . 2 ) , in 10 contralateral hemispheres of seven patients ( Figure 6F ) . Spatial interpolation of gamma power values mapped to contact pair location in MNI space showed a clear peak in gamma power in the dorsolateral portion of the subthalamic nucleus overlapping the subthalamic motor area ( Figure 7 ) . Correlations between averaged gamma power and each coordinate axis revealed significantly increased gamma synchronization for the lateral contact pair position ( X-axis ) and trends for more anterior ( Y-axis ) and dorsal ( Z-axis ) contact pair positions ( Spearman’s ρ = 0 . 37/0 . 21/0 . 20 , p=0 . 002/0 . 06/0 . 06 for X/Y/Z axis positions , permutation tests , FDR corrected for multiple comparisons ) . Before further discussing the significance of our findings in detail , some of the limitations of this study should be emphasized . First and foremost , the experiments were conducted in patients with Parkinson’s disease only; given the invasive character of intracranial recordings , no healthy control group could be tested which limits the generalizability of the reported results . However , as patients performed the task on their regular medication , the physiological basal ganglia activity was restored to the greatest possible extent . Previous studies that report similar movement-induced oscillatory changes across several nuclei of the basal ganglia in patients with different pathologies such as tremor or dystonia ( Androulidakis et al . , 2007; Brücke et al . , 2012 , 2013 ) and healthy non-human primates ( Courtemanche et al . , 2003; Connolly et al . , 2015 ) further suggest generalizability to the physiological brain . While the study was designed to explore changes in gamma activity with movement amplitude and velocity , grip strength was not monitored . Therefore , varying grip forces across trials are may have added to the association of movement velocity and gamma activity . However , rotation of the handle required a coordinated movement of proximal and distal arm muscles . Maximal movement velocity correlated across trials and patients with gamma activity , which is unlikely related to sole changes in grip strength . Regarding the comparison to neuronal activation patterns in the OFF medication condition , it has to be considered that recordings shortly after deep brain stimulation surgery may be confounded by a transitional symptom alleviation due to electrode implantation and consecutive edema , the so-called stun effect ( Chen et al . , 2006; Mann et al . , 2009 ) . Indeed , 4 out of 11 patients recorded OFF medication were not included in the direct ON/OFF comparison , as they showed no relevant deterioration in symptom severity . Another inherent limitation is the lack of histological verification of correct electrode position inside the subthalamic nucleus in deep brain stimulation patients . Nevertheless , electrode placement was guided by intraoperative microelectrode recordings and controlled postoperatively using the Lead-DBS toolbox ( Horn and Kühn , 2015 ) . Moreover , effective intraoperative macrostimulation and a mean improvement of ~54% in motor symptoms ( as assessed by UPDRS-III score ) during chronic deep brain stimulation provide further evidence for correct deep brain stimulation lead location . Bipolar recordings of adjacent contact pairs additionally assure the focal origin of the analyzed local field potentials by avoiding volume conduction from more distant areas . The recordings from DBS-macro electrodes did not allow assessment of multi-unit activity . The overlap of changes in gamma power with asynchronous bursts of local spiking activity thus cannot be addressed conclusively . Indeed , broad gamma synchronization spreading between 30 and 200 Hz can be contaminated by parallel increases in neuronal firing rate ( Pesaran et al . , 2002; Mukamel et al . , 2005; Rasch et al . , 2008 ) . However , the narrowband gamma peaks in averaged power spectra during movement and the spectrally distinct ( 40–90 Hz ) correlations to maximal velocity hint toward modulations in rhythmic activity instead of an uncoordinated increase in local neuronal firing ( Fries et al . , 2007; Manning et al . , 2009; Buzsáki and Wang , 2012 ) . The basal ganglia are suggested to play a pivotal role in both movement initiation and execution . Because gamma synchrony of cortical and subcortical structures accompanies movement rather than preceding it ( Androulidakis et al . , 2007; Brücke et al . , 2012 ) , it seems less related to motor planning than to monitoring or execution of ongoing movement without primarily depending on somatosensory feedback ( Fischer et al . , 2017 ) . Velocity-dependent scaling of gamma band activity has been interpreted as physiological motor function in the internal globus pallidus ( Brücke et al . , 2012 ) , the major efferent target of the subthalamic nucleus . Interestingly , subthalamic efferences have recently been shown to entrain the activity in downstream structures ( Deffains et al . , 2016 ) and gamma synchrony seems to play an important role in such network communication ( Fries et al . , 2007 ) . The motor circuit is coupled in a narrow gamma band while movement is executed , a coupling that is strengthened by levodopa and shows an inverse correlation with symptom manifestation ( Litvak et al . , 2012 ) . Thus , movement execution is likely dependent on gamma-band synchronization in local and distant basal ganglia-cortex motor networks and its disturbances may lead to reduced motor vigor in Parkinson’s disease . Motor vigor describes the gain of the parameters for a purposive action , that is , speed or amplitude ( Turner and Desmurget , 2010; Panigrahi et al . , 2015; Tan et al . , 2015 ) and increasing evidence from animal models suggests that the encoding of motor vigor , with special emphasis on movement speed , is strongly dopamine dependent ( Panigrahi et al . , 2015; Yttri and Dudman , 2016 ) . Several studies have shown in patients with Parkinson’s disease ON medication that subthalamic gamma synchrony scales with kinematic parameters albeit primarily focusing on force instead of velocity modulated movements ( Anzak et al . , 2012; Joundi et al . , 2012; Tan et al . , 2013; Tan et al . , 2016; Fischer et al . , 2017 ) . Given that patients with Parkinson’s disease OFF medication display increasingly bradykinetic and decrementing movements as signs of reduced motor vigor , the question arises whether dopamine affects the scaling capacity within the gamma frequency range . Indeed , ON-OFF medication comparisons in this regard have so far been missing , especially regarding movements that the patients are actually impaired in performing in the dopamine-depleted state . In the present study , the monitoring through gamma synchrony scaling turned out to be more precise ON than OFF medication as the general power decrease OFF medication was accompanied by a less discriminative modulation of gamma power that may have contributed to the slowing in peak-trial velocity in large and fast movements . In the basal ganglia , synchronization in the beta band has been shown to primarily rely on transient bursting activity that increases in rate during striatal post-movement beta synchronization in task-time-specific patterns ( Feingold et al . , 2015 ) . In the parkinsonian resting state , beta burst duration seems to be pathologically prolonged ( Tinkhauser et al . , 2017 ) . So far , subcortical gamma synchronization as seen during motor performance has not been investigated in this regard while cortical gamma bursts seem to vary remarkably in rate during memory processing which has been attributed to a selective routing potential near the onset of oscillatory synchrony ( Lundqvist et al . , 2016; Palmigiano et al . , 2017 ) . Here , we provide first evidence for discrete oscillatory dynamics in the gamma frequency range of the subthalamic nucleus during motor processing that showed increasing burst amplitude , duration and rate compared to baseline activity with the gamma burst rate correlating best with averaged increases of gamma power and showing similar scaling with movement velocity . Most interestingly , the movement-related bursting rate correlated with the patients’ clinical state and was significantly reduced in the dopamine-depleted state while burst amplitude and duration remained stable in both medication states . As complementary gamma and beta burst occurrence has been described in working memory tasks ( Lundqvist et al . , 2016 ) , one could hypothesize that the afore evoked disadvantageous distribution toward long beta bursts in Parkinson’s disease ( Tinkhauser et al . , 2017 ) would be more pronounced in the dopamine-depleted state thereby impeding fast repetitive synchronization in the gamma band that is needed for increasingly effortful movements . This could , however , not be tested here , as beta burst duration in Parkinson’s disease has been reported to rise above 1000 ms and performed movements did not exceed 600 ms . The basal ganglia are functionally segregated into parallel cortico – subcortical loops ( Alexander and Crutcher , 1990 ) that result in partially overlapping segments of structural connectivity . We have recently reproduced this segregation on the level of resting oscillatory activity by demonstrating that the dorsolateral part of the subthalamic nucleus , most connected to cortical motor regions , was dominated by beta resting activity ( Accolla et al . , 2016 ) , whereas higher resting alpha power was associated with a more anterior recording location in PD patients ( Horn et al . , 2017b ) . This study extends these findings to the precise localization of movement-related gamma synchronization . The significant correlation between contact pair location and movement-related gamma power indicates a spatially confined oscillator within the dorsolateral part of the subthalamic nucleus . Extracellular recordings of subthalamic neurons have previously characterized a somatotopic organization of the human subthalamic nucleus that showed higher local activity with arm movements localized more laterally than leg-related neurons ( Rodriguez-Oroz et al . , 2001 ) . Given that subjects in this study performed rotatory arm movements , the significant increase of gamma power toward more lateral electrode positions may relate to the somatotopic organization of the subthalamic nucleus . Interestingly , the percentage of units oscillating at gamma frequency correlates negatively with the bradykinesia scores ( Sharott et al . , 2014 ) which additionally points toward a link between gamma oscillations and multi-unit activity as interconnected prokinetic features in the subthalamic network . Disease-specific oscillations have reliably and repeatedly been associated to symptom severity – most prominently regarding the level of subthalamic resting beta synchrony in Parkinson’s disease ( Silberstein et al . , 2003; Kühn et al . , 2006; Neumann et al . , 2016b; Kühn and Volkmann , 2017 ) . Its potential role in motor slowing is further supported by direct beta frequency stimulation effects at different nodes of the motor network leading to increased bradykinesia in Parkinson’s disease patients ( Fogelson et al . , 2005; Chen et al . , 2007; Eusebio et al . , 2008 ) . In contrast , synchronization in the gamma band using motor cortical transcranial alternating current stimulation speeds up movement in healthy subjects ( Joundi et al . , 2012 ) and deep brain stimulation at individual gamma band frequency improves motor symptoms in Parkinson’s disease ( Tsang et al . , 2012 ) . Our results provide additional evidence for a pathophysiological link between reduced movement-related gamma activity and Parkinson’s disease by showing a negative correlation between gamma synchronization during movement and motor impairment across 16 Parkinson’s disease patients . Moreover , our findings hint toward a potential marker for adaptive deep brain stimulation , considering that we were able to show that movement-related gamma synchronization indicates the current motor state in a spatially confined area while other disease-specific patterns such as the beta peak attenuates significantly during movement . Indeed , subthalamic activity in the gamma band has recently been used to decode the temporal profile of gripping force ( Tan et al . , 2016 ) , which further supports the notion of gamma activity comprising detailed information about specific aspects of movement performance rather than just encoding motor effort . This hypothesis is further supported by our finding that subthalamic gamma synchronization occurs in oscillatory bursts and seems to convey information by varying the rate of gamma bursts instead of their mere amplitude or duration which has been proven to be especially efficient in organizing information in distant networks ( Lundqvist et al . , 2016; Palmigiano et al . , 2017 ) . The exact pattern of these discrete dynamics and their coding capacity within specific motor programs should be investigated in future studies . Sixteen patients with idiopathic Parkinson’s disease ( mean disease duration 11 . 8 years , range 4–20 years; mean age 59 . 5 years , range 39–75 years; four women; further clinical details given in Table 1 ) took part in this study . All participants provided written informed consent which was approved by the local review boards of the Charité- Universitätsmedizin Berlin and Hannover Medical School and in accordance with the standards set by the Declaration of Helsinki . All patients underwent stereotactic functional neurosurgery for bilateral implantation of deep brain stimulation electrodes in the subthalamic nucleus . The surgical procedure has been described previously ( Kühn et al . , 2005 ) . Deep brain stimulation electrode extension cables were externalized for 1–7 days , offering access to subthalamic recording sites . The permanent quadripolar macroelectrode used was model 3389 ( Medtronic Neurological Division , Minneapolis , MN , USA ) with four platinum-iridium cylindrical surfaces ( 1 . 27 mm diameter , 1 . 5 mm length ) and a center-to-center separation of 2 mm . Its contacts are numbered 0 , 1 , 2 , and 3 , with 0 being the most caudal and 3 the most cranial contact . The intended coordinates at the tip of the electrode ( contact 0 ) were 12 mm from the midline , 0–4 mm behind the midcommissural point and 4–5 mm below the anterior-posterior commissure determined by T2-weighted magnetic resonance images adjusted to the individual patient’s anatomy . Electrode placement was refined by microelectrode recordings displaying typical activity patterns once the dorsal border of the subthalamic nucleus was reached ( Hutchison et al . , 1998 ) , effective intra-operative macro-stimulation and postoperative stereotactic imaging . In all cases except of case 10 and 16 , the electrode position was additionally controlled using the Lead-DBS toolbox ( Horn and Kühn , 2015 ) based on post-operative imaging . Specific methodology is described elsewhere ( Horn and Kühn , 2015; Horn et al . , 2017a , 2017b ) . Briefly , postoperative acquisitions were linearly coregistered to preoperative acquisitions and normalized to standard stereotactic ( ICBM 2009b NLIN Asym; ‘MNI’ ) space using DARTEL ( Ashburner , 2007 ) as implemented in the Statistical Parametric Mapping toolbox ( SPM12; http://www . fil . ion . ucl . ac . uk/spm/software/spm12/ ) . Contacts can then be related to a version of the Morel atlas ( Morel , 2013 ) that was likewise defined in MNI space ( Jakab et al . , 2012 ) . All patients had at least one contact within the subthalamic nucleus based on postoperative imaging ( except for the right electrode in Case 1 ) . Specifically , 69/84 contact pairs ( 82% ) of 14 patients had at least one contact inside the subthalamic nucleus . The remaining 15 contact pairs from 7 patients either lay below the edge of the target ( Case 6: L01 ) , dorso-lateral to the subthalamic border ( Case 2: R23 , L23; Case 8: R23 , L23; Case 9: R23 , L23; Case 13: R23 , L23; Case 14: R23 , L12 , L23 ) or outside the subthalamic nucleus ( Case 1: R01 , R12 , R23 ) . Contact pairs outside the subthalamic nucleus ( n = 15 ) were excluded from further analysis . For an overview of all electrode localizations and a single example of a reconstructed electrode pair position , see Figure 7—figure supplement 1 . Local field potentials were recorded while patients engaged in a reaction time task in which they were asked to perform forearm pronation movements of 3 amplitudes ( 30° , 60° and 90° ) in response to imperative visual cues . They were comfortably seated in front of a 15-inch laptop screen and held a light , rotatable grip with an integrated potentiometer that was able to sense rotatory movement . The investigator visually controlled for a steady posture and unchanged holding of the rotatory device throughout the recording session . On screen , the current handle position was shown as a red dot , the requested angular route as a black semicircle that changed to blue once the patient reached the target position . After 3 s , the color changed back to black , indicating that the patient should return to the starting position . The cues , labeled as ‘small’ , ‘medium’ and ‘large’ , were presented in randomized order and with inter-trial periods of 4 s ( see Figure 1 ) . Subjects were instructed to perform the task as quickly and accurately as possible . Before the LFP-recording a brief training session of 5–10 trials was performed . For each arm , patients performed at least 45 trials ( 15 per condition ) during the recording session . The amplitude and duration of the presented cues and the consecutive rotation movements were recorded in parallel to local field potential signals via a 1401 DA converter using Spike 2 software . It has previously been shown in patients with idiopathic isolated dystonia that the three executed movement amplitudes also reflect different movement speed entities ( Brücke et al . , 2012 ) , which we verified for the tested Parkinson’s disease patient cohort . Recordings were made 1–4 days postoperatively while the deep brain stimulation electrodes were externalized . All patients performed experiments in 30–60 min following the intake of their regular dopaminergic medication ( ON medication state ) . In 11/16 patients , experiments were also performed after they had been at least 12 hours withdrawn from dopaminergic medication . Patients were considered OFF medication when they showed >30% UPDRS III score increase compared to UPDRS III score under dopaminergic substitution . 4 patients did not meet this criterion despite medication withdrawal , most likely due to the post-operative stun effect and were therefore not considered for further comparisons between the ON and OFF state . 10 patients performed the paradigm with both hands consecutively , four patients only with their left and two patients only with their right hand . In the OFF-medication state , an additional 2 patients performed it only with the left hand due to severe tremor or rigidity that obviated test performance in those patients . This led to a total of 26 sides from 16 patients in the ON-medication and 11 sides from 7 patients in the OFF-medication state that were finally analyzed . From the 7 patients included in the ON-OFF comparison , three performed the task first ON then OFF medication and 4 patients the other way around to limit order effect . Local field potentials were recorded in a bipolar montage between adjacent contact pairs ( 0–1 , 1–2 , 2–3 ) . Signals were band-pass filtered between 1 and 250 Hz , amplified ( x 50 . 000 ) using a D360 amplifier ( Digitimer Ltd , Welwyn Garden City , Hertfordshire , UK ) , and sampled at 1 kHz in parallel with the movement traces using a 1401 AD converter . Analyses of both behavioral and electrophysiological data were performed in MATLAB ( version R2016a; The MathWorks , Natick ) using custom MATLAB code based on the Statistical Parametric Mapping ( Litvak et al . , 2011bLitvak et al . , 2011 ) ( SPM ) and Fieldtrip ( Oostenveld et al . , 2011 ) toolboxes ( https://github . com/RoxanneLofredi/stn ) . Continuous recordings were divided into epochs of 5 s ( −2 to +3 s around movement onset of each trial ) . Fifth order butterworth , high-pass ( >1 Hz ) and notch filters ( 48–52 Hz ) were applied to limit effects of line noise . Epochs were rejected from further analysis if they contained artifacts and saturation in the local field potentials trace or abnormalities in the movement trace , leaving a mean ( ±SEM ) of 27 . 6 ± 10 . 8 remaining trials in the ON-medication state and 27 . 0 ± 9 . 0 ( mean ±SEM ) in the OFF-medication state per hand and per condition for final analysis . The time points of the maximal amplitude ( maxA ) , maximal velocity ( maxV ) and movement onset ( MO ) of each trial were automatically detected on the basis of the movement trace . Maximal amplitude was set as the highest point in the movement trace per trial . The velocity trace was defined as the first derivative of the movement trace and its highest point as maximal velocity . Movement onset was automatically defined as the time point where 10% of the peak-trial velocity were reached . All automatically defined time points were visually checked and adjusted if necessary . Reaction time was calculated as the interval between the imperative visual cue and movement onset . Accuracy was defined as the percentage deviation of a reference scaling movement , performed before starting the task . To investigate the changes in local field potential activity in the time-frequency domain , a time-frequency decomposition based on Morlet wavelets with seven cycles was applied to the local field potential recordings of all contact pairs from each trial . Local field potentials were analyzed over frequencies between 1 and 100 Hz with a frequency resolution of 1 Hz . Event-related local field potential power change was subsequently normalized in percentage relative to an averaged pre-trial baseline period ( −2 to 0 . 5 s preceding the visual cue ) , smoothed over 6 Hz and 200 ms and averaged using robust averaging across trials and contact of pairs of the same patient . Normalized time frequency plots of each movement condition ( small , medium , large ) were aligned to movement onset , maximal velocity and maximal amplitude , averaged across contacts , hemispheres ( ipsi- and contralateral ) and patients for visualization . If not otherwise indicated , analyses were separately performed for both subthalamic nuclei ipsi- and contralateral to the moved hand . For the main analysis , distinct frequency ranges and a time period of interest were designated . Frequency bands were determined according to significant cluster of movement-related modulation in the grand average from all patients during movement ( 0–500 ms after movement onset; theta-band 2–8 Hz and beta-band 13–30 Hz ) . As the study focuses on velocity-related modulation of gamma activity , the gamma band was restricted to the frequency range that showed modulation between conditions ( gamma-band 40–90 Hz ) , see Figure 2—figure supplement 1 . In order to compare varying movement characteristics , mean power changes were averaged around a time period of 500 ms following movement onset , which included significant differences in amplitude and velocity as seen in the behavioral results while assuring an ongoing movement in all conditions . Thus , if not indicated otherwise , power changes averaged over time and frequency refer to the mean power changes in the theta , beta or gamma band from time points 0 to 0 . 5 s after movement onset . We examined whether power changes averaged across all patients differed from baseline during movement , and compared time-frequency bins between the 3 task conditions ( small , medium , large ) . We then tested if movement-related oscillatory patterns differed between the ipsi- and contralateral subthalamic nucleus to the moved hand by comparing averaged relative power changes in the predefined theta , beta and gamma bands . Mean relative power changes as seen when aligned on movement onset were juxtaposed to an alignment on the time points maximal velocity and maximal amplitude , in order to identify the condition that reflected best the maximal relative power change . Then , mean relative power changes in the 3 predefined time-frequency ranges were compared between conditions to determine an eventual parametric modulation of event-related changes in oscillatory activity . To evaluate the temporal and spectral specificity of movement-related oscillatory activity , the within-patient correlation between maximal inter-trial velocity and power changes in each time-frequency bin was calculated on a single-trial level ( Spearman’s ρ ) and results were tested for significance using one sample tests . In the same way , oscillatory activity was also correlated with single-trial amplitude , accuracy and reaction time . On a group level , correlation coefficients of the entire power spectra from all patients were tested against the null hypothesis using permutation tests and False Discovery Rate ( FDR ) correction . Relative power changes in the subthalamic nucleus contralateral to the moved side were averaged across conditions for each of the three frequency bands of interest and correlated across all patients with the UPDRS-ON score using non-parametric Spearman rank correlation . In a subpopulation of 7 patients , the influence of dopamine on movement-related modulation of oscillatory activity was assessed by matched group comparison of the mean relative power changes for the theta , beta and gamma time-frequency bands with and without dopaminergic medication . For visualization , averaged time-frequency plots in the ON- and OFF- medication state were shown separately for this subgroup . Only results of the subthalamic nucleus that was contralateral to the moved hand are shown . For burst determination , the criteria that have previously been established by Tinkhauser et al . ( 2017 ) for beta burst analysis in patients with Parkinson’s disease were applied . The raw local field potential recording was band pass filtered ( Fifth order butteworth filter ) around the gamma band ( 40–90 Hz ) , rectified and smoothed with moving average gaussian smoothing kernel of 100 ms length . A gamma burst was determined when the resulting instantaneous power exceeded the 75th percentile of the signal amplitude distribution of all data in each electrode . Threshold crossings lasting shorter than one gamma cycle were excluded from the analysis ( see Figure 7A ) . For ON-OFF comparisons ( n = 7 ) , the threshold of the ON medication state was applied to the OFF recordings of the matching hemisphere . The amplitude of a gamma burst was defined as the area under the curve between signal and threshold line . The density of gamma bursts , in the following ‘burst rate’ , was defined as the number of bursts occurring in 0 . 5 s . The change of burst rate therefore serves as an estimate of the probability of bursting during movement ( 0 . 5 s following movement onset ) compared to that in 0 . 5 s of baseline activity ( 2–1 . 5 s before movement onset ) . We also compared averaged burst duration and amplitude during movement with that during baseline activity . In a second step , we investigated which movement-related burst property changes ( amplitude , duration or rate ) correlated most with the movement-related increase of power in the gamma band . Correlation coefficients between power changes and changes in burst properties were calculated for each bipolar recording , and mean values from contact pairs of the same electrode ( n = 26 ) were averaged on the group level and tested for differences using permutation tests . For the property that correlated best , we investigated whether it also exhibited scaling with task condition ( small , medium , large ) by averaging across hemispheres and patients ( n = 16 ) as we did for averaged gamma power . Additionally , its relative change averaged across conditions was correlated with the clinical state as assessed by the UPDRS-III ON score , available in 14 patients ON medication . In the7 patients that performed the task both ON and OFF dopaminergic medication , gamma burst rate , amplitude and duration were averaged across contact pairs and ON-OFF comparison was performed separately for each hemisphere ( n = 10 ) using permutation testing . To create a spatial map of oscillatory changes in normalized three-dimensional anatomical space , all contact pair MNI coordinates from electrodes contralateral to the moved hand were extracted from electrode localization results following the subcortical electrophysiology mapping approach described in Horn et al . ( 2017b ) . Briefly , contact pair coordinates from the left hemisphere were flipped to the right hemisphere , averaged gamma power ( see above ) was then assigned to the respective contact pairs for each movement condition ( small , medium , large ) , aligned to the time point of maximum velocity and visualized as a scattered point cloud in MNI space . A scattered interpolant was fit to these data points and values were projected to an equidistant grid using natural neighbor interpolation . This resulted in a volumetric description of gamma power distribution , which was then projected to the surface of the subthalamic nucleus using Surf Ice ( Open Source; https://www . nitrc . org/projects/surfice/ ) . In addition , a binary mask of the subthalamic nucleus motor part ( Ewert et al . , 2017 ) was projected onto the surface of the subthalamic nucleus for visual comparison . Non-parametric rank-based Spearman correlations were calculated between averaged gamma power across all movement conditions and respective contact-pair locations on the X , Y , Z axes to test for a gamma gradient in MNI standard space . This was done to show a potential group effect of a statistical relation between recording site and movement-related gamma synchronization . Thereby potential group effects of contact pair location on movement-related gamma synchronization are verified , remaining unaffected by the described masking and interpolation . Non-parametric Monte Carlo permutation tests were used for statistical results reported in this study . Permutation tests do not rely on assumptions about the underlying data distribution . In brief , the sample of tested values was interchanged randomly 5000 times to generate a probability distribution in which the observed original sample rank is reported as the p-value . All tests were multiple comparisons corrected by controlling the false discovery rate for an α-level of α = 0 . 05 . Only p-values smaller than the false discovery rate-corrected threshold were considered significant . When investigating power changes , each time-frequency bin was tested separately and only significant p-values that extended over a cluster of at least 150 time-frequency bins are shown . Rank-based Spearman correlations were calculated if data deviated significantly from a normal distribution as assessed by Kolmogorov-Smirnov tests . Otherwise , linear Pearson correlations were conducted .	Parkinson’s disease is a disorder of the nervous system that affects more than 1% of people over the age of 60 . Symptoms include uncontrollable shaking or tremor , and difficulty with large or fast movements . These symptoms occur when neurons that produce the chemical dopamine die . The loss of dopamine disrupts the activity of structures deep within the brain called the basal ganglia , which normally control movement . Some patients with Parkinson’s disease benefit from a treatment known as deep brain stimulation . Electrodes are lowered into the brain to stimulate part of the basal ganglia called the subthalamic nucleus . But we can also use these electrodes to record the activity of neurons . Doing so reveals that during movement , neurons in the subthalamic nucleus coordinate their firing at a frequency of 40 to 90 hertz . This is known as gamma synchronization . Lofredi et al . now reveal that patients with Parkinson’s disease , who do not take any medication , show reduced gamma synchronization . The greater the loss of synchronization , the more slowly patients move . Gamma synchronization does not occur continuously during a movement , but instead occurs in brief bursts . Patients with Parkinson’s disease show a reduction in the number of bursts , but not in their duration or intensity . Measuring bursts of gamma synchronization may help reveal what is happening inside the brain of a patient with Parkinson’s disease in real time . This could lead to improvements in deep brain stimulation therapy . At present , electrodes stimulate the basal ganglia continuously , but this can lead to side-effects . In the future , it may be possible to apply stimulation only when there is too little synchrony . This could reduce side effects and make the treatment more effective .	[ "Abstract", "Introduction", "Results", "Discussion", "Methods", "and", "materials" ]	[ "medicine", "neuroscience" ]	2018	Dopamine-dependent scaling of subthalamic gamma bursts with movement velocity in patients with Parkinson’s disease
Earlier reports showed that hyperplasia of sympathoadrenal cell precursors during embryogenesis in Nf1-deficient mice is independent of Nf1’s role in down-modulating RAS-MAPK signaling . We demonstrate in zebrafish that nf1 loss leads to aberrant activation of RAS signaling in MYCN-induced neuroblastomas that arise in these precursors , and that the GTPase-activating protein ( GAP ) -related domain ( GRD ) is sufficient to suppress the acceleration of neuroblastoma in nf1-deficient fish , but not the hypertrophy of sympathoadrenal cells in nf1 mutant embryos . Thus , even though neuroblastoma is a classical “developmental tumor” , NF1 relies on a very different mechanism to suppress malignant transformation than it does to modulate normal neural crest cell growth . We also show marked synergy in tumor cell killing between MEK inhibitors ( trametinib ) and retinoids ( isotretinoin ) in primary nf1a-/- zebrafish neuroblastomas . Thus , our model system has considerable translational potential for investigating new strategies to improve the treatment of very high-risk neuroblastomas with aberrant RAS-MAPK activation . Neuroblastoma , a malignant embryonic tumor of childhood , arises in neural crest-derived dopaminergic neuroblasts that generate the peripheral sympathetic nervous system ( PSNS ) . This cancer is the most common noncranial solid tumor in childhood , accounting for 8–10% of all childhood malignancies and 15% of all cancer deaths in children ( Maris and Matthay , 1999 ) . Neuroblastoma is genetically heterogeneous , with multiple interacting genetic mutations required to generate fully transformed malignant tumors . As in many other pediatric cancers , resequencing studies have documented a very low frequency of somatic mutations in neuroblastoma ( Cheung et al . , 2012; Molenaar et al . , 2012; Pugh et al . , 2013 ) , including ALK ( ~9 . 2% of cases ) , PTPN11 ( ~2 . 9% ) , ATRX ( ~2 . 5% ) and NF1 ( ~1% ) . Amplification and overexpression of the MYCN oncogene occurs in about one-third of patients and is a principal indicator of a poor prognosis ( Brodeur et al . , 1984; Hansford et al . , 2004 ) . Genomic aberrations that inactivate the NF1 tumor suppressor gene , including loss-of-function mutations and deletions , as well as decreased expression levels of the gene , have been identified in 6% of primary neuroblastomas and are predictive of a poor outcome ( Hölzel et al . , 2010 ) , suggesting an important role for NF1 loss in neuroblastoma tumorigenesis . The NF1 gene encodes neurofibromin , a 2818 amino acid protein whose main functional domain is the ~330 amino acid GTPase-activating protein-related domain ( GRD ) , which negatively regulates RAS signaling by catalyzing the hydrolysis of RAS-GTP into RAS-GDP ( Nur-E-Kamal et al . , 1993 ) ; thus , one consequence of NF1 loss is the aberrant activation of RAS signaling ( Maertens and Cichowski , 2014 ) . Loss of NF1 in neuroblastoma cells has been shown to mediate resistance to retinoic acid via hyperactive RAS signaling , which can be abolished by enforced expression of NF1-GRD ( Hölzel et al . , 2010 ) . However , it is known that NF1 has other functions in PSNS development besides the downmodulation of RAS signaling , because Nf1 mutant mice die at birth with evidence of massive overgrowth of neural crest tissues , including the sympathetic ganglia , while overexpression of the GRD domain is unable to reverse this overgrowth ( Ismat et al . , 2006 ) . In addition , studies showed identical or only modestly elevated RAS-GTP levels in NF1-deficient human neuroblastoma cells , in contrast to highly elevated RAS-GTP levels in NF1-deficient Schwannoma tumor cells ( Johnson et al . , 1993; The et al . , 1993 ) . These results , coupled with the numerous mutations of NF1 that cause the disease neurofibromatosis type 1 , but do not appear to affect protein stability or GAP function ( Abernathy et al . , 1997; Fahsold et al . , 2000 ) , argue that functional domains outside the GRD may mediate important aspects of neurofibromin function in neuroblastoma tumor suppression . In earlier work , we identified two separate duplicated nf1 zebrafish genes , nf1a and nf1b , and generated multiple loss-of-function nf1 mutant zebrafish lines affecting both of these alleles ( Lee et al . , 2010; Padmanabhan et al . , 2009; Shin et al . , 2012 ) . Mutant larvae carrying at least one wild-type nf1a or nf1b allele are viable , fertile , and show no obvious phenotypes during early development . To gain insight into the cellular and molecular consequences of NF1 loss in neuroblastoma , we used transgenic zebrafish models of neuroblastoma that overexpresses human MYCN in the PSNS ( Zhu et al . , 2012 ) . Here , we report that loss of the nf1a orthologue , which is much more highly expressed than nf1b during early PSNS development , greatly accelerates the onset of neuroblastoma induced by MYCN overexpression , with nearly complete penetrance by 5 weeks of age in nf1-deficient zebrafish . Loss of nf1 led to the aberrant activation of RAS signaling in MYCN-induced neuroblastoma , promoting both tumor cell survival and proliferation . We also show that the very aggressive growth properties of MYCN-induced neuroblastomas with loss of nf1 are due to aberrant activation of RAS signaling , because the increased penetrance and rapid growth could be suppressed by overexpressing the intact NF1 GRD domain . These findings establish nf1-deficient zebrafish that overexpress MYCN as an ideal animal model system for investigating new strategies to improve treatment of very high-risk neuroblastomas with aberrant RAS-MAPK activation . In vivo structure-function analysis with both the wild-type and inactive GRD domain of NF1 revealed that the GAP activity of the GRD domain is required for the tumor suppressor function of NF1 in neuroblastoma . By contrast , the wild-type GRD domain failed to rescue the hypertrophy of sympathoadrenal cells in nf1 mutant embryos , indicating that the role of NF1 in suppressing neuroblastoma tumorigenesis differs from the mechanism that prevents PSNS hyperplasia during normal development . Forced expression of the NF1 GTPase-activating protein-related domain ( GRD ) has been used to restore GAP activity in NF1-deficient human and mouse cells ( Hölzel et al . , 2010; Ismat et al . , 2006; Keutmann et al . , 1983 ) ; however , this domain did not rescue the developmental overgrowth of neural crest-derived PSNS tissues that is observed in Nf1-deficient mice ( Brannan et al . , 1994; Gitler et al . , 2003; Ismat et al . , 2006 ) , supporting an alternative activity of Nf1 as the mediator of growth regulation in PSNS neuronal progenitors . For our studies in the zebrafish model , we began with experiments to confirm this surprising result in zebrafish PSNS precursor cells by studying the effect of loss of nf1 on growth of the superior cervical ganglia ( SCG ) during the normal development of early embryos ( Figure 1 ) . We bred the nf1a+/-; nf1b+/- mutant zebrafish line ( Shin et al . , 2012 ) with transgenic fish overexpressing either EGFP or mCherry in the PSNS under control of the dβh promoter ( dbh:EGFP or dbh:mCherry ) ( Zhu et al . , 2012 ) . As in the mouse , complete loss of nf1 led to increased cell numbers in the SCG at 6 dpf ( compare Figure 1D with panel A , also panel E ) . Homozygous loss of nf1a led to the same level of increase in SCG neuronal cell number as homozygous loss of nf1a plus nf1b , while the loss of nf1b had little effect on SCG cell numbers ( Figure 1B and G ) , which is consistent with the fact that nf1a is expressed at a much higher level than nf1b in sympathetic neurons as well as the whole embryo during the first week of zebrafish embryonic development ( Figure 1F ) . Later in development at 3 , 4 and 6 weeks of life , we observed lower relative nf1a and nf1b levels in RNA from the whole fish , and at these time points the expression levels of nf1a and nf1b were similar to each other , without evidence of the predominance of nf1a that was observed at 1 week of age . 10 . 7554/eLife . 14713 . 003Figure 1 . The GRD domain of NF1 cannot rescue the PSNS overgrowth in nf1 deficient zebrafish . ( A–D ) Development of superior cervical ganglia ( SCG , highlighted by dotted circles ) in representative embryos of nf1a+/+;nf1b+/+;GFP , nf1a+/+;nf1b-/-;GFP , nf1a+/-;nf1b-/-;GFP and nf1a-/-;nf1b-/-;GFP genotypes at the age of 6 days postfertilization ( dpf ) . ( E ) Quantification of GFP+ cells in the SCG of embryos of nf1a+/+;nf1b+/+;GFP ( n = 7 ) , nf1a+/+;nf1b-/-;GFP ( n = 7 ) , nf1a+/-;nf1b-/-;GFP ( n = 12 ) and nf1a-/-;nf1b-/-;GFP ( n = 9 ) genotypes at the age of 6 dpf . ( F ) Quantitative RT-PCR showing the relative expression levels of nf1a and nf1b in the PSNS of zebrafish embryos and juveniles . 1-week-old dbh:mCherry embryos ( n = 200 ) were pooled , dissociated and FACS sorted to obtain PSNS ( mCherry+ ) and non-PSNS ( mCherry- ) cells for analysis . RNA of whole embryos from the same clutch of eggs were also analyzed . RNAs from whole juvenile zebrafish at the ages of 3 , 4 , and 6 weeks were also examined by quantitative RT-PCR . ( G ) Quantification of GFP+ cells in the SCG of embryos of nf1a+/+;nf1b+/+;GFP ( n = 9 ) , nf1a+/-;nf1b+/+;GFP ( n = 15 ) and nf1a-/-;nf1b+/+;GFP ( n = 13 ) genotypes at the age of 6 dpf . ( H ) Quantification of GFP+ cells in the SCG of embryos of nf1a+/+;nf1b+/+;GFP ( n = 6 ) , nf1a-/-;nf1b+/+;GFP ( n = 6 ) , nf1a+/+;nf1b+/+;GFP;wt-GRD;mCherry ( n = 5 ) , nf1a+/-;nf1b+/+;GFP;wt-GRD;mCherry ( n = 12 ) and nf1a-/-;nf1b+/+;GFP;wt-GRD;mCherry ( n = 11 ) genotypes at the age of 6 dpf . p<0 . 01 , p<0 . 001 by two-tailed unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 003 Given the high conservation between the GRD domains of the human and zebrafish neurofibromin proteins ( Padmanabhan et al . , 2009 ) , we introduced the GRD domain of human NF1 into our nf1 mutant zebrafish using a dbh:GRD; dbh:mCherry stable transgenic zebrafish line that expresses both this domain and the dbh:mCherry fluorescent marker in the PSNS . In agreement with studies in the mouse , we found that the functionally active wild-type GRD domain ( designated 'wt-GRD' ) did not rescue SCG overgrowth in nf1 mutant fish , as the levels of SCG overgrowth in nf1a-/-; nf1b+/+;EGFP embryos transgenic for dbh:wt-GRD;dbh:mCherry were similar to those of nf1a-/-; nf1b+/+;EGFP embryos at 6 dpf ( Figure 1H ) . Thus , nf1 is required to control the growth of PSNS precursor cells during embryologic development in zebrafish as well as mice , and this activity is independent of the nf1 GRD in zebrafish as well as in mice . Loss of Nf1 has been shown to increase the penetrance of MYCN-induced neuroblastoma in MYCN transgenic mice ( Weiss et al . , 1997 ) , and for our studies in zebrafish we determined the extent to which loss of nf1 synergizes with overexpression of the MYCN oncogene in this disease . Thus , we bred the nf1a+/-; nf1b+/- mutant zebrafish line ( Shin et al . , 2012 ) with transgenic fish overexpressing EGFP-MYCN in the PSNS under control of the dβh promoter ( designated 'MYCN' in this article ) ( Zhu et al . , 2012 ) . The nf1a+/-;nf1b+/-;MYCN zebrafish line was then bred with a nf1a+/-;nf1b+/-;dbh:EGFP transgenic line to obtain transgenic lines harboring one , two , or three nf1 mutant alleles , as well as the overexpression MYCN transgene in the PSNS ( Figure 2figure supplement 1 ) . The resultant offspring were monitored every 2 weeks , beginning at 4 weeks postfertilization for evidence of fluorescent tumors in the PSNS . The fish were genotyped at 8 weeks of age , with each of the 16 expected genotypes represented in the offspring of this cross: ( 1 ) nf1a+/+;nf1b+/+;EGFP; ( 2 ) nf1a+/-;nf1b+/+;EGFP; ( 3 ) nf1a+/+;nf1b+/-;EGFP; ( 4 ) nf1a-/-;nf1b+/+;EGFP; ( 5 ) nf1a+/-;nf1b+/-;EGFP; ( 6 ) nf1a+/+;nf1b-/-;EGFP; ( 7 ) nf1a-/-;nf1b+/-;EGFP; ( 8 ) nf1a+/-;nf1b-/-;EGFP; ( 9 ) nf1a+/+;nf1b+/+;EGFP;MYCN; ( 10 ) nf1a+/-;nf1b+/+;EGFP;MYCN; ( 11 ) nf1a+/+;nf1b+/-;EGFP;MYCN; ( 12 ) nf1a-/-;nf1b+/+;EGFP;MYCN; ( 13 ) nf1a+/-;nf1b+/-;EGFP;MYCN; ( 14 ) nf1a+/+;nf1b-/-;EGFP;MYCN; ( 15 ) nf1a-/-;nf1b+/-;EGFP;MYCN; ( 16 ) nf1a+/-;nf1b-/-;EGFP;MYCN . Larval fish harboring homozygous loss of both alleles of nf1a and nf1b were not obtained from this cross , because they die between 7 and 9 days of age ( Shin et al . , 2012 ) . Transgenic fish lines that expressed EGFP in the PSNS but not the MYCN transgene did not develop neuroblastoma , regardless of the mutational status of nf1 , indicating that loss of up to three alleles of nf1 was insufficient to induce neuroblastoma on its own ( Figure 2A and Figure 4C ) . Over the 20-week course of this experiment , neuroblastoma developed in one of the 14 transgenic MYCN-positive zebrafish with wild-type nf1 alleles ( Figure 2C ) , consistent with the relatively low penetrance and late onset of these tumors in the wild-type background ( Zhu et al . , 2012 ) . Histopathologically , the tumor masses that developed in MYCN transgenic fish with nf1 mutations consisted of small , undifferentiated tumor cells with distinct single nucleoli ( Figure 2B ) , which were diagnostic of neuroblastoma and indistinguishable from the tumors that develop in MYCN transgenic zebrafish with wild-type nf1 alleles ( Zhu et al . , 2012 ) . The first tumors arose in the interrenal gland ( IRG ) of 4-week-old animals with loss of one or more of the nf1a alleles , representing the earliest onset we have observed in this MYCN transgenic model ( Figure 2C ) . 10 . 7554/eLife . 14713 . 004Figure 2 . Loss of nf1 accelerates disease onset and increases the penetrance of MYCN-induced neuroblastoma . ( A ) Representative fish of nf1a+/+;nf1b+/+;GFP , nf1a+/+;nf1b+/+;MYCN;GFP , nf1a-/-;nf1b+/+;GFP and nf1a-/-;nf1b+/+;MYCN;GFP genotypes at 5 weeks of age . The interrenal glands ( IRGs ) are highlighted with dashed circles . ( B ) H&E-stained sagittal sections of a 10-week old nf1a+/+;nf1b+/+;GFP fish ( left ) and a 10-week old nf1a-/-;nf1b+/-;GFP;MYCN fish ( middle ) and 63-fold magnified tumor cells ( right ) , which are magnified from the area in the small white box in the middle panel . ( C ) Cumulative frequency of neuroblastoma in MYCN transgenic zebrafish representing all eight nf1 genotypes generated by the breeding of the nf1a+/-;nf1b+/-;MYCN line with the nf1a+/-;nf1b+/- zebrafish line ( p<0 . 0001 nf1a-/-;nf1b+/+ vs . nf1a+/+;nf1b+/+ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 00410 . 7554/eLife . 14713 . 005Figure 2—figure supplement 1 . Breeding scheme to obtain zebrafish lines harboring mutated nf1 and overexpression of GFP and the MYCN transgene in the PSNS for this study . nf1 mutant zebrafish ( gray ) were bred with dbh:EGFP-MYCN ( yellow ) or dbh:EGFP ( light green ) transgenic lines to incorporate the overexpression of both MYCN and EGFP in the PSNS ( green ) into the nf1 mutant lines . Fish with the nf1a-/-;nf1b-/- genotype are embryonic lethal ( indicated with blue letters ) . The viable genotypes used in this study are highlighted in the red box . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 005 Tumor penetrance in the progeny of this cross varied widely depending on the nf1 genotype ( Figure 2C ) . Three genotypes that retained two copies of functional nf1a were associated with similar tumor onsets , as shown in blue in Figure 2C , with a neuroblastoma penetrance of ~5 . 5% , suggesting that loss of nf1b has much less impact than nf1a on the pathogenesis of neuroblastoma in the PSNS . This is consistent with the fact that nf1b is expressed at a much lower level than nf1a in the PSNS during the first week of zebrafish development ( Figure 1F ) . By contrast , the highest neuroblastoma penetrance levels at 4 weeks of age ( 82 . 6% ) was observed in MYCN transgenic fish with homozygous loss of nf1a and heterozygous loss of nf1b , which produced the earliest onset and highest penetrance of neuroblastoma that we have observed in this model system . Finally , the penetrance of nf1a-/- fish with wild-type nf1b was still very high at 62 . 5% , indicating that of the two paralogues , the nf1a gene is primarily responsible for tumor suppression in the developing zebrafish sympathetic nervous system . To evaluate the histopathology of neuroblastoma tumors derived from PSNS cells with loss of nf1 , we focused on the genotypes with mutant nf1a alleles and wild-type nf1b alleles . Neuroblastomas that arose in fish with nf1a+/-;nf1b+/+;EGFP;MYCN and nf1a-/-;nf1b+/+;EGFP;MYCN genotypes were all strongly immunoreactive for tyrosine hydroxylase ( TH ) and the pan-neuronal marker Hu , comparable to the fish with wild-type nf1 alleles ( nf1a+/+;nf1b+/+;EGFP;MYCN; Figure 3 ) , indicating derivation from sympathetic neuroblast precursors ( TH+ , Hu+ ) . This is consistent with our previous report in nf1 wild-type fish that MYCN-induced neuroblastoma tumors arise from adrenal sympathetic neuroblasts that are prevented from differentiation into chromaffin cells by the overexpression of MYCN ( Zhu et al . , 2012 ) . Predominately TH+ , Hu- chromaffin cells were detected in the IRG of nf1a mutant fish lacking MYCN overexpression ( Figure 3—figure supplement 1 ) , indicating that loss of nf1 alone does not contribute to developmental arrest of PSNS neuroblasts at the TH+ , Hu+ stage . Taken together , these data demonstrate that loss of nf1a synergizes with overexpression of MYCN in the initiation of neuroblastoma , but that loss of nf1a is insufficient to initiate neuroblastoma on its own , possibly because loss of nf1a by itself does not lead to a block of terminal differentiation in PSNS neuroblasts in the IRG . 10 . 7554/eLife . 14713 . 006Figure 3 . Neuroblastomas arise from sympathetic neuroblast precursors in MYCN transgenic fish with loss of nf1 . H&E staining of sagittal sections through the tumors in the IRG of nf1a+/+;nf1b+/+;MYCN;GFP ( A ) , nf1a+/-;nf1b+/+;MYCN;GFP ( B ) and nf1a-/-;nf1b+/+;MYCN;GFP ( C ) fish at the age of 6 weeks . Immunohistochemical analysis of neuroblastoma markers tyrosine hydroxylase ( TH , D–F ) and Hu ( G–I ) expression on sagittal sections through tumors in the IRG of nf1a+/+;nf1b+/+;MYCN;GFP ( D , G ) , nf1a+/-;nf1b+/+;MYCN;GFP ( E , H ) and nf1a-/-;nf1b+/+;MYCN;GFP ( F , I ) fish at the age of 6 weeks . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 00610 . 7554/eLife . 14713 . 007Figure 3—figure supplement 1 . No TH+ , Hu+ neuroblasts was detected in the IRG of nf1a mutant zebrafish which had no overexpression of MYCN . H&E staining and immunohistochemical analysis of TH and Hu expression on sagittal sections through the IRG of nf1a+/+;nf1b+/+;GFP , nf1a+/-;nf1b+/+;GFP and nf1a-/-;nf1b+/+;GFP fish at the age of 6 weeks . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 007 Taking advantage of the optical transparency of our zebrafish neuroblastoma model , which allows us to monitor fluorescent tumor cell progression in vivo , we further investigated the compound effects of loss of nf1 and gain of MYCN on the kinetics of neuroblastoma progression . For these studies , we focused on loss of nf1a , because this orthologue is the most highly expressed ( Figure 1F ) , and its loss synergizes most prominently with overexpression of MYCN during tumorigenesis ( Figure 2C ) . Expansion of EGFP+ sympathoadrenal cells was observed in the IRG of nf1a heterozygous zebrafish ( nf1a+/-;nf1b+/+;MYCN;EGFP , Figure 4A middle panel ) , with an average increased expansion of 1 . 73 fold by the end of the 3-week period and a tumor induction rate of 43% ( 9/21 fish had tumors before they reached 6 weeks of age; Figure 4B ) . By contrast , the EGFP+ sympathoadrenal cells in the IRG of nf1a-/- homozygous zebrafish expanded much more rapidly , with an average 6 . 61-fold expansion rate within the 3 week period ( nf1a-/-;nf1b+/+; MYCN;EGFP ) and tumor development in 9 of 10 fish before 6 weeks of age ( Figure 4B ) . These data demonstrate that the loss of nf1a promotes MYCN-induced neuroblastoma onset and progression in vivo in zebrafish in a dose-dependent manner . 10 . 7554/eLife . 14713 . 008Figure 4 . Loss of nf1 promotes neuroblastoma tumor progression in MYCN transgenic fish . ( A ) Neuroblastoma development in representative fish of the nf1a+/+;nf1b+/+;MYCN;GFP , nf1a+/-;nf1b+/+;MYCN;GFP and nf1a-/-;nf1b+/+;MYCN;GFP genotypes over 3 to 6 weeks of age . Each fish was imaged weekly for 3 continuous weeks from 3 weeks of age . ( B ) Quantification of GFP+ sympathoadrenal cells in the IRG of fish with nf1a+/+;nf1b+/+;MYCN;GFP ( n = 10 ) , nf1a+/-;nf1b+/+;MYCN;GFP ( n = 21 ) and nf1a-/-;nf1b+/+;MYCN;GFP ( n = 15 ) genotypes , demonstrating tumor progression over 3 weeks . Tumors were scored when the GFP+ signals in the IRG exceeded the threshold defined by the dotted line . ( C ) IRG development in nf1a mutant zebrafish lacking overexpression of MYCN . Quantification of GFP+ sympathoadrenal cells in the IRG of fish with nf1a+/+;nf1b+/+;GFP ( n = 10 ) , nf1a+/-;nf1b+/+;GFP ( n = 10 ) and nf1a-/-;nf1b+/+;GFP ( n = 10 ) genotypes . The same tumor threshold line shown in panel B is included for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 008 Wild-type zebrafish transgenic for MYCN exhibit hyperplasia of sympathetic neuroblast precursors in the IRG from 3 to 5 weeks of life , followed by a developmentally-timed apoptotic response , resulting in regression of the hyperplastic sympathoadrenal cells in most fish . This leads to a low penetrance , with tumors developing in 515% of animals , and a relatively long latency of 12 to 20 weeks ( Figure 2C ) ( Zhu et al . , 2012 ) . In the current study , we observed positive staining for cleaved caspase-3 at 6 weeks of age in nf1 wild-type MYCN transgenic fish , consistent with our previous findings , but we did not detect staining for cleaved caspase-3 in the IRG of either nf1a+/-;nf1b+/+;MYCN;EGFP or nf1a-/-;nf1b+/+;MYCN;EGFP fish ( Figure 5 ) . Thus , loss of either one or both alleles of nf1a blocked the apoptotic response due to overexpression of MYCN in sympathetic neuroblasts , thus promoting tumor cell survival and leading to the development of neuroblastoma at high penetrance ( Figure 2 ) . The effects of loss of NF1 on cell survival therefore appear similar to the effects of mutational activation of the ALK tyrosine kinase ( Zhu et al . , 2012 ) , which also promotes activated signaling through the RAS pathway . 10 . 7554/eLife . 14713 . 009Figure 5 . Loss of nf1 suppresses apoptosis and enhances proliferation of tumor cells in MYCN-driven neuroblastoma . Immunohistochemical analysis of sagittal sections through tumors in the IRG of nf1a+/+;nf1b+/+;MYCN;GFP ( A , D ) , nf1a+/-;nf1b+/+;MYCN;GFP ( B , E ) and nf1a-/-;nf1b+/+;MYCN;GFP ( C , F ) fish at the age of 6 weeks , using antibodies against cleaved caspase-3 ( CC3; A–C ) and proliferating cell nuclear antigen ( PCNA; D-F ) . Dotted lines indicate tumor boundaries . Quantification of CC3-positive ( G ) and PCNA-positive ( H ) cells . Values are means + s . e . m . per section ( three nf1a+/+;nf1b+/+;MYCN;GFP , five nf1a+/-;nf1b+/+;MYCN;GFP and four nf1a-/-;nf1b+/+;MYCN;GFP tumors ) . p<0 . 05 , p<0 . 01 , *p<0 . 001 by two-tailed unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 00910 . 7554/eLife . 14713 . 010Figure 5—figure supplement 1 . Quantitative RT-PCR showing the relative expression levels of key genes involved in control of the cell cycle in neuroblastoma tumors with wild-type or mutant nf1a . RNA was extracted from tumors of 25-week old nf1a+/+;nf1b+/+;MYCN;GFP ( n = 2 ) and nf1a-/-;nf1b+/+;MYCN;GFP ( n = 3 ) zebrafish , in which the tumor mass were composed of ~80% GFP-positive cells . Note that ccnd3 was also examined , which was undetectable in these tumors . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 010 We also analyzed the effect of nf1 loss on the proliferative capacity of MYCN-overexpressing sympathetic neuroblast precursors by evaluating the fraction of cells expressing the proliferating cell nuclear antigen ( PCNA ) . We found that neuroblastomas arising in the nf1a+/- and nf1a-/- background exhibited a progressively higher percentage of nuclei with PCNA staining ( 30% for nf1a+/- and 80% for nf1a-/- ) compared to PCNA staining of 20% of nuclei in the wild-type nf1a+/+ genotype ( Figure 5 ) . Thus , loss of nf1a resulted in an increased neuroblastoma cell proliferative fraction , consistent with the observation that neuroblastoma tumors grow much more rapidly in fish with homozygous loss of nf1a ( Figure 4B ) . Together , our results support an important role of the tumor suppressor nf1 in restricting both the proliferative rate and survival of MYCN-overexpressing hyperplastic sympathetic neuroblast precursors . They also provide important cellular mechanisms that underlie the striking synergy between nf1 loss and overexpression of MYCN in neuroblastoma pathogenesis . We then isolated established neuroblastoma tumors of nf1a+/+;nf1b+/+;MYCN and nf1a-/-;nf1b+/+;MYCN zebrafish to examine expression of key genes involved in control of the cell cycle , including ccna1 , ccna2 , ccnb1 , ccnd1 , ccnd2 , ccnd3 , ccne , cdk2 , cdk4 , cdk6 and e2f1 . We did not detect significant differences in mRNA levels of these genes ( Figure 5—figure supplement 1 ) , suggesting that further experiments are required to decipher the molecular mechanism through which the loss of nf1a promotes increased proliferation of MYCN-overexpressing neuroblastoma tumor cells . Given the well-described role of neurofibromin as a negative regulator of RAS signaling , we postulated that nf1a loss in our neuroblastoma model would lead to activation of effector pathways downstream of RAS . To test this hypothesis , we first assessed the MYCN-independent activation of RAS effector pathways in the IRG of nf1-deficient fish by immunohistochemistry ( wild-type nf1a+/+;nf1b+/+;EGFP fish vs . nf1a+/-;nf1b+/+;EGFP and nf1a-/-;nf1b+/+;EGFP fish ) . The IRG of wild-type nf1a+/+;nf1b+/+;EGFP fish did not exhibit detectable phosphorylated ERK ( pERK ) , phosphorylated AKT ( pAKT ) or phosphorylated S6 ( pS6 ) , demonstrating a very low level of basal RAS activity in the normal sympathoadrenal cells of the IRG ( Figure 6—figure supplement 1 ) . We did not observe notable increases in pERK , pAKT or pS6 in the nontransformed IRG of nf1a mutant fish that lacked MYCN overexpression ( Figure 6—figure supplement 1 ) . Thus , the loss of nf1 did not cause a detectable increase in the basal levels of endogenous RAS signaling in normal sympathoadrenal cells of the IRG , probably because in the absence of MYCN overexpression they rapidly differentiate into mature chromaffin cells ( Zhu et al . , 2012 ) . We then assessed the role of nf1 in suppressing RAS effector pathways in the neuroblastoma cells of wild-type fish overexpressing MYCN . By immunohistochemistry and measurement of the area of the diaminobenzidine-peroxidase ( DAB ) -stained tumor cells , we detected high levels of pERK , pAKT and pS6 in neuroblastomas from 6-week-old fish ( Figure 6A , D , G and J ) , indicating activation of RAS effector pathways in the proliferating neuroblasts of wild-type nf1a+/+; nf1b+/+;MYCN fish that were blocked from terminal differentiation by overexpression of MYCN . In this context , it is not surprising that we observed much higher levels of activation of these effector pathways in neuroblastomas arising in nf1a-deficient fish , determined by significant increases of positive DAB-stained tumor areas ( Figure 6 B , C , E , F , H , I and J ) . Thus , the overexpression of MYCN blocks the differentiation of sympathoadrenal precursor cells , leaving them vulnerable to the potentiating effects of nf1 loss on RAS pathway activation , which in turn leads to the marked increased in neuroblastoma penetrance and growth rate . 10 . 7554/eLife . 14713 . 011Figure 6 . Loss of nf1 results in aberrant ERK/Akt/mTOR signaling in MYCN-driven neuroblastoma . Immunohistochemical analysis of sagittal sections through tumors in the IRG of nf1a+/+;nf1b+/+;MYCN;GFP ( A , D , G ) , nf1a+/-;nf1b+/+;MYCN;GFP ( B , E , H ) and nf1a-/-;nf1b+/+;MYCN;GFP ( C , F , I ) fish at the age of 6 weeks , using antibodies against phosphorylated ERK1/2 ( pERK , A–C ) , phosphorylated AKT ( pAKT , D–E ) and phosphorylated S6 ( pS6 , G– I ) . Dotted lines indicate the tumor boundaries . The quantification of pERK- , pAKT- and pS6-positive tumor areas are shown in ( J ) , with the red bars representing the median values . ns p>0 . 05 , p<0 . 05 , *p<0 . 01 by two-tailed unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 01110 . 7554/eLife . 14713 . 012Figure 6—figure supplement 1 . No aberrant ERK/AKT/mTOR signaling was detected in the IRG of nf1a mutant fish that lacked MYCN overexpression . ( A ) Immunohistochemical analysis of sagittal sections through the IRG of nf1a+/+;nf1b+/+;GFP , nf1a+/-;nf1b+/+;GFP and nf1a-/-;nf1b+/+;GFP fish at the age of 6 weeks , using antibodies against phosphorylated ERK1/2 ( pERK ) , phosphorylated AKT ( pAKT ) , phosphorylated S6 ( pS6 ) , cleaved Caspase-3 ( CC3 ) and proliferating cell nuclear antigen ( PCNA ) . The quantification of pERK- , pAKT- and pS6-positive IRG areas are shown in ( B ) , with the red bars representing the median values . ns p>0 . 05 , p<0 . 05 , **p<0 . 01 by two-tailed unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 012 To test the ability of the GRD domain to rescue tumor suppression in neuroblastoma , we bred the nf1a-/-;nf1b+/+;dbh:wt-GRD;dbh:mCherry transgenic line ( as in Figure 1H ) with the nf1a-/-;nf1b+/+;MYCN;EGFP fish . Only 12 of 31 nf1a-/-;nf1b+/+;MYCN;EGFP;wt-GRD;mCherry fish developed neuroblastoma tumors by the age of 12 weeks ( 38 . 7% , Figure 7A ) , compared to 35 of 41 nf1a-/-;nf1b+/+;MYCN;EGFP fish ( 85 . 4% , p<0 . 0001; Figure 7A ) , indicating that restoration of the NF1 GAP activity significantly reduced the penetrance of neuroblastoma in the nf1a-/-;nf1b+/+;MYCN;EGFP background . As a control , we also stably expressed the inactive GRD R1276P mutant ( designated 'mut-GRD' ) into nf1a-/-;nf1b+/+;MYCN;EGFP fish , which led to a frequency of neuroblastoma at 12 weeks of age that was identical to that in fish not transgenic for either GRD construct ( 90 . 0% , Figure 7A ) . Because the GRD R1276P mutant specifically disrupts the ability of the GRD to mediate the NF1 GAP activity , this result indicates that NF1 normally suppresses the development of neuroblastoma by downmodulating oncogenic signals from RAS . This finding stands in marked contrast to the inability of the same zebrafish line expressing the same level of GRD to rescue overgrowth of the SCG in normal development , indicating that even though a novel mechanism is responsible for restriction of PSNS neuron growth during normal embryogenesis , NF1 still acts as a classical GAP protein to suppress RAS-MAPK signaling for the suppression of neuroblastoma . 10 . 7554/eLife . 14713 . 013Figure 7 . GRD domain is required for the tumor suppressor function of nf1 in MYCN-driven neuroblastoma tumorigenesis . ( A ) Tumor penetrance of stable transgenic zebrafish with the genotypes of nf1a-/-;nf1b+/+;GFP;wt-GRD;mCherry ( n = 31 ) , nf1a-/-;nf1b+/+;GFP;mut-GRD;mCherry ( n = 20 ) and nf1a-/-;nf1b+/+;GFP;mCherry ( n = 41 ) at the age of 12 weeks . ( B ) nf1a-/-;nf1b+/+;MYCN;GFP fish injected with dbh:wt-GRD;dbh:mCherry ( n = 58 ) showed a significantly lower mCherry+ mosaic tumor rate compared with the sibling fish injected with dbh:mut-GRD;dbh:mCherry ( n = 25 ) at the age of 7 weeks ( Fisher’s exact test ) , indicating the tumor suppression function of the NF1 GRD domain . Most early tumors arose in nf1a-/-;nf1b+/+;MYCN;GFP fish injected with dbh:wt-GRD; dbh:mCherry only expressed GFP ( C ) but not mCherry fluorescent protein ( D ) . A significant subset of early tumors arose in nf1a-/-;nf1b+/+;MYCN;GFP fish injected with dbh:mut-GRD; dbh:mCherry did express both GFP ( E ) and mCherry ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 01310 . 7554/eLife . 14713 . 014Figure 7—figure supplement 1 . Scheme of the mosaic structure-function analysis using NF1-GRD . An equal mixture of dbh:mCherry and dbh:GRD was injected into the nf1a-/-;nf1b+/+;MYCN;GFP embryos at the one-cell stage . The injected transgenes randomly integrated into different cells and resulted in the mosaic expression of mCherry/GRD in their PSNS . A subpopulation of early tumors is expected to be mCherry+ , GFP+ if the tumorigenesis capacity of the mCherry+ PSNS cells is not suppressed . When the tumor suppression function of nf1 is restored , the mCherry+ PSNS cells are not capable of progress into early neuroblastoma tumor and all early tumors are expected to be mCherry- , GFP+ . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 014 To control for possible founder effects in our stable transgenic lines , we constructed primary mosaic GRD transgenic fish by coinjecting the dbh:wt-GRD or dbh:mut-GRD with the dbh:mCherry fluorescent marker into one-cell stage nf1a-/-;nf1b+/+;MYCN;EGFP embryos . Coinjected constructs integrate together in cells of the primary injectant , producing fish with mosaic coexpression of the transgenes ( Langenau et al . , 2008 ) . The MYCN-driven neuroblastomas that arise in these fish express the EGFP that is fused to MYCN , but mCherry is detected only if the tumors arise from a precursor cell that has integrated and expresses the mCherry and GRD transgenes ( Figure 7figure supplement 1 ) . As shown in Figure 7B , D and F , only 2 of 58 neuroblastoma tumors in fish coinjected with the functionally active wt-GRD domain and the mCherry transgene exhibited mCherry fluorescence at 7 weeks of age ( 3 . 5% , Figure 7B ) . By contrast , coinjection of the mut-GRD domain and the mCherry transgene into nf1a-/-;nf1b+/+;MYCN;EGFP fish yielded strong mCherry fluorescence in 5 of 25 tumors that developed by 7 weeks of age ( 20 . 0%; p=0 . 0237; Figure 7B , D , and F ) . This experiment in primary injectants confirms our results in stable GRD transgenic lines , demonstrating that NF1 functions differently in normal development and tumorigenesis of the PSNS , and the tumor suppression function of NF1 in MYCN-induced neuroblastoma is mediated through the GAP activity of the GRD domain . Because very aggressive growth properties of MYCN-induced neuroblastomas in zebrafish with loss of both alleles of nf1a ( See Figure 2 ) are attributable to aberrant activation of RAS-MAPK signaling ( see Figures 6 and 7 ) , we evaluated the importance of MEK inhibition for the treatment of these tumors . For these experiments , we used the FDA-approved MEK inhibitor trametinib to treat primary neuroblastoma tumors that were growing in 3-week-old nf1a-/-;nf1b+/+;MYCN; EGFP fish ( Figure 8 ) . After 1 week of treatment with trametinib , we observed that the drug significantly reduced the rate of tumor growth , but did not by itself cause shrinkage of the tumor ( Figure 8A ) . 10 . 7554/eLife . 14713 . 015Figure 8 . The MEK inhibitor trametinib synergizes with isotretinoin in suppressing nf1-deficient neuroblastoma in vivo . ( A ) Neuroblastoma growth in nf1a-/-;nf1b+/+;MYCN;GFP fish treated with vehicle control , trametinib , isotretnoin or combinations of trametinib and isotretnoin . Representative fish are shown in ( C ) . ( B ) Synergistic effects of trametinib and isotretinoin on tumor suppression were analyzed by isobologram analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 14713 . 015 The real impact of inhibition of RAS-MAPK signaling by trametinib was evident when it was tested together with isotretinoin ( Figure 8 ) . Isotretinoin , also known as 13-cis retinoic acid , is a retinoid used as frontline therapy for childhood neuroblastoma and improves the event-free survival of combination chemotherapy ( Matthay et al . , 2009; Park et al . , 2009 ) . In our nf1a-/-;nf1b+/+;MYCN; EGFP fish , isotretinoin alone leads to reduced tumor growth , with essentially stable tumor size over 7 days of treatment at the highest dosages ( Figure 8A ) . However , the combination of isotretinoin and trametinib was much more active in reducing tumor size , in many fish reducing the GFP fluorescence from the dβh promoter to the level of the normal IRG at 4 weeks of life ( Figure 8A and C ) . Importantly , synergistic antitumor effects of trametinib and isotretinoin were documented in vivo at several different dosage combinations by isobologram analysis ( Figure 8B ) , indicating that inhibition of RAS-MAPK signaling can significantly improve the treatment of this very aggressive form of neuroblastoma when it is combined with the inhibition of other key pathways . Because of the very high penetrance and rapid onset of neuroblastoma in our nf1-deficient , MYCN-transgenic zebrafish model , it becomes one of a very few systems in which extensive analysis of the synergistic activity of two or more drugs can be evaluated in primary tumors in vivo . This capability is especially valuable given the recent evidence documenting the profound impact of RAS-MAPK hyperactivation in therapy-resistant neuroblastoma of childhood ( Eleveld et al . , 2015 ) . In this study , we use a zebrafish model to demonstrate that NF1 loss can potentiate the tumorigenic effects of MYCN-overexpression in high-risk neuroblastoma . We show that nf1 deficiency in the fish leads to aberrant activation of RAS-MAPK signaling in MYCN-induced neuroblastoma , which promotes both tumor cell survival and proliferation , leading to marked acceleration of tumor onset and increased tumor penetrance . To establish the underlying mechanism of these observations , we performed in vivo structure-function analyses to show that the tumor suppression function of NF1 is mediated through the GAP activity of its GRD domain . The same rescue construct does not suppress the PSNS overgrowth that accompanies nf1 loss during development , supporting earlier work implicating a different and still-to-be-identified activity of NF1 in the restriction of the normal growth of PSNS neuroblasts during embryogenesis . We and others have shown that inactivation of the NF1 gene in human neuroblastoma and melanoma cell lines does not result in increased levels of the GTP-bound form of Ras compared to similar lines that express normal levels of the NF1 protein ( Johnson et al . , 1993; The et al . , 1993 ) . Ostensibly , this would seem inconsistent with our current findings in the zebrafish model implicating downregulation of the RAS pathway in NF1-mediated tumor suppression . However , a recent whole-genome sequencing study revealed that loss of NF1 is one of a strikingly high frequency of diverse mutations that hyperactivate the RAS-MAPK signaling pathway in relapsed neuroblastoma ( Eleveld et al . , 2015 ) . Thus , our inability to demonstrate increased GTP on RAS in NF1-deficient neuroblastoma cell lines may reflect the strong activation of the RAS-MAPK signaling pathway in essentially all high-risk neuroblastomas , if not by NF1 loss then by direct mutation of RAS itself or other types of mutations . This interpretation becomes more appealing when one realizes that neuroblastomas able to grow in vitro as continuous cell lines are invariably from patients with the very highest risk of treatment failure , and often are obtained from tumor biopsies at the time of relapse . The same relationship likely applies to melanoma cell lines as well , since most of these tumors harbor activating mutations in the RAS-MAPK pathway , including BRAF V600E and NRAS Q61L/R , as well as the loss of NF1 in a subset of cases ( Cancer Genome Atlas Network , 2015; Hodis et al . , 2012 ) . Studies of neural crest and PSNS development in Nf1 null mice were the first to implicate a GAP-independent activity of Nf1 in restricting the growth of sympathoadrenal cells , as the GRD domain was not able to rescue the overgrowth of sympathetic neuroblasts in the adrenal medulla ( Ismat et al . , 2006 ) . Our results support those studies , in that the GRD domain expressed in the same zebrafish line that suppressed neuroblastoma pathogenesis failed to suppress sympathetic neuroblast overgrowth in nf1-deficient zebrafish embryos . Thus , GAP-independent activities of NF1 appear to restrict the growth of normal neural crest-derived tissues during development , as the PSNS overgrowth phenotype was not rescued by the same level of GRD restoration that suppressed the accelerated onset of MYCN-driven neuroblastoma ( Figures 1 and 7 ) . The recent discovery that mutations leading to activated RAS-MAPK signaling occur in nearly 80% of relapsed neuroblastomas has clear implications for the selection of targeted therapy for this tumor ( Eleveld et al . , 2015 ) . Indeed , it suggests that MEK inhibitors and other agents targeting the RAS-MAPK pathway might suppress the outgrowth of resistant clones and therefore should be investigated as part of the initial combination therapy for high-risk neuroblastoma . Our zebrafish model of MYCN-driven neuroblastoma with nf1 loss is ideally suited for the rapid in vivo analysis of the effects of candidate small-molecule inhibitors selected for this purpose . Neuroblastomas can be detected in most MYCN-transgenic fish with nf1a loss by 3 weeks of age , when the fish are very small , making it feasible to test the effectiveness of many drugs and drug combinations for their ability to kill primary neuroblastoma cells in vivo . This advantage of the model is readily apparent in our demonstration of marked antitumor synergy between the MEK inhibitor trametinib and the retinoid isoretinoin in MYCN-overexpressing fish with loss of nf1 ( Figure 8 ) . Studies using our model should be a valuable asset in devising new therapeutic strategies for neuroblastomas with mutations affecting the RAS-MAPK pathway , which appear to be a major cause of relapse in children with this devastating tumor . The previously described nf1a and nf1b mutant zebrafish lines ( Shin et al . , 2012 ) were crossed with transgenic lines including Tg ( dbh:EGFP ) and Tg ( dbh:EGFP-MYCN ) ( Zhu et al . , 2012 ) for this study . The compound mutant fish were fin-clipped and genotyped for the nf1a and nf1b as previously described ( Shin et al . , 2012 ) . All zebrafish studies and maintenance of the animals were performed in accordance with Dana-Farber Cancer Institute IACUC-approved protocol #02–107 . The DNA construct for dbh:wt-GRD , dbh:mut-GRD and dbh:mCherry was subcloned using the Multisite Gateway System ( Invitrogen , Carlsbad , CA ) as previously described ( Invitrogen ) ( Ismat et al . , 2006; Zhu et al . , 2012 ) . Embryos were injected with these DNA constructs at the one-cell stage as previously described ( Zhu et al . , 2012 ) and grown to adulthood . Fin clips from the offspring were genotyped for the stable integration and germline transmission of the transgenes . nf1a+/-; nf1b+/-; dbh:EGFP and nf1a+/-; nf1b+/-; dbh:EGFP-MYCN mutant zebrafish were crossed , and offspring were screened every 2 weeks , starting from the age of 4 weeks , for fluorescent EGFP expressing cell masses indicative of tumors ( Zhu et al . , 2012 ) . Once an EGFP-positive cell mass was identified , the individual fish were separated and carefully monitored weekly for at least 4 weeks for tumor progression . Only the fish with progressing EGFP-positive cell masses were scored as tumor fish and analyzed further by H&E staining and immunohistochemical assays . All fish were genotyped for nf1a and nf1b at the age of 8 weeks . Wild-type dbh:mCherry embryos ( n = 2000 ) were collected at the age of 1 week , and dissociated into single-cell suspensions using 0 . 05% trypsin-EDTA ( Life Technologies , Carlsbad , CA ) . Cells were filtered through a 40-μm cell strainer ( Falcon , Corning , NY ) and resuspended in PBS . Fluorescence-activated cell sorting was performed on a BD FACSAria II CORP UV ( Dana-Farber Cancer Institute Hematologic Neoplasia Flow Cytometry Core ) , and mCherry-positive cells , as well as the mCherry-negative control cells , were collected for RNA extraction . For juvenlie zebrafish at the ages of 3 , 4 , and 6 weeks , 5 wild-type fish at each age were homogenized in a tissue grinder for RNA extraction . Total RNA was extracted with Trizol reagent ( Invitrogen ) and purified with the Qiagen RNeasy kit ( Qiagen , Santa Clarita , CA ) according to the manufacturer’s instructions . cDNA was synthesized from total RNA using the iScript cDNA synthesis kit ( Biorad , Hercules , CA ) according to the manufacturer’s instructions ( Biorad ) . The Q-RT-PCR reactions were performed with the SYBR green PCR Core Reagents kit ( Applied Biosystems , Foster City , CA ) and ViiA 7 Real-Time PCR System ( Life Technologies ) according to the manufacturer’s instructions . All Q-RT-PCR assays were performed in triplicate , with β-actin used as an endogenous control . Q-RT-PCR primer sequences spanning exon-exon junctions were as follows: nf1a forward , 5’-AAATTCCAGACTACGCCGAGC-3’ and reverse , 5’- TATAAACTATAGGGCCCTCTGGGGA-3’; nf1b forward , 5’- TGGCGCAGAAGTTTGCATTTCAATA -3’ and reverse , 5’- GCAATGACTGTGGCTTCGATT-3’; β-actin forward , 5’- TTCCTGGGTATGGAATCTTGCG -3’ and reverse , 5’- GTGGAAGGAGCAAGAGAGGTG -3’; ccna1 forward , 5’- TGGCTCAGGGTCATTTATGG-3’ and reverse , 5’- TAACTTCGCATTCACGCAGG-3’; ccna2 forward , 5’-TCCACTGGAGGCCAGTTTTG-3’ and reverse , 5’- GACTTGACCTCCATTTCCCG-3’; ccnb1 forward , 5’-ATTCTCCTCAGTGTTTCTCCAGTC-3’ and reverse , 5’-AAGTGTAGATGTCTCGCTCATATTC-3’; ccnd1 forward , 5’-TTGCTGCGAAGTGGATACCATAAG-3’ and reverse , 5’-AGGCACAATTTCTTTCTGAACACAC-3’; ccnd2 forward , 5’- CCGTCCTGATCCGAATCTTCTG-3’ and reverse , 5’-GCCACCATCCTCCGCATAAAG-3’; ccnd3 forward , 5’-ACGGCTACAGAGCTGAAGTT-3’ and reverse , 5’-CATCTGCTCGGCGCTAACA-3’; ccne forward , 5’-ACAACCTGCTCGGAAAAGACAAG-3’ and reverse , 5’-CACAAACCTCCATTAGCCAGTCC-3’; cdk2 forward , 5’-TCGCGCTGAAGAAAATCCGA-3’ and reverse , 5’-ACGCAACTTGACTATGTTAGGGT-3’; cdk4 forward , 5’-TGAGCCAGTAGCAGAGATCG-3’ and reverse , 5’-AGTGGGAGTCCGTCCTGATT-3’; cdk6 forward , 5’-TCTCACCGTGTGGTTCATCG-3’ and reverse , 5’-ATGTCACAACCACCACGGAA-3’; e2f1 forward , 5’-ACAACATCCAGTGGCTAGGG-3’ and reverse , 5’-TTCGTCCAGTTTCTCCTCGG-3’ . Zebrafish were euthanized in tricaine anesthetic , fixed in 4% paraformaldehyde at 4°C for 2 days , and decalcified with 0 . 25 M EDTA , pH 8 . 0 , for at least 24 hr . Paraffin sectioning followed by hematoxylin and eosin ( H&E ) staining or immunohistochemistry ( IHC ) was performed at the Dana-Farber/Harvard Cancer Center Research Pathology Core . Primary antibodies included Phospho-p44/42 MAPK ( ERK1/2 ) ( Thr202/Tyr204 , Cell Signaling #4370; 1:150 ) , Phospho-AKT ( Ser473 , Cell Signaling #4060 ) , Phospho-S6 ribosomal protein ( Ser240/244 , Cell Signaling #4838 ) , PCNA ( PC10 , EMD Millipore; 1:100 ) , cleaved Caspase-3 ( Cell Signaling #9664; 1:250 ) , TH ( Pel-Freez # P40101 , 1:500 ) and HuC/D ( Invitrogen #A-21271 , 1:200 ) . Antibody binding was detected with a diaminobenzidine-peroxidase ( DAB ) visualization system ( EnVision+ , Dako , Carpinteria , CA ) . Mayer’s hematoxylin was used for counterstaining . For brightfield DIC images , a Zeiss Axio Imager . Z1 compound microscope equipped with an AxioCam HRc was used . For quantification of immunohistochemistry staining , brightfield images were taken with the Mantra Quantitative Pathology Workstation ( Perkin Elmer , Norwalk , CT ) and analyzed with ImageJ . The color deconvolution plugin of ImageJ and the 'H DAB' vector was used to separate the hematoxylin and DAB stains ( Ruifrok and Johnston , 2001 ) , and the kidney tubules adjacent to the IRG or tumor region were applied as internal references to define threshold of DAB staining for each IRG or tumor region . For live imaging , zebrafish and embryos were anaesthetized using 0 . 016% tricaine ( ) and mounted in 4% methycellulose ( ) . A Nikon SMZ1500 microscope equipped with a Nikon digital sight DS-U1 camera was used for capturing both the bright field and fluorescent images from live zebrafish and embryos . For PSNS and neuroblastoma quantification , all animals in the same experiments were imaged under the same conditions and the acquired fluorescent images were quantified using the ImageJ software by measuring the EGFP covered area . For neuroblastoma quantification , the fluorescent area was normalized against the surface area of the fish head as fish size was variable . Overlays were created using ImageJ and Adobe Photoshop 7 . 0 . 1 . nf1a-/-; nf1b+/+; dbh:EGFP; dbh:EGFP-MYCN zebrafish with GFP+ tumor were imaged individually at the age of 3 weeks , separated and treated with trametinib ( ) and/or isotretinoin ( Selleck Chemicals , Houston , TX ) with refreshment every 2 days . Drug synergism was evaluated using the CalcuSyn Software . Statistical analysis was performed with Prism 5 software ( GraphPad ) . Kaplan-Meier methods and the log-rank test were applied to assess the rate of tumor development in Figure 2 . Fish that died before they had evidence of EGFP-positive masses were censored . Fisher’s exact test was used to assess the difference between tumor rate in wild-type versus mutant fish in Figure 7 . A two-tailed unpaired t-test with confidence intervals of 95% was used for the analyses in Figures 5 , 6 and 7 . The quantitative data in Figures 4 and 7 are reported as means with standard errors of means ( s . e . m ) . For Figure 8 , a Mann Whitney test with confidence intervals of 95% was used for the analysis and the quantitative data are reported as median .	Neuroblastoma is one of the most common childhood cancers and is responsible for about 15% of childhood deaths due to cancer . The neuroblastoma tumors arise in cells that develop into and form part of the body’s nervous system . Many researchers have studied the genetics of this disease and have recognised common patterns . In particular , neuroblastomas can occur when a protein called MYCN is over-produced and a tumor suppressor protein called NF1 is lost . NF1 is a large protein with several distinct parts or domains . The most studied domain of NF1 is called the GRD , and it is mainly responsible for dampening down signals that cause cells to grow , specialize and survive . However , experiments in mice have revealed that this protein uses its other domains to control the normal development of part of the nervous system . He et al . wanted to know which domains of NF1 are important for suppressing the growth of neuroblastomas . The experiments were conducted in zebrafish that had been engineered to produce an excess of the human version of MYCN . When He et al . also deleted the gene for the zebrafish’s version of NF1 , the fish quickly developed neuroblastomas . Supplying the zebrafish with just the GRD of NF1 was enough to supress the growth of the tumors . These experiments show that NF1 uses different domains and signalling pathways to regulate the normal development of part of the nervous system and to prevent formation of neuroblastoma . These engineered zebrafish represent an animal model of neuroblastoma that mimics the human disease in many ways . This model will make it possible to test new drug combinations and to find more effective treatments for neuroblastoma patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "cancer", "biology" ]	2016	Synergy between loss of NF1 and overexpression of MYCN in neuroblastoma is mediated by the GAP-related domain
Migraine is the sixth most prevalent disease worldwide but the mechanisms that underlie migraine chronicity are poorly understood . Cytoskeletal flexibility is fundamental to neuronal-plasticity and is dependent on dynamic microtubules . Histone-deacetylase-6 ( HDAC6 ) decreases microtubule dynamics by deacetylating its primary substrate , α-tubulin . We use validated mouse models of migraine to show that HDAC6-inhibition is a promising migraine treatment and reveal an undiscovered cytoarchitectural basis for migraine chronicity . The human migraine trigger , nitroglycerin , produced chronic migraine-associated pain and decreased neurite growth in headache-processing regions , which were reversed by HDAC6 inhibition . Cortical spreading depression ( CSD ) , a physiological correlate of migraine aura , also decreased cortical neurite growth , while HDAC6-inhibitor restored neuronal complexity and decreased CSD . Importantly , a calcitonin gene-related peptide receptor antagonist also restored blunted neuronal complexity induced by nitroglycerin . Our results demonstrate that disruptions in neuronal cytoarchitecture are a feature of chronic migraine , and effective migraine therapies might include agents that restore microtubule/neuronal plasticity . Migraine is an extremely common neurological disorder that is estimated to affect 14% of the global population , making it the third most prevalent disease worldwide ( Global Burden of Disease Study 2013 Collaborators , 2015; GBD 2016 Disease and Injury Incidence and Prevalence Collaborators , 2017 ) . One particularly debilitating subset of migraine patients are those with chronic migraine , which is defined as having more than 15 headache days a month ( Headache Classification Committee of the International Headache Society ( IHS ) , 2013 ) . Despite its high prevalence , migraine therapies are often only partially effective or are poorly tolerated , creating a need for better pharmacotherapies ( Ford et al . , 2017 ) . Recent clinical success of antibodies and antagonists against calcitonin gene-related peptide ( CGRP ) and CGRP receptor demonstrate the effectiveness of targeted migraine therapeutics . While there has been more research into understanding the molecular mechanisms of migraine , there remains much to be discovered . Neuroplastic changes play an important role in a variety of chronic neuropsychiatric conditions ( Descalzi et al . , 2015 ) , and epigenetic alterations through histone deacetylases ( HDACs ) are frequently investigated . HDACs are best characterized for their ability to deacetylate histones , promoting chromatin condensation and altered gene expression ( Krishnan et al . , 2014 ) . Intriguingly , some HDACs can also deacetylate non-histone targets , including proteins involved in cytoarchitecture and dynamic cellular structure . Due to its cytoplasmic retention signal , HDAC6 is primarily expressed in the cytosol ( Valenzuela-Fernández et al . , 2008 ) , and one of its primary targets for deacetylation is α-tubulin ( Valenzuela-Fernández et al . , 2008; d'Ydewalle et al . , 2012 ) . α- and β-tubulin form heterodimers that make up microtubules , which are a major component of the cytoskeleton , and regulate intracellular transport , cell morphology , motility , and organelle distribution ( Janke and Bulinski , 2011; Janke and Montagnac , 2017 ) . Microtubules undergo multiple cycles of polymerization and depolymerization , creating a state of dynamic instability ( Janke and Bulinski , 2011 ) . Tubulin displays a variety of post-translational modifications , including α-tubulin acetylation , which occurs endogenously through α-tubulin N-acetyltransferase I ( αTAT1 ) , and it is correspondingly deacetylated by HDAC6 ( Janke and Montagnac , 2017 ) . Tubulin acetylation is associated with increased flexibility and stability of microtubules ( Xu et al . , 2017 ) . In contrast , deacetylated microtubules are more fragile and prone to breakage or increased cycling between assembly and disassembly ( Xu et al . , 2017 ) . Microtubules are important for cellular response to injury and play a role in neurite branching ( Gallo , 2011 ) ; and microtubule dynamics influence neuronal signaling and mediate axonal transport ( Covington et al . , 2009; Braun et al . , 2011; Schappi et al . , 2014; Jin et al . , 2017; Van Helleputte et al . , 2018 ) . Importantly , changes in cellular structure such as alterations in dendritic spine density , have been implicated in disease chronicity ( Jochems et al . , 2015; Forrest et al . , 2018; Singh et al . , 2018; Nestler and Lüscher , 2019 ) . The aims of this study were to determine if altered neuronal cytoarchitecture facilitates the chronic migraine state and whether modifying this by inhibition of HDAC6 would be effective in mouse models of migraine . We observed decreased neuronal complexity in headache-processing brain regions in the nitroglycerin ( NTG ) model of chronic migraine-associated pain . We further demonstrated that treatment with HDAC6 inhibitor reversed these cytoarchitectural changes and correspondingly decreased cephalic allodynia . These studies were extended to a mechanistically distinct model of migraine , cortical spreading depression ( CSD ) , which is thought to be the electrophysiological correlate of migraine aura . Again , we observed decreased neuronal complexity in migraine related sites , which was reversed by a HDAC6 inhibitor . To investigate the translational implication , we also tested the effect of olcegepant , a CGRP receptor inhibitor , and found that it alleviated chronic allodynia induced by NTG , and restored cytoarchitectural changes associated with chronic migraine-associated pain . These results suggest a novel mechanism for migraine pathophysiology and establish HDAC6 as a novel therapeutic target for this disorder . Changes in the structural plasticity of neurons have been observed in a number of neuropsychiatric disorders and can serve as a marker of disease chronicity ( Forrest et al . , 2018; Nestler and Lüscher , 2019 ) . To investigate if this was also the case in migraine , we treated male and female C57BL/6J mice every other day for 9 days with NTG or vehicle and tested for periorbital mechanical responses on days 1 , 5 , and 9 ( Figure 1A ) . NTG induced severe and sustained cephalic allodynia as measured by von Frey hair stimulation of the periorbital region as compared to vehicle animals on the same day ( Figure 1B ) . Mice were sacrificed on day 10 , 24 hr after the final NTG/VEH treatment , and neuronal size and arborization were examined through a Golgi staining procedure in a key cephalic pain processing region , the trigeminal nucleus caudalis ( TNC ) ( Figure 1C; Goadsby et al . , 2017 ) . We observed a dramatic decrease in neuronal complexity after NTG treatment ( Figure 1D ) . Neurons of chronic NTG-treated mice had significantly fewer branch points ( Figure 1E ) , and shorter neurites resulting in decreased overall length of the neurons ( Figure 1F ) . Further examination of the complexity of the neurons using Sholl Analysis , showed a significant decrease in the number of intersections following NTG treatment ( Figure 1G–I ) . In addition to the TNC we also determined if other brain regions related to central pain processing were affected by chronic NTG treatment . We examined the somatosensory cortex ( SCx ) and periaqueductal gray ( PAG ) of these mice and found similar results , where neurons from NTG-treated mice had fewer branch points , were shorter in length , and had fewer intersections ( Figure 1—figure supplement 1A–D and E–H , respectively ) . To ensure that this effect was associated with migraine-pain processing and not a non-specific effect of NTG we also analyzed neuronal complexity in the nucleus accumbens shell ( NAc ) , a region more commonly associated with reward , and found no alteration in number of branches , total neuron length , or Sholl analysis for cells in this region ( Figure 1—figure supplement 1I–L ) . Furthermore , we also examined the dorsal horn of the lumbar and cervical spinal cord , important sites for pain processing in the peripheral nervous system . No differences in NTG versus vehicle controls in number of branches , total neuron length , or Sholl analysis were observed ( Figure 1—figure supplement 1M–T ) . These results suggest that decreased neuronal complexity may be a feature that maintains the chronic migraine state; a previously undiscovered phenomenon . We hypothesized that if we could restore migraine-compromised cytoarchitectural complexity , this might also relieve cephalic pain . Recent studies demonstrate that increased tubulin acetylation facilitates microtubule flexibility and prevents microtubule breakage ( Janke and Montagnac , 2017; Portran et al . , 2017; Xu et al . , 2017 ) . Thus , we hypothesized that inhibiting HDAC6 to promote microtubule stability may restore the neuronal complexity observed following chronic NTG . We tested the selective HDAC6 inhibitor , ACY-738 , in the chronic NTG model . Mice were treated with NTG or VEH for 9 days . On day 10 , mice were injected with ACY-738 , after 4 hr , tissue was processed and analyzed by western blot . ACY-738 treatment resulted in a significant increase in the ratio of acetylated α-tubulin to total tubulin within the trigeminal ganglia ( TG ) , TNC , SCx , NAc , and spinal cord ( Figure 2—figure supplement 1 ) . A separate group of mice were analyzed by neuronal tracing in the TNC , 4 hr post-ACY-738 ( Figure 2A ) . Again , chronic NTG treatment caused a decrease in branch points ( Figure 2B ) , combined neurite length ( Figure 2C ) , and number of intersections ( Figure 2D–F ) . In contrast , treatment with ACY-738 led to a significant increase in these measures in both chronic vehicle and NTG groups ( Figure 2A–F ) . We next determined if this restored neuronal complexity would affect behavioral outcomes . Mice were treated with chronic intermittent NTG or vehicle for 9 days . On day 10 , baseline cephalic allodynia was observed in mice treated chronically with NTG but not vehicle ( Figure 3A , baselines ) . Mice were then injected with ACY-738 or vehicle and tested 4 , 24 , or 48 hr later . ACY-738 significantly reversed cephalic allodynia in NTG treated mice for up to 24 hr post-injection . Mechanical responses in ACY-738 treated animals , not treated with NTG were unaffected ( Figure 3A , VEH-ACY ) , suggesting that augmentation of neuronal complexity by HDAC6 inhibitor in a pain-free animal does not alter endogenous pain processing . Interestingly , the half-life of ACY-738 is only 12 min ( Jochems et al . , 2014 ) , thus short-term inhibition of HDAC6 still produced long-lasting behavioral and cytoarchitectural changes . To explore further the correlation between neuronal complexity and allodynia , we examined the cytoarchitecture of neurons following treatment with ACY-738 at the 24 and 48 hr time points ( Figure 3B ) . A separate cohort of animals were treated with NTG/Veh for 9 days and on day 10 received vehicle/ACY-738 and were sacrificed 24 or 48 hr post-injection for neuronal analysis . At 24 hr post-ACY-738 , NTG-Veh animals continued to show significantly fewer branch points as compared to the Veh-Veh controls ( Figure 3C ) . In contrast , the NTG-ACY group had a significantly higher number of branches compared to NTG-Veh . Interestingly , at 24 hr we no longer observed increased neuronal complexity in the vehicle-ACY-738 group , as was observed at the 4 hr time point ( Figure 2 ) . Similar results were observed for total neuronal length ( Figure 3D ) and interactions ( Figure 3E ) . These data further strengthen the correlation between cytoarchitectural complexity and decreased allodynia , as at both 4 and 24 hr post-injection ACY-738 has anti-allodynic effects . At 48 hr post-ACY-738 we observed that compared to the VEH-VEH controls , the number of branch points were now significantly lower in both the NTG-vehicle and NTG-ACY groups ( Figure 3F ) ; and similar results were observed for total neuronal length ( Figure 3G ) . Only the NTG-Veh group showed significantly fewer interactions , although there was a trend in the NTG-ACY group ( Figure 3H ) . Taken together , these data show that the anti-allodynic effects of ACY-378 correspond with the times at which it also restores neuronal complexity . We confirmed that this behavioral effect resulted from changes in HDAC6 inhibition . We first tested two pan-HDAC inhibitors: the well-characterized inhibitor trichostatin A ( TSA , Figure 4A ) ; and a novel brain-penetrant pan-HDAC inhibitor , RN-73 ( Abdelkarim et al . , 2017; Figure 4B ) . Both significantly reversed chronic NTG-induced allodynia , albeit for a much shorter duration than ACY-738 . In contrast , when we tested the Class I , HDAC1 and 2 selective inhibitor , ASV-85 ( Supplementary file 1 ) , we did not observe any change in NTG-induced chronic allodynia relative to vehicle controls ( Figure 4C ) . These data further support our finding that chronic migraine-associated pain can be blocked specifically by HDCA6 inhibition , and that this effect is not likely due to acetylation of histones in the cell nucleus . Considering the acute effectiveness of ACY-738 , we next determined if sustained HDAC6 inhibition could block the establishment of NTG-induced hypersensitivity . In this case , ACY-738/VEH was injected 2 hr before NTG/VEH administration every other day for 9 days . Basal cephalic mechanical thresholds were assessed before drug treatment on days 1 , 5 , and 9 . As seen previously , chronic NTG resulted in the development of a chronic allodynia ( Figure 4D , VEH-NTG vs VEH-VEH ) ; and concurrent treatment with ACY-738 prevented the development of this allodynia ( VEH-NTG vs ACY-NTG ) . These data demonstrate that chronic HDAC6 inhibition can prevent the development of chronic migraine-associated pain , further supporting the potential of HDAC6 inhibitors as a therapeutic target for migraine . HDAC6 expression is enriched in certain brain regions , such as the dorsal raphe ( Espallergues et al . , 2012 ) ; and to the best of our knowledge HDAC6 expression in head pain processing regions is not well characterized . In situ hybridization using RNAScope and immunohistochemical analysis ( Figure 4—figure supplement 1A , B ) revealed abundant expression of HDAC6 transcripts in TG , TNC , and SCx . Gene expression analysis revealed that , of these regions , chronic NTG treatment increased HDAC6 expression in the TG ( Figure 4-figure supplement 1C ) , which are the first-order cells regulating cephalic pain processing . Thus , HDAC6 is expressed and regulated dynamically in regions that are critical for migraine-associated pain processing . CSD is an electrophysiological property thought to underlie migraine aura . It is mechanistically and etiologically distinct from the NTG model of migraine pain , and reduction of CSD events is a feature of many migraine preventives ( Ayata et al . , 2006 ) . Thus , we examined whether CSD propagation was also affected by HDAC6 inhibition . Briefly , the skull was thinned in an anesthetized animal to reveal the dural vasculature and cortex underneath ( Figure 5A ) . Two burr holes were made , and the more rostral was used to continuously drip KCl onto the dura to induce CSD , while local field potentials ( LFPs ) were recorded from the caudal burr hole . The somatosensory/barrel cortex was targeted , as it is more sensitive to CSD induction ( Bogdanov et al . , 2016 ) . Throughout the 1 hr recording , CSDs were identified by visual shifts in light and sharp decreases in the LFP ( Figure 5B–C ) . Pretreatment with ACY-738 resulted in significantly fewer CSD events relative to vehicle controls ( Figure 5D ) , indicating that HDAC6 inhibition is also effective in this mechanistically separate migraine model . We next examined the neuronal complexity of pyramidal neurons within the somatosensory cortex following CSD induction . Sham mice that underwent anesthesia and surgery , but did not receive KCl , were used as controls . Mice were pretreated with ACY-738 or vehicle , underwent CSD or sham procedure , and were immediately sacrificed for Golgi staining of the SCx ( Figure 6A–B ) . In the somatosensory cortex , CSD evoked a significant decrease in branch points ( Figure 6C ) and total length of neurons ( Figure 6D ) . In addition , CSD also resulted in a significant reduction in the number of branches in neurons of the TNC ( Figure 6—figure supplement 1 ) , a region that is known to be activated following CSD events ( Zhang et al . , 2011 ) . In contrast , ACY-738 increased neuronal complexity in the cortex in both sham and CSD groups . Sholl analysis demonstrated a dramatic decrease in neuronal complexity after CSD , while ACY-738 treatment had the opposite effect ( Figure 6E–G ) . These results demonstrate that decreased neuronal complexity is also observed in a second , mechanistically distinct model of migraine , and that HDAC6 inhibition can prevent these changes in neuronal cytoarchitecture and decrease CSD events . We next sought to determine if migraine-selective therapies could influence neuronal cytoarchitecture; and we tested the small molecule CGRP receptor antagonist , olcegepant , in the chronic NTG model ( Olesen et al . , 2004 ) . Mice developed a sustained allodynia to repeated NTG treatment ( Figure 7A ) . On day 10 , 24 hr after the final NTG injection , baseline mechanical responses were determined and mice were treated with olcegepant or vehicle . Olcegepant significantly inhibited NTG-induced cephalic allodynia ( Figure 7B ) , similar to previously published reports ( Christensen et al . , 2019 ) . Subsequent Golgi analysis of TNC revealed cytoarchitectural alterations in this cohort of animals ( Figure 7C ) . As was observed previously , chronic NTG treatment decreased the number of branch points ( Figure 7D ) , combined neurite length ( Figure 7E ) , and number of intersections using Sholl analysis ( Figure 7F–H ) . Interestingly , olcegepant treatment restored neuronal complexity induced by chronic NTG , but had no effect in chronic vehicle-treated mice ( Figure 7C–H ) . These data reinforce the concept that altered neuronal complexity could be a feature of chronic migraine , and that restoration of these changes may be a marker of effective migraine treatment . Our results indicate that in models of chronic migraine and aura there is a dysregulation of cellular plasticity resulting in decreased neuronal complexity . We found that following the establishment of chronic cephalic allodynia in the NTG model of migraine-associated pain , there was a decrease in the number of branch points , combined neurite length , and interactions of neurons within the TNC , PAG , and somatosensory cortex . With this newly discovered phenomenon we sought to mitigate this decrease through inhibition of HDAC6 which we found to reciprocally restore neuronal complexity and inhibit allodynia . We found that the cytoarchitectural changes were not just induced by NTG but were also prominent following CSD . Reduction in neuronal complexity was also observed in this model of migraine aura , and again HDAC6 inhibition restored neuronal plasticity and decreased the number of CSD events . The latter effect is a hallmark of migraine preventive drugs . Furthermore , we found that a migraine specific treatment , CGRP receptor inhibition , also restored cytoarchitectural changes . Together our results demonstrate a novel mechanism of chronic migraine and reveal HDAC6 as a novel therapeutic target for this disorder ( Figure 7I ) . We used the NTG model in this study , as it is a well-validated model of migraine ( Demartini et al . , 2019 ) . NTG is a known human migraine trigger and is used as a human experimental model of migraine ( Schytz et al . , 2010 ) . Similar to humans , NTG produces a delayed allodynia in mice ( Bates et al . , 2010 ) , as well as photophobia and altered meningeal blood flow ( Markovics et al . , 2012; Greco et al . , 2011 ) . Chronic intermittent administration of NTG is used to model chronic migraine ( Pradhan et al . , 2014a; Farajdokht et al . , 2018; Long et al . , 2018; Christensen et al . , 2019; Zhang et al . , 2020 ) . Compared to humans , much higher doses of NTG are required to produce allodynia . However , NTG-induced hypersensitivity in mice is inhibited by migraine-specific medications , such as sumatriptan ( Bates et al . , 2010; Pradhan et al . , 2014a; Pradhan et al . , 2014b ) and CGRP targeting drugs ( Christensen et al . , 2019 ) , as well as the migraine preventives propranolol and topiramate ( Tipton et al . , 2016; Greco et al . , 2018 ) . Further , mice with human migraine gene mutations are more sensitive to NTG ( Brennan et al . , 2013 ) . Systemic administration of NTG also causes cellular activation throughout nociceptive pathways including in the TNC and brainstem ( Tassorelli and Joseph , 1995a; Tassorelli and Joseph , 1995b; Ramachandran et al . , 2012; Greco et al . , 2018 ) . Correspondingly , we also observed changes in neuronal complexity in the TNC , as well as in the PAG and somatosensory cortex , regions heavily involved in pain processing . Alterations in these regions could contribute to allodynia or interictal sensitivity observed in chronic migraine patients . Previous studies have shown an increase in TNC activity in headache models ( Oshinsky and Luo , 2006; Akerman et al . , 2013 ) . While our data show an overall decrease in complexity within these neurons , they do not necessarily contradict these previous findings . For example , decreased neuronal flexibility could encourage the strengthening of excitatory synapses and/or prevent the formation of inhibitory synapses , as the cell is in a more fixed state . Within the TNC , there are both inhibitory and excitatory neuronal populations; and future studies will determine which populations are altered in migraine models and how these changes directly affect neuronal activity . Alterations in neuronal complexity in response to NTG appeared to be limited to brain regions involved in pain processing . We did not observe any alterations in the nucleus accumbens which is commonly associated with reward and motivation . RNA-Seq and proteomic experiments from our lab have also revealed that the nucleus accumbens has very different responses to chronic NTG relative to parts of the trigeminovascular system ( Jeong et al . , 2018; Krishna et al . , 2019 ) . We also observed no change in neuronal cytoarchitecture in the lumbar spinal cord , a region largely involved in peripheral but not cephalic pain processing . Importantly , we found no cytoarchitectural alterations in the cervical spinal cord , a region that is also involved in head/neck pain processing . These data suggest that decreased neuronal complexity in response to migraine states may be limited to central sites that regulate headache and pain processing . We also observed decreased neuronal complexity in CSD , a migraine model mechanistically distinct from NTG . CSD is thought to underlie migraine aura and reflects changes in cortical excitability associated with the migraine brain state ( Charles and Baca , 2013; Brennan and Pietrobon , 2018 ) . Previous studies also support the idea of cytoarchitectural alterations accompanying spreading depression/depolarization events and focused mainly on dendritic morphology . Neuronal swelling ( Takano et al . , 2007 ) and dendritic beading ( Steffensen et al . , 2015 ) were observed following spreading depression events . CSD also resulted in alterations in dendritic structure ( Takano et al . , 2007; Eikermann-Haerter et al . , 2015 ) and volumetric changes ( Takano et al . , 2007 ) . Further , Steffensen et al . showed decreased microtubule presence in dendrites following spreading depression in hippocampal slices , again implying alterations in cytoarchitectural dynamics ( Steffensen et al . , 2015 ) . Microtubules have been shown to disassemble in response to increased intracellular calcium ( Schliwa et al . , 1981 ) ; and the increased calcium influx induced by spreading depolarization ( Basarsky et al . , 1998 ) may facilitate this breakdown . Tubulin acetylation is associated with increased flexibility and stability of microtubules ( Xu et al . , 2017 ) . One way in which HDAC6 inhibitors could attenuate CSD is through increased tubulin acetylation , thus counteracting microtubule disassembly produced by CSD events . In addition , CSD waves pass through the neuron in phases , from apical dendrites , to somatodendritic sites , and finally to proximal dendrites ( Pietrobon and Moskowitz , 2014 ) . Along with preventing the dendritic alterations that occur in response to calcium influx , it is possible that HDAC6 inhibition could redistribute or disturb the phasic movement of CSD events , which may decrease and/or elongate CSD events . Furthermore , multiple reports indicate that CSD can activate the trigeminovascular complex , and evoke cephalic allodynia in rodents ( Bolay et al . , 2002; Fioravanti et al . , 2011; Zhang et al . , 2012; Noseda and Burstein , 2013; Melo-Carrillo et al . , 2017; Filiz et al . , 2019 ) . We also observed decreased neurite branching in the TNC following CSD , further linking CSD to head pain processing . Combined , these data suggest that CSD has an impact on neuronal morphology that contributes to migraine pathophysiology . Proper acetylation of microtubules is necessary for a variety of cellular functions including appropriate neurite branching ( Gallo , 2011 ) , cell response to injury ( Gallo , 2011 ) , mitochondrial movement ( Braun et al . , 2011; Jin et al . , 2017 ) , anchoring of kinesin for microtubule mediated transport ( Gibbs et al . , 2015 ) and regulation of synaptic G protein signaling ( Schappi et al . , 2014; Singh et al . , 2018; Singh et al . , 2020 ) . Disruption of this process can have significant physiological effects . Knockout of the α-tubulin acetylating enzyme , α-TAT1 , in peripheral sensory neurons results in profound deficits in touch ( Morley et al . , 2016 ) . Further , Charcot-Marie-Tooth ( CMT ) disease is a hereditary axonopathy that affects peripheral nerves resulting in damage to both sensory and motor function . Mouse models of CMT reveal deficits in mitochondrial transport in the dorsal root ganglia ( DRG ) due to reduced α-tubulin acetylation , and HDAC6 inhibition ameliorate CMT-associated symptoms ( d'Ydewalle et al . , 2011; Benoy et al . , 2018 ) . We observed that ACY-738 broadly increased neuronal complexity , including in vehicle and sham controls . However , we did not see any alterations in mechanical thresholds or CSD events in response to this upregulation in these control groups . Furthermore , this effect of ACY-738 appeared to be short lasting and was no longer present at the 24 hr timepoint . Other groups using ACY-738 or other HDAC6 inhibitors also did not observe general disruption in mechanical or temperature sensitivity with HDAC6 inhibition ( Krukowski et al . , 2017; Van Helleputte et al . , 2018; Ma et al . , 2019; Sakloth et al . , 2020 ) . Additionally , constitutive knockout of HDAC6 produces viable offspring with few phenotypic changes ( Zhang et al . , 2008 ) . These findings suggest that HDAC6 inhibition does not appear to generally cause a loss of sensation , but in a migraine state , where mechanical responses are decreased , they can have anti-allodynic effects . While these results are the first of their kind to demonstrate cytoarchitectural changes in models of chronic migraine , alterations in neuronal plasticity have been described previously in models of neuropathic pain . A mouse model of chronic constriction injury of the sciatic nerve reduced neurite length in GABA neurons within lamina II of the spinal cord ( Zhang et al . , 2018 ) . Another group observed that following spared nerve injury , there were decreases in the number of branches and neurite length of hippocampal neurons but increases in spinal dorsal horn neurons ( Liu et al . , 2017 ) . These studies , along with our results , suggest that adaptations in response to chronic pain can culminate in alteration of neuronal cytoarchitecture within the central nervous system . HDAC6 inhibitors have been studied in other models of pain . They were shown to effectively reduce chemotherapy-induced allodynia following treatment with vincristine ( Van Helleputte et al . , 2018 ) or cisplatin ( Krukowski et al . , 2017 ) in mice . In addition , both groups found that chemotherapy blunted mitochondrial transport in sensory neurons , an effect that was restored by HDAC6 inhibition . Another study also found that HDAC6 inhibitors were effective in models of inflammatory and neuropathic pain ( Sakloth et al . , 2020 ) . Together , these studies highlight the importance of cytoarchitectural dynamics in relation to pain sensation and the ability of HDAC6 inhibition to promote relief from allodynia/hyperalgesia . We chose to focus on the role of HDAC6 in tubulin acetylation and microtubule dynamics , as we observed changes in neuronal complexity in migraine models . However , HDAC6 also regulates Hsp90 and cortactin ( Valenzuela-Fernández et al . , 2008 ) . HDAC6 deacetylates Hsp90 , which plays an important role in glucocorticoid receptor maturation and adaptation to stress ( Kovacs et al . , 2005 ) . A previous study showed that in social defeat stress HDAC6 knockout or inhibition decreased Hsp90-glucocorticoid receptor interaction and subsequent glucocorticoid signaling , thus encouraging resilience ( Espallergues et al . , 2012 ) . In line with these findings , HDAC6 inhibitors also show antidepressant-like effects ( Covington et al . , 2009; Jochems et al . , 2014 ) , and membrane-associated acetylated tubulin is decreased in humans with depression ( Singh et al . , 2020 ) . Further , HDAC6 also directly deacetylates cortactin , a protein that regulates actin-dependent cell motility ( Zhang et al . , 2007 ) . Future studies will explore the contribution of these other mechanisms by which HDAC6 may impact neuronal complexity in migraine . We investigated whether a current migraine treatment strategy , CGRP receptor inhibition , could ameliorate the cytoarchitectural changes induced by chronic migraine-associated pain . We found a good correlation between the anti-allodynic effects of olcegepant and its ability to restore neuronal complexity in the chronic NTG model . In contrast to ACY-738 , olcegepant had no effect on vehicle-treated mice and only recovered , but did not increase neuronal branching , length , or intersections in the NTG-treated group . Considering that CGRP receptors are not known to directly affect microtubule dynamics , these results suggest that there are multiple ways through which migraine therapies can affect neuronal plasticity . Previous studies have shown that activation of various Gα subunits , including αi , αo , αs , can inhibit microtubule assembly ( Roychowdhury et al . , 1999 ) . Therefore , it is possible that an increase in CGRP , which was found to be present following NTG ( Greco et al . , 2018; Moye et al . , 2021 ) , could result in altered microtubule assembly . Inhibition of the CGRP receptor could therefore reverse this process allowing for elaboration of microtubules . Olcegepant was previously shown to poorly cross the blood brain barrier; and it may result in cytoarchitectural changes in the central nervous system by blocking nociceptive signals from the periphery , resulting in upstream changes in the TNC and other central regions . Further , the effect of olcegepant along with the finding that NTG did not alter complexity in the spinal cord or nucleus accumbens , help to confirm that the changes in neuronal cytoarchitecture following NTG are associated with migraine mechanisms . This study suggests a possible mechanism in which recovered neuronal complexity is a marker of effective migraine medication . Our results reveal a novel cytoarchitectural mechanism that may underlie chronic migraine and imply that this disorder results from attenuation of neurite outgrowth and branching . Human imaging studies reveal decreased cortical thickness ( Magon et al . , 2019 ) , and gray matter reductions in the insula , anterior cingulate cortex , and amygdala of migraine patients ( Valfrè et al . , 2008 ) . Interestingly , a significant correlation was observed between gray matter reduction in anterior cingulate cortex and frequency of migraine attacks ( Valfrè et al . , 2008 ) . These structural changes could reflect decreased neuronal complexity in combination with other factors . We propose that strategies targeted toward pathways regulating neuronal cytoarchitecture may be an effective approach for the treatment of chronic migraine . Our results suggest that HDAC6 inhibitors may restore cellular adaptations induced by chronic disease states but may not otherwise affect healthy physiological function; and such compounds could contribute to the migraine therapeutic armamentarium . Experiments were performed on adult male and female C57BL/6J mice ( Jackson Laboratories , Bar Harbor , ME . USA ) weighing 20–30 g . Mice were group housed in a 12 h-12h light-dark cycle , where the lights were turned on at 07:00 and turned off at 19:00 . Food and water were available ad libitum . All experiments were conducted in a blinded fashion by 1–3 experimenters . Weight was recorded on each test day for all experiments . All experimental procedures were approved by the University of Illinois at Chicago Office of Animal Care and Institutional Biosafety Committee , in accordance with Association for Assessment and Accreditation of Laboratory Animal Care International ( AAALAC ) guidelines and the Animal Care Policies of the University of Illinois at Chicago . All results are reported according to Animal Research: reporting of In Vivo Experiments ( ARRIVE ) guidelines . No adverse effects were observed during these studies , and all animals were included in statistical analysis . Different groups of animals were used for each experiment . Mice were counter-balanced into groups following the first basal test for mechanical thresholds . Mice were tested in a behavior room , separate from the vivarium , with low light ( ~35–50 lux ) and low-noise conditions , between 09:00 and 16:00 . Mice were habituated to the testing racks for 2 days before the initial test day , and on each subsequent test days were habituated for 20 min before the first measurement . The cephalic region was tested throughout this study , except for the RN-73 experiment , where the hind paw was tested . For cephalic measures mice were tested in four oz paper cups . The periorbital region caudal to the eyes and near the midline was tested . For experiments testing peripheral mechanical responses , the intraplantar region of the hindpaw was assessed . Testing of mechanical thresholds to punctate mechanical stimuli was tested using the up-and-down method . The selected region of interest was stimulated using a series of manual von Frey hair filaments ( bending force ranging from 0 . 008 g to 2 g ) . A response of the head was defined as shaking , repeated pawing , or cowering away from the filament . In the hind paw , a response was lifting of the paw , shaking , or licking the paw after stimulation . The first filament used was 0 . 4 g . If there was no response a heavier filament ( up ) was used , and if there was a response a lighter filament ( down ) was tested . The up-down pattern persisted for four filaments after the first response . To decrease bias in testing , researchers were blinded to treatment groups at time of testing . While the same researcher performed both the mechanical threshold testing and injections these measures were recorded in different places and at separate time points . Nitroglycerin ( NTG ) was purchased at a concentration of 5 mg/ml , in 30% alcohol , 30% propylene glycol and water ( American Reagent , NY , USA ) . NTG was diluted on each test day in 0 . 9% saline to a concentration of 1 mg/ml for a dose of 10 mg/kg . Mice were administered NTG or vehicle every other day for 9 days . Animals used in cephalic experiments were tested on days 1 , 5 , and 9 . On test days a basal threshold was measured then animals were treated with either NTG or vehicle and then put back in the testing racks and subsequently tested 2 hr later for the post-treatment effect . The procedure for the cortical spreading depression ( CSD ) model is based on previously published work ( Chen and Ayata , 2017 ) that is commonly used to screen potential migraine preventives and further used in our own work ( Pradhan et al . , 2014b; Dripps et al . , 2020; Bertels et al . , 2021 ) . Mice were grouped into sham and CSD groups and then further subdivide into ACY-738 ( 50 mg/kg , IP ) or vehicle ( i . e . Sham-ACY , Sham-Veh , CSD-ACY , CSD-Veh ) . To make the thinned skull cortical window , mice were anesthetized with isoflurane ( induction 3–4%; maintenance 0 . 75% to 1 . 25%; in 67% N2 / 33% O2 ) and placed in a stereotaxic frame on a homoeothermic heating pad . Core temperature ( 37 . 0 ± 0 . 5℃ ) , non-peripheral oxygen saturation ( ∼ 99% ) , heart rate , and respiratory rate ( 80–120 bpm ) were continuously monitored ( PhysioSuite; Kent Scientific Instruments , Torrington , CT , USA ) . Mice were frequently tested for tail and hind paw reactivity to ensure that the anesthesia plane was maintained . To verify CSD events , optical intrinsic signal ( OIS ) imaging was primarily used and electrophysiological recordings were recorded as previously described ( Pradhan et al . , 2014b ) . Briefly , following anesthesia , the skin from the skull was detached and a rectangular region of ~2 . 5×3 . 3 mm2 ( ~0 . 5 mm from sagittal , and ~1 . 4 from coronal and lambdoid sutures ) of the right parietal bone was thinned to transparency with a dental drill ( Fine Science Tools , Inc , Foster City , CA , USA ) . Mineral oil application improved transparency of cortical surface parenchyma and vasculature for video recording . A green LED ( 530 nm ) illuminated the skull throughout the experiment ( 1-UP; LED Supply , Randolph , VT , USA ) . Cortical surface reflectance detected by OIS was collected with a lens ( HR Plan Apo 0 . 5 × WD 136 ) through a 515LP emission filter on a Nikon SMZ 1500 stereomicroscope ( Nikon Instruments , Melville , NY , USA ) . Images were acquired at 1–5 Hz using a high-sensitivity USB monochrome CCD camera ( CCE-B013-U; Mightex , Pleasanton , CA , USA ) with 4 . 65-micron square pixels and 1392 × 1040 pixel resolution . Lateral to the thinned window two burr holes were drilled around the midpoint of the rectangle . These burr holes were deeper than the previously drilled skull region such that the dura was exposed but not broken . To record local field potentials ( LFPs ) an electrode ( in a pulled glass pipette filled with saline ) was inserted into one burr hole and attached to an amplifier . A separate ground wire , placed underneath the skin caudal to the skull , grounded this set up and LFPs were recorded for an hour to ensure a stable baseline and recovery from any surgically induced CSDs . After an hour of stabilization , a second pulled glass pipette was filled with 1 M KCl and placed into the more rostral burr hole , avoiding contact with the brain or the surrounding skull . An initial flow of KCl was pushed to begin and then an even flow was held so that there was a constant small pool of KCl that filled the burr hole . Excess liquid was removed with tissue paper applied next to the burr hole . Regardless of grouping the CSD recording continued for 3600 s after the initial drip of KCl . Mice were euthanized by anesthetic overdose followed by decapitation . Golgi staining was performed according to the FD Rapid Golgi Stain kit ( FD Neurotechnologies ) . For NTG or Veh-treated mice , they underwent the chronic NTG model and on day 10 , 4 , 24 , and 48 hr after ACY-738 treatment or vehicle , mice were anesthetized with isoflurane , euthanized , brain/spinal cord was rapidly removed , and tissue was rinsed in ddH2O . Tissue was then placed in the impregnation solution that was an equal amount of solutions A and B that was prepared at least 24 hr in advance . After the first 24 hr the brain was placed in new impregnation solution and then stored for 1 week in the dark . The brains were then transferred to solution C , which was also replaced after the first 24 hr . After replacing solution C the brains were stored at room temperature for 72 hr more . Following solution C , brains were flash frozen in 2-methyl butane and cryostat cut at −20°C into 100 µm slices . The slices were mounted onto gelatin coated slides and secured by a drop of solution C placed onto each slice . These slides were then left to dry naturally in the dark . After processing , images were taken at ×20 magnification and a Z-stack was created based on different levels of focal plane . After the Z-stack was created the FIJI program Simple Neurite Tracer was used to trace the processes of the neuron . While many of the neurons had some overlap with other analyzed neurons , Z-stacks of varying focus levels allowed for clearer tracing . A sample gif file of a Z-stack from a traced neuron is included and demonstrates how the change in focus allow for better determination of branching from overlapping neurons ( Animation 1 ) . Furthermore , after tracing the neurons were analyzed using Simple Neurite Tracer ( Longair et al . , 2011 ) software to assess the number of branch points from each neuron , overall length of the neuron , and Sholl Analysis . Sholl Analysis was performed by placing a center ROI point at the center of the soma and producing consecutive circles every 20 pixels/0 . 377 µm , for the entire body of the neuron . Intersections were counted based on the number of times a neurite crossed each of these consecutive circles . These data were compiled per neuron and then brought into one Masterfile . Male and female mice were used for a majority of studies , and no significant differences were observed in any of the key findings based on sex . Throughout tracing all tracers were blinded to which group the images belonged to . For all brain regions analyzed , six to eight relatively isolated neurons were randomly chosen per mouse . The selected neurons were fully impregnated with Golgi stain and relatively complete . An atlas was used along with clear anatomical markers to ensure the neurons were being taken from their described region of interest . Neurons characterized for the trigeminal nucleus caudalis region were taken only from the outer lamina of caudal sections . Neurons analyzed for somatosensory cortex were all taken from layer IV of the primary somatosensory barrel cortex . To ensure a homogenous cell population , only pyramidal cells were selected . The most complex neurons were chosen for analysis in all regions . Previously , it was shown that dendritic complexity was directly correlated to soma size . To ensure that the NTG group where not just smaller in size we directly compared soma diameter of neurons in the NTG and Veh group . There was no significant difference in soma size between these two groups ( Veh 9 . 258 ± . 2515 and NTG 9 . 192 ± . 2782 , student run t-test p=0 . 8608 ) . Three individuals traced all cells . Interrater reliability was determined by having each tracer trace five neurons in their entirety . Pearson product correlations were accessed in three measures; number of branches , total dendritic length , and total intersection number through Sholl analysis and found to be 0 . 91 , 0 . 94 , and 0 . 95 , respectively . All tracings of neurons were re-examined by the primary tracer ( Z . B . ) to assure quality control . Original neuronal traces can also be viewed at NeuroMorpho . org ( http://neuromorpho . org/KeywordResult . jsp ? count=837&keywords=%22bertels%22 ) . All injections were administered at 10 ml/kg volume , intraperitoneally ( IP ) , unless otherwise indicated . ACY-738 was dissolved in a 5% DMSO saline solution , which was used as the vehicle control . RN-73 was dissolved in 10% DMSO , 10% Tween-80 , and 80% saline and was injected 1 mg/kg or 10 mg/kg , this mixture was also used as the vehicle control group . ASV-85 was dissolved in 15% DMSO , 15% Tween-80 , and then 70% saline , this mixture was also used as the vehicle control group . ASV-85 was injected at 1 mg/kg dose . TSA was dissolved in 20% DMSO solution in 80% 0 . 01M PBS and injected at a dose of 2 mg/kg , this same solution was used for the vehicle . Olcegepant was dissolved in saline solution and was injected at a 1 mg/kg dose . For the CSD experiments ACY-738 was injected 3 hr before starting the surgery so that it would reach its peak efficiency of 4 hr by the time the CSD event started . Total RNA was isolated from flash frozen brain punches using the RNeasy Plus Mini kit from Quiagen . RNA samples were reverse transcribed to single-stranded cDNA . cDNA transcription was used following the protocol from Superscript III ( Life Technologies ) and the TaqMan Gene Expression Assay system ( Applied Biosystems ) . Glyceraldehyde-3-phosphate dehydrogenase ( GAPDH , Hs02758991_g1 ) was used as a housekeeping gene . The threshold cycle ( CT ) of each target product was determined and CT values between HDAC6 transcripts and housekeeping genes were calculated ( ΔCT ) . The fold change ( 2- ΔΔCT ) for each was calculated relative to the median ΔCT from the saline control animals . Mice were anesthetized with Somnasol ( 100 µl/mouse; 390 mg/mL pentobarbital sodium; Henry Schein ) and perfused intracardially with 15 ml of ice-cold phosphate-buffered saline ( 0 . 1 M PBS , pH 7 . 2 ) and subsequently 50 mL of ice-cold 4% paraformaldehyde ( PFA ) in 0 . 1M PBS ( pH 7 . 4 ) . Whole brain and trigeminal ganglia ( TG ) were harvested and overnight left to post-fix in 4% PFA/0 . 1M PBS at 4°C . Brain and TG were then cryoprotected in 30% sucrose in 0 . 1M PBS until they sunk . Brains were then flash frozen using 2-methyl butane over dry ice . Coronal sections of the trigeminal nucleus caudalis ( TNC ) and the somatosensory cortex were sliced on a cryostat at 20 µM and TG at 16 µM and immediately mounted onto slides . Slides were washed with PBST , then incubated with a blocking solution containing 5% normal donkey serum with PBST for 1 hr at room temperature . Slides were then incubated overnight at RT with the primary rabbit anti-HDAC6 antibody ( 1:500 , courtesy of Tso-Pang Yao at Duke University ) diluted in 1% NDSDT . Slides were subsequently washed with 1% NDST and then the secondary antibody was added for 2 hr at room temperature ( donkey anti rabbit IgG , 1:2000 ) . Slides were washed with 0 . 1 M phosphate buffer , and cover slipped with Mowiol-DAPI mounting medium . Images were taken by in a blinded manner using the EVOS FL Auto Cell Imaging system , using a ×40 objective . Samples were taken from chronically treated NTG or Vehicle mice , which received an injection of ACY-738 or Vehicle on day 10 . Samples were collected 4 hr post-ACY/VEH . TG , TNC , and SCtx was analyzed using traditional western blot analysis while Nac and spinal cord samples were analyzed at a later time using the ProteinSimple Wes . Spinal cord samples were a combination of cervical and lumbar spinal cord sections . Protein concentrations were assessed using a Nanodrop 2000c spectrophotometer and equal quantities were loaded onto each Stain-Free acrylamide gel for SDS-PAGE ( Bio-Rad , Hercules , CA , USA ) . The gels were subsequently transferred to nitrocellulose membranes ( Bio-Rad , Hercules , CA USA ) for western blotting . The membranes were blocked with 5% non-fat dry milk diluted in TBS-T ( 10 mM Tris-HCl , 159 mM NaCl , and 0 . 1% Tween 20 , pH 7 . 4 ) for 1 hr . Following the blocking step , membranes were washed with Tris-buffered saline/Tween 20 and then incubated with an anti-acetyl-α-tubulin antibody ( Lysine-40 ) ( Sigma Clone 6-11B1 ) , α-tubulin ( Sigma ) , overnight at 4°C . Membranes were washed with TBS-T and incubated with a secondary antibody [HRP-linked anti-mouse antibody IgG F ( ab′ ) two or HRP-linked anti-rabbit antibody IgG F ( ab′ ) 2] ( Jackson ImmunoResearch , West Grove , PA , USA , catalog #115-036-072 for mouse , and catalog #111-036-047 for rabbit , ) for 1 hr at room temperature , washed , and developed using ECL Luminata Forte chemiluminescent reagent ( Millipore , Billerica , MA , USA ) . Blots were imaged using a Chemidoc computerized densitometer ( Bio-Rad , Hercules , CA , USA ) and quantified by ImageLab 3 . 0 software ( Bio-Rad , Hercules , CA , USA ) . In all experiments , the original gels were visualized using BioRad stainfree technology to verify protein loading . For the spinal cord and nucleus accumbens , samples were prepared and run on the ProteinSimple Instruments Wes System according to the manufacturer’s instructions . Images for these samples were also measured and visualized using the same system . Sample size was calculated by power analysis and previous experience . Since we investigated changes at the cellular level , an individual neuron represented a single sample ( Espallergues et al . , 2012; Moonat et al . , 2013 ) . Data analysis was performed using GraphPad Prism version 8 . 00 ( GraphPad , San Diego , CA ) . The level of significance ( α ) for all tests was set to 0 . 05 . Post hoc analysis was conducted using Holm-Sidak post hoc test to correct for multiple comparisons . Post hoc analysis was only performed when F values achieved p<0 . 05 . All values in the text are reported as mean ± SEM . Detailed statistical analysis can be found in Supplementary file 2 .	Migraines are a common brain disorder that affects 14% of the world’s population . For many people the main symptom of a migraine is a painful headache , often on one side of the head . Other symptoms include increased sensitivity to light or sound , disturbed vision , and feeling sick . These sensory disturbances are called aura and they often occur before the headache begins . One particularly debilitating subset of migraines are chronic migraines , in which patients experience more than 15 headache days per month . Migraine therapies are often only partially effective or poorly tolerated , making it important to develop new drugs for this condition , but unfortunately , little is known about the molecular causes of migraines . To bridge this gap , Bertels et al . used two different approaches to cause migraine-like symptoms in mice . One approach consisted on giving mice nitroglycerin , which dilates blood vessels , produces hypersensitivity to touch , and causes photophobia in both humans and mice . In the second approach , mice underwent surgery and potassium chloride was applied onto the dura , a thick membrane that surrounds the brain . This produces cortical spreading depression , an event that is linked to migraine auras and involves a wave of electric changes in brain cells that slowly propagates across the brain , silencing brain electrical activity for several minutes . Using these approaches , Bertels et al . studied whether causing chronic migraine-like symptoms in mice is associated with changes in the structures of neurons , focusing on the effects of migraines on microtubules . Microtubules are cylindrical protein structures formed by the assembly of smaller protein units . In most cells , microtubules assemble and disassemble depending on what the cell needs . Neurons need stable microtubules to establish connections with other neurons . The experiments showed that provoking chronic migraines in mice led to a reduction in the numbers of connections between different neurons . Additionally , Bertels et al . found that inhibiting HDAC6 ( a protein that destabilizes microtubules ) reverses the structural changes in neurons caused by migraines and decreases migraine symptoms . The same effects are seen when a known migraine treatment strategy , known as CGRP receptor blockade , is applied . These results suggest that chronic migraines may involve decreased neural complexity , and that the restoration of this complexity by HDAC6 inhibitors could be a potential therapeutic strategy for migraine .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2021	Neuronal complexity is attenuated in preclinical models of migraine and restored by HDAC6 inhibition
Changes in chromosome number impair fitness by disrupting the balance of gene expression . Here we analyze mechanisms to compensate for changes in gene dose that accompanied the evolution of sex chromosomes from autosomes . Using single-copy transgenes integrated throughout the Caenorhabditis elegans genome , we show that expression of all X-linked transgenes is balanced between XX hermaphrodites and XO males . However , proximity of a dosage compensation complex ( DCC ) binding site ( rex site ) is neither necessary to repress X-linked transgenes nor sufficient to repress transgenes on autosomes . Thus , X is broadly permissive for dosage compensation , and the DCC acts via a chromosome-wide mechanism to balance transcription between sexes . In contrast , no analogous X-chromosome-wide mechanism balances transcription between X and autosomes: expression of compensated hermaphrodite X-linked transgenes is half that of autosomal transgenes . Furthermore , our results argue against an X-chromosome dosage compensation model contingent upon rex-directed positioning of X relative to the nuclear periphery . Abnormalities in chromosome number ( aneuploidy ) have the potential to disrupt the balance of gene expression and thereby decrease organismal fitness and viability ( Siegel and Amon , 2012 ) . Aneuploidy occurs in most solid tumors and is a major cause of severe developmental defects and spontaneous abortions ( Siegel and Amon , 2012 ) . Unlike pathological imbalances in chromosome dose , the disparity in X-chromosome dose between 1X males and 2X females caused by sex-determination mechanisms has evolved to be well tolerated ( Charlesworth , 1996 ) . How this tolerance came about remains poorly understood . Of particular relevance is whether chromosome-wide regulatory mechanisms evolved to modulate the relationship between X-chromosome gene dose and gene product . Here we dissect the function and significance of gene regulatory strategies in the nematode C . elegans to achieve two goals: ( 1 ) elucidate mechanisms by which the X-chromosome dosage compensation process balances X expression between the sexes; ( 2 ) determine whether an X-chromosome-wide regulatory mechanism balances gene expression between X chromosomes and autosomes to facilitate X-chromosome evolution . The need for X-chromosome-wide control of gene expression is illustrated by a description of sex-chromosome evolution ( Figure 1A ) . For humans , although the X and Y sex chromosomes are genetically distinct , both originated from a single pair of homologous autosomes ( Charlesworth , 1996 ) . The differentiation of an autosome pair into two different sex chromosomes was proposed to begin when one homolog acquired a male-determining gene , thereby converting the homologs into a proto-Y and a proto-X ( Charlesworth and Charlesworth , 2000; Bachtrog , 2013 ) . As recombination ceased between the proto-Y and proto-X , the proto-Y could accumulate other male beneficial alleles , but the recombination isolation would cause it to degenerate into a gene-poor Y chromosome that did little more than specify male sexual fate . This process would result in males with one Y chromosome and one X chromosome and females with two X chromosomes . In the nematode C . elegans , sex chromosomes also likely arose from a pair of ancestral autosomes through a similar mechanism , but with Y-chromosome degradation progressing until the Y was lost completely ( Charlesworth , 1996 ) . The demise of Y as a male-determining chromosome was enabled by the emergence of a different sex-determination mechanism , one that utilized the ratio of X chromosomes to sets of autosomes ( ploidy ) to specify male ( 1X:2A ) vs . hermaphrodite ( 2X:2A ) sexual fates ( Nigon , 1951 ) . 10 . 7554/eLife . 17365 . 003Figure 1 . Sex chromosome evolution and its impact on gene expression . ( A ) Sex chromosome evolution . In mammals , the X and Y sex chromosomes were derived from a single pair of homologous autosomes referred to as the ancestral autosomes ( green ) . Before the evolution of sex chromosomes , genes represented by gene A ( black ) were present on both ancestral autosomes ( AA ) . During sex chromosome formation , one autosome acquired a dominant male-determining gene ( * ) , thereby converting an ordinary autosome pair into a proto-X ( pX ) chromosome and a male sex-determining proto-Y ( pY ) chromosome ( step 1 ) . As recombination ceased between the proto-Y and proto-X , and the proto-Y accumulated other male beneficial alleles , the proto-Y degenerated into the present-day gene-poor Y chromosome that specified male fate ( step 2 ) . Loss of genes from Y ( e . g . gene A ) caused genes from the ancestral autosome to be present in only one copy in males instead of two copies on the ancestral autosomal pair . While most mammalian sex chromosomes progressed only through steps 1 and 2 , the nematode sex chromosomes were proposed to have evolved by a similar route but then to have undergone an additional step in which Y chromosome degradation progressed until the Y was lost completely ( step 3 ) , giving rise to XX hermaphrodites and XO males . Demise of Y was enabled by the emergence of a sex-determining mechanism that utilizes the ratio of X chromosomes to sets of autosomes ( X:A signal ) to determine sex rather than the dominant masculinizing gene that initiated sex-chromosome evolution . ( B , C , D ) Predictions for X-chromosome gene expression with and without Ohno's upregulation mechanism ( B ) Prediction for X-linked gene expression when the dosage compensation mechanism increases expression of X in males , a case not requiring Ohno's hypothesis . If the dose-sensitive gene A were expressed at a level of 1 when present in two copies on the ancestral autosomes , it would be expressed at a level of 0 . 5 in present-day males with only one copy on the single male X , and at a level of 1 . 0 in present-day females , which carry one copy on both X chromosomes . If gene A were haploinsufficient , its reduced expression could have deleterious consequences for the male . To compensate for reduced gene expression in males , a dosage compensation mechanism arose to balance X expression between the sexes . Drosophila melanogaster increases X-linked gene expression two-fold in males , thereby balancing the level of gene expression with that in present-day females and that in the ancestral species prior to sex-chromosome evolution . ( C ) Prediction for X-linked gene expression when the dosage compensation mechanism reduces X gene expression in females without an accompanying upregulation mechanism proposed by Ohno . X-chromosome dosage compensation in mammals and C . elegans occurs by mechanisms different from that of Drosophila , even though sex chromosomes may have evolved by a similar route . These species compensate for the imbalance in X-chromosome dose between the sexes by reducing X-linked gene expression in females/hermaphrodites by half , causing both sexes to express gene A at half the level of the ancestral species prior to the evolution of sex chromosomes . ( D ) Prediction for X-linked gene expression when Ohno's mechanism of upregulation operates and the dosage compensation mechanism reduces X gene expression in females . Recognizing that reducing X-chromosome gene expression in females as a mechanism of dosage compensation between sexes might create a deleterious reduction in X-chromosome products for both sexes , Susumo Ohno proposed a two-step mechanism for the regulation of X gene expression . After the degeneration of Y began but before the evolution of dosage compensation , a mechanism would arise to increase X-chromosome gene expression two-fold in both sexes ( step 1 ) . This upregulation of X expression would make expression from the male X equal to that of the ancestral autosomes but would cause a two-fold overexpression of X-linked genes in females relative to the ancestral autosomes . The overexpression in females would then be offset by an X-chromosome dosage compensation process that reduced X expression in females , thereby balancing X expression between males and females , as well as balancing expression between female X chromosomes and the ancestral autosomes ( step 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 00310 . 7554/eLife . 17365 . 004Figure 1—figure supplement 1 . Average gene expression levels vary across autosomes . ( A ) Boxplots depict normalized RNA-seq counts per kb for each autosome ( black ) , for X ( purple ) , and for the aggregate level of gene expression on all autosomes ( red ) as measured from RNA made from mixed-stage embryos . Gene expression levels were normalized between experiments and also by gene length in kb . The analyses included all genes with an expression value greater than 0 . The median and mean expression levels are listed above each boxplot , and the number of genes included in each analysis is indicated below the graph . Outliers are not shown . The boxplots show that the mean level of gene expression varies widely across the five autosomes ( by 2 . 2-fold for all expressed genes ) . ( B ) Boxplots as in part A but only with genes having expression levels in the top 90% . The mean gene expression varies by 1 . 9-fold across the autosomes . ( C ) Global run-on sequencing ( GRO-seq ) used to quantify gene expression levels in mixed-stage embryos is reported as chromosome-wide boxplots , as in previous figures . Gene expression values were normalized to the number of million mapped reads and to the gene length in kb . As seen with RNA-seq measurements , the GRO-seq measurements show that the average level of gene expression varies widely ( by two-fold ) across the autosomes . The X:A expression ratio is 0 . 8 ( D ) Metagene analysis of gene expression on individual chromosomes as assayed by GRO-seq . Scaling of genic regions was as follows: the gene bodies were scaled to 1 kb , but neither the 5' ends ( −1 kb to +500 bp of the transcription start site [arrow] ) nor 3' ends ( 500 bp upstream to 1 kb downstream of 3' end ) were scaled . This analysis shows that the variability in average expression among the autosomes is evident as a difference in the distribution of engaged polymerases across the gene bodies and regulatory regions of all genes on each chromosome . ( E ) Boxplots depict the fold change in gene expression between glp-4 ( q224 ) germlineless L4 XX animals and control L4 XX animals . Shown are the modENCODE data for all expressed genes on each autosome ( black ) , the X chromosome ( X , purple ) , and the 276 genes that are X-linked in C . elegans ( X translocation , purple ) but autosome-linked in P . pacificus . These box plots show that silencing of germline expression is preferential to X chromosomes . Also , the 276 genes that are X-linked in C . elegans but not P . pacificus undergo germline silencing in C . elegans , meaning that analysis of gene expression in young adults is not valid for comparing gene expression between C . elegans and P . pacificus , as done in Albritton et al . ( 2014 ) . ( F ) Boxplots depict the fold change in gene expression in sdc-2 ( y93; RNAi ) embryos relative to control embryos for all genes on each autosome ( black ) or the X chromosome ( X , purple ) . A boxplot also depicts the fold change in gene expression observed for X-linked reporter transgenes in sdc-2 ( RNAi ) L1s vs . control L1s ( X transgenes , purple ) . The single outlier for the X reporter transgenes is shown . The X-linked reporters all have elevated expression in dosage-compensation-defective mutants , as do endogenous genes on X vs . those on autosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 004 The evolution of a male sex with only one X chromosome had the potential to impair male fitness . In the absence of any compensating mechanisms , genes present in one copy on the single male X would express half the level of gene products as genes present in two copies on the female X chromosomes and two copies on the ancestral autosomes . Reduced expression of dose-sensitive genes on X would likely decrease male viability ( Figure 1A ) . For the XX/XY species Drosophila melanogaster , potential complications caused by males having one X chromosome were averted by the co-evolution of a dosage compensation mechanism that operates by doubling gene expression from the single male X chromosome . Elevating X expression in males balanced gene expression with that from the two female X chromosomes , while maintaining similar expression between the single X and the two ancestral autosomes ( Figure 1B ) ( Lucchesi and Kuroda , 2015 ) . In contrast , other animals including placental mammals and the nematode C . elegans developed mechanisms of dosage compensation that equalized X-chromosome expression between the sexes by reducing X-linked gene expression by half in XX animals ( Figure 1C ) ( Meyer , 2010; Galupa and Heard , 2015 ) . Although decreasing X-chromosome gene expression in XX animals balances X expression between the sexes , it causes females to have the same problem as males: insufficient levels of X-chromosome products relative to those of ancestral autosomes ( Figure 1C ) . For mammals , Susumo Ohno proposed that loss of genes from the degenerating Y chromosome was compensated via two sequential evolutionary steps ( Figure 1D ) ( Ohno , 1967 ) . In the first step , mechanisms would arise to increase expression of each X-linked gene in both sexes by approximately two fold . While this increase in gene expression would offset the X-chromosome dose deficiency in males , it would cause overexpression of X-linked genes in females . In a second step , overexpression of genes on female X chromosomes would be offset by inactivating one of the two X chromosomes . Controversy has surrounded the question of whether mammals and other organisms such as C . elegans , which reduce female X-chromosome gene expression to equalize X expression between the sexes , do indeed employ a separate compensating mechanism to increase gene expression on X chromosomes of both sexes ( Xiong et al . , 2010; Deng et al . , 2011; Kruesi et al . , 2013; Albritton et al . , 2014 ) . The controversy in mammals caused Ohno's hypothesis to be re-evaluated by a multi-species approach ( Julien et al . , 2012 ) . The central prediction of Ohno's two-step hypothesis is that the single active X chromosome of males and females will be expressed at the same level as the combined ancestral autosomes . Although expression of ancestral autosomes cannot be measured directly for mammals , it can be estimated , because mammalian sex-chromosome evolution occurred after the divergence of birds and mammals . Thus , expression of orthologous autosomal genes in chickens serves as an estimate for expression of genes on the mammalian proto-X . Comparison between the extant mammalian X chromosome and the orthologous chicken autosome failed to reveal evidence for X-chromosome-wide upregulation in placental mammals ( Julien et al . , 2012 ) . In these species , genes on the single active X chromosome in males and females are expressed , on average , at half the level of the orthologous pair of autosomes , contrary to Ohno's hypothesis . Although the experimental approach failed to identify a chromosome-wide transcriptional mechanism of X upregulation , it left open the possibility that regulatory mechanisms might have arisen on a gene-by-gene basis to compensate for low activity of critical X-linked genes caused by chromosome-wide reduction of X expression . In contrast , evidence in favor of Ohno's hypothesis exists in marsupials , suggesting that X-chromosome upregulation may have accompanied sex-chromosome evolution in some lineages but not others ( Julien et al . , 2012 ) . For C . elegans , tests of Ohno's upregulation hypothesis have faced two major obstacles . First , limited information about the orthology of nematode genes relative to other species makes it premature to estimate the level of gene expression for the C . elegans proto-X chromosome from the expression level of autosomal orthologs in other species . In the absence of information about orthology , the assumption was made in some studies that the average overall expression of all genes on extant autosomes would serve as a proxy for expression of the proto-X chromosome ( Deng et al . , 2011; Kruesi et al . , 2013 ) . However , this assumption is undermined by the unexpected observation we report here that average gene expression varies widely ( 1 . 9 fold ) among the five different autosomes ( Figure 1—figure supplement 1A–D ) . Hence the previous finding that X expression is equivalent to the average level of autosomal expression does not confirm Ohno's hypothesis . Second , X-chromosome gene expression undergoes transcriptional silencing in germ cells of XX and XO animals ( Figure 1—figure supplement 1E ) ( Reinke et al . , 2000; Kelly et al . , 2002; Deng et al . , 2011; Gaydos et al . , 2012 ) , which comprise 68% of all adult cells ( Crittenden et al . , 2006; Morgan et al . , 2010 ) , causing tests of Ohno's hypothesis that quantify adult gene expression ( Xiong et al . , 2010 ) to underestimate X expression by about 50% . Thus , no compelling evidence supports or refutes Ohno's hypothesis for C . elegans . Here in a different exploration of Ohno's hypothesis , we asked whether a chromosome-wide mechanism operates in C . elegans to upregulate X-linked gene expression in both sexes . Our approach quantified gene expression specifically in somatic cells and did not rely on untested assumptions about expression levels of ancestral autosomes . We quantified expression in L1 larvae of the same transgenes integrated in single copy on either the X chromosome or autosomes . The L1 developmental stage occurs prior to the onset of germline proliferation , thereby preventing germline silencing from interfering with our quantification of X-chromosome expression . Moreover , monitoring expression of the same gene in the same species while varying only its location within the genome enabled a direct comparison of X and autosomal expression levels that tests one attractive molecular mechanism ( a chromosome-wide mechanism ) for upregulating X-chromosome transcription to balance gene expression between X and autosomes . Our transgene approach also enabled us to determine whether the dosage compensation process , which equalizes X expression between the sexes , acts chromosome-wide to control gene expression all along X or instead acts locally on a gene-by-gene basis . In C . elegans , as in mammals , not all genes on X are dosage compensated ( Carrel and Willard , 2005; Jans et al . , 2009; Kruesi et al . , 2013 ) , and the factors that determine whether a gene becomes dosage compensated or escapes from dosage compensation are not known . In particular , it has been difficult to tease apart whether a gene's local DNA sequence , its proximity to a binding site for the dosage compensation machinery , its position on the chromosome , its location within the nucleus , or a combination of such factors influences the dosage compensation process . By monitoring the expression of identical transgenes integrated at various locations along the X chromosome and autosomes , with and without a co-integrated binding site for the dosage compensation machinery , we eliminate the contribution of gene-specific differences in DNA sequence on gene expression and assess the role of chromosome location and proximity to a binding site on the regulation of X-chromosome gene expression , thereby differentiating a global chromosome-wide process from a local gene-by-gene process . Balancing X-chromosome gene expression between the sexes is achieved in C . elegans by a dosage compensation complex ( DCC ) that is homologous to condensin ( Csankovszki et al . , 2009; Mets and Meyer , 2009; Meyer , 2010 ) , a conserved protein complex that controls the compaction and resolution of all mitotic and meiotic chromosomes prior to their segregation ( Wood et al . , 2010; Hirano , 2016 ) . The DCC is recruited to both X chromosomes of hermaphrodites by cis-acting regulatory elements distributed throughout X called recruitment elements on X ( rex sites ) . These sites include DNA motifs that are highly enriched on X chromosomes and important for DCC binding ( Csankovszki et al . , 2004; McDonel et al . , 2006; Ercan et al . , 2007; Jans et al . , 2009; Pferdehirt et al . , 2011 ) . Once bound to X , the DCC remodels the topology of X , while reducing the expression from both hermaphrodite X chromosomes by half to balance gene expression between the sexes ( Crane et al . , 2015 ) . We first show here that all transgenes integrated onto C . elegans X chromosomes are dosage compensated , regardless of their position on X and hence their proximity to an endogenous rex site . Thus , the X chromosome is broadly permissive for the transcriptional repression that achieves dosage compensation . Furthermore , integration of the same transgenes onto autosomes , either with or without an adjacent DCC-bound rex site , failed to elicit DCC-mediated repression in hermaphrodites . Thus , DCC binding to a nearby rex site is not sufficient to trigger dosage compensation of a gene , nor is it necessary . These data reinforce a model of dosage compensation in which the DCC acts through multiple rex sites to induce chromosome-wide changes in X structure that influence expression of endogenous and engineered genes ( Crane et al . , 2015 ) . While our transgene approach demonstrates a robust chromosome-wide mechanism to balance X gene expression between the sexes , it provides strong evidence against an analogous , chromosome-wide mechanism that would fulfill Ohno's hypothesis for balancing gene expression between X chromosomes and autosomes . We show that in dosage-compensated ( i . e . down regulated ) XX animals , the per-copy expression of X-linked transgenes is half , not equivalent to , the per-copy expression of their counterparts on autosomes . In addition , the per-copy expression of hemizygous X-linked transgenes in XO animals is equivalent to , not double , the per-copy expression of their autosomal counterparts . Both findings are inconsistent with a chromosome-wide mechanism of upregulation . Our results suggest that if upregulation did occur to compensate for gradual loss of genes during X-chromosome evolution , it proceeded by the emergence of diverse gene-specific mechanisms that would compensate for their reduced dose . Finally , our analysis of X-chromosome regulation , combined with chromosome localization studies , allowed us to evaluate a recent , speculative model of X-chromosome dosage compensation , which proposes that rex sites target X chromosomes to the nuclear periphery in males to increase gene expression , while DCC binding to rex sites in hermaphrodites relocates X to the interior , thereby reducing gene expression to achieve dosage compensation ( Sharma et al . , 2014; Sharma and Meister , 2015 ) . Results presented here provide strong evidence against this model of dosage compensation . Together , our studies offer key insights into mechanisms by which abnormalities in chromosome number can evolve to be well tolerated . To analyze mechanisms that regulate gene expression across X , we examined the expression of 28 reporter genes integrated at 12 different sites on X and 36 reporter genes integrated at 14 different sites dispersed among the five autosomes ( Figure 2A ) . Single-copy transgene cassettes , each containing two reporters , were integrated into the genome using either targeted or random Mos1-mediated insertion ( Frøkjær-Jensen et al . , 2008 , 2014 ) . Cassettes included both Cbr-unc-119 , a neuronally expressed gene from the sister Caenorhabditid C . briggsae , and a fluorescent reporter ( gfp alone , gfp fused to histone H2B , or tdTomato fused to histone H2B ) driven by the promoter of a ubiquitously expressed C . elegans gene ( dpy-30 , eft-3 , or eft-4 ) ( Figure 2B ) . dpy-30 is an essential autosomal gene that acts independently in the DCC and the COMPASS complex , which makes the active chromatin modification H3K4me3 ( Hsu and Meyer , 1994; Miller et al . , 2001; Nagy et al . , 2002; Pferdehirt et al . , 2011; Hsu et al . , 1995 ) . eft-3 ( autosomal , also called eef-1A . 1 ) and eft-4 ( X-linked , also called eef-1A . 2 ) encode essential paralogous translation elongation factors ( Maciejowski et al . , 2005 ) . Use of multiple promoters and reporters with different expression levels and tissue specificities allowed us to test diverse gene regulatory scenarios for responsiveness to dosage compensation . 10 . 7554/eLife . 17365 . 005Figure 2 . Schematic diagram of transgenes integrated throughout the genome . ( A ) Chromosomal locations of transgene cassettes are represented by square flags marking the insertion sites on all six C . elegans chromosomes . ( B ) A transgene cassette is composed of two distinct reporters integrated at each site: Cbr-unc-119 and a fluorescent reporter with one of three C . elegans promoters . The flag color in ( A ) corresponds to the composition of the cassettes ( shown in B . Transgene cassettes containing Peft-3:gfp ( green ) and Peft-4:gfp ( red ) were inserted in the same four sites using targeted Mos1-mediated Single Copy Insertion ( mosSCI ) . Transgene cassettes containing Pdpy-30:gfp:H2B ( blue ) and Peft-3:tdTomato:H2B ( orange ) were inserted randomly throughout the genome using miniMos . Cassettes of each type are numbered sequentially , from left to right , along a chromosome . For example , in subsequent figures , the first 'green cassette' on the left end of chromosome I will be indicated by a green flag and 'Chr I , site 1'; the second 'green cassette' will be 'Chr 1 , site 2' . The first 'green cassette' on the left end of X will be indicated by a green flag and 'Chr X , site1' . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 005 Reporter gene expression was quantified from populations of L1 larvae that had been synchronized to within three hours of hatching . This strategy conferred two advantages . It eliminated any confounding influence of X-chromosome silencing in germ cells , since the L1 stage of development occurs before the onset of germline proliferation . It also minimized gene expression differences due solely to the activation or repression of genetic pathways operating at different developmental stages . To determine whether transgenes integrated on X are regulated by the DCC , we compared the overall gene expression levels in homozygous wild-type XX animals ( 2 copies of transgenes ) , homozygous dosage-compensation-defective XX animals ( 2 copies of transgenes ) , and hemizygous wild-type XO animals ( 1 copy of transgenes ) using quantitative reverse-transcriptase PCR ( qRT-PCR ) . To be considered dosage compensated , a transgene should have increased expression in DCC-defective XX animals compared to control XX animals , and it should have the same overall level of expression from the single copy in XO males as the two copies in XX hermaphrodites . That is , the single transgene copy in the male should be expressed at twice the level as either of the two copies in the wild-type hermaphrodite . To disrupt dosage compensation , we used RNAi to deplete SDC-2 , the sole hermaphrodite-specific DCC subunit that triggers assembly of DCC subunits onto X ( Dawes et al . , 1999 ) . sdc-2 ( RNAi ) causes overexpression of X-linked genes and XX-specific lethality ( Nusbaum and Meyer , 1989 ) . Depletion of SDC-2 activity not only increases expression of X-linked genes , it mildly reduces expression of about 30% of autosomal genes ( Jans et al . , 2009; Kruesi et al . , 2013 ) ( Figure 1—figure supplement 1F ) , making it essential to identify autosomal genes not affected by sdc-2 ( RNAi ) for use in normalizing gene expression . For normalization candidates , we selected 12 autosomal genes that had similar expression levels between control and sdc-2 ( RNAi ) animals , as assayed by GRO-seq , microarray , and RNA-seq experiments ( Jans et al . , 2009; Kruesi et al . , 2013 ) . We then followed the geNorm approach ( Vandesompele et al . , 2002 ) to identify the three most stably expressed autosomal genes ( cdc-42 , H06O01 . 1 , and Y38A10A . 5 ) from three replicates of control and sdc-2 ( RNAi ) animals . To verify that our normalization approach recapitulated RNA-seq data , we quantified gene expression of two dosage compensated genes on X ( F41E7 . 5 and F47B10 . 2 ) , one non-compensated gene on X ( C15C7 . 5 ) , and one autosomal gene ( F19F10 . 91 ) in both control and sdc-2 ( RNAi ) worms by qRT-PCR . The two DCC-regulated genes were significantly upregulated in sdc-2 ( RNAi ) L1s compared to control L1s , and both the autosomal gene and the non-dosage-compensated X gene were not significantly affected by sdc-2 ( RNAi ) ( Figure 3—figure supplement 1A ) , thus validating our qRT-PCR approach for assessing the dosage compensation status of any gene . Quantification of reporter mRNA levels to assess whether transgenes integrated across X were dosage compensated revealed that all 28 X-linked reporters were repressed by the DCC . All had increased expression in sdc-2 ( RNAi ) versus control L1s , regardless of their location on X and the origin of their promoter , whether from the C . elegans X chromosome ( eft-4 ) , C . elegans autosomes ( eft-3 and dpy-30 ) , or a C . briggsae autosome ( Cbr-unc-119 ) ( Figure 3A ) . The extent of DCC-mediated repression ranged from 1 . 3 to 3 . 7-fold , consistent with the range observed for endogenous dosage-compensated X-linked genes assessed by RNA-seq ( Figure 1—figure supplement 1F ) . 10 . 7554/eLife . 17365 . 006Figure 3 . Transgenes integrated on X but not autosomes are regulated by the DCC , which balances X expression between sexes . ( A ) Quantification in control RNAi XX ( light ) or sdc-2 ( RNAi ) XX ( dark ) L1 larvae of mRNA levels for the two reporters in each transgene cassette on X chromosomes ( blue ) and autosomes ( gray ) . The bars represent the average level of expression among at least three biological replicates for each reporter in a cassette . Data for reporters are presented in the same order , from left to right , as the order of transgene cassettes along a chromosome ( see Figure 2 ) . The fold change in gene expression between sdc-2 ( RNAi ) XX ( dark ) and control XX animals ( light ) is shown above each transgene , with the number asterisks indicating the p-value: p≤0 . 05 , one asterisk; p≤0 . 01 , two asterisks; p≤0 . 0001 , four asterisks ( Student's t-test ) . Error bars show the standard error of the mean for at least three biological replicates . All reporters on X show significant elevation in gene expression in the dosage-compensation-defective sdc-2 ( RNAi ) XX L1s compared to control XX L1s . In contrast , none of the reporters on autosomes exhibit a significant increase in expression in sdc-2 ( RNAi ) XX L1s . A few autosomal reporters ( see Peft-3:tdT:H2B and Peft-4:gfp ) exhibit slight but significant reduction in gene expression , consistent with previous genome-wide measurements of gene expression in dosage compensation mutants ( Jans et al . , 2009; Kruesi et al . , 2013; Crane et al . , 2015 ) . ( B ) Comparison of mRNA levels in XX vs . XO L1/L2 animals for transgene cassettes integrated on X . Total reporter mRNA levels were quantified in XX animals that were homozygous for the transgene cassettes ( 2 copies of each reporter ) and XO animals that were hemizygous for the transgene cassette ( 1 copy of each reporter ) . Reporters in cassettes selected for this experiment are designated by § or ‡ in panels A and B . The fold change in gene expression is indicated above each pair of measurements in XX and XO animals . The expression levels were not statistically different between the two copies in XX animals vs . the single copy in XO animals , indicating that the reporters in each transgene cassette became dosage compensated ( Student's t-test ) . Error bars show the standard error of the mean for at least three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 00610 . 7554/eLife . 17365 . 007Figure 3—figure supplement 1 . Transgene expression levels are highly consistent for sites across the X chromosome and for sites across an autosome in XX animals . ( A ) Data from qRT-PCR recapitulate RNA-seq and GRO-seq data . The expression levels of two dosage compensated genes ( F41E7 . 5 and F47B10 . 2 ) , one non-dosage-compensated gene on X ( C15C7 . 5 ) , and one non-dosage-compensated gene on an autosome ( F19F10 . 9 ) were determined by qRT-PCR in sdc-2 ( RNAi ) XX animals and control RNAi XX animals . Responses of the three classes of genes to sdc-2 ( RNAi ) indicate that our qRT-PCR approach to assess transgene expression recapitulates the gene expression results obtained from RNA-seq and GRO-seq experiments . Known compensated genes have elevated expression in sdc-2 ( RNAi ) animals and non-compensated genes do not . ( B ) Comparison of transgene expression at different sites along a chromosome reveals only minimal position effects . Each black circle shows the average expression level of the Cbr-unc-119 reporter transgene from all four transgene cassettes at distinct positions in the genome . Reporters are grouped by the chromosomal locations of the transgene cassette , including all sites on autosomes , all sites on chromosome V , and all sites on chromosome X . The expression levels of Cbr-unc-119 reporters on X are provided for both control RNAi and sdc-2 ( RNAi ) XX animals . Each circle represents the average of at least three biological replicates . The red bar indicates the mean , and the blue bars indicate the standard error of the mean . The variation in Cbr-unc-119 expression is low across all insertion sites , as indicated by the absolute variation in expression levels and the coefficient of variation for each group . The number above each pairwise combination indicates the fold change in gene expression , and the asterisks indicate the level of significance: p≤0 . 001 , three asterisks; p≤0 . 0001 , four asterisks ( Student's t test ) . The low expression variation among transgenes on X and autosomes permits the robust conclusion that expression of X-linked transgenes in XX animals is significantly lower than expression of autosomal transgenes , contrary to predictions of a chromosome-wide mechanism of X-chromosome gene upregulation . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 007 We also found that the X-linked reporters had equivalent expression in XX and XO animals at the L1/L2 stage . That is , the total level of transgene expression from the two X chromosomes of hermaphrodites was not statistically different from the total level of transgene expression from the single X of males ( Figure 3B ) . Together these data show that the dosage compensation process creates a chromosome-wide environment that permits repression of transgenes integrated all along the X chromosome , resulting in equivalent transcription between the sexes . In contrast to transgenes on X , expression of 36 transgenes on autosomes was not increased by disrupting dosage compensation ( Figure 3A ) . One Peft-3:tdTomato:H2B autosomal reporter and two Peft-4:gfpautosomal reporters had a slight but significant decrease in gene expression upon sdc-2 depletion FC ( Fold Change ) = 0 . 86 , p=0 . 03; FC = 0 . 76 , p=0 . 03; FC = 0 . 8 , p=0 . 008 ) . These decreases were consistent with the effect of sdc-2 mutations on 30% of autosomal genes in prior genome-wide experiments ( Jans et al . , 2009; Kruesi et al . , 2013 ) . Prior genome-wide studies found that DCC binding near an endogenous X-linked gene was neither necessary nor sufficient for the dosage compensation of the gene ( Jans et al . , 2009; Kruesi et al . , 2013 ) . We re-examined this issue using the transgenes . The DCC is recruited to endogenous X chromosomes by sequence-dependent recruitment elements on X ( rex sites ) and spreads to lower-affinity bindings sites , called dox sites ( dependent on X ) , located in promoters of actively transcribed genes ( Csankovszki et al . , 2004; McDonel et al . , 2006; Jans et al . , 2009; Pferdehirt et al . , 2011 ) . DCC occupancy at dox sites correlates directly with the expression level of the gene ( Jans et al . , 2009; Pferdehirt et al . , 2011 ) . To assess the relationship between transgene repression and proximity to a DCC binding site , we used chromatin immunoprecipitation ( ChIP ) to quantify levels of the DCC components DPY-27 ( an SMC condensin subunit ) and SDC-3 ( a zinc finger protein required for condensin loading ) bound at Peft-3:gfp and Cbr-unc-119 on X . Binding of both DPY-27 and SDC-3 to the transgenes was negligible compared to DCC binding at the strong rex sites rex-1 and rex-32 ( Figure 4A ) . The finding that transgenes lacking DCC binding are nonetheless repressed by the DCC indicates that local DCC binding is not required for dosage compensation . 10 . 7554/eLife . 17365 . 008Figure 4 . Dosage compensation of transgenes does not require local DCC binding . ( A ) Quantification of mRNA levels and DCC binding for the Peft-3:gfp and Cbr-unc-119 reporters integrated at position 15 . 6 Mb on chromosome X . Gene expression is represented as in Figure 3 . ChIP was conducted using antibodies against DCC subunits DPY-27 ( green ) or SDC-3 ( orange ) , and the negative control IgG ( gray ) . ChIP enrichment was calculated using quantitative PCR with primers for the Peft-3:gfp and Cbr-unc-119 reporters , the strong rex sites rex-1 or rex-32 , and a region of X that does not recruit the DCC . Enrichment is expressed relative to an autosomal region that does not recruit the DCC and is normalized to input . A schematic diagram of the DCC is shown . The two reporters became dosage compensated even though no DCC complex was detected at either reporter in the integrated transgene cassette . ( B ) A schematic diagram of relative positions for X-linked transgene cassettes and endogenous rex sites on the X chromosome . Cassette locations are represented as flags , as in Figure 1 . Colored bars on X indicate the positions and strength of endogenous rex sites . The 25 strongest rex sites are shown in red; all other rex sites are shown in green . rex-site strength was assessed by SDC-3 occupancy in ChIP-seq experiments . The diamond indicates the Peft-3:gfp and Cbr-unc-119 transgene cassette tested for DCC binding in part A . ( C and D ) Scatter plots compare the fold change in gene expression of reporters vs . the distance of the nearest rex site of any strength ( C ) or the nearest strong rex site ( D ) . No correlation was found between the extent of a reporter's increase in expression in DCC-defective animals and its proximity to a rex site . Linear regression lines are shown for each transgene . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 00810 . 7554/eLife . 17365 . 009Figure 4—figure supplement 1 . Transgenes do not require close proximity of a dox site to become dosage compensated . Scatter plots compare the fold change in gene expression of reporters vs . the distance of the nearest dox site . No correlation was found between the extent of a reporter's increase in expression in DCC-defective animals and its proximity to a dox site . Plots show transgenes in the X intervals of 10 . 3–12 . 7 Mb and 4 . 2–4 . 4 Mb , because those are the intervals on X for which every DCC binding peak ( 2 kb ) determined by ChIP-chip was assayed in vivo for whether it recruited the DCC in an autonomous manner ( Jans et al . , 2009 ) . For the single case in which the values for Peft-3:gfp and Cbr-unc-119 overlap , the circle is half green and half black . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 009 We then asked whether a transgene's distance from a rex site is correlated with its ability to undergo DCC-mediated repression ( gene expression fold change in sdc-2 ( RNAi ) vs . control worms ) ( Figure 4B–D ) . We found that for transgenes on X , the increase in expression in sdc-2 ( RNAi ) animals was not correlated with their proximity to a rex site . Thus , a nearby rex site is not essential for the compensation of a gene . Furthermore , DCC binding to a nearby dox site is also not essential for the compensation of a gene ( Figure 4—figure supplement 1 ) . Together , these data indicate that the DCC can act at a distance to control expression of foreign genes integrated across the X chromosome . The finding that nearby DCC binding is not necessary for a transgene on X to become dosage compensated caused us to ask whether a closely linked rex site could elicit DCC-mediated repression of transgenes on autosomes in XX animals . We analyzed expression of Cbr-unc-119 and Peft-3:gfp in transgene cassettes that were integrated with and without rex-32 at four sites on autosomes . We first showed the DCC binds to these ectopic rex sites . SDC-3 binding at the ectopic rex-32 sites on autosomes was similar to , or greater than , binding at the endogenous rex-32 site on X ( Figure 5B and C ) . Similarly , DPY-27 occupancy at the ectopic rex-32 site on chromosome I was equivalent to DPY-27 occupancy at the rex-32 site on X ( Figure 5B ) . 10 . 7554/eLife . 17365 . 010Figure 5 . Comparison of transgene expression on X and autosomes with and without a co-inserted strong rex site . ( A ) Each graph depicts expression levels of either Cbr-unc-119 or Peft-3:gfp , integrated at the same site on either X or an autosome with ( TG + rex-32 ) or without ( TG ) a co-inserted copy of the strong DCC binding site rex-32 , in both control RNAi XX ( light ) and sdc-2 ( RNAi ) ( dark ) XX animals . The specific insertion site is indicated above the graph and corresponds to the schematic on the right . Expression of autosomal transgenes is shown in light and dark gray , and X-linked transgenes in light and dark blue . Numbers above the graphs show the fold change in gene expression ( red lines ) between transgenes with and without the co-inserted rex site in either control RNAi or sdc-2 ( RNAi ) animals . Also shown is the fold change in expression ( gray lines ) of transgenes in control RNAi vs . sdc-2 ( RNAi ) animals , either with or without rex-32 . The number of asterisks indicates the p-value: p≤0 . 05 , one asterisk; p≤0 . 01 , two asterisks; p≤0 . 001 , three asterisks; p≤0 . 0001 , four asterisks . Proximity to a co-inserted rex site does not increase gene expression in sdc-2 ( RNAi ) vs . control RNAi animals , nor does it generally decrease expression significantly on autosomes in control animals . Error bars show the standard error of the mean for at least three biological replicates . ( B ) Binding of DPY-27 ( green ) , SDC-3 ( orange ) , and IgG ( gray ) at Chr I , site 1 was assayed by ChIP-qPCR at the co-inserted copy of rex-32 . For this site , ChIP was conducted in an engineered strain lacking the endogenous copy of rex-32 , and the graph depicts the enrichment of the two DCC components at the center of the co-inserted rex-32 . Similar ChIP experiments were conducted for the endogenous rex-32 site in a wild-type strain . Enrichment is expressed relative to an autosomal region that does not recruit the DCC and is normalized to input . ( C ) For the designated sites on X and autosomes , SDC-3 ( orange ) and IgG ( gray ) ChIP were quantified at the inserted copy of rex-32 in strains carrying the endogenous wild-type copy of rex-32 by using primers that recognize the unique junction between rex-32 and Peft-3::gfp . For four ectopic rex sites , the level of SDC-3 binding was similar to its level at the endogenous rex-32 site . The ectopic rex-32 inserted on Chr II bound significantly more SDC-3 than the endogenous copy on X . Error bars show the standard error of the mean for at least two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 010 Despite strong DCC binding to rex sites closely linked to the autosomal reporters , expression of seven of eight reporters was not significantly reduced compared to expression of the same reporters integrated at the same autosomal sites without a rex site ( Figure 5A ) . Furthermore , in sdc-2 ( RNAi ) XX animals with very low levels DCC binding , expression of six of eight autosomal reporters with closely linked rex sites was not elevated compared to expression of the same rex-linked autosomal reporters in control XX animals with high levels of DCC binding . These results indicate that strong DCC binding adjacent to a gene is generally not sufficient to regulate its expression . Furthermore , they are consistent with results showing that close proximity of DCC binding to either an endogenous X-linked gene or an engineered X-linked transgene was not necessary for the dosage compensation of the gene . Our results strongly support a model in which DCC binding causes global changes to the X chromosome , likely by remodeling X topology ( Crane et al . , 2015 ) to elicit chromosome-wide gene repression . Having characterized a set of X-linked and autosomal transgenes thoroughly , we could use them to assess whether a chromosome-wide mechanism of upregulation functions in C . elegans to balance expression between X chromosomes and autosomes , consistent with Ohno's hypothesis . If an X-linked transgene is controlled by both a dosage compensation mechanism , which halves X expression in XX animals , and an upregulation mechanism , which doubles X expression in both sexes as Ohno hypothesized , the per-copy expression of the X transgene in XX animals will be similar to that of an autosomal transgene ( Figure 1D and Figure 6A ) . However , if the X-linked transgene is downregulated by the dosage compensation machinery in XX animals but is not controlled by an upregulation mechanism that operates in both sexes , the transgene will be expressed at half the level of an autosomal transgene ( Figure 1C and Figure 6A ) . Lastly , if the X chromosome is controlled by an Ohno-like upregulation mechanism , an X-linked transgene in a dosage-compensation-defective XX mutant or an X transgene that escapes dosage compensation in wild-type XX animals will be expressed at twice the level of an autosomal transgene . Without X-chromosome upregulation , these X and autosomal transgenes will be expressed at a similar level ( Figure 6A ) . 10 . 7554/eLife . 17365 . 011Figure 6 . Transgenes on X are expressed at half the level of transgenes on autosomes . ( A ) Predicted vs . observed transgene expression levels for dosage compensated transgenes on X relative to transgenes on autosomes for two models of X-chromosome regulation . ( Left ) Under an X-chromosome-wide model of upregulation , dosage-compensated transgenes on hermaphrodite X chromosomes ( 2 copies ) are predicted to have similar average total expression levels as transgenes on autosomes ( 2 copies ) , despite the hermaphrodite-specific repression by the DCC . Moreover , because the DCC reduces gene expression on X by about half , sdc-2 ( RNAi ) animals would be predicted to have two-fold more transgene expression on X relative to autosomes , if X-chromosome upregulation occurred . ( Middle ) Under a model of no X-chromosome-wide upregulation , dosage compensated transgenes would be expressed at half the level of transgenes on autosomes due to repression by the DCC . In DCC-defective XX animals , transgene expression on X ( 2 copies ) would increase to the level of transgene expression on autosomes ( 2 copies ) . ( Right ) The results of comparing the average expression level of all transgenes on X and autosomes argue against a chromosome-wide model of X-chromosome upregulation . For each reporter , data were normalized to the average autosomal expression level and then combined . Numbers above the graph show the fold change in expression between transgenes on X and on autosomes in control RNAi animals or between transgenes on X in sdc-2 ( RNAi ) animals and transgenes on autosomes in control RNAi animals . The normalized expression level of all transgenes on the X chromosome ( light blue ) is only 56% of the normalized expression level of all transgenes on autosomes ( gray ) . Expression of transgenes on X is increased to 93% of transgene expression on autosomes in animals treated with RNAi against sdc-2 ( dark blue ) . p≤0 . 0001 , four asterisks . Error bars show the standard error of the mean . ( B ) Comparison of averaged mRNA expression for all reporter transgenes of each type located at all sites on X or autosomes as quantified in ( A ) , except that expression levels were not normalized to the average autosome expression level . For example , in the first panel , averaged expression from four Pdpy-30:gfp:H2B reporter transgenes inserted on autosomes is compared with averaged expression from four Pdpy-30:gfp:H2B reporter transgenes inserted on X . Expression levels of the Cbr-unc-119 reporter included with each transgene cassette are shown below the expression levels of the fluorescent partner transgenes . The number of asterisks indicates the p-value: p≤0 . 05 , one asterisk; p≤0 . 01 , two asterisks ( Student's t-test ) . Error bars show the standard error of the mean . ( C ) Comparison between XX and XO L1/L2 larvae of total expression levels for transgenes inserted on X vs . autosomes . For the transgene cassette Pdpy-30:gfp:H2B and Cbr-unc-119 , the autosomal cassette was on chromosome IV at site 1 , and the X cassette was at site 3 . For the transgene cassette Peft-3:gfp and Cbr-unc-119 , the autosomal cassette was on chromosome I at site 1 and the X cassette at site 1 . Shown below each bar are the sex of the animals in which gene expression was quantified , the copy number of the reporter transgene , and the position of the quantified reporter transgene ( either on X or autosomes ) . For all transgenes , we compared the total level of expression from two reporter copies in XX animals and one reporter copy in XO animals , regardless of whether the reporters were on X or an autosome . The fold change in gene expression between reporters on X and autosomes is given above the graphs . The number of asterisks indicates the p-value: p≤0 . 05 , one asterisk; p≤0 . 01 , two asterisks ( Student's t-test ) . Error bars show the standard error of the mean for at least three biological replicates . Expression of two copies on X was about half the expression of two copies on autosomes . Similarly , expression of the single copy on males was not different from the single copy on autosomes , meaning that expression of one copy on the male X would be half the expression of two copies on male autosomes . Results in ( B , C ) argue against an Ohno-like mechanism of X-chromosome upregulation in which chromosome-wide transcription of X is increased in expression . The results do not exclude the possibility that diverse gene-specific mechanisms might have arisen to elevate expression of individual X-linked genes with reduced dose . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 011 We found that the average total expression of all X-linked transgenes in XX animals ( 2 copies ) was 56% of the average total expression of all transgenes on autosomes ( 2 copies ) ( p<0 . 0001 , 95% CI of the mean between 0 . 497 and 0 . 620 ) ( Figure 6A , right panel ) . Thus , the relative level of transgene expression on X compared to autosomes differs significantly from the ratio of 1 predicted by an Ohno-like model of X chromosome-wide upregulation ( One Sample t-test , p<0 . 001 ) and is not significantly different from the ratio of 0 . 5 predicted by the lack of a general upregulation mechanism ( One Sample t-test , p=0 . 06 ) . Moreover , when we analyzed the data by the category of reporter gene ( Pdpy-30:gfp:H2B , Peft-3:tdT:H2B , Peft-3:gfp , Peft-4-gfp , or Cbr-unc-119 ) and compared the average expression level for each reporter at all sites on X and on autosomes , we found that the average expression was also reduced for reporters on X compared to autosomes ( Figure 6B ) , like the combined expression data for all reporters . Depending on the reporter , X chromosome transgenes were expressed on average between 44% and 68% of their autosomal counterparts ( Figure 6B ) , again arguing against a chromosome-wide mechanism to increase gene expression on X chromosomes . For six of eight reporters tested , the reduction in gene expression on X vs . autosomes was statistically significant . The two remaining reporters Peft-3:gfp and Peft-4:gfp had reduced expression when integrated on X vs . autosomes , but the reduction was not statistically significant ( fold change = 0 . 52 , p=0 . 14; fold change = 0 . 68 , p=0 . 13 ) , likely due to the small number of these reporters on X . Together , these data argue against a chromosome-wide mechanism of X upregulation that increases gene expression on the two hermaphrodite X chromosomes to balance expression with that of autosomes . In a separate analysis of Ohno's hypothesis , if a chromosome-wide mechanism were to elevate expression of genes on X , a transgene on the single male X chromosome would be expressed at twice the level as a transgene on one of a pair of homologous autosomes . We found to the contrary that expression of a single transgene on the male X was not different from expression of a single transgene on one of the autosomal homologs ( Figure 6C ) . Thus , while our results demonstrate a DCC-mediated , chromosome-wide mechanism to equalize X gene expression between the sexes by reducing expression of endogenous and ectopic genes on hermaphrodite X chromosomes , our results argue against an analogous chromosome-wide mechanism that increases transcription of X chromosomes in both sexes to balance expression between X chromosomes and autosomes . We were able to reach the robust conclusion that expression of transgenes on X is significantly lower than expression of transgenes on autosomes because our reporters had only minimal variability in expression when integrated at different sites along the X chromosome or autosomes . As examples , despite the diversity of insertion sites for the 14 Cbr-unc-119 transgenes integrated across X or the 18 Cbr-unc-119 transgenes integrated across autosomes , the variation in expression was low as indicated by the absolute variation in expression and the coefficient of variation ( Figure 3—figure supplement 1B ) . Our analysis of X-chromosome regulation enabled us to evaluate an attractive but speculative model of X-chromosome dosage compensation , which posits that repression of X-linked gene expression in XX animals by the DCC is merely the result of escaping from a chromosome-wide mechanism that upregulates X expression ( Sharma et al . , 2014; Sharma and Meister , 2015 ) . In particular , these authors proposed that rex sites target X to the nuclear periphery in males to increase chromosome-wide gene expression , while DCC binding to rex sites in hermaphrodites blocks the peripheral localization , relocating X to the interior and hence reducing X gene expression . The minimal evidence that led to this model included the following: ( 1 ) low-resolution DamID studies suggesting that X associates with nuclear pore proteins more frequently in males than in hermaphrodites , and FISH experiments suggesting that X associates with the nuclear periphery more frequently in XO than XX embryos; ( 2 ) FISH experiments suggesting that ectopic insertion of a truncated rex site at one autosomal location targets the locus to the nuclear periphery more frequently in XO and DCC-defective XX animals than in wild-type XX animals; ( 3 ) FISH experiments suggesting that seven X-linked rex sites had enriched association with the nuclear periphery in males vs . random nuclear positioning . However , only 3 of the rex sites ( at the center of X: rex-33 , rex-28 , and rex-8 ) showed a significant increase in peripheral localization in males versus hermaphrodites . No gene expression studies tested the validity of the model , and no DCC binding studies tested whether the truncated rex site integrated into the autosome recruited the DCC in XX embryos . The truncated autosomal rex site lacked a full-length DNA motif important for robust DCC binding , making the site unlikely to be a strong DCC recruiter in single copy ( McDonel et al . , 2006; Jans et al . , 2009 ) . In fact , the truncated site only partially recruited the DCC in vivo when present in multiple copies ( McDonel et al . , 2006 ) . Because our X-linked transgenes are fully responsive to the dosage compensation process , they are valid tools for assessing this untested model . Regulation of transgene expression should fulfill expectations of the model if the model is correct . Instead , results from four different experimental approaches , including transgene expression and chromosome localization studies , are inconsistent with expectations of this model . First , the nuclear positioning model predicts that in males the transgenes on X should have elevated expression relative to transgenes on autosomes due to a rex-dependent association of X with the nuclear periphery . However , we found that in males , expression of single-copy transgenes on the sole X chromosome was not different from expression of single-copy transgenes inserted on only one of two homologous autosomes ( Figure 6C ) , contrary to the nuclear positioning model of dosage compensation ( Sharma et al . , 2014; Sharma and Meister , 2015 ) . Second , the nuclear positioning model predicts that in XX animals defective in DCC binding , expression of a transgene co-inserted with a closely linked rex site should be higher than expression of the same transgene inserted at the same locus without a rex site . To test this prediction , we analyzed expression of the reporters Cbr-unc-119 and Peft-3:gfp that were co-inserted with and without rex-32 at four sites on autosomes and two sites on X . DCC binding to the ectopic rex sites was strong in the transgenic XX animals ( Figure 5B , C ) . We found that in XX animals , depletion of the DCC did not significantly increase the expression of any of the 12 rex-linked reporter transgenes on X chromosomes or autosomes compared to their expression at the same locations without a rex site ( Figure 5A ) . These findings challenge the proposal ( Sharma et al . , 2014 ) that rex sites enhance expression of a nearby gene in the absence of DCC binding . Furthermore , loss of DCC binding in XX animals did not significantly elevate expression for six of eight autosomal reporters with an adjacent rex site ( Figure 5A ) . Thus , DCC binding adjacent to a gene is not sufficient to induce repression of the gene , unlike the expectation from the nuclear positioning model . Third , using the same approach as ( Sharma et al . , 2014 ) but contrary to their single result , we found that rex sites in autosomes , validated for DCC binding in wild-type XX animals , did not preferentially recruit flanking autosomal DNA to the nuclear periphery under conditions in which DCC binding was prohibited ( Figure 7 A–C ) . We checked rex localization in young embryos of the age used by ( Sharma et al . , 2014 ) and also older embryos to be certain that embryo age did not affect our conclusion . Specifically , the autosomal regions of chromosome I and IV that flank the integrated ectopic rex-32 site did not localize more frequently to the nuclear periphery ( zone 1 vs . zones 2 and 3 ) in either younger ( 50–140 cells ) or older ( >200 cells ) XO embryos ( both lacking DCC binding at rex-32 ) compared to age-matched XX embryos ( both proficient in DCC binding at rex-32 ) ( Figure 7 A–C ) . In addition , autosomal loci adjacent to a rex insertion were not localized more frequently to the nuclear periphery in younger or older XO embryos than the same autosomal loci in age-matched XO embryos lacking the rex insertion . For comparisons involving younger embryos , p>0 . 5; for those involving older embryos , p>0 . 1 ( chi-square tests ) . Thus , an ectopic rex site lacking DCC binding is insufficient to relocate flanking autosomal DNA to the nuclear periphery . 10 . 7554/eLife . 17365 . 012Figure 7 . Proximity to an endogenous or ectopic rex sites does not determine the positions of X or autosomal loci relative to the nuclear periphery . ( A ) Illustration of approaches for quantifying the radial positions of FISH signals , the same approach as used by Sharma and Meister ( 2014 ) . ( Top ) A stack of confocal images determines the location of FISH signals in 3D . The nucleus is divided into three concentric zones with equal area . A random distribution would result in 33% of the FISH signals in each zone , as marked by the gray line in ( B–D ) . ( Middle ) , For each FISH signal , the ratio between its distance to the nuclear periphery and the nuclear radius determines the zone in which the site resides . ( Bottom ) , Representative confocal image showing position of FISH signals in a nucleus . Blue , DAPI; Red , Nuclear Pore Complex; Green , rex FISH; Scale bar , 1 μm . ( B , C ) Histograms show the fraction of autosomal FISH signals in each of three zones in XX or XO embryos of two age groups ( 50-140-cell stage or > 200-cell stage ) from wild-type strains or ectopic rex-insertion strains . Genomic locations of insertion sites ( Chr 1 , site 1 at position 2 , 85 , 041 Mb or Chr IV , site 1 at position 5 , 014 , 698 Mb ) for transgenes with and without ectopic rex sites are shown on chromosome maps above the histograms . For both autosomal loci examined , nuclear positioning was not statistically different between ( 1 ) XX and XO embryos with an ectopic rex insertion; ( 2 ) between XO embryos with or without a rex insertion; ( 3 ) between XO embryos with a rex insertion and XX embryos without a rex insertion; or ( 4 ) between XX and XO embryos without a rex insertion , regardless of embryo age ( p>0 . 5 for younger embryos and p>0 . 1 for older embryos , chi-square test ) . N is the total number of autosomal or rex FISH signals quantified for XX or XO embryos in the two age groups of each genotype . FISH probes were 30 kb . ( D ) Histograms show the fraction of endogenous rex FISH signals ( rex-32 , rex-23 , rex-33 , rex-47 , or rex-8 ) at the nuclear periphery ( zone 1 ) of wild-type XX or XO embryos of two age groups ( 50-140-cell stage or > 200-cell stage ) . Genomic locations of the 25 highest DCC-occupied rex sites on X are shown above the histograms ( red , rex sites examined; black , all others ) . The positioning of rex sites at the nuclear periphery ( zone 1 ) was not greater in XO versus XX embryos of either age group ( chi-square test , see text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 01210 . 7554/eLife . 17365 . 013Figure 7—figure supplement 1 . Radial position of rex sites on X relative to the nuclear periphery is not different between XX and XO embryos . ( A ) Cartoons illustrate methods for quantifying the radial positions of rex FISH signals as explained in the legend to Figure 7A . ( B ) Histograms show the fraction of endogenous rex FISH signals ( rex-32 , rex-23 , rex-33 , rex-47 , or rex-8 ) in each of three zones in wild-type XX or XO embryos of two age groups ( 50-140-cell stage or > 200-cell stage ) . Genomic locations of the 25 highest DCC-occupied rex sites on X are shown above the histograms ( red , rex sites examined; black , all others ) . The nuclear positioning of rex sites in each zone is not statistically different between XO and XX embryos of either age group ( chi-square test , see text ) . Mb , megabases . N is the total number of rex FISH signals quantified for XX or XO embryos in the two age groups . A random distribution would result in 33% of the FISH signals in each zone , as indicated by the gray line . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 01310 . 7554/eLife . 17365 . 014Figure 7—figure supplement 2 . Examples of XO embryos hybridized with FISH probes to assess nuclear positioning of ectopic rex sites integrated onto autosomes and endogenous rex sites on X . ( A and B ) Confocal images of wild-type or transgenic XO embryos stained with DAPI ( blue ) and hybridized with various FISH probes ( red ) as indicated below each image ( red ) . Each image represents a 1 μm projection of 12 slices . Scale bars , 5 μm ( A ) Embryos to examine nuclear positioning of ectopic rex sites . Strains TY5726 and EG6136 contain rex sites integrated onto autosomes at locations indicated on the chromosome maps above the images . Probes are 30 kb fosmid probes that are adjacent to or flank the rex insertion sites so that the same probes can be used for wild-type and transgenic embryos . ( B ) Embryos to examine nuclear positioning of endogenous rex sites . Shown above the images is a map of X with genomic locations of the 25 highest DCC-occupied rex sites . rex sites examined by FISH are shown in red; all other sites are in black . Probes for rex-32 and rex-23 are 30 kb fosmid probes and probes for rex-47 and rex-8 are 5 kb probes . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 01410 . 7554/eLife . 17365 . 015Figure 7—figure supplement 3 . Examples of XO and XX embryos hybridized with FISH probes to assess nuclear positioning of rex-33 on X . Confocal images of wild-type XO and XX embryos stained with DAPI ( blue ) and two overlapping 30 kb fosmid probes each labeled with a different fluorescent dye ( red or green as indicated ) . Each image represents a 1 μm projection of 12 slices . Scale bars , 5 μm . Only foci with signals from both fosmid probes ( indicated by arrows ) were quantified to unambiguously identify bona fide rex-33 FISH signals . Genomic locations of the 25 highest DCC-occupied rex sites on X are shown above the images ( rex-33 , red; all other rex sites , black ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17365 . 015 Fourth , all five endogenous rex sites tested on X , including three in the center ( rex-33 , rex-47 and rex-8 ) and two toward one end ( rex-32 and rex-23 ) , were not preferentially recruited to the nuclear periphery in XO embryos in either the 50-140-cell stage or the > 200-cell stage compared to age-matched XX embryos ( p>0 . 1 , chi-square test ) ( Figure 7D and Figure 7—figure supplement 1 A , B ) . Indeed , for rex-23 the opposite was found: older XX embryos showed more peripheral localization than older XO embryos ( p<0 . 001 , chi-square test ) , but both sexes showed less peripheral localization than expected by random chance ( p<0 . 0001 ) . For rex-47 , younger embryos of both sexes had enrichment in the peripheral-most zone ( p=0 . 008 ) , but peripheral enrichment in XO embryos was not greater than in XX . These results argue against a rex-dependent nuclear positioning model of dosage compensation . Furthermore , the inconsistency in rex nuclear localization across X in the Sharma et al . ( 2014 ) data set is difficult to reconcile with a spatial positioning model that is proposed to control gene expression across the entire X ( see Discussion ) . We conclude that while we have not ruled out the interesting possibility that X chromosome nuclear positioning might play a role in C . elegans dosage compensation , we have provided compelling evidence that the explicit model of ( Sharma et al . , 2014 ) cannot account for the mechanism of C . elegans dosage compensation . Whether DCC binding nearby a gene is necessary or sufficient for dosage compensation of the gene has been debated , leaving open the question of whether the dosage compensation mechanism acts on a gene-by gene basis or primarily through a chromosome-wide process that changes fundamental properties of X . Prior genome-wide studies supported a chromosome-wide mechanism by demonstrating no correlation between DCC binding in or near an endogenous gene and the dosage compensation status of the gene ( Jans et al . , 2009; Kruesi et al . , 2013 ) . Hence , DCC binding near a gene could not be the sole determinant of dosage compensation . An alternate interpretation by others posited that DCC binding near a gene is essential for its repression , but elevated expression in dosage-compensation mutants of genes lacking a DCC binding site was an indirect consequence of the mis-regulation of other genes on X , instead of the attenuation of a chromosome-wide mechanism ( Strome et al . , 2014 ) . Our experiments provide strong evidence against this alternative interpretation . First , dosage compensation of transgenes integrated on X did not require local DCC binding or even close proximity to a strong DCC recruitment site . Dosage compensation was judged by two criteria: the increased expression of transgenes in dosage-compensation-defective XX mutants vs . wild-type XX animals , and the statistically indistinguishable levels of overall transgene expression between wild-type homozygous transgenic XX and hemizygous transgenic XO animals . The former criterion averted any potential differences due to sex-specific gene expression , and the latter criterion averted any complications that might arise from the disruption of dosage compensation and consequent elevation of X expression . Second , only transgenes on X were responsive to the dosage compensation process . Indirect effects caused by the disruption of dosage compensation would be predicted to affect transgenes on autosomes as well . Third , close proximity of strong DCC recruitment sites to transgenes on autosomes did not elicit gene repression . That is , in wild-type XX embryos , expression of transgenes on autosomes was not lower in the presence of nearby rex sites than in their absence , and transgenes co-inserted with rex sites were not generally increased in expression in dosage-compensation-defective mutants , further demonstrating that local DCC binding is not sufficient to establish dosage compensation . The promoter of an X-linked gene also does not dictate the dosage compensation status . Although the eft-4 gene is dosage compensated at its endogenous location on X , and X-linked transgenes driven by the eft-4 promoter are dosage compensated , eft-4-driven transgenes on autosomes are not . Thus , the promoter does not transmit responsiveness to the dosage compensation process . Our data support the model that rex sites act in concert and over long distance to establish an X-chromosome environment that promotes reduction of gene expression across the entire chromosome , even for endogenous and engineered genes that lack DCC binding sites . Our results are in strong agreement with recent studies demonstrating that DCC binding at rex sites controls the topology of the entire X chromosome ( Crane et al . , 2015 ) . All engineered transgenes on X were responsive to dosage compensation yet some endogenous X-linked genes escape dosage compensation in C . elegans . These results suggest a model in which endogenous X-linked genes that escape regulation may have special features that insulate them from repression by the dosage compensation machinery . These features likely operate locally because genes that escape dosage compensation are in close proximity to and interspersed with genes that undergo dosage compensation ( Jans et al . , 2009; Kruesi et al . , 2013; Crane et al . , 2015 ) . As in C . elegans , most X-chromosome transgenes in both Drosophila melanogaster and mice are subject to XX/XY dosage compensation ( Scholnick et al . , 1983; Spradling and Rubin , 1983; Krumlauf et al . , 1986; Dandolo et al . , 1993; Tan et al . , 1993; Yang et al . , 2012 ) , but in both species , some endogenous genes on X escape . In support of an insulation model , the territory surrounding the mouse gene Kdm5c has two separable regulatory domains that influence X inactivation in opposite ways: one causes Kdm5c to escape X inactivation at its endogenous site on X and at ectopic sites on X , and a second region prevents X-linked genes nearby Kdm5c from escaping X inactivation ( Horvath et al . , 2013 ) . Whether similar insulators exist in C . elegans is yet to be determined . Upregulation of X-chromosome gene expression was proposed to be an essential step in the evolution of sex chromosomes from a pair of ancestral autosomes ( Ohno , 1967 ) . However , compelling evidence has neither validated nor refuted this hypothesis for most species , with the plausible exception of placental and marsupial mammals , which had different outcomes ( Julien et al . , 2012 ) . Our analysis in C . elegans overcame limitations in prior studies and provided strong evidence against a chromosome-wide mechanism that increases transcription on X to balance gene expression between X chromosomes and autosomes during sex-chromosome evolution . These results rule out one plausible molecular mechanism by which Ohno's hypothesis might work , suggesting that if upregulation does occur , it operates through multiple , diverse gene-specific mechanisms , as discussed later . In hermaphrodites , the X-linked dosage-compensated transgenes ( 2 copies ) were expressed at 56% of their autosomal counterparts ( 2 copies ) , a value significantly different from the X-to-A expression ratio of 1 predicted by a model of X-chromosome-wide upregulation . In males , expression of transgenes on X ( 1 copy ) was indistinguishable from expression of transgenes on one homolog of an autosome pair ( 1 copy ) , in contrast to the doubled X expression predicted by X-chromosome-wide upregulation . Both observations are inconsistent with a chromosome-wide mechanism for elevating X-linked gene expression . We propose that the discrepancy in conclusions about X-chromosome-wide upregulation between our current study and previous studies ( Deng et al . , 2011; Kruesi et al . , 2013 ) results from an incorrect prior assumption that the average expression of all present-day autosomes serves as a reliable proxy for expression of proto-X chromosomes . That assumption was shown to be incorrect for placental mammals ( Julien et al . , 2012 ) , and our observation that the average expression for each of the five autosomes varies by 1 . 9-fold undermines that assumption for C . elegans . Even under a conservative assumption that the proto-X chromosome ( pX ) would be expressed within the range of individual present-day autosomes , the predicted X/pX expression ratio would vary from 0 . 56 to 1 . 22 . An X/pX expression ratio of 0 . 5 would refute Ohno's hypothesis , and a ratio of 1 would support it , making the expression level of present-day autosomes too dissimilar to estimate pX expression for a test of Ohno's hypothesis . In a different test of Ohno's hypothesis , a recent C . elegans study compared the expression of 276 genes located on chromosome I in the nematode Pristionchus pacificus but on chromosome X in Caenorhabditis due to a chromosome translocation that occurred after the species' divergence ( ~ 300 MYA ) ( Albritton et al . , 2014 ) . The study found that in XO animals , the new genes on the C . elegans X chromosome were expressed , on average , at about half the level of their autosomal orthologs in P . pacificus , a result the study concluded is inconsistent with chromosome-wide upregulation of X . While this study bypassed the need to approximate the proto-X expression level , it did not take into account X-chromosome silencing during germ cell proliferation . Expression of X chromosomes and autosomes was measured in young adults , leaving open the possibility that the reduction in expression of 276 genes on the C . elegans X chromosome was due to X-chromosome-specific germline silencing , rather than lack of X-chromosome upregulation in somatic cells . Consistent with this interpretation , we found that expression of the 276 genes was higher in somatic cells of C . elegans germlineless XX L4 larvae than in the combination of somatic and germ cells found in fertile , wild-type L4 XX larvae ( Figure 1—figure supplement 1E ) . Therefore , germline silencing , which causes an underestimate of somatic gene expression in both sexes ( Deng et al . , 2011 ) , is a plausible cause for the apparent reduction in expression of genes naturally translocated to X during evolution compared to their expression when sited on autosomes . Furthermore , without knowledge of whether the autosomal P . pacificus genes are subject to germline silencing , the data also cannot be used to conclude that X upregulation occurs . Thus , the experimental design appears to have precluded a reliable assessment of gene expression needed to assess Ohno's hypothesis . In other experiments bearing on nematode X-chromosome upregulation , the histone mark H4K16ac was used as a proxy for gene activity ( Wells et al . , 2012 ) . In Drosophila , elevation of H4K16Ac is the hallmark of an upregulated male X chromosome . In the C . elegans studies , H4K16ac was found to be depleted on X chromosomes vs . autosomes of wild-type hermaphrodites and enriched on X vs . autosomes of dosage-compensation-defective hermaphrodites , suggesting an involvement of H4K16Ac in upregulation . However , males showed no X-chromosome enrichment of H4K16Ac , a result that precludes a role for H4K16Ac in X-chromosome upregulation . Thus , no prior evidence validates or convincingly refutes Ohno's hypothesis in C . elegans . Our current work provides strong evidence against an X-chromosome-wide mechanism of transcriptional upregulation that would fulfill Ohno's hypothesis . The lack of global X upregulation in C . elegans is consistent with findings in placental mammals ( Julien et al . , 2012 ) , suggesting that many organisms tolerated the evolution of sex chromosomes without a global compensation mechanism to correct for reduced X-chromosome gene expression . ( See discussion below for alternative mechanisms of compensation ) . Existence of an X-chromosome-wide mechanism to upregulate gene expression was a key assumption in a model proposed to explain X-chromosome dosage compensation ( Sharma et al . , 2014; Sharma and Meister , 2015 ) . In this nuclear positioning model , rex sites in males were proposed to promote interactions between X chromosomes and nuclear pore proteins at the nuclear periphery that would induce chromosome-wide upregulation of X expression . In hermaphrodites , DCC binding to rex sites was proposed to block interactions between X chromosomes and nuclear pore proteins , thereby reducing expression by preventing upregulation of both Xs . Key predictions of this nuclear positioning model were not fulfilled by our studies . Transgenes on X that were shown to be fully responsive to the dosage compensation process were not regulated in a manner consistent with the nuclear positioning model of dosage compensation: for example , X-linked transgenes were not expressed at higher levels in males than transgenes on autosomes . Furthermore , ectopic rex sites on X chromosomes or autosomes were insufficient to alter gene expression of nearby transgenes or to relocate flanking DNA to the nuclear periphery . Moreover , all five rex sites tested on X failed to show enhanced peripheral localization in males compared to hermaphrodites . Included among them were three rex sites in the center of X , contrary to the positive results of Sharma and Meister , and two rex sites at the end of X , consistent with their negative results . Since the experimental approach was similar for both laboratories , reasons for differences in rex localization patterns are not apparent . More confounding for the nuclear positioning model is the finding by Sharma and Meister that differences in rex nuclear localization patterns between the sexes were not uniform across X , unlike the expectation for a robust mechanism of dosage compensation that acts chromosome wide . Without a compensating mechanism to upregulate X-chromosome-wide gene expression , how did organisms tolerate the reduction in dose of neo-X-linked genes during sex chromosome evolution caused by the loss of homologous genes on Y or the complete demise of Y ? We consider several possibilities for C . elegans . First , hemizygosity of many X-linked genes might not have been deleterious during sex chromosome evolution . Precedent for this possibility comes from studies of the newly formed sex chromosomes ( neo-X and neo-Y ) of Drosophila miranda . In D . miranda , a fusion between the Y chromosome and an autosome initiated the formation of these new sex chromosomes about 1 MYA ( Bachtrog and Charlesworth , 2002 ) . Since the fusion , almost half of the genes on the neo-Y chromosome were lost or accumulated inactivating mutations ( Bachtrog et al . , 2008; Zhou and Bachtrog , 2012 ) . The neo-X chromosome acquired dosage compensation by upregulating X-chromosome expression in males using the same dosage compensation machinery as in D . melanogaster ( Bone and Kuroda , 1996; Marín et al . , 1996; Alekseyenko et al . , 2013 ) . However , compensation in D . miranda is far less complete than that for D . melanogaster ( Zhou et al . , 2013 ) . Many genes that were lost from the neo-Y chromosome are not yet upregulated on the D . miranda neo-X chromosome , suggesting that D . miranda can tolerate hemizygosity of many X-chromosome genes in the male XY sex . A second mechanism to compensate for reduced X-chromosome gene dose might have been to alter the genetic content of X chromosomes during sex chromosome evolution to favor genes whose lowered dose in males would be tolerated . In C . elegans , gene content on X differs from that on autosomes: the X is significantly depleted of essential and haploinsufficient genes compared to autosomes ( Kamath et al . , 2003; de Clare et al . , 2011; Albritton et al . , 2014 ) . Furthermore , a few genes on the C . elegans X chromosome maintain a functional paralog on an autosome ( Maciejowski et al . , 2005 ) . Together , these features of X have been proposed to prevent problems caused by germline silencing of X , but changes in X-chromosome gene content may equally well have played a role during sex chromosome evolution to accommodate the reduced dose of X-linked genes . Relocating haploinsufficient or essential genes from X to autosomes may have relaxed the pressure to balance gene expression between X chromosomes and autosomes . Third , instead of one chromosome-wide mechanism of gene regulation , several different gene-specific mechanisms might have compensated for the reduced dose of haploinsufficient genes on X . Precedent for this model comes from studies of aneuploidy in yeast and dose-sensitive genes in placental mammals . A balance hypothesis , borne out in yeast , posits that gene dose is especially important for genes that encode subunits of protein complexes , because changes in gene dose can disrupt complex formation by altering subunit stoichiometry ( Papp et al . , 2003 ) . In recent studies of aneuploid laboratory yeast strains , post-translational mechanisms , especially protein degradation , were found to attenuate the increase in protein levels caused by the increased dose of specific classes of genes , particularly those encoding subunits of multi-protein complexes ( Torres et al . , 2010; Dephoure et al . , 2014 ) . In wild yeast strains , chromosomal amplifications have been found , but a debate exists over whether an active mode of dosage compensation specifically reduces transcript levels of amplified genes ( Hose et al . , 2015; Gasch et al . , 2016; Torres et al . , 2016 ) . In placental mammals , which lack global X upregulation , reduced X-chromosome dose can be compensated in some cases by increasing expression of dose-sensitive genes on X , particularly those producing subunits of large protein complexes ( Pessia et al . , 2012 ) , or by decreasing expression of autosomal genes that produce subunits of macromolecular complexes containing proteins encoded by X-linked genes ( Julien et al . , 2012 ) . While the general mechanisms that govern local changes in mammalian gene expression are not well known , one source of gene regulation is the mammalian histone methyl transferase complex called MOF , which acetylates histone H4 on lysine 16 to upregulate expression of a small set of genes on X ( Deng et al . , 2013 ) . In addition , for some X-linked genes , evidence also exists for enhanced mRNA stability or enhanced translational efficiency via increased ribosome density as possible mechanisms to compensate for reduced gene dose ( Deng et al . , 2013; Faucillion and Larsson , 2015 ) . Some combination of mechanisms similar to those used in yeast and mammals might also function in C . elegans to compensate for dose-sensitive genes on X . Indeed , if loss of genes from sex chromosomes were gradual , the genome would have had the opportunity to respond in a gene-by-gene fashion to compensate for the reduced X gene dose , thereby rendering a global mechanism of X-chromosome upregulation unnecessary . All C . elegans strains were derived from the Bristol N2 variant and were maintained as described in ( Brenner , 1974 ) . Supplementary file 1 contains a complete list of strains used in this study . Supplementary file 2 contains a complete list of oligos used in this study . Strains were constructed and transgene copy number was analyzed as in ( Frøkjær-Jensen et al . , 2008 , 2014 ) . The hallmark of multi-copy transgenes and transgenes resulting from imprecise insertion events is the incorporation of DNA from the cloning vector backbone into the worm genome . Therefore to obtain strains with single-copy transgenes , we performed PCR with primers specific to the vector backbone to identify and eliminate strains that carried complex transgene insertion events . Each 50 ml LB culture was supplemented with ampicillin ( 10 μg/ml ) and inoculated with Ahringer feeding library bacteria bearing an sdc-2 plasmid or , as a control , a plasmid with no insert ( Kamath et al . , 2001 ) . Cultures were grown at 37°C for 12–16 hr and concentrated 10-fold . RNAi plates ( 1 mM IPTG , 25 μg/ml carbenicillin ) were inoculated with 200 μl concentrated bacteria and incubated at 25°C for 24 hr . Gravid hermaphrodites of the appropriate genotype were bleached , and 800 embryos were plated on each control and sdc-2 ( RNAi ) plate and incubated at 20°C . After 4 days , RNAi plates were washed twice with 5 ml M9 ( first for 3 min , then for 1 min ) to remove all hatched worms but retain embryos . Plates were then returned to 20°C for three hours to permit embryos to hatch . The hatching synchronized L1s were collected by washing the plate with 5 ml M9 . To remove any embryos that may have become dislodged from the plate during the final wash , L1s were concentrated and contaminating embryos were removed by mouth pipetting while viewing the animals with a dissecting microscope . L1s were frozen on liquid nitrogen and stored at -80°C . For each transgene , data were collected from at least three independent biological replicates . To isolate males bearing a transgene of interest , approximately 200 L4 hermaphrodites of genotype him-8 ( e1489 ) IV; unc-58 ( e655dm ) X were mated for 2 days at 20°C with approximately 200 males carrying the appropriate transgene . Since unc-58 ( e655dm ) causes a dominant paralyzed phenotype , and him-8 ( e1489 ) hermaphrodites produce nullo-X oocytes due to X-chromosome non-disjunction , only transgene-bearing XO male cross progeny that inherited a paternal X chromosome will be mobile due to the lack of the unc-58 mutation . To collect these mobile males , embryos were bleached and spotted on the empty half of a plate that had OP50 bacteria cultured on the other half . After 18 hr , L1/L2 male cross-progeny were isolated by cutting the plate in half and collecting only the larvae that had crawled onto the bacteria . To collect stage-matched homozygous hermaphrodite controls , embryos were collected by bleaching non-mated gravid hermaphrodites and isolated as above . Larvae were frozen on liquid nitrogen and stored at -80°C . RNA isolation was conducted as described in ( Baugh et al . , 2003 ) and at www . mcb . harvard . edu/hunter with two modifications: 1 ml of TriZOL reagent ( Life Technologies [Carlsbad , CA] 15596–026 ) was used instead of 300 μl , and the pellets were resuspended in 12 μl of nuclease-free water . cDNA was made from 6 μl RNA for the L1 hermaphrodites ( 10 μl RNA for the L1/L2 hermaphrodites and males ) using the QuantiTect Kit ( Qiagen [Hilden , Germany] 205313 ) . A no-reverse-transcriptase control was generated using 2 μl RNA . Gene expression was analyzed using SYBR green ( BioRad [Hercules , CA] iQ SYBR Green Supermix 170–8886 ) on a BioRad CFX384 Real-Time System . Standard curves were generated from genomic DNA and expression levels were determined from the appropriate standard curve by CFX detection software . Plus and minus reverse-transcriptase reactions were diluted 1/6 for L1 hermaphrodites and 1/2 for L1/L2 hermaphrodites and males , and 2 μl diluted cDNA was used as template in each 10 μl qPCR reaction . Each cDNA sample was quantified in triplicate with normalization and query primer sets . To select normalization genes that are stably expressed in control RNAi and sdc-2 ( RNAi ) conditions , we required that normalization gene candidates meet two conditions . First , the gene must be located on an autosome . Second , the normalization genes must not be significantly different in sdc-2 ( y93 , RNAi ) embryos as assessed by RNA-seq , GRO-seq , or microarray expression analysis ( Jans et al . , 2009; Kruesi et al . , 2013; Crane et al . , 2015 ) . For the 12 candidate normalization genes that met the first two criteria , we used the geNorm approach ( Vandesompele et al . , 2002 ) to narrow normalization genes to those that are the most stably expressed in L1s , as isolated above . This approach indicated that cdc-42 , H06O01 . 1 , and Y38A10A . 5 were the best normalization genes . gfp , Cbr-unc-119 , and tdTomato expression levels were normalized to the geometric mean of three normalization primers . To quantify changes in gene expression , we compared the average of at least three biological replicates of control or sdc-2 ( RNAi ) animals . Error bars represent the standard error of the mean and p values were generated using a Student's t-test ( Graphpad prism ) . To quantify average transcript levels on X and autosomes we used RNA-seq data generated in ( Crane et al . , 2015 ) and similar protocols for analysis . Libraries were sequenced with Illumina’s HiSeq 2000 platform . Reads were required to have passed the CASAVA 1 . 8 quality filtering to be considered further . To remove and trim reads containing the sequencing barcodes , we used cutadapt version 0 . 9 . 5 ( https://cutadapt . readthedocs . org/ ) ( Martin , 2011 ) . Reads were aligned to the WS220 transcriptome using GSNAP version 2012-01-11 ( Wu and Nacu , 2010 ) . Uniquely mapping reads were assigned to genes using HTSeq version 0 . 5 . 4p3 using the union mode ( Anders et al . , 2015 ) . We used DESeq to calculate normalization factors and for significance testing ( Anders and Huber , 2010 ) . To calculate expression level of individual genes , the normalized expression values from DESeq were divided by gene length ( kb ) . Gene expression boxplots were generated using either all genes with a normalized expression level greater than zero or genes within the top 90% of expressed genes . Raw reads from RNA-seq experiments in L4 animals with and without germlines were downloaded from the NIH short read archive ( Deng et al . , 2011 ) . For the N2 transcriptome , we used the files SRR023579 . sra , SRR023580 . sra , and SRR023581 . sra . For the glp-1 ( q224 ) germlineless transcriptomes , we used the files SRR031122 . sra and SRR031123 . sra . Reads were processed using the approach above . Wild-type N2 animals were grown on NG agar plates with HB101 bacteria . Mixed-stage embryos were harvested from gravid hermaphrodites , and cross-linked with 2% formaldehyde for 10 min . Cross-linked embryos were resuspended in 1 ml of FA buffer ( 150 mM NaCl , 50 mM HEPES-KOH ( pH 7 . 6 ) , 1 mM EDTA , 1% Triton X-100 , 0 . 1% sodium deoxycholate , 5 mM DTT , protease inhibitor cocktail , 1 mM PMSF ) for every 1 g of embryos . This mixture was frozen on liquid nitrogen , then ground under liquid nitrogen by mortar and pestle until few intact embryos were visible with a dissecting microscope . Chromatin was sheared by the Covaris S2 sonicator ( 20% duty factor , power level 8 , 200 cycles per burst ) for a total of 30 min processing time ( 60 s ON , 45 s OFF , 30 cycles ) . To perform the ChIP reactions , extract containing approximately 40 μg of DNA was incubated in a microfuge tube with 6 . 6 μg of anti-DPY-27 , anti-SDC-3 or random IgG antibodies overnight at 4°C . A 50 μl bed volume of protein A Sepharose beads was added to the ChIP for 4 hr . ChIPs were washed for 5 min at room temperature twice with FA buffer ( 150 mM NaCl ) , once with FA buffer ( 1 M NaCl ) , once with FA buffer ( 500 mM NaCl ) , once with TEL buffer ( 10 mM Tris-HCl ( pH 8 . 0 ) , 250 mM LiCl , 1% NP-40 , 1% sodium deoxycholate , 1 mM EDTA ) , and twice with TE buffer ( 10 mM Tris pH8 , 1 mM EDTA ) . Protein and DNA were eluted twice with 1% SDS , 250 mM NaCl , 1 mM EDTA at 65°C for 15 min . After reversing crosslinks overnight at 65°C , and treatment with proteinase K and RNAse A , DNA was isolated using the Qiagen PCR purification kit . Using quantitative PCR , DCC enrichment was calculated at a query locus relative to a non-DCC bound autosomal control and is plotted relative to input DNA . To assess the nuclear positioning of the ectopic rex insertions oxSi239 and oxSi246 on autosomes , FISH probes were made using the following fosmids ( BioScience LifeSciences , Nottingham , United Kingdom ) that are adjacent to or flanking the transgene insertion sites: fosmid WRM062dG04 for oxSi239 at chromosome I: 2 , 813 , 818–2 , 846 , 292 Mb and fosmid WRM0633bA08 for oxSi246 at chromosome IV: 5 , 005 , 044–5 , 0426 , 76 Mb . The probes were labeled with Alexa-594 using FISH Tag DNA Kit ( Invitrogen ) . To assess the nuclear positioning of endogenous rex sites in older embryos , FISH probes were made from PCR products corresponding to ~5 kb regions surrounding rex-32 , rex-23 , rex-47 or rex-8 . The primers used are as follows: rex-23 F ( gcccattcaacccattgtcc ) ; rex-23 R ( gcactcgcatattccaaaacg ) ; rex-32 F ( cgcagctggccgttaaatg ) ; rex-32 R ( cattgcaggtgcgttcacaac ) ; rex-47 F ( ccgaaacacaacaacaatgc ) ; rex-47 R ( agactggcgaagaggaacaa ) ; rex-8 F ( tgtgatgcaagccagagttgg ) ; rex-8 R ( cattgagccgaatttccaaagg ) . For younger embryos , the FISH probes were made from 30 kb fosmid probes as follows: rex-47 , WRM0631aB04; rex-23 , WRM0626cG08; rex-32 , WRM0638aF07 . To assess the nuclear position of the endogenous rex-33 site in both younger and older embryos , FISH probes were made to two overlapping fosmids ( WRM0615aA09 and WRM063cD06 ) that were labeled with different fluorescent dyes . Quantification was performed only on FISH spots that had signals from both fosmids to unambiguously identify bona fide FISH spots . C . elegans embryos were obtained for Figure 7B–C by dissecting three different strains of mated gravid adult hermaphrodites: wild-type ( N2 ) , TY5726 , and EG6136 . Older C . elegans embryos were obtained for Figure 7D and Figure 7—figure supplement 1B by dissecting gravid adult hermaphrodites from strain CB1489 him-8 ( e1489 ) . Younger embryos were obtained by dissecting gravid adult wild-type hermaphrodites that had been mated with wild-type males . FISH was performed on the embryos as described previously ( Crane et al . , 2015 ) . Following FISH , immunostaining with rabbit DPY-27 antibody ( rb699 ) ( Chuang et al . , 1994 ) and Alexa-Fluor-647 goat anti-rat antibody ( Invitrogen ) was performed to determine the sex of embryos for experiments involving ectopic rex sites on autosomes and older embryos for endogenous rex sites on X . Embryos were determined to be XX if they exhibited punctate DPY-27 staining on both X chromosomes and XO if they lacked DPY-27 staining . Sex was determined for all embryos examining rex-33 localization and all younger embryos examining localization of endogenous rex sites on X by counting the number of FISH spots in the nuclei . Embryos were determined to be XX if the nuclei had two FISH spots and XO if nuclei had one FISH spot . The age of embryos was determined by counting the number of DAPI-stained nuclei using the FindPoints function in Priism software ( Chen et al . , 1996 ) . Confocal image stacks with a 51 . 5 nm XY pixel size and an 83 . 9 nm Z-spacing were obtained on a Leica TCS SP8 microscope using a 63× , 1 . 4 NA objective lens . Our Z-spacing was smaller than that used in ( Sharma et al . , 2014 ) , making our point picking more precise . Their XY pixel size was not available for comparison . Image deconvolution with a theoretical point spread function was performed using Huygens Professional Software ( Scientific Volume Imaging , The Netherlands ) . FISH spots were identified using the FindPoints function . To unambiguously select bona fide FISH spots , we excluded nuclei that showed more than one or two spots for XO and XX embryos , respectively . The nuclear positioning of the FISH signals was measured using a previously described approach ( Meister et al . , 2010 ) , but with the Pick Points function in Priism software . Spots at the top or bottom of nuclei were excluded .	DNA inside cells is packaged into structures called chromosomes , each of which contains numerous genes . Many organisms , including humans , have two copies of most chromosomes in their cells . If the process of cell division goes awry , cells can end up with too many or too few copies of their chromosomes , which can cause serious illnesses . Sex chromosomes pose a conundrum for cells . In humans , females have two copies of the X chromosome , whereas males only have one . This means that males have half the copy number ( dose ) of genes on the X chromosome . Human cells correct this imbalance by suppressing the activity , or expression , of most of the genes on one of the X chromosomes in females . “Dosage compensation” also occurs in the roundworm species Caenorhabditis elegans , because male worms have one X chromosome whilst hermaphrodites have two . The dosage compensation mechanism in roundworms differs from that in humans . It involves turning down the expression of both hermaphrodite X chromosomes by half . The process is enacted by a dosage compensation complex that binds to specific sites along both hermaphrodite X chromosomes . Dosage compensation mechanisms that reduce X chromosome expression in females cause sex chromosomes to have lower gene expression than non-sex chromosomes . Modern sex chromosomes evolved from a pair of non-sex chromosomes , and males lost one copy of all of the genes located on those ancestral chromosomes . This evolutionary history causes both sexes to have lower gene expression from X chromosomes than the other chromosomes , raising the question of whether a mechanism exists to balance out the difference in gene expression between sex chromosomes and non-sex chromosomes . Wheeler et al . now show that the expression of any foreign gene artificially added to the X chromosomes of C . elegans is equalized between males and hermaphrodites despite the difference in gene dose . The equalization works regardless of where on the X chromosome the new gene is added . The foreign gene does not need to be adjacent to a binding site for the dosage compensation complex . These results indicate that dosage compensation mechanisms regulate gene expression on a chromosome-wide scale . Wheeler et al . also show that genes added to X chromosomes are expressed at half the level as the same genes added to non-sex chromosomes . These results mean that no chromosome-wide mechanism balances gene expression levels between the X chromosome and the non-sex chromosomes . It remains unknown how C . elegans , and many other living organisms , evolved to tolerate a lower level of gene expression from the sex chromosomes . Instead of a chromosome-wide mechanism , it is likely that individual genes evolved different ways to alter their expression levels . Working out what these mechanisms are remains a challenge for further research .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2016	Chromosome-wide mechanisms to decouple gene expression from gene dose during sex-chromosome evolution
Recent breakthroughs in 3-dimensional ( 3D ) organoid cultures for many organ systems have led to new physiologically complex in vitro models to study human development and disease . Here , we report the step-wise differentiation of human pluripotent stem cells ( hPSCs ) ( embryonic and induced ) into lung organoids . By manipulating developmental signaling pathways hPSCs generate ventral-anterior foregut spheroids , which are then expanded into human lung organoids ( HLOs ) . HLOs consist of epithelial and mesenchymal compartments of the lung , organized with structural features similar to the native lung . HLOs possess upper airway-like epithelium with basal cells and immature ciliated cells surrounded by smooth muscle and myofibroblasts as well as an alveolar-like domain with appropriate cell types . Using RNA-sequencing , we show that HLOs are remarkably similar to human fetal lung based on global transcriptional profiles , suggesting that HLOs are an excellent model to study human lung development , maturation and disease . Several reports have demonstrated that directed differentiation of human pluripotent stem cells ( hPSCs ) , which include embryonic ( hESCs ) and induced ( iPSCs ) stem cells , is one of the most efficient approaches to achieving differentiation of a cell or tissue of interest ( D'Amour et al . , 2005; Kroon et al . , 2008; Si-Tayeb et al . , 2009; Spence et al . , 2011; Wong et al . , 2012 ) . Using this approach , differentiation of hPSCs into lung lineages has been achieved using diverse methodology with varying degrees of success ( Kadzik and Morrisey , 2012; Longmire et al . , 2012; Mou et al . , 2012; Wong et al . , 2012; Ghaedi et al . , 2013; Huang et al . , 2013; Firth et al . , 2014 ) . Thus far , the majority of efforts to differentiate lung lineages from hPSCs have focused on using 2-dimensional ( 2D ) monolayer cultures . Several recent advances in generating 3-dimensional ( 3D ) organ-like tissues , called ‘organoids’ , have been reported ( Meyer et al . , 2011; Spence et al . , 2011; Nakano et al . , 2012; Takebe et al . , 2013; Lancaster et al . , 2013; McCracken et al . , 2014 ) . Such 3D models offer several advantages; they often possess structural organization similar to the native organ , cell types from multiple germ layers ( for example , mesoderm and endoderm ( Spence et al . , 2011; McCracken et al . , 2014; Wells and Spence , 2014 ) , and multiple cellular lineages , making them a physiologically complex model to study developmental processes , tissue homeostasis and pathological conditions in vitro . Previous work has demonstrated that activation of FGF and WNT signaling synergistically drives CDX2+ intestinal lineage commitment in hPSC-derived endoderm and also drives ‘morphogenesis in a dish’ , where the 2D tissues self-organize into 3D spheroids comprised of mesenchymal and polarized epithelial layers that detach from the adherent cell layer ( Spence et al . , 2011 ) . It has also been demonstrated that inhibition of BMP and TGFβ signaling is able to drive tissue into a SOX2+ foregut lineage ( Green et al . , 2011; McCracken et al . , 2014 ) . Building on these previous studies , we show that simultaneous stimulation of WNT and FGF signaling while inhibiting BMP/TGFβ signaling pathways in hPSC-derived endoderm cultures prevents intestinal lineage commitment , and instead , favors a SOX2+ anterior foregut fate while also robustly generating SOX2+ anterior foregut 3D spheroid structures . In order to further restrict foregut spheroids to the lung lineage , the current study focused on manipulating FGF and HH signaling . In the mouse , high levels of Fgf signaling have been shown to induce Shh expression in the lung endoderm ( Hebrok et al . , 1998; Morrisey and Hogan , 2010; Rankin and Zorn , 2014 ) which is accompanied by induction of the Nkx2 . 1+ lung progenitor field ( Hebrok et al . , 1998; Serls , 2004 ) . Shh then signals from the endoderm to the mesoderm , and mutations in Shh , Gli2 or Gli3 lead to perturbed lung development , with Gli2/Gli3 double knockout mice showing lung agenesis ( Bellusci et al . , 1997a; Motoyama et al . , 1998; Li et al . , 2004 ) . Our results demonstrate that FGF2 induces NKX2 . 1 , PAX8 , and SHH in human foregut endoderm cultures . By using pharmacological inhibitors of FGF and HH signaling we demonstrate that SHH is required for NKX2 . 1 expression downstream of FGF2 , and that FGF2 also induces PAX8 independently of HH signaling . These observations suggest a paradigm where FGFLo/HHHi conditions preferentially induce PAX8Lo/NKX2 . 1Hi lung progenitors and FGFHi/HHLo conditions favor a PAX8Hi/NKX2 . 1Lo fate . Given that Pax8 is required for thyroid development , we focused on defining the most robust conditions to induce NKX2 . 1 while minimizing PAX8 expression ( Kimura et al . , 1996; Mansouri et al . , 1998; Yuan et al . , 2000; Vilain et al . , 2001; Li et al . , 2004; Kusakabe et al . , 2006; Carré et al . , 2009; Narumi et al . , 2012 ) . By applying HHHi conditions during generation of foregut spheroids we were able to enhance NKX2 . 1 expression in foregut spheroids and subsequently expand spheroids in media containing FGF10 , allowing them to grow into organoids . Organoids persisted in culture for over 100 days and developed well-organized proximal-like airway epithelial structures that included many cell types found in the proximal lung epithelium , including basal and ciliated cells along with rare club cells . Moreover , proximal airway structures were often surrounded by smooth muscle actin ( SMA ) positive mesenchymal tissue . Organoids also possessed distal-like epithelial cells that co-expressed progenitor markers , SFTPC/SOX9 and HOPX/SOX9 , consistent with early bipotent alveolar progenitor cells seen in mice ( Desai et al . , 2014; Treutlein et al . , 2014 ) . To support the idea that organoids may be more similar to a developing lung with abundant progenitor cells , we used RNA-sequencing to compare the global transcriptional profile of organoids to the human fetal and adult lung , undifferentiated hESCs and definitive endoderm . Principal component analysis , hierarchical clustering and Spearman's correlation all show that organoids have striking similarity to the human fetal lung . Taken together , our data demonstrates an efficient and robust in vitro system to generate complex , 3D human lung organoids that are immature/fetal in nature . We anticipate that this model will serve as an unparalleled model for the study of human lung development , maturation and disease . We and others have reported efficient induction of human endoderm using ActivinA ( D'Amour et al . , 2005; Zhang et al . , 2010; Spence et al . , 2011 ) , and a further lineage restriction into SOX2+ anterior foregut endoderm using inhibition of BMP and TGFβ signaling ( Green et al . , 2011; Loh et al . , 2014 ) . We have recently demonstrated that inhibition of BMP signaling during intestinal lineage induction with WNT and FGF ligands is sufficient to inhibit intestinal CDX2 and induce SOX2+ posterior foregut spheroids capable of giving rise to human gastric ( antral ) organoids ( McCracken et al . , 2014 ) . Given that the lung is derived from the anterior foregut , we sought to define conditions to generate ventral anterior foregut spheroids . To do this , we tested if dual inhibition of BMP and TGFβ was able to anteriorize cultures , as previously described ( Green et al . , 2011 ) . We treated hESCs with ActivinA ( 100 ng/ml ) for 4 days to induce endoderm , followed by 4 days of Noggin ( NOG , 200 ng/ml ) and the small molecule TGFβ inhibitor , SB431542 ( SB , 10 µM ) . We confirmed that these conditions were able to induce robust mRNA and protein expression of SOX2 , which co-expressed the endodermal marker FOXA2 , while repressing the intestinal lineage marker CDX2 ( Figure 1A–C , Figure 1—figure supplement 1A ) . QRT-PCR analysis also showed that compared to controls ( in which endoderm was induced but was not exposed to NOG/SB ) , exposure to NOG/SB robustly induced ventral anterior foregut genes NKX2 . 1 and PAX8 , while the posterior foregut transcript , PDX1 was reduced . HHEX , which is expressed in the developing liver , biliary system and thyroid , but is absent from the lung primordium , remained unchanged ( Figure 1B ) . Given that NKX2 . 1 is expressed in the lung and thyroid primordium , and PAX8 is expressed in the thyroid primordium , these results suggest that 4 day ActivinA treatment followed by a 4 day NOG/SB treatment biases the cultures towards ventral-anterior foregut lineages . 10 . 7554/eLife . 05098 . 003Figure 1 . Generation of three-dimensional ventral anterior foregut spheroids from endoderm monolayers . ( A ) hESCs were differentiated into foregut endoderm by treating cells with 4 days of Activin A ( ACTA ) followed by 4 days of NOG+SB . ( B ) Foregut endoderm ( NOG+SB ) had high expression of the foregut marker SOX2 while the hindgut marker CDX2 was significantly reduced compared to untreated endoderm controls ( End ) . NOG+SB monolayers had high expression of ventral anterior foregut genes NKX2 . 1 and PAX8 while the posterior foregut marker PDX1 was reduced . The foregut marker HHEX is expressed in the developing liver , biliary system , and thyroid and remained unchanged . ( C ) The majority of cells in NOG+SB treated cultures were SOX2 positive ( green ) compared to the control , in which only scattered clusters of cells were SOX2 positive . The scale bar represents 200 µm . ( D ) hESCs were differentiated into foregut spheroids by treating cells with 4 days of ACTA and then additional 4–6 days of NOG+SB+FGF4+Ch . Representative images of a spheroid in a matrigel droplet are shown as a whole mount image . Scale bar represents 100 µm . ( E ) Foregut spheroids ( NOG+SB+FGF4+Ch ) had high expression of the foregut marker SOX2 while the hindgut marker CDX2 was significantly reduced compared to untreated endoderm control ( End ) ( top panel ) . Spheroids had high expression of anterior foregut genes NKX2 . 1 and PAX8 while the posterior foregut marker PDX1 was reduced and HHEX was unchanged ( bottom panel ) . p < 0 . 05 , error bars represent SEM . ( F ) The majority of cells in foregut spheroids are FOXA2+ ( green , left panel ) and SOX2+ ( white , right panel ) and ECAD+ ( red , right panel ) . Scale bar represent 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 00310 . 7554/eLife . 05098 . 004Figure 1—figure supplement 1 . Monolayer cultures express lung specific markers . Immunohistochemistry for markers expressed in endoderm , ventral foregut or lung epithelium were assessed ( SOX2 , FOXA2 , NKX2 . 1 , SOX9 ) in endoderm controls , foregut controls or foregut cultures treated with SAG or SAG+SU . ( A ) All conditions express endoderm marker FOXA2 ( red ) , but the foregut ( NOG+SB ) control , SAG and SAG+SU treated cultures have co-expression of FOXA2 ( red ) and SOX2 ( green ) in the majority of cells . ( B ) All conditions expressed endoderm marker FOXA2 ( red ) , but only foregut endoderm treated with SAG and SAG+SU have robust NKX2 . 1+ cells ( green ) that also express FOXA2 ( red ) . ( A–B ) Scale bars represent 200 µm and apply to all images . ( C ) Only foregut endoderm treated with SAG and SAG+SU have robust NKX2 . 1+ cells ( green ) with the majority of cell co-expressing with lung epithelial marker SOX9 ( red ) . Scale bar represents 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 00410 . 7554/eLife . 05098 . 005Figure 1—figure supplement 2 . Foregut spheroids co-express endoderm and lung specific markers . ( A ) NOG/SB/FGF4/Ch spheroids have weak NKX2 . 1 ( green ) expression which co-expresses with endoderm marker FOXA2 ( red ) . ( B ) The majority of cells in the spheroid express SOX2 ( green ) and co-stain with FOXA2 ( red ) . Scale bars represent 50 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 00510 . 7554/eLife . 05098 . 006Figure 1—figure supplement 3 . Foregut spheroids consist of both epithelial and mesenchymal cells . NOG/SB/FGF4/Ch spheroids have a minor population of Vimentin ( VIM , white ) positive mesenchymal cells , while the majority of cells are epithelial and express ECAD ( red ) . Scale bar represents 50 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 00610 . 7554/eLife . 05098 . 007Figure 1—figure supplement 4 . NOG+SB+FGF4+Ch spheroids do not express neural markers . hESCs were differentiated into endoderm by treating with 4 days of ActivinA ( ACTA ) and spheroids were generated with an additional 4 days of NOG+SB+FGF4+Ch . Neural cultures were not treated with ACTA , but were treated with NOG+SB for 8 days . Compared to foregut spheroids ( NOG+SB+FGF4+Ch ) , NOG+SB neural cultures had a significant increase in neural markers NESTIN , SOX1 , and PAX6 and significant decrease in FOXA2 , which is highly expressed in endoderm . p < 0 . 05 , error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 007 Addition of FGF4 plus WNT3A ( or Chir99021 , a GSK3β inhibitor that enhances β-catenin dependent WNT signaling ) promotes CDX2 intestinal lineage commitment and 3D spheroid formation in endoderm cultures ( Spence et al . , 2011; Xue et al . , 2013; Chen et al . , 2014b ) . Based on our results in Figure 1B–C , we hypothesized that combining FGF , Chir99021 , NOG and SB would result in the generation of SOX2+ ventral-anterior foregut spheroids . To test this , we generated endoderm ( 4 days ACTA ) and added no growth factors ( Endoderm controls ) or NOG , SB , FGF4 , and Chir99021 ( NOG/SB/F/Ch ) ( Figure 1D ) . Addition of all four factors resulted in the generation of 3-dimensional SOX2+ , CDX2− spheroids ( Figure 1E , F ) . SOX2+ spheroids also expressed the endodermal protein FOXA2 , and were epithelial , co-expressing E-Cadherin ( ECAD ) ( Figure 1F , Figure 1—figure supplement 2 ) . In addition to SOX2 , spheroids exhibited higher mRNA expression of anterior foregut lineage markers NKX2 . 1 and PAX8 compared to endoderm controls , suggesting that they are ventral-anterior foregut spheroids ( Figure 1E ) , however , immunofluorescence revealed that levels of NKX2 . 1 protein were just above the detection threshold ( Figure 1—figure supplement 2 ) . Spheroids also possess a minor population of cells that are mesodermal in origin staining positive for Vimentin protein ( VIM ) ( Figure 1—figure supplement 3 ) . Given that neural tissues also express NKX2 . 1 , PAX8 , SOX2 , and FOXA2 , and that neural induction protocols use dual BMP and TGFβ inhibition , we wanted to exclude the possibility that spheroids were neural in nature . To do this , we generated endoderm control cultures , foregut spheroids ( ActivinA followed by NOG/SB/F/Ch ) , and induced neural tissue by adding NOG/SB to hESC cultures that were not treated with ActivinA ( Chambers et al . , 2009 ) . By examining induction of the neural markers NESTIN , SOX1 , and PAX6 , we confirmed that these transcripts were highly induced in dual NOG/SB neural cultures , but were low in ventral foregut spheroid cultures . In contrast , FOXA2 , which is expressed in the foregut ( Monaghan et al . , 1993; Ang and Rossant , 1994; D'Amour et al . , 2005 , 2006; Kroon et al . , 2008; Si-Tayeb et al . , 2009; DeLaForest et al . , 2011 ) and in some neural tissues ( Stott et al . , 2013 ) , had high expression in ventral foregut spheroids , but was significantly reduced in dual NOG/SB neural conditions ( Figure 1—figure supplement 4 ) . Taken together , these results strongly suggest spheroids are indeed foregut , and not of neural origin . Many signaling pathways are important for lung induction and development ( reviewed in Min et al . , 1998; Weaver et al . , 2000; Morrisey and Hogan , 2010; Rankin and Zorn , 2014 ) . High levels of Fgf signaling have been shown to induce Shh and Nkx2 . 1 expression in the foregut endoderm in mice ( Hebrok et al . , 1998; Serls , 2004 ) ; furthermore , Gli2/3 null mouse embryos fail to form lungs ( Motoyama et al . , 1998 ) and Hh signaling is important for lung mesenchyme proliferation in vivo ( Bellusci et al . , 1997a ) . These data confirm that Fgf and Hh signaling are critical for lung specification and ligands from both signaling pathways have been applied to hPSC derived lung lineages in 2D cultures ( Wong et al . , 2012; Huang et al . , 2013 ) . In our cultures we have reported that approximately 85–95% of cells are endoderm , but a portion of the remaining cells are mesodermal and this small mesodermal population is maintained in the spheroids and organoids ( Spence et al . , 2011; McCracken et al . , 2014 ) ( Figure 1—figure supplement 3 ) . Therefore , based on mouse and hPSC studies , we hypothesized that FGF and/or HH signaling would induce an NKX2 . 1+ lung lineage in anterior foregut endoderm . To test our hypothesis we initially focused on adherent endoderm monolayer cultures to optimize induction conditions . Cultures were treated for 4 days with ActivinA followed by an additional 4 days with NOG/SB ( referred to as Foregut ) . Controls consisted of ActivinA treatment only followed by no additional growth factors ( Endoderm controls ) , or ActivinA followed by NOG/SB , followed by no additional factors ( Foregut controls ) . All experimental groups were compared to both endoderm and foregut controls ( Figure 2 ) . We first tested the ability of FGF2 to induce SHH , NKX2 . 1 and PAX8 by exposing foregut cultures to low and high concentrations of FGF2 ( 50 , 500 ng/ml ) ( Figure 2A ) . We observed a robust concentration dependent increase in SHH and PAX8 mRNA expression compared to foregut or endoderm controls , and a modest increase of NKX2 . 1 expression at the highest dose of FGF2 ( 500 ng/ml ) ( Figure 2A ) . We also observed that dual NOG/SB inhibition in endoderm cultures induced robust NKX2 . 1 and PAX8 expression without adding FGF2 ( Figures 1B , 2A ) . Thus , we wanted to determine if NKX2 . 1 expression in foregut cultures was due to endogenous FGF and/or HH signaling . To test this , we inhibited the FGF or HH pathway with small molecules SU5402 ( SU , 10 µm ) and Sant-2 ( 10 µm ) respectively ( Figure 2B–C ) . Treating foregut cultures with the FGF inhibitor SU caused a significant , robust reduction in PAX8 and a modest reduction in SHH , while NKX2 . 1 expression was unchanged compared to foregut controls ( Figure 2B ) . Conversely , inhibition of HH signaling caused a significant reduction in NKX2 . 1 expression , but not PAX8 compared to untreated foregut . When FGF2 was added to the cultures , we observed a modest increase in NKX2 . 1 expression , and when FGF was added along with Sant-2 , NKX2 . 1 expression was significantly reduced ( Figure 2C ) . Together our results suggest a hierarchy where FGF is upstream of SHH and PAX8 , and where SHH is upstream of NKX2 . 1 . To test if HH signaling was able to induce NKX2 . 1 in foregut cultures , we added the Smoothened agonist , SAG ( 1 µM ) to foregut cultures . The addition of SAG induced a 6 . 5-fold increase of NKX2 . 1 expression above foregut controls ( Figure 2D ) . However , SAG alone did not reduce PAX8 expression . 10 . 7554/eLife . 05098 . 008Figure 2 . Induction of NKX2 . 1 in anterior foregut endoderm by modulating FGF and HH signaling . ( A ) hESCs were differentiated into endoderm ( End ) or anterior foregut with NOG+SB ( For ) . Anterior foregut was treated with low ( 50 ng/ml ) and high ( 500 ng/ml ) concentrations of FGF2 . FGF2 caused a dose-dependent increase in SHH and PAX8 expression with a modest increase in NKX2 . 1 expression compared to untreated endoderm controls . Note that NKX2 . 1 expression is increased by NOG+SB exposure alone ( no FGF2 ) . ( B ) Addition of the FGF inhibitor SU5402 ( SU ) to NOG+SB foregut cultures ( For ) caused a significant reduction of SHH and PAX8 expression , but NKX2 . 1 , GLI1 , and PTCH1 were not significantly different compared to the foregut controls , in which no growth factors were added after SB+NOG . ( C ) Addition of the HH inhibitor Sant-2 caused a significant reduction in NKX2 . 1 compared to foregut control . Similarly when FGF2 ( 500 ng/ml ) and Sant-2 were added simultaneously , the modest NKX2 . 1 induction caused by FGF2 was significantly reduced whereas PAX8 expression remained unchanged . ( D ) Foregut endoderm treated with SAG or SAG+SU for 8 days had a 6 . 5-fold and 21-fold increase of NKX2 . 1 expression , respectively , compared to untreated foregut controls . PAX8 expression was unchanged in the SAG treated cultures whereas SAG+SU treated cultures demonstrated a 41-fold decrease in PAX8 expression . End = endoderm; For = foregut in all panels . p < 0 . 05 , error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 00810 . 7554/eLife . 05098 . 009Figure 2—figure supplement 1 . Robust induction of NKX2 . 1 in foregut endoderm with HH stimulation and FGF inhibition . ( A ) Immunohistochemistry of NKX2 . 1 and PAX8 in endoderm controls , foregut controls or foregut cultures treated with SAG or SAG+SU . Treatment of foregut cultures with SAG or SAG+SU resulted in more NKX2 . 1+ cells compared to endoderm and foregut controls . Scale bars represent 200 µm and apply to all images . ( B ) Quantification showed that 20% ± 4% of cells in foregut controls were NKX2 . 1+ , whereas 72% ± 3% cells were positive in SAG+SU treated cultures ( p < 0 . 05 ) . All error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 009 Based on these results , we further hypothesized that enhancing HH signaling would result in increased NKX2 . 1 expression downstream of FGF , and that simultaneous inhibition of FGF signaling would reduce PAX8 expression; therefore , we inhibited endogenous FGF signaling with SU while activating HH with SAG ( Figure 2D ) . This combination caused an additional increase in NKX2 . 1 expression ( 21-fold vs 6 . 5-fold with SAG only , when compared to foregut ) and a concomitant decrease in PAX8 mRNA ( Figure 2D ) . Importantly , immunofluorescence was correlated with QRT-PCR data showing an increased number of NKX2 . 1+ cells with the addition of SAG only . SAG+SU treated cultures showed a further increase in the number of NKX2 . 1 expressing cells , with ∼77% of all cells expressing NKX2 . 1 compared to ∼20% in foregut controls , and almost no PAX8 expressing cells ( Figure 2—figure supplement 1 ) . SAG and SAG+SU treated cells also co-expressed FOXA2 and SOX2 confirming their endodermal origin ( Figure 1—figure supplement 1 ) . Based on the observations that stimulating HH and inhibiting FGF signaling strongly enhances NKX2 . 1 expression while reducing PAX8 expression ( Figure 2 ) , we tested multiple conditions of HH activation and FGF inhibition to induce NKX2 . 1HI/PAX8LO foregut spheroids ( NOG/SB/F/Ch ) ( Summarized in Figure 3—figure supplement 1 ) . Consistent with the important roles of FGF signaling in lung growth and branching morphogenesis ( Hebrok et al . , 1998; Min et al . , 1998; Weaver et al . , 2000; Abler et al . , 2009; Morrisey and Hogan , 2010; Rankin and Zorn , 2014 ) , we found that conditions where FGF inhibition was used led to a reduction of epithelial tissue relative to mesenchymal tissue , which could be due to a loss of epithelium or an overgrowth of mesenchyme; this suggests that endogenous FGF signaling is necessary to maintain the epithelial tissue in 3D cultures ( Figure 3—figure supplement 2 ) . Therefore , we also tested several conditions that stimulated HH signaling using SAG only , without FGF inhibition . We found that the most efficient method to enhance NKX2 . 1 expression was by adding SAG during the foregut spheroid phase ( Figure 3A ) . Comparing foregut spheroids ( NOG/SB/F/Ch ) with those treated with SAG ( NOG/SB/F/Ch/SAG ) , we observed a substantial decrease in SOX2 expression compared to NOG/SB/F/Ch spheroids and a significant increase in NKX2 . 1 mRNA . Additionally , nuclear NKX2 . 1 protein expression was found in ECAD+ epithelium which co-expressed endoderm epithelial markers FOXA2 and SOX2 ( Figure 3B , C , Figure 3—figure supplement 3 ) . Interestingly , during lung specification in mice , the gut tube initially expresses Sox2 throughout the endoderm , but Sox2 is down-regulated in the lung field during lung specification and Nkx2 . 1 induction ( Hebrok et al . , 1998; Serls , 2004; Domyan et al . , 2011 ) . Thus , concomitant down-regulation of SOX2 and increased NKX2 . 1 observed in SAG treated foregut spheroids is consistent with early transcriptional changes that occur during lung specification in mice . 10 . 7554/eLife . 05098 . 010Figure 3 . HH-induced ventral foregut spheroids give rise to lung organoids . ( A ) hESCs were differentiated into foregut spheroids by treating cells with 4 days of ACTA and then an additional 4–6 days of NOG+SB+FGF4+Ch with the addition of the HH agonist SAG . Representative whole mount images of spheroids in a matrigel droplet are shown at low ( left , scale bar 200 µm ) and high magnification ( right , scale bar 100 µm ) . ( B ) The addition of SAG to the NOG+SB+FGF4+Ch spheres caused a reduction in SOX2 and CDX2 transcripts ( top panel ) and a significant increase of NKX2 . 1 transcript ( bottom panel ) compared to NOG+SB+FGF4+Ch spheres ( without SAG ) . Other foregut lineages ( PAX8 , PDX1 , HHEX ) were not significantly different when SAG was added . ( C ) The majority of the cells in NOG+SB+FGF4+Ch+SAG spheres expressed FOXA2 , SOX2 and NKX2 . 1 protein . Scale bars represent 50 µm . ( D ) Timeline showing NOG+SB+FGF4+Ch+SAG induced foregut spheroids grown and maintained in FGF10 . Note that Day 1 is the day spheroids were plated in Matrigel . The scale bar represents 100 µm . ( E ) Organoids express lung markers in a manner consistent with mouse lung development . All expression is shown relative to undifferentiated pluripotent stem cells ( hPSC ) , and adult human lung is shown as a reference . Lung progenitor markers NMYC and ID2 were very low in adult lung , and were expressed at high levels in early organoid cultures , but were reduced over time ( D = Days in culture ) , whereas NKX2 . 1 expression remained relatively constant . In contrast , SFTPC is known to be expressed at low levels in distal lung progenitors , but increases and is highly expressed in AECII cells . Consistently , SFTPC is highly expressed in adult human lungs and increases over time in organoid cultures and the AECI marker HOPX is also highly expressed in adult human lung and increases over time in organoids . p < 0 . 05 . All error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 01010 . 7554/eLife . 05098 . 011Figure 3—figure supplement 1 . Overview of conditions tested to generate human lung organoids . hPSCs are OCT4 and NANOG positive . After 4 days of 100 ng/ml Activin A , definitive endoderm ( FOXA2 and SOX17 positive ) was generated and then treated with two different conditions . In the top branch , NOG+SB+FGF4+Ch spheroids were generated , and different conditions were tested to promote lung organoid differentiation . In the bottom branch , NOG+SB+FGF4+Ch+SAG spheroids were generated , and different conditions were tested to promote lung organoid differentiation . Ultimately , we determined that spheroids generated with NOG+SB+FGF4+Ch+SAG and that were subsequently embedded in Matrigel and expanded in FGF10 gave rise to ‘Human Lung Organoids’ ( HLOs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 01110 . 7554/eLife . 05098 . 012Figure 3—figure supplement 2 . FGF-low culture conditions cause a loss of organoid epithelium over time . ( A ) NOG+SB+F+Ch foregut spheroids were generated and then cultured in SAG+SU for 10 days followed by 1% FBS ± FGF10 . Timeline images show organoids cultured in 1% FBS . By day 20 , 3D structures appeared ‘fuzzy’ , which indicates an outgrowth of mesenchymal tissue . Scale bar represents 200 µm . ( B ) NOG+SB+F+Ch foregut spheroids treated with SAG+SU and maintained in 1% FBS showed an increase in Vimentin ( VIM , green ) immunofluorescence over time . Scale bar represents 50 µM . ( C ) NOG+SB+F+Ch foregut spheroids treated with SAG+SU and maintained in 1% FBS ( upper panel ) or 1% FBS+FGF10 ( lower panel ) had a significant increase of VIM expression starting at day 20 ( D20 ) compared to hPSCs and showed very weak E-CADHERIN ( CDH1 ) expression compared to D20 HLOS ( optimized conditions , as described in Figure 3 ) . Lastly , both conditions appeared to lose NKX2 . 1 expression over time . ( D ) NOG+SB+F+Ch+SAG spheroids maintained in 1% FBS ( basal media ) also appear to lose epithelial structures over time . Scale bar represents 200 µm . ( E ) By day 20 ( D20 ) the tissue had very few epithelial structures expressing ECAD ( white , left panel ) and there was robust VIM expression ( green , right panel ) at both time points . Scale bar represents 50 µm . HLO p < 0 . 05 . All error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 01210 . 7554/eLife . 05098 . 013Figure 3—figure supplement 3 . Foregut spheroids express lung and foregut specific markers . ( A ) NOG/SB/FGF4/Ch/SAG spheroids coexpress NKX2 . 1 ( green ) and the endoderm marker FOXA2 ( red ) . ( B ) The majority of the cells in the spheroid co-expresses SOX2 ( green ) and FOXA2 ( red ) . Scale bars represent 50 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 01310 . 7554/eLife . 05098 . 014Figure 3—figure supplement 4 . Ventral foregut spheroids do not express appreciable levels of PAX8 protein . Although NOG+SB+FGF4+Ch+SAG ventral foregut spheroids expressed PAX8 mRNA ( Figure 3B ) , we did not detect PAX8 protein in spheroids using immunofluorescence , whereas PAX8 protein in FGF2 8 day treated foregut monolayers ( ACTA followed by NOG/SB ) was readily detectable . Left panel: scale bar represents 50 µm . Right panel: scale bar represents 200 µm , inset scale bar represents 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 01410 . 7554/eLife . 05098 . 015Figure 3—figure supplement 5 . Foregut spheroids consist of both epithelial and mesenchymal cells . NOG/SB/FGF4/Ch/SAG spheroids have a minor population of Vimentin ( VIM , white ) positive mesenchymal cells , while the majority of cells are epithelial and express ECAD ( red ) . Scale bar represents 50 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 01510 . 7554/eLife . 05098 . 016Figure 3—figure supplement 6 . Lung organoids contain both proximal and distal domains . NOG/SB/FGF4/Ch/SAG spheroids cultured for 15 days with FGF10 express the distal lung epithelium marker SOX9 ( green ) and proximal marker SOX2 ( white ) as separate domains in the epithelium labeled by ECAD ( red ) . Z-stack images are shown every 40 µm sections through the HLO . Scale bar represents 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 016 We also observed a slight , but non-significant increase in PAX8 transcript level in NOG/SB/F/Ch/SAG treated foregut spheroids ( Figure 3B ) . Importantly , PAX8 protein expression was undetectable in NOG/SB/F/Ch/SAG treated foregut spheroids and expression remained low/undetectable throughout time in culture . ( Figure 3—figure supplement 4 ) . Similar to NOG/SB/F/Ch treated spheroids , the NOG/SB/F/Ch/SAG treated spheroids had a minor population of cells within the spheroids of mesodermal in origin , expressing Vimentin ( VIM ) ( Figure 3—figure supplement 5 ) . NOG/SB/F/Ch/SAG treated foregut spheroids were embedded in Matrigel to provide a 3D growth environment . Spheroids maintained in basal media ( see ‘Materials and methods’ ) supplemented with 1% FBS lost ECAD+ epithelial structures and were mainly comprised of mesenchyme within 20 days of 3D culture ( Figure 3—figure supplement 2D , E ) . FGF10 is essential for branching morphogenesis and maintenance of lung progenitor cells during development as well as tissue homeostasis in the adult lung ( Bellusci et al . , 1997a; Min et al . , 1998; Weaver et al . , 2000; Volckaert et al . , 2013 ) . We observed that the addition of FGF10 ( 500 ng/ml ) allowed spheroids to expand and be passaged for over 100 days . FGF10 promoted the maintenance of ECAD+ epithelial structures with less mesenchymal contributions compared to both basal and FGF inhibitor conditions ( Figure 3D ) . NOG/SB/F/Ch/SAG cultured for 15 days in FGF10 possessed abundant ECAD+ epithelium that expressed the proximal lung marker SOX2 and distal lung marker SOX9 . SOX2+ domains and SOX9+ domains were distributed throughout the entire HLO as determined by whole mount immunofluorescence and confocal Z-sections . ( Figure 3—figure supplement 6 ) . FGF10 treated foregut spheroids maintained NKX2 . 1 expression over time; however , consistent with mouse development , distal progenitor markers , NMYC and ID2 mRNA expression decreased over time while distal Alveolar Type I and II cell markers , HOPX and SFTPC increased over time ( Okubo , 2005; Rawlins et al . , 2009 ) ( Figure 3E ) . These data suggest that HLOs pass through a stage resembling early fetal lung development in mice . HLOs cultured longer than 2 months had striking epithelial structures resembling proximal airways , expressing proximal cell type-specific markers , including basal cells ( P63 ) , ciliated cells ( FOXJ1 , ACTTUB ) and club cells ( SCGB1A1 ) ( Figure 4 ) . Proximal-like airway tissues were often surrounded by a smooth muscle actin positive ( SMA+ ) mesenchyme compartment . Although P63 mRNA expression is maintained throughout culture ( Figure 4A ) , it is only in prolonged cultures ( >2 months ) where the P63+ cells are spatially arranged along the basal side of the epithelial tube-like structures , adjacent to SMA+ mesenchyme , similar to human bronchi and bronchioles ( Figure 4B ) ( Boers et al . , 1998; Nakajima et al . , 1998; Evans et al . , 2001; Rock et al . , 2009 ) . By 65 days in vitro ( D65 ) proximal-like epithelial structures form a cyst-like structure that expresses P63 , as determined by whole mount immunofluorescence staining and confocal z-stacks . Moreover , SMA expression is strongest at the periphery of the HLO ( Figure 4—figure supplement 1 ) . P63+ proximal airway-like cells also co-express SOX2 and NKX2 . 1 as determined on serial sections ( Figure 4—figure supplement 2 ) . Located on the luminal surface of HLO proximal airway-like structures are cells expressing the multi-ciliated cell transcription factor FOXJ1 ( Figure 4B ) . Very few cells expressed the club cell marker SCGB1A1 , and this protein was observed in a pixilated expression pattern ( Figure 4D ) . Multi-ciliated and club cell specific mRNAs , FOXJ1 and SCGB1A1 respectively , were significantly increased in prolonged HLO culture ( Figure 4A ) . Although the goblet cell marker MUC5AC mRNA expression was detected , protein expression was not detected by immunofluorescence ( Figure 4A and data not shown ) . 10 . 7554/eLife . 05098 . 017Figure 4 . Lung organoids form proximal airway-like structures . ( A ) Genes expressed in the proximal airway were examined in organoids across time . The proximal airway cell marker SOX2 decreased over time in HLOs cultures compared to D10 HLOs . Compared to undifferentiated hPSCs , organoids expressed high levels of the basal cell marker P63 at all time points , while expression of the club cell marker SCGB1A1 and ciliated cell marker FOXJ1 increased significantly in prolonged cultures ( compared to D10 HLOs ) . There was an increasing but non-significant trend in goblet cell MUC5AC expression over time in culture . ( B ) D65 HLOs had structures resembling the proximal airway , in which the epithelium ( β-catenin , red ) possesses P63+ basal cells ( green ) , and is surrounded by SMA+ ( white , upper and lower left panel ) mesenchymal tissue . Adjacent to the P63 positive basal cell layer ( green , lower , right panel ) were FOXJ1 positive cells ( white ) . Scale bars represent 50 µM ( top ) and 10 µM ( bottom ) . ( C ) Proximal airway-like epithelium ( β-catenin , green ) co-stained for ACTTUB on the apical side of the cell ( red ) . Scale bars represent 50 µM ( top ) and 10 µM ( bottom ) . ( D ) . Proximal airway-like epithelium ( E-cadherin , red ) also co-stained with Club cell marker CC10 ( white , right panel ) . Scale bars represent 50 µM ( top ) and 10 µM ( bottom ) . ( E ) Acellular human lung matrix was seeded with spheroids and cultured for 40 days ( D40 ) . Matrices had abundant proximal airway-like structures that had multi-ciliated cells on the apical surfaced labeled by ACTTUB ( red , top panel ) in low ( scale bar 50 µM ) and high magnification ( scale bar 10 µM ) . Serial sections showed that cells were also FOXJ1 positive ( white , lower panel ) with the epithelium outlined in ECAD ( green ) in low ( scale bar 50 µM ) and high magnification ( scale bar 10 µM ) . ( B–D ) ‘L’ in high magnification images indicates the lumen . *p < 0 . 05 . All error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 01710 . 7554/eLife . 05098 . 018Figure 4—figure supplement 1 . Lung organoids have P63+ epithelium throughout the organoid . ( A ) Confocal Z-slices taken at every 40 µm show P63+ ( green ) and ECAD+ ( white ) structures through the D65 HLO . ( B ) Z-slices taken at every 40 µm show SMA ( white ) surrounding the periphery the HLO with P63 ( green ) staining within the HLO . Scale bars represent 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 01810 . 7554/eLife . 05098 . 019Figure 4—figure supplement 2 . P63+ cells have an NKX2 . 1+ lung identity . ( A ) Serial sections were stained with NKX2 . 1 and P63 respectively . The adjacent sections expressed ECAD ( white ) and NKX2 . 1 ( green ) in the first section and P63 ( green ) in the second section . ( B ) P63+ cells ( green ) co-expressed the proximal lung marker SOX2 ( red ) in the epithelium labeled by ECAD ( white ) . Scale bars represent 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 019 Although the multi-ciliated cell transcription factor FOXJ1 was abundant in proximal airway-like structures , we observed that ACTTUB was localized to the apical side of these cells , but did not appear to be localized to cilia on the apical cell surface ( Figure 4C ) , suggesting that this may represent a cell that has not fully differentiated . Others have demonstrated that robust differentiation of multi-ciliated cells from hPSCs require modified culture conditions to promote differentiation of functional cell types ( Firth et al . , 2014 ) . Thus , it is possible that the HLO environment , such as Matrigel or media rich in FGF10 , does not promote terminal differentiation of all cell types . In order to alter the HLO environment , we seeded NOG/SB/F/Ch/SAG foregut spheroids onto an acellular human lung matrix ( Booth et al . , 2012 ) . Spheroids seeded on slices of acellular lung matrix predominantly gave rise to proximal airway-like structures in which stereotypical tufts of ACTTUB positive ciliated structures on the apical surface of cells were observed facing into a lumen . In serial sections , these airways had abundant FOXJ1+ cells ( Figure 4E ) . Thus , HLOs have the capacity to generate more mature ciliated cells given the correct stimulus or environment . As noted , proximal airways are often closely associated with the SMA+ mesenchyme ( Figure 4B ) whereas in the adult murine lung , proximal airways are also associated with Pdgfrα+ and Vim+ mesenchymal cells ( Boucherat et al . , 2007; Hinz et al . , 2007; Chen et al . , 2012 ) . Thus , we investigated the mesenchymal population within the HLOs in more detail . Immunofluorescence revealed that D65 HLOs have both PDGFRα+/VIM+ double positive and PDGFRα−/VIM+ cell populations , which is indicative of myofibroblasts and fibroblasts respectively ( Figure 5A ) . Adult murine myofibroblasts also co-express Sma and Pdgfrα whereas differentiated smooth muscle is Sma+/Pdgfra− ( Leslie et al . , 1990; Low and White , 1998; Boucherat et al . , 2007; Hinz et al . , 2007; Chen et al . , 2012 ) , and we observe PDGFRα+/SMA+ and PDGFRα−/SMA+ populations of cells indicating that HLOs possess myofibroblasts and smooth muscle cells ( Figure 5B ) . The HLOs did not stain positive for Safranin O indicating there is no cartilage tissue , whereas iPSC derived teratomas had abundant cartilage ( Figure 5C ) . Taken together , the HLO mesenchymal population is diverse with myofibroblasts , fibroblasts , and smooth muscle cells . 10 . 7554/eLife . 05098 . 020Figure 5 . Lung organoids possess multiple types of mesenchymal cells . ( A ) D65 HLOs have PDGFRα+ ( green ) VIM+ ( white ) double-positive myofibroblasts and PDGFRα−/VIM+ fibroblasts . Scale bar represents 50 µm . ( B ) D65 HLOs also possesses PDGFRα+ ( green ) SMA+ ( white ) double-positive myofibroblasts and PDGFRα−/SMA+ smooth muscle and myofibrblasts . Scale bar represents 50 µm . ( C ) D65 HLOs do not contain any cartilage whereas positive control iPSC derived teratoma had clear Safranin O staining specific to cartilage . Fast green marks the cytoplasm and hematoxylin the nuclei of both tissues . Scale bar represents 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 020 The distal lung epithelium in mouse and human make up the gas-exchanging alveoli , consisting of type I and type II alveolar epithelial cells ( AECI , AECII ) . During development , the distal lung epithelium expresses progenitor markers including SOX9 , ID2 , and NMYC ( Okubo , 2005; Rawlins et al . , 2009; Chang et al . , 2013; Rockich et al . , 2013 ) . All distal markers are present in the HLOs; however , ID2 and NMYC are expressed at high levels in early cultures , but are down regulated in prolonged culture ( Figure 3F ) while SOX9 expression remains consistent across time in culture ( Figure 6A ) . 10 . 7554/eLife . 05098 . 021Figure 6 . Lung organoids possess abundant distal bipotent progenitor cells . ( A ) The expression of the distal progenitor marker SOX9 remained unchanged over time and expression of the AECI marker PDPN was low in HLO cultures . ( B ) The majority of SFTPC+ cells ( green , left panel ) co-expressed SOX9 ( red ) . Similarly , many cells expressing the AECI early marker HOPX+ ( green , right panel ) co-expressed SOX9 ( red ) . Few , scattered cells expressed the late AECII marker SFTPB ( white , second panel ) or the AECI marker , PDPN ( third panel , white ) . Few PDPN+ cells also showed elongated , squamous morphology seen in the adult lung . ( C ) Human lung AECII cells labeled with SFTPC ( green , left panel ) did not co-express SOX9 . SFTPB+ cells ( white , second panel ) in the adult human lung have similar morphology to SFTPB+ cells in HLOs . Human lung AECI cells expressed PDPN ( white , third panel ) , and show characteristic AECI cell shape . Human AECI cells express HOPX ( green , right panel ) , but did not co-express SOX9 . ( B–C ) Scale bar in lower magnification images in B ( upper panel ) represent 50 µM and the scale bars in higher magnification images in B , C ( lower panel ) represent 10 µM . ( D ) D50 HLOs contain lamellar bodies which are organelles specific to AECII cells . Scale bars represent 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 02110 . 7554/eLife . 05098 . 022Figure 6—figure supplement 1 . SFTPC+ cells express lung specific markers . D65 HLOs express lung epithelial markers NKX2 . 1 ( green ) and SOX9 ( red ) and the adjacent section expresses SFTPC ( green ) and SOX9 ( red ) . Scale bar represents 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 022 Recently , there have been major advances in mice toward defining a bipotent alveolar progenitor population during the late fetal/early neonatal period ( Desai et al . , 2014; Treutlein et al . , 2014 ) , and this work has highlighted the fact that many markers previously considered terminal differentiation markers are co-expressed in the bipotent progenitors . Specifically , the AECII marker SftpC and AECI marker Hopx can be co-expressed in a bipotent progenitor before becoming committed to one lineage or the other . Moreover , we have shown that Sox9 marks an early progenitor population in the developing mouse lung and Sox9 also marks the bipotent progenitor in late fetal life ( Rockich et al . , 2013; Treutlein et al . , 2014 ) . In HLOs grown in prolonged culture ( >2 months ) , we observed that AECII ( SFTPC , SFTPB ) and AECI ( PDPN , HOPX ) cell-type markers were present ( Figure 6A–B ) . However , we also observed that SFTPC levels were very low ( Figure 3F ) , and that SFTPB+ cells were rare ( Figure 6B ) . This suggested that the distal airway cells present in HLOs might be a progenitor-like population . To test this possibility , we co-stained SFTPC ( AECII ) or HOPX ( AECI ) with SOX9 and found abundant SFTPC/SOX9 and HOPX/SOX9 double positive cells ( Figure 6B ) . Co-staining in serial sections suggests that SFTPC/SOX9 double positive cells are also NKX2 . 1+ ( Figure 6—figure supplement 1 ) . In contrast these co-expressing cells were not found in the adult human lung ( Figure 6C ) . Although rare , the few SFTPB+ observed in HLOs resemble AECII cells seen in the adult human lung , and PDPN+ cells resembled the elongated AECI cells in the human lung ( Figure 6B–C ) . In order to improve confidence that cells expressing AECII markers are AECII cells , we used transmission electron microscopy ( TEM ) to determine if HLOs possessed cells containing lamellar bodies , which are necessary for surfactant protein trafficking and secretion ( Schmitz and Müller , 1991; Stahlman et al . , 2000; Weaver et al . , 2002 ) . Using TEM , we observed lamellar bodies both in cells within HLOs , and in open spaces between cells , indicating that lamellar bodies are being secreted ( Figure 6D ) . Taken together , our data suggests that HLOs predominantly possess undifferentiated alveolar progenitor cells with rare differentiated AECI and AECII cells interspersed throughout the distal-like tissue . We have shown that HLOs have both proximal-like and distal-like epithelium in addition to surrounding mesenchymal tissue . In order to better gauge the composition of HLOs , we performed a detailed quantitative analysis of cell types and structures . We sectioned 48 individual HLOs , and examined them for P63+ proximal airway-like structures ( as shown in Figure 4B–D ) , and distal-airway like structures ( as shown in Figure 6—figure supplement 1 ) . We found that 39/48 ( 81% ) of the HLOs have proximal airway epithelial structures while 48/48 ( 100% ) of HLOs have distal airway-like structures ( Figure 7A ) . We then calculated the average cross-sectional area comprised of P63+ proximal airway-like and P63−/SFTPC+ distal airway-like tissue and found that proximal structures comprised 14 . 5% ( ±0 . 6% ) of the entire area of the HLO , whereas 85 . 5% ( ±0 . 6% ) were distal in nature ( including epithelium and mesenchyme ) ( Figure 7B ) . To determine the percentage of certain cell types within an HLO , we sectioned and stained 15 individual HLOs ( n = 15 ) and counted cells positive for specific markers , and the total number of Dapi+ nuclei within a section ( Figure 7C–G ) . On average , 57% of all cells in the HLOs were NKX2 . 1+ ( Figure 7C ) , 39% of all cells were P63+ , 3% were FOXJ1+ , 5% were SFTPC+ , and 4% of all cells were HOPX+ ( Figure 7D–G ) . 10 . 7554/eLife . 05098 . 023Figure 7 . Quantitative assessment of the composition of lung organoids . ( A ) HLOs were assessed for proximal airway-like structures ( P63+ ) and distal airway-like structures ( P63−/SFTPC+ ) . 81% of HLOs have proximal airway-like epithelium while 100% have distal airway-like epithelium ( n = 48 individual HLOs ) . ( B ) The average cross-sectional area within an HLO that is comprised of P63+ proximal airway-like and P63−/SFTPC+ distal airway-like epithelium was calculated . Proximal structures comprised 14 . 5% ( ±0 . 6% ) of the entire area of the HLO ( P63+ ) , whereas 85 . 5% ( ±0 . 6% ) of HLO was distal-like epithelium and mesenchyme ( P63− ) . ( C–G ) The percent of specific cell markers present in an organoid was determined by dividing by the total number of DAPI+ nuclei within the same section ( n = 15 individual HLOs ) . Each point represents the data from an individual HLO while the open bar represents the average percent of cells . ( C ) On average , 57% of all cells in the HLOs were NKX2 . 1+ , ( D ) 39% of all cells were P63+ , ( E ) 3% were FOXJ1+ , ( F ) 5% were SFTPC+ , ( G ) 4% of all cells were HOPX+ . ( B–G ) Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 023 Accumulating evidence suggests that HLOs are immature . For example , distal progenitor markers are initially robustly expressed whereas SFTPC expression is very low across time in HLOs ( Figure 3E ) , FOXJ1+ cells do not appear to form mature multi-ciliated structures until placed onto a decelluarized lung matrix ( Figure 4B , E ) and rare SCGB1A1+ cells do not resemble mature club cells ( Figure 4D ) . Moreover , the majority of the distal-like epithelium expresses bipotent progenitor markers ( Figure 5 ) . In order to directly address the maturity of HLOs , we conducted RNA-sequencing ( RNAseq ) on HLOs ( n = 6; 3 D65 HLOs , 3 D110 HLOs ) , on undifferentiated hESCs and on definitive endoderm . We also took advantage of publicly available RNAseq datasets for human fetal lung representing a range of gestational stages , and for adult human lung ( Supplementary file 1 ) . In order to determine global similarity among these tissues relative to HLOs , we conducted principal component ( PC ) analysis ( Figure 8A , B ) ( Ringnér , 2008 ) , hierarchical clustering ( Figure 8C ) ( Eisen et al . , 1998 ) and Spearman's rank-order correlation ( Jiang et al . , 2004 ) matrix analysis ( Figure 8D ) of the complete tabulated FPKM matrix generated from RNA sequences datasets and representing the total gene expression complement in each sample . Consistent across all three types of informatics analysis , transcriptional activity in the HLOs shares the greatest degree of similarity to human fetal lung . These data strongly suggest that global transcription of HLOs is highly similar to human fetal lung , and support the idea that HLOs are in a less differentiated , fetal state when grown in the conditions described here . 10 . 7554/eLife . 05098 . 024Figure 8 . RNA sequencing analysis associates HLOs with fetal lung tissue . 6 HLOs ( n = 3 D65 HLOs and n = 3 D110 HLOs ) were compared to the undifferentiated H9 stem cells ( SC ) and definitive endoderm ( Def End ) and publicly available datasets of adult and fetal human lungs ( see Supplementary file 1 ) . ( A–B ) Plot of the first three principle components generated in the principle component ( PC ) analysis as pairwise 2-dimensional plots ( A ) or as an aggregate 3-dimensional projection ( B ) , ( C ) hierarchical clustering , and ( D ) Spearman's correlation all demonstrate that HLOs are most closely related to the fetal lung . DOI: http://dx . doi . org/10 . 7554/eLife . 05098 . 024 To date , a number of groups have defined methods to generate lung specific cell types utilizing 2D culture systems ( Green et al . , 2011; Longmire et al . , 2012; Mou et al . , 2012; Wong et al . , 2012; Ghaedi et al . , 2013; Huang et al . , 2013 ) . Although lung lineage cells have been generated with varying efficiency ( ∼30–80% NKX2 . 1+ cells [Wong et al . , 2012; Huang et al . , 2013; Firth et al . , 2014] ) and can generate both both proximal ( ∼5–36% of cells [Wong et al . , 2012; Huang et al . , 2013; Firth et al . , 2014] ) and distal cell types ( up to ∼50% of cells [Huang et al . , 2013] ) , proper spatial organization of the cell types and specific tissue morphology have not been reported in 2D systems . Here , we show that HLOs possess both mesenchymal and lung epithelial ( ∼60% NKX2 . 1+ ) cells with proximal airway-like structures that possess P63+ ( ∼40% ) and FOXJ1+ cells ( ∼3% ) along with distal airway-like structures that possess SFTPC+ ( ∼5% ) and HOPX+ ( ∼4% ) cells . It is currently unclear if 2D culture systems described have the capability to give rise to mesodermal lineages . Thus , HLOs allow one to address questions regarding spatial tissue organization and epithelial-mesenchymal interactions . Since HLOs form organized structures that resemble bronchi and bronchioles with adjacent mesenchyme , these complex , organized tissues may allow exploration , for example , of airway remodeling after injury . Moreover , the spatial arrangement of specific cell types will be critical to study proximal airway dynamics during homeostasis and injury . For example , the location of P63+ cells adjacent to FOXJ1+ cells in the HLOs will be necessary to study basal cell differentiation into different proximal airway cell types during homeostasis or after injury . In addition to tissue morphology and structure during prolonged culture , the HLOs consist of both epithelium and mesenchyme in early cultures that are maintained over time . Since lung development requires extensive cross talk between the epithelium and mesenchyme in order to regulate developmental processes , proliferation and differentiation , HLOs may be an ideal in vitro system to study these complex tissue–tissue interactions . Recently , there has been a push to define progenitor populations during lung development and adult homeostasis in order to better understand differentiation and the transition between branching and alveolarization . Two groups have defined a bipotent progenitor population in the embryonic/neonatal lung that gives rise to both AECI and AECII cells ( Desai et al . , 2014; Treutlein et al . , 2014 ) . These bipotent cells express the distal progenitor marker Sox9 along with differentiation markers of AECI and AECII cells , including SftpC , HopX , and Pdpn . We demonstrate that HLOs expressed both AECI and II markers; however , the majority of these cells also expressed SOX9 suggesting that the majority of the distal epithelium is comprised of bipotent progenitors . Thus , HLOs will allow us to gain insight into this bipotent population , explore how bipotent progenitors are regulated , and define the mechanisms of how fate-decisions are made as terminal differentiation occurs . The evidence supporting that HLOs are fetal in nature could reflect the fact that a block to full maturation exists in vitro , as is the case with other endoderm lineage organoids ( intestinal and gastric ) , which appear to be immature . That is , while they possess committed lineage-specific cell types , the cells may not exhibit fully matured adult-like function ( McCracken et al . , 2014; Watson et al . , 2014 ) . This is also the case for pancreatic β-like cells and hepatocyte-like cells generated in vitro ( Si-Tayeb et al . , 2009; Hrvatin et al . , 2014 ) . Alternatively , the progenitor state may reflect the high levels of FGF10 in the culture media , since FGF10 is known to maintain progenitor cells in the lung ( Ramasamy et al . , 2007; Nyeng et al . , 2008 ) . Given that HLOs are similar to human fetal lung , this tissue is an ideal model to study lung maturation of both the proximal and distal epithelium along with epithelial-mesenchymal interactions in a developmental context . While the multi-lineage , multi-cellular composition of HLOs is a major advantage , one of the caveats to this system is that HLOs do not appear to undergo bona fide branching morphogenesis or possess transitional zones found in the adult lung , such as the bronchioalveolar ductal junction ( BADJ ) . The HLOs possess proximal SOX2+ domain and distal SOX9+ domains observed during branching morphogenesis , but this regionalization occurs without setting up the stereotyped branching pattern . This may be due to the fact that the organoids are surrounded by media supplemented with FGF10 compared to the in vivo situation where FGF10 is expressed in a dynamic , spatially restricted manner in the distal mesenchyme ( Bellusci et al . , 1997b; Nyeng et al . , 2008; Abler et al . , 2009 ) . However , it has recently been demonstrated that localized expression FGF10 is not required for branching ( Volckaert et al . , 2013 ) , so this may not explain the lack of branching . Alternatively , similar to other endoderm-derived organoid models , HLOs lack several components of the native organ , including immune cells , vasculature , and innervation . Thus , it is possible that cellular inputs important for branching morphogenesis are missing from HLOs . Indeed , recent reports have shown that innervation is required for proper branching ( Bower et al . , 2014 ) , and while vasculature may not be important for lung branching ( Havrilak and Shannon , 2015 ) , others have shown the vascular endothelium is required to induce a branching-like program of isolated airway epithelium in 3D cultures ( Franzdóttir et al . , 2010 ) . Lastly , the microenvironment is essential for branching morphogenesis to occur including dynamic changes in the extracellular matrix around branching lung bud tips where the ECM is constantly changing and interacting with the cytoskeleton of the branching epithelium in order to facilitate cell movement and branching bifurcations ( Moore et al . , 2005; Kim and Nelson , 2012; Wan et al . , 2013 ) . It is possible that in the future , co-culture with additional cellular inputs may prove to enhance HLO branching . Taken together , we describe here a novel system to generate human lung organoids from human pluripotent stem cells . HLOs possess both mesenchymal and epithelial lineages , as well as organized proximal airway structures with multiple cell types and surrounded by mesenchyme . HLOs also possess distal epithelial cells that are reminiscent of a bipotent alveolar progenitor cell recently described in mice which is likely a reflection of the similarities of HLOs to the human fetal lung . We believe that HLOs will be an excellent new human system to model lung differentiation , homeostasis and disease in vitro . Human ES cell lines H1 ( NIH registry #0043 ) and H9 ( NIH registry #0062 ) were obtained from WiCell Research Institute . Human ES line UM77-2 ( NIH registry #0278 ) was obtained from the University of Michigan . iPSC lines 3-5 and 20-1 were generated at Cincinnati Children's Hospital and have been previously described ( Spence et al . , 2011 ) . Stem cells were maintained on Matrigel ( BD Biosciences , San Jose , CA ) in mTeSR1 medium ( STEMCELL Technologies , Vancouver , Canada ) . HESCs were passaged as previously described ( Spence et al . , 2011 ) . Differentiation into definitive endoderm was carried out as previously described ( D'Amour et al . , 2005; Spence et al . , 2011 ) . Briefly , a 4-day Activin A ( R&D systems , Minneapolis , MN ) differentiation protocol was used . Cells were treated with Activin A ( 100 ng ml−1 ) for 3 consecutive days in RPMI 1640 media ( Life Technologies , Grand Island , NY ) with increasing concentrations of 0% , 0 . 2% and 2% HyClone defined fetal bovine serum ( dFBS , Thermo Scientific , West Palm Beach , FL ) . After differentiation into definitive endoderm , foregut endoderm was differentiated , essentially as described ( Green et al . , 2011 ) . Briefly , cells were incubated in foregut media: Advanced DMEM/F12 plus N-2 and B27 supplement , 10 mM Hepes , 1× L-Glutamine ( 200 mM ) , 1× Penicillin-streptomycin ( 5000 U/ml , all from Life Technologies ) with 200 ng/ml Noggin ( NOG , R&D Systems ) and 10 µM SB431542 ( SB , Stemgent , Cambridge , MA ) for 4 days . For long term maintenance , cultures were maintain in ‘basal’ foregut media without NOG and SB , or in the presence of growth factors including 50 , 500 ng/ml FGF2 ( R&D systems ) , 10 µM Sant-2 ( Stemgent ) , 10 µM SU5402 ( SU , Stemgent ) , 100 ng/ml SHH ( R&D systems ) , and SAG ( Enzo Life Sciences , Farmingdale , NY ) for 8 days . After differentiation into definitive endoderm , cells were incubated in foregut media with NOG , SB , 500 ng/ml FGF4 ( R&D Systems ) , and 2 µM CHIR99021 ( Chiron , Stemgent ) for 4–6 days . After 4 days with treatment of growth factors , three-dimensional floating spheroids were present in the culture . Three-dimensional spheroids were transferred into Matrigel to support 3D growth as previously described ( McCracken et al . , 2011 ) . Briefly , spheroids were embedded in a droplet of Matrigel ( BD Bioscience #356237 ) in one well of a 24 well plate , and incubated at room temperature for 10 min . After the Matrigel solidified , foregut media with 1% Fetal bovine serum ( FBS , CAT#: 16000–044 , Life Technologies ) or other growth factors and small molecules were overlaid and replaced every 4 days . Organoids were transferred into new Matrigel droplets every 10–15 days . Immunostaining was carried out as previously described ( Spence et al . , 2009; Rockich et al . , 2013 ) . Antibody information and dilutions can be found in Supplementary file 2 . All images were taken on a Nikon A1 confocal microscope or an Olympus IX71 epifluorescent microscope . RNA was extracted from monolayers , spheroids , and organoids using a MagMAX-96 Total RNA Isolation Kit ( Life Technologies ) and MAG Max Express ( Applied Biosystems , Grand Island , NY ) . RNA quantity and quality were determined spectrophotometrically , using a Nano Drop 2000 ( Thermoscientific ) . Reverse transcription was conducted using the SuperScript VILO kit ( Invitrogen , Grand Island , NY ) , according to manufacturer's protocol . Finally , qRT-PCR was carried out using Quantitect Sybr Green MasterMix ( Qiagen ) on a Step One Plus Real-Time PCR system ( Life Technologies ) . For a list of primer sequences see Supplementary file 3 . Human lungs deemed to be unsuitable for lung transplantation were obtained from beating-heart ( or warm autopsy ) donors through Gift of Life Michigan and lungs were decellularized as previously described ( Booth et al . , 2012 ) . Slices were prepared using a sterile tissue punch ( Fisher ) and sterilized with 0 . 18% peracetic acid and 4 . 8% EtOH . Matrix slices were placed in a 96 well plate and approximately 50 NOG+SB+F+Ch+SAG spheres were pipetted directly onto the matrices . Samples were centrifuged for 2 min at 2000 rpm and then incubated at 37°C for 30 min without media . Foregut media supplemented with 1% FBS and 500 ng/ml FGF10 was then added to the matrices . Media was changed daily . D50 HLOs were processed as previously described ( Prasov et al . , 2012; Rockich et al . , 2013 ) . 70 nm sections were sections were imaged using a Philips CM-100 electron microscope . HLOs with P63+ cells were counted as having proximal airway-like epithelium and HLOs with SFTPC+ cells were counted as having distal airway-like epithelium . The area of proximal epithelium was determined by P63+ECAD+ staining . Area was measured using ImageJ software . Cell quantification of NKX2 . 1 , P63 , and DAPI was counted by Metamorph cell counting software . FOXJ1 , SFTPC , and HOPX were counted in ImageJ using the cell counter plugin . All immunofluorescence and qRT-PCR experiments were carried out at least two times with three ( n = 3 ) independent biological samples per experiment . The only exceptions to this were experiments that included human adult lung samples in the analysis . For these experiments , n = 1 biological human lung sample was used in statistical replicates ( triplicates ) whereas all other samples used biological replicates ( n = 3 ) . For quantification in Figure 7 , a total of 48 different HLOs ( n = 48 ) were counted for HLO composition . For the proximal epithelial area , 29 different HLOs were counted ( n = 29 ) . For cell quantification , 15 different HLOs were counted ( n = 15 ) . Statistical differences between groups were assessed with Prism software , using multiple t tests . All error bars represent SEM . Results were considered statistically significant at p < 0 . 05 . Sequencing of HLOs ( n = 3 D65 , n = 3 D110 ) was performed by the University of Michigan DNA Sequencing Core , using the Illumina Hi-Seq platform . Sequencing of H9 Stem Cells ( SC ) and Definitive Endoderm ( DE ) was performed by the University of California , San Francisco DNA Sequencing Core using the Illumina Hi-Seq platform . All sequences were deposited in the EMBL-EBI ArrayExpress database using Annotare 2 . 0 and are catalogued under the accession number E-MTAB-3339 for the HLOs and E-MTAB-3158 for SC and DE . The University of Michigan Bioinformatics Core obtained the reads files and concatenated those into a single ‘ . fastq’ file for each sample . The Bioinformatics Core also downloaded reads files from EBI-AE database ( Adult lung Samples ) and NCBI-GEO ( SRA ) database ( Fetal lung samples ) ( Supplementary file 1 ) . The quality of the raw reads data for each sample was evaluated using FastQC ( version 0 . 10 . 1 ) to identify features of the data that may indicate quality problems ( e . g . , low quality scores , over-represented sequences , inappropriate GC content , etc ) . Initial QC report indicated over-representation of Illumina adapter sequences in samples from EBI-AE data set and NCBI-GEO data set . Adapter sequences were trimmed from the reads using Cutadapt ( version 0 . 9 . 5 ) ( Chen et al . , 2014a ) . Briefly , reads were aligned to the reference transcriptome ( UCSC hg19 ) using TopHat ( version 2 . 0 . 9 ) and Bowtie ( version 2 . 1 . 0 . 0 ) ( Langmead et al . , 2009 ) . Cufflinks/CuffNorm ( version 2 . 2 . 1 ) was used for expression quantitation and differential expression analysis ( Trapnell et al . , 2012 ) , using UCSC hg19 . fa as the reference genome sequence and UCSC hg19 . gtf as the reference transcriptome annotation . For this analysis , we used parameter settings: ‘–multi-read-correct’ to adjust expression calculations for reads that map in more than one locus , as well as ‘–compatible-hits-norm’ and ‘–upper-quartile –norm’ for normalization of expression values . Normalized FPKM tables were generated using the CuffNorm function found in Cufflinks . Transcriptional quantitation analysis in Cufflinks was conducted using the 64-bit Debian Linux stable version 7 . 8 ( ‘Wheezy’ ) platform . The complete FPKM matrix , containing frequency counts for all 24 , 010 genes contained in the reference genome for all 23 RNAseq samples , was evaluated using unscaled principle component analysis ( PCA ) to visualize and quantify multi-dimensional variation between samples ( Ringnér , 2008 ) . Of the 24 , 010 genes annotated in the reference genome , 2815 ( 11 . 7% ) were not detected in the RNAseq analysis of any of the 23 samples . Principle components were calculated using the function ‘prcomp’ found in the R ( version 3 . 1 . 2 ) statistical programming language ( http://www . R-project . org/ ) and plotted using the R package ‘ggplot2’ ( Wickham , 2009 ) . Hierarchical cluster analysis based on the Canberra distance ( Eisen et al . , 1998 ) between FPKM vectors was used to classify discrete RNAseq samples according to the degree of total transcriptional dissimilarity indicated by the normalized FPKM values . Bootstrap analysis was used to assess the uncertainty in the assigned hierarchical clustering relationships . 10 , 000 bootstraping iterations were generated by repeatedly randomly sampling the FPKM dataset . The bootstrap probability ( BP ) of a cluster is defined as the frequency of a given relationship among the bootstrap replicates . Multiscale bootstrap resampling was used to calculate an approximately unbiased ( AU ) p-value for a given relationship , with AU > 95 indicating a high degree of statistical significance . Analyses were conducted using R package ‘pvclust’ ( Suzuki and Shimodaira , 2006 ) . Spearman correlation was applied as an additional assessment of the cumulative degree of correlation among RNAseq datasets . In addition , we computed Spearman rank correlation coefficients ( ρ ) in a pairwise manner among all 23 RNAseq samples using the complete normalized FPKM data . The Spearman coefficients were plotted as a heatmap using the function ‘heatmap . 2’ in the R package ‘gplots’ ( http://CRAN . R-project . org/package=gplots ) . Complete data analysis scripts are available at https://github . com/hilldr/HLO_eLife2015 .	Cell behavior has traditionally been studied in the lab in two-dimensional situations , where cells are grown in thin layers on cell-culture dishes . However , most cells in the body exist in a three-dimensional environment as part of complex tissues and organs , and so researchers have been attempting to re-create these environments in the lab . To date , several such ‘organoids’ have been successfully generated , including models of the human intestine , stomach , brain and liver . These organoids can mimic the responses of real tissues and can be used to investigate how organs form , change with disease , and how they might respond to potential therapies . Here , Dye et al . developed a new three-dimensional model of the human lung by coaxing human stem cells to become specific types of cells that then formed complex tissues in a petri dish . To make these lung organoids , Dye et al . manipulated several of the signaling pathways that control the formation of organs during the development of animal embryos . First , the stem cells were instructed to form a type of tissue called endoderm , which is found in early embryos and gives rise to the lung , liver and other several other internal organs . Then , Dye et al . activated two important developmental pathways that are known to make endoderm form three-dimensional intestinal tissue . However , by inhibiting two other key developmental pathways at the same time , the endoderm became tissue that resembles the early lung found in embryos instead . This early lung-like tissue formed three-dimensional spherical structures as it developed . The next challenge was to make these structures develop into lung tissue . Dye et al . worked out a method to do this , which involved exposing the cells to additional proteins that are involved in lung development . The resulting lung organoids survived in laboratory cultures for over 100 days and developed into well-organized structures that contain many of the types of cells found in the lung . Further analysis revealed the gene activity in the lung organoids resembles that of the lung of a developing human fetus , suggesting that lung organoids grown in the dish are not fully mature . Dye et al . 's findings provide a new approach for creating human lung organoids in culture that may open up new avenues for investigating lung development and diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2015	In vitro generation of human pluripotent stem cell derived lung organoids

Subsets and Splits

No community queries yet

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