News Article | May 24, 2017
No statistical methods were used to predetermine sample size. The investigators were not blinded to allocation during experiments and outcome assessment. Stable cell lines expressing affinity-tagged bait proteins were created according to protocols described previously in detail4. In brief, C-terminally HA–Flag-tagged clones targeting human bait proteins were constructed from clones included in version 8.1 of the human ORFeome (http://horfdb.dfci.harvard.edu)14. All expression clones used in this study are available from the Dana Farber/Harvard Cancer Center DNA Resource Core Facility (http://dnaseq.med.harvard.edu/). After sequence validation, clones were introduced into HEK293T, HCT116, or MCF10A cells (all from American Type Culture Collection) via lentiviral transfection. Cells were expanded under puromycin selection to obtain five 10-cm dishes per cell line before AP–MS. Bait proteins were selected from the ORFeome for high-throughput AP–MS analysis in batches corresponding to individual 96-well plates. Plates were selected for processing in random order. For AP–MS experiments in MCF10A cells, 1.15 × 106 cells per 15 cm dish were collected after 3 days (sub-confluent) or after 14 days in culture (contact inhibited) to allow for expulsion of YAP1 from the nucleus and Hippo pathway activation. MCF10A cells were grown in DMEM/F12 media supplemented with 5% horse serum, 20 ng ml−1 EGF, 10 μg ml−1 insulin, 0.5 μg ml−1 hydrocortisone, 100 ng ml−1 cholera toxin, 50 U ml−1 penicillin, and 50 μg ml−1 streptomycin. All cell lines were found to be free of mycoplasma using Mycoplasma Plus PCR assay kit (Agilent). Karyotyping (GTG-banded karyotype) of HeLa, HCT116, and HEK293T cells for cell line validation was performed by Brigham and Women’s Hospital Cytogenomics Core Laboratory. All AP–MS experiments were performed as presented previously in full4. In brief, cell pellets were lysed in the presence of 50 mM Tris-HCl pH 7.5, 300 mM NaCl, 0.5% (v/v) NP40, followed by centrifugation and filtration to remove debris. Immunoprecipitation was achieved using immobilized and pre-washed mouse monoclonal anti-HA agarose resin (Sigma-Aldrich, clone HA-7) that was incubated with clarified lysate for 4 h at 4 °C before removal of supernatant and four washes with lysis buffer followed by two washes with PBS (pH 7.2). Complexes were eluted in two steps using HA peptide in PBS at 37 °C and subsequently underwent TCA precipitation. Baits were processed in batches corresponding to 96-well plates in the ORFeome collection; plates were processed in random order. In preparation for LC–MS analysis, protein samples were reduced and digested with sequencing-grade trypsin (Promega). Peptides were then de-salted using homemade StageTips30 and approximately 1 μg of peptides were loaded onto C18 reversed-phase microcapillary columns and analysed on Thermo Fisher Q-Exactive mass spectrometers. Data acquisition methods were approximately 70 min long, including sample loading, gradient, and column re-equilibration. Tandem mass spectrometry (MS/MS) spectra were acquired in data-dependent fashion targeting the top 20 precursors for MS2 analysis. Unless noted otherwise, a single biological replicate of each bait was subjected to affinity purification followed by technical duplicate LC–MS analysis. For a complete description of data acquisition parameters, see ref. 4. A brief synopsis of our methods for identifying peptides and proteins from LC–MS data and distinguishing bona fide interacting proteins from background is provided here. For full details, refer to ref. 4. The BioPlex 2.0 network was generated by reanalysing Sequest search results from the BioPlex 1.0 dataset, combined with additional new AP–MS datasets. Sequest31 was used to match MS/MS spectra with peptide sequences from the Uniprot20 human protein database supplemented with sequences of green florescent protein (GFP) (our negative control), our Flag–HA affinity tag, and common contaminant proteins. This version of the UniProt database includes both SwissProt and Trembl entries and was current in 2013, at the outset of this project when the first AP–MS data were collected and searched. All protein sequences were included in forward and reversed orientations. Only fully tryptic peptides with two or fewer missed cleavages were considered, and precursor and product ion mass tolerances were set to 50 p.p.m. and 0.05 Da, respectively. The sole variable modification considered was oxidation of methionine (+15.9949). Target-decoy filtering32 was applied to control FDRs, using a linear discriminant function for peptide filtering and probabilistic scoring at the protein level33. Linear discriminant analysis considered Xcorr, D-Cn, peptide length, charge state, fractions of ions matched, and precursor mass error to distinguish correct from incorrect identifications. Peptide-spectral matches from each run were filtered to a 1% protein-level FDR with additional entropy-based filtering4 to reduce the final dataset protein-level FDR to well under 1%. Protein identifications supported by only a single peptide were discarded as well. These additional post-search filters further reduced the dataset-level FDR by over 100-fold. Scoring to identify HCIPs was performed in multiple stages after combining technical duplicate analyses of each AP–MS experiment and mapping all protein identifiers to Entrez Gene identifiers to minimize technical issues due to protein isoforms. Protein abundances in each immunoprecipitation were quantified using spectral counts averaged across technical replicates. The CompPASS algorithm34, 35 compared abundances of the proteins detected in each immunoprecipitation with their average levels across all other immunoprecipitations, returning a z score that quantified the extent to which a protein’s abundance exceeds its average levels across the dataset as well as the empirical NWD-score that accounted for a protein’s abundance, frequency of detection, and consistency across duplicate analyses. Subsequent filtering based on PSM counts, entropy scoring, and each protein’s frequency of detection within each batch of samples minimized false positives, liquid chromatography carryover, and technical artefacts. Putative bait–prey interactions were further filtered using CompPASS-Plus4, a naive Bayes classifier that learns to distinguish true interacting proteins from non-specific background and false positive identifications on the basis of CompPASS scores and several other metrics described previously. The algorithm modelled true interactions using examples from STRING36 and GeneMania37 databases. False positive protein identifications were modelled using decoy identifications that had survived previous filters. All remaining data were used to model background. Cross-validation was applied by batch, with each 96-well plate of immunoprecipitations scored using a model trained on ~57 different plates. Bait–prey interactions were then assembled across immunoprecipitations to produce a single network, combining scores of reciprocal interactions to increase their weight. BioPlex 2.0 was obtained by pruning this network to retain only those interactions that earned scores above 0.75, as described previously4. See Supplementary Table 1 for a list of baits as well as a complete list of interactions. BioPlex 2.0 interaction data were compared with data from BioGRID38, CORUM15, STRING36, GeneMania37, and MINT39 databases as described previously4. Because the BioPlex 2.0 dataset incorporates the contents of BioPlex 1.0 and data from this project have been deposited directly into BioGRID, released to the scientific community via the project website (http://bioplex.hms.harvard.edu), and otherwise distributed40 at intervals throughout the project, snapshots of these databases predating public disclosure of any BioPlex data were used to ensure that no interactions derived from BioPlex were included in the comparison. In Extended Data Fig. 1a, several data sources were used to determine the fractions of various protein families included as baits or preys in BioPlex 1.0 or 2.0. The list of human kinases was downloaded from kinase.com (http://kinase.com/web/current/human/; December 2007 update). Mitochondrial proteins were taken from MitoCarta 2.0 (ref. 41). Lists of transcription factors and chromatin-remodelling factors were drawn from http://www.bioguo.org. Drug target lists were taken from http://www.drugbank.ca. Cancer genes were taken from ref. 42. Disease genes were extracted from the curated set of disease–gene associations in the DisGeNET database25. ‘Essential’ genes were taken from recent papers describing clustered regularly interspaced palindromic repeat (CRISPR)–Cas9 screening to identify human genes that confer a fitness advantage6, 7. In each case, protein identifiers were converted to Entrez Gene identifiers, if necessary, and compared against those gene products included in either interaction network. Each of these analyses was performed exactly as described previously4. Brief summaries follow. Subcellular localization predictions relied upon localization information provided for a subset of proteins by the UniProt website (http://www.uniprot.org) in March 2016. These localization terms were manually condensed to 13 core localizations: nucleus, cytoplasm, cytoskeleton, endosome, endoplasmic reticulum, extracellular, Golgi, lysosome, mitochondrion, peroxisome, plasma membrane, vesicle, and cell projection. Fisher’s exact test was used to calculate the enrichment of each term among each protein’s primary and secondary neighbours, with multiple testing correction43. Predictions were made when enrichments were significant at an adjusted FDR of 1%. Localization predictions are provided in Supplementary Table 3. Domain–domain associations were uncovered by mapping PFAM domains onto the 56,553 protein–protein interactions in the BioPlex 2.0 network. After counting the numbers of interactions involving each domain individually and the number of interactions in which the domains were brought together within separate proteins, Fisher’s exact test was used to evaluate significance with subsequent correction for multiple hypothesis testing. Domains were considered significantly associated at an adjusted P value less than 0.01. Significant domain–domain associations are summarized in Supplementary Table 4. The enrichment of GO44 terms and PFAM22 domains was determined among each protein’s immediate neighbours and for each network community using Fisher’s exact test with multiple testing correction43. GO and PFAM data were downloaded from the UniProt website (http://www.uniprot.org) in March 2016. Only terms occurring at least twice were considered. Enrichments of GO terms and PFAM domains among each protein’s neighbours are summarized in Supplementary Table 5. The MCL algorithm5 was used to partition the BioPlex 2.0 network into communities of tightly interconnected proteins, using an implementation provided by the algorithm’s creator, S. van Dongen, at http://micans.org/mcl/. The option –force-connected=y was used to ensure that final clusters correspond to connected components. The MCL algorithm requires specification of one parameter, the inflation parameter, which controls the granularity of the clusters that are produced. Clustering of BioPlex 2.0 was repeated for several values of the inflation parameter between 1.5 and 2.5. After comparing experimentally derived clusters with known protein complexes, an inflation parameter of 2.0 was selected for final clustering. Clusters containing fewer than three proteins were discarded, producing a final list of 1,320 protein communities. Each cluster and its members are summarized in Supplementary Table 6; GO terms and PFAM domains enriched in each community are provided in Supplementary Table 7. One important question has been the extent to which each of the clusters observed in BioPlex 2.0 is also visible in BioPlex 1.0. To address this question, we mapped each cluster detected in BioPlex 2.0 onto the BioPlex 1.0 network. If a given cluster was also reflected in the BioPlex 1.0, then we would expect to see an enrichment of interactions; conversely, if interactions were not enriched among the relevant set of proteins above background, then there would be no evidence to support the indicated cluster. After mapping each cluster of tightly interconnected proteins from BioPlex 2.0 onto the BioPlex 1.0 network, we used a binomial test to evaluate the enrichment of BioPlex 1.0 interactions among matching proteins. The probability of interaction was estimated from the fraction of all possible interactions in the BioPlex 1.0 network that was actually detected (8.08 × 10−4); the number of trials was taken to be the maximum number of interactions possible among those proteins within the cluster that were part of the BioPlex 1.0 network; the number of interactions actually observed in this portion of BioPlex 1.0 was taken as the number of successes. A one-sided binomial test was performed and a correction for multiple testing was applied43. Overall, 45% of complexes detected in BioPlex 2.0 did not show any enrichment for protein interactions in BioPlex 1.0, suggesting that these were macromolecular complexes not covered in the first interaction network. Moreover, although the remaining 55% of complexes were at least partly reflected in BioPlex 1.0, the density of their coverage consistently increased with incorporation of additional AP–MS data into the BioPlex 2.0 network. In addition to using MCL clustering to partition the BioPlex 2.0 network into individual clusters of tightly interconnected proteins, we also wanted to explore patterns of interconnection within the network that related these clusters to each other. For this purpose, we searched for pairs of clusters that were connected to each other through interactions among their constituent proteins more often than would be expected. First, the full set of 56,553 interactions was trimmed to include only those interactions connecting one cluster with another, and the set of all cluster pairs connected by one or more interactions was identified. For each of these pairs of clusters, the number of interactions connecting the pair was determined, as were the numbers of interactions involving each cluster individually. Fisher’s exact test was used to identify pairs of clusters that were enriched for interactions among them, followed by multiple testing correction43. The 929 cluster–cluster associations that were accepted at a 1% FDR are displayed in Fig. 3a and Extended Data Fig. 9 and provided in Supplementary Table 6. GO and PFAM enrichments for each community are summarized in Supplementary Table 7. The first step towards examining network properties of fitness proteins was to combine lists of proteins associated with increased cellular fitness from refs 6, 7 into a single composite list. For our purposes, we used the union of both lists to define the set of fitness proteins. Entrez Gene identifiers were associated with proteins on this list and mapped onto the BioPlex 2.0 network. To assess network properties of fitness proteins, the composite list of proteins associated with increased cellular fitness was superimposed onto the BioPlex network, effectively subdividing all proteins in the network into two groups corresponding to fitness and non-fitness proteins. Vertex degrees, local clustering coefficients, and eigenvector centralities were then computed and averaged across all fitness proteins. To evaluate whether these values differed for fitness proteins compared with randomly selected protein subsets of equivalent size, fitness and non-fitness labels were scrambled across the network and a new average was calculated for the randomized list of fitness proteins. This process was repeated 10,000 times to define null distributions for each statistic. Since these distributions were normally distributed, Gaussian distributions were fitted to each and used to assign z scores and P values for each statistic associated with the true set of fitness proteins. To evaluate graph assortativity, the BioPlex network was subdivided into fitness and non-fitness proteins and the assortativity of the partitioned graph was calculated. This process was repeated 10,000 times, randomizing fitness and non-fitness labels, and the resulting distribution was fitted to a Gaussian distribution and used to determine a z score and P value associated with the true assortativity. A second goal was to identify clusters enriched with fitness proteins. For this purpose, a one-sided hypergeometric test was used to evaluate the enrichment of fitness proteins, taking into account the size of the cluster, the size of the BioPlex network, and the fraction of network proteins that were associated with increased cellular fitness. Only clusters containing two or more fitness proteins were considered for this analysis. Once a multiple testing correction43 was applied, 53 communities were found to be enriched with fitness proteins at a 1% FDR. These clusters are summarized in Extended Data Fig. 9. Levels of enrichment are summarized for those communities containing two or more cellular fitness proteins in Supplementary Table 8. To assess the tendency for clusters containing fitness proteins or enriched for fitness proteins to be centrally located within the cluster–cluster association network (Fig. 3a), all clusters were sorted according to their eigenvector centralities. The Kolmogorov–Smirnov test was used to compare distributions of clusters enriched and not enriched with fitness proteins within the ranked list of all clusters. This process was repeated to compare distributions of clusters containing multiple fitness proteins with clusters containing 0 or 1 fitness proteins, as shown in Fig. 3d. The basis for our study of protein complexes and disease was the DisGeNET database of disease–gene associations25. For our analysis we used the full database that relates over 16,000 genes with 13,000 partly redundant disease classifications. Each disease state and its associated proteins were then mapped onto each BioPlex 2.0 complex and evaluated for enrichment using a hypergeometric test, taking into account the size of the complex, the number of disease proteins in the complex, the number of disease proteins within the network, and the total network size. This process was repeated for each community and for each disease state. After multiple testing correction43, those complexes enriched with proteins involved with each disease at a 1% FDR were deemed associated. The resulting disease–complex associations were assembled into a network in which clusters and disease states are both represented as nodes, with edges connecting clusters with significantly associated disease states, depicted in full in Fig. 4a. All significant disease-cluster associations are provided in Supplementary Table 8. The eigenvector centralities assigned to disease states within the composite disease-community network were used to compare across a range of disease states. Disease classifications were taken from the DisGeNET database as reported in their SQLite download. All disease states in the network were ranked according to increasing eigenvector centrality. For each disease classification (for example, ‘neoplasms’), a Kolmogorov–Smirnov test was used to compare the distributions of matching and non-matching disease states within the entire ranked list. After multiple testing correction, disease states that appeared differentially distributed with respect to eigenvector centrality at a 1% FDR were identified and are highlighted in Fig. 4b. HEK293T cells were transfected with Flag–HA–GFP control plasmid, C13orf18–GFP, GFP–BECN1, or RUFY1–Flag–HA plasmids, and, after 48 h, cells were collected in lysis buffer (50 mM Tris pH 7.5, 150 mM NaCl, 1% NP-40), with protease and phosphatase inhibitors (Roche) on ice. Lysates were cleared by centrifugation, and subjected to affinity purification using anti-GFP antibodies (Chromotek, GFP–Trap, GTMA-20) or anti-Flag magnetic beads (Sigma-Aldrich, A2220)) for 2 h at 4 °C. Beads were washed four times with lysis buffer, and subsequently subjected to SDS–PAGE and immunoblotting with the following antibodies: BECN1 (Cell Signaling, clone D40C5), GFP (Roche, mouse IgG clones 7.1 and 13.1), C13orf18 (Proteintech, 21183-1-AP), and HA (Biolegend, clone HA.11). For validation of Hippo pathway interactions within BioPlex 2.0, we performed AP–MS experiments in MCF10A cells. Unlike HEK293T cells, MCF10A cells undergo contact inhibition and activate the Hippo signalling pathway; therefore we used cells under both sub-confluent and confluent conditions wherein YAP1 expulsion from the nucleus was verified by immunofluorescence (see section on ‘Clone construction and cell culture’). Affinity purification was performed essentially as described previously34, but eluted anti-HA immune complexes (Sigma-Aldrich, clone HA-7) were analysed in two ways. First, immune complexes for PDLIM7, MAGI1, YAP1, WWC1, NF2, and MPP5 (replicate 1) were subjected to LC–MS/MS analysis on an LTQ-Velos instrument and HCIPs identified using CompPASS34 in combination with a false positive background dataset derived in MCF10A cells45. The second replicate set for PDLIM7, MAGI1, YAP1, WWC1, NF2, and MPP5, as well as both replicates for PTPN14 and INADL, were processed identically to the first set except that the HA-eluted proteins were reduced and alkylated with DTT and iodoacetamide before trypsin digestion, and all the digested peptides corresponding to one sub-confluent and one confluent anti-HA immunoprecipitation were labelled heavy and light respectively, by reductive dimethylation46. Sub-confluent and confluent sample pairs corresponding to each bait were mixed to normalize the amount of bait present in each heavy and light fraction to 1:1 and analysed on an Orbitrap Elite Hybrid Ion Trap-Orbitrap Mass Spectrometer (ThermoFisher). Complexes from each growth condition were deconvolved using linear discriminant analysis parameters that filtered for either heavy-only or light-only labelled peptides. The heavy- or light-specific search results were subsequently imported into CompPASS for protein interaction analysis. Spectral count and CompPASS score data for the MCF10A dataset is provided in Supplementary Table 10. Anti-PTPN14 antibodies were from Sigma-Aldrich (GW21498A). We used CRISPR–Cas9 gene editing to knockout KIAA0196 using the gRNA sequence (GTCTAAGCCATTTAGACCAA) as described47. The KIAA0196 ORF (a gift from C. Clemen, University of Cologne) was cloned into pLenti-NTAP-IRES-Puro and expressed in KIAA0196−/− cells after selection using puromycin (1 μg ml−1). Immunoprecipitation with anti-Flag (Sigma-Aldrich, M2) antibodies, trypsinization, tandem mass tagging labelling, analysis by mass spectrometry, and quantification were performed as described previously4. Parallel immune complexes or whole-cell lysates were subjected to immunoblotting with anti-WASH1 (Sigma-Aldrich, SAB4200373), anti-KIAA0196 (Santa Cruz Biotechnology, sc-87442), anti-KIAA1033 (Bethyl Labs, A304-919A), anti-CCDC53 (Proteintech, 24445-1-AP), anti-PCNA (Santa Cruz Biotechnology, sc-56), or anti-actin (Santa Cruz Biotechnology, sc-69879) and immunoblot signals quantified using Protein Simple M in biological triplicate. HeLa cells (American Type Culture Collection) were plated on glass coverslips (Zeiss) and transiently transduced with lentiviral vectors expressing C-Flag–HA-tagged baits. At 48 h after infection, cells were fixed with 4% paraformaldehyde for 15 min at room temperature. Cells were washed in PBS, then blocked for 1 h with 5% normal goat serum (Cell Signaling Technology) in PBS containing 0.3% Triton X-100 (Sigma-Aldrich). Coverslips were incubated with anti-HA antibodies (mouse monoclonal, clone HA.11, BioLegend) or anti-HA plus anti-TOMM20 (rabbit polyclonal mitochondrial marker, Santa Cruz Biotechnology, clone FL-145, catalogue number 11415) for 2 h at room temperature in a humidified chamber. Cells were washed three times with PBS, then incubated for 1 h with appropriate Alexa Fluor-conjugated secondary antibodies (ThermoFisher). Nuclei were stained with Hoechst, and cells were washed three times with PBS and mounted on slides using Prolong Gold mounting media (ThermoFisher). All images were collected with a Yokogawa CSU-X1 spinning disk confocal scanner with Spectral Applied Research Aurora Borealis modification on a Nikon Ti-E inverted microscope using a 100 × Plan Apo numerical aperture 1.4 objective lens (Nikon Imaging Center, Harvard Medical School). Confocal images were acquired with a Hamamatsu ORCA-AG cooled CCD (charge-coupled device) camera controlled with MetaMorph 7 software (Molecular Devices). Fluorophores were excited using a Spectral Applied Research LMM-5 laser merge module with acousto-optic tuneable filter (AOTF)-controlled solid-state lasers (488 nm and 561 nm). A Lumencor SOLA fluorescence light source was used for imaging Hoechst staining. z series optical sections were collected with a step size of 0.2 μm, using the internal Nikon Ti-E focus motor, and stacked using MetaMorph to construct maximum intensity projections. We performed three major validation experiments using (1) analysis of a dozen bait proteins in both HCT116 colon cells and HEK293T cells to examine overlap in interaction partners, (2) reciprocal AP–MS experiments directed at interacting proteins for a set of 14-3-3 proteins, and (3) analysis of the PDLIM7–PTPN14–YAP1 adhesion network in MCF10A cells. As a validation approach, we selected 12 largely unstudied proteins displaying a range of interaction partners from 1 to 25 in HEK293T cells and performed AP–MS in HCT116 cells, a cell line of distinct tissue origin from HEK293T cells. After identification of HCIPs for proteins in HCT116 cells, we determined the interactions in common with HEK293T cells (Extended Data Fig. 1b–m). Over the 12 bait proteins identified, we observed 30–100% validation of interactions seen for individual baits in HEK293T cells. Cumulatively, this reflected an overall 60% validation (92 of 147 interactions seen in HEC293T cells were seen in HCT116). This rate of validation is comparable to that seen in focused studies examining F-box protein interactors in these two cell lines (51%)48. Thus, a substantial fraction of interactions seen in HEK293T cells are recapitulated in HCT116 cells. The 14-3-3 proteins represent a well-studied group of seven proteins (YWHAB, YWHAE, YWHAZ, YWHAH, YWHAQ, YWHAG, and SFN) that typically associate with phosphorylated proteins. Thirty-nine baits in BioPlex 2.0 were found to interact with one or more of these 14-3-3 proteins, with YWHAZ being detected most frequently (35 baits) and SFN being detected the least frequently (4 baits) (Extended Data Fig. 2). Seventeen of these proteins are not known to interact with 14-3-3 proteins on the basis of BioGrid. Because only the atypical 14-3-3 protein SFN had been targeted as a bait in BioPlex 2.0, the remaining six 14-3-3 proteins were submitted to our standard AP–MS pipeline using ORFeome 8.1 clones; while the clone for YWHAE failed at the sequence validation stage, the remaining five 14-3-3 proteins were processed successfully, identifying 130–360 HCIPs (Supplementary Table 2). While eight of 39 BioPlex 2.0 baits that had been observed to interact with one or more 14-3-3 proteins were not detected in HEK293T cells and thus may be impossible to detect in reciprocal immunoprecipitations, 63% of interactions eligible for reciprocal detection were confirmed (Extended Data Fig. 2a–c). This demonstrates that BioPlex 2.0 may reliably reveal novel reciprocally interacting partners even for proteins as well studied as 14-3-3 proteins. PTPN14 is a protein phosphatase that has recently been found to associate with several proteins within the Hippo pathway involving the transcription factor YAP1. The Hippo pathway is regulated by contact inhibition, and promotes YAP1 sequestration in the cytoplasm49. BioPlex 2.0 contains a highly connected group of proteins centred on PTPN14, MAGI1, MPP5, LIN7A/C, and INADL (Extended Data Fig. 2d). This network contained several interactions not seen in BioGrid. To validate these interactions, we performed an AP–MS analysis or immunoprecipitation–western analysis of PTPN14, MAGI1, MPP5, PDLIM7, INADL, WWC1, NF2, and YAP1 after stable expression in MCF10A cells in both sub-confluent and confluent states. This series of experiments strongly validated interactions seen in HEK293T cells (Extended Data Fig. 2d, f) with 65% of eligible interactions being seen in both cell lines, further validating our method and the ability of BioPlex 2.0 to robustly identify interactions. Furthermore, 63% of interactions identified in both BioPlex 2.0 and MCF10A cells were novel, having not been previously described in several previous interaction profiling experiments (Extended Data Fig. 2g). Overall, these three lines of study indicate the ability of BioPlex 2.0 to identify interactions that can be validated reciprocally or in other cell lines. The BioPlex 2.0 network and its underlying data are available in several formats. First, all interactions in the BioPlex network have been deposited in the BioGRID protein interaction database. Second, we have created a website devoted to the project (http://bioplex.hms.harvard.edu) which provides tools to download (1) the interactions that make up BioPlex 1.0 and 2.0, (2) a customized viewer that enables browsing of either network to examine the interactions of specific proteins, (3) an interface for download of nearly 12,000 individual RAW files containing mass spectrometry data from individual AP–MS experiments, and (4) an R package and web-based tool for performing CompPASS analyses. Third, the BioPlex 2.0 network as bait–prey pairs has been incorporated into NDEx40, a web-based platform for biological Network Data Exchange. Fourth, our RAW files have been submitted for inclusion in ProteomicsDB50. Finally, all RAW files (3 Tb) from this study will be provided to investigators upon request using investigator-provided hard drives. Finally, a table in.tsv format containing all proteins and spectral count information for all 5,891 AP–MS experiments reported here is available for download at the BioPlex website. All other data are available from the corresponding authors upon reasonable request.
News Article | April 17, 2017
Breaks in DNA can wreak havoc in the body, giving rise to cancer and other health problems. Yet sometimes cells rupture their own DNA for a good reason. During meiosis, when cells divide to become sperm and eggs, making and repairing DNA breaks helps lock together pairs of chromosomes so they can exchange genetic material and continue on their reproductive journey. But even “good” breaks need to be controlled before they get out of hand, and so, once chromosomes have been paired up, something tells the DNA-snapping machinery to shut down. What exactly gives the command, however, has eluded researchers—until now. Studying the reproductive organs of tiny worms called Caenorhabditis elegans, a team of Harvard Medical School scientists has identified a trio of proteins that staff the DNA-break control center. If the same proteins operate the controls in humans, the researchers say, the finding could suggest new ways to rein in runaway DNA breaks throughout the body to avert cancer, infertility, miscarriages and birth defects. Genetics professor Monica Colaiácovo, postdoctoral fellow Saravanapriah Nadarajan and colleagues reported their discoveries in the journal eLife. The team found that a pair of enzymes, polo-like kinases 1 and 2, sense when two chromosomes attach at a DNA break site. The enzymes then begin to sound the “no more breaks needed” alarm by sticking a chemical tag onto proteins called SYP-4. SYP-4 is part of a zipper-like structure that holds chromosome pairs together during meiosis. The researchers watched through a microscope as a wave of this tagging, known as phosphorylation, started at the break site, shown above in green, and spread out, shown in pink, in both directions along the zipper until it reached the ends of the chromosomes. “We think this makes the chromosomes less accessible to the machinery that makes the DNA breaks,” said Colaiácovo. The researchers discovered that phosphorylation not only blocks additional DNA breaks, it also helps stabilize the zipper. “Having a more stable zipper probably helps disseminate the ‘stop’ signal,” said Colaiácovo. Further experiments showed that “when you mess up the ability to modify SYP-4, the cells never stop making double-strand breaks,” Colaiácovo added. As a result, worms with uncontrolled DNA breaks had problems with their eggs that led to infertility or sterility, Nadarajan revealed. Having answered a fundamental question about how DNA breaks are controlled, the researchers are now wondering whether their discoveries apply to humans. A look at sperm and egg precursor cells in mice and humans turned up a promising lead: Proteins that form the equivalent zipper are similarly phosphorylated by polo-like kinases. Colaiácovo and Nadarajan collaborated with two core facilities at HMS to capture and analyze microscopic images of the worms: the Nikon Imaging Center, including director Jennifer Waters and advanced microscopy research associate Talley Lambert (both co-authors of the study), and the Image and Data Analysis Core, including director Hunter Elliott.
Nayak T.R.,National University of Singapore |
Andersen H.,National University of Singapore |
Makam V.S.,National University of Singapore |
Khaw C.,Nikon Imaging Center |
And 7 more authors.
ACS Nano | Year: 2011
Current tissue engineering approaches combine different scaffold materials with living cells to provide biological substitutes that can repair and eventually improve tissue functions. Both natural and synthetic materials have been fabricated for transplantation of stem cells and their specific differentiation into muscles, bones, and cartilages. One of the key objectives for bone regeneration therapy to be successful is to direct stem cells proliferation and to accelerate their differentiation in a controlled manner through the use of growth factors and osteogenic inducers. Here we show that graphene provides a promising biocompatible scaffold that does not hamper the proliferation of human mesenchymal stem cells (hMSCs) and accelerates their specific differentiation into bone cells. The differentiation rate is comparable to the one achieved with common growth factors, demonstrating graphenes potential for stem cell research. © 2011 American Chemical Society.
Heinrich D.,Leiden University |
Heinrich D.,Fraunhofer Institute for Silicate Research |
Ecke M.,Max Planck Institute of Biochemistry |
Jasnin M.,Max Planck Institute of Biochemistry |
And 2 more authors.
Biophysical Journal | Year: 2014
Membrane pearling in live cells is observed when the plasma membrane is depleted of its support, the cortical actin network. Upon efficient depolymerization of actin, pearls of variable size are formed, which are connected by nanotubes of ∼40 nm diameter. We show that formation of the membrane tubes and their transition into chains of pearls do not require external tension, and that they neither depend on microtubule-based molecular motors nor pressure generated by myosin-II. Pearling thus differs from blebbing. The pearling state is stable as long as actin is prevented from polymerizing. When polymerization is restored, the pearls are retracted into the cell, indicating continuity of the membrane. Our data suggest that the alternation of pearls and strings is an energetically favored state of the unsupported plasma membrane, and that one of the functions of the actin cortex is to prevent the membrane from spontaneously assuming this configuration. © 2014 Biophysical Society.
Castello M.,Instituto Italiano Of Tecnonogia |
Castello M.,University of Genoa |
Lanzano L.L.,Instituto Italiano Of Tecnonogia |
Coto Hernandez I.,Instituto Italiano Of Tecnonogia |
And 6 more authors.
Progress in Biomedical Optics and Imaging - Proceedings of SPIE | Year: 2015
In a stimulated emission depletion (STED) microscope the region from which a fluorophore can spontaneously emit shrinks with the continued STED beam action after the excitation event. This fact has been recently used to implement a versatile, simple and cheap STED microscope that uses a pulsed excitation beam, a STED beam running in continuous-wave (CW) and a time-gated detection: By collecting only the delayed (with respect to the excitation events) fluorescence, the STED beam intensity needed for obtaining a certain spatial resolution strongly reduces, which is fundamental to increase live cell imaging compatibility. This new STED microscopy implementation, namely gated CW-STED, is in essence limited (only) by the reduction of the signal associated with the time-gated detection. Here we show the recent advances in gated CW-STED microscopy and related methods. We show that the time-gated detection can be substituted by more efficient computational methods when the arrival-times of all fluorescence photons are provided. © 2015 SPIE.
Rebscher N.,University of Marburg |
Lidke A.K.,University of Marburg |
Ackermann C.F.,Nikon Imaging Center
EvoDevo | Year: 2012
Background: In the polychaete Platynereis, the primordial germ cells (PGCs) emerge from the vasa, piwi, and PL10 expressing mesodermal posterior growth zone (MPGZ) at the end of larval development, suggesting a post-embryonic formation from stem cells.Methods: In order to verify this hypothesis, embryos and larvae were pulse labeled with the proliferation marker 5-ethynyl-2'-deoxyuridine (EdU) at different stages of development. Subsequently, the PGCs were visualized in 7-day-old young worms using antibodies against the Vasa protein.Results: Surprisingly, the primordial germ cells of Platynereis incorporate EdU only shortly before gastrulation (6-8 hours post fertilization (hpf)), which coincides with the emergence of four small blastomeres from the mesoblast lineage. We conclude that these so-called 'secondary mesoblast cells' constitute the definitive PGCs in Platynereis. In contrast, the cells of the MPGZ incorporate EdU only from the pre-trochophore stage onward (14 hpf).Conclusion: While PGCs and the cells of the MPGZ in Platynereis are indistinguishable in morphology and both express the germline markers vasa, nanos, and piwi, a distinct cluster of PGCs is detectable anterior of the MPGZ following EdU pulse-labeling. Indeed the PGCs form independently from the stem cells of the MPGZ prior to gastrulation. Our data suggest an early PGC formation in the polychaete by preformation rather than by epigenesis. © 2012 Rebscher et al; licensee BioMed Central Ltd.
Hernandez I.C.,Italian Institute of Technology |
Hernandez I.C.,University of Genoa |
Buttafava M.,Polytechnic of Milan |
Boso G.,Polytechnic of Milan |
And 6 more authors.
Biomedical Optics Express | Year: 2015
Stimulated emission depletion (STED) microscopy provides fluorescence imaging with sub-diffraction resolution. Experimentally demonstrated at the end of the 90s, STED microscopy has gained substantial momentum and impact only in the last few years. Indeed, advances in many fields improved its compatibility with everyday biological research. Among them, a fundamental step was represented by the introduction in a STED architecture of the time-gated detection, which greatly reduced the complexity of the implementation and the illumination intensity needed. However, the benefits of the time-gated detection came along with a reduction of the fluorescence signal forming the STED microscopy images. The maximization of the useful (within the time gate) photon flux is then an important aspect to obtain super-resolved images. Here we show that by using a fast-gated single-photon avalanche diode (SPAD), i.e. a detector able to rapidly (hundreds picoseconds) switch-on and -off can improve significantly the signal-to-noise ratio (SNR) of the gated STED image. In addition to an enhancement of the image SNR, the use of the fast-gated SPAD reduces also the system complexity. We demonstrate these abilities both on calibration and biological sample. The experiments were carried on a gated STED microscope based on a STED beam operating in continuous-wave (CW), although the fast-gated SPAD is fully compatible with gated STED implementations based on pulsed STED beams. © 2015 Optical Society of America.
Saliba A.-E.,University Pierre and Marie Curie |
Saias L.,University Pierre and Marie Curie |
Psychari E.,University Pierre and Marie Curie |
Psychari E.,French Institute of Health and Medical Research |
And 15 more authors.
Proceedings of the National Academy of Sciences of the United States of America | Year: 2010
We propose a unique method for cell sorting, "Ephesia," using columns of biofunctionalized superparamagnetic beads self-assembled in a microfluidic channel onto an array of magnetic traps prepared by microcontact printing. It combines the advantages of microfluidic cell sorting, notably the application of a well controlled, flow-activated interaction between cells and beads, and those of immunomagnetic sorting, notably the use of batch-prepared, well characterized antibody-bearing beads. On cell lines mixtures, we demonstrated a capture yield better than 94%, and the possibility to cultivate in situ the captured cells. Asecond series of experiments involved clinical samples - blood, pleural effusion, and fine needle aspirates - issued from healthy donors and patients with B-cell hematological malignant tumors (leukemia andlymphoma). The immunophenotype and morphology of B-lymphocytes were analyzed directly in the microfluidic chamber, and compared with conventional flow cytometry and visual cytology data, in a blind test. Immunophenotyping results using Ephesia were fully consistent with those obtained by flow cytometry.We obtained in situ high resolution confocal three-dimensional images of the cell nuclei, showing intranuclear details consistent with conventional cytological staining. Ephesia thus provides a powerful approach to cell capture and typing allowing fully automated high resolution and quantitative immunophenotyping and morphological analysis. It requires at least 10 times smaller sample volume and cell numbers than cytometry, potentially increasing the range of indications and the success rate of microbiopsy-based diagnosis, and reducing analysis time and cost.
Hegge S.,University of Heidelberg |
Munter S.,University of Heidelberg |
Steinbuchel M.,University of Heidelberg |
Heiss K.,University of Heidelberg |
And 4 more authors.
FASEB Journal | Year: 2010
Adhesion of eukaryotic cells is a complex process during which interactions between extracellular ligands and cellular receptors on the plasma membrane modulate the organization of the cytoskeleton. Pathogens particularly rely often on adhesion to tissues or host cells in order to establish an infection. Here, we examined the adhesion of Plasmodium sporozoites, the motile form of the malaria parasite transmitted by the mosquito, to flat surfaces. Experiments using total internal reflection fluorescence microscopy and analysis of sporozoites under flow revealed a stepwise and developmentally regulated adhesion process. The sporozoite-specific transmembrane proteins TRAP and S6 were found to be important for initial adhesion. The structurally related protein TLP appears to play a specific role in adhesion under static conditions, as tlp(-) sporozoites move 4 times less efficiently than wild-type sporozoites. This likely reflects the decreased intradermal sporozoite movement of sporozoites lacking TLP. Further, these three sporozoite surface proteins also act in concert with actin filaments to organize efficient adhesion of the sporozoite prior to initiating motility and host cell invasion. © FASEB.
Castello M.,Italian Institute of Technology |
Castello M.,University of Genoa |
Diaspro A.,Italian Institute of Technology |
Diaspro A.,Nikon Imaging Center |
Vicidomini G.,Italian Institute of Technology
Applied Physics Letters | Year: 2014
Time-gated detection, namely, only collecting the fluorescence photons after a time-delay from the excitation events, reduces complexity, cost, and illumination intensity of a stimulated emission depletion (STED) microscope. In the gated continuous-wave- (CW-) STED implementation, the spatial resolution improves with increased time-delay, but the signal-to-noise ratio (SNR) reduces. Thus, in sub-optimal conditions, such as a low photon-budget regime, the SNR reduction can cancel-out the expected gain in resolution. Here, we propose a method which does not discard photons, but instead collects all the photons in different time-gates and recombines them through a multi-image deconvolution. Our results, obtained on simulated and experimental data, show that the SNR of the restored image improves relative to the gated image, thereby improving the effective resolution. © 2014 AIP Publishing LLC.