Defining the Human Deubiquitinating Enzyme Interaction Landscape

Systematic proteomics of deubiquitinating enzymes

Interaction proteomics provides the most direct approach available for identifying physiologically relevant protein complexes and networks but is often difficult to apply to large collections of proteins due to the experimental resources required and the large number of nonspecific interacting proteins that typically dominate the mass spectral analysis of immune complexes (Ewing et al., 2007). In order to address these issues and begin to define the landscape of the Dub interactome we have established a facile platform for interaction proteomics which includes a readily scalable expression and purification strategy using a retroviral mediated expression library of 75 Flag-HA tagged Dubs (Table S2) in combination with new methodology for defining HCIPs from parallel LC-MS/MS datasets (Figure 1A and Supplemental Methods). Sixty-nine of the 75 Dubs analyzed in this study are endogenously expressed in human embryonic kidney (HEK) 293 cells (Figure S1), indicating that this is a suitable system in which to systematically identify proteins associated with Dubs. Each Dub was purified using anti-HA antibody coupled resin and trypsinized complexes were subjected directly to LC-MS/MS in duplicate (Supplemental Methods) to create a database of Dub associated proteins. Flag-HA-Dub expression varied over a wide range, yet no correlation was observed between Dub protein levels and their number of HCIPs determined as described below (a correlation would be expected if overexpression consistently led to increases in non-physiological interactions) (Figure 1B,C). This also demonstrates that we were able to identify HCIPs over a wide range of Dub expression levels.

The identification of bona fide interactions remains an outstanding problem for proteomic analysis of immune complexes and towards this end, multiple methods have been developed (Collins et al., 2007; Ewing et al., 2007; Gavin et al., 2002; Krogan et al., 2006; Sardiu et al., 2008). Although these methods have been successfully applied to specific data, certain limitations exist, especially with regards to non-reciprocal datasets such as observed with our Dub dataset (see Supplemental Text). As an alternative to these approaches, we have developed an unbiased comparative approach to identify HCIPs in parallel proteomic datasets using a software platform called CompPASS. CompPASS performs better on non-reciprocal datasets than existing methods and does not rely on a training set of “gold standard” or previously reported interactions, and does not require the expense associated with the use of stable isotopic labeling with amino acids in culture (SILAC) for background protein identification (For a comparison of CompPASS and other methods, see Supplemental Text, Figures S4, S5, and Table S1).

CompPASS: A Software Platform for Comparative Proteomic Analysis

CompPASS is composed of an automated MS/MS data processing component, a protein function/annotation component, and an interaction network analysis component, which together, form an integrated platform for analyzing parallel proteomic data (Figures 1D,E, S2 and Supplemental Methods). The centerpiece of CompPASS is the data processing component, which employs an unbiased comparative methodology to assign scores to proteins identified within parallel proteomic datasets.

Our approach employs two scoring metrics which are calculated based on a “stats table” where the columns are individual IP-MS/MS experiments, the rows are bait-associated proteins, and each element is populated with the total spectral counts (TSCs; used to approximate protein abundance (Liu et al., 2004)) for each identified protein from each IP-MS/MS experiment (Figure 1E). The first metric is the conventional Z-score (Figure 1E, equations 1 and 2), which is most useful when analyzing proteins that are present in multiple immune complexes but are found at much higher levels in a subset of these. A drawback of the Z-score is that it equally weights unique interactors regardless of their TSC. To address this issue, we have devised a new metric, which we term the D-score (Figure 1E, equation 3). The D-score incorporates the uniqueness, the abundance of the interactor (TSC), and the reproducibility of the interaction to assign a score to each protein within each IP (the same protein in two different IPs may have distinct D-scores since its TSC and reproducibility could differ). In this way, the highest scores are given to proteins in each IP that are found rarely, found in duplicate runs, and have high TSCs – all characteristics of proteins that would be considered candidate interactors. A global D-score threshold (D^T) is determined and all raw D-scores (D^R) are normalized to this value, producing D^N-scores (Figure 1E, equation 4, and Figure S3; Supplemental Methods). Interactors in each IP with a D^N-score ≥ 1 are considered HCIPs while those with D^N-score < 1 are less likely to be bona fide interactors (examination of Z-scores in this case can facilitate the identification of candidate interactors). All data reported here were calculated from a stats table populated with data derived from 76 immune complexes (75 Dubs and EGFP as a control), analyzed in duplicate (Table S3).

Application of CompPASS to Macromolecular and Distributive Complexes

The ability of Z- and D^N-scores to identify HCIPs within the context of a macromolecular complex is exemplified by analysis of proteins associated with COPS5 and COPS6, active and inactive JAMM motif components of the COP9/signalosome complex, respectively (Cope and Deshaies, 2003). When ranked by D^N-score, the 9 known subunits of the signalosome (COPS1-6, COPS7A and COPS7B, and COPS8) are found within the top 20 (of 284) COPS6-associated proteins, with 6 of the top 10 proteins being core signalosome components (Figure 2A,B). The best understood role of the signalosome is to remove the ubiquitin-like protein Nedd8 from the Cullin family of E3 ubiquitin ligases, an event catalyzed by the COPS5 subunit (Cope and Deshaies, 2003), and accordingly, 42 of the top 50 D^N-score ranked non-core signalosome proteins identified in association with COPS6 include Cullins, Nedd8, and Cullin-based substrate adaptors (including F-Box and BTB proteins not previously known to bind CUL1 or CUL3; Figure 2D,H). The cullin adaptors captured here may represent either the most abundant adaptors or those with the highest residence time on the signalosome. While D^N-scores are useful in identifying HCIPs from relatively unique interacting proteins, Z-scores allow for the identification of candidate interacting proteins that are found frequently across multiple baits. For example, CUL3 is detected in 32% of all Dub LC-MS/MS datasets as well as the EGFP control dataset, but its Z-score (5.79) identifies CUL3 as a candidate interactor for COPS6 but not for EGFP (Figure 2B and Table S3). Analysis of HCIPs for COPS5 revealed all but 3 of the proteins found in the COPS6 analysis (Fig 2F), demonstrating the usefulness of our platform in identifying constituents of multi-protein complexes in a reciprocal manner.

We evaluated several other Dubs known to form multi-protein complexes to assess the accuracy of our scoring criteria. While this work was in process, USP22 was shown to associate with several subunits of the SAGA transcriptional regulatory complex (Zhang et al., 2008; Zhao et al., 2008). Of the 28 HCIPs from 384 total proteins identified in association with USP22, 14 were either known SAGA components, paralogs of known SAGA components (ATXN7L2), or proteins that interact with SAGA components in the yeast two-hybrid (Y2H) system (CCDC101 and CNOT1) (Figure 2G). Further validating our approach, all 21 subunits of the 19S regulatory sub-complex of the proteasome (Pickart and Cohen, 2004) were identified in HCIPs from USP14, PSMD14, UCHL5, and PSMD7, with the exception of the ADRM1 subunit which was found in 3 of 4 proteasome Dub complexes (Figure 2H, Table S3). With USP14, for example, 23 of the top 25 (from 232) proteins ranked by D^N-Score were 19S or 20S subunits (Table S3). For a discussion of 4 additional well-studied Dubs, see Supplemental Text and Figure S12.

To evaluate CompPASS for the identification of HCIPs within the context of distributive complexes (i.e. baits with a variety of interactors that tend not to interact with each other) we investigated USP11 using 4 independent biological replicates, and three additional Dubs with biological replicates. On average, complete datasets from biological replicates showed a 65% overlap with each other but only a 35% overlap with EGFP control complexes (Figure 2C). Examination of HCIPs derived from the USP11 and 3 additional Dub datasets revealed an 80% overlap among the biological replicates (Figure 2C) and greater than 50% of HCIPs were found in at least 3 out of the 4 USP11 biological replicates (Figure 2D,E). As expected, proteins with higher TSCs were more reproducibly identified (Figure 2D), but reproducibility itself does not accurately specify a candidate interactor since the majority of proteins found in all four USP11 replicates were deemed non-specific by our metrics (Figure 2E). Taken together, these results demonstrate the proteomic methodology used here, together with CompPASS, allows us to filter out common false positive interacting proteins and identify a subset of associated proteins, HCIPs, that are candidates for bona fide interacting proteins in both macromolecular and distributive complexes. In practice, the combined use of Z- and D^N-scores provides the most effective approach for identifying HCIPs (Figures 2A, S7).

An Overview of the Human Dub Interaction Landscape

Out of the 2,458 uniquely identified proteins in our Dub dataset, 774 were found to be HCIPs associated with 75 Dubs (Figures 3A, S8, and Table S3), with an overall validation rate of 68% as described below. The diversity of interacting proteins among Dub complexes, as well as the specificity for individual Dubs, is illustrated by hierarchical clustering of HCIPs and Dubs (Figure 3A). The vast majority of HCIPs are found only in complexes with a single Dub, the exceptions being Dubs associated with the signalosome, the EIF3 complex, and the proteasome. Additionally, two sets of Dubs were found to share a significant number of HCIPs: USP12 and USP46 share 5 of their top ranked interacting proteins (WDR48, WDR20, DMWD, PHLPP, and PHLPPL), while USP4, USP15, and USP39 share 14% of their HCIPs, all of which are linked to mRNA processing (Figure 3A). Analysis of HCIPs that cluster together for an individual Dub can reveal proteins that participate in a complex, such as found with USP22 (Figure 3A). Similarly, Dubs that associate with multiple members of a protein family can also be identified, such as USP49, which interacts with all three members of the Centrin protein family and three members of the SAPS domain family of phosphatase activators (Figure 3A).

Network Topology of the Dub Interaction Landscape

We next used protein-protein interaction (PPI) database tools within CompPASS (Figure 1D) to organize Dubs into distinct topological groups based on the number of HCIPs for each Dub and the number of interactions among HCIPs (Figures 3B and S9). Seven topological groups, each representing different network profiles, were created and can provide insight into the biological context in which member Dubs likely function. On one extreme are Group 1 Dubs (28 members), which have ≤6 HCIPs, and no known interactions among HCIPs (Figures 3B and S9D). Group 1 Dubs include USP28, which is known to form a heterodimer with 53BP1 (Zhang et al., 2006). On the other extreme are Group 2 and Group 3 Dubs, which have >2 dozen HCIPs with significant interconnectivity among them (Figures 3B and S9E,F). These include Dubs that interact with the proteasome (USP14, UCHL5, PSMD7, PSMD14), the EIF3 complex (EIF3S3, EIF3S5), the SAGA complex (USP22), and newly identified interactions between Dubs and mRNA processing (USP4, USP15, USP39), transcriptional (JOSD3), and protein phosphatase scaffolding (ZRANB1/TRABID) complexes (Figure 3B). Group 6 Dubs contain ≥9 HCIPs with little known connectivity (Figures 3B and S9I). However, we demonstrate below that HCIPs for the Group 6 Dubs USP12 and USP46 form multi-protein complexes, suggesting that the interconnectivity among HCIPs for distributive Dubs may be underestimated.

Gene Ontology Analysis of the Dub Interaction Landscape

In order to begin to place these 75 Dubs within a putative biological context, we used tools within CompPASS to organize GO process and component descriptions for each of the Dub’s associated HCIPs (Figure 4A–C, Supplemental Methods and Table S8). GO process analysis linked Dubs to a wide variety of functions, with 34% of Dubs containing ≥3 HCIPs having GO terms associated with the ubiquitin system (Figure 4C). Five-hundred and fifty-two unique protein interaction domains (Pfam) were identified among 774 HCIPs including some domain types (e.g. WD40) that were highly enriched among HCIPs (Figure S15). As described below, the WD40 protein WDR48 has recently been demonstrated to function as an activator of several Dubs (Cohn et al., 2008; Cohn et al., 2007).

Placement of Dubs Within Biological and Cellular Modules

Given the GO analysis discussed above and the extent of validation observed, we developed a cellular map of where and in what processes Dubs are likely to function (Figure 4D). Functional modules were ascribed based on the presence of ≥3 HCIPs being in a particular GO category. However, given the array of interacting proteins identified for certain Dubs (such as Group 6 and 7 Dubs), many Dubs are likely involved with multiple distinct pathways. Dubs with limited HCIPs or HCIPs with disparate GO classifications were localized within the cell by stable expression of GFP-Dub fusions in either Hela or 293T cells (Figures 4D and S16). Interestingly, one third of the Dubs examined in this study can be placed within the nucleus based on either GFP localization or GO assignment (Figure 4A,D). Among these are 6 Dubs (USP11, USP22, USP7, JOSD3, BRCC36, and BAP1) that can be placed within transcriptional or DNA damage modules (Figures 4D, S12, and Supplemental Text). In addition, 3 Dubs (USP39, USP4, and USP15) are implicated in mRNA processing by virtue of their association with U5/U6-snRNP components. Other Dubs, including TNFAIP3/A20, USP21, USP46, USP12, ZRANB1, and USP43 are linked with cytoplasmic phosphorylation-based signaling systems (Figures 4D, S12, and Supplemental Text).

Dub Interactome Validation

Three independent approaches were used to validate our Dub interaction dataset. First, we stably expressed 25 Flag-HA-tagged HCIPs found among 18 Dubs and performed IP-MS/MS on purified complexes. Of the 40 binary interactions represented by this subset of HCIPs, 33 reciprocal interactions (83%) were identified (Figure 5; Table S4). Second, a further set of 20 Myc-tagged HCIPs (associated with 5 Dubs) or an additional set of 5 Myc-tagged Dubs were expressed in either 293T cells stably expressing the appropriate HA-tagged Dub or 293T cells for detection of the endogenous HCIP. Immunoblotting of anti-Myc immune complexes validated 14 of 29 expected interactions (Figures 5 and S10). In 6 of 6 cases tested, both IP-MS/MS and IP-Western methods gave identical results (Figure 5). Third, we also validated 3 out of 6 binary interactions using endogenous co-IP methods (Figures 5 and S10). In total, 45 of 66 (68%) tested interactions were validated using at least one independent method. Of note is the recent demonstration that other commonly used interaction mapping methods such as Y2H are only able to detect 30% of known ‘gold-standard’ binary interactions (Braun et al., 2009).

To determine what percentage of published Dub interacting proteins we identify in our Dub interaction landscape, we employed the BioGRID PPI database (Stark et al., 2006) as well as manual literature curation to identify reported interactions with Dubs. In total, this identified 332 interactions associated with 51 Dubs (Table S7). Due to the inherent differences in the amount of false-positive identifications among various interaction mapping methodologies (Cusick et al., 2009), we categorized these data by experimental system to separate interactions determined by co-purification and reconstitution methods from those identified using Y2H. Using the most stringent criteria (endogenous Co-IP or Co-purification), 71% of 91 interactions in this category were found in our analysis, and >90% of these had D^N-scores ≥1 (Figure S11, Table S7). When the less stringent category of overexpression Co-IP was examined, 36% of previous interactions were found. In contrast, only 4.6% of interactions that were previously observed only by Y2H were found in our dataset (Figure S11). This value is somewhat lower than the overlap seen between independent large-scale Y2H screens (~11–15%) (Futschik et al., 2007), as expected given that many interactions are identified in only a single screen.

As a final test of the robustness and reproducibility of our methodology, we performed IP-MS/MS experiments on 11 Dubs stably expressed in HCT116 cells. Fifty-four percent of interactors that have D^N-scores ≥1 in our 293T dataset are also present when the same Dub was isolated from HCT116 cells (Figure S6). This is increased to 63% when considering proteins with p-values < 0.01 (Figures S3, S6, Table S5, and Supplemental Text). Closer examination of distributive Dub complexes like USP11 and USP20 reveals that even interacting proteins originally identified with <10 TSCs are identified when the same Dub was isolated from HCT116 cells (Figure S6C,D). Taken together, these validation experiments demonstrate the efficacy of our methodology and the ability of our metrics in CompPASS to strongly enrich for bona fide interacting proteins.

The Dub Interactome Reveals New Regulatory Networks Linked with Core Cellular Functions

In order to explore the connectivity of interaction networks containing both Dubs and their associated proteins, we performed a reverse proteomic analysis of HCIPs found in several Dub immune-complexes. One such network, which we refer to as the DPW (Dub, Phosphatase, WD40) network, contains 3 Dubs whose catalytic domains are most closely related to each other in primary sequence (USP1, USP12, USP46), WDR48, 2 related protein phosphatases (PHLPP, PHLPPL) previously reported to dephosphorylate Akt (Mendoza and Blenis, 2007), and 2 closely related but previously unstudied WD40 proteins (WDR20 and DMWD) (Figure 6A and Table S4). Both USP12 and USP46 were found to individually associate with all 3 WD40 proteins and both phosphatases, while USP1 associated with WDR48 and PHLPP but not WDR20, DMWD, or PHLPPL. Reciprocal IP-MS/MS suggests the existence of multiple distinct Dub complexes in which USP12 (or USP46) interacts with WDR48, WDR20 (or DMWD), and PHLPP (or PHLPPL) (Figure 6A, Table S4). In contrast, USP1 forms complexes with WDR48 and PHLPP with exclusion of WDR20 and DMWD. WDR48/UAF1 is an activator of USP1 (Cohn et al., 2007) and more recently has been shown to activate USP12 and USP46 (Cohn et al., 2008). The finding that 36% of Dubs are associated with WD40 proteins (Figure S15) suggests that this type of activating mechanism may be more prevalent than currently appreciated.

Previous studies have revealed the involvement of ubiquitin conjugation and removal in the mRNA splicing pathway via the U5/U6-snRNP complex in budding yeast (Bellare et al., 2008). Consistent with these observations, we identified 3 Dubs that associate with components of the U5/U6-snRNP as well as other components involved in mRNA processing (Figure 6B, Table S3). USP39 was previously shown to associate with BCDIN3, a 7SK snRNA methylphosphate capping enzyme that associates with U5/U6-snRNP (Jeronimo et al., 2007). USP39 complexes contain 11 known subunits of U5/U6-snRNP suggesting a role for USP39 in mRNA splicing. In contrast, USP4 and USP15 associated with 5 subunits of the U5/U6-snRNP (PRPF31, PRPF3, BCDIN3, PRPF4, and PPIH/cyclophilin H) as well as 5 and 4 subunits, respectively, of the LSM mRNA binding complex (LSM2-8). Interestingly, both USP4 and USP15 associated with Terminal Uridylyl Transferase (TUT1) previously implicated in 3′ uridylation of U6 snRNA (Trippe et al., 2006), but not known to associate with the U5/U6-snRNP complex. In reciprocal LC-MS/MS experiments, SART3 complexes contained the majority of U5/U6-snRNP components as well as USP39 and USP4, and PRPF4 complexes contained USP39 and USP15 (Figure 6B, Table S4). Differences in associated proteins suggest that these Dubs play distinct roles in ubiquitin-dependent control of mRNA splicing and/or decay.

The ubiquitin system plays a major role in the response of cells to DNA damage. BRCC36 and its regulatory subunit BRE (BRCC45) have been shown to associate with BRCA1 as well as the Abraxis complex containing CCDC98/Abraxis and UIMC1/RAP80 (Sobhian et al., 2007; Wang et al., 2007) which uses its ubiquitin interaction motifs (UIMs) to target it to ubiquitinated proteins at DNA double-strand breaks. We identified two additional proteins in association with BRCC36 not previously linked to the BRCC36 complex: KIAA0157 (ABRO), a protein related to Abraxis (Wang et al., 2007), and HSCP142, a protein of unknown function lacking previously described protein interaction domains (Figure 6C). IP-MS/MS analysis of BRE, HSPC142, and UIMC1 revealed extensive cross-talk between components of the Abraxis complex (Figure 6C). Recently, HSPC142 was independently identified and demonstrated to localize to sites of DNA damage and to be required for loading of the Abraxis complex on double-strand DNA breaks (Wang, 2009), providing further functional validation of our metrics. Finally, a number of proteins were found associated with 2 or more components of the BRCC36 complex, suggesting a potential role for these proteins in the DNA damage response (Figure 6C).

Budding yeast Ubp3p and its putative substrate targeting subunit Bre5p control autophagic degradation of ribosomes (ribophagy) (Kraft et al., 2008). We found that USP10, the human ortholog of Ubp3p, associates with the Bre5 ortholog G3BP1, as previously reported (Table S7), and also associates with the G3BP1 paralog G3BP2 (Figure 6D). Proteomic analysis of G3BP1 and 2 revealed USP10 in addition to 19 interacting proteins in common with G3BP1 and 2 (Figure 6D). Based on these four examples, we hypothesize that Dub activity may be important for a larger fraction of core cellular functions than previously appreciated.

Functional Validation of VCP Interacting Dubs Reveals a Role for USP13 in ERAD

A reoccurring theme throughout our Dub interaction landscape is the association of Dubs with components of the ubiquitin conjugation machinery, signifying a high degree of cross-regulation within the ubiquitin system (Figures 7A,B, S13). Indeed, 26 Dubs contain HCIPs with Pfam domains linking them with ubiquitin conjugation, including HECT, RING, and cullin-based E3 ubiquitin ligases (Figure 7A,B). It is conceivable that Dubs act broadly upon particular E3s and/or their targets to promote their stabilization or activity, as has been suggested for MDM2-USP7, APC^CDC20-USP44, and KPC1-USP19 (Lu et al., 2009; Stegmeier et al., 2007a; Ventii and Wilkinson, 2008). Subsets of Dubs also interact with other components of the ubiquitin system (Figure 7A–C). A case in point is the AAA ATPase VCP/p97, which plays roles in ubiquitin binding, conjugation, and deconjugation, and is essential for ER-associated degradation (ERAD) of misfolded proteins (Vembar and Brodsky, 2008). VCP peptides were identified in 32 Dub immune-complexes, but was particularly abundant (>15 TSCs) in YOD1, USP13, USP50, and VCPIP1 datasets (Figure 7C). Among the Dubs we found in complex with VCP, VCPIP1 and Ataxin-3 have been previously demonstrated to interact with VCP (Uchiyama et al., 2002; Wang et al., 2006) and YOD1 is the human ortholog of S. cerevisiae Otu1 which has also been shown to bind VCP (Rumpf and Jentsch, 2006).

We therefore focused our attention on USP13 as it had no known cellular role and was isolated not only with VCP, but also with three VCP interacting proteins, UFD1, NPL4, and ETEA/UBDX8 (Figure 7D), which are thought to serve as adaptors to link VCP to targets (Vembar and Brodsky, 2008). Based on these proteins being identified as HCIPs for USP13, we predicted that USP13 would likely play a role within the ERAD pathway and tested this using a stable cell line expressing TCRαGFP, a well-established model ERAD substrate (DeLaBarre et al., 2006). As expected, knockdown of VCP, PSMD7, and PSMD14 all resulted in a more than 8-fold increase in TCRαGFP levels (Figure 7E,F). Depletion of USP13 using two separate siRNAs resulted in a 1.5–2 fold accumulation of TCRαGFP relative to control siRNA transfections, while depletion of Hrd1, the known E3 for TCRαGFP (Kikkert et al., 2004) only caused a 1.2–1.5 fold increase (Figure 7E,F). Depletion of the other VCP interacting Dubs, USP50, YOD1, and VCPIP1, as well as USP10 (which did not interact with VCP) had no effect on TCRαGFP levels, while depletion of Ataxin-3 caused a small, but consistent 1.25-fold decrease in TCRαGFP, as expected (Wang et al., 2006) (Figure 7E; depletion was validated by western blotting shown in Figure S14E).

Interestingly, knockdown of USP13, VCP, PSMD7, or PSMD14 all resulted in elevated fluorescence levels of a ubiquitin-dependent cytoplasmic GFP reporter based on the CL1 degron (Bennett et al., 2005) (Figure S14B) consistent with an additional role for a VCP complex in the degradation of non-ER associated substrates (Wang et al., 2008). In contrast, depletion of USP13 or VCP had no effect on a cytoplasmic ubiquitin-independent GFP reporter based on the ornithine decarboxylase degron (Figure S14A). Consistent with destruction of this reporter via the proteasome, siRNAs targeting PSMD7 and PSMD14 resulted in increased fluorescence of this reporter similar to that seen with MG132 treatment (Figure S14A).

To further investigate the role of USP13 in ERAD, we utilized a cell viability assay in combination with tunicamycin, a drug that inhibits glycosylation within the ER and initiates the unfolded protein response leading to cell death (Malhotra and Kaufman, 2007). Depletion of PSMD7, PSMD14, VCP, or USP13 resulted in hypersensitivity to tunicamycin-mediated cell death relative to the amount detected with control siRNA transfections (Figure 7G). Depletion of VCPIP1 or USP10 had no significant effect on cell viability after tunicamycin treatment; however depletion of Ataxin-3 resulted in a greater than 2.5-fold resistance to tunicamycin (Figure 7G). The phenotypes of USP13 and Ataxin-3 depletion in the tunicamycin sensitivity and TCRαGFP stability assays indicated that Ataxin-3 and USP13 have opposing functions within ERAD. Our results, as well as previously published reports regarding Ataxin-3 (Wang et al., 2006), indicate that Ataxin-3 directly deubiquitinates ERAD substrates such as TCRαGFP and thus removal of this Dub would allow increased E3 ligase activity, accelerated protein turnover, and increased flux through the ERAD pathway. Because loss of USP13 phenocopies loss of VCP we hypothesize that USP13 functions in regulating the activity of the VCP complex itself and/or other ubiquitin ligases that promote TCRαGFP turnover.

Concluding Remarks

This study provides the first glimpse of the Dub interactome by identifying stably associated proteins for 75% of the Dubs encoded by the human genome; significantly increasing the number of validated and candidate Dub associated proteins. The majority of Dubs examined in this study can either be placed within a large multi-subunit complex or in the context of a cellular process based on established annotation for their interactors. Indeed, only one-third of the Dubs in our study cannot be assigned a candidate functional role due to either a small number of HCIPs or the absence of GO annotation. The finding that Dubs are frequently associated with multi-protein complexes was somewhat unexpected. On one hand, Dubs associated with such complexes may function to regulate the activity of the complex by removing regulatory ubiquitination events. Alternatively, Dub associated proteins may regulate the activity or substrate specificity of the Dub, as appears to be the case with WDR48. Importantly, our analysis failed to identify substrates for the small number of Dubs where such targets are known (Ventii and Wilkinson, 2008), likely due to the transient nature of the Dub-substrate interaction and/or the low steady-state abundance of these proteins. Additional methods, such as utilizing catalytically inactive Dubs as a basis for complex identification, a method previously used to identify histone H2A as a substrate of USP3 (Nicassio et al., 2007), will likely be required to identify Dub substrates. In this regard, initial experiments using this methodology identified histone H2A as an HCIP for catalytically inactive USP3^C168S (unpublished data), whereas H2A was not found in the wild-type USP3 IP-MS/MS analysis. It is also possible that distinct Dub associated proteins may have been missed in our analysis if specific stimuli are required for this interaction to occur.

Finally, the analysis of the Dub interactome has been accomplished through the development and use of CompPASS, an integrated platform for analyzing proteomic data. Utilizing protein network analysis tools contained within CompPASS, we predicted a role for USP13, a Dub with no previously known cellular role, within the ERAD pathway. We tested and validated this prediction through functional studies. The tools within CompPASS are applicable to proteomic investigations ranging from focused studies on a small number of selected proteins to the analysis of entire protein families or biological regulatory networks.

Protein Expression, Purification, and Functional Validation

Open reading frames for Dubs and associated proteins (Table S2) were cloned into constitutive or inducible retroviral vectors and expressed in HEK cells (see Supplemental Experimental Procedures). For purification, ~10⁷ cells were lysed in 4 ml of lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCI, 0.5% Nonidet P40, and protease inhibitors). Cleared lysates were filtered through 0.45 μM spin filters (Millipore Ultrafree®-CL) and immunoprecipitated with 30 μl of anti-HA resin (Sigma). Complexes were washed with lysis buffer, exchanged into PBS, eluted with HA peptide and precipitated with 10% TCA. Detailed methods, including validation of interactions, microscopic localization of GFP-tagged proteins, and functional analysis of ERAD pathway are provided in Supplemental Methods.

Mass Spectrometry

TCA-precipitated proteins were trypsinized, purified using Empore C18 extraction media (3M), and analyzed via LC-MS/MS with a LTQ linear ion trap mass spectrometer (ThermoFinnigan) using an 18 cm × 125 μm (ID) C18 column and a 50 minute 8%–26% acetonitrile gradient. Spectra were searched using Sequest against a target-decoy human tryptic peptide database and these results were loaded into CompPASS for further processing and analysis (see Supplemental Methods).

CompPASS and Bioinformatics

Details of how CompPASS processes and analyzes MS/MS data are provided in the Supplementary Methods. CompPASS is accessible at http://pathology.hms.harvard.edu/labs/harper/Welcome.html.