A Data Set of Human Endogenous Protein Ubiquitination Sites*

Lysine ubiquitination is an important and versatile protein post-translational modification. Numerous cellular functions are regulated by ubiquitination, suggesting that extensive numbers of proteins, if not all, are modified with ubiquitin at certain times. However, proteome-wide profiling of ubiquitination sites in the mammalian system is technically challenging. We report the design and characterization of an engineered protein affinity reagent for the isolation of ubiquitinated proteins and the identification of ubiquitination sites with mass spectrometry. This recombinant protein consists of four tandem repeats of ubiquitin-associated domain from UBQLN1 fused to a GST tag. We used this GST-qUBA reagent to isolate polyubiquitinated proteins and identified 294 endogenous ubiquitination sites on 223 proteins from human 293T cells without proteasome inhibitors or overexpression of ubiquitin. Mitochondrial proteins constitute 14.7% of this data set, implicating ubiquitination in a wide range of mitochondrial functions.

Lysine ubiquitination is an important and versatile protein post-translational modification. Numerous cellular functions are regulated by ubiquitination, suggesting that extensive numbers of proteins, if not all, are modified with ubiquitin at certain times. However, proteome-wide profiling of ubiquitination sites in the mammalian system is technically challenging. We report the design and characterization of an engineered protein affinity reagent for the isolation of ubiquitinated proteins and the identification of ubiquitination sites with mass spectrometry. This recombinant protein consists of four tandem repeats of ubiquitin-associated domain from UBQLN1 fused to a GST tag. We used this GST-qUBA reagent to isolate polyubiquiti- Post-translational modification of proteins by ubiquitination at lysine residues plays regulatory roles in a broad spectrum of cellular processes including cell cycle progression, DNA damage, and immune response (1). Deregulation of ubiquitination is linked to many human diseases including cancer and neuronal disorders (2,3). In the canonical ubiquitination reaction, a substrate is covalently conjugated with ubiquitin by an enzymatic cascade involving ubiquitin-activating enzyme (E1), ubiquitin conjugation enzyme (E2), and a ubiquitin ligase (E3). Conversely, the ubiquitin moiety can be cleaved off from the substrates by deubiquitinases (DUBs). 1 The human genome encodes at least two E1, 53 E2, and ϳ500 E3 enzymes and more than 100 DUBs (4,5); such diversity of the ubiquitination system rivals that of the kinome and offers multiple gateways for regulation of ubiquitination dynamics in a cell.
The diversity of the cellular ubiquitination system or ubiquitome is increased by formation of polyubiquitin (poly-Ub) chains with variable lengths and linkages. It is now known that all seven internal lysine residues and the N terminus of ubiquitin can form poly-Ub chains that may impart different functions. For example, proteasome-mediated protein degradation is a well known consequence of Lys-48-linked poly-Ub modification (6). Other linkages may also lead to protein degradation (7) or may have non-proteolytic consequences. The Lys-63 linkage is best characterized in the inflammatory response where it aids in activation of kinase cascades (8,9), the N-terminal linear poly-Ub has been shown to activate the IB kinase (10), and Lys-33 linkage has recently been shown to regulate specific signal transduction in a proteolysis-independent manner (11).
Thorough understanding of this complex regulatory system requires identification of ubiquitinated substrates and, preferably, mapping of their ubiquitination sites, which is technically challenging. The most common approach used by laboratories to date is site-directed mutagenesis of putative lysine targets to infer the ubiquitination site(s). Although this is necessary for determining whether ubiquitination has a significant effect on a particular biology of the protein in question, it does not provide direct chemical evidence of ubiquitination. Erroneous conclusions can be made if mutation of lysine residues affects protein structure, folding, or docking site for E2 and E3 enzymes. Mass spectrometry (MS) has emerged as an indispensable tool for direct measurement of the lysine ubiquitination site(s) that complements mutagenesis studies. The ubiquitin conjugation generates an isopeptide bond between the -amine of the modified lysine on the substrate and the C terminus of ubiquitin. After trypsin cleavage, ubiquitination can be detected by mass spectrometry as a 114.043-Da mass shift (from the Gly-Gly remnant of the ubiquitin C terminus) on the modified peptides. The primary limitation in proteomewide identification of ubiquitination sites is the lack of high affinity reagents for isolation of ubiquitinated peptides. Additionally, there are three confounding factors that limit our abilities to enrich ubiquitinated peptides. First, only a small percentage of a given protein is ubiquitinated in the steady state. Second, DUBs have sizable enzymatic activity and further decrease levels of ubiquitinated proteins upon cell lysis. Lastly, ubiquitin is the most abundant ubiquitinated protein in the cell due to the prevalence of poly-Ub chains, masking the identification of other substrates by mass spectrometry. Despite these inherent barriers, several studies have shown moderate success in cataloguing the lysine ubiquitination sites (12)(13)(14)(15)(16). For example, Peng et al. (12) pioneered a yeast ubiquitin deletion strain expressing His 6 -tagged ubiquitin to isolate ubiquitinated substrates and identified 110 ubiquitination sites in yeast proteins with an LCQ mass spectrometer. Meierhofer et al. (15) used a similar approach in HeLa cells that stably overexpress tagged ubiquitin and identified ϳ50 ubiquitination sites in human proteins with an LTQ-XL-Orbitrap.
Eukaryotic cells have evolved protein domain structures, ubiquitin binding domains (UBDs), that can recognize and bind to ubiquitin modifications. More than 20 different families have been identified to date, and most of them bind poly-Ub relatively weakly (17,18). The ubiquitin-associated domain (UBA) is the first identified UBD and is one of the best poly-Ub binders (19,20). Although a single UBA domain has been successfully used as an affinity reagent to quantify the poly-Ub chains in a mouse model of Huntington disease (21), it bears moderate ubiquitin binding affinity that may not be sufficient for proteome-wide isolation of ubiquitinated proteins. Recently, several studies showed that tandem UBDs display avidity in poly-Ub binding (22)(23)(24). We have utilized this concept to develop and test the suitability of a tandem UBA protein for large scale characterization of ubiquitination substrates.
Here, we report using recombinant GST fusion of four tandem ubiquilin-1 (UBQLN1) UBA domains (GST-qUBA) as a bait to isolate endogenous ubiquitinated proteins from human 293T cells without ubiquitin overexpression and proteasome inhibition. We confirmed that tandem GST-qUBA is more efficient at binding poly-Ub chains than the single GST-UBA. Using GST-qUBA as an affinity purification reagent, we were able to detect ϳ300 lysine ubiquitination sites that are supported by high quality mass spectra. Both abundant and low abundance cellular proteins including tumor suppressors and regulators of apoptosis and NF-B pathways were identified. We demonstrate that GST-qUBA is a powerful tool for the affinity purification and identification of endogenous protein ubiquitination sites at the proteome scale.

EXPERIMENTAL PROCEDURES
DNA Subcloning, Protein Purification, and GST-Protein Immobilization-GST-UBA was subcloned by PCR amplification of the UBA domain (amino acids 540 -589) of UBQLN1 and inserted into the pGEX-4T-1 vector using BamHI/EcoRI sites. To generate GST-qUBA, four repeats of the DNA sequence encoding the UBA domain were synthesized (Genscript) and subcloned into pGEX-4T-1 vector using EcoRI/SalI sites. Three glycine residues were inserted as a linker between each UBA domain. For protein production, transformed BL21 cells were induced with isopropyl 1-thio-␤-D-galactopyranoside (0.8 mM) at ϳ0.6 A 600 nm for 5 h. Cells were sonicated in lysis buffer (0.1% Nonidet P-40 in PBS), and recombinant proteins were purified with Glutathione-Sepharose 4B (GSH) beads (GE Healthcare).
Cell Culture and Poly-Ub Pulldown Assay-Twenty 150-mm dishes of 293T cells were grown to confluence before harvesting. Cells were briefly sonicated and lysed on ice in NETN buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.5% Nonidet P-40) supplemented with protease inhibitor mixture (GenDEPOT, P3100) and DUB inhibitors (1 mM iodoacetamide and 8 mM 1,10-o-phenanthroline). Protein extracts were centrifuged twice at 100,000 ϫ g for 15 min and then incubated with 200 l of immobilized GST-qUBA beads at 4°C for 40 min. After incubation, the beads were washed four times with 1 ml of ice-cold NETN buffer with DUB inhibitors and then eluted in 50% acetonitrile in 0.1% formic acid (FA) or boiled in SDS-PAGE loading buffer.
Sample Preparation for Mass Spectrometric Analysis-One-half of the eluted proteins were resolved by 4 -20% SDS-PAGE, and the gel was sliced into 22-24 bands based on molecular weight and subjected to in-gel trypsin digestion as described previously (25). The other half of the sample was dried in a Savant SpeedVac, reconstituted with 100 mM NH 4 HCO 3 , and digested by trypsin (1:50, w/w) at 37°C overnight. The digested peptides were then centrifuged at 13,000 ϫ g for 5 min, and the supernatant was subjected to isoelectric focusing (IEF) in an Agilent 3100 OFFGEL fractionator (Agilent, G3100AA) according to the manufacturer's instructions. Peptide mixtures collected from the 12 IEF fractions were desalted and concentrated with C 18 stop and go extraction tips. Mass spectrometry was performed with an LTQ-Velos-Oribitrap mass spectrometer (Thermo Fisher) equipped with a nanospray ion source and HPLC system. Purified peptides were injected in an in-house-built C 18 column (75-m inner diameter) and eluted with a 120-min linear gradient of 95% solvent A (0.1% FA in water) to 35% solvent B (0.1% FA in acetonitrile) with a flow rate of 400 nl/min. Raw data were acquired in a data-dependent mode in which full MS spectra (m/z ϭ 350 -1,300) were collected in the Orbitrap with targeted resolution of 100,000, and the top 20 most intense ions were fragmented by collision-induced disassociation (CID) and analyzed in the LTQ-Velos. The maximum ion injection time was set at 30 ms. A total of ϳ2 ϫ 10 6 MS/MS spectra were recorded in which ϳ70,000 redundant spectra were identified at a 1% false discovery rate (FDR) by Mascot.
Quantitative MS for Poly-Ub Quantification-293T cells were cultured in stable isotope-labeled DMEM (Invitrogen; [ 13 C 6 ]arginine and [ 13 C 6 ]lysine) for six generations before harvesting (26). To compare relative ubiquitin binding efficiency between GST-qUBA and GST-UBA, a similar amount of recombinant proteins (5-10 mg) was used to isolate polyubiquitinated proteins from whole cell protein extracts (ϳ0.4 mg/ml; 1 ml of protein extracts) of either "heavy" or "light" labeled cells. After binding, the beads were mixed and washed with NETN buffer. To account for potential difference in SILAC, the pulldown was repeated, but the order of the isotope label was switched. Proteins that bound the beads were resolved by SDS-PAGE, and the gel region above 50 kDa was digested with trypsin and subjected to LC-MS/MS for quantification with the Orbitrap mass spectrometer.
Database Search of CID Spectra for Ubiquitinated Peptides and Protein Identification-Raw CID spectra were searched against a target-decoy (27) human RefSeq database (June 27, 2009; 39,165 entries) with the Mascot 2.2 (Matrix Science) algorithm embedded in the Proteome Discoverer 1.2 interface (Thermo Fisher). Key parameters used for the search were as follows: precursor mass tolerance was confined within 10 ppm with a fragment mass tolerance of 0.5 dalton and a maximum of two missed cleavages allowed. Dynamic modifications were allowed for lysine ubiquitination (ϩ114.0429 Da) and methionine oxidation; fixed modification was set to carbamidomethylation of cysteine. Identified peptides were filtered with 5% FDR and subjected to rigorous manual verifications as described in the text. Identical parameters were used by Sequest (Thermo Fisher) for ubiquitination identification and compared with Mascot.
Biochemical Validation of Ubiquitination-Two ubiquitinated substrates were verified by pulldown using GST-qUBA followed by immunoblot analysis using antibodies specific to Diablo (Sigma-Aldrich, S0941) and ␤-catenin (Cell Signaling Technology, CS-95268).
Bioinformatics Analysis-A list of 223 substrates was submitted to the Database for Annotation, Visualization and Integrated Discovery (DAVID, version 6.7) Bioinformatics Resources (28) for enrichment analysis. For ubiquitination interaction network analysis, the substrate proteins were first searched against the STRING database (29) (version 8.2) to identify protein-protein interaction network. The ubiquitination interaction networks extracted by STRING were then analyzed for densely connected regions with the MCODE algorithm (30). MCODE is a part of the toolkit for network analysis and visualization in the Cytoscape software (31) (version 2.63). Only the top four interaction modules were selected.
The relative density map for ubiquitinated lysines was constructed in an in-house-built FileMaker-based application. Briefly, the identified lysine ubiquitination sites and the 6 flanking amino acid residues on either side of the modified lysine were used to generate a positionspecific matrix. The ubiquitinated lysine was fixed at position 0. The frequency of amino acid occurrence at each position from Ϫ6 to ϩ6 was calculated; enrichment against the natural abundance of amino acids in the human proteins was shown as an intensity map.

RESULTS AND DISCUSSION
Comparison of Poly-Ub Binding Efficiencies of qUBA and UBA with Quantitative Mass Spectrometry-We took advantage of the natural ability of UBA to bind poly-Ub with moderate affinity (19,20) and made a recombinant affinity reagent for ubiquitin binding with tandem repeats of the UBA domain (qUBA), reasoning that it should have higher affinity to poly-Ub. We first compared relative binding efficiency of qUBA and of the single UBA domain to polyubiquitinated proteins from 293T cell lysate. To distinguish poly-Ub isolated with each UBA reagent by MS, we used similar amounts of heavy and light isotope-labeled cell lysates for binding with qUBA and UBA. We then mixed the samples isolated by the different UBAs, resolved them by SDS-PAGE, and in-gel-digested the region that is above 50 kDa with trypsin. The resulting peptides were measured by an LTQ-Velos-Orbitrap mass spectrometer for relative quantification of poly-Ubs. To check for a potential bias of heavy isotope labeling, we repeated the experiment with reversed heavy and light lysates for the UBAs (Fig. 1, A and B).
When a limited amount of cell lysate was used, Lys-6-, Lys-11-, Lys-48-, and Lys-63-linked poly-Ubs were readily detected from the sample pulled down by the GST-qUBA, whereas only the Lys-48 linkage was detectable for GST-UBA, and other Ub linkages were below the detection limit of the mass spectrometer. The enrichment for Lys-48 linkage by qUBA over UBA was ϳ500-fold as judged by the peak areas, whereas the enrichment for Lys-6, Lys-11, and Lys-63 linkages was estimated to be greater than a 1,000-fold (Fig. 1D). Other poly-Ub linkages (Lys-27, Lys-29, and Lys-33) were not detected in this experiment; however, they were detected when more cell lysate was used. We conclude that qUBA is more efficient in poly-Ub binding than a single UBA, consistent with the conclusion from the previous report using synthetic poly-Ub chains (24).

Large Scale Identification of High Confidence Endogenous Ubiquitination Sites in 293T Cells Using Automated Search with FDR Threshold Followed by Manual Verification-Next
we sought to utilize qUBA for identification of ubiquitination sites in 293T cells that were not preincubated with proteasome inhibitors or overexpressing exogenous ubiquitin. To minimize the loss of ubiquitination by the DUB activity, both cysteine and serine protease inhibitors were included during cell lysis (32). Ubiquitinated proteins were isolated with the GST-qUBA immobilized on GSH-agarose beads, digested in-gel after separation by SDS-PAGE, or digested in solution and further separated by IEF. Extracted tryptic peptides were subjected to LC-MS/MS analysis (Fig. 2).
Two LC-MS/MS runs were searched with Mascot and Sequest to compare various search parameters that may increase the reliability of ubiquitination identification. Using a target-decoy database search with 1 and 5% FDRs, Mascot identified 20 and 50 unique sites, respectively. Among the 20 sites identified at 1% FDR, 19 sites were manually verified, and one site was rejected. In contrast, among the 50 sites identified at 5% FDR, only 27 sites were manually verified, and 23 sites were rejected (Fig. 3A). Thus, a 1% FDR search with Mascot may produce some false negative identification, whereas a 5% FDR search led to high levels of false positive identifications. Under the same FDR conditions, Sequest was able to identify more ubiquitination sites than Mascot; however, many of them could not be manually verified (Fig. 3B). Potential differences exist between these two popular search engines as only 18 spectra were identified by both algorithms at 1% FDR (Fig. 3C).
Theoretically, one can combine the results obtained by the two different algorithms: in our case, only 25 of the 46 combined spectra identified by Sequest and Mascot searched with a 1% FDR passed our manual verifications (Fig. 3D). Thus, it is unsafe to gauge the validity of ubiquitination site identification using a 1% FDR as the sole determinant. We conclude that manual verification is the most reliable means for the identification of ubiquitination sites, and the FDR alone might not be accurate enough to estimate the overall reliability.
Identification of 294 Unique Lysine Ubiquitination Sites in 293T Cells without Proteasome Inhibition or Ub Overexpression-A total of ϳ70,000 redundant spectra were identified at a 1% FDR by Mascot in three replicate experiments in which 7.5% of the spectra were from ubiquitin itself. To identify "provisional" ubiquitinated peptides, we used the Mascot search engine with a 5% FDR and then subjected spectral matches to vigorous manual verification. Several criteria for manual verification were used to eliminate false positive assignments. First, in the correct assignment, major MS 2 fragment peaks must be unequivocally assigned and follow a pattern of a continual stretch of y and b ions and the appearance of dominant fragment ions N-terminal to proline and C-terminal to aspartic acid for arginine-containing peptides (33)(34)(35). Second, only when both the ubiquitinated and unmodified peptides of the same protein are recovered is the ubiquitinated peptide assignment accepted. Lastly, we found that mass accuracy of the full mass spectra provides a good constraint on the fidelity of the ubiquitination assignment where good assignments that pass manual verification are within a Ϯ2-ppm window from the median value of the mass accuracy distribution in the LC-MS run (supplemental Fig. 1). Thus, we rejected assignments that fall outside of this window.
A total of 294 endogenous ubiquitination sites on 223 substrates were identified and manually verified (supplemental Table 1) of which Ͼ95% of the ubiquitination sites were not previously reported in the Swiss-Prot database. Greater than 96% of the ubiquitinated peptides are either doubly or triply charged, and less than 4% of peptides with 4ϩ and 5ϩ charges were assigned with sufficient confidence (supplemental Fig. 2A). It is noteworthy that ubiquitination on internal, but not C-terminal, lysine was detected in our study, indicating that trypsin is indeed very specific and only cleaves the unmodified lysine residue. The 292 unique ubiquitination spectra (containing 294 lysine ubiquitination sites) were annotated and combined into a spectral library file provided as supplemental Fig. 5. Two substrates were randomly selected for biochemical validation (supplemental Fig. 2B). Ubiquitinated substrates were isolated by GST-qUBA from human 293T cells and immunoblotted with antibodies against Diablo or ␤-catenin. In agreement with MS identification, the detection of smears of these proteins on the SDS-PAGE gel is consistent with ubiquitination and further supports the conclusion that GST-qUBA specifically enriches the polyubiquitinated forms of these proteins.
Ubiquitination on seven internal lysines as well as Lys-6/ Lys-11 double linkage (or "forked structure") was identified (supplemental Table 1), highlighting the complexity of poly-Ub topology in vivo. The N-terminal linear Ub linkage was not detected. Poly-Ub linkages on all internal lysine residues were abundantly detected on most regions of the SDS-PAGE gel, suggesting that modification of proteins with Ub linkages other than Lys-48 is widespread in the cells.
Motif-X analysis (36) identified a putative ubiquitination motif of KXLXD (where X denotes any amino acid) (supplemental Fig. 3A). This sequence appears over 4,000 times in human proteome. Strikingly, ubiquitin itself contains this sequence at Lys-48. Although the appearance of this sequence motif is statistically significant, further tests will show whether it is indeed a functional motif used by cells to select the ubiquitination site to build a Lys-48 Ub linkage for protein degradation. We also constructed a density map to visualize distribution and overall enrichment of amino acid residues surrounding the ubiquitinated lysine (supplemental Fig. 3C).
Because both ubiquitination and acetylation are the major forms of lysine modification, we compared our data set with the recently published lysine acetylation sites (37) and found  that ϳ15% (43 sites) of the ubiquitinated lysines are also acetylated. Such coincidence of ubiquitination and acetylation on the same lysine suggests a competitive relationship between the two modification pathways. Abundant proteins that belong to protein translational factors, the Ub-proteasome system, cytoskeleton, heat shock proteins, RNA splicing, and histones are well represented in our data set (supplemental Fig. 4 and Table 1). Importantly, low abundance proteins including a cell cycle regulator (cyclin B1), transcription factor (ID2), tumor suppressor (␤catenin), and oncoprotein (c-Myc) were also detected (supplemental Table 1). The identification of low abundance proteins that are involved in tumorigenesis is particularly noteworthy. It is possible that these regulatory proteins are tightly regulated by the ubiquitination system for the maintenance of cellular homeostasis.
Approximately 20% of the substrates in our data set have been linked to various human diseases (supplemental Table 2). For example, ubiquitination sites of CASP8, a proapoptosis protein involved in multiple cancer types (38,39), and UBE3A, whose mutations have been linked to Angelman syndrome (40), are among the substrates we identified. These observations further support that ubiquitination pathways play important roles in regulating various cell functions and that deregulation of these pathways may lead to human disease.
Ubiquitination Is Enriched in Apoptosis, NF-B Pathway, and Mitochondria-As shown in Fig. 4A, ubiquitination sites for several critical components of both intrinsic and extrinsic apoptosis pathways were identified. There are three critical decision points in the intrinsic apoptosis pathway: 1) the release of cytochrome c, somatic (CYCS) to active the apoptosome (APAF1/CASP9), 2) the relief of inhibition by SMAC/ DIABLO of X-linked inhibitor of apoptosis protein on caspase 3, and 3) the translocation of AIFM1 to the nucleus for DNA fragmentation (41). Surprisingly, ubiquitination sites on proteins in all three pathway components were mapped. Similarly, ubiquitination sites on the upstream regulators caspase 8 and protein kinase A that mediate apoptosis in response to survival and growth cues in the extrinsic pathway were also identified.
Proteins in the NF-B pathway are well represented in the data set. As shown in Fig. 4B, multiple components proximal to the TNF␣ receptor are ubiquitinated at multiple lysine residues including several well known ubiquitin E3 ligases of both positive and negative regulators in the pathway; an E2 enzyme (UBE2N), which specifically processes a Lys-63 poly-Ub chain; and a DUB (OTUD5). Together, these observations highlight the importance of ubiquitination in the regulation of the NF-B pathway and activation of the immune response (8). Intriguingly, most of the substrates identified in the NF-B pathway are themselves enzymes that play regulatory roles in ubiquitination reaction, thereby demonstrating a delicate self-controlled system in this pathway. Because 14.7% of the identified substrates (33 proteins) are mitochondrial, our data also reveal possible roles of ubiquitination in a wide range of mitochondrial functions. These proteins localize to inner and outer membranes as well as matrix of the mitochondria (Table I). Because the mitochondrion is well known as the hub of apoptosis (41) in which ubiquitination plays an essential role, we speculate that some substrates identified here regulate apoptosis. Our data raise several interesting questions regarding the mitochondrial ubiquitination reaction. For example, do the ubiquitination events happen inside or outside of mitochondria? Will there be any functional difference or poly-Ub linkage preferences of mitochondrial ubiquitination versus ubiquitination elsewhere in the cell? Answering these questions with further biological analysis will expand our understanding of mitochondrial functions.
In summary, we have provided evidence that the GST-qUBA binds poly-Ub chains more efficiently than GST-UBA and can be used as an affinity reagent to isolate ubiquitinated substrates. Using a combination of FDR and manual verification, we were able to map 294 endogenous lysine ubiquitination sites that are supported by high quality MS data. Ubiquitination sites on both abundant and scarce regulatory proteins were identified. Our ubiquitination site data set contains a significant number of substrates that are localized to the mitochondria, implicating a role of ubiquitination in a wide range of mitochondrial functions. Overall, our data set thus provides direct and accurate chemical evidence for ubiquitination on the substrates, and the information builds a solid foundation for future functional analysis of these substrates. Because this method does not rely on Ub overexpression or proteasome inhibition, it will find applications for ubiquitination profiling in physiological or pathological samples. Although this method is very efficient for isolation of ubiquitinated substrates, it does not directly enrich ubiquitinated peptides and is less than optimal for mapping ubiquitination sites. Affinity reagents that specifically recognize and enrich tryptic peptides containing a Gly-Gly isopeptide bond could be more useful for this purpose. We envision that the combination of qUBA pulldown followed by enrichment with the recently described ubiquitin antibody (42) will allow for an even deeper exploration of the ubiquitome.