A schematic of our methodology is shown in Figure 1A. Biochemical reactions are performed ‘on-chip’ by overlaying the protein microarrays with a purified conjugating enzyme or extract prepared from a biological specimen (e.g. cell line or pathological specimen) and all required co-factors. The protein microarrays are spotted with >8,000 different human recombinant proteins in duplicate which serve as substrates for PTM conjugation. The substrates are subsequently ‘tagged’ by conjugation of a labeled-modifier (e.g. biotin) present in the reaction mixture. Following a stringent wash to remove non-covalent substrate-modifier interactions, the PTM-conjugated substrates are then detected using ‘binders’ (e.g. antibodies or streptavidin) labeled with fluorescent dyes and the protein microarrays analyzed using a fluorescence slide scanner.

We first tried various configurations of ubiquitylation reactions using cellular fractions (S-100) and rabbit reticulocyte lysate to determine the optimal assay conditions. We evaluated different slide surface chemistries, reaction buffers, assay conditions, and detection methods. PATH slides (glass slides coated with nitrocellulose) proved to be superior to epoxy or hydragel-coated slides in reducing background (data not shown). The addition of 0.1% Tween-20 to both the reaction and wash buffers also significantly limited background and did not adversely affect PTM conjugation activity. Furthermore, the addition of inhibitors of de-conjugating enzymes (e.g. ubiquitin-aldehyde) to the reactions was found to increase PTM conjugation activity (data not shown). Moreover, washing the protein microarrays with 1 M NaCl +0.1% Tween-20 in PBS appeared to be sufficient for removing the non-covalent binding of modifiers to substrates since washes with 8 M urea, which is known to reduce non-covalent ubiquitin binding, was found to give an identical conjugation profile (data not shown). Although the use of protein microarrays to detect substrates of ubiquitylation has been previously reported [7], [8], we evaluated the reproducibility of our optimized ‘on-chip’ ubiquitylation reactions by statistically analyzing the results of three independent ubiquitylation reactions using whole-cell HeLa cell extract. Figure 1B shows an enlarged view of the same sub-array region from each of the three protein microarray replicates, wherein those spots producing fluorescent signals over background were found to be present on all three sub-arrays. Statistical analysis of the pair-wise scatter plots, plotting the signal intensity of each protein for each biological replicate, revealed a high degree of reproducibility between experiments (p<2.2 E-16; Fig. 1B).

As proof of principle, we first sought to determine whether our assay system could be used to faithfully identify substrates of a purified E3 ubiquitin ligase. For these experiments, we utilized the ubiquitin ligase SCF^Skp2 which has a well-defined role in human tumorigenesis [9]. SCF^Skp2 is known to ubiquitylate several different substrates including the cyclin-dependent kinase (Cdk) inhibitor p27^Kip1 [10], [11]. This reaction is molecularly complex and requires 1) substrate phosphorylation, 2) association of the substrate with cyclin A-Cdk2 complexes, and 3) the co-factor protein Cks1. We performed ‘on-chip’ ubiquitylation reactions that included recombinant human SCF^Skp2 isolated from Sf9 insect cells, purified E1 and E2 enzymes, ATP regeneration system, ubiquitin, and biotin-labeled ubiquitin. The results of these experiments showed that p27^Kip1 could be efficiently ubiquitylated on the protein microarray by SCF^Skp2, and ubiquitin conjugation activity was only present when all the required components were added to the mixture, recapitulating the reaction conditions in vivo (Fig. 1C). In addition to p27^Kip1, we identified several novel substrates of SCF^Skp2 (Fig. 1C; see also Substrate validation section).

We next sought to determine whether this methodology could be used to accurately profile the PTM activity of complex biological mixtures, such as cellular extracts or pathological specimens. Using a 2-fold change as a cutoff over negative controls that lacked cellular extract, ubiquitylation reactions performed with rabbit reticulocyte lysate and S-100 fraction of HeLa cells revealed robust conjugation activities with 239 and 119 substrates identified, respectively (Table 1). Sixty-six substrates were found to be common to both the rabbit reticulocyte lysate and HeLa S-100 fraction (Table 2). Of these substrates, several were previously shown to either bind ubiquitin (e.g. LIVIN [12], RNF4 [13], ZNF364 [14]), contain ubiquitin binding domains (e.g. CUED1C [15], RAD23A [16]), or were known substrates of ubiquitylation activity (e.g. FLT1 [17], JAK2 [18], INSR [19]), lending strong support that this methodology faithfully detects true substrates of ubiquitin conjugation activity. We next profiled whole-cell extracts prepared from various cultured cell lines of both human and mouse origin and found these extracts efficiently ubiquitylated many (>120) different substrates on the protein microarrays (Table 1; data not shown). Approximately half of these substrates were found to be consistently ubiquitylated by all the cellular extracts analyzed. Collectively, these results demonstrate that this methodology could be used to profile biologically relevant PTM activity in complex biological specimens of various species origins.

A clinically relevant application of this methodology is comparative profiling, wherein disease-associated changes in PTM activity are compared to the normal state. To this end, we applied this methodology to identify changes in ubiquitylation activity that occurs during the progression of human tumors to more advanced and life-threatening disease. Remarkably, we found that human breast tumor specimens that had been kept frozen at −80°C for >10 years contained robust ubiquitin conjugation activity (Table 1) comparable to that observed for cellular fractions or whole-cell extracts prepared from cultured cells. We next pooled extracts prepared from 5 low-grade and 5 high-grade breast tumors and performed ‘on-chip’ ubiquitylation reactions with these extracts. Using a 1.6-fold change as a cutoff over negative control reactions that lacked tumor lysate, we identified several differentially ubiquitylated substrates between the low-grade and high-grade specimens. These results are visually represented as a heat map in Figure 2 (fold changes are listed in Table S1). Interestingly, the majority of the differentially ubiquitylated substrates were found to have defined roles in several processes implicated in tumor progression. One of the proteins showing increased ubiquitylation in high-grade tumors was RAD23A [20], [21]. RAD23 is implicated in DNA repair and is known to interact with the E3 ligase E6AP, suggesting that its degradation by ubiquitylation may contribute to tumor progression through impairment of the DNA repair process. Moreover, TRIM52, a protein that possesses intrinsic E3 ubiquitin ligase activity, demonstrated increased ubiquitylation in high-grade tumors suggesting that it may also be targeted for degradation by ubiquitylation. In support of this, we found that TRIM52 is indeed a target of the ubiquitin proteasome pathway (see Substrate validation section).

We next determined whether this methodology could be easily adapted to other complex PTMs, such as SUMO1 (small ubiquitin-like modifier 1) and NEDD8 (neural precursor cell expressed and developmentally down-regulated 8). SUMO1 and NEDD8 are conjugated to substrates in multi-step enzymatic reactions similar to but distinct from ubiquitylation [22]. Reaction conditions used in our assay were similar to those used for the conjugation to ubiquitin (described above) except for the substitution of the relevant reaction buffer, E1 enzyme, aldehyde derivative, and biotin-labeled modifier. The results of these experiments showed that HeLa cell extracts efficiently conjugated SUMO1 and NEDD8 to many substrates on the protein microarrays (Table 2). Of the putative SUMOylated substrates identified, HIPK3 [23] and RNF4 [13] were previously shown to bind SUMO1 and the majority of the remaining substrates contained consensus SUMO1 targeting sequences (yKxE/D) [24]. Although only a few substrates of NEDDylation have been reported in the literature [25], [26], [27], our screen did detect LGALS3, which was previously shown to be NEDDylated using an alternative proteomic approach [27]. Of note, we failed to detect NEDDylation of the well-known NEDD8 target cullin protein family with our assay (cullins 1, 3, 4a, and 4b are displayed on the protein microarrays but the level of conjugation activities did not meet our 2-fold cutoff criteria). This lack of activity could be due to a number of factors. Although it is readily accepted that cullins are NEDDylated on the Lys in the conserved sequence IVRIMKMR [28], the accessory factors required for promoting cullin NEDDylation may be molecularly complex and is still an area of active investigation. In vitro evidence shows that the RING finger protein Rbx1 is required for cullin NEDDylation [29], [30], [31], while in vivo NEDDylation is enhanced by DCN1 [32]. Moreover, the ability to detect cullin protein NEDDylated may be influenced by de-NEDDylase activities (e.g. COP9 Signalosome) [33]. Therefore, it is plausible that the activity of Rbx1 or DCN1 present in our reactions was limiting or de-NEDDylase activity was dominant in our assay. Alternatively, these proteins may not be appropriately folded or pre-modified in insect cells (used for recombinant protein expression) and cannot be appropriately recognized by the NEDDylation machinery using our reaction conditions.

To determine the accuracy of our assay system in detecting true PTM conjugation activities, we first randomly selected c-Src, a SCF^Skp2 substrate identified using our assay but not reported in the literature, and determined if it was indeed a substrate of SCF^Skp2 in vivo. c-Src is a non-receptor tyrosine kinase that plays an important role in regulating cell proliferation and its augmented expression promotes tumor cell invasion and metastasis [34]. To validate c-Src as a novel SCF^Skp2 substrate, we transduced SKP2^−/− knockout MEFs with retroviruses that express Skp2 and found this induced the down-regulation of c-Src protein levels, consistent with its enforced degradation (Fig. 3A). Moreover, immunoprecipitation of Skp2 from these cell extracts revealed that endogenous c-Src associates with Skp2 in vivo (Fig. 3B). Furthermore, ectopic expression of Skp2 in HEK293T cells was found to stimulate c-Src ubiquitylation in vivo (Fig. 3C). Collectively, these results are consistent with SCF^Skp2 regulating the degradation of c-Src through ubiquitin-dependent proteolysis.

To further validate the accuracy of our methodology, we randomly selected 10 substrates which were shown to be ubiquitylated on the protein microarrays (by both rabbit reticulocyte lysate and HeLa S-100 fraction) but whose modification was not reported in the literature and attempted to verify whether they were substrates of ubiquitylation in vivo. HEK293T cells were co-transfected with plasmids that express HA-tagged ubiquitin and the Myc- or GST-tagged substrates activin A receptor-type IB (ACVR1B), beta-adrenergic receptor kinase 2 (ADRBK2), IL2-inducible T-cell kinase (ITK), protein kinase C-gamma (PRKCgamma), ephrin type-A receptor 1 (EPHA1), serine/threonine protein kinase PIM2, 90 kDa ribosomal protein S6 kinase 5 (RPS6KA5), kinesin family member 2C (KIF2C), ephrin type-A receptor 5 (EPHA5), or tripartite motif-containing protein 52 (TRIM52) (Fig. S1). To determine whether these substrates were covalently conjugated to ubiquitin in vivo, we subjected the HEK293T extracts to denaturing immunoprecipitation, which included lysis of cells in buffer containing 1% SDS and boiling the samples prior to immunoprecipitation [35]. Of the 8 substrates that were expressed and immunoprecipitated at detectable levels all were found to be ubiquitylated in vivo (Fig. 4A; data not shown). Bayesian statistical testing [36], [37] of these results verified that substrates that were ubiquitylated on the protein microarrays had a high-probability of being true substrates of ubiquitylation in vivo (the null hypothesis was tested H₀∶ p = 0.5 against the probability P* = 0.63 derived from our validation data and rejected with evidence ev = 0.89). To confirm that the observed ubiquitylation in vivo was not due to substrate overexpression, we immunoprecipitated endogenous YY1 protein, a putative substrate of ubiquitylation identified by our assay and regulator of the MDM2 ubiquitin ligase that controls the ubiquitin-dependent proteolysis of p53 [38], from HEK293T cell extracts using denaturing conditions and analyzed its ubiquitylation status by Western blot analysis. These experiments clearly showed that endogenous YY1 was indeed ubiquitylated in vivo (Fig. 4B).

We next tested the accuracy of our assay in profiling SUMO1 and NEDD8 conjugation activities using similar experimental strategies. Immunoprecipitation of endogenous insulin-like growth factor 1 receptor (IGF-1R), a receptor tyrosine kinase that mediates IGF1 signaling [38], from HEK293T cell extracts using denaturing conditions confirmed that it was covalently conjugated to SUMO1 in vivo (Fig. 5A). Furthermore, p21^Cip1-activated kinase 3 (Pak3), which is associated with non-syndromic mental retardation in humans [39], and Musk, a receptor tyrosine kinase that plays a role in neuromuscular junction organization [40], were found to be covalently conjugated to NEDD8 in vivo (Fig. 5B).

Several versions of the ProtoArray Human Protein Microarray (Invitrogen) were utilized in this study. Profiling experiments performed with purified ligases, whole-cell extracts, and tumor extracts utilized version 4 arrays. These protein microarrays display >8000 purified human proteins (in duplicate) on a nitrocellulose-coated glass slide. Each of the >8000 human proteins are derived from human open reading frames (ORF) that were expressed in Sf9 insect cells as an N-terminal GST fusion protein.

Cell lines (HeLa, mouse and human fibroblasts) and tumor (fresh-frozen human breast cancer tissue) specimens were suspended in lysis buffer (20 mM Hepes (pH 7.4), 2.5 mM MgCl₂, 0.5 mM DTT, 5 mM NaF, 1 mM sodium orthovanadate, 1 mM PMSF, 2 µg/ml aprotinin, 1 µg/ml pepstatin, and 1 µg/ml leupeptin) on ice for 15 min and then sonicated briefly. The extracts were clarified by centrifugation for 15 min at 14,000× g and snap-frozen in liquid nitrogen until use. Rabbit reticulocyte lysate and HeLa S-100 fraction were purchased (Boston Biochem).

Human SCF^Skp2 complexes were produced in Sf9 insect cells as described previously [11]. Recombinant Cks1 was produced in bacteria and purified as described [51]. Cyclin A-Cdk2 complexes were purchased (Life Technologies).

Antibodies used in this study included: anti-ubiquitin (Biomol, PW8805); anti-SUMO1 (Zymed, 33-2400); anti-NEDD8 (Zymed, 34-1400); anti-p27^Kip1 (BD pharmingen); anti-c-Src (Biosource); anti-Skp2 (Zymed), anti-YY1 (Santa Cruz Biotechnology); anti-IGF-1R (Zymed); anti-HA (Covance); anti-Flag (Sigma); anti-GST (Santa Cruz Biotechnology); and anti-Myc (9E10, Santa Cruz Biotechnology).

Extracts (2-100 µg in 40 µl of lysis buffer) were combined with either 4 µM of ubiquitin aldehyde (Boston Biochem) to prevent the action of deubiquitylating enzymes in the ubiquitylation reactions, SUMO1 aldehyde to inhibit SUMO-specific isopeptidases (SENPs) (Boston Biochem) in SUMOylation reactions, or NEDD8 aldehyde to inhibit deNEDDylating and NEDD8 processing enzymes in NEDDylation reactions (Boston Biochem), and then incubated at 25°C for 15 min. The reactions were then supplemented with modifier (1.25 µg/ml), biotin-labeled modifier (50 ng/ml), Tween-20 (0.1%), energy regenerating system (Boston Biochem), and 1× reaction buffer (ubiquitylation, SUMOylation, NEDDylation; Boston Biochem) in a final volume of 100 µl. Proteasome inhibitor MG132 (5 µM) was added to ubiquitylation reactions. For SCF^Skp2 experiments, reaction conditions were as described [11]. The reaction mixtures were applied to the protein microarrays, covered with glass coverslips equipped with rubber gaskets to avoid leakage (Life Technologies), and then incubated at 37°C for 1 hr in a humidified chamber. The arrays were then washed in PBS-Tween (0.1%, PBST) containing 1 M NaCl for 10 min, 2×10 min in PBST, and then incubated with Streptavidin Alexa Fluor 647 (100 ng/ml; Life Technologies) for 1 hr at 25°C. The arrays were then washed 3×10 min in PBST and spun dry. Imaging was performed using a GenePix 4000B Slide Imager (Molecular Devices) and fluorescent spots analyzed using GenePix Pro software. Gal files (which contain array production information and spot location, identification, and quantification) were downloaded from www.invitrogen.com and used with GenePix Pro software to analyze the median intensity of each spot. All data evaluations were done using the statistical program R [52]. Specifically, we first filtered the data with a cutoff threshold of 5000 counts for the fluorescence values, and then applied the Benjamini-Hochberg procedure [53] with control of the false discovery rate (FDR) set at the 5% level. The resulting set of proteins was used to mine UniProt and PubMed using the BioConductor modules of R.

SKP2^−/− MEFs were transduced with control (pBabe) or 3× FLAG-Skp2 (pBabe-Skp2) retrovirus, and used for validation of c-Src as a SCF^Skp2 substrate. All in vivo validation experiments were performed using a technique that preserves the substrate modification and limits co-purification with non-covalently bound modifiers of modified interacting proteins [35]. Briefly, HEK293T cells (ATCC) were lysed under denaturing conditions in 1% SDS (containing 20 mM N-ethyl-maleimide (NEM)) and boiled briefly to disrupt non-covalent interactions, and then the buffer adjusted to 1× RIPA (0.1% SDS, 0.5% NP40, 20 mM NEM, 50 mM Tris (pH 8.5), 150 mM NaCl, 5 mM EDTA, 5 mM NaF, 1 mM sodium orthovanadate, 1 mM PMSF, 2 µg/ml aprotinin, 1 µg/ml pepstatin, and 1 µg/ml leupeptin). The expressed or endogenous putative substrates of ubiquitin, SUMO1, or NEDD8 were then immunoprecipitated from the extracts as indicated. In all cases, immunoprecipitation of extracts with IgG antibodies of the same species served as control.