DIA-MS and neural network-based data processing boost ubiquitinome coverage

Even though our revised lysis protocol quantified up to 30,000 K-GG peptides in 125min LC-MS runs, due to the semi-stochastic sampling inherent to DDA, only about 50% of those identifications were without missing values in replicate samples (Supplementary Fig. ^1d). Consequently, the number of robustly quantified K-GG peptides would be greatly reduced in large sample series. To address this problem, we explored data-independent acquisition (DIA), an acquisition technique that is less susceptible to run-to-run variability²¹. Being actively developed, DIA allows the identification and quantification of very high numbers of peptides and proteins, for either regular proteomes or phosphoproteomes^22,23. Recently we introduced DIA-NN, a deep neural network-based software, which significantly increases proteomic depth and quantitative accuracy for DIA, especially for samples of high complexity²⁴. To improve the analysis of DIA data for ubiquitinomics, we expanded DIA-NN with an additional scoring module that ensures confident identification of modified peptides, including K-GG peptides (Methods).

Using a medium-length (75min) nanoLC gradient, we set up optimized MS methods (Supplementary Data ⁴) and benchmarked our DIA workflow against state-of-the-art label-free DDA. For DIA data processing we used DIA-NN in “library-free mode” (that is searching against a sequence database without an experimentally-generated spectral library), whereas for DDA data processing we used MaxQuant²⁰, with match-between-runs (MBR) enabled (Fig. 2a). While DDA quantified 21,434 K-GG peptides on average per sample from proteasome inhibitor-treated HCT116 cells, DIA more than tripled this number, to 68,429 K-GG peptides (Fig. 2b). Besides increased coverage, DIA showed excellent quantitative precision and reproducibility. The median CV for all quantified K-GG peptides was about 10%, and 68,057 peptides were quantified in at least three replicates (Fig. 2c, d). 88% of ubiquitinated peptides detected by DDA were also identified by DIA (Fig. 2e).

Open in a separate window

Fig. 2

Side-by-side comparison of DDA and DIA for ubiquitinomics.

a Schematic of data-dependent acquisition (DDA) and data-independent acquisition (DIA) strategies. A K-GG peptide sample was prepared from MG-132-treated HCT116 cells and four replicates were acquired with each scan mode. The DDA data were processed with MaxQuant (MBR enabled). For DIA data processing in DIA-NN, the same DIA-MS raw files were processed with either library-free (no experimental spectral library) or library-based search (experimental spectral library obtained through high pH reversed-phase fractionation of peptides prior to K-GG peptide enrichment). b Quantified K-GG peptides with 2mg of total protein as input for each replicate, in a side-by-side comparison of DDA and DIA-MS. Four samples (injection replicates) were acquired with 75min single-shot LC-MS runs and processed with MaxQuant or DIA-NN. Library-based DIA data analysis was performed with an ultra-deep K-GG peptide library generated by library-free DIA analysis of high-pH fractions, as detailed in “Methods” (LF=library-free, lib=experimentally-generated spectral library). c Coefficients of variation (CVs, in %) for K-GG peptides as shown in b are plotted for each method. Continuous lines demarcate the median and dashed lines the upper and lower quartiles. d Number of identifications with 0, 1, 2, or 3 missing values in four replicates (samples shown in b) for DDA and DIA. e Venn diagram showing overlap of ubiquitinated peptides (K-GG) identified by DDA and by DIA (DIA-NN’s library-free (LF) and spectral library-based (lib) approaches). f Scatter plot of Log2-transformed K-GG peptide intensities obtained from library-free and spectral library-based search options in DIA-NN (n=4, injection replicates). Values indicate Pearson correlation coefficients. g Raw data of MG-132-treated HEK293 cells from Hansen et al. (Fig. 1d in the original manuscript, three workflow replicates, each injected in duplicate and processed with Spectronaut’s directDIA mode¹⁶) were processed with DIA-NN (library-free mode). Plots show identified K-GG peptides (unique peptide sequences) for both Software packages. Source data are provided as a ^{Source Data} file.

Although DIA-NN uses a rigorous approach to determine the false discovery rate (FDR)^24,25, it has not been validated for the identification of K-GG peptides. We therefore determined the identification confidence specifically for K-GG peptides experimentally, and showed that it is comparable to that of the DDA workflow and to a different DIA processing software (Supplementary Note ¹ and Supplementary Fig. ¹⁰). We also confirmed the excellent quantitative accuracy and dynamic range of our DIA workflow, using a pool of synthetic K-GG peptides spiked into a yeast tryptic digest at different concentrations (Supplementary Note ² and Supplementary Fig. ¹¹).

We next benchmarked the performance of DIA-NN when using it in combination with an ultra-deep spectral library generated by high-pH reversed-phase fractionation, consisting of 146,626 K-GG peptides (Methods). This yielded similar results as library-free analysis, both in terms of coverage and reproducibility (Fig. 2b–f). Finally, to directly compare DIA-NN’s performance for analyzing ubiquitinomics data to another software, we processed six raw files from Hansen et al.¹⁶ with DIA-NN, revealing that DIA-NN identified on average 40% more K-GG peptides (Fig. 2g).

DIA-MS ubiquitinomics for low protein amounts and high-throughput analyses

When studying intracellular protein ubiquitination by mass spectrometry, proteasome inhibitors are often used to prevent degradation of the target of interest, thus conserving and boosting the ubiquitin signal^9,19,26. However, such compounds are highly cytotoxic and result in an accumulation of newly synthesized, misfolded proteins, thus activating the unfolded protein response, one of the major cellular anti-stress mechanisms²⁷. Moreover, proteasome inhibition globally perturbs protein turnover by disconnecting protein ubiquitination from degradation⁸. To determine whether our DIA-MS method is suitable for high coverage ubiquitinomics in the physiological context and in case of low protein input, we quantified K-GG peptides derived from different protein input amounts of either MG-132- or DMSO-treated cells. This revealed that only 1mg of total protein was sufficient to reach saturation of about 70,000 quantified K-GG peptides in MG-132-treated cells. In DMSO-treated cells, however, increasing the input protein amount to 2mg augmented the number of quantified K-GG remnants from 39,000 to 46,000 (Fig. 3a and Supplementary Data ⁵). Notably, with just 250µg of protein, we quantified about 18,000 K-GG peptides in the absence of proteasome inhibition, demonstrating the feasibility of in-depth ubiquitinomics in the physiological context, even for low protein amounts. As expected, a large fraction of K-GG peptides increased in intensity upon proteasome inhibition and about 20,000 peptides increased >4-fold. On the other hand, about 1,500 and 6,200 K-GG peptides decreased in intensity >4-fold and >2-fold, respectively. Among those were K-GG sites mapping to the C-terminus of the histones H2A and H2B (K119 and K121, respectively), both of which are well-studied ubiquitination sites involved in the regulation of transcription and in DNA repair^28,29. Besides these known sites, we also identified uncharacterized K-GG sites on histones that were strongly reduced in intensity upon MG-132 treatment, such as H2B-K44 (Fig. 3b).

Open in a separate window

Fig. 3

DIA-MS ubiquitinomics for untreated samples and for high-throughput applications.

a HCT116 cells were treated with DMSO or 10µM MG-132 for 6h and K-GG peptides from different protein inputs quantified by 75min single-run DIA-MS (library-free mode). Four samples were processed independently for each tested protein input. b Dot plot of Log2-transformed K-GG peptide ratios (MG-132 vs DMSO) with 2mg of protein input, as shown in a. Selected K-GG sites are highlighted. c Number of quantified K-GG peptides with 15, 30, 45, 60, and 75min DIA-MS runs. A K-GG peptide pool was injected on the same instrument for each method (n=4, injection replicates). The data were processed with library-free DIA. d Coefficients of variation (CVs) distributions for K-GG peptides quantified in c. Continuous lines in the violin plots demarcate the median and dashed lines upper and lower quartiles. e Shown are K-GG peptide numbers from DMSO-treated U2OS cells processed by Hansen et al. (Fig. 3 in the original manuscript, three biological replicates, each injected twice¹⁶) and the same raw files processed with DIA-NN (library-free mode). Source data are provided as a ^{Source Data} file.

To evaluate the potential for higher throughput applications, we set up DIA-MS methods for 15, 30, 45, and 60min LC gradients (Supplementary Data ⁴), and measured K-GG-enriched HCT116 samples obtained from 2mg of protein input. This showed that identification numbers improve massively from 15 to 30min and from 30 to 45min LC gradient length (+81% and +31%, respectively), while the gain in K-GG peptide IDs was moderate when further increasing MS acquisition time (8%, for 45–60min and 6%, for 60–75min). Even the 15min LC-MS runs consistently quantified more than 20,000 K-GG peptides with good precision, demonstrating the potential for high-throughput ubiquitinomics applications (Fig. 3c, d and Supplementary Data ⁶). To benchmark DIA-NN’s performance for analyzing samples not treated with proteasome inhibitors, we processed six raw files (DMSO-treated) from Hansen et al in library-free mode. Also in this comparison, DIA-NN clearly outperformed another DIA data processing software, quantifying on average almost 75% more K-GG peptides (Fig. 3e).

In-depth quantitative ubiquitin site profiling for USP7 substrate identification

Currently, there are six clinically approved drugs targeting different components of the UPS and several more in preclinical testing⁴. The unbiased analysis of ubiquitination and deubiquitination events by MS-based ubiquitinomics represents a powerful approach to develop or challenge a drug-targeting hypothesis. However, key aspects such as the modulation and the time-resolved interplay of protein ubiquitination and degradation upon drug treatment have been difficult to address on a truly proteome-wide scale. To meet this challenge, we sought to profile the selective USP7 inhibitor FT671³⁰ with our DIA workflow in a series of time-course experiments. The deubiquitinase USP7 is a potential anticancer drug target that is of particular interest for the treatment of hematological malignancies^12,13,31. For example, USP7 deubiquitinates and stabilizes the E3 ligase Mdm2. As Mdm2 promotes p53 degradation, inhibition of USP7 leads to p53-dependent tumor growth suppression^32,33. However, recent reports indicate that the ubiquitin signaling network regulated by USP7 may be much more complex than initially thought^30,33–35. In this context, global proteomics approaches connecting early protein ubiquitination with later changes in protein abundance represent an elegant tool for pinpointing putative USP7 substrates with high confidence.

First, we quantified the ubiquitinomes and the proteomes at five individual time points after USP7 inhibition with FT671 and in the absence of proteasome inhibitors, to a depth of 45,000 K-GG peptides and over 10,000 proteins in each sample (Fig. 4a, Supplementary Fig. ² and Supplementary Data ⁷). Principal component analysis (PCA) of the ubiquitinome captured the variance of the different treatment times and revealed a clear separation of DMSO and FT671 treatments at each time point.

Open in a separate window

Fig. 4

Identification of USP7 substrates in a time-course experiment.

a Schematic of the USP7 inhibitor time-course experiment. HCT116 cells were treated with DMSO or 10µM of FT671 and harvested in SDC buffer at the indicated time points. After tryptic digestion of 2mg of total protein per sample, both the ubiquitinome and the proteome were acquired in single-shot DIA mode and the resulting raw files processed using DIA-NN. b Volcano plots of the ubiquitinome (15min) and the proteome (6h) after FT671 treatment are shown on the left. Significantly-regulated proteins and K-GG peptides (LIMMA⁵², 5% FDR) are colored (Log2 fold changes −1<x<1 in blue and x < −1orx>1 in red). The Venn diagram shows significantly and >2-fold ubiquitinated proteins at 15min (red), and ubiquitinated proteins that were significantly downregulated by more than 20% over the 6h time course (blue). The overlapping proteins were significantly enriched for categories such as zinc finger and ubiquitin-protein ligase activity. BH FDR=Benjamini–Hochberg false discovery rate. c Heat map of proteins that were significantly downregulated by more than 20% upon FT671 treatment (or more than 50%, in magenta) and that showed significant, and more than twofold induction of at least one ubiquitination site at 15min. Ubiquitin peptide profiles were averaged and both data were matched based on gene level. Topors was included in the heatmap - its protein intensities were below detection limit (missing values, in gray) because it was rapidly degraded upon FT671 treatment. d Profiles of Log2 fold changes (FT671/DMSO) at different time points for different cell cycle regulators upon USP7 inhibition. Plotted are mean±SEM. n=4 biological replicates. e After transfection with different siRNAs (Control, USP7_1, USP7_2) for 48h, HCT116 cells were treated with DMSO/10µM FT671 for 5min. Shown are Log2 fold changes (FT671/DMSO) of all quantified K-GG remnants for siRNAs targeting USP7 (y-axis on both plots) or a control siRNA (x-axes). Selected targets found to be regulated in the time course experiment (Fig. 4a–d) are colored. f Significantly upregulated ubiquitinated peptides at 15min of FT671 treatment were mapped onto a BioGrid³⁷ network for reported USP7 interacting proteins (filtered for four evidences). The proteins were colored according to the peaking time of the ubiquitin signal (averaged profiles for significantly upregulated peptides). Source data are provided as a ^{Source Data} file.

We mapped at least one ubiquitination site on 8,087 proteins. Already after 15min of USP7 inhibition, we found 1,243 ubiquitinated peptides (mapping to 552 proteins) that were significantly upregulated by more than twofold (Supplementary Fig. ³). However, only 42 of these rapidly ubiquitinated proteins were significantly downregulated by more than 20% over the 6h time course (Supplementary Data ⁷). These 42 factors most likely represent direct USP7 targets marked for proteasomal degradation, and gene annotation enrichment analysis revealed that they were overrepresented in transcriptional regulation (Zinc finger) and the ubiquitin conjugation pathway (Fig. 4b, c and Supplementary Data ⁸). We detected upregulated ubiquitination sites for many reported USP7 targets, such as Trim27, UvssA, or Mdm2^32,34,35. Besides these, we identified protein ubiquitination that was followed by degradation for a number of proteins without any reported connection to USP7 (e.g., Rnf220, L3mbtl2, Pcgf6). However, the majority of ubiquitinated proteins, such as USP7 itself or the USP7 interaction partner Daxx³⁶, were not degraded, even though multiple sites were strongly upregulated throughout the time course (Supplementary Fig. ⁴). Our time-resolved quantification of both the proteome and the ubiquitinome in the absence of proteasome inhibition thus distinguishes ubiquitination triggering apparent protein degradation from nondegradative ubiquitin signatures.

Although five Mdm2 sites were significantly upregulated already after 15min, its total protein level only transiently decreased to return to baseline thereafter (Fig. 4d). This most likely resulted from a p53-controlled increase of Mdm2 transcription, as expected and reported previously³³. In fact, ubiquitination sites on p53 initially decreased in intensity, resulting in protein stabilization at later time points, which coincided with an increased expression of its effector p21 (CDKN1A). Interestingly, besides these tumor suppressors, the proto-oncogenes c-Fos and c-Jun were both upregulated upon USP7 inhibition (Fig. 4d).

To confirm that the observed increase in protein ubiquitination is USP7-dependent, we knocked down USP7 expression using two different siRNA pools, treated the cells with FT671 for 5min and extracted the proteins for ubiquitinome analysis. Again, we quantified over 40,000 K-GG peptides and close to 10,000 proteins in each sample with very good precision and MS-based quantification and Western blotting showed a fourfold reduction in USP7 expression upon knockdown (Supplementary Fig. ^5a–f). While the compound treatment alone did not significantly alter the expression of any of the detected proteins at 5min of treatment time, knockdown of USP7 induced profound changes in protein expression, including a stabilization of p53 (Supplementary Fig. ^{5g, h}). As expected, we found that upon USP7 knockdown, for multiple targets identified in our earlier time course experiment the measured protein ubiquitination after FT671 treatment was strongly attenuated. This confirmed that FT671 is selectively targeting USP7 and further validated multiple previously unknown putative USP7 substrates (Fig. 4e and Supplementary Data ⁹).

To further increase the confidence of defining primary USP7 targets and to get more insights into USP7-mediated signaling, we mapped our ubiquitinomics data onto a USP7 interaction network retrieved from BioGrid³⁷. We identified significantly upregulated sites at 15min of FT671 treatment on 63% of high-confidence physical USP7 interactors (evidence from four experiments), strengthening the evidence that these proteins likely represent direct USP7 targets (Fig. 4f).

The strikingly high regulation of protein ubiquitination already at 15min of USP7 inhibition pointed to a rapidly regulated, complex signaling network. To further dissect which of the upregulated ubiquitination sites are directly controlled by USP7, we performed a high-resolution time-course experiment with much faster sampling (at 2, 6 and 10min), again quantifying both the ubiquitinome and the proteome in single-shot mode. At these early time points, we did not observe significant protein changes except for the downregulation of the E3 SUMO/ubiquitin ligase Topors at 10min (Supplementary Fig. 6a). Topors was strongly degraded also in our initial time-course experiment and was below detection limit already 15min after USP7 inhibition (Fig. 4c). In contrast, even at 2min of FT671 treatment, we found 927 ubiquitinated peptides that were significantly and more than twofold induced, with the strongest up-regulation seen for 22 K-GG peptides originating from USP7 itself, pointing at auto-regulation. Orthogonal validation by enriching ubiquitinated proteins with ubiquitin-binding domains, followed by either Western blotting or MS analysis, independently confirmed these results (Supplementary Fig. ⁷). We further detected 151 ubiquitinated peptides mapping to 33 of those proteins that were ubiquitinated and degraded at later time points (15min–6h, Fig. 4), again suggesting that these proteins are bonafide USP7 targets (Supplementary Fig. ^6b and Supplementary Data ¹⁰).

Cell culture, drug treatments, transfection, and cell lysis

HCT116, MM.1S, and Jurkat 6.1 were from ATCC. HCT116 were cultured in DMEM, 10% FCS, 4 mM L-glutamine and 1mM sodium pyruvate. Jurkat 6.1 and MM.1S were grown in RPMI, 10% FCS, 2mM L-Glutamine, and 1mM Sodium pyruvate. MG-132, FT671, FT827, GNE-6640, and GNE-6776 were dissolved in DMSO to prepare a 10mM stock solution. Cells were treated with either DMSO or the indicated compounds, washed twice with ice-cold PBS, and harvested using freshly prepared SDC buffer (room temperature) or urea buffer (chilled to 4°C). The SDC buffer contained 1% SDC, 10mM TCEP, 40mM CAA, 75mM Tris-HCl at pH=8.5 and its pH was adjusted to 8.5 with 1N NaOH. The urea buffer contained 8M urea, 1mM EDTA, 1mM CAA, 150mM NaCl, 50mM Tris-HCl pH 8.0, 2µg/ml aprotinin, 10µg/ml leupeptin, 50µM PR-619 and 1mM PMSF, which was added immediately before use. The SDC lysates were heated to 95°C for 10min while shaking at 750rpm in a Thermomixer (Eppendorf) and then sonicated for 10min (10×30s on/off cycles) using a Bioruptor^® Pico sonication device (Diagenode) (for volumes <2ml) or a ultrasonic probe device (Bandelin Sonopuls, 1min, energy output ~40%, for volumes >2ml). For the SDC/urea comparison in Fig. 1, the urea lysates were processed as described previously¹⁸. Briefly, the cell pellet was homogenized by pipetting up and down and the lysate cleared by centrifugation at 20,000×g for 10min at 4°C. The supernatant was then transferred to a new tube.

For siRNA-mediated knockdown of USP7, HCT116 cells were transfected with either a control siRNA (QIAGEN, 1022076) or two different siRNAs targeting USP7 (siUSP7_1= QIAGEN FlexiTube, 1027417, siUSP7_2= Sigma Aldrich, EHU131171) with RNAiMAX (Thermo Scientific), following the vendor’s protocol. The cells were transfected twice (at day 1 (0h) and day 2 (24h)) and drug treatment (DMSO or 5µM FT671 for 5min) followed by cell lysis performed at 48h post-transfection.

Ubiquitin domain pulldown experiments

HCT116 cells were treated with 10µM MG-132, alone or in combination with 10µM FT671 for 15min. Cells were harvested in ice-cold lysis buffer (50mM Tris-HCl pH 7.5, 0.15M NaCl, 1mM EDTA, 1% NP-40, 10% glycerol, freshly supplemented with 50µM PR-619 and 10mM NEM (N-ethylmaleimide) and spun at 14,000g for 10min. The supernatant was transferred to a fresh tube and the protein concentration determined using the BCA assay (Thermo Scientific, 23225). 20µl of TUBE2 agarose matrix (LifeSensors, UM402) were added to 2mg of total lysate, followed by incubation at 4°C for 1h on a rotor wheel. The beads were collected by centrifugation and washed three times each with 1ml of TBS (20mM Tris-HCl pH 8.0, 0.15M NaCl)-0.1% Tween-20. The proteins were eluted using a 5× SDS-sample buffer (200mM Tris pH 6.8, 5% (v/v) ß-Mercaptoethanol (freshly added), 5% (w/v) SDS, 50% Glycerol, 0.25% (w/v) Bromphenolblau), separated on SDS-PAGE followed by Western blotting. In short, proteins were transferred to a nitrocellulose membrane and the membrane was blocked in 5% milk/TBS-0.1% Triton for 1h at room temperature. After washing for three times with TBS-0.1% Triton, the membrane was incubated overnight with a primary USP7 antibody (Cell signaling technology, 4833, 1:1000 in 2% BSA-TBS-0.1% Triton) at 4°C. The following day, the membrane was washed, incubated with a secondary HRP-linked (Rockland (611–1302), 1:10,000 in 2% BSA-TBS-0.1% Triton) anti-rabbit antibody and USP7 detected by chemiluminescence (ChemiDoc, Biorad). For MS analysis, proteins were digested on-bead. After washing with 3 × 1ml TBS-0.1% Tween-20 and 1 × 1ml TBS, 100µl of SDC buffer were added to the beads, followed by incubation at 95°C for 10min. After cooling to RT, 0.5µg of trypsin was added to each sample, followed by overnight incubation at 37°C in a thermomixer. The following day, 100µl of 99% isopropanol/1% TFA were added and the peptides were desalted using SDB-RPS STAGE tips. STAGE tips were activated using 100µl of 99% isopropanol/1% TFA, then washed with 75µl of 80% ACN/5% NH₄OH, followed by two washes with 99% isopropanol/1% TFA (75µl each). The samples were then loaded onto the tips and washed twice with 99% isopropanol/1% TFA (200µl each) and additional two times with 0.2% TFA (200µl each). Finally, the peptides were eluted with 75µl of 80% ACN/5% NH₄OH, speed-vac dried, and resupended in 20µl of 0.1% formic acid. After peptide concentration estimation on the nanodrop (Thermo Scientific), the concentration was set to 0.4µg/µl and 2µl were injected into the mass spectrometer (800ng).

Crosslinking of K-GG antibody

Crosslinking was performed as described previously¹⁸. In brief, antibody-bound beads were washed with 3 × 1ml of 100mM sodium tetraborate (pH 9.0) and then crosslinked by incubating with 1ml of 20mM DMP/100mM sodium borate (pH=9.0) for 30min at room temperature. The crosslinking buffer was removed and the reaction stopped by washing with 2 × 1ml of 200mM ethanolamine (pH=9.0), followed by incubation with 1ml of 200mM ethanolamine (pH=9.0), with end-over-end rotation at 4°C for 2h. Finally, the beads were washed with 3 × 1ml of IP buffer and either directly used or conserved for up to 2 weeks in 1× phosphate-buffered saline (PBS)/0.02% sodium azide.

K-GG peptide enrichment and LC-MS/MS sample preparation

Protein concentrations were determined using the BCA assay (for experiments in Fig. 1a) or the 660nm assay (all other experiments, Thermo Fisher Scientific). 2mg of urea lysate for each replicate was reduced with 5mM DTT for 45min and then alkylated with 10mM CAA for 30min in the dark before digestion. Proteins were digested with trypsin/Lys-C mix overnight at 37°C (SDC lysates) or at room temperature (urea lysates) with a 1:50 enzyme to protein ratio. For urea samples, the digestion was stopped by adding TFA to a final amount of 0.1% (v/v) and peptides were desalted using C18 cartridges (Sep-Pak tC18, WAT036790) as follows: (a) conditioning with 5ml of ACN; (b) conditioning with 5ml of 50% ACN/0.1% FA; (c) equilibration with 4 × 5ml of 0.1% TFA; (d) loading of the sample, (e) washing with 4 × 5ml 0.1% TFA, (f) elution with 2 × 3ml 50% ACN/0.1% FA. For SDC-lysed samples, the digestion was stopped by adding two volumes of 99% ethylacetate/1% TFA, followed by sonication for 1min using an ultrasonic probe device (energy output ~40%). The peptides were desalted using 30mg (for<1mg of input, 8B-S029-TAK) or 100mg (for up to 2mg of input, 8B-S029-EBJ) Strata-X-C cartridges (Phenomenex) as follows: (a) conditioning with 1ml/3ml (for 30mg and 100mg cartridges, respectively) of isopropanol; (b) conditioning with 1ml/3ml of 80% ACN/5% NH₄OH; (c) equilibration with 1ml/3ml of 99% ethylacetate/1% TFA; d) loading of the sample; e) washing with 2 × 1ml/3ml of 99% ethylacetate/1% TFA; f) washing with 1ml/3ml of 0.2% TFA; g) elution with 2 × 1ml/3ml of 80% ACN/5% NH₄OH. The eluates were snap-frozen in liquid nitrogen and lyophilized overnight. K-GG peptide enrichment was performed as described with some minor modifications¹⁸. Briefly, peptides were resuspended in 1ml of cold immunoprecipitation (IP) buffer (50mM MOPS pH 7.2, 10mM Na₂HPO₄, 50mM NaCl) and incubated with 40µl of a 25% slurry of cross-linked K-GG antibody-bead conjugate (corresponding to 10µl beads/IP) for 2h at 4°C with end-over-end rotation. Beads were then washed twice with 1ml of IP buffer and an additional time with cold Milli-Q^® water. After removing all the supernatant, the beads were incubated with 200µl of 0.15 % TFA at room temperature while shaking at 1,400rpm. After briefly spinning, the supernatant was recovered and desalted using in-house prepared, 200µl two plug StageTips⁴⁶ with SDB-RPS (3M EMPORE^TM, 2241; for SDC) or C18 (3M EMPORE^TM, 2215; for urea). SDB-RPS StageTips were conditioned with 60µl isopropanol, 60µl 80% ACN/5% NH₄OH, and 100µl 0.2% TFA. The K-GG enrichment eluate (0.15% TFA) was directly loaded onto the tips followed by two washing steps of 200µl 0.2% TFA each. Peptides were eluted with 80% ACN/5% NH₄OH. For C₁₈, StageTips were equilibrated with 60µl ACN and 2 × 60µl 0.1% FA before sample loading. StageTips were washed with 2 × 60µl or 0.2% TFA and peptides eluted with 60µl of 50% ACN/0.1% FA. Peptides were Speedvac dried and then resupended in 10µl of 0.1% FA, of which 4µl were injected into the mass spectrometer. For total proteome measurements, a 50µl aliquot of desalted peptide eluate was transferred to a 0.5ml tube, Speedvac dried and resuspended in 15µl of 0.1% FA. The peptide concentration was estimated using a Nanodrop^TM device (Thermo Fisher Scientific) and adjusted to 0.4µg/µl with 0.1% FA, of which 2µl (800ng) were injected into the mass spectrometer.

High-pH reversed-phase fractionation

HCT116 cells were treated with MG-132 (10µM) or FT671 (10µM) for 6h before lysis with SDC buffer (see “Cell culture, drug treatments and cell lysis” for details). 40mg of each lysate were combined before overnight digestion (37°C) with Trypsin/LysC in a 1:50 enzyme to protein ratio. The resulting peptides were desalted using Strata-X-C cartridges (Phenomenex) as described in “K-GG peptide enrichment and LC-MS/MS sample preparation”. The lyophilized peptides were resuspended in 0.1% FA and fractionated using a Zorbax 300SB-C18 column (Agilent Technologies) on an ÄKTA HPLC system (GE Healthcare). Fractionation was performed with a flow rate of 3ml/min and with a constant flow of 10% 25mM ammonium bicarbonate, pH 10. Peptides were separated using a linear gradient of ACN from 5% to 35% over 45min, followed by a 5-min increase to 60% ACN and ramping to 70% over 3min. Fractions were collected at 60-s intervals in a 48-well plate to a total of 36 fractions and then pooled to obtain 12 fractions (A1-C1-E1, A2-C2-E2 etc.). All fractions were acidified by adding FA to a final amount of 0.1% and then lyophilized. Peptides were subsequently resuspended in 1000µl 0.1% TFA and desalted using Strata-X-C cartridges (Phenomenex) as described. Lyophilized peptides of each fraction were resuspended in 1ml of IP buffer and enriched for K-GG peptides. Each fraction was injected trice into the mass spectrometer (36 measurements in total) and the data were acquired with a 75min LC-MS method (DIA). The data were processed in DIA-NN, using the library-free setup.

LC-MS/MS measurements

Peptides were loaded on 40cm reversed-phase columns (75µm inner diameter, packed in-house with ReproSil-Pur C18-AQ 1.9µm resin [ReproSil-Pur^®, Dr. Maisch GmbH]). The column temperature was maintained at 60°C using a column oven. An EASY-nLC 1200 system (ThermoFisher) was directly coupled online with the mass spectrometer (Q Exactive HF-X, ThermoFisher) via a nano-electrospray source, and peptides were separated with a binary buffer system of buffer A (0.1% formic acid (FA) plus 5% DMSO) and buffer B (80% acetonitrile plus 0.1% FA plus 5% DMSO), at a flow rate of 250nl/min (for 75min and 125min gradients). For the 60min method, a flow rate of 300nl/min was used. For the 30min and 45min gradients, a flow rate of 350nl/min was used and for the 15min DIA method, a 15cm column and a flow rate of 500nl/min were used. The mass spectrometer was operated in positive polarity mode with a capillary temperature of 275°C.

The DDA method consisted of a MS1 scan (m/z=300–1650, R=60,000, maximum injection time= 20ms, spectrum data type=profile) followed by TopN MS/MS scans (N=15). These were acquired at R=15,000, AGC target=1e5, maximum injection time= 28ms, isolation window 1.4 Th, NCE=27 and a scan range of 200–2000m/z.

The DIA methods consisted of a MS1 scan (m/z=300-1,650) with an AGC target of 3×10⁶ and a maximum injection time of 60ms (R=120,000). DIA scans were acquired at R=30,000, with an AGC target of 3×10⁶, “auto” for injection time and a default charge state of 4. The spectra were recorded in profile mode and the stepped collision energy was 10% at 25%. The number of DIA segments was set to achieve an average of 4–5 data points per peak (Supplementary Data ⁴).

Scoring of ubiquitination sites implemented in DIA-NN

A set of scores that reflect MS2-level evidence for detection of the modified peptide were implemented in DIA-NN. In DIA proteomics, peptides are matched not just to a single spectrum, but to a series of consecutive spectra, wherein each peptide fragment ion gives rise to an elution profile. In the absence of interfering signals, originating from co-fragmenting peptides that share fragments with the same masses, the elution profiles of fragments of a peptide correlate. Detecting these correlations is the primary method that enables matching peptides to the data in DIA proteomics^24,47,48. In DIA-NN, this concept is explored to boost the confidence of K-GG peptide identification. When assessing the candidate elution peaks, which could potentially originate from a particular peptide, DIA-NN always designates one of its fragments as the “best”, based on which fragment has an elution profile that best correlates with elution profiles of other fragments. The elution profile of the “best” fragment is then considered representative of the true elution profile of the peptide itself²⁴. The “quality” of signals corresponding to all other fragments as well as to the extracted elution profile of the precursor itself (MS1 level) is then assessed based on how well their elution profiles correlate with the elution profile of the “best” fragment, with high correlation being indicative that the fragments likely originate from the same precursor. DIA-NN leverages this principle for confident K-GG detection. A number of scores are calculated, using sums of such correlations, for all the fragments that contain/do not contain the modified residues/other potential modification sites (in different combinations of these conditions), and used to determine the confidence of modified peptide identification (Supplementary Fig. ⁸).

Raw data processing

MS raw files acquired with DDA mode were analyzed using MaxQuant²⁰, version 1.6.17.0 (maxquant.org), whereas files acquired in DIA mode were processed using DIA-NN²⁴, version 1.8 (https://github.com/vdemichev/DIA-NN). Reviewed UniProt entries (human, SwissProt 11-2020 [9606], 42,395 entries; yeast [559292], SwissProt 11-2020, 6,750 entries and E. coli [83333], SwissProt 01-2021, 4,532 entries) were used as protein sequence database for both MaxQuant and DIA-NN searches. For MaxQuant, the standard settings were used (enzyme=trypsin/P, two missed cleavages, carbamidomethylation of cysteines as fixed modification and protein N-terminus acetylation and oxidation of methionine as variable modifications) and K-GG was added as a variable modification. “Match between runs” (MBR) was enabled where indicated. For DIA-NN, one missed cleavage and a maximum of two variable modifications per peptide were allowed (acetylation of protein N-termini and oxidation of methionine). Carbamidomethylation of cysteines was set as fixed modification and K-GG was added in case of ubiquitinomics. Unless specified, all ubiquitinomics data analyses were carried out using library-free analysis mode in DIA-NN. All proteome data analyses were performed in library-free mode. For details on setup of DIA-NN searches, see Supplementary Fig. ⁹. For library-free searches, “deep learning-based spectra and RTs prediction” was enabled. MBR was enabled. This instructs DIA-NN to follow the two-step workflow as described previously²⁴. Briefly, DIA-NN first generates a spectral library from DIA data using all identifications in the specified raw files. The library generation is performed using both global (experiment-wide) and run specific precursor FDR filters of 1%. This library is then used as a spectral library in a second search. The quantification strategy was set to “Robust LC (high precision)”. Spectronaut (v. 14.10.201222.47784) was run with default settings, except PTM localization was enabled and set to 0, while protein q-value filtering was disabled.

Bioinformatic data analysis

Modification-specific peptides of MaxQuant outputs (DDA) were filtered to exclude reverse hits and contaminants. Counts of K-GG and non-K-GG peptides, missing values, and CVs were computed using the Perseus software⁴⁹. Specifically, the modification-specific peptide output table from MaxQuant was filtered for unique peptide sequences for both K-GG and non-K-GG peptides.

For DIA-NN ubiquitinomics outputs, precursors were aggregated to peptides by averaging normalized precursor intensities. Coefficients of variation were computed on the K-GG peptide raw intensities and the plots were created with R (version 3.6.1). The GO term enrichment analysis was done with Perseus⁴⁹.

DIA-NN outputs for Figs. 4 and ​and55 and associated supplementary figures were further processed with R. K-GG precursor intensities were aggregated to peptides using the MaxLFQ⁵⁰ algorithm, as implemented in the DIA-NN R package (https://github.com/vdemichev/DIA-NN-rpackage/). K-GG peptide to site mapping was done using reviewed entries of the human UniProt database (SwissProt, release 11-2020, 42,395 entries). The peptide intensities were normalized by median sample scaling before differential expression analysis.

The proteome raw data were processed with the library-free search modey in DIA-NN. Instead of using the protein groups created by DIA-NN, protein inference was performed in-house following the logic of the ID Picker⁵¹ algorithm.

Significance testing of log2-transformed intensities was performed with LIMMA⁵² on all peptides/proteins present in more than 50% of samples; inhibitor treatments were compared to their corresponding DMSO controls at each time point. Multiple testing-corrected q-values <0.05 were considered as significant. For comparing proteome and ubiquitinome data, protein groups of each dataset were disaggregated into individual UniProt identifiers and then mapped between the data sets. For individual plots, the protein groups of the ubiquitinome data sets along with K-GG site positions were used. Hierarchical clustering was performed using Ward’s method on Euclidean distances, separately for proteomes and ubiquitinomes. The data was visualized in Spotfire (version 7.12.0).

Cytoscape (version 3.7.2, cytoscape.org) was used for mapping the combined interactomics data from BioGrid³⁷ to the ubiquitinomics data from this study. BioGrid interaction data (version 3.5.186) was filtered for human proteins with physical interactions with USP7 in at least four studies. Ubiquitination sites were mapped onto USP7 and its interactors, if they showed significant upregulation at 15min. Proteins without significant ubiquitination regulation are not shown.

We thank all members of Evotec München GmbH and especially S. Blencke, A. Kirgis, T. Lanzinner, J. Linnemann, S. Marx, J. Modrow, R. Jörg, R. Putzhammer and D. Turnblad-Phillips for technical assistance. M. Segura-Lepe, D. Bader, C. Kluger and B. Kracher for help in developing the DIA data analysis pipeline. J. Drechsel for help with drawing of chemical structures. A. Beling, T. Gaitanos and M.G. Steger for proofreading the manuscript. M. Mann, F. Hansen and Ö. Karayel for sharing raw data and for exchange of ideas.