Evaluation of a Pooling Chemoproteomics Strategy with an FDA-approved Drug Library

Abstract

Chemoproteomics is a key platform for characterizing the mode of action (MoA) for compounds, especially for targeted protein degraders such as proteolysis targeting chimerics (PROTACs) and molecular glues. With deep proteome coverage, multiplexed tandem mass tag-mass spectrometry (TMT-MS) can tackle up to 18 samples in a single experiment. Here, we present a pooling strategy to further enhance the throughput, and apply the strategy to an FDA-approved drug library (95 best-in-class compounds). The TMT-MS-based pooling strategy was evaluated in the following steps. First, we demonstrated the capability of TMT-MS by analyzing over 15,000 unique proteins (>12,000 gene products) in HEK293 cells treated with five PROTACs (two BRD/BET degraders and three degraders for FAK, ALK, and BTK kinases). We then introduced a rationalized pooling strategy to separate structurally similar compounds in different pools, and identified the proteomic response to 14 pools from the drug library. Finally, we validated the proteomic response from one pool by re-profiling the cells under individual drug treatment with sufficient replicates. Interestingly, numerous proteins were found to change upon drug treatment, including AMD1, ODC1, PRKX, PRKY, EXO1, AEN and LRRC58 by 7-Hydroxystaurosporine; C6orf64, HMGCR and RRM2 by Sorafenib; SYS1 and ALAS1 by Venetoclax; and ATF3, CLK1 and CLK4 by Palbocilib. Thus, the pooling chemoproteomics screening provides an efficient method for dissecting the molecular targets of compound libraries.

Introduction

Chemoproteomics plays an important role in the process of drug discovery and development, including biomarker discovery, target identification, and characterization of mechanisms of drugs^1–4. Some strategies aim to directly identify protein targets by compound-based affinity methods, such as compound-centric chemical proteomics (CCCP)⁵, and activity-based protein profiling (ABPP)⁶, while other strategies explore proteome-wide structural changes that may be exploited to implicate drug targets, such as thermal proteome profiling (TPP)⁷, limited proteolysis-coupled mass spectrometry (LiP–MS)^{8 9}, drug affinity-responsive target stability (DARTS)¹⁰, native tandem mass tag (TMT) labeling¹¹, and covalent protein painting (CPP)¹². More recently, proteomic profiling has also expanded to the field of analyzing the emerging therapeutic modalities for targeted protein degeneration (TPD) to discover the degrader targets and the corresponding ligases (e.g., lenalidomide¹³ and CDK inhibitor CR8¹⁴). Most of the published work, however, covers only 5,000–9,000 proteins in the cellular proteome, which would miss proteins of low abundance^15–17. With the advance of multiplexed isobaric labeling-based proteomics, as exemplified by 18-plex isobaric tandem mass tag (TMT), the comparison of more than 10,000 proteins across multiple conditions or samples has been achieved^{18–20,21, 22,23, 24}, providing an unbiased way to deepen the analysis of proteomic change upon compound perturbation.

Compared with traditional small molecule inhibitors that usually occupy the activity sites of protein targets, TPD degraders leverage the cellular degradation mechanism to continuously eliminate their targets. For example, a heterobifunctional proteolysis targeting chimeric (PROTAC) compound consists of two active ligands with an intermediate linker: one binds to the protein of interest (POI), and the other binds to an E3 ubiquitin ligase^25–29. The PROTAC compound bridges the POI to the ligase and induces the degradation by hijacking the ubiquitin–proteasome system (UPS)^{30, 31}. Moreover, heterobifunctional degraders have also been developed to utilize other UPS enzymes (e.g., deubiquitinating enzymes³²), as well as lysosomal³³ and autophagy-mediated degradation³⁴, which expand POI to cover membrane-associated proteins and extracellular proteins. Alternatively, molecular glue is a simple monomeric degrader that enhances degradation by binding to the POI-E3 interface instead of individual proteins. These emerging degraders significantly expand the scope of targeting previously undruggable proteins and bring unprecedented opportunities in clinical treatment. However, there is no systematic approach to identify molecular glues, as most molecular glue compounds are serendipitously discovered^{13, 14, 35–37}, suggesting that other compounds, even those approved by the FDA through other mechanisms of mode of action, might have a hitherto overlooked potential to induce protein degradation.

Here, we describe an efficient pooling chemoproteomics strategy for dissecting drug-mediated protein degradation. To benchmark our optimized chemoproteomics platform of TMT-MS, we used it to analyze PROTAC-induced protein degradation and identify more than 15,000 proteins to confirm the expected protein degradation. We further designed a pooling strategy to improve the throughput and applied it to an FDA-approved drug library. Finally, we selected one drug pool and validated the proteomic results by individual drug screening. The data demonstrate that it is a practical approach to couple the pooling and individual drug chemoproteomics screening.

Materials and Methods

Reagents.

Quality control was thoroughly performed on all compounds used for this study. Mass spectrometry was used to confirm compound identity, and UVTWC_ELSD was used to confirm compound purity.

Pooling design

To prevent interactions between compounds during pooling, we designed a pooling strategy as follows: (i) control the number of compounds to 6–7 per pool, and (ii) pool dissimilar compounds based on structure. The structural similarity of all drugs was analyzed by using ChemMine Tools (https://chemminetools.ucr.edu/ )³⁸. Briefly, each compound of the 95 FDA-approved drug library was input by its SMILES ID. All compounds were then clustered by hierarchical clustering using the single linkage method to generate a ranked compound list. Compounds with a ranking difference of cluster were pooled together to make a total of 14 pools.

Cell viability assay and drug treatment for proteome profiling

HEK293 cells were cultured under the standard growth media DMEM (Cellgro) supplemented with 10% (w/v) fetal bovine serum (Thermo Scientific) at 37°C in a 5% CO₂ humidified incubator. In a cell viability assay, HEK293 cells were harvested and plated with 5000 cells in 100μL media per well in 96-well plates. After overnight seeding, 25 μL media containing a 5X dosing concentration of the compounds or DMSO was added to each well. 5 hours after treatment of compounds at 37°C in a humidified 5% CO₂ atmosphere, 12.5 μL 10X alamarBlue (Bio-Rad) was added to each well. Cells were then incubated at 37°C for another 2 hours. The optical density was recorded at 570 nm and 600 nm by plate reader, and the relative viability (RV) was measured as follows:

$$RV=\frac{117216\times {OD}_{570}ofsample-80586\times {OD}_{600}ofsample}{117216\times {OD}_{570}ofvehicle-80586\times {OD}_{600}ofvehicle}$$

To minimize protein changes induced by cell death, the treatment concentration of each compound in the 95 FDA-approved drug library was adjusted to ensure more than 90% viability.

For proteome profiling, cells were seeded in a 6-well plate at 1 × 10⁶ cells per well and cultured overnight. The following days, cells were treated with fresh medium containing pooled or individual compounds (final DMSO concentration at 0.1%) for 5 hours. Cells were treated with DMSO as negative control, and the concentration of each compound used for pooling and individual screening was determined according to the cell viability.

Cell lysis and sample preparation

When harvesting cells, medium was removed from each well, and cells were carefully washed twice with cold PBS. Lysis buffer (0.15 mL, 50 mM HEPES, pH 8.5, fresh 8 M urea, 0.5% sodium deoxycholate (NaDoc), and 1 mM DTT) was subsequently added and liquid was transferred into a tube with glass beads for complete lysis in a Bullet Blender. The protein concentration was measured by BCA (Thermo Scientific, Ref No 23250) with BSA as a standard. Protein digestion was performed as previously reported²². Briefly, 100 μg proteins were proteolyzed with Lys-C (peptide: enzyme = 1:100 w/w) for 2 hrs and further digested by trypsin (1:50 w/w) overnight at room temperature in 2 M urea. Following digestion, the digested peptides were reduced and alkylated before being desalted by C18 spin-columns (Harvard Apparatus, #74–7206) according to the manufacturer’s instruction.

TMT labeling and basic pH fractionation

Desalted peptides were resuspended in 100 μL of 50 mM HEPES (pH 8.5) and labeled by TMTpro (peptide: TMT reagent = 1:1.5 w/w) for 30 min at 21°C. After peptides were fully labeled with TMT reagents, the reaction was quenched by 5% hydroxylamine for 15 min. We carefully generated mixtures in two steps to obtain equal pooling of TMT samples. First, half of each TMTpro sample was pre-mixed and a small aliquot of the mixture was quantified by LC-MS/MS to examine the pooling ratio. Next, the amount of each sample added was adjusted based on the quantified ratio of the pre-mixed sample to achieve an equal TMT pool. Pooled peptide samples were desalted and fractionated by offline basic pH reverse-phase RPLC with an XBridge C18 column (1.7 μm particle size, 3.0 mm × 15 cm, Waters; buffer A: 10 mM ammonium formate, pH 8.0; buffer B: 90% AcN, 10 mM ammonium formate, pH 8.0). The peptides were eluted in a 120 min gradient of 15–45% buffer B, and 360 fractions were collected every 20 s and concatenated to 96 or 48 fractions.

Acidic pH LC-MS/MS analysis

Each fraction from basic pH RPLC was dissolved in 5% formic acid for LC-MS/MS analysis using a previously optimized platform with modification ³⁹. Samples were analyzed with acidic pH reverse-phase LC-MS/MS (75 μm × 15 cm with 1.9 μm C18 resin from Dr. Maisch GmbH, heated at 65°C to reduce back pressure) coupled with a Q Exactive HF Orbitrap MS. The peptides were eluted in a ~60 min gradient from a total 90 min run at 250 nL/min (buffer A: 0.2% formic acid, 3% DMSO; buffer B: buffer A plus 65% AcN) with the following gradient: buffer B was started at 5% and raised to 10% in 8 min, further increased to 15% at 10 min at 500 nL/min, then gradually raised to 46% over 70 min at 250 nL/min. It was then rapidly increased to 95% in 5 min at 500 nL/min and maintained for 1 min before going back to 5% in 1 min. Another round composed of a 95%−5% shift was used for column cleaning, and the final 11 min was used for column equilibration at 5%. MS settings included MS1 (60,000 resolution, 450–1600 m/z range, 1 × 10⁶ AGC, and 50 ms maximal ion time) and MS2 scans (starting from 120 m/z, 1 × 10⁵ AGC, 105 ms maximal ion time, 20 data-dependent MS2 scans, 1.0 m/z isolation window with 0.2 m/z offset, 32 normalized collision energy in HCD, and 10 s dynamic exclusion).

Protein identification and quantification by the JUMP software suite

The JUMP software suite combines pattern matching with de novo sequencing during database searching to improve sensitivity and specificity ^{40, 41} and can handle multiple TMT batches for protein rescue and normalization. All raw files were directly converted to mzXML format and searched against a human target-decoy database (combining Swiss-Prot, TrEMBL, and UCSC databases) for estimation of the false discovery rate (FDR)^{42, 43}. Major search parameters included precursor and product ion mass tolerance (10 ppm), fully tryptic restriction, two maximal missed cleavages, static TMTpro modification (+304.20715), dynamic Met oxidation (+15.99491), and static modifications of Cys carbamidomethylation (+57.02146). Following the database search for the identification of peptide-to-spectrum matches (PSMs), the resulting PSMs were filtered by mass accuracy and grouped by the cutoffs of JUMP-based matching scores to reduce FDR below 1% for proteins. During the quantification of proteins, the TMT reporter ion intensities for each PSM were determined by the strongest signal within a ±6 ppm mass window surrounding the theoretical m/z for each protein.

The MS proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the iProX partner repository with the dataset identifier PXD033780.

Statistical analysis to define differential expressed proteins

We determined differential expressed proteins by the limma R package⁴⁴ to calculate the p values by moderated t-test, and FDR values by the Benjamini-Hochberg procedure. To estimate experimental variations, we compared the quantified sample replicates by fitting Log₂(fold change) data to Gaussian distribution to generate standard deviations. These standard deviations were further averaged to obtain a mean value. Statistically significant changes were usually accepted with FDR cutoff (0.01) and Log₂ (fold change) cutoff (two-fold standard deviations).

Results and Discussion

Evaluation of the TMT-MS chemoproteomics platform by PROTACs

As PROTACs are identified to induce protein degradation by the ubiquitin-proteasome system, resulting in the decrease of protein abundance, we decided to assess if our TMT-MS platform (Fig. 1A) could detect targeted protein changes by five PROTACs, including BET degrader-2, MS4078, MT-802, FAK degrader1, and A1874. (Fig. 1B and Supporting Table S1). HEK293 cells were separately treated with DMSO (a negative control) or each PROTAC for 5 hrs – the short time course was selected to avoid secondary effects (downstream transcriptional changes) on protein abundance⁴⁵. The treated cells were harvested and analyzed by our fully optimized TMT-MS platform^{40–42, 46–48}, including protein extraction and digestion, TMT labeling, extensive basic pH reverse phase liquid chromatography (RPLC), acidic pH RPLC and high-resolution tandem MS, and highly sensitive data processing by JUMP software. Finally, we collected 4.0 million MS2 spectra, leading to 1,526,372 peptide-spectrum matches (PSMs), 219,104 unique peptides, and 15,571 proteins (12,285 genes, Supporting Table S2). To our knowledge, this is one of the most comprehensive quantitative analyses of PROTAC-mediated protein degradation in a single experiment.

Figure 1.

Quantitative proteomic profiling of protein abundance changes in PROTACs. A, Scheme of quantitative chemoproteomic profiling in screening drugs’ effects on protein abundance. HEK293 cells were cultured and treated with drugs for 5 hrs. Treated cells were harvested and digested into peptides, followed by TMT labeling. TMT pooled peptides were further fractionated in basic pH RPLC and analyzed by acidic pH LC-MS/MS. Protein identification and quantification were conducted by JUMP software suite for differential analysis. B, Summary of PROTACs for proteome profiling. C, Heatmap comparing relative foldchange in protein responses to PROTACs’ treatment.

Compared with negative control (DMSO treatment), three PROTACs (BET degrader-2/A1874/FAK degrader1) identified the corresponding degraded targets (BRDs/BRD2/PTK2). BET degrager-2 decreased the levels of BRD2/3/4 by 1.6, 2.0 and 2.7-fold, respectively. Consistent with the literature⁴⁹, BET degrager-2 showed selectivity on BRD2/3/4 compared to other identified BRDs (BRD1/7/8/9) that were not changed under this condition (Supporting Table S2). FAK degrader 1 degraded its targets (PTK2/PTK2B kinase) to 3.0/4.2-fold, respectively. However, the targets (ALK and BTK) of MS4078 and MT-802 were not detected in this analysis, possibly due to the selection of HEK293 cells in which ALK and BTK may not be expressed or expressed at extremely low levels, highlighting the importance of deep proteomic profiling. Instead, these two PROTACs were found to degrade other tyrosine kinases (e.g., FER, CSK and FRK). This profiling strategy thus shows the capability to efficiently discriminate PROTAC-targeted proteins according to measured protein abundance.

Enhanced chemoproteomics throughput by TMT multiplexity and FDA drug pooling

In spite of deep proteome coverage, proteomics profiling is not widely used for high-throughput screening (HTS) due to limited throughput and high cost. To improve the throughput and reduce the cost of proteomics-based screening, we introduced a drug pooling strategy^50–52 and combined it with the multiplexed TMT methods. We selected an FDA-approved drug library (95 best-in-class drugs) that consists of 2–3 compounds of each “Mechanisms of Action” in the major druggable cancer pathways⁵³ to test the pooling strategy.

To design a drug pooling strategy, we attempted to minimize drug interaction by distributing similar drugs into different pools. In total, 14 drug pools were built from the library, each containing 6–7 compounds of dissimilar chemical structures (details in Method, Supporting Table S3). After treatment with the pooled drugs, HEK293 cells were harvested and digested into peptides, followed by TMTpro labeling and LC/LC-MS/MS analysis. Based on the pooling analysis, we focused on one pool showing obvious proteomic alterations, and then analyzed each drug individually by the TMT-MS platform (Fig. 2A).

Figure 2.

Pooling screening of protein abundance changes. A, Scheme of high-throughput quantitative proteome profiling combined with pooling and individual screening. 95 FDA drugs were divided into 14 drug pools for proteome profiling. Based on preliminary analysis of proteome in drug pools, the selected pool was progressed to the next individual stage. Proteomic data analysis coming from an individual drug was then integrated with that from drug pools. B, Heatmap of quantified proteins under treatments with different drug pools using the top 1% variable proteins.

We also performed cell viability experiments to determine the concentration of each compound for the pooling (Supporting Fig. S1). The default concentration was set at 5 μM, considering the degradation ability of PROTACs at 1 μM. To minimize the massive proteomic changes during drug-induced cell death, 77 drugs were used at 5 μM, while the other 18 drugs of high toxicity were used at 1 μM or 0.2 μM to achieve >90% cell viability. In addition, the default treatment time was 5 hrs to minimize the secondary effects common to prolonged drug exposure⁴⁵.

After optimizing the compound concentration to treat HEK293 cells, we performed a large-scale TMT-MS proteomic analysis following the workflow in Fig 2A, identifying 15,693 proteins (12,381 genes, Supporting Table S4) in the proteomes affected by drug pools. The heatmap in Fig 2B demonstrates that the proteome displays distinct effects by each drug pool. Interestingly, the majority of variable proteins in drug pools 1/4/11/12/13 were upregulated, and those in drug pools 6/7/8/14 were downregulated, compared with the DMSO treatment in this experiment. For example, drug pool 13 induced the upregulation of ATF4, GRM5, DDIT4, DUSP1 and MIDN, while drug pool 7 induced the downregulation of HMGCR, C6orf64, PRKX, EXO1 and AMD1. These data clearly indicate that the drug pools exhibit significant impacts on cellular proteome with different patterns, consistent with their functional differences.

Validation of pooling data by profiling individual drug-treated cells

To validate the proteomic changes by the pooling strategy, we selected pool 7 for individual screening in HEK293 cells by the TMT-MS approach (Fig. 3A), because it displayed a high number of downregulated proteins in the pooled screening. There are seven individual drugs in pool 7: one DNA methylation reagent (Temozolomide) and six kinase inhibitors (Venetoclax/Bcl-2, Vemurafenib/B-RAF, Sorafenib/VEGFR, Palbocilib/CDK4/6, 7-Hyddroxystaurosporin/pan-kinase, and AZD4547/FGFR) (Fig. 3B, Supporting Table S5). All drugs were applied with a default concentration of 5 μM, except that AZD4547 was used at 1 μM due to its high toxicity. We had two replicates for each drug and the DMSO control (totaling 16 samples) and profiled 15,215 proteins (11,983 gene) by mass spectrometry (Supporting Table S6). Clustering analysis of the 16 samples shows high reproducibility of the replicates in the heatmap (Fig. 3C).

Figure 3.

Individual screening of protein abundance changes in drug pool 7. A, Summary of individual screening. B, Chemical structures of individual drugs in drug pool 7. C, Heatmap comparing relative foldchange in protein responses to treatments with different individual drugs.

Moreover, we performed four-replicated screening experiments for drug pool 7 and all 7 individual drugs to achieve statistical power, on account of possible experimental variations in these large-scale proteomics studies, and the evaluation of false discovery rate (FDR) to address the multiple comparison problem^{21, 22} (Supporting Table S7). With the new dataset, we sought to determine differentially expressed (DE) proteins in each drug treatment versus DMSO by an ANOVA analysis. The DE proteins (FDR < 0.01 and Log₂FC (fold change) > 2 SD (standard deviation)) caused by drug pool 7 (n = 96), and each individual drug (summed n = 57) are shown in Fig. 4A, in which 33 proteins overlap (Fig. 4B). For example (Fig. 4C, left panel), AMD1 was downregulated 1.5-fold in the individual screening by UCN-01, and detected to be downregulated 1.7-fold in the pooling screening. Consistent results were also obtained for upregulated proteins such as CYR61, CCN1, and ATF3. In another scenario, different drugs may have opposite effects on the same protein target, and therefore the summed effect is observed in the drug pool. For example, we observed that C6orf64 was downregulated in Sorafenib, but upregulated in UCN-01, and so the change in the pooled screen was attenuated. Moreover, we compared this new dataset (pool 7 and individual results, Fig. 4A) with the previous pool 7 (Fig. 2) and individual data (Fig. 3), and found highly consistent patterns (Fig. 4C, right panel). Correlation analysis of pool 7 in the previous and new datasets shows a Pearson correlation coefficient of 0.96 (R² of 0.92, Fig. 4D).

Figure 4.

Integrated analysis of consistent protein abundance changes in pooling screening and individual screening. A, The number of DEs in pooled and individual drug treatments. The cutoffs were FDR < 0.01, and Log2FC > 2 SD. B, The overlapping DEs between pool 7 and individual drugs. Overlapping DEs in pool 7 were compared DEs of pool 7 with summed DEs of 7 individual drugs. Overlapping DEs in individual drug were compared DEs of each individual drug with DEs of pool 7. C, Heatmap comparing relative foldchange in protein responses to pooling and individual drug treatment. D, Correlation analysis of protein Log2FC (pool 7/DMSO) ) between the new dataset of validation and the dataset of pooled screen in Fig. 2. E, Volcano plot to show the total proteome comparison of UCN-01 to DMSO. The cutoffs were FDR < 0.01, and Log2FC > 2 SD. Shown in blue dots are downregulated proteins and red dots are upregulated proteins.

We further selected 7-Hydroxystaurosporin, a prototypical ATP-competitive kinase inhibitor, as an example to examine the most altered DE proteins. Of the 14,918 quantified proteins, 33 proteins (0.22%) exceeded a stringent cutoff (FDR < 0.01, Log₂FC > 2 SD, Fig. 4E). Although 7-Hydroxystaurosporin binds to many kinases with high affinity but little selectivity⁵⁴, only MELK and two cAMP-dependent protein kinases, PRKX and PRKY, were downregulated among 418 detected kinases (Supporting Table S7). Other downregulated proteins include S-adenosylmethionine decarboxylase 1 (AMD1) and ornithine decarboxylase 1 (ODC1), both of which participate in polyamine biosynthesis that is essential for oncogenicity⁵⁵. Conversely, cysteine-rich angiogenic inducer 61 (CYR61) and CCN family member 1 (CCN1) were significantly upregulated. Overall, the high-coverage proteomic profiling reveals some novel drug targets in the HEK293 cells, providing clues for potential drug repurposing of these FDA-approved compounds.

Conclusion

In summary, we have developed an efficient and sensitive pooling chemoproteomics platform to profile protein responses to drug treatment. We showed the robustness of the TMT-MS platform to confirm known targets of commercially available PROTACs, and applied it to the analysis of an FDA best-in-class drug library in a proof-of-principle study. The proteomic screening of the drug pools indeed uncovered proteomic changes, some of which were subsequently validated in individual drug analysis with sufficient replicates. While 6–7 dissimilar compounds were pooled to fit all 16 TMTpro channels in this pilot experiment, more compounds might be mixed as long as individual compounds can reach their effective concentrations, and there is minimal target competition among mixed compounds.

The protein changes in the FDA compound treated cells may stem from one or a combination of the following possibilities: (i) rapid change of transcription and/or translation, (ii) perturbation of protein degradation mediated by the ubiquitin–proteasome system, such as E1/E2/E3 enzymes (e.g., PROTAC or molecular glue mechanism with E3s)⁵⁶, deubiquitinase (e.g., DUBTACs)³², proteasome, or even ubiquitin-binding proteins, (iii) interference of protein interaction with the lysosome/autophagy system, (iv) alteration of protein localization⁵⁷, or protein modifications that indirectly influence protein levels. Although additional studies are required to understand the molecular mechanisms underlying the MoA of these drugs, this feasible pooling strategy can be extended to other cell lines and a large compound library for probing novel protein targets.

Abbreviations:

MoA

mode of action

MS

mass spectrometry

TMT

tandem mass tag

TPD

targeted protein degeneration

PROTAC

proteolysis targeting chimeric