Reimagining high-throughput profiling of reactive cysteines for cell-based screening of large electrophile libraries

Abstract

Current methods used for measuring amino acid side-chain reactivity lack the throughput needed to screen large chemical libraries for interactions across the proteome. Here we redesigned the workflow for activity-based protein profiling of reactive cysteine residues by using a smaller desthiobiotin-based probe, sample multiplexing, reduced protein starting amounts and software to boost data acquisition in real time on the mass spectrometer. Our method, streamlined cysteine activity-based protein profiling (SLC-ABPP), achieved a 42-fold improvement in sample throughput, corresponding to profiling library members at a depth of >8,000 reactive cysteine sites at 18 min per compound. We applied it to identify proteome-wide targets of covalent inhibitors to mutant Kirsten rat sarcoma (KRAS)^G12C and Bruton's tyrosine kinase (BTK). In addition, we created a resource of cysteine reactivity to 285 electrophiles in three human cell lines, which includes >20,000 cysteines from >6,000 proteins per line. The goal of proteome-wide profiling of cysteine reactivity across thousand-member libraries under several cellular contexts is now within reach.

Fig. 1 |. SLC-ABPP using minimal sample input, reduced instrument time and TMT sample multiplexing.

a, Overview of the SLC-ABPP method based on TMT11- or TMTpro16-plex sample multiplexing. Steps include treatment of lysate or cells with compounds (blue circles), treatment with DBIA probe (red circles), trypsin digestion, TMT labeling, combination, avidin enrichment for biotinylated cysteine-containing peptides and analysis of CR by real-time, search-enabled, MS3-based mass spectrometry (RTS-SPS-MS3). b-d, Multiple technical improvements were made and compared to existing ABPP mass spectrometry-based methods, which included cysteine enrichment using a custom, smaller DBIA (ΔM = 239 Da) (b), sample multiplexing using TMT reagents (c) and a fivefold reduction in total input material needed for screening each fragment (d). e, Cysteine profiling from a single 3-h gradient, resulting in fourfold less instrument time needed per experiment. In combination with 11-plex TMT labeling, SLC-ABPP reduced the average instrument time needed to profile each small-molecule fragment to 18 min, a 42-fold improvement. MudPIT, multidimensional protein identification technology.

Fig. 2 |. Benchmarking proteome-wide SLC-ABPP using lead compounds with known targets.

a, The SLC-ABPP experimental design facilitates the profiling of multiple concentrations with replicates within the same experiment. b, Mutant KRAS-specific G12C inhibitor ARS-1620 was profiled at three concentrations in intact HCC44 cells (4-h treatment), and SLC-ABPP quantified >8,000 cysteine sites within 3 h. Cysteine-12 of KRAS^G12C was the most significantly liganded site, and ARS-1620 showed dose dependence toward KRAS^G12C cysteine-12. c, SLC-ABPP profiling of the cyclin-dependent kinase inhibitor THZ1 in HCT116 cell lysates using multiple concentrations quantified >8,500 cysteine sites. Target cysteine-312 on CDK7 was quantified as the most liganded cysteine. d, Ligandable proteome from spleen tissue extracted from C57BL/6 mice treated intraperitoneally with increasing concentrations of ibrutinib for 1 h. BTK cysteine-481 was identified as significantly liganded from >9,000 quantified cysteine sites. All experiments were performed with the following biological replicates: n = 2 DMSO, n = 3 for compound treatments. Data are represented as means ± s.d. Dotted lines represent a CR threshold of 4 (75% reduction in probe binding). IAA, iodoacetic acid.

Fig. 3 |. High-throughput screening of a small-molecule, fragment-based electrophilic library using SLC-ABPP.

a, Library compounds (285 electrophiles), consisting of 128 chloroacetamide- and 157 acrylamide-containing fragment electrophiles, were screened at 25 μM using HCT116 cells grown in 48-well plates generating ~30 μg of protein per TMT channel. In total, 285 small-molecule fragments were screened in triplicate (n = 3 biological replicates) in HCT116 cells, in <3 d of acquisition time per replicate. With the current methodology, it is possible to screen 3,600 electrophiles with 30 d of total instrument time. b, Chloroacetamide-containing fragments showed variable reactivity across the cysteineome, with a median of 14 ligandable (CR ≥ 4) cysteines per fragment. c, Acrylamide-containing fragments displayed lower reactivity, with a median of ~2 ligandable cysteines per fragment. d, No difference in quantitative depth for cysteine measurements was observed between chloroacetamide- (n = 128 compounds) and acrylamide-containing fragments (n = 157 compounds). However, chloroacetamides liganded fivefold more sites and showed fivefold increased reactivity rates compared to acrylamide-containing fragments. e, In sum, 23,364 unique cysteine sites were quantified across 6,145 protein groups, of which 1,743 were ligandable by at least one fragment on 1,134 proteins. Data are represented as means ± s.d.

Fig. 4 |. Small-molecule electrophiles can display high specificity for their protein targets.

a–d, Data distribution skewness (x axis) for global cysteine sites by protein class versus maximum CR (y axis). The distribution skewness value provides a measure of the specificity of each cysteine site’s ligandability. The maximum CR, capped at 20, is the strongest detected ligandable event for each cysteine site across all 285 small-molecule electrophiles. Examples of cysteine sites across protein kinases (a), transcription factors (b), ubiquitin ligases (E3s) (c) and DUBs (d) are labeled with both maximum CR and skewness. e-g, Examples of protein-focused cysteine-reactivity maps showing the results for hundreds of compounds across all reactive cysteines for a single protein. e, Protein-focused map of the ribosomal kinase RPS6KA1 showed cysteine-441, adjacent to the ATP binding pocket, as the most ligandable among four total cysteine sites. f, Protein-focused map of specific active site cysteine ligandability of ubiquitin ligase UBE3A by chloroacetamide CL123. g, Protein-focused cysteine map for the highly reactive protein ACAT1 showing four reactive cysteine sites liganded by 24 different fragments. PDB structure of ACAT1 (2IB8) showing the structural position of each reactive cysteine (red). A highly reactive cysteine pocket containing both active site cysteines (−413 and −126) is shown. h, Paired active site ligandability for cysteines-126 (orange) and −413 (teal) displays many compounds that preferentially bind one cysteine over the other within the same binding pocket.

Fig. 5 |. Proof-of-concept rescreening of 285 compound electrophiles in two additional cell lines.

a-e, SLC-ABPP was used to profile the entire 285-compound library in two additional cell lines (HEK293T and PaTu-8988T at 25 μM in duplicate). a, Reproducibility of reactive cysteine proteome (>70% of all cysteine sites were measured in two of three cell lines). b, In total, 29,603 unique sites were quantified across 6,903 unique protein groups, of which 3,688 were ligandable by at least one fragment on 2,071 proteins in at least one cell line. c, Heatmap displaying cysteine-145 reactivity across chloroacetamide-containing electrophiles in HCT116, PaTu-8988T and HEK239T cells. MGMT cysteine-145 was significantly liganded by 66 and 67 unique electrophiles in HCT116 and PaTu-8988T cells, respectively, and showed excellent reproducibility with a rehit rate of ~80% across compounds. Note that no sites for MGMT were quantified in HEK293T cells, due to low/absent expression. d, Structure of MGMT (Protein Databank (PDB): 1QNT) showing catalytic cysteine-145 (blue). e, Cell-line-averaged, protein-focused, cysteine-reactivity map for MGMT shows preferential binding and ligandability to cysteine-145 for all chloroacetamides over two additional cysteine sites. f,g, Correlation between MGMT activity as measured using the DNA-based chemosenor NR-1 and the CR for cysteine-145 for chloroacetamides using ligandable information generated from HCT116 (f) and PaTu-8988T cells (g). Dotted lines indicate the linear equation used to determine the correlation coefficient. Compounds with high CRs showed decreased MGMT activity. h, Example workflow toward selection of the optimal starting point for synthesis of a covalent MGMT inhibitor. Fragment CL59 was found to be highly potent (inhibited >90% of MGMT activity), significantly liganded MGMT in both HCT116 and PaTu-8988T cells and was highly selective. i, Fragment CL59’s ligandable cysteine map shows remarkable specificity toward MGMT cysteine-145. Dotted lines in b represent a CR of 4, which was used to determine ligandable cysteine sites. Data are represented as means of triplicate measurements (n = 3 biological replicates). ND, not determined.

Fig. 6 |. Selective and specific SRC engagement of p-loop cysteine (C-280) by a small-molecule electrophile.

a, P-loop cysteine-280 ligandability on SRC kinase was measured across 285, 270 and 285 compounds in HCT116, HEK293T and PaTu-8988T cells, respectively. Only one chloroacetamide compound, CL126, significantly liganded cysteine-280. b, Heatmap showing all significantly liganded cysteine targets of CL126 across the three cell lines from Fig. 5, with high reproducibility between cell lines. CL126 also significantly engaged YES1, another Src family kinase, with an identical ATP binding region. c, Quantitative reproducibility between replicates for each cell line shows high reproducibility, with >90% occupancy of SRC cysteine-280 (n = 3 HCT116, n = 2 HEK293T, n = 2 PaTu-8988T biological replicates). CL126 showed tenfold higher specificity compared to the average of all other compounds. d, Cellular engagement assay by NanoBRET in HEK293T cells expressing SRC-Nluceferase and treated by either CL126 at 50 and 25 μM (n = 4 replicates) or the covalent SRC inhibitor SM-71 (n = 5 replicates) for 2 h. CL126 significantly decreased the ability of SM-71 to inhibit BRET at both 50 and 25 μM (P = 1.40⁻⁷ and P = 1.28⁻⁶). e, Representative immunoblot (n = 2 replicates) for the specified proteins after streptavidin enrichment, examining target engagement of SRC in three cell lines treated with CL126 and SM-71 for 2 h followed by incubation with 5 μM TL13–68 (biotin-containing SM-71). f,g, The library screens revealed a class of compounds containing a common cyclic sulfone core that significantly engaged cysteine-113 on the PIN1 PPIase active sites with CR values ≥ 4 in all three cell lines, suggesting SAR (f). Structure of cyclic sulfone core containing fragments with highlighted similar features (g). h, Example of CL71 ligandability map in HCT116 cells showing targeting of PIN1 cysteine-113. i, Heatmap showing all targets with hits to compounds containing a cyclic sulfone in HCT116 cells. j, PIN1 cysteine-113 is one of four targets hit by CL71 in all three cell lines. k, Example proteins from proteome-wide TMT-based expression profiling in HCT116 cells treated with CL71 at three doses for 24 h (n = 3 DMSO, n = 3 for 25 or 10 μM, n = 2 for 1 μM biological replicates). CL71 affects the stability of MYC as well as two MYC pathway proteins, PDE4B and ODC1. Statistical significance was determined using one-way analysis of variance. Data are represented as means ± s.d. ***P < 0.001. ND, not determined; NS, not significant.