Decrypting the molecular foundation of cell drug phenotypes by dose-resolved expression proteomics

Most tablets act on proteins^1,2 and it has been known since the times of Paracelsus that tablets exert their effects in a dose-dependent manner. The molecular processes leading to a drug-induced change in cell phenotype can be broadly divided into (1) target binding, (2) pathway engagement, and (3) cellular reprogramming to reach a new viable state or cell death, together constituting the mechanism of action (MoA) of a drug^2,3. Today, quantitative mass spectrometry is the most comprehensive approach for proteome-wide characterization of drugs at all three levels due to its ability to assay thousands of proteins in complex cellular backgrounds in parallel². The technology does not require any preconceived hypotheses about which proteins a drug might target, which pathways it might perturb, or what the proteomic composition of the new cellular state might be. While phenotypic dose–response measurements have long been standard in pharmacology, there remains a lack of proteomic studies that take dose into account as arguably the most critical characteristic of a drug.

Potent drugs in total engage their cellular targets within minutes, or sometimes hours if they have an especially slow off-rate^4,5. Among the most successful approaches for proteome-wide target deconvolution are activity- and affinity-based proteome profiling. Every method to measure the interaction of a drug with its target(s) directly^6,7,8,9. When performed in a dose-dependent model, they also enable the determination of apparent interaction constants^10,11. Various methods measure drug-induced changes in other biophysical or biochemical properties of proteins such as solubility at elevated temperature^12,13 or in the presence of natural solvents^14,15 sensitivity to oxidizing reagents¹⁶ or susceptibility to partial enzymatic hydrolysis^17,18. While useful, these methods generally require high levels of target engagement to produce measurable effects. Moreover, the observed effects often extend beyond the target itself, thus complicating the distinction between direct and indirect drug effects.

Because many cell pathways are regulated by reversible posttranslational protein modifications (PTMs), mass spectrometry could also be used to measure if a drug engages pathways downstream of the target¹⁹. The timeframe here can often be in the minute to few hours range¹⁹. Published studies generally report large numbers of observable PTM changes due to the use of arbitrary and often high single doses of a drug or because the data were collected after many hours of treatment²⁰. Again, the interpretation of such data can be complicated because both direct and indirect effects are contained in the data. It has only very recently been demonstrated that measuring drug effects on PTMs in a dose- and time-dependent manner is a more valuable approach to pathway engagement measurements because it allows prioritizing the data by drug potency¹⁹.

The variation of a cell to a new functional state in response to a drug is a complex process, generally involving changes in gene expression, messenger RNA (mRNA) and/or protein stabilization or degradation over the course of many hours or even days². The L-1000 connectivity map project²¹ has addressed the transcriptional perspective of drug perturbations and, more recently, many studies have extended such investigations to the level of the proteome^22,23,24. Such data are valuable because they represent the molecular consequences that underlie the cellular endpoint (phenotype) of a drug treatment. However, to the best of our knowledge, a systematic evaluation of the dose–response characteristics of drug-induced proteome expression changes has not yet been undertaken, limiting insights into the molecular basis that drives and informs the observed phenotypic changes.

Here, we close this gap by introducing a method termed decryptE, capable of measuring the dose–response characteristics of expression changes of ~8,000 proteins in human cells in response to a drug. We demonstrate the feasibility and utility of the approach by characterizing 144 drugs with various MoA and highlighting several notable findings, including the repression of (Jurkat) T cell activation in response to histone deacetylase (HDAC) inhibitors. The collective data comprise >1 million dose–response curves, which are accessible via ProteomicsDB and the custom-built decryptE web application for further exploration.

DecryptE for dose-resolved expression proteomics

The decryptE ability (Fig. 1) became developed using Jurkat acute T cell leukemia cells as a mannequin draw and is exemplified by analyzing 144 tablets from 16 drug courses (Supplementary Desk 1). These comprise accredited (53) and part III (15) tablets to boot to part I/II investigational or continually ancient tool compounds (76). Snappy, cells had been grown in forty eight-smartly plates and treated for 18 hours with 5 drug doses in elephantine log₁₀ steps between 1 and 10,000 nM and automobile adjust (dimethylsulfoxide, DMSO). Metabolic insist and cytotoxicity, to boot to cell morphology, had been particular for all tablets across the identical dose vary in parallel and had been simplest marginally affected within the timeframe of the experiment (Supplementary Desk 1 and Extended Data Fig. 1a) whereas observed proteomic drug effects had been most pronounced (Extended Data Fig. 1b–d). Proteins had been extracted by SDS-containing buffer and digested into peptides on a robotic platform following the single-pot, sturdy-part-enhanced sample preparation protocol (SP3) ability²⁵. We previously demonstrated that microflow-liquid chromatography with tandem mass spectrometry (LC–MS/MS) permits excessive-throughput proteome measurements²⁶ and, right here, extended the ability by incorporating an ion mobility dimension (excessive-field uneven ion mobility spectrometry, FAIMS) to carry out a proteome protection of>7,000 proteins per hour (Extended Data Fig. 1e–i). Your entire drug show veil required 768 hours of instrument time (a lot like 5.3 h per drug) and led to the identification and quantification of 8,892 proteins using MaxQuant and Prosit rescoring^27,28. In step with forty eight DMSO replicates, a median quantitative precision of 19% coefficient of variation (CoV) became particular for the assay (Extended Data Fig. 2a) with a excessive level of data uniformity (Extended Data Fig. 2b). Dose–response curves had been fitted to the information offering data on drug efficiency (effective focus required to carry out 50% of the cease, EC₅₀) and cease dimension (situation underneath the curve or fold replace over DMSO). The statistical energy of the dose–response data enabled sturdy classification of 1,133,847 dose–response curves (regulated or no longer) that fashioned the thought that for all extra diagnosis. DecryptE data had been reproducible with 69.5% of all particular EC₅₀ values within half a log₁₀ of drug focus (Extended Data Fig. 2c–e). Moreover, the CoVs of regulated proteins had been invariably better than for no longer regulated proteins (Extended Data Fig. 2f).

To facilitate the spend of this helpful resource by the neighborhood, the information can even be explored in ProteomicsDB (https://proteomicsdb.org/decryptE)²⁹ to boot to in a custom-constructed Shining App (https://decrypte.proteomics.ls.tum.de/) in which dose–response curves can even be visualized and when put next. Extra data on cell morphology, cell metabolic insist, cytotoxicity, protein half-lives and protein targets of compounds and drug-aim affinity (where accessible) are supplied to inspire decoding the observed effects.

High-level analysis of decryptE profiles

A total of observations were immediately apparent from a global analysis of the data. First, the abundance of most proteins did not change in response to any drug within the timeframe of the experiment (18 h; n = 982,824 dose–response curves; 87%) (Fig. 2a). Dose-dependent up-regulation occurred in 73,299 conditions and dose-dependent down-regulation was observed in 77,724 conditions. Second, the extent to which any of the 144 compounds altered the proteome of Jurkat cells varied greatly. Some compounds regulated the expression of >1,000 proteins, others only a few (Fig. 2b). Similarly, some compounds showed very potent effects, others only at high concentrations (Extended Data Fig. 3a). Both aspects are important in order to attribute the observed phenotypic (here morphology, metabolic state and cytotoxicity) and molecular (here protein expression) response of a cell to the MoA of a particular compound. As one might expect, compounds targeting common cellular processes elicited many changes. For example, HDAC inhibitors such as vorinostat or panobinostat alter transcriptional processes and, as a result, the expression of many proteins. The proteasome inhibitor carfilzomib also showed large effects because it inhibits a critical protein degradation machinery in cells. Many changes were also observed for the HSP90 inhibitor geldanamycin because it inactivates a key component of the protein folding machinery. More specifically, geldanamycin strongly up-regulated proteins (up to 50-fold) involved in the unfolded protein response (for example, DNAJB1, HSPA1B) presumably because of a cellular attempt to counteract the drug-induced loss of protein folding capacity (Extended Data Fig. 3b). In stark contrast, some compounds elicited only minor proteomic changes. Among these were the histone lysine methyltransferase inhibitor lirametostat or the dual c-MET and ALK kinase inhibitor crizotinib. The former suggests that interfering with dynamic histone lysine methylation in cultured Jurkat cells did not incur any consequences within the timeframe of the experiment and the latter implies that the viability of Jurkat cells is not dependent on ALK and MET activity.

It is generally accepted that drug-induced proteome expression changes can be used for the deconvolution of drug targets^12,22,24. A key takeaway from the realm of data analysis is that this is often not the case. First, while the list of 8,892 detected proteins includes 66% of all known targets of the drugs investigated here (expression regulated or not, Extended Data Fig. 3c), known targets for about 34% of all drugs were missing, and simulations showed that this number increased as proteome coverage decreased (Supplementary Fig. 1). Second, only about 25% of all drugs altered the expression levels of their known protein target(s) (Fig. 2c). Third, even when this occurred, the actual drug target generally did not stand out from the data in terms of potency or effect size, as illustrated by thymidylate synthetase (TYMS). Although the protein was up-regulated by its known binders methotrexate (MTX) and pemetrexed (Extended Data Fig. 3d), TYMS levels were also regulated by 63 other compounds that are not reported to target TYMS (Fig. 2d). Another example is the HSP90 inhibitor tanespimycin. HSP90 levels were regulated by the drug (Extended Data Fig. 3e), but so were hundreds of other proteins, many more potently and with larger effect sizes than HSP90 itself (Fig. 2e). When generalizing this analysis, more often than not, a protein showed drug-induced expression changes even though it was not the target of that drug (Fig. 2f). Therefore, it seems unlikely that a drug target can generally or clearly be identified from drug-induced protein expression changes alone.

Multi-omics diagnosis of drug-prompted cell remodeling

To be taught whether drug-prompted protein expression modifications are rooted in altered transcriptional purposes or pre-, co- and/or posttranslational mechanisms, we performed dose-dependent RNA sequencing (RNA-seq) experiments for seven selected tablets in the identical cell line and underneath the identical drug remedy stipulations. As evident from Fig. 3aloads of concordant and discordant effects had been observed. For instance, protein and mRNA ranges of HDAC1 remained unchanged in response to the HDAC inhibitor vorinostat. In distinction, protein and mRNA ranges of the cell cycle regulated protein RRM2 had been equipotently diminished in response to the CDK4/6 inhibitor palbociclib. This could be explained by the dose-dependent lengthen in the amount of cells enthralling at a stage of the cell cycle where RRM2 ranges are low. Conversely, the proteasome inhibitor carfilzomib up-regulated both transcript and protein ranges of the cochaperone BAG3 with identical efficiency but with very diversified cease sizes, suggesting that BAG3 protein ranges simplest quite lengthen in cells on drug remedy. Any other challenge is offered by the DNA methyltransferase DNMT1 for which protein but no longer transcript ranges had been diminished in response to decitabine. Here is essentially essentially based on literature reporting that decitabine, when integrated into DNA, covalently traps DNA methyltransferases, in turn, main to their degradation³⁰. A identical behavior became observed for molecular glues similar to pomalidomide that led to a potent and dose-dependent discount of the protein IKZF1 but no longer its mRNA level (Extended Data Fig. 4a). The aforementioned drug MTX led to an spectacular and dose-dependent lengthen in protein ranges of its notify aim DHFR whereas mRNA ranges remained unchanged (Fig. 3a). This clearly points to a posttranscriptional occasion. Old in vitro experiments have shown that DHFR binds its own mRNA to repress its translation and that addition of MTX abolishes this repression³¹. This mechanism could maybe be an clear set apart off of the observation that MTX also induces a extremely sturdy thermal or solvent balance shifts for DHFR when sure to MTX^12,13,15.

Any other case is provided by the twin specificity protein kinases CLK1–4 that showed potent up-regulation of both mRNA and protein levels on treatment with the kinase inhibitors brigatinib, abemaciclib and milciclib (Fig. 3b and Extended Data Fig. 4b–e). Published target deconvolution data confirmed that these proteins are direct targets of all three drugs¹⁰. In addition to the full-length protein, CLK1 also exists in two shorter variants that contain the N terminus but lack the kinase domain either due to intron 4 retention or exon 4 skipping. The different forms of the protein arise from the ability of CLK1 to regulate its own splicing by phosphorylating specific splicing factors³². Quantitative PCR with reverse transcription (RT–qPCR) data shown here for CLK1–4 confirmed that the ratio of N-terminal to full-length transcripts shifted in favor of the full-length transcript at higher drug concentrations, in turn, leading to higher levels of full-length protein. While an interesting observation, it remains unclear whether this has any functional consequences in cells as these drugs block kinase activity at the same time. These selected conditions highlight several discrepancies between drug-regulated transcriptome and proteome changes that may stem from different cellular mechanisms. Many more conditions are in the data and often with very good drug-specific differences, both in absolute and relative terms (Fig. 3c). It is also evident from these data that the direction of regulation is not always concordant at mRNA and protein levels (Extended Data Fig. 5a). Opposing regulation events are uncommon for many of the drugs studied. Nevertheless, carfilzomib treatment up-regulated components of the protein folding machinery at the mRNA level while down-regulating the respective proteins (Extended Data Fig. 5b–d), presumably in an attempt to maintain proteostasis.

Drug response phenotypes group tablets by feature

Whereas individual tablets may have diverse targets, they could result in identical cell and molecular drug phenotypes³³. To determine whether decryptE profiles can group tablets in this manner, we performed gene ontology (GO) enrichment analysis for up- or down-regulated proteins for each compound individually, followed by hierarchical clustering of the results for all 144 tablets (Fig. 4a and Supplementary Table 2). Indeed, compounds leading to cell cycle arrest formed two mirrored clusters (C1 and C2) characterized by up- or down-regulation, respectively, of enriched GO terms linked to, for example, sister chromatid separation, mitosis and/or meiosis or cytokinesis. Closer inspection revealed that this analysis identified compounds that arrest cells in G1/S or G2/M (Fig. 4b). Examples of proteins that drive this clustering are the strong up- and down-regulation of the cell cycle-regulated proteins PLK1 and ANLN, respectively. When summarizing this knowledge for all proteins that are up-regulated or down-regulated, respectively, for all three tablets, they showed a congruent distribution of pEC₅₀ values (Fig. 4c–f). On this basis, and when following a guilt-by-association argument, mitotic properties could be assigned to proteins not yet annotated in this process, and this approach could apply to other molecular properties present in other clusters. The pEC₅₀ plots also ranked tablets by potency, identifying paclitaxel as the most potent mitotic inhibitor in the set.

The two aforementioned HSP90 inhibitors formed a small but distinct cluster (C3) driven by GO terms linked to the up-regulation of the unfolded protein response (Supplementary Table 2). The PI3K/mTOR inhibitor GSK-1059615, the DNA crosslinker oxaliplatin and the p53 activator serdematan formed a distinct cluster indicative of down-regulated ribosome biogenesis (C4) (Supplementary Table 2). The three platinum-containing drugs oxaliplatin, carboplatin and cisplatin did not cluster. And indeed, their decryptE profiles were quite different as exemplified by the down-regulation of ribosomal proteins by oxaliplatin but not the others, implying different cellular modes of action (Extended Data Fig. 6a,b)³⁴.

HDAC inhibitors impair T cell activation

Without warning, HDAC inhibitors formed a cluster (C5) with strong links to T cell proliferation and activation (Fig. 4a). For instance, panobinostat down-regulated the expression of many key elements of the T cell receptor (TCR) with low nanomolar efficiency (Fig. 5a), particularly the TCR itself and its coreceptors (Fig. 5b). Cell viability was only marginally affected within the timeframe of the experiment (Extended Data Fig. 7a and Supplementary Table 1). The other HDAC inhibitors showed the same qualitative result (Extended Data Fig. 7b) and the dose-dependent RNA-seq data for vorinostat indicated a concerted transcriptional mechanism rather than protein degradation (Extended Data Fig. 7c). These results show that the reduction of TCR components can be directly attributed to the loss of HDAC activity. This also resulted in the reduction of anti-CD3 and/or CD28 antibody-mediated T cell activation in genetically engineered Jurkat TCR and/or CD3 effector cells that express luciferase in response to T cell activation (Fig. 5c).

To examine whether HDAC inhibition also diminishes protein expression of TCR components in primary human T cells, we isolated CD4 and CD8+ T cells from healthy donors and exposed untreated (referred to as ‘naïve’) and anti-CD3/CD28-activated cells to various HDAC inhibitors (Fig. 5d). Live-cell imaging confirmed that drug-treated primary cells exhibited a reduced ability to bind to beads carrying anti-CD3/CD28 antibodies (Fig. 5e and Extended Data Fig. 7d). Furthermore, all tested HDAC inhibitors recapitulated the findings of the in vitro Jurkat cell line experiments in all four ex vivo cell populations, exemplified by the dose-dependent loss of CD247, CD3D and CD3E (Fig. 5f). Among many other proteins, a dose-dependent reduction of the transcription factor TCF7, the master regulator of naïve T cell differentiation, was observed in naïve cells after HDAC inhibitor treatment. In activated cells, we observed a reduction of granzyme B levels, a key regulator of T cell activation and proliferation (Extended Data Fig. 7e,f). These results clearly demonstrate that HDAC inhibition impacts T cell activation and differentiation with yet unknown functional consequences, but potentially significant implications for the use of HDAC inhibitors as anticancer agents or as tools to study T cell biology (Discussion).

DecryptE particularly addresses the longer-term dose-dependent response of a cell to a drug (or other bioactive agent) like many phenotypic assays. The difference is that decryptE yields thousands of molecular readouts rather than just one (for example, cell viability or morphology changes). As such, the ability should not be confused with proteomic technologies aiming to identify the targets of a drug or illuminate the signaling pathways that lead to a cellular endpoint. These well-known aspects of drug MoA may be contained in decryptE profiles, but they may not be apparent from the data without substantial prior knowledge. Instead, decryptE profiles reflect the third component of cellular drug MoA, referred to in the introduction, which is the transition of the proteomic makeup to a drug-tailored (new) cellular state. There are two critical new technical aspects in the present work. First, showing that the combination of microflow-LC and FAIMS yields deep proteomic coverage and quantitative data (dose–response curves) for one drug and approximately 8,000 proteins in just over 5 hours of analysis time. Second, demonstrating that dose–response measurements provide information not obtainable from single doses. The decryptE approach thus paves the way for large-scale proteome-wide drug perturbation screens that could be further enabled by combining faster and more sensitive mass spectrometers than used here with data-independent acquisition or precise isotope multiplexing by tandem mass tags^22,35. With more than 1 million dose–response curves, our data already provide a valuable resource for the scientific community that can be analyzed in many ways not covered here. For instance, we considered only sigmoidal dose–response trends because these are regarded as the best-understood drug–protein interactions. However, the data could additionally contain nonsigmoidal drug-induced behaviors that might, for example, indicate pharmacological switches in a cell.

Whereas decryptE profiles faithfully describe modifications in protein expression in response to a drug, these could arise by various mechanisms that add well-known data relating to drug MoA. In light of the comparisons made here between mRNA and protein–drug profiles, we suggest measuring transcriptomes and proteomes systematically in a dose-dependent model in parallel to better understand to what extent transcription itself or splicing events play a role. Similarly, including proteomic measurements that address protein synthesis and degradation, for instance by pulse-labeling using stable isotopes^36,37will provide additional valuable insights. The latter is particularly important given the considerable interest in current drug discovery in the development of chemical degrader molecules such as proteolysis targeting chimeras or molecular glues.

Although not investigated here, we note that protein level changes induced by a drug may be cell-type specific. DecryptE profiling of immunomodulatory imide drugs (IMiDs) such as thalidomide, pomalidomide, lenalidomide and iberdomide did not show changes in protein levels for components of the E3 ligase complex itself (CRBN, DDB1, CUL4a) (Extended Data Fig. 8a,d). This suggests that the ubiquitin ligase complex acts as a conventional enzyme that releases its neo-substrates after ubiquitin transfer and that the molecular glue functions as a catalyst. Endogenous CRBN substrates (GLUL, ORAI1) were unaffected by IMiD treatment (Extended Data Fig. 8b,c). DecryptE profiles further confirmed that three of the four IMiDs degraded the neo-substrate IKZF1 in Jurkat cells in a dose-dependent manner (Extended Data Fig. 4a). However, this was not the case for other reported neo-substrates including IKZF2, IKZF4 and PATZ1 (ref. ³⁸). RAB28 was identified as a novel neo-substrate of Iberdomide in Jurkat cells (Extended Data Fig. 8f)³⁹. Such apparent discrepancies with the literature likely arise from molecular differences in the ubiquitin ligase machinery present in a particular cellular model.

We ticket, that observed drug effects no longer simplest depend upon the mannequin draw ancient but also on time, that are diversified for every compound. Here is supported by the mountainous differences in both the absolute amount of rules to boot to which proteins show drug response when comparing decryptE profiles with revealed single dose data^22,23,24 (Supplementary Fig. 2 and data deposited on MassIVE).

Future extensions of decryptE must consist of PTMs whatever the truth that long-term drug responses could maybe end result in complex PTM datasets that can even be complicated to elaborate. Particular conditions from the present work illustrating this need are pemrametostat and onametostat. Every are inhibitors of the protein arginine methyltransferase PRMT5 main to diminished methylation ranges of aim proteins. By including methylation as a variable modification in a old skool database glance for protein identification, it became most likely to measure in-cell inhibition of enzymatic insist with low nanomolar efficiency by monitoring methylation sites on the PRMT5 substrate SNRPB in response to the 2 tablets (Extended Data Fig. 8f). This would have long previous disregarded if the PTM level became no longer regarded as.

The maybe most fun pharmacological end result of the present work is the observation that HDAC inhibitors led to sturdy and potent down-regulation of the TCR with a concomitant discount of the flexibility to mount a T cell response. This will most likely smartly demonstrate the efficacy of HDAC inhibitors in the remedy of TCR signaling-driven T cell lymphoma or the attenuation of TCR signaling observed in animal models of obvious autoimmune illnesses^40,41. At the identical time, attributable to TCR insist is excessive for T cell lineage choice, antigen specificity, effector feature and survival, a repressed expression of TCR complex elements could maybe became detrimental for the remedy of so-known as ‘sizzling’ tumors which would per chance be characterized by immune cell infiltration, and which generally answer to immune checkpoint inhibition remedy. In this context, clinical trial designs would per chance be known as into request that mix immune checkpoint inhibition with HDAC inhibitors⁴². Nonetheless, there could additionally be worthwhile eventualities. Continual excessive antigen stimulation can lead to the phenomenon of T cell exhaustion, diminishing the flexibility of the immune draw to fight a tumor^{43,44,forty five,46}. In such conditions, and reckoning on the immune attach of the tumor, it must be most likely that repressed expression of TCR complex elements in response to HDAC inhibition reduces the absolute level of TCR stimulation to a level that reinvigorates exhausted T cell responses. Clearly, extra purposeful research are required to better understand such most likely HDAC inhibitor-linked effects in sufferers and the most likely of HDAC inhibitors as research instruments in the context of studying T cell exhaustion.

Taken together, the results obtained in this study suggest that dose-dependent and proteome-wide measurements of drug-induced protein expression changes should become a standard tool alongside dose-dependent target deconvolution and pathway engagement studies. The combined data could be very valuable for basic research as well as preclinical and clinical drug discovery because it provides a deeper understanding of the molecular mechanisms of bioactive compounds, from chemical probes to human medicines.

Cell culture

Human Jurkat cells Clone E6.1 (ATCC TIB-152) had been cultured in RPMI-1640 containing 10% fetal bovine serum (FBS) at 37 °C and 5% CO₂. Culture medium became refreshed every 2–3 days and cells had been saved at densities between 0.5 × 10⁶ and a pair of × 10⁶ cells per ml unless lysis or drug remedy.

Cell line authentication was performed using single nucleotide polymorphism profiling (Multiplexion).

Compound data

The information on the target sites of the 144 compounds included in this study was obtained from DrugBank Online (as of July 2023) and vendor specifications. Data regarding the clinical stage the compounds were in at the time the study was conducted were retrieved from ChEMBL (as of July 2023).

Compound remedy

Compounds had been prediluted in DMSO and extra in culture medium interior a forty eight-deep-smartly plate. Per forty eight-deep-smartly plate, three DMSO controls had been added. For remedy, 4 × 10⁶ cells in RPMI-1640 medium supplemented with 10% FBS had been added on top of every compound predilution main to a remaining quantity of 2 ml and remaining remedy concentrations beginning from 10 µM to 1 nM in elephantine log₁₀ steps, main to 5 doses for every drug (10 µM, 1 µM, 100 nM, 10 nM, 1 nM). Cells had been incubated for 18 h if no longer acknowledged otherwise at 240 rpm, 37 °C and 5% CO₂. The next day, cells had been subjected to viability evaluation and lysis.

Confluency, viability and metabolic activity assessment

For determination of cell viability and metabolic activity after compound treatment, 100 µl of cell suspension per well was added to a 96-well plate containing 50 µl of IncuCyte Cytotox Dye (250 nM final concentration, Sartorius) and alamarBlue Cell Viability Reagent (10% final concentration (v/v), Invitrogen). The plate was placed into the IncuCyte live-cell imaging system (37 °C and 5% CO₂) and cells were analyzed for cytotoxicity over a time course of 3 h (×10 magnification, scan mode was standard with 5 images per well, channel selection was phase contrast and fluorescence (300 ms acquisition time), scan interval was every hour). The integrated tool of IncuCyte (Standard Analyzer) was used for confluency and cytotoxicity analysis. After 3.5 h, metabolic activity was determined by fluorescence measurement of the AlamarBlue reagent using the fluorescence readout on the microplate reader FluoStar Omega (l_ex = 544 nm and l_em = 584 nm, BMG Labtech).

For confluency and metabolic activity evaluation, the resulting values had been normalized to the mean values for the DMSO control. For cytotoxicity, the values had been corrected for differences in confluency, before normalizing the values to the mean values for the DMSO controls. Dose–response curves had been fitted to the data as described below (section ‘Curve fitting’).

Any small images displayed in the paper or in other locations had been exported from the IncuCyte system as shown and not further modified.

Cell lysis for protein extraction

To prepare cell lysate from untreated cells (for optimization purposes), cell suspension was centrifuged at 172g for 5 min at room temperature, washed with PBS (phosphate buffered saline, without calcium or magnesium) and pelleted before resuspension in lysis buffer (2% SDS, 40 mM Tris/HCl, pH 8, 95 °C).

Lysis of compound-treated cells became performed in 96-deep-smartly plates. Therefore, after 18 h of remedy time, forty eight-deep-smartly plates had been centrifuged (172g10 min, 4 °C), supernatant became discarded, cell pellets had been resuspended in PBS and transferred to a 96-deep-smartly plate. Cell pellets had been washed two more events with PBS and centrifuged to discard the supernatant sooner than lysis in 100 µl of lysis buffer.

For hydrolysis of DNA, lysate was heated to 95 °C for 10 min while shaking at 172g and trifluoroacetic acid was added to a final concentration of 1% (v/v) and incubated for 1 min while shaking. As a result, N-methylmorpholin (NMM) was added for neutralization to the hot lysate to a final concentration of 2% (v/v). Lysate was stored at −20 °C until further use.

Tissue and bacteria sample preparation

Mouse muscle (M. musculus) and Arabidopsis thaliana (A. thaliana) tissue samples were snap frozen in liquid nitrogen before homogenization using the TissueLyser II (Quiagen, 5 min, 30 Hz, using one stainless steel bead with a 5 mm diameter). Lysis buffer (4% SDS, 40 mM Tris/HCl, pH 8) was added after removing the bead and samples were sonicated using the Bioruptor Pico (Diagenode, 25 cycles with 30 s on/off). DNA hydrolysis was performed as described above using final concentrations of 2% trifluoroacetic acid and 4% N-methylmorpholin, respectively. Lysates were cleared by centrifugation (60 min, 4 °C, 21,000g). Supernatant lysate was stored at −20 °C until further use.

Escherichia coli (E. coli) and Pseudomonas aeruginosa (P. aeruginosa) had been grown in a shaker culture in Luria-Bertani medium at 37 °C, 300 rpm. When reaching an optical density of 0.5 and 0.6, respectively, cultures had been harvested by centrifugation (172g for 60 min, 4 °C) and washed twice with PBS. Lysis buffer was added to the pellet, followed by DNA hydrolysis as described above. Lysate was sonicated using the Bioruptor Pico (above) before clearance by centrifugation (60 min, 4 °C, 21,000g). Cleared lysate was stored at −20 °C until further use.

Isolation and sorting of T cells from healthy donors

Thrombocyte-depleted blood samples had been obtained from two healthy, voluntary human donors (male, age 26) after they gave written and informed consent. This study was approved by a vote from the ethics committee of the University Hospital München rechts der Isar (564/18S). Samples were transferred into 50 ml Falcon tubes, with each tube containing approximately 15 ml of blood. The Falcon tubes were then filled up to a total volume of 37.5 ml with PBS, and the blood was thoroughly mixed. To isolate peripheral blood mononuclear cells (PBMCs), a 12 ml layer of Pancoll was meticulously underlaid using a 24 ml syringe with an extended needle (G 20 × 2 3/4’; Ø 0.9 × 70 mm). As a result, the blood samples were subjected to centrifugation using a programmed gradient (acceleration of 7, deceleration of 1, 2, 7g for 20 min at room temperature). Following the gradient centrifugation, the plasma portion was discarded, and the PBMC-containing buffy coat was carefully collected. The PBMCs were then washed with 50 ml of PBS using centrifugation (441g for 5 min, at room temperature).

For cell separation, 10⁷ PBMCs were resuspended in 40 µl MACS buffer (PBS, 1% FCS, 2 mM EDTA) and incubated with 10 µl antihuman CD4 beads for 15 min at 4 °C. As a result, PBMCs were washed with 15 ml of MACS buffer and centrifuged. CD4 T cells were positively enriched using the autoMACS Pro Separator. The flowthrough was retained and used for the isolation of CD8 T cells according to the CD4 T cell isolation protocol. Isolated T cells were cultured in RPMI-1640 containing 10% FBS and 1% penicillin and streptomycin (37 °C, 5% CO₂) and were either subjected to HDACi treatment immediately or activated as described below.

HDACi treatment of peripheral T cells from healthy donors

For every population (CD4⁺/CD8⁺) a portion of cells was activated using Dynabeads Human T-Activator CD3/CD28 for T Cell Growth and Activation (Invitrogen) and incubated for forty-eight hours (37 °C, 5% CO₂) before HDAC inhibitor (HDACi) treatment. Naïve T cells were subjected to treatment immediately after isolation and sorting. Regardless of activation status, cells were treated with various HDACi (5 doses for each drug: 10 µM, 1 µM, 100 nM, 10 nM and 1 nM) for 18 h, followed by viability, confluency and cytotoxicity evaluation as described above. Cell lysis, protein extraction followed by proteomic workflow and LC–FAIMS–MS/MS analysis was performed as described in the respective sections. For samples, where available material was limited, protein input was adjusted for tryptic digestion and obtained peptides were loaded on Evotips and analyzed on an Evosep-FAIMS-Exploris setup as described previously⁴⁶ (for a complete list of instrument settings, see Supplementary Table 3).

Transcriptome sample preparation and diagnosis

For transcriptome diagnosis, Jurkat cells were treated according to the protocol described above. After 18 h, cells were lysed and total RNA was extracted using the ReliaPrep RNA Cell Miniprep Kit (Promega), according to the manufacturer’s protocol, and evaluated on a 2100 Bioanalyzer (Agilent Technologies). RNA library preparation was performed using the 3′ mRNA-Seq Library Prep Kit FWD with Unique Dual Indices (Lexogen) and was sent to Lexogen for gene expression profiling. Alignment of obtained reads was performed using the data processing pipeline provided by the manufacturer using the QuantSeq FWD pipeline and Homo sapiens (H. sapiens) genome annotation. The obtained alignments were trimmed, reads were counted and normalized. Dose–response curves were fitted to the data as described below (section ‘Curve fitting’).

SP3 sample preparation and tryptic digestion

Protein yield was determined by Thermo Pierce BCA (bicinchoninic acid) protein assays. All steps were performed according to the manufacturer’s protocol.

Before tryptic digest, detergent was removed by single-pot SP3 clean-up, following the protocol first described by Hughes et al.²⁵ tailored to a Bravo Agilent liquid handling platform. In brief, lysate containing 200 µg of protein was mixed with 1 mg SP3 beads (50:50 mixture of Sera-Mag carboxylate-modified magnetic bead forms A and B (Cytiva Europe)) in a 96-deep-well plate and proteins were precipitated onto the beads in 70% ethanol in ddH₂O (double distilled water).

The beads had been washed three times with 80% ethanol in ddH₂O and once with 100% acetonitrile (ACN). Disulfide bonds had been reduced with 10 mM dithiothreitol for forty five min at 37 °C, followed by alkylation of cysteines with 55 mM CAA (2-chloroacetamide) for 30 min at room temperature in 100 µl of digestion buffer (2 mM CaCl₂ in 40 mM Tris-HCl, pH 7.8). Trypsin (1:50 (wt/wt) enzyme-to-protein ratio) was added and proteins had been digested off the beads at 37 °C and 1,200 rpm overnight. For peptide recovery, the beads had been settled on magnets and the supernatant was transferred to a new 96-well plate. Beads had been washed by addition of 100 µl 2% formic acid in ddH₂O and the supernatant was transferred to the collection plate. As a result, the samples had been desalted as described below.

Desalting and drying of peptides

Earlier than LC–MS/MS diagnosis samples had been desalted using hydrophilic-lipophilic balanced (10 mg of N-vinylpyrrolidon-divinylbenzol porous particles 30 μm, Macherey-Nagel) 96-smartly plates using centrifugation at 7g for 1 min unless specified otherwise. For this, hydrophilic-lipophilic balanced discipline fabric became primed with 500 µl of isopropanol, ACN and solvent B (0.1% formic acid in 70% ACN in ddH₂O) and equilibrated with 1,000 µl of solvent A (0.1% formic acid in ddH₂O) sooner than sample loading (by gravitation, 5 min). The sample flowthrough became reapplied to the plate and sure peptides had been washed with 1,000 µl of solvent A. Peptides had been eluted with 250 µl of solvent B (3 min, 7g; 1 min, 172g). Samples had been frozen at −80 °C, dried by vacuum centrifugation and saved at −20 °C unless LC–MS/MS dimension.

High pH reversed-phase fractionation

Here, 50 µg of peptides (A. thaliana for Extended Data Fig. 1i and Jurkat for Fig. 3b and Extended Data Fig. 4d–e) had been fractionated by frequent pH reversed-phase solid-phase extraction (reversed-phase sulfonate cartridge tips; 5 μl of polystyrene-divinylbenzene (PS-DVB) resin, Agilent) into six fractions using the Agilent AssayMAP Bravo pipetting system. The reversed-phase sulfonate cartridges had been primed, washed and equilibrated in accordance with the manufacturer’s protocol. Peptides had been reconstituted in 100 μl of 25 mM ammonium formate (pH 10) and loaded onto the cartridges. Peptides had been fractionated by increasing ACN concentrations (5, 10, 15, 20, 25, 30, 80%). The seven elution steps had been either combined into six fractions, combining the 5 and 80% fractions, or into four fractions. For four fractions, the 5 and 25%, the 10 and 30%, the 15 and 80%, and the 20% ACN fractions and the flowthrough had been combined. All fractions had been acidified with formic acid to a final concentration of 1%. Samples had been frozen at −80 °C, dried by vacuum centrifugation and stored at −20 °C unless LC–MS/MS analysis.

Microflow-LC–(FAIMS)–MS/MS measurements

All samples (with the exception of where indicated otherwise) had been analyzed on a microflow-LC–MS/MS draw using a Vanquish Neo extremely excessive-efficiency LC draw (Thermo Fisher Scientific) coupled to an Orbitrap Eclipse Tribrid mass spectrometer (Thermo Fisher Scientific) with or without installed FAIMS Pro Interface (Thermo Fisher Scientific). For a elephantine listing of ancient instrument tool, behold Supplementary Desk 3Materials.

Prior to analysis, samples were reconstituted in 0.1% formic acid, 2% ACN. For run optimization, the peptide concentration was determined using a Nanodrop instrument (Thermo Fisher Scientific) and the amount of peptide required for each run was injected accordingly. For drug profiling samples, half of the samples were injected per run (50 µg). For fractionated samples, all material was injected.

Chromatographic separation became performed by technique of notify injection on a 15 cm Acclaim PepMap 100 C18 column (2 µm, 1 mm inner diameter × 15 cm, Thermo Fisher Scientific) at a plug with the hump rate of 50 µl min⁻¹. The column temperature became set apart to 55 °C. Solvent A became 0.1% formic acid in 3% DMSO in ddH₂O, and solvent B became 0.1% formic acid and 3% DMSO in ACN. The gradients for diversified lengths can even be found in Supplementary Desk 3LC gradients.

Incorporation of FAIMS into microflow-LC–MS/MS

Because micro-LC separations generate noteworthy sharper peaks than nano-LC, the incorporation of FAIMS into microflow-LC–MS/MS draw critical to be evaluated from the bottom up. We first characterized the tool for peptide transmission at diversified compensation voltage (CV) values using a tryptic digest. With these data in hand, we next simulated how many and which CV values wants to be mixed for wonderful proteome protection. Simulations had been experimentally examined using LC gradient lengths between 15 and 180 min and we systematically when put next efficiency with and without FAIMS. For gradient events of 15, 30 and 60 min, simplest one CV atmosphere can even be meaningfully ancient attributable to CV switching takes sizable amounts of time. Regardless of LC events, FAIMS elevated the amount of identified protein groups at a given time or halved the MS time critical to carry out the identical depth of diagnosis when put next to the identical LC set apart-up but without using FAIMS.

Dimension without FAIMS installed

The OptaMax NG ion source (Thermo Fisher Scientific) with a heated electrospray ionization probe became ancient to spoil the information. The sprayer became positioned at middle spot in the x axis (left to handsome), at spot 1 in the y axis (front to abet) and between positions M and L in the z axis (probe peak).

The mass spectrometer was operated in data-dependent MS/MS, positive ion mode, using a spray voltage of 3.5 kV, a funnel radio-frequency lens value of 40, an ion transfer tube temperature of 325 °C and vaporizer temperature of 125 °C. The flow rates for sheath gas, auxiliary gas and sweep gas were set to 32, 5 and 0 l min⁻¹ respectively.

A full-scan (MS1) was recorded from 360 to 1,300 m/z with a resolution of 120,000 in the Orbitrap in profile mode. The MS1 AGC target was custom set to 100% and the maxIT was set to 50 ms. Following the full scans, precursors were targeted for the MS/MS scans (MS2) if the isotope envelope was peptidic (monoisotopic precursor selection), the charge was between 2 and 6 and the intensity exceeded 1 × 10⁴. The MS2 quadrupole isolation window was set to 0.4 m/z. Peptide fragmentation occurred in the ion routing multipole by HCD with a fixed collision energy mode, the collision energy normalized to the precursor m/z and set with a collision energy of 28%. The MS2 scan was acquired in the Ion Trap with rapid scan rate in centroid mode and an defined first mass of 100 m/z. Additional MS2 properties as well as cycle events for different gradient lengths can be found in Supplementary Table 3MS settings.

Dimension with FAIMS installed

The identical ion source and probe as above were used, making use of the identical spray conditions. The mass spectrometer was operated in data-dependent MS/MS, positive ion mode, using a spray voltage of 4 kV, a funnel radio-frequency lens setting of 40, an ion transfer tube temperature of 325 °C and vaporizer temperature of 300 °C. The plug with the flow rates for sheath gas and auxiliary gas had been set to 40 and 5 l min⁻¹ respectively. FAIMS was operated with standard temperature (inner and outer electrode 100 °C) and a static carrier gas flow with the flow of 3.5 l min⁻¹. Dimension parameters were unchanged and the respective FAIMS CV was set to the critical value. For measurements of drug perturbed samples, the 60 min gradient was used with a series CV of −30 V.

If multiple internal CVs were used (for optimization), separate experiments were specified for the different CVs in the Tune method with the identical settings, except for the different CV value (the CV values used can both be read directly from the figures or the raw file names). This results in the MS looping through the specified experiments of the method, switching after every MS cycle (MS1 scan + MS2 scans). To ensure the data points and thus quantification quality were sufficient, the cycle time stated above was divided by the number of internal CVs used, leading to 0.75 s for 60 min (two CVs), 1.4 s for 120 and 180 min (two CVs) and 0.8 s for 120 and 180 min (three CVs).

Database hunting

The uncooked MS data information had been processed with MaxQuant v.1.6.2.10 (ref. ²⁷) using the integrated Andromeda search engine and searched in opposition to the respective reference database (H. wise: downloaded from UniProt containing canonical and isoforms 24 August 2020; 75,776 entries, E. coli: downloaded from UniProt containing canonical and isoforms 1 July 2021; 4,713 entries, P. aeruginosa: downloaded from UniProt containing canonical and isoforms 1 July 2021; 5,563 entries, M. musculus: downloaded from UniProt containing canonical and isoforms 1 July 2021; 25,381 entries, A. thaliana: Araport11 genome unlock downloaded from Arabidopsis.org containing canonical and isoforms 16 June 2020; forty eight,359 entries).

Raw data from runs with multiple internal FAIMS CVs had to be split into separate files based on CV values before MaxQuant searches. These separate files were specified as different fractions, as for the normal reverse-phase fractions, of the same experiment in MaxQuant. Multiple injections of the same sample were specified as the same experiment. Standard MaxQuant search parameters were used. Trypsin/P was specified as protease, allowing up to a maximum of two missed cleavages. Carbamidomethylation of cysteine was specified as a fixed modification, whereas oxidation of methionine and protein N-terminal acetylation were considered as variable modifications. Where specified, mono- and di-methylation of arginine and lysine were enabled as variable modifications. The label-free quantification (LFQ) algorithm, with a default LFQ minimum ratio count setting of 1, as well as the iBAQ (intensity-based absolute quantification) algorithm, with log fit, were enabled where required. Where used, the Match-Between-Runs algorithm was enabled with default settings (0.7 min and 5 min for matching and retention time alignment window, respectively). The false discovery rate (FDR) was set to 1% at the protein and peptide spectral match level. For Prosit rescoring, the FDR was set to 100% at the protein and peptide spectral match level. The respective MaxQuant msms .txt and .raw files were rescored by Prosit. Peptides with q values ≤0.01 were retained and proteins were grouped based on the selected FDR method⁴⁷. For MaxQuant output, proteins for which no unique peptide was found and thus were not distinguishable were aggregated into protein groups. For selected FDR protein group output, proteins are grouped at the gene level and only unique peptides are considered. For clarity, we refer to all such entities as proteins in the figures. Data analysis and visualization were performed using R (v.4.1.0) in RStudio (see Supplementary Table 3 for a complete list of all software used) and Microsoft Excel 365. Further refinement of plots was carried out in Adobe Illustrator CS6. Information on whether a dataset was rescored or not can be found on MassIVE (Data availability section).

Data processing and analysis

Curve becoming

For every protein–drug pair, the LFQ intensity relative to the median protein intensity in the DMSO controls was calculated for all drug concentrations. The same was done for every transcript–drug pair of the transcriptomic data using read counts. For the various viability metrics, the data were processed as described above. To these normalized data, a sigmoidal four-parameter log-logistic model (equation (1)) was fitted using the dose–response curve R package (v.3.0-1), where x is the log₁₀ of the drug concentration, pEC₅₀ is the negative log of the inflection point of the curve (denoted as the effective concentration 50; EC₅₀), t is the top or low-dose plateau, b is the bottom or high-dose plateau, s is the curve slope between the plateaus and Y(x) is the observed protein ratio relative to the DMSO control at concentration x.

For each resulting model, descriptive parameters were extracted and reported. These included the optimized slope (s), top (t), bottom (b) and inflection point (EC₅₀), as well as the area under the curve, the coefficient of determination (R²), mean absolute deviation, the predicted y value of the fitted curve at the highest concentration (end of curve, fold change) and the slope of a linear model fitted to the data.

Curve classification

To avoid manual annotation of more than 1 million dose–response curves, a random forest classifier was trained using the ranger R package (v.0.14.1). As a ground truth dataset, curves of two compounds had been manually annotated as up-, down- and nonregulated. The dataset was split into 80:20 for training and validation, respectively (training 11,562, validation 2,883, total 14,409). The input features were created from the values described above, along with the relative LFQ intensities and number of unique peptides for all concentrations and abundance percentile of the respective protein in the DMSO control. After hyperparameter tuning, the final model was trained with 1,200 trees, randomly selecting 15 independent variables at each split and splitting only nodes with a minimum size of 3. Variable importance mode was set to impurity and the Gini split rule was used. The model’s performance and quality were evaluated using the validation dataset, calculating precision, confusion matrices and ROC curves. The resulting classifier was used as a prefilter, plotting curves into separate PDFs and writing data into separate .txt files based on the predicted classes, thereby facilitating manual examination of all drug datasets. The same classifier was used for the dose–response curves of the drug perturbed transcriptome dataset. These regulated proteins were further analyzed to determine the mode of action of drugs.

Extra filtering

For further analysis, a protein was considered up- or down-regulated if it was classified as such and the fold change exceeded 1.5 for up-regulation and 0.7 for down-regulation. The same criteria were applied to all transcripts, retaining only observations where read counts were above 50 for all concentrations.

GO term enrichments

For the heatmap clustering of medication with identical effects, a GO term enrichment diagnosis became performed for every drug for my portion using the clusterProfiler R bundle (v.4.2.2.)^{forty eight}. Every drug dataset became examined for enrichment of GO terms on all ranges (cell compartment, molecular feature and natural course of) both in up- and down-regulated proteins with the entire drug dataset because the background. P values had been corrected using the FDR ability and the q ticket in the discount of-off became set apart to 1. The enrichment outcomes for up- and down-regulation had been mixed, preserving the more critical entry for duplications. After combining the enrichment outcomes for all tablets, the q values had been log transformed, multiplied by −1 for GO terms enriched in down-regulation and z-scored for every GO term for my portion. The heatmap depicts the mixed, preprocessed GO term enrichment outcomes after hierarchically clustering of both rows and columns using Pearson correlation as a distance metric and Weighted Pair Crew Plot with arithmetic point out because the agglomerative formula. The GO term enrichment outcomes displayed in Extended Data Fig. 6a had been taken from the realm GO term enrichment diagnosis described above. For Extended Data Fig. 5d a brand new GO term enrichment diagnosis became executed (P ticket in the discount of-off, 0.05; P ticket correction, FDR ability; Subontology, Molecular Feature; total H. wise database as background).

Dose-dependent methylation

The search results for lysine and arginine methylation were challenging for dose–response curve fitting, similar to the process described for proteins and transcripts above. However, for each peptide–concentration–inhibitor combination, the intensity ratio of methylated to unmethylated form was calculated. The resulting value was then normalized to the respective DMSO control before proceeding as described above (section ‘Curve fitting’).

Simulation of target coverage relative to proteomic depth

For the simulation of target coverage across captured proteomic depth, we ranked all >8000 proteins in this study by their iBAQ values in all DMSO controls in descending order. To simulate varying proteomic depths, this list was truncated at the indicated ranks (number of identified proteins). For each drug, we then assessed how many of its targets were included in the resulting list and calculated the fraction of targeted proteins that were detected.

Replicate diagnosis

For the volcano plot displayed in Extended Data Fig. 2b assessing the quantitative reproducibility, the forty-eight DMSO controls had been randomly assigned into two equally sized groups. After median centering normalization of the LFQ intensities of the picked FDR gene neighborhood output and filtering for completeness in the dataset, a two-sided Student’s t-test was performed for all 4,694 proteins. P values had been corrected for multiple hypothesis testing using the FDR method using the R package fdrtool (v.1.2.17).

For the comparison of quantitative reproducibility between unregulated and regulated proteins using the 5 individual doses for every inhibitor as replicates, the LFQ intensities of the picked FDR gene neighborhood output had been normalized by median centering. The CoV was calculated across the 5 doses for every drug for every protein that was both classified as up- or down-regulated, or unregulated.

To evaluate the reproducibility of EC₅₀ determinations, the curves for each protein for each drug replicate had been fitted as described above. For proteins being classified as up- or down-regulated in three out of four replicates per drug, the standard deviation of the pEC₅₀s was calculated.

True-time RT–qPCR

For RT–qPCR diagnosis, cells were treated according to the protocol described above. After 18 h, cells were lysed, and total RNA was isolated using the Monarch Total RNA Miniprep Kit (New England Biolabs) according to the manufacturer’s instructions. RNA yield was determined using the Qubit fluorometer (Thermo Fisher Scientific). Complementary DNA (cDNA) was generated from 2 µg of RNA from each sample using the LunaScript RT SuperMix Kit (New England Biolabs) according to the manufacturer’s protocol. Additionally, no-reverse transcriptase controls were generated for each sample by omitting the reverse transcription step. After reverse transcription, the cDNA was diluted ~66-fold with nuclease-free ddH₂O. qPCR was performed in triplicates on a CFX384 Touch Real-Time PCR Detection System (Bio-Rad Laboratories, Inc.) using 10 ng of cDNA per sample, the Luna Universal qPCR Master Mix (New England Biolabs) and the primer pairs as shown in Supplementary Table 3Primer listing. No-reverse transcriptase controls were measured in pools of all samples on each plate. Nuclease-free ddH₂O was used as the no-template control for each assay. Cycling parameters were set to 95 °C (1 min), 40 cycles of 95 °C (15 s) and 60 °C (30 s with plate read on SYBR channel) each, and finally a melt curve was recorded from 60 to 95 °C with an increment of 0.5 °C per 5 s and SYBR channel plate reads after each increment. All samples treated with the same drug in addition to the DMSO control were measured on the same plate.

Analysis of RT–qPCR outcomes

Quantification cycle (C_q) and melting temperature (T_m) values were determined using the CFX Manager v.3.1 tool (Bio-Rad Laboratories, Inc.). The regression formula of the tool was used for C_q evaluation with baseline correction and curve fit enabled. The fold change in expression after treatment and the ratio of truncated to full-length transcript were calculated in Microsoft Excel 365 from the obtained C_q values for each sample using the 2^-∆∆Cq formula⁴⁹.

T cell activation assay

Activation status of HDACi-treated Jurkat cells was analyzed using TCR and/or CD3 effector cells (nuclear factor of activated T cells or NFAT) from a T Cell Activation Bioassay (Promega) with minor adaptations of the manufacturer’s protocol. Briefly, TCR/CD3 effector cells (NFAT) were incubated with HDACi (5 doses for each drug: 10 µM, 1 µM, 100 nM, 10 nM and 1 nM) for 16 h, followed by nonspecific activation via CD3 and/or CD28 using the Human Anti-CD3/CD28 T Cell Activation Kit (Cell Signaling Technology). After 5 h, receptor-mediated signaling was measured by luminescence using a microplate reader FluoStar Omega (BMG Labtech). Thus, the intensity of the luminescence signal corresponded to the intensity of receptor-mediated signaling. To assess the level of T cell activation, the luminescence signals were normalized to the DMSO control. Dose–response curves were fitted to the data as described in the section ‘Curve fitting’.

T cell aggregation diagnosis

The usage of the intense gentle photos of residing activated human T cells, bought using the IncuCyte live-cell diagnosis draw as described above, cell aggregates had been assigned and quantified (rely and situation in µm²). To this end photos had been processed by ilastik⁵⁰a supervised machine studying image diagnosis tool bundle. The frequent aggregate dimension became calculated for every image by summing up the detected aggregate areas and dividing by the rely of aggregates per image, treating the 5 photos bought per smartly as replicates. To evaluate statistical significance of the HDACi prompted discount of moderate aggregate dimension, an diagnosis of variance take a look at became performed for every inhibitor for my portion, followed by a Tukey handsome critical differ ences put up hoc take a look at.

Reporting abstract

Additional information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

1. Singh, S., Malik, B. K. & Sharma, D. K. Molecular drug targets and structure-based drug design: a holistic approach. Bioinformation 1314–320 (2006).

   Article PubMed PubMed Central Google Scholar

2. Meissner, F., Geddes-McAlister, J., Mann, M. & Bantscheff, M. The rising feature of mass spectrometry-essentially essentially based proteomics in drug discovery. Nat. Rev. Drug Discov. 21637–654 (2022).

   Article CAS PubMed Google Scholar

3. Geoffrey, M. C. Pharmacology, part 1: introduction to pharmacology and pharmacodynamics. J. Nucl. Med. Technol. 4681 (2018).

   Article Google Scholar

4. Swinney, D. C. Biochemical mechanisms of drug action: what does it take for success? Nat. Rev. Drug Discov. 3801–808 (2004).

   Article CAS PubMed Google Scholar

5. Tonge, P. J. Drug-target kinetics in drug discovery. ACS Chem. Neurosci. 929–39 (2018).

   Article CAS PubMed Google Scholar

6. Backus, K. M. et al. Proteome-wide covalent ligand discovery in native biological systems. Nature 534570–574 (2016).

   Article CAS PubMed PubMed Central Google Scholar

7. Bantscheff, M. et al. Quantitative chemical proteomics uncovers mechanisms of action of clinical ABL kinase inhibitors. Nat. Biotechnol. 251035–1044 (2007).

   Article CAS PubMed Google Scholar

8. Liu, Y., Patricelli, M. P. & Cravatt, B. F. Process-based protein profiling: the serine hydrolases. Proc. Natl Acad. Sci. USA 9614694–14699 (1999).

   Article CAS PubMed PubMed Central Google Scholar

9. Patricelli, M. P. et al. Purposeful interrogation of the kinome using nucleotide acyl phosphates. Biochemistry 46350–358 (2007).

   Article CAS PubMed Google Scholar

10. Klaeger, S. et al. The aim panorama of clinical kinase tablets. Science 358eaan4368 (2017).

    Article PubMed PubMed Central Google Scholar

11. Sharma, K. et al. Proteomics technique for quantitative protein interaction profiling in cell extracts. Nat. Methods 6741–744 (2009).

    Article CAS PubMed Google Scholar

12. Gaetani, M. et al. Proteome integral solubility alteration: a high-throughput proteomics assay for target deconvolution. J. Proteome Res. 184027–4037 (2019).

    Article CAS PubMed Google Scholar

13. Savitski, M. M. et al. Monitoring cancer drugs in living cells by thermal profiling of the proteome. Science 3461255784 (2014).

    Article PubMed Google Scholar

14. Van Vranken, J. G., Li, J., Mitchell, D. C., Navarrete-Perea, J. & Gygi, S. P. Assessing target engagement using proteome-wide solvent shift assays. eLife 10e70784 (2021).

    Article PubMed PubMed Central Google Scholar

15. Zhang, X. et al. Solvent-prompted protein precipitation for drug target discovery on the proteomic scale. Anal. Chem. 921363–1371 (2020).

    Article CAS PubMed Google Scholar

16. Strickland, E. C. et al. Thermodynamic diagnosis of protein-ligand binding interactions in complex natural mixtures using the sensitivity of proteins to oxidation rates. Nat. Protoc. 8148–161 (2013).

    Article CAS PubMed Google Scholar

17. Lomenick, B. et al. Target identification using drug affinity responsive target stability (DARTS). Proc. Natl Acad. Sci. USA 10621984–21989 (2009).

    Article CAS PubMed PubMed Central Google Scholar

18. Schopper, S. et al. Measuring protein structural modifications on a proteome-wide scale using limited proteolysis-coupled mass spectrometry. Nat. Protoc. 122391–2410 (2017).

    Article CAS PubMed Google Scholar

19. Zecha, J. et al. Decrypting drug actions and protein modifications by dose- and time-resolved proteomics. Science 38093–101 (2023).

    Article CAS PubMed PubMed Central Google Scholar

20. Pan, C., Olsen, J. V., Daub, H. & Mann, M. Global effects of kinase inhibitors on signaling networks revealed by quantitative phosphoproteomics. Mol. Cell Proteom. 82796–2808 (2009).

    Article CAS Google Scholar

21. Subramanian, A. et al. A next generation connectivity map: L1000 platform and the first 1,000,000 profiles. Cell 1711437–1452 e1417 (2017).

    Article CAS PubMed PubMed Central Google Scholar

22. Mitchell, D. C. et al. A proteome-wide atlas of drug mechanism of action. Nat. Biotechnol. 41845–857 (2023).

    Article CAS PubMed Google Scholar

23. Ruprecht, B. et al. A mass spectrometry-based proteome map of drug movement in lung cancer cell lines. Nat. Chem. Biol. 161111–1119 (2020).

    Article CAS PubMed Google Scholar

24. Saei, A. A. et al. ProTargetMiner as a proteome signature library of anticancer molecules for purposeful discovery. Nat. Common. 105715 (2019).

    Article CAS PubMed PubMed Central Google Scholar

25. Hughes, C. S. et al. Single-pot, sturdy-part-enhanced sample preparation for proteomics experiments. Nat. Protoc. 1468–85 (2019).

    Article CAS PubMed Google Scholar

26. Bian, Y. et al. Sturdy, reproducible and quantitative diagnosis of thousands of proteomes by micro-plug with the hump LC-MS/MS. Nat. Common. 11157 (2020).

    Article CAS PubMed PubMed Central Google Scholar

27. Cox, J. & Mann, M. MaxQuant permits high peptide identification rates, individualized ppb-level mass accuracies and proteome-wide protein quantification. Nat. Biotechnol. 261367–1372 (2008).

    Article CAS PubMed Google Scholar

28. Gessulat, S. et al. Prosit: proteome-wide prediction of peptide tandem mass spectra by deep learning. Nat. Methods 16509–518 (2019).

    Article CAS PubMed Google Scholar

29. Wilhelm, M. et al. Mass-spectrometry-based draft of the human proteome. Nature 509582–587 (2014).

    Article CAS PubMed Google Scholar

30. Datta, J., Ghoshal, K., Motiwala, T. & Jacob, S. T. Unusual insights into the molecular mechanism of movement of DNA hypomethylating agents: feature of protein kinase C δ in decitabine-prompted degradation of DNA methyltransferase 1. Genes Cancer 371–81 (2012).

    Article PubMed PubMed Central Google Scholar

31. Ercikan-Abali, E. A. et al. Dihydrofolate reductase protein inhibits its own translation by binding to dihydrofolate reductase mRNA sequences within the coding space. Biochemistry 3612317–12322 (1997).

    Article CAS PubMed Google Scholar

32. Uzor, S. et al. Autoregulation of the human splice factor kinase CLK1 by exon skipping and intron retention. Gene 67046–54 (2018).

    Article CAS PubMed Google Scholar

33. Hopkins, A. L. Community pharmacology: the next paradigm in drug discovery. Nat. Chem. Biol. 4682–690 (2008).

    Article CAS PubMed Google Scholar

34. Bruno, P. M. et al. A subset of platinum-containing chemotherapeutic agents kills cells by inducing ribosome biogenesis stress. Night. With. 23461–471 (2017).

    Article CAS PubMed PubMed Central Google Scholar

35. Demichev, V. et al. dia-PASEF data analysis using FragPipe and DIA-NN for deep proteomics of low sample amounts. Nat. Commun. 133944 (2022).

    Article CAS PubMed PubMed Central Google Scholar

36. Schwanhausser, B. et al. Global quantification of mammalian gene expression adjustment. Nature 473337–342 (2011).

    Article PubMed Google Scholar

37. Zecha, J. et al. Peptide level turnover measurements enable the study of proteoform dynamics. Mol. Cell Proteom. 17974–992 (2018).

    Article CAS Google Scholar

38. Sievers, Q. L. et al. Defining the human C₂H₂ zinc finger degron targeted by thalidomide analogs by means of CRBN. Science 362eaat0572 (2018).

    Article PubMed PubMed Central Google Scholar

39. Donovan, K. A. et al. Thalidomide promotes degradation of SALL4, a transcription factor implicated in Duane Radial Ray syndrome. eLife 7e38430 (2018).

    Article PubMed PubMed Central Google Scholar

40. Hamminger, P., Rica, R. & Ellmeier, W. Histone deacetylases as targets in autoimmune and autoinflammatory diseases. Adv. Immunol. 1471–59 (2020).

    CAS PubMed Google Scholar

41. Borcoman, E. et al. HDAC inhibition to prime immune checkpoint inhibitors. Cancers 1466 (2021).

    Article PubMed PubMed Central Google Scholar

42. Jiang, W. et al. Exhausted CD8⁺ T cells in the tumor immune microenvironment: new pathways to remedy. Entrance. Immunol. 11622509 (2021).

    Article PubMed PubMed Central Google Scholar

43. Mueller, S. N. & Ahmed, R. High antigen ranges are the set apart off of T cell exhaustion correct by technique of persistent viral an infection. Proc. Natl Acad. Know USA 1068623–8628 (2009).

    Article CAS PubMed PubMed Central Google Scholar

44. Utzschneider, D. T. et al. Early precursor T cells establish and propagate T cell exhaustion in chronic infection. Nat. Immunol. 211256–1266 (2020).

    Article CAS PubMed Google Scholar

45. Wherry, E. J. T cell exhaustion. Nat. Immunol. 12492–499 (2011).

    Article CAS PubMed Google Scholar

46. Eckert, S. et al. Overview of disposable trap column nanoLC-FAIMS-MS/MS for the proteomic diagnosis of FFPE tissue. J. Proteome Res. 205402–5411 (2021).

    Article CAS PubMed PubMed Central Google Scholar

47. The, M., Samaras, P., Kuster, B. & Wilhelm, M. Reanalysis of ProteomicsDB using an accurate, sensitive, and scalable false discovery rate estimation approach for protein groups. Mol. Cell. Proteom. 21100437 (2022).

    Article CAS Google Scholar

48. Yu, G., Wang, L. G., Han, Y. & He, Q. Y. clusterProfiler: an R package for comparing functional themes among gene clusters. Omics 16284–287 (2012).

    Article CAS PubMed PubMed Central Google Scholar

49. Livak, K. J. & Schmittgen, T. D. Analysis of relative gene expression data using real-time quantitative PCR and the 2^−ΔΔCT formula. Methods 25402–408 (2001).

    Article CAS PubMed Google Scholar

50. Berg, S. et al. ilastik: interactive machine studying for (bio)image diagnosis. Nat. Systems 161226–1232 (2019).

    Article CAS PubMed Google Scholar

51. Lechner, S. et al. Target deconvolution of HDAC pharmacopoeia unearths MBLAC2 as frequent off-target. Nat. Chem. Biol. 18812–820 (2022).

    Article CAS PubMed PubMed Central Google Scholar

52. Prokofeva, P. et al. Merits of diazirine photo-immobilization for target profiling of natural products and cofactors. ACS Chem. Biol. 173100–3109 (2022).

    Article CAS PubMed PubMed Central Google Scholar

53. Reinecke, M. et al. Chemical proteomics reveals the landscape of 1,000 kinase inhibitors. Nat. Chem. Biol. https://doi.org/10.1038/s41589-023-01459-3 (2023).

Download references

Creator notes

1. These authors contributed equally: Stephan Eckert, Nicola Berner.

Authors and Aff iliations

1. Chair of Proteomics and Bioanalytics, College of Life Sciences, Technical University of Munich, Freising, Germany

   Stephan Eckert, Nicola Berner, Karl Kramer, Annika Schneider, Julian Müller, Severin Lechner, Sarah Brajkovic, Amirhossein Sakhteman, Stephanie Wilhelm & Bernhard Kuster

2. German Cancer Consortium (DKTK), accomplice situation Munich, a partnership between DKFZ and College Heart Technical College of Munich, Munich, Germany

   Stephan Eckert, Nicola Berner & Bernhard Kuster

3. German Cancer Evaluate Heart (DKFZ), Heidelberg, Germany

   Stephan Eckert & Nicola Berner

4. Chair of Animal Physiology and Immunology, College of Life Sciences, Technical College of Munich, Freising, Germany

   Christian Graetz & Michael W. Pfaffl

5. Institute of Molecular Immunology and Experimental Oncology, College of Treatment and Health, Technical College of Munich, Munich, Germany

   Jonas Fackler, Michael Dudek & Percy Knolle

Contributions

S.E., N.B. and B.K. conceptualized the study. S.E. and N.B. developed the methodology. S.E., N.B., K.K., A. Schneider, S.B., C.G. and J.F. performed experiments. S.E. and N.B. curated the data. S.E., N.B. and B.K. wrote the paper with input from all authors. S.E. developed the tool.

Corresponding creator

Correspondence to Bernhard Kuster.

Competing pursuits

B.K. is founder and shareholder of OmicScouts and MSAID. He has no operational role in either firm. All other authors declare no competing interests.

Scrutinize overview data

Nature Biotechnology thanks the anonymous reviewers for their contribution to the peer review of this work.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions