Chemoproteomic mapping of the glycolytic targetome in cancer cells

Hyperactivated glycolysis is a metabolic hallmark of most cancer cells. Although sporadic information has revealed that glycolytic metabolites possess nonmetabolic functions as signaling molecules, how these metabolites interact with and functionally regulate their binding targets remains largely elusive. Here, we introduce a target-responsive accessibility profiling (TRAP) approach that measures changes in ligand binding-induced accessibility for target identification by globally labeling reactive proteinaceous lysines. With TRAP, we mapped 913 responsive target candidates and 2,487 interactions for 10 major glycolytic metabolites in a model cancer cell line. The wide targetome depicted by TRAP unveils diverse regulatory modalities of glycolytic metabolites, and these modalities involve direct perturbation of enzymes in carbohydrate metabolism, intervention of an orphan transcriptional protein’s activity and modulation of targetome-level acetylation. These results further our knowledge of how glycolysis orchestrates signaling pathways in cancer cells to support their survival, and inspire exploitation of the glycolytic targetome for cancer therapy. A chemoproteomic approach is developed that examines changes in ligand binding-induced accessibility by globally labeling reactive proteinaceous lysines, revealing the cellular targets of glycolytic intermediates.


Article
https://doi.org/10.1038/s41589-023-01355-w to CDP, in line with their different affinities.This model demonstrated that TRAP allows identification of ligand-binding sites.
Nevertheless, ligand engagement can also induce allostery, which drove us to test whether TRAP can detect ligand binding-induced conformational changes in targets.Thus, we incubated purified PKM2, a multidomain enzyme that metabolizes phosphoenolpyruvate (PEP) to pyruvate (Pyr), with its allosteric activator fructose 1,6-bisphosphate (FBP) 29 (Fig. 1b and Extended Data Fig. 1f).We found that the TRP carrying labeled K433, the exact FBP-binding site, indeed conferred the greatest decrease in accessibility upon FBP incubation (R high-FBP/control = 0.07, Fig. 1c).Meanwhile, TRPs carrying labeled K270 and K337 lysines located close to the PKM2 active site also showed markedly decreased accessibility (Fig. 1b,c and Extended Data Fig. 1g).Such a decrease agrees with the partially closed conformation of the active site following FBP activation 29 .Simultaneously, intersubunit contact is promoted and thus explains the decreased accessibility at K422 (R high-FBP/control = 0.20), which is located near the C-C′ subunit interface (Fig. 1b and Extended Data Fig. 1f).These results suggested that TRAP can map ligand-induced allostery.Notably, these TRPs displayed pronounced accessibility changes with high-dose FBP relative to low-dose FBP (Fig. 1c), prompting dose-responsive TRAP analysis for further study.

Benchmarking TRAP for target identification from cell milieu
Next, the applicability of TRAP to perform target identification with a complex cell milieu should be leveraged.We chose a model ligand, TEPP-46, an anticancer reagent known to specifically bind to and activate PKM2 (ref.30) (Extended Data Fig. 2a).We first prepared HCT116 cell lysates in nondenaturing lysis buffer with and without TEPP-46 and then separately labeled the proteome with TRAP labeling.Next, the labeled samples were processed via label-free quantification (LFQ) proteomics to survey the intensity changes of labeled lysine-carrying peptides and determine TRPs.Remarkably, lysines of the bona fide target PKM2 were mostly accessible to TRAP labeling reagents (Extended Data Fig. 2b).Subsequent quantitative analysis showed that the TRP harboring the labeled PKM2-K305 was markedly downregulated by TEPP-46 incubation, implying greatly reduced accessibility at this residue (R TEPP-46/control = 0.08; Fig. 1d).Given that TEPP-46 engages with PKM2 in proximity to K305 (Extended Data Fig. 2a), this model prompted us to believe that TRAP is competent in identifying target and sites of binding of the ligand in a complex cell milieu.Consistently, the same TRP was recognized by TRAP with recombinant PKM2 (Extended Data Fig. 2c,d) and even across different biological models, including A549 lung carcinoma cells and Escherichia coli (Fig. 1d and Extended Data Fig. 2e); thus, TRAP exhibits wide applicability toward diverse species.
Then, a multiplexed-TRAP workflow was developed to further increase the throughput of target identification (Fig. 1e).Multiplexed-TRAP involves the division of cell lysates into aliquots, followed by incubation with ligands of interest and the matching solvent.After protein-level TRAP labeling, proteins were analyzed following a multiplexed proteomics workflow using six-plex tandem mass tags (TMTs) (Fig. 1e).Owing to the improved proteome coverage conferred by multiplexed proteomics, we noted that multiple TRPs were detected for the same target (Fig. 1f).Hence, we developed a composite TRAP score that considered both the TRAP ratio and the significance (score = −log 10 (P value) × abs[log 2 (R treated/control )]), by which multiple TRPs can be ranked and prioritized to represent the accessibility change of an individual protein.Furthermore, to profile lysine accessibility as widely as possible, we expanded the criteria of TRP candidates from only considering labeled lysine-bearing peptides to lysine-relevant peptides.Specifically, quantified peptides that fit into the three categories (Fig. 1f), including type A peptides containing TRAP-labeled lysines that are not located at the N/C terminus, type B peptides bearing unmodified lysines in the C terminus and type C peptides produced following tryptic cleavage at unmodified lysines, were considered lysine relevant and were used to leverage In addition to causing stability shifts, upon binding to its target protein, a ligand can also change the accessibility of the proteinaceous residues of the target to covalent labeling reagents, due to direct engagement-created steric hindrance or binding-induced allostery.Thus, identifying and quantifying peptides bearing covalently labeled residues via high-throughput proteomics analysis has become a stalwart in the arsenal of structural mass spectrometry (MS) tools to uncover l ig an d-bi nding sites and to profile binding-induced c o n fo r m at ional c h a n g e s 24 .A mo ng t h e v e r sa tile l a b el ing m e t ho ds, exemplified by amino acid-specific labeling 24 and indiscriminate hydroxyl radical footprinting 25 , lysine dimethylation is highly efficient and has been successfully used to profile changes in lysine accessibility due to ligand engagement for purified proteins 26 ; however, the method is rarely used for biologically complex systems.Recently, this method was extended to recognize protein misfolding in complex brain lysates and living cells by covalent protein painting 27 .Here, we first exploited lysine dimethylation toward target discovery of endogenous metabolites in a complex cell milieu.Specifically, we developed a TRAP approach that probes proteome-level accessibility changes following ligand engagement via lysine dimethylation and then assigns the responsive proteins as potential binding targets of the assayed ligands.We first benchmarked TRAP by corroborating its capability for identifying the bona fide targets for model ligands and demonstrating its complementarity to extensively used stability-based target-discovery approaches.Then, we applied TRAP to map the accessibility-responsive targetome for glycolytic metabolites, and uncovered 913 target candidates and 2,487 possible metabolite-target interactions in HCT116 human colorectal cancer cells.Through validating the deciphered glycolytic targetome, we found that the assayed glycolytic metabolites contribute to regulating metabolism, influencing transcriptional protein activity, modulating posttranslational modification (PTM) status and tuning pharmacological responses through direct target engagement.

TRAP detects ligand binding-induced accessibility changes
TRAP was established on the premise that, due to ligand binding and the concomitantly increased steric hindrance or induced allostery, reactive lysines of the target protein show pronounced accessibility changes to covalent labeling reagents.The simple, fast, reductive dimethylation labeling method involving the use of isotope-coded formaldehyde (CD 2 O) and borane-pyridine complex (BPC) is used for TRAP labeling and confers a mass shift of +32.06 Da to the lysine residue; this avoids interference from endogenous lysine dimethylation.Then, proteome-wide accessibility changes induced by ligand engagement were leveraged by quantifying labeled lysine-containing peptide with and without ligand incubation, which is defined as the TRAP ratio (R treated/control ).Peptides displaying marked changes in R treated/control are assigned as target-responsive peptides (TRPs) and concomitantly designate the responsive proteins as target candidates (Fig. 1a).
First, we set out to benchmark the TRAP approach using purified ribonuclease A (RNase; Extended Data Fig. 1a) with its ligands (cytidine diphosphate (CDP) and cytidine triphosphate (CTP)), and a stronger affinity of CTP over CDP was revealed by native MS and microscale thermophoresis (Extended Data Fig. 1b,c).Then, TRAP labeling was conducted with the unbound, CDP-and CTP-incubated RNase.Time-resolved MS measurement confirmed that TRAP labeling was impeded upon ligand engagement (Extended Data Fig. 1d).Subsequent bottom-up analysis pinpointed the peptide bearing labeled K41, which was located in proximity to the ligand-binding pocket 28 , as a sensitive TRP, since its abundance notably decreased upon CDP/CTP binding, suggesting the marked decreased accessibility at K41 (Extended Data Fig. 1e).In contrast, peptides bearing labeled K91 and K104 lysines located remotely from the binding pocket exhibited constant intensities and, hence, R treated/control (Extended Data Fig. 1e).Notably, the K41-bearing TRP responded in a more pronounced manner to CTP than were assigned as potential targets and are highlighted in red.Each protein is represented by a single data point, corresponding to the peptide with the greatest TRAP score.e, Multiplexed-TRAP analysis using six-plex TMT reagents.f, Schematic classification of lysine-relevant peptides.g, TMT-based MS 3 quantitative proteomics analysis identified a TRAP-labeled K305-bearing TRP of PKM2 for TEPP-46 (1 µM).h, Volcano plot of the multiplexed-TRAP results identifying the targets of TEPP-46 (1 µM) in HCT116 cell lysates (n = 3 biologically independent samples).Proteins carrying TRPs (R TEPP-46/control > 2 or <0.5, P < 0.001 by unpaired two-sided Student's t-test) were assigned as potential targets and are highlighted in red.i, Dose-responsive TRAP analysis identified the TRP type A and TRP type B , which both suggested dose-dependently changed accessibility at PKM2-K305 following TEPP-46 incubation.TMT reagents were used to map accessibility for aliquoted HCT116 cell lysates incubated with serial concentrations of TEPP-46.The experiment was conducted once.

Article
https://doi.org/10.1038/s41589-023-01355-w the increased TRP type B GDLGIEIPAEK (R treated/control = 2.39, score = 6.28), which is the unmodified counterpart of the former.The TRP type B was chosen to represent the responsiveness of PKM2 to TEPP-46 in the volcano plot (Fig. 1h).
In addition to the bona fide target PKM2, we assigned a few potential targets; thus, we sought to assess the likelihood of identifying false positives when TRAP was conducted using a single concentration of TEPP-46.A dose-responsive TRAP experiment was devised, which involved incubating aliquoted lysates with serial concentrations of TEPP-46, and employing multiplexing reagents to quantify lysine-relevant peptides and examine proteome-wide accessibility changes (Fig. 1i).As such, we found that both TRPs reflecting accessibility at PKM2-K305 showed dose-responsive changes (Fig. 1i), whereas the TRPs enabling the assignment of other target candidates in the single-dose TRAP experiment (Fig. 1h), but without dose-responsive changes (Extended Data Fig. 2f), were removed as false positives.

TRAP complements classic target-discovery approaches
Since TRAP identifies targets on the basis of accessibility changes, we queried whether it complements thermal stability-based target-discovery approaches, such as thermal proteome profiling (TPP) 18 and cellular thermal shift assay (CETSA) 19 , owing to their distinct working mechanisms.Interestingly, analysis of the melting behaviors of the human proteome using HEK293T cells (identifier PXD011929; ref. 31)  revealed the existence of proteins highly resistant to heat denaturation (defined as nonmelters) (Fig. 2a and Extended Data Fig. 3a,b).Through categorizing the nonmelters on the basis of their biological process (BP) terms of gene ontology (GO), their enrichment in proteasomal proteins was revealed (Fig. 2b and Extended Data Fig. 3c).This subset includes essential components of the 20S proteasome: proteasome ɑ and β subunits, with the β1 subunit (PSMB1) being the most thermally stable (Fig. 2c).Analysis of their melting behaviors in additional human cell lines (exemplified by PSMB1; Fig. 2d) corroborated their thermally stable nature.We speculate that these nonmelters, which are not amenable to thermal stability-based approaches, may instead be applicable to TRAP.To test this speculation, we incubated HCT116 cell lysate with bortezomib 32 , a known ligand of PSMB1 (Fig. 2e), and performed TRAP analysis.Excitingly, a TRP type A bearing labeled PSMB1-K164 was identified for bortezomib (Fig. 2f), in accord with the crystallographic structure illustrating their proximity (Fig. 2e).Since a few other protein target candidates were also assigned, we conducted dose-responsive TRAP analysis to examine false positives.We first confirmed that the labeled TRP type A bearing PSMB1-K164 dose-dependently decreased following bortezomib incubation (Fig. 2g), and the paired TRP type C generated by cleavage at the C terminus of K164 dose-dependently increased (Extended Data Fig. 3d); both results suggested blocked access of K164 to covalent labeling, which is in line with the bortezomib-binding topography (Fig. 2e).In comparison, TRPs of other target candidates assigned by single-dose TRAP failed to show dose responsiveness and were removed as false positives (Extended Data Fig. 3e).
In addition to TPP and CETSA, limited proteolysis-small-molecule mapping (LiP-SMap) is another widely used, derivatization-free chemoproteomic approach enabling target discovery 6 .It globally maps changed susceptibility of proteins to a LiP step using nonspecific proteinase K (PK) upon ligand engagement, followed by complete digestion with trypsin and ensuing quantitative proteomics analysis.Changed susceptibility is assessed by comparing abundance ratios of fully tryptic (FT) and half tryptic (HT) peptides for individual proteins with and without ligand incubation; then, proteins with apparent changes are designated as target 6 .As such, proteins that exhibit high resistance to the PK-mediated LiP step, manifested by the lack of HT peptides, do not fit for the LiP method.To investigate whether such proteins (LiP-resisters) exist, we analyzed four public LiP-SMap datasets (identifier PXD015446; ref. 33) and calculated the fraction of FT peptides against all identified FT and HT peptides for each protein (FT%) to estimate the resistance of an individual protein to LiP.Overall, FT% distribution plots showed low to medium levels (Extended Data Fig. 4a).Nevertheless, we also noticed proteins that generated mostly FT peptides and hence displayed high FT%; this subset was designated as LiP-resisters (Supplementary Dataset 1).A representative LiP-resister is ATP1A1, the catalytic α subunit of a classic drug target Na + /K + -ATPase (NKA), which actively exchanges Na + /K + ions across cell membranes 34 (Extended Data Fig. 4b).Since ATP1A1 is known to be tightly bound and inhibited by cadiotonic steroids, such as digoxin 35 , we conducted TRAP analysis using ATP1A1-present HCT116 cell lysates with digoxin.Markedly increased labeled K91-bearing TRP type A was identified from ATP1A1, designating this LiP-resister as a TRAP-responsive target (Extended Data Fig. 4c).Cryo-electron microscopy suggested that this TRP links the cytosolic headpiece with the transmembrane M1 helix and is crucial for gating ion transport 34 , thereby explaining its responsiveness to digoxin treatment.Notably, this TRP was not covered in any of the reanalyzed LiP-proteome datasets 33 (Extended Data Fig. 4d).Together, our results reiterated the complementarity of TRAP to current target-discovery approaches.

TRAP mapped the glycolytic targetome in human cancer cells
Upon benchmarking the TRAP approach, we set out to map the glycolytic targetome.TRAP analysis of ten glycolytic metabolites (Fig. 3a) was carried out in two batches of multiplexed-TRAP experiments with HCT116 colorectal cancer cells (Fig. 3b).Each glycolytic metabolite was assayed at a single concentration in this study, and the administered concentrations were tailored to match their intracellular concentrations.As a result, a total of 160,459 peptides mapping to 6,926 proteins were identified, and 159,823 peptides belonging to 6,913 proteins were quantified.We assessed the coverage of lysine-specific TRAP labeling and noted that ~57.42% of lysine residues carried the designated modification (Fig. 3c and Extended Data Fig. 5a).Consistently, the labeled fraction of the quantified protein reached ~57.34% (Extended Data Fig. 5b,c).In addition to the relatively wide coverage, a distribution plot of the labeled lysines further showed that TRAP labeling was not preferential toward any secondary structure elements (Fig. 3d and Extended Data Fig. 5d).
Overall, we tentatively identified 913 proteins as glycolytic metabolite target candidates and 2,487 potential interactions from HCT116 cells (Fig. 3e, Extended Data Fig. 5e and Supplementary Dataset 2).A total of 128 previously known metabolite-target interactions for Homo sapiens documented in BRENDA 36 were quantified in this study, and TRAP identified 29 known interactions involving 16 targets (Supplementary Dataset 3).This delivered a true positive rate of 22.66%, comparable to that from a previous stability-based approach 6 .This is expected to be improved by overcoming proteome undersampling, since the false negatives showed significantly lower sequence coverage than the true positives (Extended Data Fig. 5f,g).
Then, we explored the functional ramifications of the TRAP-assigned TRPs and found that the responsive lysines in TRPs are more conserved across organisms than all the quantified lysines, indicating their potential functionality (Fig. 3f).Since molecular function (MF) analysis of the TRAP-assigned targetome showed the enrichment in enzymes (MF: catalytic activity) (Extended Data Fig. 6a), we then focused on analyzing lysines in the TRPs of these enzyme targets.We found that, compared to lysines in all quantified peptides, these residues are consistently closer to the functional sites of given enzymes (Fig. 3g), reiterating their functionality.Considering that multiple glycolytic metabolites are well-recognized enzymatic ligands or substrates 6,36 , we asked whether the assayed glycolytic metabolites can regulate the activities of the TRAP-assigned enzyme targets (Fig. 3h).

Regulation of enzymes in carbohydrate metabolism
Since the TRAP-assigned enzyme targets were further grouped to the carbohydrate metabolism pathway at the highest frequency via KEGG Article https://doi.org/10.1038/s41589-023-01355-wpathway annotation network analysis (Fig. 4a,b, Extended Data Fig. 6b and Supplementary Dataset 4), we focused on these carbohydrate metabolism enzymes (Fig. 4b) and set out to examine whether the TRAP-revealed glycolytic metabolite engagement tended to influence their active sites and activities.We first defined the boundary of an active site detectable by TRAP as 5.69 Å, on the basis of the Euclidean distances measured between the detected TRPs from known enzymatic targets that use these assayed metabolites as substrates and the corresponding active sites (Extended Data Fig. 6c and Supplementary Dataset 5).Therefore, when the distance from TRPs of certain metabolites to the active site of an enzyme falls within 5.69 Å, altered enzymatic activities may be expected.We chose PKM2, a highly promiscuous target according to TRAP analysis (Fig. 4c), to test the proposed regulatory model.Based on comparative analysis of the accessibility change patterns of the assayed metabolites, changes caused by F6P and G6P resembled that of FBP, the classic activator of PKM2 (ref.29) (Extended Data Fig. 6d,e).Specifically, markedly decreased accessibility at K270, the residue in the PKM2 active site, was observed following F6P/G6P/ FBP incubation (Fig. 4c).These changes agreed with the results of the PKM2 activity assay, which suggested that F6P and G6P are two novel activators in addition to the well-known FBP (Fig. 4d); this occurred because the engagement of activators such as FBP promotes a partially closed conformation in the PKM2 active site 29 and hence prevents this region from being accessed by labeling reagents.In contrast to the patterns of activators, 3PG is a unique ligand that harbors a TRP located distant from the active site of PKM2 (~26.78Å) (Fig. 4c).Thus, once the engagement between 3PG and PKM2 was validated with surface plasmon resonance (SPR) (Fig. 4e), we hypothesized that 3PG may not directly modulate PKM2 activity (Fig. 4d).Intriguingly, TRAP analysis showed that a K433-containing TRP of 3PG is shared by FBP (Extended Data Fig. 6d,f), indicating that 3PG may compete with FBP to bind the K433 proximal region.This inference was substantiated by the counteracted PKM2 activation upon coadministering 3PG with FBP (Fig. 4f).Thus, 3PG hardly altered PKM2 enzymatic activity but inhibited FBP-induced activation.Taken together, we propose that PKM2 activity may be precisely tuned by the composition ratios of multiple glycolytic metabolites.

Glycolytic metabolites modulate druggable enzyme activity
Metabolic enzymes constitute a major class of drug targets.Theoretically, the activities of these enzymes may be influenced by their endogenous ligands.Thus, it would be interesting to test whether the assayed glycolytic metabolites can modulate the activity of their enzyme targets and affect the efficacy of drugs that act on the same targets.We first categorized the TRAP-assigned targetome based on their druggability using DrugBank (Extended Data Fig. 6g).Among the druggable subset of the glycolytic targetome, we noticed that nicotinamide phosphoribosyltransferase (NAMPT), an essential protein for NAD + biosynthesis in cancer cells and hence a therapeutic target for cancer therapy 37 , was assigned as a potential binding target of FBP.Excitingly, its enzymatic activity assay showed that FBP can dose-dependently inhibit NAMPT (Fig. 4g).More importantly, coadministering FBP with FK866, a promising NAMPT inhibitor that is under clinical investigation 37 , synergized in inhibiting NAMPT (Fig. 4g).Mechanistically, TRAP analysis detected that FBP reduced the chemical accessibility at NAMPT-K189, a residue in proximity to FK866 (~3.42 Å) according to crystallography 38 (Fig. 4h), suggesting that FBP binding may promote this region to adopt a more closed and stabilized conformation.Such conformational change may strengthen the FK866-NAMPT engagement and hence enhance the inhibition.Although these in vitro measurements only provided a glimpse into the effect of glycolytic metabolites on drug-target action, we anticipate that new avenues will open up to harness the regulation of glycolysis as adjuvants for cancer therapy through conducting in vivo and clinical experiments in the future.

Transcriptional protein activity influenced by lactate
In addition to enzymes, we noted that the glycolytic targetome constitute proteins ascribed to transcriptional activity according to the

Article
https://doi.org/10.1038/s41589-023-01355-w Human Transcription Factor Database 39 (Fig. 5a).TRIM28 remains an orphan transcriptional regulator without established ligands, and, here, we suggested that TRIM28 is the binding partner for multiple glycolytic metabolites.The association of TRIM28 with tumorigenesis has been reported by mechanistic studies 40 and supported by consistent upregulation in multiple cancer types and a negative association with overall survival 41 (Extended Data Fig. 7a,b).Among the TRAP-identified ligands, lactate most changed the chemical accessibility of TRIM28 (Supplementary Dataset 2), indicating its stronger binding to TRIM28 compared with others.We first confirmed the lactate-TRIM28 interaction by thermal shift assay (Fig. 5b) and SPR (Fig. 5c).Subsequent TRAP analysis of cellular and recombinant TRIM28 against lactate both identified a labeled K337/K340-bearing TRP (Fig. 5d and Extended Data Fig. 7c), suggesting that lactate engagement may occur in proximity to K337/K340.These lysines are located in the coiled-coil domain 42 (Fig. 5e), through which TRIM28 interacts with the Kruppel-associated box domain of zinc finger proteins and exerts transcriptional silencing activity 42 .To corroborate the binding sites, we mutated K337/K340 and confirmed that this mutation weakened the lactate-TRIM28 interaction (Fig. 5c).
Given the established role of TRIM28 as a transcriptional regulator 43 , we then speculated whether lactate engagement would influence its transcriptional activity and induce transcript-level changes in a TRIM28-dependent manner.Therefore, we performed RNA sequencing (RNA-seq) analysis on TRIM28-adequate and TRIM28-deficient HCT116 cells, which were developed by treatment with scrambled small interfering RNA (siCtrl) and siRNA targeting TRIM28 (siTRIM28), in the presence and absence of lactate (Fig. 5f).Pairwise comparison led us to note a subset of 89 genes that were significantly regulated (38 upregulated and 51 downregulated; Supplementary Dataset 6) in response to lactate, but only in TRIM28-adequate rather than in TRIM28-deficient HCT116 cells (Fig. 5g).GO analysis revealed that these genes are linked to a variety of BP, including negative regulation of phosphoptidylinositol 3-kinase (PI3K) signaling, PIWI-interacting    RNA metabolic process and G protein-coupled receptor (GPCR) signaling pathway (Fig. 5h); these results provide a molecular basis for how lactate may influence essential signaling pathways in cancer cells through modulating TRIM28.Intriguingly, genes classified into the BP of negative regulation of PI3K signaling, including NLRC3, CRYBA1 and PLGB1, consistently showed TRIM28-dependent decrease in response to lactate (Fig. 5i).In contrast, TRIM28-dependent upregulation was detected for transcripts such as CXCL1 and CXCR3 (BP: GPCR signaling) in response to lactate, both of which have been positively associated with cancer chemoresistance and metastasis 44,45 .

Metabolite engagement affects target acetylation levels
In addition to transcriptional regulation that controls downstream genes and protein abundance, PTM serves as another core mechanism for tuning protein activity.Driven by the question of whether glycolytic metabolites may affect the PTM levels of their targetome, we

Article
https://doi.org/10.1038/s41589-023-01355-wretrieved PTM evidence 46 for lysines in TRPs and noted that, compared to lysines in the total quantified proteome, the responsive lysines are more likely to carry PTMs (Extended Data Fig. 8a).Notably, acetylation was enriched approximately twofold for lysines of TRPs (Fig. 6a).To exclude the possibility that the differences in the frequency of acetylated lysines resulted from the heightened percentage of acetylation in the assigned targetome, we compared the frequency of lysine acetylation for lysines of TRPs with those of non-TRPs, yet from identical target proteins, and also noted enriched acetylation (Extended Data Fig. 8b).GO analysis showed that these proteins were not distributed in exclusive subcellular locations or specific classes compared to those of the quantified proteome (Extended Data Fig. 8c,d).On the basis of these results, we propose a regulatory principle of glycolytic metabolites on protein lysine acetylation, that is, direct engagement shelters the acetylation sites and may block acetylation.Since G3P is the top-ranked glycolytic metabolite that bears the most documented acetylated lysines in TRPs, aside from lactate (Extended Data Fig. 8e), we sought to validate whether G3P engagement may downregulate the acetylation levels of its target proteins.We first identified α-enolase (ENO1) as its potential target via TRAP, as ENO1 carried a labeled K330/335-bearing TRP that sensitively responded to G3P incubation (R G3P/control = 0.36) (Fig. 6b and Extended Data Fig. 9a).Then, the G3P-ENO1 interaction was confirmed by a thermal shift assay that showed G3P stabilized ENO1 across the tested temperatures
As both K330 and K335 are documented acetylation sites 46 , we examined whether G3P binding inhibits their acetylation.We incubated ENO1 with p300, the responsible acetyltransferase 47 , with and without G3P.Subsequent quantitative proteomic analysis on the in vitro acetylated ENO1 confirmed that acetylation levels at K330/K335 were downregulated by G3P (Fig. 6d).As such, we propose that direct engagement of glycolytic metabolites with their targets may prevalently regulate the PTM levels of the proteins.
Furthermore, we sought to investigate the impact of pyruvate engagement on the acetylation level of its targetome, as pyruvate is another top-ranked metabolite (Extended Data Fig. 8e).Intriguingly, TRAP analysis indicated that pyruvate affected the accessibility of a TRP carrying K27 of histone H3.3 (Supplementary Dataset 2).To validate whether pyruvate can modulate acetylation on histone, we incubated pyruvate with recombinant histone H3.3 and its acetyltransferase p300 (ref.48) in vitro and then performed bottom-up analysis to examine the pyruvate-induced changes in histone lysine acetylation.Quantitative analysis confirmed that multiple lysines were acetylated upon the addition of p300, whereas the acetylation on K27 of histone H3 (H3K27Ac)-in agreement with the TRAP-identified responsive sites-was inhibited by pyruvate (Fig. 6e).We next asked whether the reduced H3K27Ac also occurred in cells.Upon validating that pyruvate can enter cells (Extended Data Fig. 10a,b), we treated HCT116 cells with trichostatin A (TSA), a mammalian histone deacetylase inhibitor 49 , to upregulate H3K27Ac, and then examined whether the presence of pyruvate would inhibit the induced acetylation.In agreement with the in vitro acetylation assay, immunoblotting corroborated that pyruvate administration downregulated TSA-induced acetylation at H3K27 (Fig. 6f).Projecting forward, as H3K27Ac has been reported to induce apoptosis in several cancer cell lines and suppress tumor growth 50,51 , we wondered whether the influence of pyruvate on H3K27Ac affects the pharmacological phenotypes induced by TSA.As a result, flow cytometry (Fig. 6g and Extended Data Fig. 10c) and immunoblotting analysis (Extended Data Fig. 10d) both supported that pyruvate treatment mitigated TSA-induced apoptosis.Thus, our findings represent an intrinsic death-rescuing mechanism of cancer cells executed by glycolytic metabolites.

Discussion
Although our knowledge of the metabolite-protein interaction network and its translational significance is accumulating, this network remains relatively understudied compared with the protein-protein network.Methodologies enabling small-molecule target discovery are paramount for this emerging field.Previously, this endeavor has often been resolved by utilizing chemoproteomic methodologies, including the probe-based approach [13][14][15][16] and derivatization-free, stability changes-based approaches 6,[17][18][19][20][21] .Alternatively, TRAP intuitively measures proteome-wide accessibility at peptide-level resolution, and assigns proteins with apparent changes upon ligand binding as target candidates.Reductive dimethylation is adopted by TRAP to probe the proteome-wide accessibility.Previously, this labeling method has been used to probe solvent accessibility for recognizing conformational changes and identifying ligand-binding sites only with purified proteins 24,26,52 and is posited to modify lysines on protein surfaces 27,52 .Therefore, whether it can be tailored to profile proteome-wide accessibility and enable target discovery for ligands of interest from biological samples remains unknown.Fortunately, our multiplexed-TRAP analysis of cell lysates showed that TRAP labeling achieved a wide coverage of ~60% of the inventory of lysines in the human proteome and the case study of the model target PKM2 revealed that lysines with relatively limited exposure according to solvent-accessible surface area (SASA) analysis were still modified to appreciable extents.Consequently, the lysine labeling-based TRAP approach holds promise to map proteome-wide accessibility and identify targets of metabolites from complex cell milieu on the basis of their significant accessibility changes.
With TRAP, we generated a list of proteins as candidates for the glycolytic targetome, which were then subjected to biological validation.Nevertheless, a fraction of the previously known glycolytic metabolite targets from Homo sapiens retrieved from the BRENDA database were not detected by TRAP and were thus designated as false negatives.We reasoned that these proteins may be missed due to proteome undersampling.In support of this view, we confirmed that the false negatives showed lower sequence coverage than the true positives.Improving the upfront fractionation and extending LC-MS gradients, common practices in proteomics to improve protein coverage, may help to recover targets that evade TRAP detection.Furthermore, analysis of structures of the false negatives inspired us to classify a subset of proteins in which the metabolite-interacting domains lack accessible lysines.This explains why the responsive accessibilities of these bona fide targets were not captured by TRAP and penalized righteous assignment.Future expansion to versatile labeling methods, such as those targeting amino acid residues 24 present in the ligand-binding domain at high frequency (histidine, tryptophan and tyrosine) 53 , and residue-indiscriminate methods, such as footprinting 25,54 and hydrogen-deuterium exchange 55 , may substantially broaden the accessible landscape.Furthermore, some cases occurred when we successfully identified labeled lysines in metabolite-interacting domains from known target proteins; however, no responsive accessibility was detected.Since we observed that increased concentrations of model drugs solicited more profound accessibility changes to the bound targets via dose-responsive TRAP, we inferred that raising the concentration of the assayed glycolytic metabolites may boost accessibility responses and overcome this problem.Notably, the dose-responsive TRAP is devised to profile proteome accessibility over a wide range of ligand concentrations.Through this method, false positives that are assigned when the ligand is assayed only at a single concentration can be removed: as we demonstrated for TEPP-46 and bortezomib.Thus, we anticipate that the initial metabolite-target network constructed via single-dose TRAP in this proof-of-concept study will continue to be refined by large-scale, dose-responsive TRAP analyses in the future, and a more comprehensive landscape of how analog metabolites can differentially and with overlap engage the targetome will be mapped.Another future direction of TRAP is its evolution to applications in living cells.We have recently performed a preliminary adaptation of TRAP for target discovery of an exogenous drug in living cells 56 .Although a subset of responsive proteins was identified, ongoing research is warranted to further advance TRAP to map the targetome of the metabolites in living cells-this is expected to be more physiologically relevant than lysate-level analysis.
As expected, the targetome mapped by TRAP confirmed the promiscuity of glycolytic metabolite-protein interactions and suggested the existence of a coregulatory network.This promiscuity may be attributed to the chemical mimicry among metabolites produced through the same metabolic pathway.Furthermore, the metabolites of analogous structures often display variant affinity and binding topology to the promiscuous target and may thus act differently; for example, the compounds may serve as agonists/antagonists with a spectrum of potency on the same protein-as we have learned from bile acids on farnesoid X receptor and tryptophan derivatives on aryl hydrocarbon receptor 57 .Although this is probably a prevalent mechanism by which nuclear receptors sense metabolic signals, our findings on PKM2 activity coregulated by multiple glycolytic metabolites prompted us to propose that promiscuous glycolytic targets may constitute https://doi.org/10.1038/s41589-023-01355-w

Cell culture and chemicals
Human HCT116 cells were purchased from Cell Bank/Stem Cell Bank, Chinese Academy of Sciences.Human HEK293T and A549 cells were obtained from the American Type Culture Collection.All cells were grown at 37 °C with 5% CO 2 .HCT116 cells were cultured in McCoy's 5A medium purchased from Sigma-Aldrich, HEK293T cells were cultured in DMEM and A549 cells were cultured in RPMI 1640 medium purchased from Gibco.All media were supplemented with 10% FBS (Biological Industries), 100 unit ml −1 penicillin and 100 µg ml −1 streptomycin (Gibco).LC-MS grade water and acetonitrile (ACN) were obtained from Merck.All chemicals were purchased from Sigma-Aldrich unless otherwise specified.

TRAP analysis of purified proteins
Purified proteins, including bovine RNase A and human PKM2, were incubated with their corresponding ligands or solvent for 1 h at room temperature.The proteins were then labeled with 2 µl of 0.5% CD 2 O and 15 µl of 10 mM BPC ( J&K Scientific, catalog no.121499) for 30 min, and the reaction was quenched by the addition of NH 4 HCO 3 .The resultant mixtures were filtered with a 10-kDa molecular weight cutoff (MWCO, Millipore, catalog no.UFC5010) by centrifugation at 12,000g for 15 min.Then, the enriched proteins were denatured by 8 M urea for 30 min followed by incubation with 5 mM dithiothreitol (DTT) at 56 °C for 30 min and subsequent incubation with iodoacetamide (IAM) at room temperature in the dark for 30 min.DTT was added again to neutralize the excessive IAM.The sample was then diluted with 25 mM NH 4 HCO 3 to make the concentration of urea less than 1 M and digested overnight using sequencing-grade trypsin (Promega, catalog no.V5280) in an enzyme:protein ratio of 1:40 (wt/wt) at 37 °C.The digested samples were then desalted with C 18 ZipTips (Waters), dried and stored until further LC-MS/MS analysis.

TRAP analysis of the human cancer cell line proteome
Cell lysate preparation.Cells were washed with cold phosphate buffered saline (PBS), scraped and centrifuged at 800g for 5 min.The cell pellets were resuspended in mammalian protein extraction reagent (M-PER, Pierce/Thermo Fisher Scientific, catalog no.78503) containing protease inhibitor (ApexBio Technology, catalog no.K1007) and phosphatase inhibitor (ApexBio Technology, catalog no.K1013) and lysed on ice for 30 min.The supernatants were collected by centrifugation at 18,000g for 10 min.Protein concentration was determined by bicinchoninic acid protein assay (Beyotime Biotechnology, catalog no.P0011).The cell lysates were then diluted with M-PER buffer to 3 µg µl −1 and incubated with given metabolites/drugs or solvents for 1 h at 25 °C.The dosages of administered glycolytic metabolites were set according to their intracellular concentrations quantified by following a previously described metabolomics approach 61 and slightly modified on the basis of refs.62,63.
Preparation for LFQ-TRAP analysis.Cell lysates (150 µl, 3 µg µl −1 ) were labeled by adding 6 µl of 1% CD 2 O and 90 µl of 10 mM BPC for 30 min.The reaction was quenched by 50 mM NH 4 HCO 3 .Then, methanol, chloroform and water were added to the labeled lysate according to the ratio 4:1:3:1 by volume followed by centrifugation at 12,000g for 10 min to precipitate the proteome.The flaky precipitate was washed twice with methanol and resolubilized by 8 M urea in 25 mM NH 4 HCO 3 .Proteins were then reduced by 10 mM DTT and alkylated by 40 mM IAM.Additional DTT was added to react with excessive IAM.The mixture was then digested overnight with trypsin at an enzyme:protein ratio of 1:40 (wt/wt) at 37 °C.The digestion was quenched with addition of formic acid (FA) to pH 3. The mixture was desalted with Sep-Pak C 18 cartridges (Waters, catalog no.WAT054955) and evaporated to dryness with a vacuum centrifuge (Thermo Fisher Scientific).The samples were stored at −80 °C before LFQ analysis.
TMT-based multiplexed-TRAP.Cell lysates were labeled and precipitated using the same protocol as the LFQ-TRAP workflow.The precipitate was resuspended in 8 M urea solution (in 50 mM Tris-HCl containing 10 mM EDTA, pH 8.0) followed by DTT reduction, IAM alkylation and LysC (Signalchem, catalog no.L585-31 N-05) digestion in an enzyme:substrate ratio of 1:400 (wt/wt) at 25 °C for 4 h.The mixture was diluted with 25 mM NH 4 HCO 3 so that the final concentration of urea was less than 2 M, and then trypsin was added in an enzyme:substrate ratio of 1:50 for overnight digestion at 37 °C.After quenching with FA, the mixture was desalted with Oasis HLB cartridges (Waters, catalog no.WAT094225) and dried.The sample was resuspended in 300 µl of 50 mM HEPES (pH 8.5) with the concentration of peptides determined by quantitative colorimetric peptide assay (Thermo Scientific, catalog no.23275).A 60-µg peptide aliquot of each sample was labeled with TMT reagents for 1.5 h according to the manufacturer's instructions.The reaction was quenched by incubating with 10 µl 5% hydroxylamine for 15 min.The aliquots labeled with the 6-plex TMT reagents or 10-plex TMT reagents were combined and acidified with FA followed by evaporation to dryness.The labeled proteome was then desalted again by SPE cartridges and evaporated to dryness.The proteome was dissolved in HPLC phase A buffer (10 mM ammonium formate containing 5% ACN) and then fractionated on an H class UPLC system (ACQUITY, Waters).Phase B consisted of ACN with 20% 10 mM ammonium formate aqueous buffer.The flow rate was set as 0.2 ml min −1 , and the gradient was set as follows: 0-5 min, 1% B; 5-79 min, 1-50% B; 79-81 min, 50-100% B; 81-98 min, 100% B; 98-100 min, 100-1% B; 100-120 min, 1% B. The effluent was collected every 1.5 min.Every 12 fractions were set as a cycle, and each fraction was combined with the fractions collected in the following cycles.The lyophilized fractions were dissolved in 60 µl of 0.1% FA, desalted with C 18 ZipTips and stored at −80 °C before analysis.

Mass spectrometry
Mass measurement.RNase A (100 µM) and CDP/CTP (1 mM) were both dissolved in 25 mM ammonium acetate buffer and incubated for 30 min, and native MS measurement of the formed holo-complexes was conducted on a TripleTOF 5600 system (SCIEX) by direct infusion.The instrument was set to acquire over the m/z range 100−2,000 for TOF-MS scan.For mass measurement of dimethylated RNase A with and without ligand incubation, samples were desalted by 3 kDa MWCO and analyzed on a C 4 column (4.6 mm × 150 mm, 3 µm, 300 Å, Sepax Technologies) on an LC-30 HPLC system (Shimadzu).The mobile phase consisting of 0.1% FA in water (phase A) and 0.1% FA in ACN (phase B) was delivered at a flow rate of 0.4 ml min −1 using a 15-min gradient program.The eluent was then measured over the m/z range 100-2,000 on the TripleTOF 5600 system.The spectra were combined by summing across the chromatographic peak of labeled RNase A and deconvoluted using SCIEX BioPharma View software v.3.0.E. coli lysates were acquired on a nanoACQUITY UPLC system coupled to a SYNAPT G2-Si mass spectrometer (Waters) and raw data files were collected by Masslynx (Waters, v.4.1).A C 18 trapping column (Waters, 0.18 mm × 20 mm, 5 µm, 100 Å) and an HSS T3 analytical column (Waters, 75 µm × 150 mm, 1.8 µm, 100 Å) were employed.Mobile phases A and B consisted of 0.1% FA in water and 0.1% FA in ACN, respectively.A 60-min and 120-min gradient of 1-40% ACN at a flow rate of 300 nl min −1 was used for separation of recombinant protein digests and cell lysate samples, respectively.The MS scan range was set to m/z 350-1,500 with a scan time of 0.2 s, and the MS/MS scan range was set to m/z 50-2,000 using data-dependent acquisition.The top ten abundant precursors were subjected to MS/MS fragmentation with a ramp collision energy (CE) set between low energy (14-19 eV) and elevated energy (60-90 eV) using a scan time of 0.15 s per function.

LFQ-TRAP analysis. LFQ-TRAP data of recombinant proteins and
https://doi.org/10.1038/s41589-023-01355-wLFQ-TRAP data of HCT116 and A549 lysates were collected on an Orbitrap Eclipse Tribrid mass spectrometer equipped with an EASY-nano LC 1200 liquid chromatography system (Thermo Fisher Scientific) and raw data files were collected by Xcalibur (Thermo Fisher Scientific, v.4.0).Briefly, mobile phase A consisting of 0.1% FA in water and B consisting of ACN-H 2 O (8:2 by volume) were delivered at a flow rate of 300 nl min −1 .The peptides were analyzed using a 120-min chromatography gradient from 3% to 38% phase B during 0-102 min on an Acclaim PepMap RSLC column (Thermo Fisher Scientific, 75 µm × 250 mm, 2 µm, 100 Å).For MS data acquisition, MS 1 spectra were collected in the m/z range 350-1,800 at a resolution of 120,000 in the Orbitrap with a maximum injection time of 50 ms or a maximum automated gain control (AGC) value of 4 × 10 5 .For MS 2 acquisition, fragmentation was conducted by high-energy collision-induced dissociation (HCD) with a normalized collision energy (NCE) at 32. MS 2 spectra were collected at a resolution of 30,000 in the Orbitrap with a maximum AGC of 5 × 10 4 or a maximum injection time of 54 ms.
Multiplexed-TRAP analysis.Multiplexed-TRAP analyses of glycolytic metabolites were conducted on an Orbitrap Fusion Lumos mass spectrometer equipped with an EASY-nano LC 1200 liquid chromatography system.Mobile phase A consisting of 0.1% FA in water and B consisting of 0.1% FA in ACN-H 2 O (8:2 by volume) were delivered at a flow rate of 300 nl min −1 .The 75-µm capillary column was packed with 35 cm of Accucore 150 resin (2.6 µm, 150 Å, Thermo Fisher Scientific).The peptides were analyzed using a 150-min chromatography gradient from 0% to 50% phase B during 5-79 min.For MS data acquisition, MS 1 spectra were collected at the m/z range 375-1,500 at a resolution of 120,000 in the Orbitrap with a maximum injection time of 50 ms or a maximum AGC value of 4 × 10 5 .For MS 2 acquisition, fragmentation was conducted by collision-induced dissociation with an NCE at 35.MS 2 spectra were collected in the mass range 400-1,200 in the ion trap with a maximum AGC of 1 × 10 4 or a maximum injection time of 50 ms.For accurate quantification, MS 3 was conducted for TMT reporter ion quantification by HCD with NCE at 65.The MS 3 spectra were collected over a mass range of 100-500 at a resolution of 50,000 with the maximum injection time set at 10 5 ms and AGC target value at 1 × 10 5 .
Multiplexed-TRAP analyses of TEPP-46/bortezomib were conducted on an Orbitrap Eclipse Tribrid equipped with an EASY-nano LC 1200 liquid chromatography system.The peptides were analyzed using a 120-min chromatography gradient from 8% to 37% phase B during 0-113 min on an Acclaim PepMap RSLC column (Thermo Fisher Scientific, 75 µm × 250 mm, 2 µm, 100 Å).The acquisition parameter settings were similar to those of Orbitrap Fusion Lumos except that the real-time search filter was enabled.

Proteomic data analysis and bioinformatics
Protein identification and quantification.LFQ data and TMT-MS 3 data were searched against the Homo sapiens UniProt database (version 2018) using PEAKS Studio v.8.5.Due to the necessary lysine labeling step employed by TRAP, we allowed up to two missed cleavages and semi-specific tryptic digestion.Carboxyamidomethylation on cysteines (+57.0215Da) was selected as the fixed modification, and methionine oxidation (+15.9949Da), CD 2 O-mediated dimethylation (+32.0564Da) and mono-methylation (+16.0282Da) on lysines were set as variable modifications.For LFQ data acquired on SYNAPT G2-Si, precursor mass tolerance was set to 20 ppm, and fragment mass tolerance was set to 0.1 Da.For LFQ data collected on Orbitrap, precursor mass tolerance was set to 10 ppm, and fragment mass tolerance was set to 0.02 Da.For TMT-based MS 3 quantification data, precursor mass tolerance was set to 10 ppm, MS 2 fragment mass tolerance was set to 0.6 Da and MS 3 fragment mass tolerance was set at 0.02 Da.The identified proteins were filtered with 1% FDR, and the quantified proteins must include at least one unique peptide.For peptide alignment of LFQ data, a 3-min retention time shift tolerance was a ll ow ed.

Classification of TRP candidates
We only quantified lysine-relevant peptides as TRP candidates whose accessibilities could be leveraged by TRAP.When a peptide could be classified into both type A and B/C, type A was stipulated to overrule B/C.

Determination of TRPs for drugs
For LFQ-TRAP data acquired on SYNAPT G2-Si, the screening standard of TRPs was set as peptides displaying a P value <0.01 and TRAP ratio R treated/control > 2 or <0.5 in the presence of the assayed drugs relative to the control.For LFQ-TRAP and multiplexed-TRAP data collected on Orbitrap, a more stringent cutoff was utilized for TRP screening: peptides displaying P value <0.001 and R treated/control > 2 or <0.5 in the presence of ligands relative to the control.

Determination of TRPs for metabolites
For screening of TRPs for metabolites, we first classified them into the following categories: loose and compact.The loose category refers to the TRP candidates that become more chemically accessible to TRAP labeling after the given metabolite incubation.The increased chemical accessibility is recognized by increased abundance of TRP type A candidates or decreased abundance of TRP type B/C candidates.Conversely, the compact category refers to peptides that become less accessible to TRAP labeling after the given metabolite incubation.The reduced chemical accessibility is judged by decreased abundance of TRP type A candidates or increased abundance of TRP type B/C candidates.We set the standard of TRP screening for metabolites as lysine-relevant peptides that display significant abundance changes with q value <0.03 (q values are used to adjust for multiple testing using the Benjamini-Hochberg method to control the FDR at the cutoff level of 0.03) and R treated/control > 1.5 or <0.67 for compact peptides, and R treated/control > 2 or <0.5 for loose peptides.We posit that loose peptides reflect indirect binding events or conformational changes induced by direct binding, so more stringent restriction was given for this category.

Quality assessment of TRAP assignment
We estimated the true positive rate of the TRAP approach by modifying a previous assessment method 6 .We collected the known interactions from the BRENDA repository with the species set as Homo sapiens, and retrieved 128 known enzyme-metabolite interactions.Our TRAP results detected 29 known enzyme-metabolite interactions that are classified as true positive hits, which confer a true positive rate of 22.66% (calculated from 29/128).

Volcano plot of the TRAP-identified targetome
In volcano plots, each point corresponds to a protein that is represented by a peptide selected on the basis of the TRAP scoring system.The TRAP score of each quantified peptide was calculated as follows.

Secondary structure analysis
UniProt identifiers of all quantified proteins were matched with PDB accession numbers from the Protein Data Bank (PDB) (http://www.rcsb.org/pdb/search/searchModels.do), and only entries with >90% sequence identity were retrieved for analysis.Secondary structure information of the compiled protein pool was downloaded from the DSSP (https://swift.cmbi.umcn.nl/gv/dssp)database, and a Python script was used to extract the secondary structure for each quantified lysine residue 64 .The extracted secondary structure classes were classified into four categories, namely, helix (DSSP classes H, G, I), sheet (DSSP classes B, E), loop (T, S) and no structure ('').

Conservation analysis
To estimate the sequence conservation of lysines in TRPs, we calculated the lysine sequence identity across 11 representative vertebrate species https://doi.org/10.1038/s41589-023-01355-w(human, rhesus monkey, mouse, rat, cow, dog, opossum, chicken, frog, zebrafish and fugu) using an in-house Perl script.Specifically, we downloaded multiple amino acid sequence alignment of coding sequence (CDS) regions across 100 species (multiz100way) from the UCSC Genome Browser 65 .For each gene, its CDS region alignment across the selected 11 representative vertebrate species was further extracted.For each lysine from TRPs, its sequence conservation was estimated using the percentage of sequence conservation across 11 representative vertebrate species.To further examine whether the obtained TRP lysine sites were more conserved than random expectation, we calculated the sequence conservation of all quantified lysine sites to estimate the background lysine conservation and then used the KS test to evaluate the statistical significance of excessive sequence conservation of the obtained TRP lysine residues.A KS test P < 0.05 was considered significant.

Measurements of Euclidean distances
For functional site analysis, PyMOL-Python scripts were used to measure the Euclidean distances between the atoms of lysine in TRPs and any atoms of annotated ligands (such as substrates, cofactors, products) in angstrom for enzymes assigned as glycolytic targets by TRAP using the available PDB structures.The minimum distance was recorded to represent each ligand-target pair.Furthermore, if the minimum distance is less than 10 Å, the lysine is categorized as functional and otherwise as nonfunctional.
The active site boundary detectable by TRAP was defined on the basis of the median of the minimum distances measured between the TRPs of known enzymatic targets that use the examined metabolites as substrates and their corresponding active sites.To further evaluate the influence of metabolites on the given enzyme's activities, the minimum distances between all TRPs from the identified targets involved in carbohydrate metabolism and the corresponding active sites were summarized (Fig. 4c and Supplementary Dataset 5).

GO and KEGG pathway analysis
The TRAP-identified targetome were annotated to nonoverlapping GO MF terms.MF classification was performed using the functional annotation tool of PANTHER (http://pantherdb.org/).GO MF terms, including transcription regulator activity (GO:0140110), catalytic activity (GO:0003824), transporter activity (GO:0005215), molecular transducer activity (GO:0060089), translation regulator activity (GO:0045182), structural molecule activity (GO:0005198), molecular function regulator (GO:0098772), binding (GO:0005488) and uncategorized (if not matched to any of the above), were color coded based on the above order (Fig. 3h).If a protein belonged to several classes, it was categorized with the above order of prioritization.Specifically, for proteins in the 'Catalytic activity' category, the ClueGO app included in Cytoscape (v.3.7.1) was employed to perform the KEGG pathway annotation network analysis.A group P value <0.001 and the inclusion of at least three genes in each group were used for filtering.Additionally, the GO BP terms of the ligandable glycolytic targetome were extracted using the BiNGO app included in Cytoscape with a setting of group P value <10 −15 .

LC-MS-based quantification of whole-cell concentrations of glycolytic metabolites
A Shimadzu LC-20 HPLC system equipped with QTRAP 5500 (SCIEX) was used to quantify the whole-cell concentrations of the examined glycolytic metabolites in HCT116 cells.A Waters XBridge BEH amide column (4.6 mm × 100 mm, 3.5 µm) was employed for metabolite separation.The mobile phases consisting of phase A (5% ACN in 10 mM ammonium acetate buffer, pH adjusted to 9 by ammonia) and phase B (ACN) were delivered at a flow rate of 0.4 ml min −1 .A 26-min gradient was used: 0-3 min, 85% B; 3-6 min, 85%-30% B; 6-15 min, 30%-2% B; 15-18 min, 2% B; 18-19 min, 2%-85% B; 19-26 min, 85% B. After separation, the QTRAP 5500 system was operated in negative electrospray ionization source (ESI) mode for metabolite quantification.The parameters of the ESI source were set as follows: ion spray voltage, −4,500 V; curtain gas pressure, 50 psi; ion source gas 1, 30 psi; ion source gas 2, 30 psi; ion source temperature, 500 °C; interface heater temperature, 600 °C.The multiple reaction monitoring (MRM) parameters employed on the QTRAP were set as follows: declustering potential (DP), −10 V for all metabolites; CE, −30 eV for FBP, G6P, G3P, 2/3PG and Lac, −25 eV for R5P and PEP, −12 eV for Pyr, − 40   61 .Briefly, the standard curves of each metabolite were generated, which allowed the extracted concentration (EC) of individual metabolites to be quantified for the analyzed whole-cell sample.The total moles of given metabolites in a whole-cell sample were deduced from the EC value and their corresponding sample volume.Finally, the whole-cell concentrations of each metabolite were estimated using the total moles of metabolites in a whole-cell sample, the total cell number per sample and the volume of each cell 61 .The number and diameter of the assayed HCT116 cells were measured by a Cellometer Mini Cell Counter (Nexcelom Bioscience).

siRNA transfection
The siRNAs used in this study, siCtrl and siTrim28, were both purchased from Santa Cruz Biotechnology (catalog nos.sc-37007 and sc-38550).Approximately 3 × 10 5 HCT116 cells were seeded into six-well plates for 24 h and then transfected with 10 nM siRNA using 5 µl lipofectamine RNAiMAX reagent (Invitrogen, catalog no.13778030) for 48 h according to the manufacturer's instructions.The efficiency of silencing was confirmed by immunoblotting.For RNA-seq experiments, cells were treated with 10 mM sodium lactate or the matching solvent at 24 h posttransfection, and then cultured for another 24 h before collection for RNA-seq analysis.

RNA-seq analysis
Total RNA was isolated using TRIzol reagent (Thermo Scientific, catalog no.15596026), and RNA-seq was performed by Annoroad Gene Technology.The Novaseq Control Software (v.1.7.5) was used for RNA-seq data collection.Briefly, the integrity and concentration of RNA were analyzed using an Agilent RNA 6000 Nano Kit and Agilent 2100 Bioanalyzer (Agilent Technologies).RNA-seq was performed using an Illumina Nova 6000 platform (Illumina).Raw sequenced reads in FASTQ format were filtered to achieve high-quality reads and then mapped to the human genome (GRCh38.100.chr) using HISAT2.Differentially expressed genes (DEGs) were analyzed using the R package DESeq2 with a cutoff of FC > 1.5 or <0.67 and P value <0.05.GO BP analysis of DEGs was performed by the DAVID online tool (https://david.ncifcrf.gov/).The experiment was performed using four biologically independent samples per group.

Recombinant protein expression and purification
Briefly, plasmids were transformed into BL21(DE3) E. coli cells, and the transformed bacteria were selected on LB plates containing 50 µg ml −1 kanamycin.The isolated colonies were grown in LB medium at 37 °C until the optical density (OD 600 ) reached 0.8.Protein expression was induced with 0.5 mM isopropyl β-d-thiogalactoside (IPTG) by shaking overnight at 16 °C.Cells were collected by centrifugation and then lysed by sonication.Lysates were then purified with a His-tag Protein Purification Kit (Beyotime Biotechnology, cat.no.P2226) and were concentrated by 10 kDa MWCOs.For MBP-tagged TRIM28, cultures were supplemented with LB medium containing 50 µg ml −1 ampicillin and 0.5% glucose.Protein expression was induced with 0.2 mM IPTG by shaking overnight at 16 °C.Then, cells were resuspended in lysis buffer (50 mM Tris, pH 7.4, 50 mM NaCl, 5 mM DTT, 1:10,000 (v/v) benzonase solution, 1× protease inhibitor cocktail) and lysed by sonication.The supernatant was transferred to amylose resin and purified using a New England Biolabs kit (catalog no.E8201S) according to the manufacturer's protocol.The yield of purified proteins was evaluated by separation on SDS-PAGE and visualized via Coomassie blue staining.

Enzymatic activity assay
PKM2 activity was determined using luminescent Kinase-Glo Plus reagent (Promega, catalog no.V3773) according to the manufacturer's instructions.In brief, human recombinant PKM2 was diluted in assay buffer (50 mM Tris-HCl, 100 mM KCl, 10 mM MgCl 2 , pH 7.4) and incubated with each metabolite/drug/vehicle at 25 °C for 40 min.The substrate solution was prepared by mixing ADP at 200 µM and PEP at 200 µM with the assay buffer.Then, a 50-µl aliquot of substrate solution was added to 50 µl of recombinant PKM2 solution and reacted for 10 min at room temperature in 96-well plates.The luminescence response of each sample was measured at 37 °C in a Gen5 platform (BioTek).The measurements were fitted by a nonlinear fitting algorithm (log [agonist] versus response-variable slope with four parameters) in Prism v.8.0.1 (GraphPad).
NAMPT activity was measured using an NAMPT colorimetric assay kit (CycLex, MBL, catalog no.CY 1251V2) in 96-well plates by a one-step method.In brief, recombinant NAMPT was diluted in NAMPT assay buffer and subsequently incubated with solvent/metabolites/ FK866 for 30 min at 30 °C.The reaction was initiated by the addition of 60 µl One-Step Assay Mixture to each well.NAMPT activity was measured photometrically in absorbance at 450 nm for 60 min with a 1-min interval.

Thermal shift assay and ITDR assay
Thermal shift assay of cell lysates was performed according to the previous literature 19 .Briefly, cultured HCT116 cells were lysed in 0.4% NP-40 lysis buffer supplemented with protease and phosphatase inhibitor cocktail by repetitive freeze thawing.The resultant supernatant was divided into two aliquots, with one aliquot being treated with given glycolytic metabolites and another aliquot with the solvent.After 60-min incubation, each sample was further divided into eight aliquots and heated at the designated temperatures for 3 min in a 96-well thermal cycler, followed by cooling at room temperature for 3 min.The lysates were centrifuged at 20,000g for 20 min at 4 °C, and analyzed by immunoblotting against the target proteins.The band intensities detected at increasing temperatures were normalized to that of the lowest temperature, and the Boltzmann sigmoid equation was fitted using Prism v.8.0.1.
For lysate-level ITDR experiments, HCT116 cell lysates were aliquoted and incubated with serial concentrations of G3P for 1 h, followed by heating at designated temperatures for 3 min and cooling for another 3 min at room temperature.Then, the soluble fractions were isolated and immunoblotted against ENO1.The resultant band intensities of ENO1 incubated with varying concentrations of G3P were normalized to the intensity at the highest concentration of G3P and plotted using the saturation binding curve function in Prism v.8.0.1.

Surface plasmon resonance analysis
SPR analysis was conducted on a Biacore T200 system (GE Healthcare).Briefly, target protein was diluted in 10 mM sodium acetate and immobilized via the amine coupling method on a CM5 sensor chip (Cytiva).Metabolites were dissolved in H 2 O and diluted to serial concentrations with running buffer (PBS with 0.05% v/v surfactant P20, pH 7.4).Then, the assayed metabolites were injected over the immobilized target protein at a flow rate of 30 µl min −1 at 25 °C.Data analysis was carried out on Biacore T200 evaluation software and the K D was obtained using the steady-state affinity model.

In vitro lysine acetylation (Kac) assay
Kac assay was performed according to the literature 47 .Briefly, purified ENO1 or histone H3.3 (10 µg, New England Biolabs, catalog no.M2507S) was diluted with reaction buffer (50 mM Tris-HCl, pH 8.0, 100 nM TSA, 0.1 mM EDTA, 1 mM DTT and protease inhibitor cocktail) and incubated with the assayed metabolites, respectively, for 30 min.Then, the mixtures were incubated with 2 mM acetyl-CoA (Shanghai Yuanye Bio-Technology, catalog no.Y34045) and 0.55 µg of human p300 protein (Active Motif, catalog no.81158) at 30 °C for 45 min.Acetylation was terminated by the addition of urea.Finally, sample preparation and bottom-up analysis of the acetylated proteins was performed as previously described.

Flow cytometry analysis of cell apoptosis
The cell apoptotic rate was examined using an AV-FITC kit (BD Biosciences).Briefly, cells were collected, washed with PBS, resuspended

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

R 23 RFig. 1 |
Fig.1| Benchmarking the TRAP approach for targetome mining in native cell milieu.a, Illustration of the TRAP workflow.TRAP leverages altered accessibility of proteinaceous lysine residues to covalent labeling reagents in response to ligand engagement by quantitative proteomics, and assigns proteins carrying lysines with significant accessibility changes as target candidates.b, Crystallographic structure of PKM2 bound to FBP (orange) and Pyr (red) (PDB 4YJ5).Lysines, including K270, K337, K422 and K433, are coded in blue.c, Extracted ion chromatograms (EICs) of TRPs and non-TRPs of FBP (control, light pink; low-dose FBP, pink; high-dose FBP, purple) assigned by TRAP analysis of recombinant PKM2.Experiments were performed with three biologically independent samples, and representative EICs are shown.d, Volcano plots identifying PKM2 as the target of TEPP-46 (10 µM) in HCT116 and A549 cell lysates (n = 4 biologically independent samples).Proteins carrying TRPs (R TEPP-46/control > 2 or <0.5, P < 0.001 by unpaired two-sided Student's t-test)

Fig. 2 |
Fig. 2 | TRAP can identify thermally stable protein targets and hence complements thermal stability-based target-discovery approaches.a, Reprocessing the human HEK293T cell meltome data (PXD011929; ref. 31) identified a nonmelting cluster.b, GO BP analysis of the nonmelters shown in a using a modified Fisher's exact test from DAVID bioinformatics website.c, Proteins belonging to the proteasomal protein catabolic process, exemplified by PSMB1, showed resistance to heat denaturation.The S-plot is plotted by calculating the relative abundances of all quantified proteins that remained soluble at 67 °C versus 37 °C using data in a. d, Melting curves of the representative nonmelter PSMB1 in HEK293T, colon cancer spheroids, HaCaT and HepG2 cells using data retrieved from the Meltome Atlas (PXD011929; ref. 31).One representative dataset was reanalyzed for each cell line.e, Crystallographic structure of the bortezomib (orange)-bound human 20S proteasome (PDB 5LF3, left panel) and a detailed view of the identified TRP in PSMB1 (in blue, right panel).f, Volcano plot of the multiplexed-TRAP analysis of bortezomib (1 µM) in HCT116 cell lysates (n = 3 biologically independent samples).Proteins carrying TRPs (R TEPP-46/control > 2 or <0.5, P < 0.001 by unpaired two-sided Student's t-test)were assigned as potential targets and are highlighted in red.g, Dose-responsive TRAP analysis identified a labeled K164-containing TRP type A reflecting dosedependently decreased accessibility at PSMB1-K164 following the administration of bortezomib at serial concentrations.The experiment was conducted once.

Fig. 3 |
Fig. 3 | TRAP mapped a global glycolytic targetome in HCT116 cells.a, Illustrated glycolysis pathway specifying the ten assayed glycolytic metabolites and relevant enzymes.b, The responsive targetome of glycolytic metabolites (FBP, 200 µM; F6P, 100 µM; G6P, 100 µM; R5P, 30 µM; G3P, 30 µM; 2PG, 10 µM; 3PG, 10 µM; PEP, 5 µM; Pyr, 300 µM; Lac, 2 mM) in HCT116 cell lysates were identified with two batches of multiplexed-TRAP experiments (n = 3 biologically independent samples).c, Broad coverage of lysines achieved by TRAP labeling.d, Labeling preference analysis for high-order structures using the TRAP approach.e, Volcano plots of the glycolytic targetome (highlighted in blue) assigned by TRAP.f, Conservation analysis of lysines in all quantified peptides (control) versus lysines in TRPs assigned for the assayed glycolytic metabolites (positive).Statistical significance was evaluated by the Kolmogorov-Smirnov (KS) test and CDF represents the cumulative distribution function.g, Box plot (center lines mark the median, box borders represent the first and third quartiles and the whiskers indicate the minimum and maximum values) showing the distance from the lysines in TRPs versus lysines in all quantified peptides to the functional sites of the enzymes assigned as the glycolytic targets by TRAP.P values were calculated by unpaired two-sided Student's t-test.h, GO MF classification of the TRAP-identified glycolytic targetome showing enrichment in enzymes (MF: catalytic activity).

Fig. 4 |
Fig. 4 | Functional validation of glycolytic metabolites on the carbohydrate metabolism enzymes assigned as glycolytic targets by TRAP.a, KEGG pathway annotation network analysis of catalytic proteins from the TRAPidentified glycolytic targetome.Pathways with a P value <0.001 (by two-sided hypergeometric test; see Supplementary Dataset 4) and the inclusion of at least three genes were summarized, and their enrichment in the carbohydrate metabolism pathway is highlighted (black circles).b, Fraction of the TRAPidentified enzyme targets involved in carbohydrate metabolism.c, Glycolytic metabolite binding-induced accessibility changes in the active sites of enzyme targets involved in the carbohydrate metabolism pathway.The numbers in the boxes indicate the minimal Euclidean distances from the TRPs of the identified target to the closest atom of the substrates.The corresponding R treated/control

Fig. 5 |
Fig. 5 | Lactate engagement influenced the transcriptional activity of TRIM28.a, Proteins of transcriptional activity assigned as glycolytic targets by TRAP.They were further classified by ligandability based on the presence in DrugBank.b, Immunoblotting-based thermal shift assay suggesting lactate (2 mM) stabilized TRIM28 in HCT116 cell lysates.Experiments were repeated (n = 3 biologically independent samples), with one representative sample shown.c,SPR analysis of the affinity of lactate for wild type (WT)-TRIM28 and K337E/ K340E double mutant-TRIM28.The experiment was conducted once.d, TRAP analysis of human recombinant TRIM28 following lactate (2 mM) incubation (n = 3 biologically independent samples).Peptides showing significantly changed accessibility (R treated/control > 1.5 or <0.67, P < 0.05 by unpaired two-sided Student's t-test) were determined as TRPs and are highlighted in red.e, X-ray structure pinpointing the K337/K340-bearing TRP of lactate (in cyan) located in the coiledcoil domain of TRIM28 (PDB 6H3A).f, Immunoblots of TRIM28 in HCT116 cells

Fig. 6 |
Fig. 6 | Glycolytic metabolites modulated acetylation levels of their binding targets.a, PTM category analysis of lysines in quantified peptides versus lysines in TRPs of the glycolytic targetome.b, S-plot of the TRPs of G3P suggesting ENO1 as a target candidate.c, SPR analysis of the affinity of G3P for the immobilized WT-ENO1 or K335E-mutant ENO1.The experiment was conducted once.d, Ion abundance of peptides bearing acetylated lysines from human recombinant ENO1 treated without and with G3P (100 µM) following p300 incubation.e, Ion abundance of peptides bearing acetylated lysines from human recombinant histone H3.3 treated without and with Pyr (100 µM) following p300 incubation.
https://doi.org/10.1038/s41589-023-01355-w in 1× AV binding buffer and stained with AV and PI for 15 min at room temperature in darkness.The rates of apoptosis were determined using an Accuri C6 flow cytometer (BD Biosciences).

Extended Data Fig. 1 |
Benchmarking the TRAP approach in probing ligandtarget interactions using RNase and PKM2.(a) Docking analysis of RNase with its ligand CDP (PDB: 1ROB).(b) Native MS showing CDP and CTP bound to RNase at different affinities.(c) Microscale thermophoresis (MST) showing CTP possessed stronger affinity to RNase than CDP.The experiment was conducted once.(d) Time-resolved intact MS analysis showing impeded mass shift of TRAP-labeled RNase in response to CDP/CTP binding (n indicates the number of TRAP-labeled lysines on RNase).(e) Summarized accessibility changes of lysines (R treated/control ) in RNase in response to CDP/CTP binding based on quantitative analysis of labeled lysine-containing peptides via TRAP.(f) Crystal structure of FBP (red, sphere)-induced PKM2 tetramerization (grey, cartoon) by an allosteric mechanism (PDB: 4B2D).The measured minimal Euclidean distances suggested that the PKM2-K422 residue (blue, stick) is distant from both FBP and pyruvate (in dashed lines).(g) Summarized accessibility changes of lysines (R treated/control ) in human recombinant PKM2 in response to FBP incubation based on quantitative analysis of labeled lysine-containing peptides via TRAP.For (e) and (g), data represent the mean ± SEM (n=3 biologically independent samples) and P values were determined using an unpaired two-tailed Student's t-test.Extended Data Fig. 2 | TRAP identified PKM2 as the target of TEPP-46 and pinpointed the binding site.(a) Crystal structure suggesting the binding site of TEPP-46 (red, sphere) as PKM2-K305 (PDB: 3U2Z).The TRAP-assigned TRP bearing K305 was colored in blue.(b) Correlation analysis between SASA and labeling occupancy of lysine residues of PKM2 without (left panel) and with TEPP-46 (right panel).The Pearson's correlation coefficients (r) and statistical significance of correlation (P) determined by unpaired two-tailed Student's t-test are shown.In agreement with the decreased SASA, labeling occupancy of PKM2-K305 (in red), the binding site of TEPP-46, was also markedly reduced following TEPP-46 incubation.Specifically, labeling occupancy of each K was estimated by calculating the ratio between abundance of the peptides bearing this labeled K residue and summed abundance of the labeled K-containing peptides as well as those carrying this unlabeled K residue.(c) TRAP identified the labeled K305 and K311-bearing TRPs in human recombinant PKM2, both of which signified markedly decreased accessibility at K305/K311 following TEPP-46 incubation.R TEPP-46/control were calculated based on n=3 biologically independent samples, while representative EICs of the TRPs and the non-TRPs bearing K115/ K188 are shown.(d) Summarized accessibility changes of lysines (R treated/control ) using data in (c).(e) R treated/control of the labeled K305-containing TRP in E. coli lysates following TEPP-46 treatment (n=5 biologically independent samples).(f) Dose-responsive accessibility change curves for TEPP-46's target candidates that were assigned by single-dose TRAP experiment shown in Fig. 1h (the bona fide target PKM2 excepted).The experiment was conducted once.For (d) and (e), data represent the mean ± SEM and P values were determined using an unpaired two-tailed Student's t-test.Extended Data Fig. 3 | TRAP complements thermal stability-based target discovery approaches.(a) Analysis of protein melting behaviors using the human HEK293T cell meltome data (PXD011929) identified five clusters as shown here and in Fig. 2a.(b) Nonmelters displaying resistance to thermal denaturation were identified in the meltome data of HaCaT, HepG2 and colon cancer spheroids cells (PXD011929).(c) GO BP analysis of the nonmelters shown in (b) using a modified Fisher's Exact test from DAVID bioinformatics website.
(d) Dose-responsive TRAP detected dose-dependent increased abundance of the TRP type C of PSMB1-K164, implying dose-dependent decreased accessibility at K164 following bortezomib incubation.The experiment was conducted once.(e) Dose-responsive TRAP curves for bortezomib's target candidates that were assigned by single-dose TRAP experiment shown in Fig. 2f (the bona fide target PSMB1 excepted).The experiment was conducted once.Extended Data Fig. 5 | Characterizing the TRAP-assigned glycolytic targetome using the multiplexed-TRAP data.(a) Wide TRAP-labeling coverage of lysines based on analysis of the second batch of the multiplexed-TRAP data.(b-c) Chemical accessibility of proteinaceous lysines assessed by the labeled fraction of lysine residues for each quantified protein using the first (b) and second (c) batches of the multiplexed-TRAP data.(d) Analysis of TRAP labeling preference for high-order structures using the second batch of the multiplexed-TRAP data.(e) Summary of the numbers of glycolytic metabolite-protein interactions detected by TRAP.(f-g) Box plots (center lines mark the median, box borders represent the first and third quartiles, and the whiskers indicate the minimum and maximum values) of sequence coverage (f) and the number of detected unique peptides (g) for known glycolytic targets (retrieved from BRENDA, species: human) that can and cannot be identified by TRAP (unpaired two-tailed Student's t-test).Extended Data Fig. 6 | Functional assessment and validation of the identified glycolytic metabolite-target interactions belonging to the carbohydrate metabolism pathway.(a) GO MF analysis of the TRAP-identified glycolytic targetome showing enrichment in proteins ascribed to catalytic activity.(b) KEGG pathway annotation summarized the enriched pathways for the targets ascribed to catalytic activity.(c) Illustration of how the boundary of active site is defined using the multiplexed-TRAP data.The minimal Euclidean distances between the TRPs of the TRAP-identified targets that use the assayed metabolites as substrates and the corresponding active site (retrievable from PDB) were measured, and the resultant median of the collected distances was used to represent the active site boundary that can be probed by TRAP.(d) Summarized R treated/control values of the TRPs in PKM2 via the multiplexed-TRAP analysis of HCT116 cell lysates following FBP, F6P and G6P incubation, respectively.Of note, the letter before K denotes the type of the classified TRPs.(e) Relative (Rel.)PKM2 activity at different FBP concentrations normalized to the activity without FBP.(f) TRAP analysis delivering the R treated/control of the TRP type B carrying PKM2-K433 following 3PG incubation.(g) DrugBank and non-DrugBank fractions of the quantified proteome (n=4778) vs. the TRAP-assigned glycolytic targetome (n=913) using the multiplexed-TRAP data.For (d, e, f), data represent the mean ± SEM (n=3 biologically independent samples).For (d, f), P values were determined using an unpaired two-tailed Student's t-test.Extended Data Fig. 7 | Functional and structural characterization of the lactate-TRIM28 interaction.(a) Gene expression profile of TRIM28 across given cancer types and paired normal tissues retrieved from the TCGA and the GTEx projects were plotted using GEPIA.(b) GEPIA-based analysis implying that high TRIM28 gene expression level is unfavored for patients' survival for the examined cancer types using Log-rank test (median cutoff).(c) Summarized R treated/control values of the TRPs assigned for lactate via the LFQ-TRAP analysis of human recombinant TRIM28 (upper panel, all peptides shown here are TRP Type A candidates) and the multiplexed-TRAP analysis of HCT116 cell lysates (bottom panel, the letter before K denotes the type of the classified TRPs).Data represent the mean ± SEM (n=3 biologically independent samples) and P values were determined using an unpaired two-tailed Student's t-test.Extended Data Fig. 8 | Glycolytic metabolites binding regulated targetome acetylation.(a) Fraction of lysines in quantified peptides vs. lysines in TRPs of the glycolytic targetome based on PTM annotation with iPTMnet.(b) Percentage of lysine acetylation in non-TRPs vs. TRPs retrieved from the same glycolytic target (n=913).Data represent the mean ± SEM and P value was determined by a paired, two-tailed Student's t-test.(c) GO analysis showing diverse subcellular locations for the fraction of TRAP-assigned glycolytic targets that have been assigned as acetylation carriers by iPTMnet.The subcellular distribution pattern resembles that of the whole glycolytic targetome.(d) Distribution pattern of MFs summarized for the whole glycolytic targetome vs. MFs of the glycolytic targetome that have been documented as acetylation carriers by iPTMnet.(e) Number of acetylated lysines in TRPs of each assayed glycolytic metabolite based on iPTMnet.Extended Data Fig. 9 | Validation of ENO1 as the binding target of G3P.(a) Summarized R treated/control values of the TRPs in ENO1 assigned for G3P using the multiplexed-TRAP data.Of note, the letter before K denotes the Type A/B/C of TRPs.Data represent the mean ± SEM (n=3 biologically independent samples) and P values were determined using an unpaired two-tailed Student's t-test.(b) Thermal shift assay validating the engagement of G3P (500 µM) with ENO1 using HCT116 cell lysates.Experiments were repeated (n=3 biologically independent samples) with one representative sample shown.(c) Thermal shift assay showing G3P-mediated stabilization of ENO1 using HCT116 cell lysates.Experiments were repeated (n=2 biologically independent samples) with one representative sample shown.(d) Volcano plot of the TRPs of G3P (500 µM) assigned by TRAP analysis of the human recombinant ENO1 (n=3 biologically independent samples).TRPs (R G3P/control >1.5 or <0.67, p <0.05 by unpaired twosided Student's t-test) were determined and highlighted in red.(e) SPR analysis showing weakened affinity of G3P to the ENO1 carrying K330E point mutation (Mutant-ENO1).The experiment was conducted once.Extended Data Fig. 10 | Metabolomic and immunoblotting analysis verified the ability of pyruvate in entering HCT116 cells and rescuing TSA-induced apoptosis.(a) EICs of intracellular pyruvate in HCT116 cells without and with pyruvate administration for 24 hr (n=3 biologically independent samples).(b) Relative ion abundance of intracellular pyruvate using data in (a).Data represent the mean ± SEM (n=3 biological samples) and P value was determined using an unpaired two-tailed Student's t-test.(c) Gating strategy to sort TSAinduced apoptotic HCT116 cells without and with pyruvate pretreatment as specified in Fig. 6g.(d) Representative immunoblots of the apoptotic protein markers, including cleaved PARP and cleaved caspase-9, in HCT116 cells treated as specified in Fig. 6g.β-Tubulin was used as the loading control.Statistical analyses of the band intensities of the protein markers are shown in the bottom panel, where data represent the mean ± SEM (n=3 independent experiments) and P values were determined using ordinary one-way ANOVA with Tukey's multiple comparisons test.β β Scatter plot of RNA-seq analysis of HCT116 cells treated with siTRIM28 or siCtrl in the presence or absence of lactate (10 mM, 24 h).Genes showing significant changes in response to lactate (fold change, FC > 1.5 or <0.67 and P < 0.05 by Wald test as implemented in DESeq2) in siCtrl-treated HCT116 cells while holding constant (P > 0.05) in the siTRIM28-treated group are highlighted in red (n = 4 biologically independent samples; see Supplementary Dataset 6).h, GO BP analysis of the genes highlighted in g using a modified Fisher's exact test from DAVID bioinformatics website.i, RNA-seq heat map of the genes highlighted in g generated by plotting the log 2 FPKM value.FPKM, fragments per kilobase of transcript per million mapped fragments; Lac, lactate (10 mM, 24 h); Con, control (matching solvent).