Abstract
The ability to understand and predict variable responses to therapeutic agents may improve outcomes in patients with cancer. We hypothesized that the basal gene-transcription state of cancer cell lines, coupled with cell viability profiles of small molecules, might be leveraged to nominate specific mechanisms of intrinsic resistance and to predict drug combinations that overcome resistance. We analyzed 564,424 sensitivity profiles to identify candidate gene–compound pairs, and validated nine such relationships. We determined the mechanism of a novel relationship, in which expression of the serine hydrolase enzymes monoacylglycerol lipase (MGLL) or carboxylesterase 1 (CES1) confers resistance to the histone lysine demethylase inhibitor GSK-J4 by direct enzymatic modification. Insensitive cell lines could be sensitized to GSK-J4 by inhibition or gene knockout. These analytical and mechanistic studies highlight the potential of integrating gene-expression features with small-molecule response to identify patient populations that are likely to benefit from treatment, to nominate rational candidates for combinations and to provide insights into mechanisms of action.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 print issues and online access
$259.00 per year
only $21.58 per issue
Buy this article
- Purchase on Springer Link
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
Data availability
All primary small-molecule screening data were downloaded from the NCI CTD2 Data Portal (https://ocg.cancer.gov/programs/ctd2/data-portal) or GDSC Portal (https://www.cancerrxgene.org/downloads), and all cell-line genomic data were downloaded from the GDSC Portal or the DepMap Portal (https://depmap.org/portal). GSK-J4 AUC data from ref. 32 were obtained from Supplementary Table 1 of the manuscript (https://doi.org/10.1126/scitranslmed.aao4680; file aao4680_Table S1.xlsx). All raw viability data in Figs. 2–6 are provided at https://doi.org/10.6084/m9.figshare.19067246. Raw data for ORF screens (Fig. 2g–i and Supplementary Table 2) are provided as Supplementary Dataset 5. Raw data for the sgRNA screen are provided as Supplementary Dataset 7. Source data are provided with this paper.
Code availability
Custom code (with worked examples) is available at https://github.com/broadinstitute/prism_data_processing and https://github.com/reesmg.
References
Konieczkowski, D. J., Johannessen, C. M. & Garraway, L. A. A convergence-based framework for cancer drug resistance. Cancer Cell 33, 801–815 (2018).
Chapman, P. B. et al. Improved survival with vemurafenib in melanoma with BRAF V600E mutation. N. Engl. J. Med. 364, 2507–2516 (2011).
Robert, C. et al. Improved overall survival in melanoma with combined dabrafenib and trametinib. N. Engl. J. Med. 372, 30–39 (2015).
Konieczkowski, D. J. et al. A melanoma cell state distinction influences sensitivity to MAPK pathway inhibitors. Cancer Discov. 4, 816–827 (2014).
Yu, C. et al. High-throughput identification of genotype-specific cancer vulnerabilities in mixtures of barcoded tumor cell lines. Nat. Biotechnol. 34, 419–423 (2016).
Corsello, S. M. et al. Discovering the anticancer potential of non-oncology drugs by systematic viability profiling. Nat. Cancer 1, 235–248 (2020).
Lehár, J. et al. Synergistic drug combinations tend to improve therapeutically relevant selectivity. Nat. Biotechnol. 27, 659–666 (2009).
Mathews Griner, L. A. et al. High-throughput combinatorial screening identifies drugs that cooperate with ibrutinib to kill activated B-cell-like diffuse large B-cell lymphoma cells. Proc. Natl Acad. Sci. USA 111, 2349–2354 (2014).
Greco, W. R., Bravo, G. & Parsons, J. C. The search for synergy: a critical review from a response surface perspective. Pharmacol. Rev 47, 331–385 (1995).
Foucquier, J. & Guedj, M. Analysis of drug combinations: current methodological landscape. Pharmacol. Res. Perspect. 3, e00149 (2015).
Iorio, F. et al. A landscape of pharmacogenomic interactions in cancer. Cell 166, 740–754 (2016).
Rees, M. G. et al. Correlating chemical sensitivity and basal gene expression reveals mechanism of action. Nat. Chem. Biol. 12, 109–116 (2016).
Seashore-Ludlow, B. et al. Harnessing connectivity in a large-scale small-molecule sensitivity dataset. Cancer Discov. 5, 1210–1223 (2015).
Barretina, J. et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature 483, 603–607 (2012).
Ben-David, U. et al. Genetic and transcriptional evolution alters cancer cell line drug response. Nature 560, 325–330 (2018).
de Waal, L. et al. Identification of cancer-cytotoxic modulators of PDE3A by predictive chemogenomics. Nat. Chem. Biol. 12, 102–108 (2016).
Winter, G. E. et al. The solute carrier SLC35F2 enables YM155-mediated DNA damage toxicity. Nat. Chem. Biol. 10, 768–773 (2014).
Tsherniak, A. et al. Defining a cancer dependency map. Cell 170, 564–576.e16 (2017).
Bersuker, K. et al. The CoQ oxidoreductase FSP1 acts parallel to GPX4 to inhibit ferroptosis. Nature 575, 688–692 (2019).
Hegi, M. E. et al. MGMT gene silencing and benefit from temozolomide in glioblastoma. N. Engl. J. Med. 352, 997–1003 (2005).
Tahir, S. K. et al. Identification of expression signatures predictive of sensitivity to the Bcl-2 family member inhibitor ABT-263 in small cell lung carcinoma and leukemia/lymphoma cell lines. Mol. Cancer Ther. 9, 545–557 (2010).
Lu, D. et al. Metabolism of the anthelmintic drug niclosamide by cytochrome P450 enzymes and UDP-glucuronosyltransferases: metabolite elucidation and main contributions from CYP1A2 and UGT1A1. Xenobiotica 46, 1–13 (2016).
Lamers, F. et al. Targeted BIRC5 silencing using YM155 causes cell death in neuroblastoma cells with low ABCB1 expression. Eur. J. Cancer 48, 763–771 (2012).
Hafner, M., Niepel, M., Chung, M. & Sorger, P. K. Growth rate inhibition metrics correct for confounders in measuring sensitivity to cancer drugs. Nat. Methods 13, 521–527 (2016).
Piccioni, F., Younger, S. T. & Root, D. E. Pooled lentiviral-delivery genetic screens. Curr. Protoc. Mol. Biol. 121, 32.1.1–32.1.21 (2018).
Yang, X. et al. A public genome-scale lentiviral expression library of human ORFs. Nat. Methods 8, 659–661 (2011).
Conrad, M. & Sato, H. The oxidative stress-inducible cystine/glutamate antiporter, system x (c) (-): cystine supplier and beyond. Amino Acids 42, 231–246 (2012).
Bachovchin, D. A. et al. Superfamily-wide portrait of serine hydrolase inhibition achieved by library-versus-library screening. Proc. Natl Acad. Sci. USA 107, 20941–20946 (2010).
Crow, J. A., Bittles, V., Borazjani, A., Potter, P. M. & Ross, M. K. Covalent inhibition of recombinant human carboxylesterase 1 and 2 and monoacylglycerol lipase by the carbamates JZL184 and URB597. Biochem. Pharmacol. 84, 1215–1222 (2012).
Ramírez, J., Mirkov, S., House, L. K. & Ratain, M. J. Glucuronidation of OTS167 in humans is catalyzed by UDP-glucuronosyltransferases UGT1A1, UGT1A3, UGT1A8, and UGT1A10. Drug Metab. Dispos. 43, 928–935 (2015).
Pattanawongsa, A., Chau, N., Rowland, A. & Miners, J. O. Inhibition of human UDP-glucuronosyltransferase enzymes by canagliflozin and dapagliflozin: implications for drug-drug interactions. Drug Metab. Dispos. 43, 1468–1476 (2015).
Lochmann, T. L. et al. Targeted inhibition of histone H3K27 demethylation is effective in high-risk neuroblastoma. Sci. Transl. Med. 10, eaao4680 (2018).
Basu, A. et al. An interactive resource to identify cancer genetic and lineage dependencies targeted by small molecules. Cell 154, 1151–1161 (2013).
Nomura, D. K. et al. Monoacylglycerol lipase regulates a fatty acid network that promotes cancer pathogenesis. Cell 140, 49–61 (2010).
Heinemann, B. et al. Inhibition of demethylases by GSK-J1/J4. Nature 514, E1–E2 (2014).
Long, J. Z. & Cravatt, B. F. The metabolic serine hydrolases and their functions in mammalian physiology and disease. Chem. Rev. 111, 6022–6063 (2011).
Mulder, K. et al. Cross-tissue single-cell landscape of human monocytes and macrophages in health and disease. Immunity 54, 1883–1900.e5 (2021).
Cottone, L. et al. Inhibition of histone H3K27 demethylases inactivates brachyury (TBXT) and promotes chordoma cell death. Cancer Res. 80, 4540–4551 (2020).
Kruidenier, L. et al. A selective jumonji H3K27 demethylase inhibitor modulates the proinflammatory macrophage response. Nature 488, 404–408 (2012).
Hashizume, R. et al. Pharmacologic inhibition of histone demethylation as a therapy for pediatric brainstem glioma. Nat. Med. 20, 1394–1396 (2014).
Johansson, C. et al. Structural analysis of human KDM5B guides histone demethylase inhibitor development. Nat. Chem. Biol. 12, 539–545 (2016).
Mohammad, H. P. et al. A DNA hypomethylation signature predicts antitumor activity of LSD1 inhibitors in SCLC. Cancer Cell 28, 57–69 (2015).
Ito, T. et al. Paralog knockout profiling identifies DUSP4 and DUSP6 as a digenic dependence in MAPK pathway-driven cancers. Nat. Genet. 53, 1664–1672 (2021).
Eckschlager, T., Adam, V., Hrabeta, J., Figova, K. & Kizek, R. Metallothioneins and cancer. Curr. Protein Pept. Sci. 10, 360–375 (2009).
Cribbs, A. et al. Inhibition of histone H3K27 demethylases selectively modulates inflammatory phenotypes of natural killer cells. J. Biol. Chem. 293, 2422–2437 (2018).
Das, A. et al. RNA sequencing reveals resistance of TLR4 ligand-activated microglial cells to inflammation mediated by the selective jumonji H3K27 demethylase inhibitor. Sci. Rep. 7, 6554 (2017).
Roy, R., Winteringham, L. N., Lassmann, T. & Forrest, A. R. R. Expression levels of therapeutic targets as indicators of sensitivity to targeted therapeutics. Mol. Cancer Ther. 18, 2480–2489 (2019).
Menden, M. P. et al. Community assessment to advance computational prediction of cancer drug combinations in a pharmacogenomic screen. Nat. Commun. 10, 2674 (2019).
Tardito, S. et al. Copper binding agents acting as copper ionophores lead to caspase inhibition and paraptotic cell death in human cancer cells. J. Am. Chem. Soc. 133, 6235–6242 (2011).
Grasso, C. S. et al. Functionally defined therapeutic targets in diffuse intrinsic pontine glioma. Nat. Med. 21, 555–559 (2015).
Kim, S. ppcor: an R package for a fast calculation to semi-partial correlation coefficients. Commun. Stat. Appl. Methods 22, 665–674 (2015).
Cowley, G. S. et al. Parallel genome-scale loss of function screens in 216 cancer cell lines for the identification of context-specific genetic dependencies. Sci. Data 1, 140035 (2014).
Ritz, C., Baty, F., Streibig, J. C. & Gerhard, D. Dose-response analysis using R. PLoS ONE 10, e0146021 (2015).
Van der Borght, K. et al. BIGL: Biochemically Intuitive Generalized Loewe null model for prediction of the expected combined effect compatible with partial agonism and antagonism. Sci. Rep. 7, 17935 (2017).
Li, W. et al. MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens. Genome Biol. 15, 554 (2014).
Dančík, V. et al. Connecting small molecules with similar assay performance profiles leads to new biological hypotheses. J. Biomol. Screen. 19, 771–781 (2014).
Acknowledgements
We thank M. Kocak for analytical support and advice, and the PRISM team for generation of screening data. We thank K. Weiskopf for advice on human macrophage expression resources and S. Nelson for assistance with chemical characterization of KDOBA67. This work was supported by a grant from The Gerstner Family Foundation.
Author information
Authors and Affiliations
Contributions
M.G.R. and C.M.J. designed the study and wrote the manuscript. M.G.R., L.B., M.D.C. and P.D. acquired and analyzed cell viability data. B.B. and M.A. acquired and analyzed LC–MS data. S.J.F. synthesized KDOBA67. E.V., F.P. and J.G.D. acquired and/or analyzed genetic screening data. D.R. and T.S. acquired PRISM data and cell data for LC–MS experiments. A.B. analyzed cell viability data and PRISM data. M.G.R., B.B., T.A.L., V.K.K., F.P., D.E.R., J.G.D. and C.M.J. designed experiments and interpreted data.
Corresponding authors
Ethics declarations
Competing interests
C.M.J. is a full-time employee and stockholder of Novartis Institutes for Biomedical Research. D.E.R. receives research funding from members of the Functional Genomics Consortium (Abbvie, Bristol-Myers Squibb, Janssen, Merck and Vir), and is a director of Addgene. F.P. is a current employee of Merck Research Laboratories.
Peer review
Peer review information
Nature Chemical Biology thanks Liron Bar-Peled, Ultan McDermott, Jason Moffat and Kris Wood for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 Identification of resistance-associated transcript.
(a) Number of transcripts (of 19,174) passing a Bonferroni-adjusted- and z-score significance cutoff for at least one small molecule in at least one of 43 cellular sub-contexts (lineage, histology, growth mode, mutation) in the CTRP dataset and number of transcripts (of 17,737) passing a Bonferroni-adjusted- and z-score significance cutoff for at least one small molecule in at least one of 51 cellular sub-contexts (lineage, histology, growth mode, mutation) in the GDSC dataset. Low variation: transcripts with a range < 3 TPM (or RMA-normalized, for GDSC) units. (b) Number of compounds associated with each gene showing a significant correlation in both the CTRP and GDSC datasets. (c) Heatmap showing scaled relative expression across cell lines of most frequently associated genes from Fig. 1b. Heatmaps were generated using Morpheus (https://software.broadinstitute.org/morpheus), with columns clustered (metric: one minus Pearson correlation; linkage method: average). (d) Results from (a) after filtering for potential confounding factors such as gene co-expression and non-specific associations.
Extended Data Fig. 2 Validation of single-agent response and protein levels.
Sensitivity after 72 hours, as measured by CellTiter-Glo, of a panel of cell lines to each nominated compound, as well as Western blots to assess protein levels. Cell viability values (GR) are normalized to DMSO treatment and to initial (D0) cell number. mRNA expression of each nominated co-target is shown by color (red to blue), along with Pearson correlation coefficients (and associated two-sided P-values) between viability and expression. For Western blots, beta actin (ACTB) or vinculin (VCL) was used as a loading control as indicated to calculate relative protein levels and samples were ordered by their relative compound sensitivity. For (a-e), mRNA expression is also shown below the blots as bar graphs aligned to protein samples and as x-y scatterplots for direct comparison with calculated protein levels.
Extended Data Fig. 3 Cropped Western blots for ORF overexpression.
Vinculin (VCL), beta actin (ACTB), or glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were used as loading controls as indicated. For uncropped gel images, see Source Data.
Extended Data Fig. 4 Supporting data for ORF overexpression experiments across additional compounds and cellular models.
Effects of lentiviral overexpression of nominated transcripts (red) on compound sensitivity after 72 hours, as measured by CellTiter-Glo, and Western blots to assess protein levels after overexpression. Parental cell lines are shown in black, with control ORFs in blue. Cell viability values (GR) are normalized to DMSO treatment and to initial (D0) cell number. The difference in corrected Akaike Information Criteria (AICc) and associated probability of separate dose-response curves explaining the underlying data are shown. Each point represents the mean of 2 (d and doxorubicin/G402/BCL2L1), 4 (f) or 3 (all other panels) technical replicates.
Extended Data Fig. 5 Co-inhibition of candidate co-targets engenders synergistic responses.
Cell viability (CellTiter-Glo) on 72-hour co-treatment of compounds with vehicle (black) and increasing concentrations of co-target inhibitors (blue to red). The difference in AICc and associated probability of separate dose-response curves explaining the underlying data are shown. Cell lines are indicated in bold (lower-left of graph). Cell viability values (GR) are normalized to DMSO treatment and to initial (D0) cell number. Each point represents the mean of 2 (a-j) or 3 (k-l) technical replicates.
Extended Data Fig. 6 Synergy calculations for co-inhibition experiments where inhibition of the co-target showed significant concentration-dependent toxicity.
(a) Legend for (b–d) where combination effect (deviation from null model) is shown by color. (b–d) Deviation from Loewe additivity (synergy) for the combinations (b) navitoclax x S63845, (c) doxorubicin x WEHI-539, and (d) PRIMA-1 x erastin across a panel of cell lines. The overall F-statistic and two-sided P-value (from meanR test) are shown for each cell line.
Extended Data Fig. 7 Combinational drug treatment specifically induces synergy in cell lines with high synergistic target expression.
(a) Legend for (b-d, f-j). (b-d, f-i) Underlying data for Fig. 5. The degree of sensitization for each drug combination is shown relative to the cell line most sensitive to that compound. (e) Combination of canagliflozin with each of 8 compounds across 12 cell lines. Degree of sensitization is shown by shading. Degree of sensitization is defined as the magnitude of GR50 shift for combination versus single-agent treatment, normalized to the overall range of single-agent GR50 values across the entire cell-line panel. Basal mRNA levels for UGT1A10 across each cell line are shown as gray bars. (j) Degree of temozolomide sensitization (change in GR at the maximum temozolomide concentration tested) between vehicle co-treatment and 5 µM O6-benzylguanine co-treatment across an expanded panel of 17 cell lines. Two-sided p-values <0.05 for Pearson correlation between basal expression and degree of sensitization are shown.
Extended Data Fig. 8 Knockout of candidate co-targets sensitizes to nominated compounds.
(a) Effects of infection of MeWo cells with control or AIFM2-targeting sgRNAs on response to ML210 and AIFM2 protein levels. The difference in AICc (and associated probability) relative to sgctrl are shown. For GSK-J4, paclitaxel, and doxorubicin, all ΔAICc < −1 and P > 0.62. Points represent the mean of 2 technical replicates. (b) Effects of infection of WM2664 cells with control or IRS2-targeting sgRNAs on response to pictilisib and IRS2 protein levels. Cas9-expressing cells (no sgRNA) are shown in gray. The difference in AICc (and associated probability) relative to sgctrl are shown. Points represent the mean of 2 technical replicates. For Western blots, glyceraldehyde 3-phosphate dehydrogenase was used as a loading control.
Extended Data Fig. 9 MGLL activity and expression are sufficient for resistance to GSK-J4.
(a) Effects of N-acetylcysteine (red) or ferrostatin-1 (blue) on co-treatment with GSK-J4, ML210, or piperlongumine in G402 cells. Cell confluence values (GR) are normalized to DMSO treatment and to initial (D0) confluence. The difference in AICc and associated probability of separate dose-response curves explaining the underlying data are shown. Each point represents the mean of 2 technical replicates. (b) Effects of GSK-J4 or bortezomib on viability and caspase induction (as measured by caspase 3/7 reagent) in G402-HcRed or G402-MGLL cells. Each point represents the mean of 3 technical replicates. (c) Sensitivity after 72 hours, as measured by CellTiter-Glo, of a panel of cell lines to GSK-J4, GSK-J4 + 1 µM JZL184, or JZL184 alone. Cell viability values (GR) are normalized to DMSO treatment and to initial (D0) cell number. (d–i) Effects on GSK-J4 sensitivity of overexpression of MGLL in HEC1A cells, or infection with sgRNAs targeting MGLL (or control guides) in SUIT2 or HEC1A cells, with and without JZL184 co-treatment, as well as Western blots probing MGLL expression. Beta actin (ACTB) or vinculin (VCL) was used as a loading control as indicated. The difference in AICc (and associated probability) relative to sgctrl are shown. Each data point represents the mean of 2 (f) or 3 (g-h) technical replicates.
Extended Data Fig. 10 GSK-J4 is a substrate for MGLL and CES1, and sensitivity to GSK-J4 is affected by extracellular copper levels.
(a) Chemical structures of the canonical MGLL substrate 2-arachidonoylglycerol and products arachidonic acid and glycerol. (b) Western blot of G402 cells overexpressing HcRed, MGLLWT, MGLLS132A, or MGLLD249N ORFs. Vinculin (VCL) was used as loading control. (c-d) Peak areas from LCMS assessment of GSK-J4 and GSK-J1 incubated in the presence and absence of MGLL (independent experiment from Fig. 6a). Points represent the mean of 3 technical replicates. (e) Effects of co-treatment on sensitivity to GSK-J4 with selective inhibitors of MGLL (KML29), CES1 (WWL123, WWL229), or MGLL + CES1 (JZL184) in G402-HcRed, G402-MGLL, or THP1 cells as measured by CellTiter-Glo. Each point represents the mean of 3 technical replicates. (f) Sensitivity to KDOBA67 in G402, G402-HcRed, G402-MGLL, G402-MGLLS132A, or G402-MGLLD249N cells. Each point represents the mean of 2 technical replicates. (g) Western blots probing total histone H3, H3K27me2, and H3K27me3 levels in CJM and CJM-MGLL cells after 72-hour treatment with the perturbations and concentrations indicated. Vinculin (VCL) was used as a loading control. (h) Effects of DMSO, 10 µM CuCl2, or 10 µM ZnCl2 on GSK-J4 sensitivity in G402-HcRed cells as measured by live-cell imaging after 6 or 72 hours. (i) Effects of 10 µM CuCl2 on GSK-J4 sensitivity in G402-HcRed cells as measured by live-cell imaging. Cell confluence values are normalized to compound in the absence of CuCl2. Each point represents the mean (n = 2 technical replicates). (j) Effects of DMSO or 10 µM FeCl2 on GSK-J4 sensitivity in G402-HcRed cells as measured by live-cell imaging after 72 hours. (k) G402-MGLLWT or S132A cells were treated with 10 µM KDOBA67 or 5 µM GSK-J4 and co-treated with 10 µM CuCl2 either immediately or after 24 hours. Results are normalized to KDOBA67 or GSK-J4 treatment alone. (l) Effects of copper chelators TTM or BCS, the iron chelator DFX, or the zinc chelator TPEN on sensitivity to GSK-J4, KDOBA67, or elesclomol in G402 cells. Each point represents mean ± SD (n = 2 technical replicates). For (e-f, h, j, l), cell confluence values (GR) are normalized to DMSO treatment and to initial (D0) confluence and the difference in AICc and associated probability of separate dose-response curves explaining the underlying data are shown.
Supplementary information
Supplementary Information
Supplementary Figs. 1–6, Dataset legends 1–7, Tables 1–5, and Note.
Supplementary Data
Supplementary Datasets 1–7.
Source data
Source Data Fig. 4
Underlying source data.
Source Data Fig. 5
Statistical source data.
Source Data Fig. 6
Statistical source data.
Source Data Extended Data Fig. 2
Unprocessed western blots.
Source Data Extended Data Fig. 3
Unprocessed western blots.
Source Data Extended Data Fig. 8
Unprocessed western blots.
Source Data Extended Data Fig. 9
Unprocessed western blots.
Source Data Extended Data Fig. 9
Underlying source data.
Source Data Extended Data Fig. 10
Unprocessed western blots.
Rights and permissions
About this article
Cite this article
Rees, M.G., Brenan, L., do Carmo, M. et al. Systematic identification of biomarker-driven drug combinations to overcome resistance. Nat Chem Biol 18, 615–624 (2022). https://doi.org/10.1038/s41589-022-00996-7
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41589-022-00996-7
This article is cited by
-
Predicting drug response from single-cell expression profiles of tumours
BMC Medicine (2023)
