Identification of the molecular mechanism of action (MoA) of bioactive compounds is a crucial step for drug development but remains a challenging task despite recent advances in technology. In this study, we applied multidimensional proteomics, sensitivity correlation analysis, and transcriptomics to identify a common MoA for the anticancer compounds RITA, aminoflavone (AF), and oncrasin-1 (Onc-1). Global thermal proteome profiling revealed that the three compounds target mRNA processing and transcription, thereby attacking a cancer vulnerability, transcriptional addiction. This led to the preferential loss of expression of oncogenes involved in PDGF, EGFR, VEGF, insulin/IGF/MAPKK, FGF, Hedgehog, TGFβ, and PI3K signaling pathways. Increased reactive oxygen species level in cancer cells was a prerequisite for targeting the mRNA transcription machinery, thus conferring cancer selectivity to these compounds. Furthermore, DNA repair factors involved in homologous recombination were among the most prominently repressed proteins. In cancer patient samples, RITA, AF, and Onc-1 sensitized to poly(ADP-ribose) polymerase inhibitors both in vitro and ex vivo. These findings might pave a way for new synthetic lethal combination therapies.
Significance: These findings highlight agents that target transcriptional addiction in cancer cells and suggest combination treatments that target RNA processing and DNA repair pathways simultaneously as effective cancer therapies.
Comprehensive assessment of mechanism of action (MoA) and drug–protein interactions in living cells constitutes a major challenge for rationalizing a drug's therapeutic and adverse effects. Multiple strategies for target deconvolution have been developed, but many of them include compound modification/labeling (1). Recently developed thermal proteome profiling (TPP) enables such an assessment in a physiologically relevant cellular environment without compound modification and at a proteome-wide level. TPP is based on the principle that the heat-induced unfolding of proteins changes upon binding to ligand. Using quantitative mass spectrometry–based proteomics, TPP measures ligand-induced thermal stabilization or destabilization of proteins (2). The changes in melting proteome reflect direct drug–protein binding but also the changes in posttranslational modifications and protein–protein interactions. Therefore, the shift in thermal stability of proteins could be used not only for the prediction of direct drug targets, but also for an unbiased identification of molecular MoA.
Cancer cells are subject to multiple stresses, including genotoxic, proteotoxic, replicative, metabolic, and oxidative stresses. Tumor cells adapt to detrimental stressful conditions by dysregulated gene expression. This creates a dependence on continuous active transcription of oncogenes (3). Transcriptional addiction emerged as a potential “Achilles' heel” of cancer cells that can be therapeutically exploited. However, targeting transcription for selective elimination of cancer cells has been considered difficult due to potential toxicity for normal tissues. The development of CDK7 and CDK12/13 inhibitors has challenged this paradigm: although acting on numerous genes, they exert a highly specific effect on gene expression (3). These compounds selectively affect cancer cells, leading to a preferential loss of oncogenic gene expression, along with DNA damage response genes (4). Attacking addiction to a single oncogene is the basis of targeted therapies (5), but invariably leads to emergence of resistance clones due to cancer cells' plasticity. Therefore, targeting addiction to multiple oncogenes, either with a single compound or combinatorial treatments, is an attractive anticancer strategy to limit acquired resistance.
In this study, we applied multidimensional approaches, including TPP, to identify the MoA and pathways affected by selective anticancer compounds RITA, aminoflavone AF, and oncrasin-1 (Onc-1) and investigated how these new findings might provide opportunities for novel combination therapies.
Materials and Methods
MCF7, T47D, A375, U2OS, OVCAR-3, 184A1, MCF10A, and BJ-hTERT were obtained from ATCC. BJ-hTert HRasV12ER-Tam cells were a kind gift from Reuven Agami (Netherlands Cancer Institute, Amsterdam, The Netherland). U2OS cMycER-Tam cells were obtained from Martin Eilers (University of Würzburg, Würzburg, Germany). CRISPR/Cas9-mediated p53 deletion was performed in MCF7 and A375 cells using sgRNA-targeting TP53 exon 4. Cell lines were authenticated by ATCC and Mycoplasma contamination was tested monthly using MycoAlert Mycoplasma Detection Kit (Lonza) according to the manufacturer's instructions. All experiments were performed within 20 passages from frozen stocks. Further details can be found in Supplementary Materials and Methods and Supplementary Table S1.
Experiments on primary cancer cells from patients with breast cancer and ascites fluid from patients with metastatic high-grade serous ovarian cancer (HGSOC) were approved by the Regional Ethical Board of Stockholm (Dnr 2016/957-31, 2017/742-32, and 2016/1197-31/1) in compliance with the declaration of Helsinki. Samples were taken from patients who signed written informed consent. Detailed information about isolation procedure and treatments can be found in Supplementary Materials and Methods.
Compounds and treatments
RITA (NSC652287) and aminoflavone (NSC686288) were obtained from the NCI, while oncrasin-1 was from Santa Cruz Biotechnology. Unless stated otherwise, cells were treated in complete medium with 1 μmol/L RITA, 1 μmol/L AF, or 5 μmol/L Onc-1. α-amanitin, ActD, BMH21, camptothecin, cycloheximide, irinotecan, MG132, nordihydroguaiaretic acid, Nutlin-3, and resveratrol were purchased from Sigma-Aldrich; cisplatin, doxorubicin, olaparib, and talazoparib from Selleckchem; and THZ531 from ApexBio Technology. For combination experiment with PARP inhibitors, cells were pretreated 2 days with olaparib or talazoparib and then treated with the combination for the indicated time.
All antibodies used in the study are detailed in Supplementary Materials and Methods.
Total RNA extraction and cDNA synthesis were performed using Aurum total RNA and iScript cDNA Synthesis Kits (Bio-Rad) according to the manufacturer's instructions. mRNA quantification was performed by qRT-PCR using Sso Advanced Universal SYBRGreen SuperMix (Bio-Rad). GAPDH and RPL13A were used as housekeeping genes. Error bars represent SD from mean of at least three independent experiments. The sequences of qPCR primers used are detailed in Supplementary Table S2.
Correlation sensitivity analysis
NCI-60 analysis tools provided by CellMiner (https://discover.nci.nih.gov/cellminer/; ref. 6) and the NCI-60 cell line panel with associated drug screens were used to calculate the drug sensitivity Pearson correlation coefficient between RITA, AF, and Onc-1, as well as the correlation between RITA, DNA-damaging agents, and topoisomerase-1 inhibitors. The 50% growth inhibitory concentration (GI50) values of drugs were used for calculations.
MCF7 cells were treated with 1 μmol/L RITA, 1 μmol/L AF, or 5 μmol/L Onc-1 for 2 hours (RITA and AF) or 6 hours (Onc-1) in biological duplicates. TPP was performed as described previously (7). Briefly, cells from each biological replicate were split equally into 10 parts; each aliquot was heated at a different temperature point ranging from 37°C to 67°C. Soluble proteins were then analyzed by quantitative LC/MS-MS. For each protein, the melting temperature (Tm) was calculated and proteins with |ΔTm| ≥ 0.5°C and P < 0.05 were considered as hits for subsequent analyses. Sample processing and data analysis are detailed in Supplementary Materials and Methods, and the complete list of TPP hits for each compound treatment in MCF7 cells can be found in Supplementary Table S3.
Gene ontology (GO) statistical overrepresentation analysis was performed using PANTHER (Protein ANalysis THrough Evolutionary Relationships 13.1 (http://pantherdb.org; ref. 8) against all human genes. PANTHER GO-Slim biological processes enriched in TPP hits with Fisher adjusted P < 0.01 were considered as affected by each treatment. The complete list of GO-Slim biological processes enriched in TPP hits can be found in Supplementary Table S4. Functional protein interaction network analysis was performed using STRING 10.5 (http://string-db.org; ref. 9). The STRING network was built with all proteins affected by at least one treatment. Only interactions with the highest score (>0.900) from experiments or curated databases were considered. The three main clusters of the network were identified by k-means clustering, and each cluster was analyzed for GO statistical overrepresentation using PANTHER to identify its biological function.
Expression profiling of MCF7 cells treated with 1 μmol/L RITA, 10 μmol/L Nutlin (8 hours), or with 1 μmol/L AF (8 hours), 5 μmol/L Onc-1 (16 hours) was performed using GeneChip HG-U219 or HT HG-U133+ PM 24-Array Plate, respectively. The datasets generated in this study are available in Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/) under accession number GSE86880 and GSE128329, and in Supplementary Table S5. Data analysis is detailed in Supplementary Materials and Methods.
Cell viability assays
Cell viability was assessed by resazurin assay. Briefly, cells were treated for 24–72 hours and incubated for 2 hours with 5 μmol/L resazurin (Sigma-Aldrich) before fluorescence measurement on a VICTOR X5 plate reader (PerkinElmer). For crystal violet staining experiments, cells were treated for 1–5 days, fixed with ethanol, and stained with 0.2% crystal violet. Quantification was performed using ImageJ software.
Reactive oxygen species measurement
Cells were treated as indicated for 8 hours, washed, and incubated for 30 minutes with 10 μmol/L DCFDA in serum-free medium. Then cells were trypsinized, washed twice with PBS, and fluorescence was analyzed by a FACSCalibur Flow Cytometer (BD Biosciences).
Cells on glass coverslips were treated for the indicated time, and 1 mmol/L 5-ethynyluridine (EU) was added to the medium 1 hour before cell fixation. EU was labeled using the Click-iT RNA Alexa Fluor 594 Imaging Kit (Thermo Fisher Scientific) according to the manufacturer's instructions.
Homologous recombination reporter assay
Homologous recombination (HR) reporter assay was performed in U2OS cells harboring stably integrated reporter construct for HR (DR-GFP) as described previously (10). To generate a DSB, U2OS DR-GFP cells were transfected with I-SceI expression vector pCMV3xnls-I-SceI (1 μg) using Fugene HD (Promega). Twenty-four hours after transfection, cells were treated with RITA, AF, or Onc-1 for 16 hours and GFP fluorescence due to HR of I-SceI cleavage site was monitored by flow cytometry.
Unless otherwise stated, statistical significance was calculated using two-tailed Student t test.
Similarity of phenotypes induced in cancer cells by RITA, AF, and Onc-1
RITA (reactivation of p53 and induction of tumor cells apoptosis) was identified by us in a screen for molecules that selectively kill cells expressing wild-type (wt) p53. RITA is a potent anticancer agent in vitro and in vivo and is highly selective for cancer, but not normal cells (11, 12). Using several pairs of isogenic cancer cells with wtp53/knockout (KO) p53, we found that the contribution of p53 to cell death upon RITA is cell type–dependent (Supplementary Fig. S1A–S1J). While in melanoma A375 cells, RITA effects were completely p53-dependent (Supplementary Fig. S1I and S1J), in MCF7 cells the contribution of p53 to the effects of RITA was quite weak, although detectable (Supplementary Fig. S1A–S1H), which is in-line with recently published data (13).
To address the p53-independent MoA of RITA, we took advantage of ProTargetMiner database, containing proteome profiles for 56 anticancer drugs in A549 lung cancer cells (14). Because protein abundances depend not only on mRNA levels, but also on translation rate and protein stability, proteome response can be highly specific to drug MoA. We compared the proteome signature upon RITA treatment to 20 other anticancer compounds. Unsupervised clustering revealed the similarity of RITA and antimetabolic drugs raltitrexed and methotrexate profiles (Fig. 1A). However, RITA profile did not cluster with DNA-damaging compounds such as bleomycin or doxorubicin, suggesting that its MoA is different from that of DNA-damaging drugs.
Next, we searched the NCI-60 cancer cell lines pharmacology database for compounds eliciting a sensitivity profile similar to RITA. Consistent with our chemical proteomics analysis, RITA sensitivity profile did not correlate with those of DNA-damaging drugs (such as cisplatin and oxaliplatin), topoisomerase inhibitors (camptothecin, irinotecan, and topotecan), or reactive oxygen species (ROS)-inducing compounds (doxorubicin; Supplementary Fig. S2A). We also compared transcriptional changes induced by RITA treatment with those upon cisplatin, doxorubicin, camptothecin, irinotecan, or Nutlin-3 in MCF7 cells (Supplementary Fig. S2B). While the response of p53-activated genes PUMA (BBC3), NOXA (PMAIP1), and p21 (CDKN1A) was similar, the repression of oncogenes MDM2, PPM1D, and MCL1 occurred only upon RITA treatment. These results were confirmed at protein level (Supplementary Fig. S2C). Furthermore, comparison of genome-wide expression upon RITA, Nutlin-3, camptothecin, cisplatin, and doxorubicin revealed a very distinct transcriptional profile upon RITA treatment (Supplementary Fig. S2D). Taken together, our analysis of proteomics, transcriptomics, and cancer cell sensitivity profiles upon RITA revealed a response different from that induced by DNA-damaging compounds, ROS inducers, or topoisomerase inhibitors.
Notably, sensitivity correlation analysis of NCI-60 database revealed a significant correlation between the sensitivity of cancer cells to RITA and the two anticancer compounds AF and Onc-1 (R2 = 0.5938 and R2 = 0.6466, respectively; Fig. 1B and C).
AF [5-amino-2-(4-amino-3-fluorophenyl)-6,8-difluoro-7-methylchromen-4-one; NSC 686288] displays an antiproliferative and proapoptotic activity in tumor cells (15). It induces replication-dependent DNA lesions and S-phase–dependent apoptosis (16), although its MoA has not been fully elucidated. Onc-1 (1-[(4-chlorophenyl)methyl]-1H-indole-3-carboxaldehyde) and its derivative NSC-743380, identified in a screen for selective targeting of mutant K-Ras cancer cells (17), exhibit antitumor activity in several tumors types (18). Onc-1 MoA remains unclear, but involves interference with RNA polymerase II (RNA Pol II) transcription and STAT3 inhibition (19, 20).
Cancer selectivity is crucial for the success of anticancer compounds. Therefore, we compared the effect of these compounds on the growth of normal and cancer cells (Fig. 1D). RITA, AF, and Onc-1 efficiently reduced cell viability in breast cancer cells MCF7 and T47D, whereas they had a limited effect on normal mammary epithelial cells 184A1 and MCF10A and nontransformed BJ-hTert fibroblasts.
Furthermore, we found that the treatment of MCF7 cells with AF or Onc-1 phenocopied RITA effects on apoptosis and DNA damage response, although the kinetics was slightly different for Onc-1 (Fig. 1E; Supplementary Fig. S3A). The three compounds activated the JNK pathway (Fig. 1E), in-line with previous studies (19, 21, 22). We found previously that RITA inhibits the antioxidant enzyme thioredoxin reductase 1 (TrxR1; ref. 23), increasing intracellular ROS level. Similarly, AF and Onc-1 induced a strong oxidative stress in different cell lines (Fig. 1F; Supplementary Fig. S3B).
Because growth suppression by RITA depends on the downregulation of several crucial oncogenes (12, 22), we tested whether AF and Onc-1 have similar effects. Indeed, MDM2, PPM1D, MCL1, and MDMX were inhibited on mRNA and protein level upon treatment with all three compounds, along with the induction of proapoptotic PUMA and NOXA (Fig. 2A and B; Supplementary Fig. S3C). Interestingly, while p21 mRNA was induced, its protein level was downregulated by AF and Onc-1, similar to the effects we have shown previously for RITA (24). Finally, as RITA, AF, and Onc-1 all stabilize p53 in MCF7 cells (Fig. 1E), we performed viability assays in p53 KO MCF7 and showed that apoptosis caused by all three compounds was largely p53-independent in these cells (Supplementary Fig. S3D).
Furthermore, we observed a remarkably similar expression profile upon RITA, AF, and Onc-1, and identified 3,154 common differentially expressed genes (DEG) in MCF7 cells (Fig. 2C and D). Strikingly, oncogenes were overrepresented among common repressed genes (Fig. 2E). Analysis of gene expression data revealed significant downregulation of oncogenes representing major oncogenic pathways, such as PDGF, EGFR, VEGF, insulin/IGF/MAPKK, FGF, Hedgehog, TGF-β, and PI3K signaling pathways (Fig. 2F and G; Supplementary Fig. S3E).
TPP identifies RNA processing as the common cell process affected by RITA, AF, and Onc-1
Because RITA, AF, and Onc-1 induced a highly similar cell response, we hypothesized that they could have a common molecular target, and/or target a common pathway. To identify pathways affected by the three compounds, we performed TPP in MCF7 cells. Identifying a direct drug target with TPP is challenging, as many variations in protein interactions measured in the cells could be secondary, indirect events. Instead, the proteome-wide variations in protein interactions provide important clues as to which pathways are affected by a drug. Therefore, we considered the changes of interactions observed upon treatment as a reflection of the key affected biological processes.
To minimize the indirect events, we analyzed proteome changes at early time points (2 hours for RITA and AF, and 6 hours for Onc-1). We identified 144, 182, and 188 hits, defined as proteins with a significant thermal shift |ΔTm| ≥ 0.5°C upon treatment with RITA, Onc-1, and AF, respectively (Fig. 3A and B), with a prominent overlap among them (Fig. 3B; Supplementary Fig. S4A). We identified 90 hits common for two treatments and eight hits common for all three treatments. We performed a GO analysis using PANTHER to identify biological processes affected (Fig. 3C). Notably, most of the GO biological processes terms enriched in TPP hits were related to RNA processing, including splicing and RNA metabolism. The overlap between the GO biological processes terms was striking, with RNA processing being the common process affected by all three treatments (Fig. 3D). Mitochondrial translation was also significantly affected upon AF treatment, but not upon RITA and Onc-1.
As TPP detects the changes of interactions for a given protein, other proteins interacting with a TPP hit are likely to be also affected. To have a more integrated view of the changes in protein complexes involved, we generated a protein–protein interaction network using STRING to visualize the functional interactions affected by RITA, AF, and Onc-1 (Fig. 3E; Supplementary Fig. S4B). Using unbiased k-means clustering, we identified three main clusters of interactions in the network (clusters A, B, and C), containing 34, 20, and 16 proteins, respectively. The proportion of proteins in the clusters affected by each treatment is shown in Fig. 3F. To identify the biological function of each cluster, we performed a GO analysis (Fig. 3G). Cluster A represents proteins involved in RNA processing, including multiple splicing factors and auxiliary splicing regulatory proteins (MAGOH, RBM8A, DHX38, SF1, SF3B1, SF3B2, SF3B4, RBM22, SUGP1, PTPB1, HNRNPM, and SNRPC), RNA Pol II subunits (POLR2A, POLR2L, and POLR2K), transcriptional elongation regulators (RDBP), and polyadenylation factors (CPSF1, CSTF1, and PAPOLA). Importantly, this cluster was affected evenly by the three treatments. Cluster C has a less defined biological function, including proteasome subunits, cell-cycle regulators (CCND1 and CDK2), and DNA damage checkpoints factors (BLM and RAD50).
Cluster B contains mostly mitochondrial ribosomal proteins affected by AF treatment, although some (such as MRPL10) were also affected by RITA and Onc-1. Using mitochondrial membrane permeabilization assay, we found that only AF caused a small, but significant change in the mitochondrial membrane potential (Supplementary Fig. S4C). We also assessed mitochondrial morphology by electron microscopy but did not observe changes upon treatment (Supplementary Fig. S4D).
In summary, TPP data assessed by both GO and network analysis approaches revealed RNA processing as a common pathway affected by the three compounds.
RITA, AF, and Onc-1 inhibit transcription in cancer cells, but not in normal cells
To investigate the effect of RITA, AF, and Onc-1 on RNA processing, we visualized the incorporation of EU. As shown in Fig. 4A and Supplementary Fig. S5A, RITA, AF, and Onc-1 completely abolished RNA synthesis, similar to known RNA Pol II inhibitors actinomycin D (ActD) and α-amanitin. Notably, the levels of total, phospho-S5, and phospho-S2 (elongating) RNA Pol II strongly decreased upon treatment in different cancer lines (Fig. 4B; Supplementary Fig. S5B and S5C). RNA Pol II level was rescued by the proteasome inhibitor MG132, suggesting proteasomal degradation after stalling (Fig. 4C). Accordingly, cycloheximide chase experiment showed a clear decrease in RNA Pol II half-life upon RITA treatment (Fig. 4D). Then, we investigated whether the inhibition of RNA synthesis was dependent on p53 activation by RITA, AF, and Onc-1. RNA synthesis was also abolished in p53 KO MCF7 cells (Supplementary Fig. S5D).
Moreover, comparison of expression profiles upon ActD and CDK12/CDK13 inhibitor THZ531 revealed similarities with the transcriptional response to RITA, AF, and Onc-1, even though different cancer cell lines were used (Supplementary Fig. S5E and S5F). We confirmed that the oncogenes involved in RITA-induced cell death were also downregulated at protein level by THZ531 and ActD treatment (Supplementary Fig. S5G).
Strikingly, RITA, AF, and Onc-1 did not prevent EU incorporation and RNA synthesis in normal human fibroblasts (Fig. 4E). In contrast, ActD and α-amanitin completely shut down transcription in nontransformed cells as well. Thus, the inhibition of transcription by RITA, AF, and Onc-1 occurs selectively in cancer cells, contrary to other RNA Pol II inhibitors.
Several splicing/RNA-binding factors were among the top common TPP hits, including CIRBP and RBM8A for all three compounds and SUGP1, SF3B2, and GRSF1 for any two of them. Therefore, we reasoned that RITA, AF, and Onc-1 might affect RNA splicing. Indeed, all three compounds induced the formation of enlarged sc35 speckle domains (splicing factor compartments, SFC), which were previously described upon RNA Pol II or splicing inhibition (Fig. 4F; Supplementary Fig. S5A; ref. 25).
Not only RNA Pol II but also Pol I was inhibited by the compounds, as evidenced by shuttling of the RNA helicase DDX21 from the nucleoli to the nucleoplasm (Fig. 4G), which is known to occur upon RNA Pol I inhibition (26). Consistently, protein level of POLR1A, the main subunit of RNA Pol I, was decreased upon treatment to the same extent as upon specific Pol I inhibitor, BMH21 (Fig. 4H). Finally, we compared the inhibition of both Pol II and Pol I by testing the expression of RNA Pol II- and RNA Pol I–regulated RNAs, MYC mRNA, and 45S preribosomal RNA, respectively, upon a broad range of compounds' concentrations in MCF7 cells (Fig. 4I). Expression of both genes was substantially decreased by the three compounds, indicating the block of both RNA polymerases. To exclude the effect of p53-mediated MYC repression (27), we used MCF7 p53KO cells and got a similar result (Supplementary Fig. S5H). RNA Pol I inhibition (i.e., upon low doses of ActD) and nucleolar stress activate p53 through the release of MDM2-binding protein, RPL11 (28), therefore we tested whether RITA had the same effect (Supplementary Fig. S5I). SiRNA-mediated RPL11 knockdown rescued p53 stabilization by 5 nmol/L ActD but not upon RITA, demonstrating that nucleolar stress is not involved in p53 activation upon RITA.
FACS analysis of synchronized cells using bromodeoxyuridine/propidium iodide staining showed that RITA-induced cell death occurs mainly in early S-phase (Supplementary Fig. S6A), suggesting replication stress, in-line with previous study (29). Thus, we addressed the question whether RNA Pol II inhibition was a consequence of replication stalling. Using cells synchronized in early G1-phase by lovastatin and treated with RITA before they entered S-phase, we found that the inhibition of transcription occurred in G1-phase and was therefore independent of replication stress. Furthermore, we demonstrated that early S-phase arrest and replication stress (as manifested by γH2AX induction) induced by double thymidine block were not sufficient to inhibit transcription (Supplementary Fig. S6B and S6C).
To investigate the mechanism of RNA polymerase inhibition, we tested whether RITA, AF, and Onc-1 can intercalate DNA by performing a thiazole orange displacement assay. No DNA intercalation was observed at doses 12–60 times higher than those used in cells (Supplementary Fig. S7A), demonstrating that these compounds are not DNA intercalating agents. Interestingly, gene set enrichment analysis of our microarray data revealed an enrichment of UV response genes among the DEGs induced by all three compounds (Supplementary Fig. S7B). This correlates with the UV-like (30), pan-nuclear γH2AX staining, different from the discrete nuclear foci associated with DNA breaks (Supplementary Fig. S7C). To investigate the possibility that Pol II stalling might be caused by DNA adducts, we performed ASCC2 foci formation analysis in MCF7 cells, a marker for cell response to DNA alkylation (31). Although we indeed observed the formation of ASCC2 foci (Supplementary Fig. S7D and S7E), they were induced to a lesser extent than foci formed upon low dose of alkylating agent, methyl methanesulfonate.
Altogether, our data confirm the results obtained via TPP analysis and demonstrate that RITA, AF, and Onc-1 inhibit RNA Pol I- and Pol II–mediated transcription and interfere with RNA processing such as splicing.
RITA, AF, and Onc-1 cancer-specific effects are dependent on oxidative stress
As discussed previously, RITA, AF, and Onc-1 induce a strong oxidative stress (Fig. 1F). This was associated with oxidative DNA damage as assessed by immunostaining with 8-oxo-7,8-dihydro-2′-deoxyguanosine (8-oxodG) antibody (Supplementary Fig. S8A). Oxidative stress–dependent growth suppression has been previously described for all three compounds (22, 32, 33) and we confirmed in our cell models that two antioxidants, resveratrol and nordihydroguaiaretic acid, rescue different cancer lines from apoptosis and growth suppression induced by RITA, AF, and Onc-1 (Fig. 5A and B; Supplementary Fig. S8B and S8C). Notably, antioxidants also prevented RNA Pol II stalling and degradation in a p53-independent way (Fig. 5C; Supplementary Fig. S8D), as well as inhibition of prosurvival genes (Fig. 5D and E; Supplementary Fig. S8E).
As cancer cells have increased ROS levels and therefore they are more sensitive to further oxidative stress (34), we hypothesized that induction of ROS could be the reason for cancer selectivity of our compounds. To test this idea, we used a model of increased ROS upon HRasV12ER-TAM induction in BJ-hTert fibroblasts. We found that RITA, AF, and Onc-1 suppressed the growth of fibroblasts upon HRasV12ER-TAM activation by tamoxifen (Fig. 5F) followed by the increase of the basal ROS level (Supplementary Fig. S8F). Moreover, we observed a significant downregulation of RNA Pol II CTD phosphorylation only upon HRasV12 induction (Fig. 5G).
Taken together, our results suggest that the cancer selectivity of transcription inhibition and growth suppression by the three compounds is dependent on oncogene-induced oxidative stress.
RITA, AF, and Onc-1 inhibit homologous recombination–mediated DNA repair
It has been suggested that DNA repair factors are among the most effected genes upon inhibition of transcription elongation (4). Indeed, upon treatments the DNA repair factors CtIP, RPA32, WRAP53, RNF168, RNF8, and RAD51 (Supplementary Fig. S9A) were downregulated, the HR-associated protein RAD51 and its partners RNF8 and RNF168 being the most prominently decreased (Fig. 6A; Supplementary Fig. S9B). The observed decrease of these proteins was associated with transcriptional repression (Fig. 6B), but not with protein stability, as measured by cycloheximide chase experiment (Supplementary Fig. S9C).
We confirmed that HR was impaired upon treatment by RITA, AF, and Onc-1 using U2OS cells harboring stably integrated reporter construct for HR (DR-GFP), where repair of I-SceI–induced break by HR leads to a functional GFP gene. As shown in Fig. 6C, treatment with RITA, AF, or Onc-1 leads to a stable decrease of GFP-positive cells, indicating impairment of the HR repair system.
Overexpression of RAD51, RNF8, and RNF168 did not rescue RNA Pol II inhibition or downregulation of prosurvival genes (Fig. 6D; Supplementary Fig. S9D), but we observed a partial rescue of γH2AX when overexpressing the three repair factors simultaneously, indicating that inhibition of RAD51 and its partners plays a role in the replication stress observed upon RITA (Fig. 6D). However, the rescue of γH2AX was not sufficient to increase cell viability (Supplementary Fig. S9E), highlighting the importance of the downregulation of prosurvival factors for apoptosis induction.
RITA, AF, and Onc-1 sensitize to PARP inhibition
As DNA repair, and more specifically HR, is impaired because of the transcription inhibition of HR factors by RITA, AF, and Onc-1, we hypothesized that these compounds could synergize with PARP inhibition to efficiently kill cancer cells. Therefore, we used sublethal doses of RITA, AF, and Onc-1 in combination with two PARP inhibitors, olaparib (FDA approved) and talazoparib (currently in clinical trials). Remarkably, we observed synthetic lethality with both inhibitors in MCF7 (Fig. 6E and F; Supplementary Fig. S9F) and OVCAR-3 cells (Supplementary Fig. S9G). We tested the combination ex vivo in primary ascites–derived cells from patients with HGSOC and observed similar effects (Fig. 6G), demonstrating the therapeutic potential of a combination of RITA, AF, or Onc-1 with PARP inhibition both in vitro and ex vivo. Furthermore, we observed synthetic lethality of RITA combination with PARP inhibitors in samples derived from patients with breast cancer in 3D conditions ex vivo. Notably, we found that the beneficial effect of this combination was evident in patient samples with high Ki67 index and tumor grade, that is, highly malignant tumors (Fig. 6H). Similarly, synthetic lethality was significantly enhanced upon activation of oncogene in cMyc-inducible U2OS cells (U2OS cMycER-Tam; Supplementary Fig. S9H). These findings might provide opportunities for novel therapeutic interventions in cancer.
Identification of small molecules' MoA has always been a challenge in drug development. It became particularly important with the rise of personalized medicine and targeted therapies. TPP has been developed as a novel approach for the identification of small molecules' targets. However, finding the correct target among the TPP hits is challenging given the number of proteins with thermal stability shifts due to secondary events. TPP is also limited to soluble proteins; additional restrictions inherent to quantitative proteomics may lead to a number of false negatives. For example, p53 and TrxR1, two previously described direct binding targets of RITA (11, 23), were not found in our TPP screen. However, our study demonstrates how comprehensive analysis of the global modulation of protein interactions at the whole-proteome level by TPP can help to decipher therapeutic MoA.
Our TPP analysis shows that mRNA maturation machinery is significantly affected by RITA, AF, and Onc-1 treatment. Applying GO biological processes terms and a more integrated view of protein–protein interaction network revealed a striking overlap between the three treatments, with RNA processing being the common process. Validation of these findings confirmed the inhibition of RNA Pol I- and Pol II–mediated transcription selectively in cancer cells. This transcription blockade leads to the loss of expression of dozens of vital oncogenes, involved in major oncogenic pathways, such as PDGF, EGFR, VEGF, insulin/IGF/MAPKK, FGF, Hedgehog, TGFβ, and PI3K signaling pathways.
Targeting transcription for anticancer therapy has been considered to be too toxic to normal cells. However, recent studies have challenged this view by showing that the expression of oncogenes required for the survival of cancer cells is highly dependent on uninterrupted active transcription. For instance, blocking RNA Pol II function affects more c-Myc–transformed cells than their normal counterparts (35). This “transcriptional addiction” of cancer cells creates a therapeutic window allowing selective anticancer growth suppression (3). Our results are in-line with these ideas. We show that inhibition of transcription in cancer cells by RITA, AF, and Onc-1 leads to the preferential loss of a crucial set of oncogenes and therefore target the “Achilles' heel” of cancer cells, exerting a highly selective effect on cancer, but not normal cells.
How to achieve selectivity for tumor cells without affecting normal tissues is one of the main challenges for anticancer drug discovery. One of the cancer cells vulnerabilities is a high basal level of ROS, leading to the dependence on high expression of their antioxidant systems. Therefore, compounds increasing the oxidative stress may induce cancer cell death while sparing normal cells, which have lower ROS levels and therefore can cope with ROS increase. Indeed, it has been shown that normal cells keep a low level of ROS upon exposure to RITA (23). RITA, AF, and Onc-1 all increase oxidative stress in cells and different antioxidants can prevent their cytotoxic effect (22, 32, 33). Moreover, head and neck cancer cells with high level of antioxidant systems are more resistant to RITA (36). Here, we show that high level of intracellular ROS is important not only for apoptosis induced by RITA, AF, and Onc-1 but also for their activity as RNA Pol inhibitors. We detected oxidative DNA damage manifested as 8-oxodG upon RITA, AF, and Onc-1, in-line with the report about AF (32). Importantly, a widespread transcriptional silencing is induced upon base excision repair (BER) of oxidative DNA damage (37). Taken together, these data suggest that the inhibition of transcription by the three compounds could be caused by BER of oxidized DNA bases. In addition, we cannot exclude the possibility that the three compounds may cause nucleotide imbalance, for instance, through oxidation of guanine, which might lead to a disruption of RNA synthesis. This hypothesis will be investigated in the future.
Overall, our data suggest that RITA, AF, and Onc-1 high selectivity for cancer cells can be explained by their simultaneous targeting of multiple cancer vulnerabilities, including oncogene addiction, increased ROS, and replication stress (as summarized in Supplementary Fig. S10). Our study paves a way to rational treatment designs with RITA, AF, and Onc-1. We found that DNA repair factors, including HR factors RAD51, RNF8, and RNF168 are strongly repressed upon RITA, AF, and Onc-1, similar to CDK12/CDK13 inhibitors (4). HR-deficient cells, such as BRCA2-mutant cells, are known to be particularly sensitive to PARP inhibitors (38). Thus, compounds such as RITA, AF, and Onc-1, which simultaneously cause replication stress and inhibit HR, are very attractive for combination therapy with PARP inhibitors to achieve synthetic lethality in cancer cells with functional HR.
Disclosure of Potential Conflicts of Interest
T. Helleday is a SAB (paid consultant) at Sierra Oncology, has ownership interest (including patents) in PARP inhibitor sales (royalty), Shareholder Oxcia AB, and has an unpaid consultant/advisory board relationship with Chair Helleday Foundation. No potential conflicts of interest were disclosed by the other authors.
Conception and design: S. Peuget, R. Zubarev, G. Selivanova
Development of methodology: M. Singh, L. Moyano-Galceran, K. Lehti, R. Zubarev, G. Selivanova
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Peuget, J. Zhu, M. Singh, M. Gaetani, X. Chen, A.A. Saei, T. Visnes, M.S. Lindström, L. Moyano-Galceran, J.W. Carlson, E. Hjerpe, U. Joneborg, K. Lehti, J. Hartman, R. Zubarev
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Peuget, J. Zhu, G. Sanz, M. Singh, M. Gaetani, X. Chen, Y. Shi, A.A. Saei, A. Rihani, E. Hjerpe, G. Selivanova
Writing, review, and/or revision of the manuscript: S. Peuget, G. Sanz, M. Singh, M. Gaetani, T. Visnes, E. Hjerpe, U. Joneborg, J. Hartman, T. Helleday, G. Selivanova
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Peuget, X. Chen, R. Zubarev, G. Selivanova
Study supervision: S. Peuget, T. Helleday, G. Selivanova
We thank all our colleagues for sharing precious reagents and material with us, especially Prof. J. Bartek (Danish Cancer Society Research Center and Karolinska Institute, Stockholm, Sweden) and Dr. M. Farnebo (Karolinska Institute, Stockholm, Sweden). This study was supported by the Swedish Cancer Society, the Swedish Research Council, and Knut and Alice Wallenberg Foundation. S. Peuget has been supported by a fellowship from Wenner Gren Foundation.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.