A multitargeted probe-based strategy to identify signaling vulnerabilities in cancers

Most cancer cells are dependent on a network of deregulated signaling pathways for survival and are insensitive, or rapidly evolve resistance, to selective inhibitors aimed at a single target. For these reasons, drugs that target more than one protein (polypharmacology) can be clinically advantageous. The discovery of useful polypharmacology remains serendipitous and is challenging to characterize and validate. In this study, we developed a non-genetic strategy for the identification of pathways that drive cancer cell proliferation and represent exploitable signaling vulnerabilities. Our approach is based on using a multitargeted kinase inhibitor, SM1-71, as a tool compound to identify combinations of targets whose simultaneous inhibition elicits a potent cytotoxic effect. As a proof of concept, we applied this approach to a KRAS-dependent non-small cell lung cancer (NSCLC) cell line, H23-KRASG12C. Using a combination of phenotypic screens, signaling analyses, and kinase inhibitors, we found that dual inhibition of MEK1/2 and insulin-like growth factor 1 receptor (IGF1R)/insulin receptor (INSR) is critical for blocking proliferation in cells. Our work supports the value of multitargeted tool compounds with well-validated polypharmacology and target space as tools to discover kinase dependences in cancer. We propose that the strategy described here is complementary to existing genetics-based approaches, generalizable to other systems, and enabling for future mechanistic and translational studies of polypharmacology in the context of signaling vulnerabilities in cancers.

Over the last two decades cancer treatment has been revolutionized by a targeted approach to therapy in which a selective agent is developed to hit a single, specific target. Prominent examples of targeted therapies include selective kinase inhibitors that target BCR-ABL in chronic myelogenous leukemia or mutant EGFR 3 and EML4-ALK in nonsmall cell lung cancer (NSCLC) (1). Unfortunately, in some tumor types, this approach is limited by the rapid emergence of drug resistance; in other cancers with multiple or conventionally undruggable driver mutations, targeted approaches can be hard to apply. Many tumors are genetically heterogeneous, harboring multiple genomic alterations in different combinations, which results in signaling plasticity and rapid evolvability. These properties of tumors have generated interest in developing drugs capable of simultaneously inhibiting multiple signaling pathways. Such inhibitors, often described as "polypharmacological agents" have long been utilized therapeutically in CNS diseases, infection, inflammatory diseases, and psychiatric disorders where selective inhibitors have failed (2,3). Examples of approved polypharmacological drugs include acetyl salicylic acid, paracetamol, clozapine, etc., which act by binding and interacting with several proteins, thereby exerting pharmacological effects that cannot be ascribed to a single molecular target (2).
In the field of kinase inhibitors, many approved drugs were developed with a particular target in mind but are now known to be multitargeted. Their multitargeting properties are a consequence of their binding mode. Kinase inhibitors have been developed to bind in conserved ATP-binding pockets and therefore have cross-reactivity toward other kinases with shared structural features in their active sites. The multitargeted nature of most approved kinase inhibitors has now been confirmed and characterized through kinome-wide profiling technologies (4 -10). The polypharmacology of kinase inhibitors has resulted in the discovery of new indications for particular compounds. For example, imatinib was initially developed as a BCR-ABL inhibitor for the treatment of chronic myelogenous leukemia, but its activity against c-KIT/PDGFR allowed it to be a successful drug for the treatment of gastrointestinal stromal tumors (11,12). Similarly, crizotinib was developed as a MET inhibitor, but its activity against ALK resulted in its approval for the treatment of EML4-ALK-positive NSCLC. In several cases, kinase inhibitor polypharmacology has been shown to be important for anticancer activity. For example, the ability of ibrutinib to simultaneously inhibit BTK and HCK makes it a superior drug for treatment of Waldenström's macroglobulinemia as compared with highly selective BTK inhibitors (13). Similarly, sorafenib, originally developed as a BRAF inhibitor, is now known to target numerous other kinases, including VEGFR, PDGFR, RET, DDR1/2, and FLT3 (4), leading to its later approval as a multikinase inhibitor for renal cell carcinoma and hepatocellular carcinoma (14). Similarly, rationally designed inhibitors that simultaneously target bromodomains and kinases have shown superior potency compared with a single target inhibition (15). A recent report suggests that Ͼ60% of Food and Drug Administration-approved kinase inhibitors, including sorafenib, dasatinib, pazopanib, and ponatinib, exert their mechanism in a multitargeted manner by targeting at least three or more kinases (16).
With a better understanding of molecular mechanisms governing tumor growth and progression, efforts are underway to rationally design drugs with precision polypharmacology. The most promising efforts combine structure-based analysis with medicinal chemistry campaigns to identify pharmacophores that potently inhibit two or more kinases. Using this approach, Apsel et al. (17) developed inhibitors that simultaneously target PI3K and tyrosine kinases to overcome resistance mediated by activation of one or the other signaling kinases. In a study that combined phenotypic and target-based drug discovery approaches, Dar et al. (18) identified inhibitors with polypharmacological profiles that exerted potent activity in a RET-kinase driven Drosophila model bearing multiple endocrine neoplasia 2.
A major challenge in rationally designing cancer drugs with polypharmacology is to identify the subset of kinases that must be simultaneously inhibited to induce potent antiproliferative effects in a particular tumor type. One way to address this is to conduct systematic phenotypic screens using drug combinations and/or gene knockout techniques (19 -24). This approach is complicated by the difficulty of achieving simultaneous knockdown or knockout of multiple targets in a single cell (such multigene knockouts are often lethal). In this study, we demonstrate an alternate strategy that uses a multitargeted kinase inhibitor, SM1-71, with well-characterized polypharmacology as a chemical tool to investigate signaling vulnerabilities in cancer cells. As a proof of concept, we explored signaling vulnerabilities in a KRAS mutant NSCLC cell line, H23-KRAS G12C , and demonstrated that dual inhibition of MEK1/2 and IGF1R/INSR is required for antiproliferative activity in these cells. Our work provides a framework for leveraging a multitargeted kinase inhibitor with known polypharmacology to identify key signal-ing pathways driving tumor cells. This further lays the path for development of active compounds with desired polypharmacology or effective combination therapies.

Investigating the cytotoxic effect of SM1-71 across multiple cancer cell lines
SM1-71 is a diaminopyrimidine kinase inhibitor that potently targets kinases both through reversible binding in the ATPbinding site and irreversible binding promoted by reaction of the SM1-71 acrylamide moiety with cysteine resides (25,26) (Fig. 1a). We synthesized the reversible analog of SM1-71, SM1-71-R, which lacks the acrylamide warhead and is thus incapable of forming covalent bonds, as a control compound for our studies of cellular effects of SM1-71 (Fig. 1a). We previously used chemical proteomic approaches to elucidate ϳ54 kinase targets of SM1-71 (45) (Table S1) and identified 24 kinases as exhibiting an IC 50 value Ͻ10 M ( Table 1) Table 2 for genotypes). For comparative purposes, we profiled investigational and clinically approved kinase inhibitors against their described nominal targets, including PI3K, MEK1/2, ERK1/2, EGFR, BRAF, ALK, MET, and IGFR1 (Table 3). Cells were plated and 24 h later treated with varying doses of compounds for a period of 72 h. The CellTiter-Glo reagent was added to the plates, which were then analyzed for cytostatic or cytotoxic effects potentially induced by the drugs. To overcome confounding effects of varying division rates between cell lines on estimates of drug potency and efficacy, we used our recently developed growth rate (GR)-corrected values (27,28). We use GR 50 as a measure of potency (analogous to IC 50 ) and GR max as a measure of maximal efficacy (analogous to E max ; File S1). A GR max value between 1 and 0 corresponds to partial growth inhibition, a value of 0 indicates complete cytostasis, and a negative value denotes cell killing (27). The GR values reported in Table S2 were computed using the online GR Calculator (http://www.grcalculator.org 4 ; see "Experimental procedures and Ref. 29).

Elucidating kinases responsible for mediating cytotoxic effects in KRAS mutant cells
From our growth inhibitory screen, we identified several cancer cell lines with different genetic backgrounds that were highly sensitive to SM1-71 while showing resistance to inhibitors designed to target single kinases. To demonstrate that SM1-71 serves as an effective multitargeted chemical tool com-

Table 1 List of kinases inhibited by SM1-71 (IC 50 value <10 M) in the multiplexed inhibitor bead (MIB) assay and their role in promoting proliferation
Kinases were identified and reported in Rao et al. (45).

YES1
We observed a concomitant increase in p-MEK1/2 S217/S221 levels after 2 and 4 h postwashout, which is common following ERK1/2 inhibition as a consequence of disrupting negative feedback regulation (32). We observed p-AKT S473 and p-ERK1/ 2 T202/Y204 levels rising 2 and 4 h postwashout with SM1-71 and SM1-71-R. Several reports have previously shown that kinases within the MAPK and PI3K pathways are reactivated in response to specific inhibitors that suppress negative feedback loops (33)(34)(35). We thus predict a similar phenomenon responsible for the reactivation of p-AKT S473 and p-ERK1/2 T202/Y204 signaling upon treatment with SM1-71 and SM1-71-R. So far, signaling analysis using Western blotting demonstrated direct cellular inhibition of MAPK and PI3K signaling pathways. Our data also revealed inhibition of p-AKT S473 (PI3K pathway), which was likely affected through inhibition of an upstream receptor and not directly by targeting PI3K, AKT, or mTOR (nontargets of SM1-71; Table S1). To test this possibility, we profiled a panel of 49 receptor tyrosine kinases (RTKs) using an RTK array (R&D Systems) in which phosphorylation of RTKs (and their inhibition in the presence of SM1-71) was measured by exposing cell lysates to capture antibodies spotted in duplicates (per RTK) on a nitrocellulose membrane (see Fig.  S2 for dot-blots from the two independent experiments). H23-KRAS G12C cells were treated with 1 M SM1-71 or DMSO for 6 h and lysed after which the lysate was incubated with the RTK arrays. Phosphorylation signals were quantified for both the SM1-71-and DMSO-treated samples using the dot-blot analyzer (ImageJ software), and -fold change was calculated. These -fold change values were averaged across two independent experiments to generate an average -fold change signal (ϮS.E.) for each RTK. These average -fold change values for IGF1R (80-fold), INSR (12-fold), and MET (5-fold) have been plotted as bar graphs (Fig. 2b) (p Ͻ 0.0001, compared with INSR and MET -fold change). Our results indicate that among the 49 RTKs profiled, SM1-71 potently inhibited IGF1R, INSR, and MET. We conclude that SM1-71 is active on at least three RTKs known to lie upstream of the PI3K signaling pathway. Furthermore, we identified each of these three RTKs, IGF1R, INSR, and MET, as direct targets of SM1-71 from our previous study (Table S1) (45).

Validation of key targets driving proliferation in H23-KRAS G12C cells
To determine whether inhibition of IGF1R/INSR and/or MET is involved in down-regulation of p-AKT S473 levels, we attempted to phenocopy the effects using combinations of kinase inhibitors. The effects of 1 M SM1-71 were compared with those of an ALK/MET inhibitor (1 M crizotinib), IGF1R inhibitor (AEW541), ERK1/2 inhibitor (SCH772984), pan-PI3K inhibitor (BKM120), or DMSO. H23-KRAS G12C cells were incubated with the compound for 4 h, and phosphorylation of downstream kinases was assessed using Western blotting ( Fig. 2c; see Fig. S3 for blots from two independent experiments). We found that crizotinib reduced p-MET Y1234/1235 phosphorylation to background levels, partially reduced p-AKT S473 levels but had no discernable effect on p-ERK1/ 2 T202/Y204 levels. AEW541 reduced p-IGF1R/p-INSR Y1135/1136 levels and caused complete inhibition of pAKT S473 , also with no effect on pERK1/2 T202/Y204 (Fig. 2c). This inhibition of IGF1R/ INSR and/or MET in H23-KRAS G12C cells can down-regulate the PI3K pathway without affecting the activity of the MAPK pathway. In contrast, SM1-71 reduced not only p-IGF1R/p-INSR Y1135/1136 and p-MET Y1234/1235 levels but also p-ERK1/  . We therefore asked whether inhibition of MEK1/2 in combination with IGF1R/INSR or MET would recapitulate the cytotoxicity observed with SM1-71. We found that AZD6244 (MEK1/2 inhibitor) and AEW541 were weakly cytostatic on their own (GR 50 ϭ 0.5 M; GR max , between 0 and 1) but when combined were 5-fold more potent (GR 50 ϭ 0.08 M) and also cytotoxic as indicated by a GR max value of Ϫ0.4 ( Fig. 2d and File S1). As previously mentioned, a negative GR max value is indicative of cytotoxicity. Moreover, the MEK-IGFR1 inhibitor combination of AEW541 plus AZD6244, MEK-ERK-PI3K triple-inhibitor combination of AZD6244 plus SCH772984 plus BKM120, and SM1-71 were all similar in potency and cytotoxicity. In contrast, the MEK-MET inhibitor combination of AZD6244 plus crizotinib was only weakly cytotoxic ( Fig. 2e and File S1). Based on these data, we propose that MEK1/2 and IGF1R/INSR are critical drivers of growth and proliferation in H23-KRAS G12C cells. Furthermore, these are targeted by SM1-71, which results in inhibition of proliferation and induction of cell death.

Distinct molecular mechanisms drive different tumor types
The multitargeted nature of SM1-71 makes it a valuable tool to interrogate cancer cell signaling across different cell lines and tumor types. Having investigated pathways responsible for driving growth in a sensitive cell line (H23-KRAS G12C ), we wished to further apply SM1-71 to understand what might be mediating resistance in some other cell lines. From our growth inhibitory screen, we identified H3122, H460, and MDA-MB-453 cells to be slightly more resistant (submicromolar/micromolar GR 50 values compared with nanomolar values for sensitive cell lines) to the action of SM1-71. H3122 is an NSCLC line harboring an EML4-ALK translocation in which exposure to ceritinib, an ALK inhibitor, strongly inhibits proliferation (GR 50 ϭ 0.05 M, GR max ϭ Ϫ0.78; Fig. 3a and Table S2). However, SM1-71 binds only weakly to the ALK oncogenic driver (data not shown). The H460 NSCLC cell line and the MDA-MB-453 TNBC cell line both harbor E545K and H1074R mutations in the PIK3CA gene (https://cancer.sanger.ac.uk/cosmic). 4 This EDITORS' PICK: Deciphering cancer signaling vulnerabilities mutation introduces an oncogenic driver downstream of the RTKs such as IGF1R/INSR, which activates the PI3K pathway. We therefore predicted that a pan-PI3K inhibitor such as BKM120 would make these cell lines sensitive to SM1-71. We found that the combination of SM1-71 plus BKM120 reduced the GR 50 for SM1-71 Ͼ3-fold in both cell lines, suggesting that resistance is possibly a consequence of PIK3CA mutation (Fig.  3, b and c). Of note, both HCT116 and H1975 harbor H1074R and G118D PIK3CA mutations, respectively, and were sensitive to SM1-71. This suggests that the PIK3CA mutation is not a sufficient oncogenic driver to confer SM1-71 resistance. Such cell line-specific differences are observed with many kinase inhibitors and arise from the specific signaling biology of the lines (36,37). However, because SM1-71 has multiple targets, we cannot fully exclude other mechanisms that might be con-tributing toward the overall sensitivity and resistance effects of the compound.

Discussion
Several challenges, including complex signaling networks, cross-talk with the tumor microenvironment, and onset of drug resistance, are associated with treating cancers driven by multiple oncogenes. It is thus increasingly appreciated that only use of drugs that affect multiple signaling nodes will result in strong antiproliferative effects and delay the onset of drug resistance. Such compounds, referred to as polypharmacological drugs, have been investigated in the past to treat polygenic diseases such as cancers, CNS disorders, and inflammatory disorders. Our current work aims to promote the rational development of such inhibitors by developing means to unravel key signaling EDITORS' PICK: Deciphering cancer signaling vulnerabilities pathways that must be simultaneously inhibited to achieve maximum antitumor effects. By utilizing a multitargeted kinase inhibitor, SM1-71, that serves as an effective tool compound we were able to elucidate molecular mechanisms driving specific cancer cell types. In our previous study, we used SM1-71 to interrogate the human kinome to identify cysteines that can be targeted because of the covalent nature of the compound (45). In the present study, we extended the use of this compound to the investigation of signaling pathways that drive cellular proliferation in cancer cells. As a proof of concept, we demonstrated the utility of this chemical tool by using a KRAS mutant NSCLC cell line, H23-KRAS G12C , and identified MEK1/2 and IGF1R/INSR as being key players in driving cellular proliferation. We corroborated our findings by using clinical and investigational kinase inhibitors to induce pharmacologic shutdown of kinases. Our results validated that dual inhibition of MEK1/2 and IGF1R/INSR led to down-regulation of the MAPK and PI3K pathways, which is responsible for inducing potent cytotoxic effects in H23-KRAS G12C cells. Our findings are further supported by the knowledge that mutant KRAS leads to constitutive activation of its downstream effector pathways, MAPK and PI3K, and thus, inhibiting both arms of the oncogene leads to potent antitumor effects (38 -40). Furthermore, preclinical and clinical studies focusing on the dual inhibition of MEK1/2 and IGF1R/INSR have demonstrated beneficial effects across different types of cancers (41)(42)(43)(44). Our study does not unequivocally prove the pharmacologically relevant targets of SM1-71; indeed, the functionally relevant targets of this compound are likely different in different cell types.
It is well established that for drugs that target a single oncogene (e.g. EGFR, BCR-ABL, BTK, etc.), gene knockout (e.g. RNAi or CRISPR-Cas9) or rescue experiments following sitedirected mutagenesis that prevents the drug from binding the kinase are powerful techniques to functionally validate the ontarget effect of the drug. However, in the case of polypharmacological agents acting on polygenic tumors, similar methods pose many challenges (3). An alternative strategy is to use combinations of selective inhibitors targeting kinases that phenocopy the effects induced by a polypharmacological inhibitor. Given the multitargeted properties of SM1-71, we adapted a similar strategy to pharmacologically validate potential targets.
Our study also draws attention to the benefits of using a multitargeted chemical probe compared with a selective chemical probe. Traditionally, most chemical probes are designed to retain selectivity and specificity toward a single target to characterize its role in a given cell type. However, in our current study, we demonstrate that chemical probes with multiple targets can serve as powerful tools to interrogate oncogenic drivers in cancer cells in lieu of systematic combinatorial screens. By using a combination of cell-based assays and pharmacologic inhibitors, we outlined a framework for adopting multitargeted kinase inhibitors with defined polypharmacology as effective chemical probes. As a proof of concept, we conducted all our analyses in the sensitive H23-KRAS G12C cells; however, a similar strategy can be adopted to interrogate other cell types. In fact, we demonstrated this by extending our evaluation toward three cell lines that were relatively resistant to SM1-71 in the growth inhibitory screen. Further investigation is required to ascertain these mechanisms as causes for the lowered activity of SM1-71 in these less sensitive cell lines. Nonetheless, direct pharmacologic inhibition of targets provides compelling evidence toward the mechanism of cytotoxicity in both sensitive and resistant cells. Together, these illustrate the generalizability of this approach and imply that such probes can serve as effective means to elucidate signaling pathways driving tumor cells.
Polypharmacological agents are also associated with several limitations when it involves improving their properties through systematic medicinal chemistry efforts. It can be extremely difficult to hone potency and selectivity toward two or more desired targets while simultaneously achieving selectivity and the desired pharmacokinetic and toxicological profile. A further challenge relates to finding suitable experimental models that can faithfully predict response in the clinic. For example, a simple cell proliferation assay may be able to predict response to a BCR-ABL inhibitor but may be a poor predictor for a kinase inhibitor such as SM1-71 that inhibits through multiple targets. Specifically, in the case of SM1-71, the ability of the compound to block PI3K pathway signaling through RTK inhibition may not be accurately modeled in simple cancer cell proliferation assays with unnatural levels of growth factors. Given that polypharmacological kinase inhibitors will continue to be discovered, we propose that these multitargeted inhibitors can serve as effective research tools to help unravel pathways that must be targeted. Their application for path finding and vulnerability identification will allow the development of rational combinations as well as potentially aide rational design of tailored polypharmacology agents.

Chemical synthesis
Detailed methodology on the synthesis of compounds has been described previously (25).

Growth inhibition assay
Cells were plated in 384-well plates (3764, Corning) at a seeding density of 2000 cells/well using the Multidrop Combi Reagent Dispenser (Thermo Fisher Scientific). Cells were treated with different doses of compounds for 72 h using the automated HP-D300 digital dispenser and normalized to 0.2% DMSO 24 h postplating. Viability was measured 72 h after treatment by adding 25 l/well CellTiter-Glo (G7572, Promega) and reading plates using the Synergy H1 microplate reader (BioTek). Each experiment was carried out in technical triplicate and biological duplicate.

Growth inhibition analysis using the online GR calculator
Growth inhibition across cell lines was analyzed using the online GR calculator (http://www.grcalculator.org/grcalculator) 4 developed by members of the LINCS-BD2K Data Coordination and Integration Center, Harvard Medical School (29), which is based on the method originally described by Hafner et al. (27). The GR metrics (GR 50 and GR max ) along with their corresponding IC 50 values were calculated using the online GR calculator based on cell division rates obtained for different cell lines. To measure cell doubling time, cells were seeded at an initial count of 500,000 in a 10-cm dish, and final count was measured 3-4 days (72-96 h) later using the TC20 automated cell counter (Bio-Rad). The doubling time was calculated using the following formula: DT ϭ T ln(2)/ln(Xe/Xb) where T is the time of growth (hours), Xe is the final cell count, and Xb is the initial cell count. The following doubling times were used: A375 (19. . Data from the CellTiter-Glo assay along with cell division rates were uploaded onto the online GR calculator that generated dose-response curves and GR 50 and GR max values along with all other statistical parameters that can be found in File S1. Codes used in the analysis can be made available upon request.

RTK array
H23-KRAS G12C cells were grown in 10-cm dishes, treated with 1 M SM1-71 or DMSO for 6 h, washed, extracted, and lysed as described under "Western blotting." Following protein quantification, 500 g of protein/sample was used to carry out the phospho-RTK array analysis according to the protocol described by the manufacturer (ARY001B, R&D Systems). Henceforth, all incubation and wash steps were accompanied by end-to-end rocking. Briefly, each array was incubated with 2 ml of Array Buffer 1 for 1 h at room temperature. After removal of this blocking buffer, 500 g of sample diluted in 1.5 ml of Array Buffer 1 was added to each array and incubated overnight at 4°C. Following washes with 1ϫ Wash Buffer, arrays were incubated with 2 ml of anti-phosphotyrosine-HRP antibody diluted in Array Buffer 2 for 2 h at room temperature. Wash steps were repeated, and arrays were visualized by adding a 1:1 ratio of the SuperSignal West Dura Extended Duration Substrate and scanning them using the myECL imager. Phosphorylation signals obtained were mapped to their respective RTKs using the reference RTK coordinates included in the kit (blots for both experiments are shown in Fig. S2). The experiment was repeated twice, and the phosphorylation signals for each RTK were quantified using the dot-blot analyzer macro in the ImageJ 1.50i software (code and documentation are available upon request). Each phospho-RTK signal had two representative spots on a given array. Fold change was calculated for SM1-71-EDITORS' PICK: Deciphering cancer signaling vulnerabilities and DMSO-treated samples for each RTK and averaged across the two independent experiments. These fold change values ϮS.E. were plotted as bar graphs for IGF1R, INSR, and MET. The GraphPad Prism 7.0 software was used to generate graphs and carry out statistical analysis.

Statistical analysis
All statistical analyses were carried out using GraphPad Prism 7.0 software. In Figs. 1c and 2b, statistical analysis was carried out using one-way ANOVA. In Fig. 1d, statistical significance between the sensitive and resistant cell lines was calculated using the two-tailed unpaired t test (p ϭ 0.0005).