ERK Inhibitor LY3214996-Based Treatment Strategies for RAS-Driven Lung Cancer

This article is featured in Highlights of This Issue, p. 623

Non–small cell lung cancers (NSCLCs) with aberrations in the rat sarcoma oncogene family (H-, N-, KRAS) represent 30% of all lung tumors (1). RAS mutations activate multiple downstream signaling pathways, among them MAPK and PI3K signaling both of which converge on the Cyclin D1/CDK4/6-RB axis (2, 3). Besides this direct mechanistic link to mutant RAS, RB signaling is activated by co-occurring genetic events such as loss of CDKN2A (encoding for p16/p14ARF) or amplification of CDK4 or CCND1 (encoding for Cyclin D1; refs. 4–6). These effector pathways play pivotal roles for cell cycle progression, proliferation, and apoptotic resistance of cancer cells (7). Yet, abemaciclib—a CDK4/6 inhibitor—had only limited single-agent activity in clinical trials (8, 9) and MEK inhibitors administered on an uninterrupted schedule exhibited toxicity and poor antitumor activity in patients with lung cancer (10, 11). Ultimately, MEK and PI3K inhibitor combinations caused significant toxicity in humans (2, 6, 12, 13).

Despite these setbacks, it remains an attractive strategy to inhibit RAS-dependent effector pathways, because direct KRASG12C and immune checkpoint inhibitor (ICI) activity is limited to subgroups of patients with RAS-mutant cancers (14–16). ERK1/2 proteins exert a bottleneck function in activating cytoplasmic and nuclear targets, and therefore ERK inhibitors are considered to be more potent in preventing bypass and feedback activation than MEK inhibitors (17–19). So far, however, only few ERK inhibitors have been tested in clinical trials for solid tumors despite demonstrating efficacy in preclinical models of solid tumors and in BRAF and/or MEK inhibitor-refractory patients with melanoma (20–24).

In this study, we used models of RAS-mutant lung cancer derived from patients pretreated with multiple treatment modalities (including chemo ± radiotherapy and ICIs) and from genetically engineered mouse models (GEMM) to investigate the efficacy of the novel ERK1/2 inhibitor LY3214996 alone and in combination with a PI3K/mTOR (LY3023414; ref. 25) or CDK4/6 inhibitor (abemaciclib; ref. 26). LY3214996 is effective in preclinical models across several cancer types (27, 28) and currently being evaluated in phase I trials as monotherapy or in combination treatments for patients with advanced cancer (NCT02857270; ref. 29).

Detailed information is presented in Supplementary Materials and Methods.

Patient-derived cell lines were generated from malignant pleural effusions or ascites collected under a Dana-Farber Cancer Institute IRB approved protocol. All patients provided written informed consent and the studies were conducted in accordance with the declaration of Helsinki. A targeted next-generation sequencing (NGS) cancer genomic assay (“Dana-Farber Cancer Institute (DFCI) OncoPanel”) was used to detect cell line specific somatic mutations, copy number variations, and structural variants in tumor DNA. All cell lines were kept at 37°C in complete media supplemented with 10% FBS. DFCI168 (30), DFCI316, DFCI366, DFCI516 cells were grown in RPMI1640 media, DFCI24, DFCI298, and DFCI332 in ACL4 media. Murine cell lines were derived from previously described genetically engineered mouse models and grown in DMEM (GEMMs; refs. 31, 32). All cell lines tested negative for mycoplasma throughout the study, were last tested on October 8, 2020, and used until passage 25 for functional assays. LY3214996 (example No. 1; ref. 33), abemaciclib (example No. 1; ref. 34), and LY3023414 (example No. 1; ref. 35) were discovered at the Lilly Research Laboratories and synthesized as described in the respective patents. Selumetinib/AZD6244 (36), SCH772984 (37), afatinib/BIBW2992 (38), and linsitinib/OSI-906 (39) were purchased from SelleckChem. Cells were lysed in RIPA buffer supplemented with protease and phosphatase inhibitors. The list of antibodies used can be found in Supplementary Materials and Methods. Western blot band intensities were quantified with ImageJ. All in vitro experiments were performed under adherent cell culture condition and live cell imaging was performed with an IncuCyte ZOOM real-time imaging system. Bliss synergy was determined and visualized with Combenefit (40). The Firefly-luciferase expressing FHRE-Luc reporter plasmid was a gift from Michael Greenberg (Addgene plasmid No. 1789; http://n2t.net/addgene:1289; RRID:Addgene_1789). A renilla luciferase-expressing pRL-CMV control plasmid was used for normalization. Luciferase activities were quantified with the Dual-Glo Luciferase Assay system. Nuclear-cytoplasmic fractionation was performed with the NE-PER Nuclear and Cytoplasmic Extraction Reagents. For patient-derived xenograft (PDX) models, female NSG (NOD SCID Gamma) and NCr nude mice (Taconic Biosciences), were housed and treated in accordance with protocols approved by the Dana-Farber Cancer Institute Animal Care and Use Committee. Tumor growth delay was calculated as described previously (41). MAPK pathway-dependent gene expression was quantified with a previously published 6-gene signature (DUSP4, DUSP6, ETV4, ETV5, SPRY2, PHLDA1; ref. 42). A P value less than 0.05 was considered statistically significant for all datasets. “*” indicates P < 0.05, “**” P < 0.01, “***” P < 0.001, and “****” P < 0.0001. Statistical analyses were performed with GraphPad Prism 8 and SAS (Version 9.3).

We sought to establish cancer cell lines reflecting the clinical distribution of RAS mutations among patients with lung cancer. In the TCGA Pan-Lung cancer dataset, about 30% of tumors exhibit aberrations in the RAS gene family including activating gene mutations (Supplementary Fig. S1). KRAS is overall more frequently affected (23%) than N- (2.6%) or HRAS (1.7%; ref. 43). Six out of seven cancer cell lines from patients who had previously undergone treatment for metastatic RAS-mutant lung cancer (Fig. 1A: bright-field microscopy images; Supplementary Fig. S2: patient histories) had KRAS mutations, one (DFCI168) had an NRAS mutation (30), and none had HRAS mutations. All patient-derived cell lines were genetically characterized by “OncoPanel” next-generation sequencing (NGS) at DFCI (44). We observed a high degree of concordance of genetic events between the available initial tumor biopsies and the established cell lines (Supplementary Table S1). Initially, we determined the absolute inhibitory concentrations (IC₅₀ values) of single-agent LY3214996 after 72 hours. DFCI168^NRASQ61K (1.1 μmol/L) and DFCI516^KRASG12C (1.5 μmol/L) exhibited the highest, DFCI316^KRASQ61H (3.6 μmol/L), DFCI24^KRASG12C (4.9 μmol/L), and DFCI366^KRASG12D (9.1 μmol/L) intermediate, and DFCI298^KRASG12C and DFCI332^KRASG12D the lowest sensitivity (>10 μmol/L) to LY3214996 (Fig. 1B). Despite a paradoxical dose-dependent increase in pERK1/2^{Thr202/Tyr204} and pMEK1/2^Ser217/221 following treatment with LY3214996, signaling downstream of pERK1/2 remained occluded as indicated by potent inhibition of phosphorylation of p90RSK^{Thr359/Ser363} and S6 ribosomal protein^Ser217/221 as well as reduction of c-MYC, DUSP4, and SPRY2 protein levels (Fig. 1C). We next established that compensatory ERK1/2 phosphorylation, which was not induced in response to the structurally different ERK inhibitor SCH772984 (45) could be prevented by selumetinib (MEK inhibitor) pretreatment (Supplementary Fig. S3). Afatinib (pan-ErbB inhibitor) and linsitinib (IGFR inhibitor) had no effect on early ERK1/2 phosphorylation excluding a major involvement of these upstream RTKs in early ERK phosphorylation. Importantly, ERK target inhibition with LY3214996 (1 μmol/L) increased proportionally to treatment duration as evidenced by decreasing effector levels over 48 hours (Fig. 1D). Pathway inhibition was accompanied by stronger accumulation of pro-apoptotic BIM in sensitive (DFCI168^NRASQ61K) and intermediate sensitive cell lines (DFCI316^KRASQ61H), which also exhibited more prominent PARP cleavage. LY3214996 treatment activated compensatory PI3K signaling (pAKT^Ser473) over baseline in all cell lines to varying extents (most evident in resistant cell lines and DFCI316^KRASQ61H cells).

Subsequently, we quantified the expression of MAPK pathway-dependent effector genes during LY3214996 (1 μmol/L) treatment, using a previously published MAPK pathway activation gene set (42) and observed a decrease in mean pathway-dependent transcription in all cell lines over 48 hours (Fig. 1E). In DFCI24 and DFCI332 cells, mean expression levels increased slightly after 48 hours over levels observed at 24 hours but did not reach levels of untreated cells. Despite similar pathway suppression, dose-dependent inhibition of cancer cell proliferation with LY3214996 differed in cell lines over 72 hours and DFCI322^KRASG12D cells were left unaffected with drug concentrations of up to 10 μmol/L (Fig. 1F). To investigate potential mechanisms of differential drug responses, we performed FOXO3a luciferase reporter assays because FOXO3a nuclear translocation and transcriptional activity has been shown to influence MEK inhibitor sensitivity (46). The more sensitive cell lines (DFCI168, DFCI24) exhibited increased FOXO3a reporter activity after 24 hours of LY3214996 (1 μmol/L) treatment, whereas reporter activity remained unchanged in resistant cell lines (DFCI366, DFCI298, DFCI332) and was reduced in intermediate sensitive DFCI316 cells (Fig. 1G). Cytoplasmic-nuclear fractionation experiments corroborated increased nuclear accumulation of FOXO3a in sensitive DFCI168 cells whereas FOXO3a levels were left unaffected in resistant DFCI332 cells (Fig. 1H).

To test the in vivo single agent activity, we treated DFCI168^NRASQ61K tumor-bearing NSG mice with LY3214996 (100 mg/kg, orally every day) and observed a significant reduction in tumor growth compared with tumors of vehicle-treated animals (Fig. 2A, n = 8 animals/group, P < 0.0001; two-way repeated measures ANOVA with standard post hoc t tests), which translated into a survival benefit despite treatment cessation after 21 days (median survival time: 49 days vs. 29.5 days, Fig. 2B, P = 0.0003, log-rank test). Pharmacodynamic analyses (PD) demonstrated inhibition of ERK targets at 4 hours, including p90RSK^{Thr359/Ser363}—a reliable PD biomarker for the extent of ERK inhibition (22)—of S6 ribosomal protein^Ser235/236 and reduction of total c-MYC, DUSP4, and SPRY2 protein levels (Fig. 2C). However, MAPK pathway activity recovered by 16 and 24 hours posttreatment. These pharmacodynamic effects correlated with plasma concentrations of LY3214996. At 4 hours, the LY3214996 plasma concentration was 3,840 ± 654 nmol/L, resulting in 66% inhibition of pRSK. As the plasma concentrations declined at 16 (189 ± 267 nmol/L) and 24 hours (9 nmol/L), pRSK inhibition declined to 33% and 34%, respectively (Fig. 2D). AKT^Ser473 phosphorylation slightly increased at 4 hours indicating PI3K pathway activation and gradually decreased again over the next 20 hours. Consistent with strong MAPK pathway inhibition at 4 hours, we detected strong transcriptional suppression of individual genes within the 6-gene signature (60%–80%; Supplementary Fig. S4A), and of the mean overall ERK transcriptional output (∼60%), whereas gene expression increased slightly over baseline after 16 and 24 hours (Fig. 2E).

Despite good antitumor activity in DFCI168 tumors, LY3214996 did not affect growth of DFCI316^KRASQ61H tumors when dosed once every day (100 mg/kg, Fig. 2F; two-way repeated measures ANOVA with standard post hoc t tests) despite similar suppression of ERK-dependent gene expression as in DFCI168^NRASQ61K tumors (Fig. 2G; Supplementary Fig. S4B). Interestingly, in contrast to LY3214996-sensitive DFCI168^NRASQ61K tumors, LY3214996-resistant DFCI316^KRASQ61H tumors exhibited markedly increased ERK^{Thr202/Tyr204} and AKT^Ser473 phosphorylation levels 4 hours after LY3214996 treatment compared with tumors of vehicle treated animals (Fig. 2H; DFCI316 vs. DFCI168, P < 0.05, Student t test; Supplementary Fig. S5). S6^Ser235/236 phosphorylation levels in DFCI316 tumors slightly increased over baseline compared with DFCI168^NRASQ61K tumors (Fig. 2H; Supplementary Fig. S5, P > 0.05, Student t test).

Because of the lack of antiproliferative activity in DFCI316^KRASQ61H tumors, the rapid MAPK pathway reactivation in two PDX models and the decrease of LY3214996 plasma levels with strong PK/PD relationship with once-daily drug dosing (Fig. 2C–H; Supplementary Figs. S4A and S4B), we subsequently tested if twice-daily dosing at 50 mg/kg is superior to once daily dosing at 100 mg/kg. Indeed, this was the case in the DFCI168^NRASQ61K PDX model (Fig. 2I, n = 8 animals/group, P < 0.0001, two-way repeated measures ANOVA with standard post hoc t tests). Twice-daily application of LY3214996 was well tolerated by NSG mice over 21 days (Supplementary Fig. S6).

Because LY3214996-induced FOXO3a reporter activities differed substantially in the patient-derived cancer cell lines (Fig. 1G), suggestive of differences in PI3K pathway activity, we next investigated the baseline characteristics of these cell lines. We observed that resistant cell lines (DFCI366, DFCI298, and DFCI332) exhibited a more mesenchymal phenotype (AXL^high, ERBB3^low, BIM^low) than sensitive cell lines (DFCI168, DFCI316 and DFCI24; Fig. 3A; Supplementary Fig. S7). Because mesenchymal cells also exhibited increased PI3K pathway activation (pAKT^S473, pS6^S235/236), we next tested the effect of combined ERK1/2 and PI3K/mTOR inhibition with LY3214996 and LY3023414. First, we determined Bliss synergy after 72 hours in all patient-derived cell lines and observed mostly additive effects (color-coded in green) with some synergy (color-coded in blue) in DFCI366^KRASG12A and DFCI516^KRASG12C cells (Fig. 3B). Western blot analyses of DFCI24^KRASG12C and DFCI316^KRASQ61H protein lysates indicated that PI3K inhibition with 1 μmol/L of LY3023414 slightly increased protein levels of ERK targets c-MYC, DUSP4, and SPRY2 compared with DMSO treated cells, whereas 1 μmol/L of LY3214996 and the combination of both drugs (both 1 μmol/L) reduced ERK target levels over 72 hours, respectively (Fig. 3C). In both models, combined ERK plus PI3K inhibition also reduced S6^Ser235/236 phosphorylation more profoundly than either drug alone. Levels of pro-apoptotic BIM increased slightly in both cell lines and PARP cleavage was detectable after 48 and 72 hours in DFCI316 cells with combined drug treatment.

Next, we quantified transcriptional changes during simultaneous ERK1/2 and PI3K/mTOR inhibition. Although LY3023414 (1 μmol/L) significantly increased mean overall MAPK pathway-dependent gene expression over 72 hours in both models compared with DMSO-treated cells (Fig. 3D; P < 0.001 for DFCI24 and P < 0.01 for DFCI316; unpaired t test), LY3214996 alone (1 μmol/L) or the combination of PI3K/mTOR and ERK1/2 inhibitor reduced the expression (P < 0.0001; unpaired t test). The reduction of overall gene expression in cells treated with single-agent LY3214996 and with the drug combination was comparable (P = 0.83 for DFCI24 and P = 0.068 for DFCI316; unpaired t test). Stronger transcriptional pathway suppression with the drug combination was accompanied by increased FOXO3a reporter activity in DFCI24 and DFCI316 cells compared with LY3214996 treatment alone (Fig. 3E) and resulted in increased nuclear FOXO3a accumulation in DFCI24 cells (Fig. 3F).

To determine the effect of co-occurring Tp53 or Lkb1 mutations (47) on efficacy of combined ERK1/2 and PI3K/mTOR inhibition, we determined Bliss synergy in cell lines derived from genetically engineered mouse models (GEMM; Kras^G12V = “K”-genotype: CHA14.1 and CHA14.2 (48); Kras^G12V;Tp53^fl/fl = “KP”-genotype: CHA9.1 and CHA9.3; Kras^G12V;Tp53^fl/fl;Lkb1^fl/fl = “KPL”-genotype: CHA487 and CHA496) after 72 hours (Fig. 3G). We observed mostly additive effects (color-coded in green), increased synergy (color-coded in blue) was detected in the presence of Tp53 and Lkb1/Stk11 comutations. Overall, the surface area indicating synergy was increased in GEMM-derived cell lines across genotypes compared with human-derived cancer cell lines (Fig. 3B).

To test the in vivo efficacy of combined ERK1/2 and PI3K inhibition, we treated NSG mice with xenotransplanted DFCI24^KRASG12C tumors for 4 weeks with vehicle, high-dose continuous LY3214996 (150 mg/kg, orally every day), intermittent split-dose LY3214996 (75 mg/kg, orally twice a day, 4 days ON, 3 days OFF), intermittent split-dose LY3023414 (15 mg/kg, orally twice a day, 4 days ON, 3 days OFF) or a combination of both drugs. Antitumor effects of single-agent treatments were comparably modest, whereas the inhibitor combination had a relatively robust antitumor effect (Fig. 4A; Supplementary Fig. S8A; P < 0.0001; two-way repeated measures ANOVA with standard post hoc t tests). We also treated the DFCI316^KRASQ61H PDX model with the same treatment schedules. We observed significant tumor growth inhibition in all treatment groups after 4 weeks, but tumor regressions were only achieved in the combination group (Fig. 4B; Supplementary Fig. S8B; P < 0.0001; two-way repeated measures ANOVA with standard post hoc t tests). All treatment schedules were well tolerated (Supplementary Fig. S9) and no differences in plasma drug concentrations from day 1/2 to day 14/15 were observed thus excluding changes in drug metabolism over time (Supplementary Fig. S10). With most animals for both PDX models having progressive disease with single-agent treatment, nearly half of the animals with DFCI24^KRASG12C tumors and all animals with DFCI316^KRASQ61H tumors achieved at least stable disease after 4 weeks of treatment in the combination group. Partial responses (PR) were observed in three of eight animals (DFCI24) and five of nine animals (DFCI316) in the combination group, respectively. Overall, treatment effects were synergistic in both PDX models (Bliss Independence method; Supplementary Figs. S11A and S11B) and translated into a stronger tumor growth delay in mice treated with the drug combination [Fig. 4C; DFCI24: 15.8 (combination) vs. 2.3 (LY3214996 BID) vs. 0.9 days (LY3023414); DFCI316: 26.8 (combination) vs. 12.2 (LY3214996 BID) vs. 17.5 days (LY3023414) compared with vehicle control].

Lysates of DFCI24^KRASG12C tumors which had been treated with vehicle, LY3214996 (75 mg/kg), LY3023414 (15 mg/kg), or a combination of both drugs for 4, 8, and 24 hours showed reduced levels of DUSP4, SPRY2, c-MYC, and phosphorylation of p90RSK^{Thr359/Ser363} as early as 4 hours after the application of LY3214996 or of the drug combination. Contrariwise, phosphorylation and total levels of these proteins were increased in tumors treated with single-agent LY3023414 (Fig. 4D). Although signaling parameters (phospho-p90RSK and phospho-S6) returned to baseline after 24 hours with single-agent treatment, the ERK/PI3K inhibitor combination induced continuous suppression of MAPK and PI3K downstream targets. ERK1/2^{Thr202/Tyr204} phosphorylation was increased in LY3214996-treated tumors at all time points and slightly increased in tumors of LY3023414-treated mice. More potent pathway inhibition with the drug combination coincided with stronger BIM accumulation and PARP cleavage at all time points. Analysis of MAPK pathway-dependent gene expression from the same tumors showed good overall pathway suppression with single-agent LY3214996 4 and 8 hours after the last drug application, whereas PI3K/mTOR inhibition alone induced pathway hyperactivation at 4 hours (Fig. 4E). The drug combination not only prevented the early PI3K/mTOR inhibitor-induced transcriptional pathway hyperactivation but was also more efficacious in counteracting pathway reactivation with single-agent ERK1/2 inhibition at 24 hours (Fig. 4E, dotted red line, P < 0.05, unpaired t test).

Because inputs from the MAPK and PI3K signaling pathways promote G₁–S cell-cycle transition via the cyclin D1-CDK4/6-RB pathway (3), we also tested the combination of LY3214996 and abemaciclib (LY2835219), a specific cyclin-dependent kinase (CDK) 4/6 inhibitor. First, we determined Bliss synergy by treating all patient-derived cell lines for 72 hours with different concentrations of both drugs. We observed mostly additive effects (color-coded in green) across the tested dose range with some synergy (color-coded in blue) in DFCI316^KRASQ61H, DFCI332^KRASG12D, and DFCI516^KRASG12C cells (Fig. 5A). Treatment effects were independent of the drug sequence (Supplementary Fig. S12) and of the used ERK inhibitor as combination partner (SCH772984 in Supplementary Fig. S13). Lysates of DFCI24^KRASG12C and DFCI332^KRASG12D cells which had been treated for up to 72 hours with DMSO, LY3214996 (1 μmol/L), abemaciclib (1 μmol/L), or the combination of both drugs showed more profound suppression of RB, ERK1/2, and S6 phosphorylation as well as of c-MYC protein levels than with either drug alone (Fig. 5B). DUSP4 and SPRY2 protein levels were equally suppressed with LY3214996 and the drug combination. Single-agent abemaciclib suppressed S6 phosphorylation, which is due to inhibition of CDK4-activated mTOR activity (49). Neither treatment led to PARP cleavage and BIM accumulation in response to LY3214996 was counteracted by the addition of abemaciclib.

To determine the impact of co-occurring Tp53 and Lkb1 mutations (47) on the efficacy of combined ERK1/2 and CDK4/6 inhibition, we determined drug synergy in GEMM-derived KRASG12V mutant lung cancer cell lines with K, KP, and KPL genotype after 72 hours and observed some drug synergy (color-coded in blue) across genotypes, which was increased in the presence of comutations in Tp53 and STK11/Lkb1 (Fig. 5C). Overall, the synergy of combined ERK1/2 and CDK4/6 inhibition was more pronounced in GEMM-derived cell lines across genotypes compared with patient-derived cell lines (Fig. 5A).

To test the in vivo efficacy of combined ERK1/2 and CDK4/6 inhibition, we treated the DFCI168^NRASQ61K (Fig. 6A) and DFCI24^KRASG12C PDX models (Fig. 6B) for 28 days with vehicle, LY3214996 (100 mg/kg, orally every day), abemaciclib (50 mg/kg, orally every day), or the combination of both drugs. We established that the well-tolerated split-dose intermittent application of LY3214996 (50 mg/kg, orally twice a day, 4 days ON, 3 days OFF) in combination with abemaciclib (50 mg/kg, orally every day) was not superior over once-daily drug application (Supplementary Fig. S14). Therefore, we chose a continuous once-daily dosing schedule for the efficacy studies. We observed tumor growth inhibition in both PDX models with LY3214996, whereas abemaciclib itself was ineffective (P < 0.001; two-way repeated measures ANOVA with standard post hoc t tests). The strongest antitumor effect was observed in animals which received both drugs (DFCI168: Fig. 6A; Supplementary Fig. S15A; DFCI24: Fig. 6B; Supplementary Fig. S15B; P < 0.0001, two-way repeated measures ANOVA with standard post hoc t tests). The effect of combining LY3214996 and abemaciclib was additive in the DFCI168^NRASQ61K and less than additive in the DFCI24^KRASG12C model (Bliss Independence method; Supplementary Figs. S16A and S16B) but translated into a longer tumor growth delay compared with single-agent treatment [Fig. 6C; DFCI168: 24.8 (combination) vs. 15.7 (LY3214996) vs. 2.9 days (abemaciclib); DFCI24: 7.2 (combination) vs. 3.2 (LY3214996) vs. 3.1 days (abemaciclib) compared with vehicle control]. In tumor lysates we observed inhibition of RB phosphorylation with LY3214996 or abemaciclib at early timepoints (4 and 8 hours) but partial pathway reactivation occurred after 24 hours (DFCI168: Fig. 6D; DFCI24: Supplementary Fig. S17). This rebound was effectively reduced by the addition of LY3214996 to abemaciclib. Single-agent abemaciclib increased phosphorylation of p90RSK (which is in contrast to p-p90RSK inhibition which we observed in vitro; Fig. 5B) and AKT as well as protein levels of Cyclin D1 compared with vehicle which was pronounced at 24 hours in both models, suggesting compensatory MAPK and PI3K pathway hyperactivation. MAPK pathway hyperactivation could be counteracted by adding LY3214996 to abemaciclib. While BIM protein levels increased in LY3214996-treated DFCI168^NRASQ61K tumors, treatment with abemaciclib, or the drug combination had the opposite effect on BIM overall and neither treatment increased PARP cleavage. Analysis of MAPK pathway-dependent gene expression from the same tumors showed good overall pathway suppression with single agent LY3214996 4 and 8 hours after the last drug application, whereas abemaciclib induced early (4 and 8 hours) pathway hyperactivation, which could be prevented by the drug combination (Fig. 6E).

Activating mutations in RAS proteins occur in about 30% of lung cancers (Supplementary Fig. S1; ref. 1). Advances have been made in the clinical development of direct KRAS and ICIs, but these strategies are so far restricted to lung cancers with KRASG12C mutations and to immunologically “hot” tumors, respectively (14–16, 50). Therefore, inhibiting RAS effector pathways which promote malignant behavior of cancer cells remains an attractive alternative. Unfortunately, single-agent treatment approaches have had no (MEK inhibitors) or only limited (CDK4/6 inhibitor) antitumor activity in clinical trials (8, 10, 11). On the basis of high cancer cell plasticity and pathway redundancy, potential resistance mechanisms include the loss of ERK-mediated negative feedback inhibition and PI3K pathway activation via receptor tyrosine kinases (RTK) for MEK (51, 52) and compensatory MAPK pathway activation for CDK4/6 inhibitors (53). In addition, co-occurring loss of function mutations of tumor suppressor genes (e.g., TP53 and STK11) increase the genetic heterogeneity of RAS-mutant tumors and impede therapeutic interventions (47). The major obstacle, however, is the toxicity observed with continuous blockade of MAPK pathway activity alone or in combination with other pathways (e.g., MEK plus PI3K inhibition), hence strongly indicating an unmet need to develop alternative combination strategies (2, 6, 10, 11, 13).

Because ERK1/2 inhibitors are more potent in preventing feedback and bypass activation compared with MEK1/2 inhibitors, toxicities intrinsic to continuous MEK inhibition could potentially be overcome by intermittent application (i.e., to include drug-free periods) of ERK inhibitors without compromising on antitumor efficacy (17, 18). Therefore, in this study, we investigated the efficacy of the ATP-competitive ERK1/2 inhibitor LY3214996 as single agent and in combination with a PI3K/mTOR (LY3023414) inhibitor (25) or CDK4/6 inhibitor (abemaciclib; ref. 26). LY3214996 has antineoplastic activity in commercially available cell lines and tumor models with MAPK pathway aberrations—many of them with KRAS mutations (27). Because no or only very limited clinical annotation is available for these models, we sought to study genetically well-characterized cell lines established from tumors of pretreated patients which are more reflective of the clinical setting in which an ERK inhibitor or ERK inhibitor-based drug combination would be initially clinically evaluated.

We established seven RAS-mutant cancer cell lines (n = 6 KRAS, n = 1 NRAS, Fig. 1A; Supplementary Table S1) from malignant effusions of in part heavily pretreated patients (Supplementary Fig. S2), in which single-agent LY3214996 treatment profoundly suppressed MAPK pathway activity (Fig. 1C–E). With the exception of DFCI332 cells, this translated into growth inhibition in all cell lines including those with STK11/LKB1 comutations (Fig. 1F), a genotype with relative intrinsic insensitivity to MEK inhibition (54), hence supporting preclinical data indicating that ERK inhibitors may have the potential to overcome this limitation of MEK inhibitors (55). LY3214996 also exhibited good antitumor activity in various PDX models (Fig. 2A; Fig. 4A and B; Fig. 6A and B) in which—similar to other ERK inhibitors—we observed a strong PK/PD relationship with once-daily drug application due to rapid plasma elimination (Fig. 2D). This pharmacokinetic trait led to an early recovery of MAPK pathway activity (Fig. 2C–E, G and H; Supplementary Figs. S4A and S4B) and consequently, the antitumor potency of LY3214996 increased with a split-dose approach without affecting the plasma drug clearance over longer treatment periods (Fig. 2I; Fig. 4B vs. Fig. 2F; Supplementary Figs. S6 and S10).

Across various cell lines and PDX models, however, the antineoplastic effect of LY3214996 differed substantially (Fig. 1B and F; Fig. 2A and F) despite comparable suppression of MAPK pathway activity (Fig. 1C–E; Fig. 2E and G). LY3214996 increased pro-apoptotic BIM in sensitive and intermediate sensitive cell lines in vitro (Fig. 1D), but failed to do so in resistant cell lines and—despite reducing tumor growth in various PDX models—tumor regressions were not observed (Fig. 2A; Fig. 4A and B; Fig. 6A and B). This possibly indicated compensatory mechanisms counteracting full execution of apoptosis. We subsequently established that cancer cell lines with reduced LY3214996 sensitivity exhibit a more mesenchymal phenotype with PI3K pathway activation and BIM suppression at baseline (DFCI298, DFCI332; Fig. 3A; Supplementary Fig. S7) or activate PI3K signaling in response to LY3214996 (DFCI316, Fig. 1C and D; Fig. 2H; Supplementary Fig. S5). PI3K pathway activation is known to prevent nuclear translocation of FOXO3a upon MEK inhibition thus preventing BIM induction and apoptosis execution (46, 56). Consistently, we observed a lack of FOXO3a nuclear translocation and transcriptional activation in resistant (DFCI298, DFCI332, DFCI366) versus sensitive (DFCI168) cell lines upon LY3214996 treatment (Fig. 1G and H).

We subsequently show that PI3K pathway-mediated LY3214996 resistance can be overcome by combined ERK1/2 plus PI3K/mTOR inhibition. The combination with LY3023414 increased downstream pathway suppression, BIM induction, and antiproliferative activity in human and GEMM-derived cancer cell lines (Fig. 3B–D and G) by promoting nuclear FOXO3a accumulation and transcriptional activation (Fig. 3D and E). This translated into synergistic growth inhibition and stronger tumor growth delay in PDX models with actual tumor regressions indicating an in vivo apoptotic response (Fig. 4A–D; Supplementary Figs. S8, S11A, and S11B). Tumor growth delay is a critically important measure of antitumor efficacy because it most closely mimics clinical endpoints that required observation of mice through the time of disease progression (41). Intermittent dosing of both drugs (i.e., 4 days treatment followed by a 3-day treatment-free period) was well tolerated by NSG mice (Supplementary Fig. S9) and did not compromise treatment efficacy which is important for the transfer into clinical applications even though toxicity profiles of ERK plus PI3K inhibitor combinations in humans remain unknown today. In light of significant toxicity observed with MEK or BRAF plus PI3K inhibitor combinations on uninterrupted schedules (6, 57), future clinical trials are therefore required to investigate if the LY3214996 plus LY3023414 drug combination is better tolerated. Compared with MEK inhibitors, ERK inhibitors lack a subset of toxicities (e.g., retinopathy; ref. 58) and LY3023414 differentiates from other PI3K pathway inhibitors in showing a short half-life of approximately two hours in clinical PK studies (59), thus potentially making the LY3214996 plus LY3023414 combination more applicable to humans to overcome overlapping toxicities such as nausea, vomiting, diarrhea, fatigue, or rash. The pronounced drug synergy that we observed in GEMM-derived cell lines with subtype-defining Tp53 and Stk11/Lkb1 comutations (Fig. 3G) may hold promise for this drug combination also in these otherwise hard to treat lung cancer subtypes (47, 54).

Although tumor regressions were observed with the ERK/PI3K inhibitor combination, combining LY3214996 with the CDK4/6 inhibitor abemaciclib (26)—a concept supported by mechanistic studies which showed synthetic lethality between KRAS and CDK4 and identified MAPK pathway activation as a CDK4/6 inhibitor resistance mechanism (5, 60)—lacked the potential to induce an apoptotic response in PDX models despite more sustained pathway inhibition and antineoplastic activity in vitro and in vivo (Fig. 5B and C; Fig. 6A–E; Supplementary Figs. S12A and S12B; Supplementary Fig. S13). This is not unexpected because a key feature of CDK4/6 inhibition is the induction of a cell-cycle inhibitory response that mimics the intrinsic senescence phenotype (5). Failure to induce cancer cell death, however, could compromise treatment efficacy, especially if rapid tumor shrinkage is required in symptomatic patients with high tumor burden.

Therefore, one major aspect that needs to be addressed in future preclinical and clinical studies is a survey to identify predictive biomarkers to aid patient stratification especially for the ERK plus CDK4/6 inhibitor combination. Here, we identify epithelial differentiation as a correlative marker for LY3214996 sensitivity. We hypothesize that apart from epithelial-to-mesenchymal transition, mutations affecting PI3K signaling (e.g., loss of PTEN or PIK3CA mutations) may exert similar effects. Interestingly, resistant DFCI332 cells which harbor a Noonan syndrome-associated SHP2/PTPN11 (c.923A>G, p.N308S) missense mutation exhibited the highest level of baseline PI3K pathway activation of all cell lines (Fig. 3A). SHP2 controls PI3K (and MAPK) pathway activity (61) and the PTPN11 c.923A>G mutation is considered pathogenic (FATHMM prediction score 1.0) in COSMIC. Furthermore of interest, DFCI168 cells with the highest LY3214996 single agent sensitivity harbor a NRAS^Q61K mutation. Zhou and colleagues reported recently, that KRASQ61H preferentially signals through the MAPK pathway conferring MEK inhibitor sensitivity (62). Even though our study is limited to draw general conclusions, certain RAS mutations may also predict increased LY3214996 sensitivity. Of note, single-agent abemaciclib was ineffective in both PDX models (Fig. 6A and B), even though both cell lines harbor nonfunctional p16 proteins which have been associated with CDK4/6 inhibitor sensitivity (63). DFCI168 cells present with a p16INK4A R80* mutation predicted to be pathogenic (FATHMM prediction score 0.88 in COSMIC) and reported to produce a truncated, nonfunctional p16 protein (64); DFCI24 cells exhibit a two-copy deletion of CDKN2A conferring loss of p16 (Supplementary Table S1). The predictive value of p16 loss, however, remains unclear (65) and in DFCI24 cells, mesenchymal differentiation with associated PI3K pathway activation (Fig. 3A) may have prevented abemaciclib from being effective (66). Overall, our experimental findings in combination with the available clinical data (8, 9) suggest that isolated CDK4/6 inhibition may be insufficient to achieve satisfactory antitumor activity in RAS-mutant lung cancer and that combined CDK4/6 plus ERK1/2 inhibition has the potential to overcome this limitation to a certain extent especially in tumors with TP53 and STK11/LKB1 comutations (Figs. 3E and 5C; refs. 47, 54).

In summary, we demonstrate the efficacy of LY3214996-based drug combinations in PDX models of RAS-mutant lung cancer (Fig. 6F summarizes the therapeutic approaches). Intermittent ERK inhibition on a “4-days-ON-3-days-OFF” schedule is well tolerated in mice without compromising on antitumor activity. Currently ongoing (NCT02857270, NCT03454035) and future clinical trials should aim to determine whether intermittent ERK inhibitor-based drug combinations can overcome toxicities associated with continuous MEK inhibition and to validate treatment predictors.

J. Köhler reported grants from Eli Lilly & Company and from the German Cancer Aid Foundation during the conduct of the study. Y. Zhao reported grants from Eli Lilly & Company during the conduct of the study. P.C. Gokhale reported grants from Elstar Therapeutics, Epizyme Inc, and Daiichi Sankyo outside the submitted work. C.P. Paweletz reported grants from Eli Lilly during the conduct of the study; C.P. Paweletz also reported personal fees from BioRad and Dropworks and nonfinancial support from Xsphera outside the submitted work. S.V. Bhagwat reported other from Eli Lilly and Company outside the submitted work. R.V. Tiu reported a patent for W02018005234A1 issued; R.V. Tiu also reports employment with Astellas Pharma and Eli Lilly & Company and is a stockholder of Eli Lilly & Company. P.A. Janne reported grants and personal fees from Eli Lilly during the conduct of the study; P.A. Janne also reported grants from AstraZeneca, Boehringer Ingelheim, Daiichi Sankyo, Takeda Oncology, Astellas, PUMA, and Revolution Medicines; personal fees from AstraZeneca, Boehringer Ingelheim Pfizer, Roche/Genentech, Chugai, Ignyta, Loxo Oncology, from SFJ Pharmaceuticals, Voronoi, Daiichi Sankyo, Biocartis, Novartis, Sanofi Oncology, Takeda Oncology, Mirati Therapeutics, Trasncenta, Silicon Therapeutics, Syndax outside the submitted work; and also receives postmarketing royalties from a DFCI-owned patent on EGFR Mutations issued and licensed to Lab Corp. No potential conflicts of interest were disclosed by the other authors.

J. Köhler: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. Y. Zhao: Data curation, formal analysis, validation, investigation, visualization, methodology, performed in vitro experiments. J. Li: Data curation, formal analysis, validation, investigation, visualization, methodology, project administration, performed in vitro experiments. P.C. Gokhale: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. H.L. Tiv: Formal analysis, validation, investigation, visualization, methodology, performed in vivo experiments. A.R. Knott: Validation, investigation, visualization, methodology, performed in vivo experiments. M.K. Wilkens: Formal analysis, validation, investigation, visualization, methodology, performed in vivo experiments. K.M. Soroko: Formal analysis, validation, investigation, visualization, methodology, performed in vivo experiments. M. Lin: Resources, validation, investigation, generated and validated patient-derived cell lines. C. Ambrogio: Resources, methodology, provided GEMM-derived cancer cell lines. M. Musteanu: Resources, methodology, generated and provided GEMM-derived cancer cell lines. A. Ogino: Resources, validation, investigation, methodology, established and validated a patient-derived cancer cell line. J. Choi: Resources, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing, established and validated patient-derived cell lines. M. Bahcall: Investigation, visualization, writing–original draft, project administration, writing–review and editing. A.A. Bertram: Resources, investigation, visualization, analyzed patient clinical histories. E.S. Chambers: Resources, methodology. C.P. Paweletz: Resources, formal analysis, methodology, writing–review and editing. S.V. Bhagwat: Resources, data curation, software, formal analysis, methodology, writing–review and editing. J.R. Manro: Data curation, software, formal analysis, methodology, writing–review and editing. R.V. Tiu: Resources, formal analysis, methodology, writing–review and editing. P.A. Jänne: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This study was funded by Eli Lilly & Company. We thank Johannes Köster for supporting this project scientifically. This study was supported in part by Eli Lilly & Company, an NCI grant R35CA220497 (to P.A. Janne), and a Mildred-Scheel postdoctoral research fellowship of the German Cancer Aid Foundation (70111755 to J. Köhler).

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.