Covalent inhibitors of KRASG12C specifically target tumors driven by this form of mutant KRAS, yet early studies show that bypass signaling drives adaptive resistance. Although several combination strategies have been shown to improve efficacy of KRASG12C inhibitors (KRASi), underlying mechanisms and predictive strategies for patient enrichment are less clear.
We performed mass spectrometry–based phosphoproteomics analysis in KRASG12C cell lines after short-term treatment with ARS-1620. To understand signaling diversity and cell type–specific markers, we compared proteome and phosphoproteomes of KRASG12C cells. Gene expression patterns of KRASG12C cell lines and lung tumor tissues were examined.
Our analysis suggests cell type–specific perturbation to ERBB2/3 signaling compensates for repressed ERK and AKT signaling following ARS-1620 treatment in epithelial cell type, and this subtype was also more responsive to coinhibition of SHP2 and SOS1. Conversely, both high basal and feedback activation of FGFR or AXL signaling were identified in mesenchymal cells. Inhibition of FGFR signaling suppressed feedback activation of ERK and mTOR, while AXL inhibition suppressed PI3K pathway. In both cell lines and human lung cancer tissues with KRASG12C, we observed high basal ERBB2/3 associated with epithelial gene signatures, while higher basal FGFR1 and AXL were observed in cells/tumors with mesenchymal gene signatures.
Our phosphoproteomic study identified cell type–adaptive responses to KRASi. Markers and targets associated with ERBB2/3 signaling in epithelial subtype and with FGFR1/AXL signaling in mesenchymal subtype should be considered in patient enrichment schemes with KRASi.
This article is featured in Highlights of This Issue, p. 2367
KRASG12C inhibitors (KRASi; AMG-510 and MRTX849) have demonstrated efficacy in KRASG12C-mutant lung cancer, however, variability in tumor response and activation of bypass signaling limit efficacy. Although combination strategies (MEK inhibitor, SHP2 inhibitor, pan-HER inhibitor, and EGFR inhibitor) with KRASG12C-specific inhibitors are being investigated in clinical trials, how to best identify responsive tumors for each combination remains a clinical challenge. Our data suggest that depicting epithelial versus mesenchymal state of KRASG12C along with basal expression of ERBB family, FGFR1, and AXL may associate with combination efficacy secondary to mechanisms of adaptive resistance. We identified that ERBB2/3 signaling causes resistance to KRASi in epithelial types, while FGFR1 or AXL signaling is more prominent in mesenchymal types. Our data highlight the importance of phosphoproteomics-based approach to identify tumor-specific adaptive rewiring, which can be utilized to aid personalized patient care in KRASG12C mutants.
Oncogenic KRAS mutations occur in nearly 25% of lung adenocarcinomas, are often associated with poor prognosis, and are notoriously refractory to conventional cytotoxic chemotherapies, as well as agents that target receptor tyrosine kinases (RTKs; refs. 1, 2). Use of MEK inhibitors with cytotoxic agents or other pathway inhibitors (PI3K/Akt) also proved to be either ineffective or toxic (3). Therefore, it is an unmet medical need to identify effective therapies for patients with KRAS-mutant lung cancer. The advancements in medicinal chemistry have produced high potency KRASG12C inhibitors (KRASi), such as ARS-1620, rapid binding to the GDP-bound inactive form, and optimized pharmacokinetics to maximize exposure and duration (4, 5). Because KRASG12C mutations occur in approximately 11% to 16% of lung adenocarcinomas, this represents a large group of patients who can potentially benefit. Subsets of KRASG12C non–small cell lung cancer (NSCLC) cells are highly sensitive to these inhibitors, and AMG-510 and MRTX-849 have demonstrated early activity in patients with KRASG12C-mutant lung cancer (6–8).
Because nucleotide exchange of RAS signaling can be mediated by upstream RTKs, recent results demonstrated that RTK inhibitors can impede the exchange reaction and sensitize cells to KRASi (9, 10). Recent studies identified several combination strategies with KRASi, which includes inhibitors of HER kinases, FGFR1, AXL, SHP2, mTOR, MEK, PI3K, and CDK4/6, among others (11–16). However, it remains unclear why one cotargeting strategy may be more effective in particular cell types, and what mechanisms underlie the observed effectiveness of some combinations. The exact RTK that mediates this effect may vary from cell line to cell line, despite the common KRAS driver mutation, thus making it difficult to determine how to precisely predict combinations that would be effective in the clinic. In one case, higher basal expression of EGFR signaling in KRASG12C colorectal cancer cell line compared with NSCLC predicted effectiveness of adding anti-EGFR therapy to KRASi (17). However, these results are somewhat perplexing given the high levels of EGFR signaling in NSCLC (18). Because KRASG12C bound to GDP competes with the small-molecule inhibitors and exchange factors (i.e., SOS), a better understanding of signaling events that control RAS signaling will be needed to optimize the use of KRASi and design precision medicine with KRASi in the clinic.
Given the diverse responses observed in both preclinical models and human patients, it is critical to understand how cells escape from targeted inhibition, which pathways contribute to resistance, and how to predict pathway utilization for escape to enable precision medicine in the form of combination therapy. Given the diversity of kinases and other signaling molecules known for adaptive resistance, a system-level analysis is the most suitable approach to uncover mechanisms of therapeutic escape to KRASi. To address these challenges, we carried out mass spectrometry–based phosphoproteomics following KRASG12C inhibition in three different lung cancer cell lines. Our data suggest a model in which inhibition of KRASG12C is overcome by adaptive signaling changes. There is growing appreciation for the role of adaptive resistance to targeted agents mediated by changes in feedback programs (19). Our phosphoproteomics data suggest different processes of adaptive rewiring and therapeutic escape to KRASi and indicate the possibility to predict cotargeting strategies that interfere with compensatory pathways.
Materials and Methods
NCI-H358, Calu1, NCI-H1373, NCI-H1792, NCI-H23, and NCI-H2122 were purchased from the ATCC. HCC-44 and HOP-62 were acquired from MD Anderson Cancer Center (Houston, TX) and NCI, respectively. LU-99 was generous gift from Dr. Sheri Moores (The Janssen Pharmaceutical Companies of Johnson & Johnson).
ARS-1620 and BI-3406 were purchased from MedChemExpress. AMG-510, erlotinib, afatinib, AZD-4547, linsitinib, imatinib, cabozantinib, ceritinib, GDC-0941, and RXDX-106 were purchased from Selleckchem. Drugs were reconstituted in DMSO to 10 mmol/L stock concentrations and stored at −80°C. The combination index (CI) was calculated using CompuSyn software (20). The coefficient of drug interaction (CDI) was calculated using the following formula: CDI = AB/(A × B) (21).
Sample preparation for phosphoproteomics and LC/MS-MS
Sample preparation for proteomics was carried out according to Cell Signaling Technology protocol (#8803). Briefly, the cells were lysed in denaturing buffer, followed by reduction/alkylation, trypsin digestion, and peptide desalting. For phosphotyrosine (pY) enrichment, anti–pY-1000 antibodies were used, followed by tandem mass tags (TMTs), isobaric tag labeling and immobilized metal affinity chromatography (IMAC) enrichment. For global phosphorylation enrichment, the peptides were first labeled with TMT and fractionated with bRPLC, followed by IMAC enrichment. A nanoflow ultra-HPLC interfaced with an Electrospray Quadrupole-orbitrap Mass Spectrometer (RSLCnano and Q Exactive HF-X, Thermo Fisher Scientific) was used for LC/MS-MS peptide sequencing and TMT quantification.
MaxQuant software (version 126.96.36.199) was used for peptide identification and reporter ion quantification (22). Data were normalized and analyzed for differential expression between treatment and control timepoints. Data were available via the PRIDE (23) for pY in ARS-1620–treated cells (PXD021607), for global phosphoproteomics after ARS-1620 treatment in H358 (PXD021611), H1792 (PXD021609), and Calu1 (PXD021608), and for pY (PXD021604) and protein expression (PXD021603) in eight KRASG12C cell lines.
Liquid chromatography-multiple reaction monitoring mass spectrometry and data analysis
Liquid chromatography-multiple reaction monitoring mass spectrometry (LC-MRM) analysis was performed in triplicate on a nanoUHPLC interfaced with an Electrospray Triple Quadrupole Mass Spectrometer (DionexRSLCnano and TSQ Altis, Thermo Fisher Scientific). For each sample, peptide mixture (5 μL) was loaded onto the trap column and the trapped peptides were eluted onto a C18 analytic column and separated using a 62.5-minute gradient. Collision energy (CE) values were optimized for this instrument in Skyline version 188.8.131.52 (24) using CE optimization equations empirically derived from data collected previously on the instrument. Skyline (v. 184.108.40.206) was used to evaluate the data (24).
TGFβ-epithelial-to-mesenchymal transition scores
The 105-probe set TGFβ-epithelial-to-mesenchymal transition (EMT) signature (25) was used to assess the levels of TGFβ-driven EMT in cell lines and tumor samples. For each dataset, rows corresponding to the signature genes were selected and principal component analysis (PCA) models were generated from these TGFβ-EMT rows. Low/high directionality of PC1 was established from the PCA loadings and their signs from the original TGFβ-EMT signature, flipping the sign of PC1 as appropriate so that increasing PC1 values corresponds to increasing levels of TGFβ-driven EMT.
For detailed experimental procedures, see the Supplementary Materials and Methods.
Quantitative phosphoproteomics identify patterns of signaling responses following KRASG12C inhibition
To understand mechanism(s) of short-term signaling adaptation to KRASi using phosphoproteomics, we selected three cell lines with differential sensitivity to ARS-1620. A heterogeneous response to ARS-1620 was observed in a panel of eight KRASG12C NSCLC cell lines in both 2D and 3D cultures (Fig. 1A; Supplementary Fig. S1A), with H358 exhibiting the highest sensitivity with IC50 value of 0.4 μmol/L (in 2D culture). Similar heterogeneity of response was also observed with AMG-510 (Supplementary Fig. S1B and S1C). Using 1 μmol/L doses, three cell line models, H358, Calu1, and H1792 cells, were selected and designated as sensitive, moderate, and resistant lines, respectively (Supplementary Fig. S1D). To select timepoints for phosphoproteomics analysis, we carried out Western blotting of key signaling molecules, p-ERK (T202/Y204) and p-Akt (S473), in a panel KRASG12C cell lines at different treatment timepoints. Signaling analysis after ARS-1620 and AMG-510 treatment revealed inhibition of p-ERK with evidence of rebound starting at 24 hours (Fig. 1B; Supplementary Fig. S1E). Similarly, suppression in p-Akt was observed only at 6 hours in H1373 and H1792, following rebound starting at 24 hours. However, moderate to less impact on Akt phosphorylation was observed in H358 and Calu1 at all treatment timepoints (Fig. 1B).
In parallel, we evaluated KRASG12C target engagement at different times after ARS-1620 treatment using LC-MRM relative quantification of tryptic peptides representing total KRAS, wild-type (WT) KRAS, free G12C KRAS, and drug-bound KRAS G12C (Supplementary Table S1). Total KRAS expression before treatment differed between cell lines: H358 > Calu1 > H1792. H358 and Calu1 cell lines showed a rebound in total KRAS expression at 24 hours, while H1792 showed a decreasing trend (P = 0.054). Baseline expression of WT KRAS also differed between the cell lines: H1792 > Calu1 > H358, while mutant protein was highly expressed in H358 compared with the other cell lines. Target engagement (%) was determined by the loss of ion signal for the free KRAS G12C protein (LVVVGAcamCGVGK); values are listed on the plot (Fig. 1C). Signal for this peptide was significantly decreased in all cell lines at both timepoints after ARS-1620 treatment, but a minor rebound in free KRAS G12C protein expression was observed in H358 cells. The G12C peptide covalently modified with ARS-1620 was also quantified and increased at both timepoints after drug treatment.
The signaling analysis and LC-MRM peptide quantification guided selection of ARS-1620 treatment timepoints (6 and 24 hours) for a multiplexed isobaric tag–based (TMT) quantitative phosphoproteomics experiment in H358, Calu1, and H1792 cells done in biological triplicates (Fig. 1D). With pY enrichment, we identified 2,275 human phosphosites in total, corresponding to 1,153 proteins in 987 protein groups, of which 433 phosphosites were differentially expressed (selected on the basis of ±1.5-fold change and P < 0.05) between any treatment timepoint and control (Supplementary Table S2). Analysis of phosphoproteomics data (pS/T/Y) identified 26,735 human phosphosites, corresponding to 5,577 proteins in 5,527 protein groups, of which 5,579 phosphosites were differentially expressed (based on ±1.5-fold change and P < 0.05) between any treatment timepoint and control (Supplementary Table S3).
Differentially expressed phosphosites were then categorized into four broad patterns of behaviors: “attenuating,” “ramping,” “stable,” and “opposite.” Furthermore, “attenuating” proteins were categorized into “attenuating-down” and “attenuating-up” (Fig. 1E). We reasoned that “attenuating” proteins indicate direct impact of ARS-1620 treatment on signaling and generated a literature network for these proteins (Fig. 1F) using MetaCore (Clarivate Analytics) pathway analysis. This analysis clustered “attenuating-down” proteins indicating inhibition of ERK signaling. The network on “attenuating-up” proteins indicated activation of insulin signaling, where IRS1, IRS2, DOK1, and PRKCD were the major activation nodes connected to insulin receptor. In addition, a network connecting RPS6KB1, RPS6KB2, EIF4B, and RPS6 nodes indicated activation of mTOR signaling. Thus, IGFR, PI3K, and mTOR served as adaptive signaling hubs in response to ARS-1620.
We next compared alterations in the phosphoproteomes of these cells following KRASG12C inhibition. Interestingly, PCA of the pY and pSTY dataset (Fig. 1G) showed separation of the three cell lines between PC1 and PC2, which indicates that the three cell lines have distinct phosphoproteomic alterations following ARS-1620 treatment. Indeed, the PC plot between PC3 and PC4 indicated more changes at 6 than 24 hours when compared with DMSO treatment (0 hours) and lesser change in magnitude between 0 and 24 hours. This analysis indicated major phosphoproteomics alterations at 6 hours, followed by signaling rebound by 24 hours, consistent with our Western blot observations.
The distribution of differentially expressed phosphosites indicates more of cell line–specific response, where only seven pY and 99 pSTY phosphosites were differentially affected across all three cell lines (Fig. 1H; Supplementary Tables S2 and S3). The decreased phosphorylation of MAPK3 (pY204) and MAPK6 (pS189) in all cells indicated inhibition of ERK signaling. Among the other proteins with decreased phosphorylation, we identified proteins associated with (i) cell-cycle regulation and DNA replication check point (RB1 pS788, WEE1 pT190, RBL2 pT401, and EMD pS120), (ii) transcription and RNA splicing (BCLAF1 pS196, SRSF4 pS431, NCBP1 pS7, RANBP2 pS1509, and SRRM1 pS452), and (iii) receptor signaling (EPHA2 pY594, pS901, and pS897, as well as CD44 pS697). The phosphosites with increased phosphorylation associated with (i) mTOR signaling (RPS6 pS247, pS244, and pS240, EIF4G1 pS1147 and RPS6KB1 pS394), (ii) cell motility and survival (SDC4 pY197, IRS2 pS620, and PRKCD pS302), (iii) negative regulation of apoptosis (HIPK3 pY359), and (iv) regulation of RAS-ABL signaling (RIN1 pY36). Despite this commonality among cell lines, these data suggest that perturbations result in unique phosphosignatures across various cell lines despite all harboring KRASG12C. It also suggested that each cell type required individual data analysis to uncover cell type–specific adaptive resistance mechanisms.
HER2/HER3 signaling drives adaptive resistance to KRASi
We observed the highest number of perturbed phosphosites in H358 cells, where 167 and 3,186 phosphosites were differentially expressed in the pY and pSTY data, respectively. We carried out MetaCore pathway analysis on phosphosites perturbed at 6 hours to understand an immediate impact of ARS-1620 on signaling. We observed many nodes densely connected in several subnetworks (Supplementary Fig. S2A). The inhibition of network surrounding MAPK1/MAPK3 indicated inhibition of ERK signaling. Among the various activating subnetworks, we observed a signaling hub surrounding RAF1, presented as a direct substrate of several kinases, including CAMKK2, CAMK2A, PRKCA, PRKCB, PRKD1, and PAK4.
Having observed the complexity of the MetaCore network, we next sought to generate a simplified network by manually curating the pathway based on information retrieved from PhosphoSitePlus (26) and experimentally affirmed phosphosites information in the literature (Fig. 2A). We observed decreased phosphorylation of MAPK1/3 and RPS6KA1, indicating inhibition of MAPK signaling. However, other signaling events, such as decreased phosphorylation of RPS6KA1-specific SOS1 site (pS1134) and increased phosphorylation of PRKAA1-specific RAF1 site (pS621), indicated feedback activation of MAPK signaling following KRASG12C inhibition (27, 28).
We next checked alterations in EGFR phosphorylation as previous studies report synergistic combination effects of EGFR/HER family kinase inhibitors with ARS-1620 (14). ARS-1620 treatment decreased phosphorylation levels of known inhibitory phosphorylation sites of EGFR at S1064 and S991 (29). Next, we sought to look in signaling perturbation in other members of ERBB family. Notably, we identified the phospho-peptide for tyrosine activation site (Y1289) of HER3 only in ARS-1620–treated cells. Enhanced phosphorylation on multiple phosphosites of HER2 (T1240) and HER3 (S1104 and S982) was also observed. Conversely, decreased phosphorylation of many phosphosites on HER2 (T701, T1166, S1078, S1174, S1107, and S998) and HER3 (S1094, S1096, and S1097) was observed. The ERK-mediated phosphorylation of T701 (old annotation T677) on HER2 mediated a negative feedback loop to control ERBB receptor dimer formation (30) and MEK inhibition induced ERBB/Akt signaling activation by suppressing HER2 pT701 phosphorylation (31). These alterations in multiple phosphosites of HER2 and HER3 indicate a significant role of HER2/HER3 signaling, beyond EGFR signaling, and could explain recent studies suggesting a lesser role of EGFR in lung cancer.
To address this, we evaluated in vitro efficacy of the ARS-1620 plus ERBB inhibitor combinations in H358 cells. When compared with more specific EGFR inhibitor (erlotinib), the ARS-1620 combination with the pan-ERBB inhibitor, afatinib, had greater impact on reducing cell viability (Fig. 2B) and presented higher synergistic effects even at lower dose (Supplementary Fig. S2B and S2C). Western blot–based signaling analysis also indicated longer suppression of p-ERK and p-AKT with the afatinib combination (Fig. 2C). Despite higher total levels of HER2/3 following KRASG12C inhibition (Fig. 2C), decreased phosphorylation on multiple serine and threonine residues suggests these sites as inhibitory phosphorylation sites, thereby driving resistance to KRASi via feedback ERBB signaling activation. ARS-1620 treatment alone increased the phosphorylation of HER3 (Y1289), although a weak and variable signal for HER2 phosphorylation (Y1248) was observed. Conversely, decreased phosphorylation of EGFR (Y1068) was observed in cells treated with ARS-1620 for longer duration. Cotreatment with afatinib suppressed HER2 expression and HER3 phosphorylation while also suppressing EGFR phosphorylation. Because higher expression of HER2 and HER3 was observed following ARS-1620 treatment, we next checked whether ARS-1620 treatment enhances HER2-HER3 dimerization. We observed higher HER2 and HER3 binding following KRASG12C inhibition, which was suppressed by the ARS-1620 plus afatinib combination (Fig. 2D). These findings indicate that activation of HER2/HER3 signaling limits the efficacy of KRASG12C inhibition and provides the mechanistic insight of responsiveness to afatinib combination with KRASG12C inhibition in xenograft models, as well as why lung cancer may be less reliant on EGFR than colon cancer (12).
HER2/HER3 signaling within an epithelial subtype associates with effectiveness of dual KRAS/pan-HER inhibition
We and others have observed that afatinib treatment as single agent had effects on cell viability in H358 cells (32). This observation indicated that some KRASG12C mutants have active ERBB signaling aiding KRASG12C signaling for tumor progression. Moreover, our results indicated that KRASi further amplifies ERBB signaling as an adaptive mechanism to sustain cell survival. Next, we determined the efficacy of the KRASi plus pan-HER inhibitor combination on cell viability and then evaluated ERBB signaling in the panel of KRASG12C cell lines. The afatinib combination with ARS-1620 had higher effect on cell viability in H2122, H1373, and H23, along with H358 cells (Fig. 3A). Consistent with our previous observation, enhanced expression and phosphorylation (Y1289) of HER3 and increase in HER2 expression were observed in H2122 following ARS-1620 treatment. On the other hand, a higher expression and phosphorylation (Y1248) of HER2 and enhanced expression of HER3 were observed in H23 (Fig. 3B).
Next, to ascertain whether cell lines responsive to the afatinib/ARS-1620 combination have enhanced ERBB signaling, we carried out TMT-based quantitative phosphoproteomics (with pY enrichment) and expression proteomics of all eight KRASG12C cell lines (Supplementary Fig. S3A; Supplementary Tables S4 and S5). We observed higher phosphorylation of multiple phosphosites on HER3 in cell lines responsive to afatinib combination (Fig. 3C). Western blot analysis of ERBB proteins indicated a varied basal expression of EGFR among KRASG12C cell lines, but higher basal expression of HER2 and HER3 was observed in H358, H2122, and H1373 (Fig. 3D).
We hypothesized that cell states, specifically epithelial or mesenchymal phenotypes, could contribute to the response of combining pan-HER inhibitor with KRASG12C inhibition. To determine EMT status, we assessed the protein ratio of epithelial marker E-cadherin (CADH1/CDH1) to mesenchymal marker N-cadherin (CADH2/CDH2). We observed significantly higher CDH1 to CDH2 ratio in four cell lines (H358, H2122, H1373, and H23), which were then designated as epithelial (E) type (Fig. 3E), and other models as mesenchymal (M) type. When we compared the CDI for afatinib plus ARS-1620 combination between epithelial and mesenchymal models, we observed higher synergistic effects in E-type KRASG12C models (Fig. 3F). The E-type KRASG12C models also showed significant lower intensities of many phosphosite(s) associated with vimentin (Y53, Y11, Y38, Y61, Y117, and Y276) and higher intensity for CDH1 Y797 when compared with M-type (Supplementary Fig. S3B). We did not see a significant difference in ERBB phosphosites in two-group (epithelial vs. mesenchymal) comparison. However, a significant difference in ERBB3 phosphosite, Y1328, was observed (Supplementary Fig. S3B) when we compared two most epithelial (H358 and H1373) with two most mesenchymal (H1792 and Calu1) models (Fig. 3E).
The LC-MRM–based quantification of EMT makers also confirmed that E-cadherin is highly expressed in H358, while N-cadherin is expressed at higher levels in Calu1 and H1792 (Fig. 3G; Supplementary Table S6). We next compared gene expression [using RNA sequencing (RNA-seq) data] of ERBB genes and EMT markers in KRASG12C cell lines using data from Cancer Cell Line Encyclopedia (CCLE; ref. 33; Fig. 3H; Supplementary Table S7). In concordance with our proteomics results, we identified KRASG12C models (H1171, H2291, H358, H2122, and H1373) with higher E-cadherin and/or lower N-cadherin expression also have higher ERBB3 gene expression and a significant difference was observed when we performed two-group (epithelial vs. mesenchymal) comparison (Supplementary Fig. S4A).
As an alternative way to score epithelial versus mesenchymal phenotypes, TGFβ-induced EMT signature (25) was translated into the protein expression data of KRASG12C cell lines to differentiate (Supplementary Fig. S4B). The PCA plot of TGFβ-EMT score among eight KRASG12C cell lines predicted similar phenotypes, where H358, H2122, and H1373 appeared to be more epithelial (low EMT score) than other cell lines (Fig. 3I; Supplementary Table S8).
We next examined whether KRASG12C human lung cancer tumor tissues also had a similar distribution of ERBB2 and ERBB3 expression and included both epithelial and mesenchymal phenotypes. Gene expression data of KRASG12C mutants (n = 59) were extracted from microarray gene expression dataset from 442 patients with lung adenocarcinoma (34). A divergent expression of ERBB2 and ERBB3 was observed in KRASG12C mutants with a subpopulation of mutants having higher expression of both ERBB2 and ERBB3 (Fig. 3J; Supplementary Table S9). Moreover, a positive correlation was also observed between ERBB2 and ERBB3 gene expression (Fig. 3K). Next, we examined EMT activity based on the TGFβ-EMT gene signature in KRASG12C mutants (25). The samples were ranked by TGFβ-105 score and divided into two groups at the median (ERBB3, P = 1.62E-04 and ERBB2, P = 1.97E-03). We observed a significant negative correlation of EMT score with ERBB2 (r = −0.50; P = 5.63E-05) and ERBB3 (r = −0.57; P = 2.95E-06; Fig. 3L; Supplementary Table S9). Therefore, KRASG12C human lung cancer exists in an ERBB2/ERBB3-overexpressing state associated with epithelial subtype.
SHP2, SOS1, and IRS1 contribute to adaptive resistance to KRASG12C inhibition with an epithelial subtype
Our data identified rebound of MAPK signaling as one of the underlying mechanisms of resistance in the epithelial subtype (exemplified by H358 cells). In one of such signaling event, we saw decreased phosphorylation of SOS1 (on S1134) only in H358 cells (phospho-peptide not detected in H1792 and Calu1), which relaxes allosteric inhibition of SOS1 (27). We hypothesized that the epithelial phenotype could also be important in affecting cell sensitivity to SHP2 and SOS1 inhibition in combination with KRASG12C inhibition and recapitulate effects observed with afatinib. SHP2 inhibition using RMC-4550 (10) and SOS1 inhibition using BI-3406 (35) as single agents incompletely suppressed ERK phosphorylation at any treatment timepoint when compared with KRASi in H358 cells (Fig. 4A). Individual inhibition of SHP2 or SOS1 was not effective to induce reduction in cell viability in KRASG12C cell lines (Supplementary Fig. S5A–S5C). Longer suppression of p-ERK with ARS-1620 was observed when combined with an SHP2 inhibitor (SHP2i) and an SOS1 inhibitor (SOS1i) (Fig. 4B), although p-ERK rebound was observed with SOS1i at 48 hours. We observed no effects on p-AKT expression, which is consistent with previous studies indicating that Akt signaling is independent of SHP2 (36). Cell viability analysis indicated that combining SHP2 or SOS1 inhibitors sensitizes H358 cells to KRASG12C inhibition, but not H1792 cells (Fig. 4C; Supplementary Fig. S5D and S5E). The cell viability analysis in other KRASG12C NSCLC cells indicated that epithelial subtypes showed greater synergistic effects to SHP2 and SOS1 inhibitor when combined with ARS-1620 (Fig. 4D and E; Supplementary Fig. S5F and S5G). Of note, complete inhibition of ERK signaling using three-drug combination (KRASi, SHP2i, and SOS1i) suppressed p-ERK expression only in epithelial subtype, while p-ERK rebound was still observed in mesenchymal cells (Fig. 4F). These results suggest targeting SHP2 and SOS1, downstream of HER signaling, also has combination efficacy within the epithelial subtype.
We also observed that KRASG12C inhibition in H358 also increased phosphorylation on IRS1, an adaptor protein for insulin-like growth factor receptor 1 (IGF1R), on multiple sites. Inhibition of IGF1R-mediated signaling with linsitinib was reported previously to sensitize KRASG12C cell lines to ARS-1620 (13). We sought to extend these findings using ceritinib, an FDA-approved inhibitor of ALK, which also has activity against IGF1R (37). Both ceritinib and linsitinib combinations with ARS-1620 demonstrated reduced cell viability (Fig. 4G). Next, we assessed the effect of the linsitinib plus ARS-1620 combination on cell viability in other KRASG12C models (Supplementary Fig. S5H). Our data suggest no significant difference in CDI between epithelial and mesenchymal models to linsitinib combination with ARS-1620, where three epithelial (H358, H23, and H2122) and one mesenchymal model (H1792) showed synergistic effects to linsitinib combination (Fig. 4H).
IRS1 is also known to interact with ERBB family kinases further activating the ERBB–PI3K–Akt axis (38, 39). KRASG12C inhibition in combination with either afatinib or linsitinib effectively suppressed cell growth in some KRASG12C lung cancer models; we hypothesized that IRS1 was common signaling node downstream of ERBB and IGFR signaling. Our proteomics data identified increased phosphorylation of IRS1 (on Y632) following ARS-1620 treatment; hence, we assessed IRS1 signaling after combining KRASi with pan-HER or IGF1R inhibitor (Fig. 4I). Afatinib and linsitinib alone or in combination with ARS-1620 were able to reduce phosphorylation of IRS1. We observed that KRASi/afatinib combination decreased phosphorylation of Akt and ERK, while KRASi/linsitinib combination only reduced Akt phosphorylation. These observations suggest that IRS1 mediates signaling downstream of activated ERBB and IGFRs regulate Akt signaling to sustain cell survival following KRASG12C inhibition.
FGFR1 activation drives resistance to ARS-1620 in H1792 cells
We next focused on H1792 cells because these cells presented with a more mesenchymal phenotype and showed modest response to dual pan-HER/KRASG12C or SHP2/KRASG12C inhibition. Phosphoproteomics analysis identified 140 pY and 1,941 pSTY modulated by ARS-1620. The MetaCore pathway analysis (Supplementary Fig. S6A) indicated inhibition of ERK signaling, but a well-connected activation network surrounding ABL1, CDK5, PRKCD, EGFR, RPS6KB1, YY1, and MYC was identified. Although many proteins appeared to be interconnected, identification of the upstream source of the signaling was quite challenging.
To gain more in-depth understanding of the signaling cascade, we manually curated the network solely on perturbed phosphosites (Fig. 5A). We identified increased phosphorylation of FGFR1 (Y653/Y654) and FGFRL1 (Y502). We also identified perturbation in other components of the FGFR signaling axis, which include increased phosphorylation of PLCG1 (Y977), ABL1 (Y393), and CDK5 (Y15). Increased phosphorylation of multiple phosphosites on RPS6, RPS6KB1, RPS6KB2, EIF4EBP1 indicated activation of mTOR signaling. These results are consistent with other observations that MEK inhibition can enhance phosphorylation of the FGFR adaptor protein, FRS2, serving a role in adaptive resistance in KRAS mutants (40, 41). We also observed increased phosphorylation of FRS2 at Y436 in H1792 following ARS-1620 treatment (Fig. 5B).
We next checked the efficacy of AZD-4547 and ARS-1620 combination on cell viability in KRASG12C lung cancer cell lines. Inhibition of FGFR1 signaling using FGFR kinase inhibitor (FGFRi), AZD-4547, inhibited tyrosine phosphorylation of FRS2 as expected (Fig. 5B). AZD-4547 effectively enhanced the sensitivity of ARS-1620 in H1792 (Fig. 5C and D; Supplementary Fig. S6B). Western blotting indicated that inhibition of FGFR1 delayed p-ERK rebound; however, decrease in p-AKT was not observed. Furthermore, combining AZD-4547 with ARS-1620 repressed mTOR signaling as indicated by decrease in RPS6 phosphorylation (Fig. 5E). Notably, we observed similar observations with AZD-4547 combination on cell viability and signaling in another mesenchymal model, LU99 (Fig. 5F–H). siRNA-based knockdown of FGFR1 in combination with KRASi also suppressed cell viability, and decreased ERK and RPS6 phosphorylation (Fig. 5I and J). Next, we checked the impact of exogenous FGF2 stimulation on KRASi sensitivity and signaling. FGF2 stimulation reduced the sensitivity of H358 cells to ARS-1620 and only moderate suppression in p-ERK was observed with ARS-1620 (Fig. 5K).
To more directly determine the impact of EMT on sensitivity to KRASG12C inhibition and different combination strategies, H358 cells were chronically treated with TGFβ (41). The induction of EMT, as indicated by enhanced expression of vimentin and suppressed expression of CDH1 (Fig. 5L), reduced the sensitivity to ARS-1620 (Fig. 5M). Our results indicate substantial rewiring of RTKs during EMT in H358 cells, where we observed reduced expression of HER2 and HER3, and enhanced expression of FGFR1 and FRS2 in the mesenchymal cells. Of note, this observation was consistent with the cell viability analysis where EMT-induced H358 cells become sensitive to both signal-agent treatment with FGFR1 inhibitor and in combination with ARS-1620 (Fig. 5N). Consistently, we observed decreased sensitivity to pan-HER inhibitor when treated standalone and with ARS-1620 combination. Consistent with our earlier results, sensitivity to dual SHP2i/KRASi and SOSi/KRASi combination treatments also reduced in the mesenchymal H358 cells (Fig. 5N). However, sensitivity to dual pan-HERi/KRASi and SHP2i/KRASi combinations remained unchanged when treated at higher doses of drugs combination (Supplementary Fig. S7A–S7D).
Next, we analyzed our proteomics and CCLE RNA-seq data (33) of KRASG12C cell lines for FGFR signaling signatures. We observed highest expression of FGF2 protein/gene expression in H1792 and LU99 cells (Supplementary Fig. S8; Supplementary Table S7). We also observed higher phosphorylation (Y502) and expression of FGFRL1 in H1792 in our proteomics data (Supplementary Fig. S8A). Next, we analyzed CCLE RNA-seq data which enabled inclusion of more KRASG12C models where a positive correlation was observed between FGFR1 and TGFβ-EMT score (r = 0.72; P = 0.002; Supplementary Table S7). Also, we observed a significant difference in the FGFR1 gene in two-group (epithelial vs. mesenchymal) comparison (Supplementary Fig. S8B). Finally, we asked whether FGFR1 expression in KRASG12C human lung cancer tumor tissues corelates with the mesenchymal subtype. We observed a positive correlation between FGFR1 and TGFβ-EMT score (r = 0.707; P = 4.04E-10; Fig. 5O; Supplementary Table S9). Our results show a significant role of FGFR1 signaling in mesenchymal type KRASG12C tumors and we show synergistic effects with dual FGFRi/KRASi combination treatment in a subgroup of KRASG12C models with mesenchymal type.
Phosphoproteomics identified AXL receptor–mediated adaptive rewiring to KRASG12C inhibition
Our results showed that inhibition of IGF1R and FGFR1 does not sensitize Calu1 to ARS-1620 treatment. In addition, combination with a pan-HER or SHP2 inhibitor only had moderate response (Fig. 6A). Immunoblot-based signaling analysis also indicated that SHP099 alone or in combination with ARS-1620 does not effectively suppress p-ERK expression (Fig. 6B). Our phosphoproteomics data identified 208 pY and 1,353 pSTY phosphosites that were modulated by ARS-1620. The MetaCore pathway analysis on phosphosites perturbed at 6 hours (Supplementary Fig. S9A) indicates inhibition of ERK signaling, yet other signaling pathways were upregulated. A well-connected signaling network of PI3K-AKT-mTOR was identified, which involves IRS1, IRS2, AKT1, AKT2, RPS6KB1, RPS6, and EIF4B among other nodes.
Resistance to combinations of ARS-1620 with drugs targeting HER, IGF1R, or FGFR in Calu1 led to more comprehensive reevaluation of perturbed phosphoproteome with manual curation of the signaling network based on differentially expressed phosphosites (Fig. 6C). This strategy also indicated activation of PI3K-AKT pathway. Interestingly, we identified increased phosphorylation of AXL (Y779). We also observed higher phosphorylation and expression (both protein and gene) of AXL in Calu1 when compared with other cell lines (Fig. 6D; Supplementary Table S7). For this reason, we sought to check whether inhibition of AXL can sensitize Calu1 to ARS-1620 treatment. The drug–response curves indicated that combination treatment with AXL inhibitor(s) enhances sensitivity to ARS-1620 in Calu1 cells. In addition, our data were in concordance with previous study, where PI3K inhibitor showed synergistic effects with KRASi in Calu1 and suppressed p-Akt expression (Fig. 6E and F; Supplementary Fig. S9B).
Next, we sought to identify whether AXL is the upstream signaling source to PI3K-AKT pathway in Calu1. Signaling analysis indicated that cabozantinib and foretinib alone or in combination with ARS-1620 were able to reduce p-AXL and p-Akt (Fig. 6G; Supplementary Fig. S9C). AXL inhibition using other small-molecule inhibitor, RXDX-106, alone or in combination with ARS-1620 was also able to reduce Akt phosphorylation in Calu1 cells (Fig. 6H), but not in H358 and H1792 cells (Supplementary Fig. S9D). The siRNA-based knockdown of AXL in combination with KRASi also suppressed cell viability (Fig. 6I), and decreased p-AKT expression (Supplementary Fig. S9E and S9F). Interestingly, ceritinib showed similar results, attributing to its potential of simultaneously inhibiting multiple pathways, including insulin, mTOR, adherens junction, and focal adhesion signaling (37). Combination with other RTK inhibitors (afatinib, AZD-4547, linsitinib, and imatinib) did not alter phosphorylation of Akt.
Next, we also determined whether AXL expression in KRASG12C human lung cancer tumor tissues associates with the mesenchymal subtype, similar to what we observed with positive association with FGFR1 and mesenchymal subtype. We observed a positive correlation (r = 0.75; P = 9.04E-12) between AXL and TGFβ-EMT score (Fig. 6J; Supplementary Table S9). Our data suggest another subtype within mesenchymal models where AXL inhibitor–mediated combination strategies with KRASi merit further investigation, and these results suggest the general utility of unbiased phosphoproteomics to elucidate adaptive resistance mechanisms to KRASi.
Although several combination regimens with KRASi have been suggested, our study describes the various mechanisms of signaling rebound and the mechanistic insight of these combination strategies (11–16). Our phosphoproteomics analysis indicated that each KRASG12C NSCLC model (e.g., H358, H1792, and Calu1 cells) underwent unique adaptive signaling events to KRASG12C inhibition. These results, along with signaling studies of tissues from KRAS-mutant genetically engineered mouse models (GEMMs), argue strongly that cell context will dictate responses to KRASG12C inhibition (42). Furthermore, comparison of proteomes and phosphoproteomes of eight different KRASG12C lung cancer models indicates that preexisting transcription states are likely to influence response to KRASi. As depicted in model (Fig. 7) describing signaling adaptations to ARS-1620 in NSCLC, the epithelial subtype typically adapted to KRASG12C-targeted agents by activating ERBB signaling. Conversely, the mesenchymal subtype showed activation of the FGFR or AXL signaling cascade. This analysis suggests a personalized combination strategy with HER2/3-targeting agents along with KRASi for epithelial subtypes, while FGFR or AXL inhibition should be considered for more mesenchymal tumors. We also observed similar differences in the transcriptome of other cell lines using CCLE data (33) and in KRASG12C tumor tissues taken directly from patients (34). Our results also raise the possibility to analyze the proteomes in rebiopsies of recurrent tumors following targeted therapy to capture adaptive resistance signaling and design personalized combination therapy.
The altered phosphorylation of SOS1, SHP2, and RAF1 showed rebound of MAPK signaling as one of the underlying mechanisms of resistance in the epithelial subtype. SHP2 inhibitors have been suggested as an effective combination strategy with KRASi to disrupt RAS activation by RTKs (16). Our results suggest that dual SHP2i/KRASi and SOS1i/KRASi strategies will provide more benefit to patients with epithelial subtype tumors and this should be examined in ongoing trials. We also observed increased phosphorylation of RAF1 (S621) following KRAS inhibition. Phosphorylation of RAF1 at S621, a site for PRKAA1, enhances its interaction with the cross-linker protein, 14-3-3, which stabilizes RAF1/BRAF heterodimers (28, 43). In the context to RAS-driven cancer, concomitant suppression of RAF1 and MEK leads to persistent ERK suppression and enhance apoptosis (44). Thus, development of specific RAF1 inhibitors and determination of their synergistic effects with KRASi on growth inhibition will be important.
The genetic ablation of IRS1 and IRS2 in KRAS-driven GEMMs of lung cancer suppresses Akt signaling and delays lung tumorigenesis (45). As compared with xenograft models, GEMMs reliably recapitulate the tumor pathology in epithelial cancers and allow better exploitation of signaling mechanisms to understand process of tumorigenesis (46, 47). Given the importance of IRS signaling downstream of ERBB and IGFRs following KRAS inhibition, our data explain the mechanistic insight of effective PI3K/AKT inhibitor combination strategies and provide a rationale for cotargeting HER2/3 or IGF1Rs along with KRASi in epithelial subtype of KRASG12C mutants.
Our analysis identified activation of FGFR1 and FGFRL1 signaling as one of the mechanism of therapy resistance in mesenchymal subtype (exemplified by H1792 cells), which provides mechanistic basis to cotarget FGFR and KRASG12C. FGFRL1 lacks an intracellular kinase domain and consists of an SH2 domain to facilitate binding to signaling adaptors. FGFRL1 was identified to interact with RASG12V (48) and can enhance ERK phosphorylation either by association with SHP-1 or by enhancing FGFR1 activation to induce MEK-independent activation of ERK1/2 (49). Because our analysis identified cell type–specific activation of FGFR1 signaling, future interrogations of tumors for FGFR pathway activation, possibly through FGFR proximity ligation assays (18) or MRM-based proteomics methods (50), may help in predicting personalized dual FGFRi/KRASi treatment strategy. On the basis of phosphoproteomic signatures, we identified activation of AXL signaling as an alternative mechanism of resistance in mesenchymal models. We identified higher FGF2 expression and FGFR1 phosphorylation on other NSCLC KRASG12C models, such as HCC44 and HOP62, but they did not respond to the FGFRi/KRASi combination. Thus, a more comprehensive phosphoproteomic analysis is required in these models to identify other mechanisms of resistance to KRASi. Our dataset is also likely to harbor other mechanisms and targets of adaptive resistance to KRASi. For example, we observed many perturbations on phosphosites associated with Wnt signaling proteins (CTNNB1, CTNNA1, CSNK1G3, and CSNK2B) and G protein-coupled receptors. Thus, phenotypic consequences of these perturbations in KRASG12C models remain important to investigate in further detail.
Our data highlighted the importance of mass spectrometry–based proteomics to read out system-wide changes in signaling events to understand the diversity of drug responses based on protein expression and posttranslational modifications. Our study provides a more comprehensive view of phosphoproteome alterations to KRASG12C inhibition in various lung cancer models with high clinical relevance to consider biomarker development strategies to suggest clinically effective combinations in diverse group of KRASG12C-mutant lung cancer. Furthermore, our analysis raises the possibilities to utilize applications of phosphoproteomics in biopsy/rebiopsy samples, organoids, and patient-derived xenograft models to aid personalized approach to cancer care.
E.B. Haura reports personal fees and nonfinancial support from Janssen Pharmaceuticals and personal fees from Revolution Medicines outside the submitted work. J.M. Koomen reports grants from NCI and other from NCI/Leidos during the conduct of the study. No disclosures were reported by the other authors.
H.S. Solanki: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration. E.A. Welsh: Data curation, software, formal analysis, investigation, methodology. B. Fang: Formal analysis, methodology. V. Izumi: Formal analysis, methodology. L. Darville: Formal analysis, methodology. B. Stone: Methodology. R. Franzese: Formal analysis, methodology. S. Chavan: Formal analysis, methodology. F. Kinose: Formal analysis, methodology. D. Imbody: Methodology. J.M. Koomen: Formal analysis, methodology, writing–review and editing. U. Rix: Writing–review and editing. E.B. Haura: Conceptualization, supervision, funding acquisition, project administration.
We thank A.C. Tan for useful discussions in designing TMT experiments. We thank Daniel Chen for his assistance in article revision. The work was supported by the State of Florida Bankhead Coley Grant (5BC07) and the Moffitt Lung Cancer Center of Excellence. This work has been supported, in part, by the Proteomics & Metabolomics Core Facility and the Biostatistics and Bioinformatics Shared Resource at the H. Lee Moffitt Cancer Center & Research Institute, an NCI-designated Comprehensive Cancer Center (P30-CA076292). LC-MRM assays for cancer signaling proteins were developed with funding from the NCI CPTAC and RAS Initiative through Leidos (14 × 270 to J.M. Koomen).
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.