SOS1 Inhibition Enhances the Efficacy of KRAS^G12C Inhibitors and Delays Resistance in Lung Adenocarcinoma

Abstract

The clinical effectiveness of KRAS^G12C inhibitors (G12Ci) is limited both by intrinsic and acquired resistance, necessitating the development of combination approaches. Here, we identified targeting proximal receptor tyrosine kinase (RTK) signaling using the SOS1 inhibitor (SOS1i) BI-3406 as a strategy to improve responses to G12Ci treatment. SOS1i enhanced the efficacy of G12Ci and limited rebound RTK/ERK signaling to overcome intrinsic/adaptive resistance, but this effect was modulated by SOS2 protein levels. G12Ci drug tolerant persister (DTP) cells showed up to a 3-fold enrichment of tumor initiating cells (TIC), suggestive of a sanctuary population of G12Ci resistant cells. SOS1i re-sensitized DTPs to G12Ci and inhibited G12C-induced TIC enrichment. Co-mutation of the tumor suppressor KEAP1 limited the clinical effectiveness of G12Ci, and KEAP1 and STK11 deletion increased TIC frequency and accelerated the development of acquired resistance to G12Ci, consistent with clinical G12Ci resistance seen with these co-mutations. Treatment with SOS1i both delayed acquired G12Ci resistance and limited the total number of resistant colonies regardless of KEAP1 and STK11 mutational status. Together, these data suggest that targeting SOS1 could be an effective strategy to both enhance G12Ci efficacy and prevent G12Ci resistance regardless of co-mutations.

Introduction

Lung cancer is the leading cause of cancer-related death (1,2). Oncogenic driver mutations in the RTK/RAS pathway occur in 75-90% of lung adenocarcinoma (LUAD), with 30-40% being driven by mutations in KRAS. Approximately 13% of LUAD harbor KRAS^G12C mutations (1,2). While KRAS was once considered constitutively active and undruggable, KRAS^G12C mutant proteins maintain a rate of hydrolysis sufficient for nucleotide exchange (3). This nucleotide cycling between the GTP-bound on and GTP-bound off states, combined with the identification of a unique and targetable binding pocket in KRAS^GDP (4), afforded the unique opportunity to target KRAS^G12C with agents that specifically bind KRAS in the GDP-bound off state and covalently modify the mutated cystine to lock mutant KRAS^G12C in the inactive GDP-bound state (4). Pre-clinical development of therapeutics based on this principle showed potent tumor suppression in cellular and animal models (5,6), culminating in the FDA approval of the KRAS^G12C inhibitors AMG-510/sotorasib (7) and MRTX849/adagrasib (8).

Unfortunately, intrinsic and acquired resistance to both adagrasib and sotorasib limit the clinical benefit in patients with KRAS^G12C-mutated LUAD. Across multiple published clinical trials, previously treated patients receiving adagrasib or sotorasib monotherapy showed response rates of 34-43% (9–12), indicating that intrinsic resistance will limit G12Ci efficacy for more than half of patients with KRAS^G12C-mutated LUAD. Intrinsic G12Ci resistance can be attributed to both pre-existing co-mutations that prevent cancer cells from responding to G12Ci (10,13,14) and rapid rewiring of intracellular signaling that reactivates RAS effectors to bypass the effects of G12Ci. This reactivation of RAS effector signaling, known as adaptive resistance, is driven both by relief of ERK-dependent negative feedback of RTKs-SOS-RAS signaling (15–18) and by pathway activation through newly translated KRAS^G12C proteins that are not yet inhibited by G12Ci (19).

Genomic analysis of >400 patients treated with adagrasib or sotorasib found that concurrent loss-of-function (LOF) mutations in KEAP1, SMARCA4, and CDKN2A are associated with inferior clinical outcomes and progression free survival (PFS) ≤ 3 months, indicative of intrinsic resistance (13). Although not associated with inferior overall outcomes, STK11 mutations were also enriched in patients showing PFS ≤ 3 months (13). KRAS-mutated cancers can be broadly categorized into three groups based on co-mutations of the tumor suppressors TP53, CDKN2A, or STK11 (20). Of these, tumors with STK11 mutations show the highest frequency of KEAP1 mutations (20). Patients with KRAS mutant tumors harboring STK11 co-mutations (21), KEAP1 co-mutations (22), or both KEAP1 and STK11 co-mutations (23) respond poorly to both conventional chemotherapy and immunotherapy and are enriched in LUAD proteogenomic subtypes with poor overall survival (24). Interestingly, the poor immunotherapy responses for patients whose tumors harbor KEAP1/STK11 co-mutations were only observed for patients with KRAS-mutated, but not KRAS WT, tumors (23). This finding highlights an important unmet therapeutic need for this patient population.

For patients whose tumors do initially respond to G12Ci, acquired resistance rapidly develops with a mean progression free survival between 5-14 months for patients receiving adagrasib or sotorasib (9–14). Acquired resistance is often associated with secondary mutations in upstream RTKs, KRAS, NRAS, or downstream RAS effector pathway (3,25,26), although, for roughly half of patients, no definitive driver mutation was observed (3,25). As tumors adapt to therapeutic pressure, a proportion of the bulk tumor, known as drug-tolerant persister cells (DTPs) (3,27–29), shows chromatin remodeling that allows cells to survive under drug pressure. These adaptations include broad up-regulation of RTK signaling (29), enhanced ability to handle redox stress (30–33), and entering a near-quiescent state (28,29,34). A subset of the DTP population, known as tumor initiating cells (TICs), has stem-like properties and is capable of asymmetric division and self-renewal. This population likely represents the therapeutic sanctuary of cells responsible for tumor re-growth after the acquisition of additional mutations leading to acquired resistance (3,27,35–37). The idea that the bulk tumor is largely composed of non-TICs is critical to the understanding of acquired resistance, since initial therapeutic efficacy and killing of the bulk population can mask the survival of the small population of TICs that survive and adapt under therapeutic pressure to drive acquired resistance.

SHP2, SOS1, and SOS2 are proximal RTK signaling intermediates critical for activation of both mutated and wild type RAS proteins; deletion of Ptp11/Shp2 (38) or Sos1 (39) inhibited tumor development in mutant KRAS-driven animal models, and Sos2^KO reduced mutant KRAS-driven transformation (40). The phosphatase SHP2 acts as an adaptor to recruit the RASGEFs SOS1 and SOS2 to the GRB2/GADs complex, allowing for RAS activation at the plasma membrane. While there are currently no SOS2 inhibitors, SHP2i (RMC-4550, SHP099, TNO155) (15–17,41–43) and SOS1i (BAY-293, BI-3406, and MRTX0902) (44–47) should help overcome intrinsic G12Ci resistance by two distinct mechanisms. First, since G12Cis bind KRAS^G12C in the GDP-bound (off) state, SHP2 (15–17,41–43) and SOS1 (44–47) inhibitors enhance G12Ci potency by increasing the amount of KRAS^G12C available for G12Ci binding and inhibition (46). Second, SHP2i (15–17,41,43), SOS1i (17,46,47), or SOS2^KO (40,48) can inhibit adaptive resistance to both G12Ci and MEKi driven by relief of ERK-dependent negative feedback of RTKs-SOS-RAS signaling.

Here, we show that the SOS1i BI-3406 inhibits both intrinsic and acquired G12Ci resistance in LUAD cells. To limit intrinsic resistance, SOS1i both enhanced the efficacy of multiple G12Cis and prevented G12Ci-induced RTK/ERK re-activation associated with adaptive resistance. To limit acquired resistance, SOS1i re-sensitized DTPs to G12Ci and limited survival of G12Ci-induced TICs. SOS1i further delayed acquired G12Ci resistance and limited the total number of resistant colonies, even in cells with KEAP1 and/or STK11 inactivation. These data suggest that SOS1i could be an effective strategy to both enhance G12Ci efficacy and prevent G12Ci resistance in KRAS-mutated LUAD.

Materials and Methods

Cell Culture

Lung cancer cell lines were purchased from ATCC; NCI-H358 (RRID:CVCL_1559), NCI-H1373 (RRID:CVCL_1465), NCI-H1792 (RRID:CVCL_1495), NCI-H2030 (RRID:CVCL_1517), NCI-H23 (RRID:CVCL_1547). We acknowledge Drs. Gazdar and JD Minna and the HHS (Department of Health and Human Services) for depositing these cells at ATCC. Once cell lines were received, they were expanded and frozen at passage 3 and 4; cells were passaged once they became approximately 80% confluent. Cells were maintained in culture for 2-3 months before a new vial was thawed since prolonged passaging can alter TIC frequency. The panel of H358 NT, STK11 ^KO, KEAP1^KO, and STK11/KEAP1 DKO cells were a generous gift from Charles Rudin (Memorial Sloan Kettering Cancer Center) (49). Cell lines were cultured at 37°C and 5% CO2. All cells were passaged in RPMI supplemented with 10% FBS and 1% penicillin/streptomycin. For 3D signaling experiments, cells were seeded in a 24-well micropatterned Aggrewell 400 low-attachment plates (StemCell) at 9 × 10⁵ cells/well in 2 mL of medium. After 24h post-plating, 1 mL of media was removed to be replaced with either new media or 2× inhibitor. Cells were treated for 0h-72h ours in the same manner, replacing 1 mL of media every day with either new media or 2× inhibitor, or fresh 1× inhibitor for those that previously were drugged. These cell lines did not undergo authentication and were not tested for mycoplasma.

To generate 21d DTPs, H358 cells were seeded at 25% confluence. 24h later, cells were treated with 30 or 100 nM adagrasib. Cells were monitored for confluence, and passaged once at day 12. At 21d, cells were either lysed or rested in the absence of G12Ci for 3d prior to assessment of G12Ci sensitivity.

For Washout studies, cells were treated with G12Ci for 72h, washed 2× with PBS and then cultured in media without G12Ci for 0 – 72h.

Generation of SOS2^KO cell lines

For SOS2^KO studies, H358, H1373, H1792, and H2030 cells were infected with lentiviruses based on pLentiCRISPRv2 (RRID:Addgene_52961) with either a non-targeting sgRNA (NT) or a sgRNA targeting SOS2 (40,50). 48h post-infection, 4 μg/mL Puromycin (Invitrogen) was used to select cells. These cells were then analyzed for SOS1 or SOS2 expression after 7-10 days post-selection and only those cultures showing >80% deletion were used for the studies.

Dose response curves

Cells were seeded at 750 cells per well in 100 μL in the inner-60 wells of 96-well ultra-low attachment round bottomed plates (FaCellitate BIOFLOAT # F202003) and allowed to coalesce as spheroids for 24-48h prior to drug treatment. For all studies, peripheral wells (rows A and H, columns 1 and 12) were filled with 200 μL of PBS to buffer inner wells and prevent evaporative effects on cells. Triplicate wells were then treated with increasing doses of G12Ci adagrasib, G12Ci sotorasib, or MEKi trametinib for 72h on a semi-log scale and assessed for cell viability using CellTiter-Glo 2.0 (30 μL/well). Luminescence was assessed using a Bio-Tek Cytation five multi-mode plate reader. Data were normalized to the maximum luminescence reading of untreated cells, and individual drug EC₅₀ values were calculated using Prism9 by non-linear regression using log(inhibitor) vs. response. For all drug-treatment studies, the untreated sample for each cell line was set to 100%. This would mask any differences in 3D cell proliferation seen between cell lines. For single dose studies in untreated vs. 3d or 21d DTPs, cells were treated with G12Ci alone or in combination with the SOS1i BI-3406 (300 nM) or the ALDHi disulfram (10 nM).

Bliss Independence Analysis and MuSyC for Synergy

Cells were seeded at 750 cells per well in 100 μL in the inner-60 wells of 96-well ultra-low attachment round bottomed plates (FaCellitate BIOFLOAT # F202003) and allowed to coalesce as spheroids for 24-48h prior to drug treatment. For all studies, peripheral wells (rows A and H, columns 1 and 12) were filled with 200 μL of PBS to buffer inner wells and prevent evaporative effects on cells. Triplicate wells were then treated with increasing doses of the G12Ci adagrasib or the SOS1i BI-3406 alone or a combination of adagrasib + BI-3406 in a 9 × 9 matrix of drug combinations on a semi-log scale for 96h. and assessed for cell viability using CellTiter-Glo 2.0 (30 μL/well). Luminescence was assessed using a Bio-Tek Cytation five multi-mode plate reader. Data were normalized to the maximum luminescence reading of untreated cells, and individual drug EC₅₀ values were calculated using Prism9 by non-linear regression using log(inhibitor) vs. response. For all drug-treatment studies, the untreated sample for each cell line was set to 100%. This would mask any differences in 3D cell proliferation seen between cell lines. Excess over Bliss was calculated as the Actual Effect – Expected Effect as outlined in (51). The SUM EOB is calculated by taking the sum of excess over bliss values across the 9 x 9 treatment matrix. EOB values > 0 indicate increasing synergy. To deconvolute synergistic efficacy versus potency, data were analyzed by Multi-dimensional Synergy of Combinations (MuSyC) Analysis (52,53) using an online tool (https://musyc.lolab.xyz). Data are presented as a fold change in G12Ci potency (log α₂) versus % change in G12Ci efficacy (β_obs).

Cell lysis and Western blotting

To prepare for cell lysis, cells were washed twice with cold PBS before RIPA buffer (1% NP-40, 0.1% SDS, 0.1% Na-deoxycholate, 10% glycerol, 0.137 M NaCl, 20 mM Tris pH [8.0], protease (Biotool #B14002) and phosphatase (Biotool #B15002) inhibitor cocktails) was added to cells for 20 minutes at 4°C and spun at 10,000 RPM for 10 min. Supernatant was collected and protein concentration was assessed via protein assay. Normalized lysates were then boiled in SDS sample buffer containing 100 mM DTT for 10 min prior to Western blotting. Proteins were resolved by sodium dodecyl sulfate-polyacrylamide (Criterion TGX precast) gel electrophoresis and transferred to nitrocellulose membranes. Western blots were developed by multiplex Western blotting using anti-SOS1 (Santa Cruz sc-256; 1:500; RRID:AB_632417), anti-SOS2 (Santa Cruz sc-258; 1:500; RRID:AB_2192448), anti-β-actin (Sigma AC-15; 1:5,000; RRID:AB_476692 or Santa Cruz Biotechnology sc-47778, 1:2000 dilution; RRID:AB_626632), anti-pERK1/2 (Cell Signaling 4370; 1:1,000; RRID:AB_2315112), anti-ERK1/2 (Cell Signaling 4696; 1:1000; RRID:AB_390780), anti-KRAS (ProteinTech 12063-1-AP; 1:1000; RRID:AB_878040), anti-pMEK (Cell Signaling 9154; 1:1000 RRID:AB_2138017 anti-MEK (Cell Signaling 4694; 1:1000 RRID:AB_390778), anti-DUSP6 (Santa Cruz 377070; 1:1000 RRID:AB_2802089), anti-AXL (Cell Signaling 8661; 1:1000; RRID:AB_11217435), anti-pIGF-1R (Cell Signaling 3918; 1:1000; RRID:AB_10548764), anti-IGF-1R (Cell Signaling 9750; 1:1000; RRID:AB_10949773), anti-tubulin (ProteinTech 66031-1-IG; 1:20,000; RRID:AB_11042766), anti-HSP90 (SC 515081; 1:500; RRID:AB_2943315), and anti-HIF1α (Cell Signaling 36169; 1:1000; RRID:AB_2799095) primary antibodies. Anti-mouse and anti-rabbit secondary antibodies conjugated to IRDye680 (Goat anti-Mouse IgG RRID:AB_10956588; Goat anti-Rabbit RRID:AB_10956166) or IRDye800 (Goat anti-Mouse IgG RRID:AB_621842; Goat anti-Rabbit RRID:AB_621843) (LI-COR; 1:20,000) were used to probe primary antibodies. Western blot protein bands were detected and quantified using the Odyssey system (LI-COR).

Transcriptome profiling

Gene expression (RNA-seq) analysis was performed as previously described (44,46). H358 cells were seeded in 6-well micropatterned Aggrewell 400 low-attachment plates (StemCell) at 3.5 × 10⁶ cells/well in 4 mL of medium. Half of the media was removed and fresh media was added daily. 72h post-plating, 2 mL of media was removed to be replaced with 2 × SOS1i, G12Ci, or SOS1i + G12Ci and incubated for either 6h or 72h. Cells were collected, pelleted, resuspended in RLT plus buffer (Invitrogen), and homogenized using a QIAshredder column (Qiagen). RNA was isolated using an RNeasy spin column (Qiagen) according to the manufacturer’s instructions. After elution, RNA was quantified with the Quant-iT Ribogreen RNA Kit according to the manufacturer’s instructions (Invitrogen). All samples were normalized to 20 ng/μL, aliquoted and frozen at −80°C For each sample, a total RNA input of 200 ng was used for library preparation using the Illumina Stranded Total RNA Prep with Ribo-Zero Plus Kit (Illumina) and IDT for Illumina RNA UD Indexes Set A, Ligation (96 Indexes, 96 Samples). Sequencing libraries were quantified by real-time PCR using KAPA Library Quantification Kit for NGS (Roche) and assessed for size distribution on a Fragment Analyzer (Agilent) to confirm absence of adapter dimers and unligated library molecules. Sequencing libraries were pooled at <40-plex and sequenced on a NovaSeq 6000 (Illumina) using a 200 cycle SBS kit (Illumina) with paired-end reads at 101 bp length targeting >60 million reads per sample.

After sequencing, the Nextflow nf-core/rnaseq (version 3.4) workflow was used to estimate a matrix of sample gene expression values using the reference human genome (GRCh37; Ensembl release 75) obtained from Illumina’s iGenomes resource (https://support.illumina.com/sequencing/sequencing_software/igenome.html). Briefly, paired-end DNA sequence FastQ files were merged by sample (two/sample) and underwent read quality control (FastQC v0.11.9) prior to adapter and quality trimming (Trim Galore! v0.6.7), then input into Salmon’s (v1.5.2) mapping-mode for gene expression matrix quantification. Further sequencing quality control was performed via subsequent trimmed read alignment using STAR (v2.6.1d), SAMtools (v1.13), and Picard (v2.25.7, http://broadinstitute.github.io/picard/), followed by RSeQC (v3.0.1), Qualimap (v2.2.2-dev), and Preseq (v3.1.1, https://smithlabresearch.org/software/preseq/). After confirming the absence of sequencing or sequencing library anomalies, we proceeded to differential expression analysis.

Gene set enrichment analysis was performed using fgsea R/Bioconductor (v1.14) package and hallmark gene sets from the molecular signatures database (MSigDB v7.5.1) (54). The resulting nominal p values were adjusted using Benjamini-Hochberg multiple testing correction method and gene sets with adjusted p value < 0.01 were considered as significant. Estimation of individual transcription factor activities based on ‘footprint based’ signatures for individual transcription factors were performed using the Discriminant Regulon Expression Analysis (DoRothEA) R/Bioconductor package and Visualization Pipeline for RNA-seq (VIPER) analysis (55). The Pathway RespOnsive GENes (PROGENy) algorithm was used to infer activation of 14 key cancer pathways (56). MAPK Pathway Activity (MPAS) scores were calculated based on expression of the 10-MAPK gene signature (57). Heatmaps of GSEA analysis were generated using ComplexHeatmap (v2.4.3) or Prism 9. All other figures from data analyses were visualized using ggplot2 (v3.3.2) R package.

Flow cytometry

Cells were plated in 6 cm tissue-culture treated plates and allowed to adhere for 24h prior to adding G12Ci at the indicated doses for 72h. After 72h of treatment, cells were harvested by trypsinization, spun down, resuspended in Aldefluor Assay Buffer (StemCell) at 1*10⁶ cells/mL, and stained for ALDH activity using the Aldefluor Assay Kit per manufacturer’s instructions+/−APC-conjugated antibodies to CD44 (Biolegend #397506), CXCR4 (Biolegend #306510), IL-6Rα (Biolegend #352806), or CD133 (Biolegend #3728060) or isotype controls (Biolegend # 400122). Three tubes of untreated cells were collected so one could be treated with DEAB, an ALDH inhibitor, to use as a negative gating control during analysis, and a second tube could be used for isotype control for APC-conjugated antibodies. Data was analyzed using FloJo with and are presented as the percentage of cells showing ALDH activity over DEAB controls and/or APC staining over isotype controls. Cell isolation was performed on a FACSaria, with gaiting based on either unstained sample, single antibody-stained samples, and/or DEAB controls.

Extreme Limiting Dilution Assays

For in situ assays, cells are seeded in a 96-well ultra-low attachment flat bottomed plates (Corning Corstar #3474). Cells are plated at decreasing cell concentrations at half-log interavals, beginning with 1000 cells/well – 1 cell/well, with 12 wells/cell concentration except 10 cells/well, where 24 wells are seeded. To assess the effect of G12C inhibition on TIC frequency, cells were left untreated or were pre-treated with the indicated doses of G12Ci for 72h, after which drugged media was removed and replaced with fresh media for 72h prior to plating. To assess the effect of SOS1i on TICs, cells were seeded ± 300 nM BI-3406. Cells were then allowed to grow for 7-10 days before being assessed for spheroid growth (>100 mM was scored as positive). The TIC frequency and significance between the different groups was calculated by ELDA website (RRID:SCR_018933) https://bioinf.wehi.edu.au/software/elda/ (58).

Resistance Assays

In situ Resistance Assays were performed as previously described (59,60). Briefly, cells were plated in 96-well tissue-culture treated plates at low-density (250 cells/well). Each plate was treated with a single dose of a G12C inhibitor ± BI-3406. Wells were assessed weekly for outgrowth and those that were >50% confluent were marked as resistant to that given dose. Plates that did not reach full confluency were fed weekly until an indicated endpoint was reached. Data are plotted as a Kaplan-Meyer survival curve; significance was assessed by comparing Kaplan-Meyer curves using Prism 9.

Animal studies

All animal studies were approved by the USUHS and Crown Biological IACUC Committees. Animal studies were performed at Crown Biological (San Diego, CA). Group size in efficacy studies have been selected after performing power analysis. 5 or 10 million cells were injected subcutaneously into NOD/SCID mice (NOD.Cg-Prkd^scid/J, Jackson Laboratory strain # 001303; RRID:IMSR_JAX:001303). 9-week-old female mice were randomized in groups (n=3 mice per treatment group each for mice injected with 5 million and 10 million cells) once tumors reached a size greater than 140 μm³. Compound treatment was initiated after randomization. Mice were treated with adagrasib (100 mg/kg/day) +/− BI-3406 (50 mg/kg BID). BI-3406 and adagrasib were dissolved in 0.5% Natrosol. The control group was treated with 0.5% of Natrosol orally in the same frequency as in treated groups. Tumor size was measured by an electronic caliper on days 3, 7, and 10 and calculated using the following equation: (longest diameter * shortest diameter²)/2 or (length*width²)/2; body weight and well-being of the mice was monitored daily.

Data availability

The RNA sequencing data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus (RRID:SCR_005012) and are accessible through GEO Series accession number GSE275534 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE275534). The remaining data generated in this study are available upon request from the corresponding author.

Results

SOS1i synergizes with G12Ci to drive transcriptional changes regulating MAPK and hypoxia pathways

G12Cis including ARS-1620, sotorasib, and adagrasib bind KRAS^G12C in the GDP-bound state. Thus, secondary treatments that enhance the level of KRAS-GDP should increase the amount of targetable KRAS^G12C, thereby enhancing the efficacy of G12Ci (17,61,62). SOS1 is a ubiquitously expressed RASGEF and contributes to the continued activation of mutated KRAS^G12C; SOS1 inhibitor MRTX0902 inhibitor is currently under investigation in combination with the G12Ci adagrasib in a clinical trial [NCT05578092]. We found in KRAS^G12C-mutated H358 LUAD cells heterozygous for the KRAS^G12C allele, the SOS1i BI-3406 synergized with both pre-clinical (ARS-1620) and clinical (sotorasib, adagrasib) G12Cis across a 9 × 9 matrix of SOS1i:G12Ci doses as assessed by an increase in the excess over Bliss across the treatment matrix in 3D spheroid (Fig. 1A–B and S1A–B) and 2D adherent cultures (Fig. S1C–D). Further deconvoluting the contributions of synergistic potency vs. efficacy showed that the major contribution of SOS1i was to enhance the potency of G12Ci, consistent with the proposed mechanisms of action for both inhibitors (Fig. 1C and S1E–F). We continued our analysis of G12Ci:SOS1i combination therapy in 3D spheroid cultures, as responses to G12Ci in 3D more accurately represent responses in vivo compared to 2D adherent cultures (5).

Figure 1.

SOS1i synergizes with G12Ci to drive transcriptional changes regulating MAPK and hypoxia pathways.

A-C, Heat map of cell viability (top) and excess over Bliss (bottom) for H358 cells treated with a 9×9 matrix of the indicated G12Ci ± SOS1i (A), sum of excess over Bliss values (B) or quantitation of the change in G12Ci potency (log α₂) vs. efficacy (β_obs) (C). D-K, RNA sequencing of 3D spheroid cultured H358 cells treated with G12Ci (10 nM) ± SOS1i (100 nM) for 6h or 72h. MPAS score (D); Venn Diagrams from GVSA analysis of Hallmark MsigDB gene sets (E-F); gene set enrichment (G) and GVSA analysis (H) of hypoxia associated genes; VIPR scores of selected transcription factor signatures (I); heat maps of GVSA analysis of selected MsigDB hypoxia signatures (J) and PROGENy scores (K). L-M, Western blots of WCLs from 3D cultured H358 cells treated with G12Ci ± SOS1i for the indicated times.

* p < 0.05, *** p< 0.001 for G12Ci treated vs. untreated; ### p < 0.001 for SOS1i treated vs. untreated. ^ p < 0.05, ^^ p < 0.01 for G12Ci vs. G12Ci + SOS1i treated for 72h.

A-C, L-M are from three independent experiments.

We next focused further assessment on the adagrasib:BI-3406 combination, as the adagrasib:SOS1i MRTX0902 combination is currently under investigation in a clinical trial. We treated H358 cells cultured as 3D spheroids with G12Ci ± SOS1i for 6h or 72h and assessed overall changes in the signaling environment by RNA sequencing. Mutant KRAS signals through the RAF/MEK/ERK cascade to drive proliferation, thus we first assessed the MAPK Pathway Activity Score (MPAS) as a global indicator of on-target transcriptional changes as well as changes in transcription of key MAPK-regulated genes (63). At 6 hours, SOS1i had a more significant impact on MAPK-regulated transcriptional activity compared to G12Ci alone, likely due to the role of SOS1 as a GEF for both mutated and wild-type RAS proteins (Fig. 1D and S2). Further, cells treated with the combination of SOS1i + G12Ci showed enhanced inhibition of MAPK-dependent transcriptional changes compared to either drug alone at both 6 and 72h.

To globally assess G12Ci ± SOS1i dependent transcriptional changes, we performed gene set enrichment analysis (GSEA) first using the 50 MSigDB hallmark gene sets (54) to identify biological processes regulated by SOS1i and G12Ci individually and in combination. At 6h, there were 11 gene sets significantly upregulated by G12Ci and/or SOS1i and only two gene sets significantly down-regulated by SOS1i alone (Fig. 1E). At 72h there were no upregulated gene sets and 15 down-regulated gene sets (Fig. 1F). Further focusing on down-regulated gene sets, we found that pathways associated with hypoxia and myogenesis were significantly downregulated by SOS1i alone (Fig. 1E), with hypoxia-associated genes showing a significant enrichment in untreated compared to SOS1i treated samples (Fig. 1G). Hypoxia pathways remained downregulated by G12Ci ± SOS1i treatment at 72h (Fig. 1F), and gene variant set analysis (GVSA score) showed that hypoxia-regulated genes were more downregulated in SOS1i+G12Ci treated cells compared to cells treated with G12Ci alone at 72h (Fig. 1H). Assessment of transcription factor signatures associated with early (ETS1, EGR1) or prolonged (MYC, JUN) EGFR/MAPK signaling showed that early transcriptional changes (ETS1, EGR1) were decreased by SOS1i alone at 6h, and this inhibition was enhanced by combined G12Ci + SOS1i at 6 and 72h (Fig. 1I). Late EGFR/MAPK regulated transcriptional changes (MYC, JUN) were only downregulated by G12Ci or G12Ci+SOS1i at 72h. In contrast, gene signatures associated with HIF1α (hypoxia signaling) or E2F1 as a marker of proliferative signaling changes were significantly changed at 72h treatment by combined G12Ci+SOS1i compared to G12Ci alone (Fig. 1I). We further assessed the GVSA for additional gene sets associated with hypoxia and found that 9/13 gene sets that were enriched in cells under hypoxic conditions were significantly decreased by SOS1i+G12Ci treatment for 72h, with all 13 showing a more marked decrease in GVSA score for samples treated with SOS1i+G12Ci compared to G12Ci alone (Fig. 1J).

We further assessed transcriptional changes in fourteen key cancer-related signaling pathways using PROGENy (56). Transcription of genes associated with EGFR and MAPK signaling was decreased by SOS1i at 6h and either SOS1i or G12Ci at 72h (Fig. 1K). Combined SOS1i + G12Ci further decreased transcription of genes associated with EGFR, JAK-STAT, and MAPK signaling at both 6 and 72h compared to single-drug treated cells, while SOS2^KO showed further pathway inhibition in the presence of a SOS1i (Fig. 1K). In contrast, the PROGENy score for genes associated with hypoxia was only significantly decreased in samples from cells treated with a SOS1i (± G12Ci), but not in samples from cells treated with a G12Ci alone (Fig. 1K). Inhibition of KRAS/ERK signaling and HIF1α abundance by combined G12Ci + SOS1i was confirmed by Western blotting (Fig. 1L and S3). Combined G12Ci + SOS1i showed decreased MEK/ERK activation, DUSP6 protein abundance, and HIF1α protein abundance compared to G12Ci alone in both 3D spheroid and 2D adherent cultures (Fig. 1L and S3); this corresponded to increased KRAS-G12Ci engagement in G12Ci + SOS1i compared to G12Ci treated cells as assessed by gel shift (Fig. 1L and S3). We further observed that HIF1α rebound was RAF/MEK/ERK pathway dependent rather than a specific KRAS effect. HIF1α protein abundance was inhibited by either a G12Ci or the MEKi trametinib, but HIF1α levels rebounded at 72 hours; HIF1α rebound was blunted by SOS1i in both G12Ci and MEKi treated cells (Fig. 1M). These results suggest a role for SOS1i in regulating both MAPK-dependent signaling associated with G12Ci intrinsic/adaptive resistance and in hypoxia pathways that are associated with the drug-tolerant persister state (64,65).

SOS1i synergizes with and prevents adaptive resistance to G12Ci

We expanded our evaluation of combined SOS1i + G12Ci in a panel of KRAS^G12C-mutated cell lines. Similar to H358 cells, SOS1i synergistically enhanced the potency of the G12Ci adagrasib in 3D spheroid-cultured H1373 and H23 cells, but not in H1792 or H2030 cells (Fig. 2A–C and S4A–D). In contrast, we observed strong synergy between SHP2i + G12Ci across all four cell lines (Fig. 2A–C and S5A–C). The phosphatase SHP2 is required for full activation of the MAPK pathway, and SHP2i can serve as a proxy for combined SOS1/SOS2 inhibition (16,17,41,66). Thus, we hypothesized that SOS2 signaling may compensate for SOS1i in H1792 and H2030 cells. To directly assess the effects of SOS2 on G12Ci:SOS1 synergy, we took two complementary approaches; we diminished the contribution of RTK-SOS2 signaling by treating cells with G12Ci ± SOS1i under either low serum conditions to reduce the overall level of RTK-SOS2 signaling (40,48) (Fig. S6A–C) or in SOS2^KO cells (Fig S7A–C). In both H1792 and H2030 cells, SOS1i:G12Ci synergy was restored in 2% serum or SOS2^KO (Fig. 2B–C). Co-mutations in the tumor suppressors CDKN2A and KEAP1 occur more commonly in patients with KRAS^G12C-mutated tumors showing intrinsic resistance to adagrasib and/or sotorasib (13,25). However, since H1373 and H1792 harbor inactivating CDKN2A mutations and H1792, H2030, and H23 harbor inactivating KEAP1 mutations (Fig. S4B), co-mutational status alone does not result in the variation seen in SOS1i:G12Ci synergy. We therefore assessed the extent to which the relative ratio of SOS1:SOS2 protein abundance determined the requirement for inhibition of SOS2 signaling to promote SOS1i:G12Ci synergy (Fig. 2D–E). Indeed, KRAS^G12C-mutated LUAD cell lines showed differential SOS1 vs. SOS2 protein abundance. H1792 and H2030 cells showing higher SOS2 protein abundance compared to H358, H1373, and H23 cells (Fig. 2D), and a plot of the relative SOS1:SOS2 ratio versus total EOB for cells treated with SOS1i+G12Ci in 10% serum showed a clear relationship between SOS1:SOS2 protein abundance and SOS1i:G12Ci synergy (Fig. 2E).

Figure 2.

SOS2 expression determines the extent of G12Ci:SOS1i synergy.

A-C, Heat map of cell viability (top) and excess over Bliss (bottom) for the indicated cells treated with a 9×9 matrix of G12Ci ± SOS1i or SHP2i (A), sum of excess over Bliss values (B) or quantitation of the change in G12Ci potency (log α₂) vs. efficacy (β_obs) (C).

G12Ci:SOS1i in NT [squares cultured in 10% (closed) or 2% (open) serum] or SOS2^KO (blue circles) cells; G12Ci:SHP2i (purple diamonds). *** p < 0.001 vs. NT cells treated with G12CI + SOS1i in 10% serum. D-E, Western blots of WCLs for SOS1, SOS2, tubulin, and β-actin (D) and plot of the relative ratio of SOS1:SOS2 versus the sum of the EOB values for G12Ci:SOS1i treatment (E). Red circles EOB > 0; purple diamonds EOB ≤ 0. F, Western blots of WCLs of H358 in which SOS2 overexpression was driven by one of four distinct promoters. G-H, Sum of excess over Bliss (G) or quantitation of the change in G12Ci potency (log α₂) vs. efficacy (β_obs) (H) for cells from (F) treated with a 9×9 matrix of G12Ci ± SOS1i. ** p < 0.01, *** p < 0.001 vs. controls; # p < 0.05, ## p < 0.01 vs. cells expressing SOS2 driven by a mCMV promoter. I-J, Western blots of WCLs of 3D spheroid-cultured H358 or H1373 (I) or NT vs. SOS2^KO H1792 or H2030 cells (J) treated with G12Ci ± SOS1i for the indicated times. Data are from three independent experiments.

To determine if increased SOS2 protein abundance was sufficient to limit G12Ci:SOS1i synergy, we expressed increasing amounts of SOS2 using lentiviral vectors in which SOS2 is driven by one of four different promoters that increase SOS2 protein abundance 2- to 30-fold (Fig. 2F) (40). We observed a SOS2 dose-dependent decrease in G12Ci:SOS1i synergy as SOS2 protein abundance was increased (Fig. 2G–H and S8A–C).

The necessity for combined SOS1/2i to enhance G12Ci efficacy in H1792 and H2030 cells further extended to inhibition of rebound ERK activation following G12Ci treatment. 3D-cultured H358, H1373, H1792, and H2030 cells show rebound ERK phosphorylation 48-72h after G12Ci treatment (Fig. 2I–J), which has been attributed both to relief of ERK-dependent negative feedback of RTKs-SOS-RAS signaling (15–18) and by pathway activation through newly translated KRAS^G12C proteins that are not yet inhibited by G12Ci (19). In H358 and H1373 cells that show relatively low SOS2 protein abundance, SOS1i was sufficient to limit rebound ERK activation following G12Ci treatment (Fig. 2I), and G12Ci+SOS1i treated cells showed greater inhibition of MEK phosphorylation and DUSP6 abundance compared to G12Ci treated cells alone (Fig. S9). In contrast, in H1792 and H2030 SOS1i alone was not sufficient to block pERK rebound signaling. Instead, combined SOS1i + SOS2^KO was required to limit ERK rebound signaling following three days of G12Ci treatment in these cells (Fig. 2J). Taken together, these data demonstrate the relative abundance of SOS2 protein determines the extent to which SOS1i can enhance G12Ci efficacy to limit intrinsic/adaptive resistance in KRAS^G12C-mutated LUAD cells.

SOS1i targets G12Ci drug tolerant persister cells

In addition to overcoming intrinsic/adaptive resistance, an optimal therapeutic combination would further delay the development of acquired resistance. Prior to the acquisition of overt secondary mutations, a subset of the bulk tumor can undergo non-genetic adaptations that alter the intracellular redox environment and allow for continued survival in the face of therapeutic pressure (30). This 'drug tolerant persister’ population (DTP) is thought to act as a therapeutic sanctuary for cells to gain additional mutations and develop acquired resistance. Although no single marker fully defines DTP cells, increased aldehyde dehydrogenase (ALDH) activity has been used as a functional marker of cells with increased ability to detoxify the effects of therapy-induced oxidative stress to promote tumor cell survival (31–33). Although DTPs have been well characterized for EGFR-TKIs in LUAD (27,29), they have not been well documented after G12Ci treatment. G12Ci treatment resulted in a 3-10-fold increase in the frequency of ALDH^high cells in H358, H1373, H1792, and H2030 cells compared to the untreated cells (Fig. 3A and S10), suggestive of an enrichment in DTPs. We further found that while SOS2^KO H1792 and H2030 cells showed fewer ALDH^high cells compared to controls, G12Ci treatment resulted in a 3-12-fold increase in the frequency of ALDH^high cells H1792 and H2030 SOS2^KO cells compared to untreated controls (Fig. S10).

Figure 3.

Isolated ALDH^high populations are resistant to G12Ci but sensitive to combined G12Ci:SOS1i.

A, Aldefluor staining for ALDH enzyme activity in H358 negative control (DEAB), untreated cells, or cells treated with 100 nM G12Ci for 72h. B, Gaiting strategy for isolating ALDH^low (green) versus ALDH^high (dark blue) populations. C, G12Ci or MEKi dose response curves for unsorted (grey), ALDH^low (green), and ALDH^high (dark blue) H358 cells Data are the mean ± sd from three independent experiments, each experiment had three technical replicates. D-G, Heat map of cell viability and excess over Bliss ALDH^high or unsorted H358 cells treated with a 9 × 9 matrix of G12Ci ± SOS1i (D), sum of excess over Bliss values (E) or quantitation of the change in G12Ci potency (log α₂) vs. efficacy (β_obs) (F, zoom in for unsorted cells in G). H, G12Ci dose response curves for naïve H358 cells (black diamonds) or cells pre-treated with G12Ci for 72h and then rested (no G12Ci) for 72h prior to assessment (blue squares). I-J, Dose response curves of naïve (I) or G12Ci pre-treated (J) H358 cells alone or in the presence of either SOS1i (red) or ALDHi (purple). In C-J, data are the mean ± sd from three independent experiments, each experiment had three technical replicates.

Because therapeutic resistance is thought to arise out of the DTP population, we further assessed the extent to which SOS1i could target ALDH^high DTPs. H358 cells were sorted into ALDH^high versus ALDH^low populations and assessed for G12Ci or MEKi sensitivity compared to unsorted populations (Fig. 3B). While unsorted and ALDH^low cells demonstrated similar sensitivity to increasing doses of G12Ci or MEKi treatment, adagrasib (G12Ci), sotorasib (G12Ci), and trametinib (MEKi) potency were markedly reduced in sorted ALDH^high populations (Fig. 3C), showing >30% survival. However, despite this relative insensitivity to G12Ci or MEKi alone, SOS1i + G12Ci or SOS1i + MEKi showed marked synergy in sorted ALDH^high cells (Fig. 3D–G and Fig. S11). We further generated 3-day and 21-day G12Ci DTPs by continuous treatment with 30 nM G12Ci, and then rested cells for three days to ensure G12Ci ‘washout’ prior (Fig. S12) to assessment of G12Ci responses. Both 3-day (Fig. 3H) and 21-day (Fig. S13A) DTPs showed reduced G12Ci potency (EC₅₀ shift) compared to parental controls. Treatment with a SOS1i enhanced G12Ci potency to both parental and DTP cells, whereas treatment with the ALDHi disulfiram enhanced G12Ci potency only in DTPs that showed increased ALDH^high staining (Fig. 3I–J and S13B–C). Western blotting for RTKs shown to be upregulated in EGFRi DTPs (AXL, IGF-1R) showed increased IGF-1R abundance at 3d and a further increase in IGF-1R abundance and phosphorylation at 21d. We further observed a marked increase in AXL abundance at 21d (Fig. S14). In contrast, assessment of RAS effector signaling showed differential signaling between 3d and 21d DTPs. MEK and ERK phosphorylation were reduced after 3d G12Ci treatment along with (ERK-responsive) DUSP6 abundance, but while DUSP6 abundance was markedly decreased in 21d DTPs, we observed robust increases in MEK, ERK, and AKT phosphorylation at 21d reflecting increased RTK and decreased DUSP6 abundance (Fig. S14). These data suggest that SOS1i can re-sensitize DTPs to G12Ci and potentially target pre-resistant LUAD populations prior to the development of acquired resistance.

Tumor initiating cells (TICs) are a subset of the drug tolerant persister population with stem-like properties that are capable of self-renewal and asymmetric division; the TIC population is thought to represent the sanctuary population responsible for tumor recurrence after treatment failure (35,36). HIF1α promotes transcription of genes responsible for TIC activity (67), and hypoxia signatures are associated with cancer stemness (64,65) and poor survival for patients with LUAD (68). As SOS1i ± G12Ci inhibited hypoxia gene signatures and HIF1α dependent transcription (Fig. 1), we hypothesized that SOS1 may regulate TIC survival. TIC frequency can be modeled in situ by assessing the frequency with which cells can grow as 3D spheroids from a single cell (tumor initiating cells; TICs). G12Ci treatment (72h) caused a 2-3-fold increase in TIC frequency in H358, H1373, H1792, and H2030 cells (Fig. 4A), indicating that tumor-promoting DTPs are enriched in G12Ci treated populations. Since SOS1i (± SOS2^KO) inhibited G12Ci adaptive resistance (Fig. 2), we assessed the ability of SOS1i to target TICs. In both H358 and H1373 cells, TIC frequency was inhibited by SOS1i in a dose-dependent manner, illustrating the dependency of TIC outgrowth on SOS1 signaling (Fig. 4B and S15A). Further, TICs were enriched in sorted ALDH^high populations, and survival of ALDH^high TICs was inhibited by SOS1i (Fig. 4C and S15B). Unlike H358 and H1373 cells, TIC frequency in H1792 and H2030 cells was not decreased by SOS1i alone, but required combined SOS1i + SOS2^KO indicating the importance of SOS2 to TIC survival in these cell lines (Fig. 4D).

Figure 4.

SOS1i ± SOS2^KO prevents G12Ci-induced TIC outgrowth.

A-E. TIC frequency from in situ extreme limiting dilution assessment (ELDA) of the indicated cell lines: pre-treated with G12Ci 72h (A); treated with increasing dose of SOS1i (B); in unsorted, ALDH^low, and ALDH^high cells ± SOS1i (C); in NT or SOS2^KO cells ± SOS1i (D); in NT or SOS2^KO cells pre-treated with G12Ci for 72h to upregulate TICs ± SOS1i (E); ## χ² < 0.01 vs. untreated for TIC upregulation; * χ² < 0.05, ** χ² < 0.01 vs. untreated for SOS1i ± SOS2^KO. F, ALDH/CD133 staining in H1373 DEAB negative control (DEAB), untreated cells, or cells treated with G12Ci for 72 hours. G, TIC frequency from G12Ci-treated cells from (F) isolated based on CD133 and ALDH staining. ** χ² < 0.01 vs. unsorted; # χ² < 0.05 vs. ALDH^high/CD133⁺.

Since SOS1i (± SOS2^KO) inhibited G12Ci adaptive resistance (Fig. 2), we assessed the ability of SOS1i to target G12Ci-induced TICs. Cells were treated with G12Ci for 3d, and then rested for 72h in the absence of G12Ci to allow turnover of KRAS so that the majority of KRAS was not G12Ci bound and MEK phosphorylation was restored to baseline levels (Fig. S12). In both H358 and H1373 cells, SOS1i inhibited G12Ci-induced TIC outgrowth (Fig. 4E). In contrast, while H1792 and H2030 cells showed a similar 3-fold increase in TIC frequency post G12Ci treatment (Fig. 4A), SOS1i + SOS2^KO (Fig. 4E) or SHP2i (Fig. S16) was necessary to effectively inhibit survival of G12Ci-induced TICs. We observed similar inhibition of MEKi-induced DTPs by SOS1i + SOS2^KO (Fig. S17), further supporting a role for SOS1/2 in limiting therapy-induced DTPs. These data illustrate that inhibition of proximal RTK signaling can target G12Ci-induced DTPs and TIC survival, thereby potentially reducing the pool of cells capable of promoting acquired resistance.

While ALDH considered as a functional TIC marker, several additional cell surface markers including CD133, CD44, EpCAM/CD326, IL-6Rα/CD126, integrin α2, and CXCR4 have been reported as lung TIC/CSC (cancer stem cell) markers (32,69–71). To investigate the extent to which these potential markers are (i) upregulated in G12Ci DTPs and (ii) mark G12Ci TICs, we first assessed the extent to which each marker was upregulated individually (Fig. S18A) or in combination with ALDH activity in H1373 cells treated with G12Ci for 3 days (Fig. 4F and S18B). We found that CXCR4, IL-6Rα, and CD133 were individually increased by G12Ci treatment, only CD133 was upregulated in combination with ALDH activity after G12Ci treatment (Fig. 4F and S18). We further isolated G12Ci-treated H1373 cells by FACS based on CD133 and ALDH staining and assessed TIC frequency by in situ ELDA. Compared to parental controls, we observed a five-fold increase in TIC frequency in isolated CD133⁺/ALDH^high cells and a corresponding 45-fold decrease in TIC frequency in isolated CD133^-/ALDH^low cells (Fig. 4G). These data suggest that a combination of surface an intracellular marker including CD133 and ALDH can be used to identify TIC versus non-TIC populations in KRAS^-G12C-mutated LUAD.

SOS1 inhibition limits G12C inhibitor resistance in KRAS^G12C-mutated cells

To directly assess the ability of SOS1i to limit the development of G12Ci acquired resistance, we performed multi-well in situ resistance assays (59,60) in KRAS^G12C-mutated LUAD cells treated with G12Ci ± SOS1i over 6-12 weeks. Here, cells are plated at low density in the inner 60 wells of a 96-well plate and treated with a single dose (or combination) of drug(s). Wells are re-fed weekly and wells showing >50% confluence are scored as resistant to that dose of drug(s) (60). We previously showed that while low doses of RTK/RAS pathway targeted inhibitors cause a proliferative delay, acquired resistance to RTK/RAS pathway inhibitors including G12Cis can be modeled using a ≥ EC₈₀ dose of G12Ci (59). For H358 cells, 10 nM adagrasib modeled G12Ci acquired resistance whereas 1, 3, or 6 nM caused a slight delay in proliferation but still showed intrinsic resistance. To determine both whether SOS1i could reduce the G12Ci dose needed to overcome intrinsic resistance and the extent to which SOS1i could delay or reduce the development of G12Ci acquired resistance, we treated H358 cells with 1, 3, 6, or 10 nM adagrasib ± increasing doses of SOS1i (10 – 300 nM) (Fig. S19A). We saw that 10 nM adagrasib was required to model acquired G12Ci resistance, which was also the adagrasib dose required to inhibit MEK phosphorylation, ERK phosphorylation, and DUSP6 abundance after 24h of treatment (Fig. S19B). While SOS1i had minimal effects at delaying colony outgrowth or inhibiting MEK/ERK signaling at 1 nM G12Ci, SOS1i caused a dose-dependent delay in colony outgrowth and MEK/ERK signaling inhibition at 3 nM and 6 nM G12Ci (Fig. S19). Indeed, at an intermediate (6 nM) adagrasib dose that only delayed H358 outgrowth by 1-2 weeks, SOS1i (100nM) inhibited the development of acquired G12Ci resistance in 80% of cultures, and SOS1i completely inhibited the development of resistance to 10 nM adagrasib (Fig. S19). We then modeled adagrasib and sotorasib acquired resistance in H358, H1373, H1792, and H2030 cells at increasing doses of G12Ci ± SOS1 (Fig. S20A–B) and at the maximum G12Ci dose with SOS1i ± SOS2^KO (Fig. 5A–B). In all four cell lines, SOS1i reduced the frequency of cultures showing G12Ci resistance (Fig. S20A–B), even at G12Ci doses sub-optimal for modeling G12Ci resistance (outgrowth at < 6 weeks). For both H1792 and H2030 cells, G12Ci doses > 1 μM were required to delay outgrowth, and we were unable to model sotorasib resistance in H2030 cells (Fig. 5B). At maximal G12Ci doses for modeling acquired resistance, 100 nM SOS1i completely inhibited acquired G12Ci resistance in H358 and H1373 cells either by delaying the time to develop resistance and/or by reducing the overall percentage of G12Ci resistant cultures in H1792 and H2030 cells (Fig. 5). Further, in H1792 and H2030 cells, the ability of SOS1i to limit G12Ci acquired resistance was enhanced by using either a higher dose SOS1i (300 nM) or by combining with SOS2^KO (Fig. 5). These data suggest that SOS1i both prolongs the window of G12Ci efficacy and limits the development of G12Ci acquired resistance.

Figure 5.

SOS1 inhibition limits the development of acquired G12Ci resistance.

A-B, Assessment of acquired resistance to the G12Ci adagrasib (A) or sotorasib (B) in the indicated NT or SOS2^KO cells treated with G12Ci alone (NT black; SOS2^KO blue) or G12Ci + SOS1i (NT red; SOS2^KO purple) at 100 nM (lite) or 300 nM (dark). Data are pooled from three independent experiments. *** p < 0.001 vs. G12Ci alone; ^^^ p < 0.001 vs. NT cells treated with SOS1i.

SOS1i targets DTPs in KEAP1^KO cells to limit acquired G12Ci resistance

H1792 and H2030 cells harbor inactivating co-mutations in the tumor suppressors KEAP1 and/or STK11 (Fig. S5B) and these co-mutations have been associated with a poorer response to G12Ci treatment and worse overall prognosis for patients (9,12,13,24,25). However, whether the differential responsiveness of KRAS^G12C-mutated cell lines to G12Ci + SOS1i is directly related to inactivation of KEAP1 or STK11 is difficult to assess due to the genetic diversity between different cell lines and the relatively low numbers of KRAS^G12C-mutated LUAD cell lines available for study. To directly assess the impact of KEAP1 or STK11 co-mutations to G12Ci + SOS1i combination therapy, we used a panel of isogenic H358 cells where KEAP1 and STK11 had been deleted individually or in combination (49). We first assessed the importance of these co-mutations in altering G12Ci:SOS1i synergy after short-term (72h) treatment of 3D spheroids; while KEAP1^KO did not alter G12Ci:SOS1i synergy, STK11^KO showed an apparent decrease in drug-drug synergy, although the overall decrease in cell number at high G12Ci + SOS1i doses did not appear appreciably different (Fig. 6A). Intriguingly, STK11^KO cells were more sensitive to G12Ci alone, causing a ½ log decrease in the EC₅₀ for adagrasib (Fig. 6B–C) which led to an apparent decrease in the excess over bliss value (Fig. 6D). Assessment of RAS effector signaling in cycling cells showed a 50% reduction in phosphorylated AKT in STK11^KO and DKO cells, and a stepwise reduction in ERK phosphorylation among the four genotypes (NT > STK11^KO > KEAP1^KO > DKO) (Fig. S21A–B), suggesting that reduced RAS effector signaling in STK11^KO cells may underly the enhanced sensitivity to G12Ci.

Figure 6.

KEAP1 and STK11 co-mutations regulate resistance to G12Ci + SOS1i.

A-C, Heat map of cell viability and excess over Bliss for the indicated cells treated with a 9×9 matrix of G12Ci ± SOS1i (A), G12Ci dose response curves (B), G12Ci EC₅₀ values (C), and sum of excess over Bliss values for (D).NT (black closed squares), STK11^KO (dk.grey closed circles), KEAP1^KO (grey open squares), and STK11/KEAP1 DKO (lt.grey open circles) H358 cells.* p < 0.05, p < 0.01 vs. NT. Data are the mean from three independent experiments, each experiment had three technical replicates. E, TIC frequency in the indicated cells ± SOS1i. ** χ² > 0.01, *** χ² < 0.001 for SOS1i treated vs. untreated; ### χ² < 0.001 vs. NT. F-G, In situ resistance assays assessing acquired adagrasib (F) or sotorasib (G) resistance in NT (black), STK11^KO (dk.grey), KEAP1^KO (grey), and STK11/KEAP1 DKO (lt.grey) cells. *** p < 0.001 vs. NT. H-I. Assessment of acquired resistance to the G12Ci adagrasib (H) or sotorasib (I) resistance in NT (black), STK11^KO (dark grey), KEAP1^KO (grey), and STK11/KEAP1 DKO (light grey) alone or in the presence of SOS1i (red). *** p < 0.001 for G12Ci vs G12Ci+SOS1i.

J-K. Waterfall plot of percent change in tumor volume from H2030 xenografts left untreated or treated with G12Ci +/− SOS1i for 7d (J) or 10d (K).

To determine whether KEAP1 and/or STK11 co-mutations affected the survival of DTPs and development of acquired G12Ci resistance, we assessed TIC frequency. Both KEAP1^KO and KEAP1/STK11 double knock-out (DKO) cells showed an increased TIC frequency compared to NT and STK11^KO cells (Fig. 6E). The increased TICs in KEAP1^KO were significantly inhibited by SOS1i, but not to levels seen in either NT or STK11^KO cells suggesting that KEAP1 mutations alter resistance to combined G12Ci + SOS1i. To directly test this possibility, we performed long-term in situ resistance assays to both adagrasib and sotorasib in the panel of H358 KO cells. When assessing G12Ci resistance alone, KEAP1^KO cells developed G12Ci resistance more rapidly than NT controls, and KEAP1/STK11 DKO cells were intrinsically resistant to 10 nM adagrasib (Fig. 6F) or 30 nM sotorasib (Fig. 6G) confirming that the H358 KO panel recapitulates the reduced responsiveness seen in patients whose tumors harbor KEAP1 ± STK11 co-mutations. Intriguingly, SOS1i completely inhibited the development of G12Ci resistance in KEAP1^KO cells and showed a significant enhancement of the treatment window in KEAP1/STK11 DKO cells (Fig. 6H), indicating that SOS1i has the potential to delay the development of G12Ci acquired resistance in the setting of tumors with KEAP1 ± STK11 co-mutations. We further evaluated the efficacy of G12Ci vs G12Ci + SOS1i in established (150 – 900 μM³) H2030 xenografts, as H2030 we previously shown to be ‘partially sensitive’ to adagrasib alone, with a cytostatic, but not cytotoxic, response (8). Unlike previously published results, we observed robust antitumor activity for G12Ci alone at 7d and 10d (Fig. 6J–K). Combined G12Ci + SOS1i showed a trend toward increased antitumor activity (p = 0.14-0.16); the robust antitumor activity of G12Ci alone limited significance of the combination in this CDX study. In support of a role for SOS1i in enhancing G12Ci anti-tumor activity, Thatikonda et al (46) recently showed enhanced anti-tumor activity against PDX models of KRASG12C in mice treated with G12Ci + SOS1i compared to G12Ci alone. Taken together, these data illustrate that co-mutations may not limit the usefulness of G12Ci + SOS1i combination therapy.

Discussion

Mutations in KRAS^G12C are responsible for nearly 13% of LUAD cases (1). For patients with KRAS^G12C-mutated LUAD, KRAS^G12C inhibitors have enormous therapeutic potential, however, both intrinsic and acquired resistance limit their overall effectiveness. Studies show that inhibition of proximal RTK signaling using SHP2 (15–17) or SOS1 (45,46) inhibitors enhances G12Ci binding to mutated KRAS^G12C and limits RTK-SOS-WT RAS signaling to overcome intrinsic G12Ci resistance, and, as a result, enhances G12Ci efficacy in vitro and in vivo. Here, we show that SOS1 signaling lies at the crossroad of intrinsic and acquired G12Ci resistance. SOS1i enhanced the potency of G12Ci and inhibited rebound RTK-SOS-WT RAS signaling to limit intrinsic resistance in a SOS2-dependent manner. SOS1i further re-sensitized drug tolerant persister cells to G12Ci, thereby prolonging the window of G12Ci effectiveness and limiting the frequency of acquired G12Ci resistance.

While resistance to targeted therapies is generally framed as being either primary / intrinsic or secondary / acquired, emerging evidence suggests that for RTK/RAS pathway targeted therapies, resistance should be evaluated on a continuum based on the survival and evolution of DTPs (Fig. 7). Understanding this continuum is key to understanding G12Ci responses in patients with co-mutations in KEAP1 and STK11 (13). Unlike primary or secondary resistance due to genetic mutations that directly circumvent G12Ci, DTPs use non-genetic chromatin remodeling that up-regulate signaling through multiple RTKs , enhance ability of cells to detoxify therapy-induced redox stress (30–33), and allow cells to enter a therapy-resistant near-quiescent state (28,29,34). Within the DTP population, TICs are capable of self-renewal and are hypothesized to represent the sanctuary population responsible for tumor recurrence after treatment failure (35,36). KEAP1 is a negative regulator of the NRF2 transcription factor and KEAP1 loss-of-function results in enhanced NRF2 metabolic alterations leading to inability to regulate responses to oxidative damage (23,49). Patients with KRAS^G12C-mutated LUAD harboring KEAP1 co-mutations have inferior clinical outcomes and more often reach progression free survival of less than 3 months (13). While this could be interpreted as indicative of intrinsic resistance, an alternative hypothesis is that KEAP1 LOF enhances DTP/TIC survival thereby limiting the overall effectiveness of G12Ci. Indeed, we observed that isogenic KEAP1^KO cells showed an increased frequency of TICs (Fig. 4) and accelerated the development of acquired G12Ci resistance in situ (Fig. 6). These data suggest that patients whose tumors harbor KEAP1 loss-of-function mutations may show G12Ci resistance due to enrichment of the DTP population.

Figure 7.

SOS1i targets the continuum of G12Ci resistant states.

(Left) Intrinsic G12Ci resistance is driven by adaptive reactivation of RTK signaling due to a loss of ERK-dependent negative feedback. SOS1i targets rebound RTK signaling to limit adaptive G12Ci resistance. (Middle) Cancer cells undergo non-genetic adaptation to G12Ci to both alter the redox environment and enhance alternative RTK signaling, both of which allow these ‘drug tolerant persister’ (DTP) cells to survive under therapeutic pressure. Within the DTP population, a subset of ‘tumor initiating cells’ (TICs) are capable of self-renewal and are thought to be the pharmacologic sanctuary driving that ultimately develop acquired resistance. SOS1i re-sensitizes DTPs to G12Ci and reduces TIC frequeny in G12Ci treated cultures. (Right) Acquired G12Ci resistance is often driven by RTK/RAS pathway reactivation by both genetic and non-genetic mechanisms. SOS1i both delayed the development of and reduced the frequency with which cultures acquired G12Ci resistance.

Unbiased assessment of signaling pathways modulated by combined G12Ci plus SOS1i treatment revealed that SOS1i inhibited both RTK/MAPK and hypoxia/HIF1α pathways in G12Ci treated cells (Fig. 1). Intriguingly, the combination of SOS1i-dependent inhibition of RTK signaling and hypoxia-associated pathways may underlie SOS1i-dependent inhibition of both DTP/TIC survival (Figs. 3–5) and acquired G1Ci resistance (Fig. 5). Following G12Ci treatment, loss of ERK-dependent negative feedback on multiple RTKs drives adaptive G12Ci resistance and chromatin remodeling of DTPs activates multiple RTKs; HIF1α promotes transcription of genes responsible for TIC activity (67) and hypoxia signatures are associated with cancer stemness (35,64), DTP survival (64,65), and poor survival for patients with LUAD (68). SOS1i both re-sensitized DTPs to G12Ci (Fig. 3) and inhibited G12Ci-induced TIC survival (Fig. 4), likely due to the combined effects of SOS1i inhibiting RTK signaling and hypoxia-associated pathways in DTPs. Previous work shows that Sos1^−/− increases mitochondrial oxidative stress in MEFs due to mitochondrial dysfunction (72,73) that SOS1 is an important regulator of redox signaling. Increased oxidative stress can lead to ROS accumulation to drive both senescence and multiple forms of cell death including ferroptosis (74). This oxidative stress is counteracted by KEAP1-dependent degradation of NRF2; tumors with KEAP1 inactivating mutations show increased NRF2 activity and preventing the initiation of ferroptosis (75). NRF2 activity is further enhanced in cells harboring STK11/KEAP1 co-mutations compared to KEAP1 mutations alone, suggesting that patients whose tumors harbor KEAP1 ± STK11 inactivating mutations are particularly insensitive to oxidative stress caused by targeted therapeutics. We found that SOS1i reduced DTPs and inhibited G12Ci resistance in KEAP1^KO and KEAP1/STK11 DKO cells (Fig. 6). SOS1i may similarly tip the balance of oxidative stress in KEAP1 LOF and/or G12Ci-treated cells to inhibit DTPs survival, thereby enhancing the effectiveness of and blocking acquired resistance to G12Ci for patients whose tumors harbor KEAP1 LOF mutations.

G12Ci effectiveness is limited by intrinsic and acquired resistance, hence, combination approaches are imperative to enhance clinical outcomes for patients with KRAS^G12C-mutated tumors. Our study provides a framework for modeling the evolution of G12Ci resistance and assessing the impact of therapeutic combinations during this evolution. Our data suggests that SOS1i can limit resistance at each step of G12Ci resistance (Fig. 7). SOS1i enhanced G12Ci potency and inhibited adaptive resistance to increase G12Ci killing of LUAD cells. SOS1i further re-sensitized DTPs to G12Ci to delay and inhibit the development of acquired G12Ci resistance in situ. Supporting our model, Thatikonda et al (46) showed that SOS1i enhanced in vivo G12Ci efficacy in xenograft models, and further that SOS1i partially re-sensitize G12Ci resistant cells to adagrasib treatment. Our finding that SOS1i can target each stage of G12Ci resistance, even in the presence of KEAP1/STK11 co-mutations that predict insensitivity to G12Ci alone, reveals the therapeutic versatility of SOS1i + G12Ci combination therapy and should inform future clinical trials with the goal of enhancing the overall effectiveness of G12Ci treatment in LUAD.

Significance.

The SOS1 inhibitor BI-3406 both inhibits intrinsic/adaptive resistance and targets drug tolerant persister cells to limit the development of acquired resistance to clinical KRAS^G12C inhibitors in lung adenocarcinoma cells.