Bcl-xL Is a Key Mediator of Apoptosis Following KRASG12C Inhibition in KRASG12C-mutant Colorectal Cancer

Abstract Novel covalent inhibitors of KRASG12C have shown limited response rates in patients with KRASG12C-mutant (MT) colorectal cancer. Thus, novel KRASG12C inhibitor combination strategies that can achieve deep and durable responses are needed. Small-molecule KRASG12C inhibitors AZ’1569 and AZ’8037 were used. To identify novel candidate combination strategies for AZ’1569, we performed RNA sequencing, siRNA, and high-throughput drug screening. Top hits were validated in a panel of KRASG12CMT colorectal cancer cells and in vivo. AZ’1569-resistant colorectal cancer cells were generated and characterized. We found that response to AZ’1569 was heterogeneous across the KRASG12CMT models. AZ’1569 was ineffective at inducing apoptosis when used as a single agent or combined with chemotherapy or agents targeting the EGFR/KRAS/AKT axis. Using a systems biology approach, we identified the antiapoptotic BH3-family member BCL2L1/Bcl-xL as a top hit mediating resistance to AZ’1569. Further analyses identified acute increases in the proapoptotic protein BIM following AZ’1569 treatment. ABT-263 (navitoclax), a pharmacologic Bcl-2 family inhibitor that blocks the ability of Bcl-xL to bind and inhibit BIM, led to dramatic and universal apoptosis when combined with AZ’1569. Furthermore, this combination also resulted in dramatically attenuated tumor growth in KRASG12CMT xenografts. Finally, AZ’1569-resistant cells showed amplification of KRASG12C, EphA2/c-MET activation, increased proinflammatory chemokine profile and cross-resistance to several targeted agents. Importantly, KRAS amplification and AZ’1569 resistance were reversible upon drug withdrawal, arguing strongly for the use of drug holidays in the case of KRAS amplification. Taken together, combinatorial targeting of Bcl-xL and KRASG12C is highly effective, suggesting a novel therapeutic strategy for patients with KRASG12CMT colorectal cancer.


Introduction
In colorectal cancer, KRAS is the most mutated RAS isoform ($86%), and mutations are most likely to occur in codon 12 (1,2). KRAS cycles between its inactive, GDP-bound and an active, GTPbound form that is regulated by either GTP-loading guanine nucleotide-exchange factors or GTPase-activating proteins (GAPs; ref. 3). KRAS mutations interfere with the rate of its intrinsic and GAPinduced GTP hydrolysis, favoring formation of the constitutively active GTP-bound form.
Although substantial advances have been made in the treatment of genetically defined subtypes, such as RAS/BRAF wild-type (4) and BRAFMT colorectal cancer (5), an effective therapeutic strategy for KRASMT colorectal cancer, the most common genetically defined subtype ($40%-45%) is still lacking. Current treatment options for KRASMT colorectal cancer are primarily based on combinations of chemotherapy with antiangiogenic agents (6). KRAS proteins have long been considered "undruggable" due to their small size and the tight binding of KRAS to GTP in its active state. Recently, unique characteristics of the KRAS G12C allele have been exploited for the design of a number of covalent inhibitors that bind specifically to the cysteine at position 12, thereby locking KRAS in its inactive state (7). KRAS G12C can be found in approximately 14% and approximately 4% of lung cancer and colorectal cancer, respectively (8,9). A recent trial with the KRAS G12C inhibitor AMG510 (sotorasib) has shown remarkable singleagent activity in KRAS G12C MT lung cancer, but efficacy was not encouraging in KRAS G12C MT colorectal cancer (10).
Here, we characterize the activity of AZ'1569 (11), a novel KRAS G12C inhibitor, in a panel of KRAS G12C MT colorectal cancer cells. Using RNA sequencing (RNA-seq), RNAi/compound screens, and mechanistic studies, we identified Bcl-xL as mediator of apoptosis and intrinsic resistance to AZ'1569. We also show that concomitant inhibition of Bcl-xL and KRAS G12C leads to marked increases in therapeutic efficacy in KRAS G12C MT in vitro and xenograft models. Using genomic and proteomic analyses, we show that AZ'1569acquired resistant models have high-level amplification of the KRAS G12C allele with marked elevation of proinflammatory environment, resulting in resistance to several targeted agents and conventional chemotherapy.
Cell culture C106, SW837, SW1463, SNU1411, LIM2099, and V481 colorectal cancer cells were kindly provided by Prof. Bardelli (12). RW7213 cells were provided by Dr. Arango (University Hospital Vall d'Hebron, Spain; ref. 13). HCT116 cells were purchased as authenticated stocks from ATCC. Frozen stocks were immediately established from early passage cells. Cells were cultured for not more than 20 passages following thawing. All cell lines were screened monthly for Mycoplasma (MycoAlert Detection Kit, Lonza). KRAS G12C status was confirmed using Sanger sequencing. See Supplementary Materials and Methods for detailed protocols.

Generation of AZ'1569-resistant cells
Concentration of AZ'1569 was increased (0.125-1 mmol/L) until a single-cell density was obtained. Surviving RW7213 cells were expanded in the presence of AZ'1569 (maximum concentration 3 mmol/L). Resistance was determined using cell viability assays.

Protein analysis and Western blotting
Western blotting has been described previously (14,15). b-actin was used as a loading control. Details of antibodies are provided in Supplementary Table S1. Absolute protein quantification of BAK, BAX, BCL2, Bcl-xL, and MCL-1 was performed as described previously (16).

Immunoprecipitation
Cells were transfected with 1 mg Myc-tagged Bcl-xL for 24 hours, after which cells were treated with AZ'1569 and/or navitoclax for a further 24 hours. Cells were lysed with SDS-free RIPA supplemented with protease inhibitors, followed by protein quantification, retrieval of inputs, and incubation with Myc-tag antibody conjugated dynabeads overnight. Dynabeads were then washed and boiled in loading buffer for 5 minutes at 95 C before Western blotting analysis.

Absolute protein quantification for correlation analysis [log (EC 50 ) to Bcl-xL/BAK ratio]
To calculate BAK and Bcl-xL protein molar concentrations, densitometry values of basal protein expression were obtained for the panel of KRAS G12C MT colorectal cancer cell lines and HCT116 cells. ImageJ software analysis of Western blots for respective proteins, normalized to b-actin (loading control) was used. Fixed molar concentrations were considered for BAK and Bcl-xL in the HCT116 cell line (Bak ¼ 677 nmol/L, Bcl-xL ¼ 604 nmol/L) as determined previously (16). The equation used was protein (nmol/L) value ¼ [(densitometry value (BAK or Bcl-xL) normalized to b-actin)/respective HCT116 value for BAK or Bcl-xL) Â respective concentration in HCT116 (nmol/L)]. Correlation analyses were performed using GraphPad Prism 9.0.

RNA/DNA extraction and RT-PCR analysis
RNA and DNA extractions were performed using RNeasy and DNeasy Blood and Tissue Kits (Qiagen). A260/280 and 260/230 ratios were used for quality control. RT-PCR was performed as described previously (15). Probes were purchased from Roche and Thermo Fisher Scientific (TFS). See Supplementary Materials and Methods for primer sequences.
Cytokine/receptor tyrosine kinase arrays and CXCL1/TGF-a ELISA Cytokine, receptor tyrosine kinase (RTK) arrays and ELISAs (R&D systems) were used according to the manufacturer's instructions, as described previously (14). Densitometry was performed using ImageJ.

RAS-GTP assay
KRAS-GTP expression was evaluated using an active RASpulldown kit (TFS) according to the manufacturer's instructions.

RNA-seq
RNA-seq of AZ'1569-treated SW837 and SNU1411 cells was performed on a NextSeq 500 using a 150-cycle High-Output kit (Illumina) as described previously (14). Additional information is provided in the Supplementary Materials and Methods.

Sanger sequencing
PCR products were cleaned up using Agencourt AM pure beads on the Hamilton Microlab STAR liquid handling robot and Sanger

MedExome sequencing and next-generation sequencing
MedExome and next-generation sequencing (NGS) of RW7213 parental and resistant cells was performed using the Illumina Novaseq 6000 and NextSeq 500 (Illumina), respectively. Additional information is provided in the Supplementary Materials and Methods.

In vitro migration assays
Migration assays have been described previously (17). Additional information is provided in the Supplementary Materials and Methods.

In vivo study
In vivo studies were conducted as previously described using 6to 8-week-old, female NOD SCID mice (Envigo; ref. 14). Details of the initial xenograft growth curves and tolerability studies are provided in the Supplementary Materials and Methods (Supplementary Fig. S5A and S5B). For the efficacy study, 2.5 Â 10 6 SNU1411 or 10 Â 10 6 SW1463 cells were injected into the flank of NOD SCID mice. Mice received vehicle (10% ethanol, 30% polyethylene glycol 400, and 60% Phosal 50 PG orally), AZ'8037 (100 mg/kg orally), navitoclax (100 mg/kg orally), or AZ'8037 (100 mg/kg) with navitoclax (100 mg/kg). Each treatment group contained eight animals. AZ'8037 was administered daily and navitoclax was administered 5 out of 6 days. Experiments were carried out according to UKCCCR guidelines under license PPL2875, in accordance with the Animals (Scientific Procedures) Act 1986, approved by the Department of Health, Social Services and Public Safety, Northern Ireland and the Animal Welfare Ethical Review Body.

Statistical analysis
Robust z-scores (rZ ¼ median/median absolute deviation) were calculated from cell viability assays. All data were plotted (mean and SD, unless specified otherwise) and analyzed using GraphPad Prism 8.0. Significance was defined as Ã , P < 0.05; ÃÃ , P < 0.01; ÃÃÃ , P < 0.001, with P > 0.05 not significant (ns). Experiments are representative of three independent repeats unless indicated otherwise. The nature of interaction between AZ'1569 and a second drug was determined by calculating combination index (CI) values according to the Chou-Talalay method (18), using CalcuSyn (Microsoft Windows). CI values <1, >1, and ¼ 1 indicate synergy, antagonism, and additive effects, respectively. R index (RI) values were used when one compound had minimal and/or no effect on cell viability. RI values >1 indicate synergy (19).

Data availability
Raw data for RNA-seq and MedExome sequencing experiments have been deposited at the relevant NCBI platforms, under the accession numbers GSE198530 and PRJNA815942, respectively.

Results
KRAS G12C MT colorectal cancer cells show differential sensitivity to the KRAS G12C inhibitor AZ'1569 To understand the mechanistic basis for the minor clinical responses to KRAS G12C inhibition in colorectal cancer (10), we analyzed the effect of AZ'1569 in a panel of 7 KRAS G12C MT colorectal cancer cells. Initially, we validated the KRAS mutational status using Sanger sequencing of exon 2, confirming that four cell lines had a homozygous KRAS G12C mutation and three had a heterozygous KRAS G12C mutation (Fig. 1A). Five cell lines were found to have a TP53 mutation, and the V481 cells, were PIK3CAMT (Q546P) with loss of PTEN, confirming the results of previous studies (refs. 20-23; https://web.expasy.org/cellosaurus/). Thus, the genetic background of these models captures some of the heterogeneity of KRASMT colorectal cancer observed in tumors (24).
Next, we explored differences in depth, duration, and feedback signaling in response to AZ'1569 treatment between our colorectal cancer models with low, intermediate, and high AZ'1569 IC 50 values. Colorectal cancer cells were treated with AZ'1569 in a time course (Fig. 1C). A similar KRAS electromobility shift, indicative of covalent compound binding, was observed following AZ'1569 treatment across the colorectal cancer panel, suggesting that differential sensitivity was unlikely due to differences in target engagement (25). KRAS G12C inhibition resulted in profound downregulation in pERK1/2 levels as early as 6 hours after treatment in all cell lines. Further decreases in pERK1/2 levels were observed in RW7213 and LIM2099 cells 24 to 48 hours after treatment, while all the other cell lines showed a rebound in pERK1/2 levels. A total of 24 hours posttreatment, AZ'1569 caused sustained suppression of pAKT levels in the three most sensitive cell lines, although minor pAKT decreases were also observed in the most resistant cell line, V481. S6 phosphorylation was transiently reduced in the most sensitive cell lines, but marked reactivation was observed in all the cell lines. Altogether, a range of effects of AZ'1569 on downstream signaling dynamics was observed, and these were not sufficient to explain the differences in viability and/or apoptosis outcome following AZ'1569 treatment.
KRAS G12C inhibition does not sensitize KRAS G12C MT colorectal cancer cells to chemotherapy 5-FU-based doublet therapies (5-FUþoxaliplatin; 5-FUþirinotecan) remain the cornerstone of treatment for patients with KRASMT colorectal cancer (6). We therefore evaluated whether AZ'1569/5-FU, AZ'1569/oxaliplatin, and AZ'1569/SN-38 combined treatments could effectively suppress the growth of KRAS G12C MT colorectal cancer cells and used the Chou-Talalay method to calculate CI values (18). CI values for combined AZ'1569/5-FU treatment were >0.7 for the majority of the concentrations, indicative of additive interactions ( Fig. 2A). Moreover, the absolute cell viability remained > 40% for the majority of combinations. Similar results were obtained for AZ'1569/oxaliplatin and AZ'1569/SN-38 combinations ( Supplementary Fig. S1B). In addition, except for combined AZ'1569/5-FU treatment in the SNU1411 cells, none of the AZ'1569/chemotherapy combinations resulted in further increases in apoptosis compared with the effect of each treatment alone (Fig. 2B).
We next sought to identify pharmacologic combinations that could overcome primary resistance to KRAS G12C inhibition. On the basis of the known roles for MAPK, AKT, STAT3, and EGFR signaling in intrinsic and/or acquired resistance to targeted therapies and their relevance in colorectal cancer, we next evaluated whether combining AZ'1569 with either MEK1/2 inhibitor AZD6244, AKT inhibitor capivasertib, JAK/STAT inhibitors ruloxitinib/AZD1480 or EGFR inhibitor cetuximab could effectively suppress colorectal cancer cell viability ( Fig. 2C and D; Supplementary Figs. S1C-S1H). Notably, only concurrent cetuximab/AZ'1569 treatment showed moderate and/ or strong synergy across all KRAS G12C MT colorectal cancer cells tested (Fig. 2C). Although combined cetuximab/AZ'1569 treatment resulted in major reduction in cell viability ( Supplementary Fig. S1G), efficient apoptosis induction occurred only in RW7213 and C106 cells (Fig. 2D).
BCL2L1 is a key regulator of apoptotic response to KRAS G12C inhibition in KRAS G12C MT colorectal cancer cells Cytostatic and cytotoxic drugs have been linked to clinically observed disease stabilization and objective responses, respectively (26). To gain further insight into the molecular mechanisms of apoptosis following AZ'1569 treatment, we performed RNA-seq analysis before the onset of cell death in SW837 and SNU1411 cells (Supplementary Fig. S2A and S2B; GSE198530). Significant downregula-tion of the MAPK pathway negative feedback mediators DUSP4/6 and SPRY4 was observed 6 hours post-AZ'1569 treatment in both SW837 and SNU1411 cells, confirming inhibition of ERK1/2 activity ( Supplementary Fig. S2C). To identify pathways that are involved in resistance to KRAS G12C inhibition, Ingenuity Pathway Analysis (IPA) was conducted using the three gene lists generated for both cell lines. IPA comparison analyses of upregulated and downregulated pathways showed that 63 pathways overlapped and were significantly deregulated across all the timepoints analyzed in both cell lines, with a significant enrichment of gene sets in cell death-related signaling pathways, including death receptor, apoptosis, necroptosis, and TP53 signaling ( Supplementary Fig. S2D and S2E).
To identify key functional genes and/or targets that, when inhibited, cooperate with KRAS G12C inhibition to decrease survival, and increase apoptosis in KRAS G12C MT colorectal cancer cells, we used an RNAi screening approach targeting proteins that lie at nodal points in the identified cell death-related signaling pathways. The effect of downregulating each of these proteins on cell viability was tested in both SW837 and SNU1411 cells, using an ON-TARGETplus siRNA library against 42 targets ( Supplementary Fig. S3A) in the absence and presence of AZ'1569 treatment and rZ values were calculated. Notably, only one of 42 siRNAs had a significant inhibitory effect on survival in the presence of AZ'1569 in both cell lines, and this was BCL2L1, the gene encoding the antiapoptotic BH3-family member Bcl-xL (Fig. 3A). To exclude cell line-specific effects, we extended these studies to a broader panel of KRAS G12C MT colorectal cancer cells and also confirmed the cytotoxic activity of the combination by using apoptotic cell death assays, described previously (Fig. 3B). BCL2L1 silencing resulted in marked increases in apoptosis when combined with AZ'1569 in all KRAS G12C MT colorectal cancer models, compared with the effects of each treatment alone. In addition, transient overexpression of Myc-tagged Bcl-xL led to marked reduction in basal and AZ'1569-induced apoptosis in SW837 cells (Fig. 3C). Similar effects were observed in the RW7213 cells.
DR_MOMP was previously developed to predict the stress dose required in a cell to induce mitochondrial outer membrane permeabilization (MOMP), as a readout of sensitivity of colorectal cancer cells to genotoxic chemotherapy (16,27). To evaluate whether sensitivity to AZ'1569 correlated with expression levels of proapoptotic and antiapoptotic BCL-2 family members or DR_MOMP stress dose, we initially determined the basal absolute BCL-2 proteins profiles (BAK, BAX, BCL-2, Bcl-xL, MCL1) in our KRAS G12C MT colorectal cancer cells ( Supplementary Fig. S3B). Not surprisingly, given the heterogeneity of KRASMT colorectal cancer, levels of BCL-2 proteins were variable across the cell line panel. Interestingly, Bcl-xL/BAK ratio correlated with response to AZ'1569 ( Supplementary Fig. S3B, r ¼ 0.54), indicating that cells with an increased Bcl-xL/BAK ratio show an unfavorable response to AZ'1569 treatment. There was no correlation between DR_MOMP calculated stress dose and sensitivity   to AZ'1569 treatment. Next, we assessed basal and AZ'1569-induced levels of the proapoptotic and antiapoptotic BCL-2 family proteins, including the MOMP effector proteins BAX and BAK ( Supplementary  Fig. S3C). Notably, BIM levels were markedly higher in the AZ'1569sensitive RW7213 and C106 cells, compared with the levels observed in the intermediate sensitive and resistant cell lines. Expression of BIM was also acutely increased following AZ'1569 treatment in all KRAS G12C MT colorectal cancer cell lines-in particular, the RW7213 and C106 cells. Immunoprecipitation of BIM confirmed that BIM levels were induced by AZ'1569, and the BH3-mimetic ABT-263 (navitoclax) completely disrupted the association of Bcl-xL with BIM under basal conditions and following BIM induction by AZ'1569 ( Supplementary Fig. S3D). Collectively, these data indicate that concomitant suppression of the antiapoptotic protein Bcl-xL, thereby "freeing" BIM, is needed for a robust apoptotic response after KRAS G12C inhibition in KRAS G12C MT colorectal cancer.

The BH3-mimetic ABT-737 potently synergizes with KRAS G12C inhibition
To complement the siRNA profiling results, we performed a focused drug screen to identify compounds that could effectively suppress viability of KRAS G12C MT colorectal cancer cells when combined with AZ'1569. We used a drug library targeting the top druggable pathways previously identified ( Supplementary Fig. S2D). On the basis of potential for clinical application, we prioritized 45 compounds ( Supplementary Fig. S4A), including activators of intrinsic and/or extrinsic cell death and cell-cycle regulators. The effect of these drugs in the absence and presence of AZ'1569 was tested in SW837 and SNU1411 cells. Positive hits were identified as compounds that resulted in rZ less than À1.5 in three independent experiments in both cell lines; this identified 12 hits (Fig. 4A). To further refine our hit list, we determined synergy between these 12 compounds and AZ'1569, using the Chou-Talalay method in SW837 and SNU1411 cells. ABT-737 and Entinostat were the most synergistic with AZ'1569 in both cell lines ( Fig. 4B; Supplementary Fig. S4B and S4C), with ABT-737 resulting in the most potent growth suppression when combined with AZ'1569 in the extended panel of KRAS G12C MT colorectal cancer cells ( Fig. 4B; Supplementary Fig. S4D and S4E). Combined ABT-737/AZ'1569 treatment resulted also in potent increases in apoptosis as indicated by increased PARP cleavage and caspase-9/8/3 processing in all KRAS G12C MT colorectal cancer cells ( Fig. 4C; Supplementary Fig. S4F). Notably, combined ABT-737/AZ'1569 treatment resulted in higher levels of apoptosis, compared with the levels observed with cetuximab/AZ'1569 ( Fig. 4D; Supplementary Fig. S4G), suggesting that the ABT-737/AZ'1569 combined strategy could have a more beneficial effect on tumor shrinkage and objective responses in a clinical setting (26).
KRAS G12C MT colorectal cancer xenograft models are sensitive to combinatorial Bcl-xL/KRAS G12C inhibition We next assessed the in vivo therapeutic efficacy of combined Bcl-xL/KRAS G12C inhibition. We selected two different KRAS G12C MT colorectal cancer models, SW1463 and SNU1411 that showed exponential growth characteristics when grown as xenografts (Supplementary Fig. S5A) and used the orally bioavailable BH3-mimetic navitoclax and the orally bioavailable KRAS G12C inhibitor and close analog of AZ'1569, AZ'8037. The SW1463 model was resistant to single-agent navitoclax treatment and exhibited slowed but persistent growth when mice were treated with AZ'8037 (Fig. 5A). Combination treatment of navitoclax/AZ'8037 resulted in marked tumor shrinkage in the treated animals. Strong pERK1/2 inhibition was observed, in particular, in the Navitoclax/AZ'8037 cotreated tumor samples. Similar to our results in the SW1463 model, single-agent AZ'8037 slowed SNU1411 tumor growth (Fig. 5B). There was also no effect of single-agent navitoclax. Although addition of navitoclax to AZ'8037 resulted in further reduction in tumor growth, there was no tumor regression in the SNU1411 xenografts. Treatment cessation resulted in tumor regrowth in AZ'8037 monotherapy and navitoclax/ AZ'8037 combination groups (Fig. 5B). The navitoclax/AZ'8037 combination was less well tolerated in this second mouse model as shown by decreases in tumor weight in week 2 of the treatment (Fig. 5B). Navitoclax was therefore given as a 5-day-on, 1-day-off schedule. Collectively, these results indicate that Bcl-xL-targeted agents may be highly effective when used in combination with KRAS G12C inhibitors in KRAS G12C MT colorectal cancer.

Generation of colorectal cancer models with acquired resistance to KRAS G12C inhibition
Although recent clinical trials of KRAS G12C inhibition have shown modest efficacy in KRAS G12C MT colorectal cancer, emergence of acquired resistance limits further clinical benefit (28). To identify mechanisms underlying acquired drug resistance to KRAS G12C Figure 6. AZ'1569-acquired resistant cells exhibit increased PD-L1 expression and a proinflammatory phenotype. A, Left: RW7213 parental and AZ'1569-resistant clones (No. 2,No. 3,and No. 4) were treated for 72 hours with indicated concentrations of AZ'1569 or sotorasib, and cell viability was determined using CTG assays. Right: Lysates from RW7213 parental and AZ'1569-resistant clones were analyzed by WB for KRAS, pEGFR Y1068 , EGFR, pMET Y1234/1235 , MET, pERK1/2 T202/Y204 , ERK1/2, pS6 S235/6 , S6, pAKT S473 , AKT, pEphA2 S897 , pEphA2 Y588 , pEphA2 Y772 , and EphA2. Active Raf1-bound Ras was isolated from RW7213 parental and resistant clones using an RAS-GTP assay and basal GTP-bound and total KRAS levels assessed by WB. LE ¼ longer exposure. KRAS mRNA was quantified using RT-PCR. Raw values were normalized to ACTB and GAPDH expression and were analyzed using the DDC T method. A one-way ANOVA was used to calculate statistical significance. Data are representative of three independent experimental repeats. Results of NGS of RW7213 Par and AZ'1569-R clones are shown. B, RW7213 parental and resistant cells were treated with SHP-099, BI-3406, 5-FU, SN-38, oxaliplatin, crizotinib, cetuximab, dasatinib, trametinib, ulixertinib, capivasertib, PF-4708671, AZD1480, ONC206, ABT-737, sabutoclax or entinostat for 72 hours, at the indicated concentrations and cell viability was assessed using CTG assays. Heatmap represents cell viability relative to control. Data are representative of three independent experimental repeats. C, Top left: Human cytokine array using conditioned medium of RW7213 parental and resistant clones. Right: Mean spot pixel density was analyzed using Image J, rZ (relative to parental cells) were calculated using densitometry data and presented in a heatmap. Bottom left: CXCL1, CD274, and IL8 mRNA in parental and resistant clones were quantified using RT-PCR. Raw values were normalized to the expression of housekeeping genes ACTB and GAPDH and were analyzed using the DDC T method. CXCL1 protein levels in the culture media of parental (Par) and resistant subpopulations were measured by ELISA. A one-way ANOVA was used to calculate statistical significance. Data are representative of three independent experimental repeats. D, Left: Dose-response curves for AZ'1569 in RW7213 cells, incubated with conditioned media from parental cells or drug-resistant clones No. 2,No. 3,or No. 4. Cells were treated for 72 hours and cell viability was determined using CTG assay. IC 50 values were calculated using a Prism software package. Dashed line indicates 50% cell viability. A representative of three independent experiments is shown. Right: A 24-well 5-mm polycarbonate Transwell insert-plate system was used. 2.5 Â 10 5 PBMCs were resuspended in 2% FCS-supplemented DMEM and were added to the top chamber. The bottom chamber was filled with conditioned medium (medium ¼ 2% FCS-supplemented DMEM) obtained from RW7213 parental and resistant cells. Cells were incubated for 4 hours, following which CellTiter-Glo was used to measure PBMC migration to the bottom chamber (RLU ¼ relative luminescence). Serum-free DMEM was used in the bottom chamber as a negative control (neg CT). Data are representative of three independent experimental repeats.
inhibition and therapeutic strategies to overcome this limitation, we generated a preclinical AZ'1569-resistant colorectal cancer model. We selected the RW7213 cell line, which shows the highest sensitivity to AZ'1569 and cultured this cell line until resistant derivatives and/or clones emerged in the presence of AZ'1569. Three independent resistant (R) RW7213 cell populations were obtained, and these were therefore indicated as resistant No. 2, No. 3, and No. 4 (Supplementary Fig. S6A). Resistance to AZ'1569 was confirmed by cell viability assays comparing parental and resistant cell derivatives (Fig. 6A). All resistant models also showed cross-resistance to the KRAS G12C inhibitors sotorasib and adagrasib (MRTX849; Fig. 6A; Supplementary Fig. S6A). Notably, both AZ'1569 and sotorasib seemed to increase the growth rate of AZ'1569-R clones, suggesting that AZ'1569-R clones had become addicted to the presence of KRAS G12C inhibition for proliferation.

Acquired AZ'1569-R clones display KRAS amplification and activation of several RTKs
Prior studies indicated that tumors with acquired resistance to KRAS G12C agents can have multiple resistance mechanisms, including alterations within the RAS/MAPK pathway and bypass activating alterations (28,29). Although all resistant cell populations retained the original KRAS G12C mutation, MedExome sequencing of the RW7213-R clones did not reveal any secondary mutations within KRAS, BRAF/MEK/ERK, PIK3CA/AKT (Supplementary Fig. S6B). Notably, total KRAS, KRAS-GTP, and mRNA levels together with pERK1/2 levels were markedly upregulated in the RW7213-R clones (Fig. 6A). Moreover, NGS confirmed that KRAS was amplified in all three clones and clone 3 had an additional amplification in EGFR ( Supplementary Fig. S6C). Of note, withdrawal of AZ'1569 for 8 weeks resulted in loss of KRAS overexpression, reversal of hyperactivated signaling to ERK1/2, and resensitization to AZ'1569 ( Supplementary  Fig. S6D).
We also assessed the phosphorylation status of 49 RTKs in parental and AZ'1569-R clones. AZ'1569-R derivatives showed increased phosphorylation of a number of RTKs such as c-MET and EphA2 ( Supplementary Fig. S6E); these were validated using Western blotting (WB) analysis (Fig. 6A). Collectively, our data revealed KRAS amplification and coincidental bypass RTK acquired alterations in our AZ'1569-resistant clones, suggesting that the cell models generated in this work have the potential to recapitulate clinically relevant resistance mechanisms.

Acquired AZ'1569-R clones overproduce a wide array of proinflammatory factors
We have previously shown that oncogenic KRAS regulates growth factor and/or cytokine shedding and ADAM17 activity (30), a protease involved in acute resistance to chemotherapy and targeted therapies (31). We therefore investigated the growth factors/cytokines released by the AZ'1569-resistant cells, using a cytokine array (Fig. 6C). Of the 105 cytokines examined by the array, 15 targets were >1.5-fold upregulated in all three resistant clones, and these included cytokines and/or chemokines involved in innate/adaptive immunity (e.g., IL8, CXCL1) and growth factors (e.g., TGFa). We validated our array results using real-time PCR and/or specific ELISAs, showing that AZ'1569-resistant clones exhibited higher levels of IL8, CXCL1, IFNg, and TGFa ( Fig. 6C; Supplementary Fig. S6G and S6H). Given the marked cytokine and/or chemokine abundance in the drugresistant lines, we also determined PD-L1 levels and found >300-fold increased levels of CD274 (encoding PD-L1) in the AZ'1569-R clones (Fig. 6C). Of note, conditioned medium of all three AZ'1569-R clones markedly reduced sensitivity of parental RW7213 cells to AZ'1569 ( Fig. 6D). In addition, exposure to conditioned medium of all three AZ'1569-R clones increased peripheral blood mononuclear cell (PBMC) migration, indicating their importance for lymphocytic infiltration (Fig. 6D). Taken together, these results show that longterm exposure to AZ'1569 dramatically increases the immunogenicity of these cells and suggests that the induction of proinflammatory factors may produce a tumor microenvironment that is conducive to increased tumor infiltration by immune cells.

Discussion
The efficacy of anticancer targeted therapies has often been compromised by the occurrence of intrinsic and acquired resistance mechanisms, involving intratumoral heterogeneity and various compensatory signaling. Recently, drugs such as sotorasib and adagrasib, which inhibit KRAS G12C , have emerged as promising targeted therapies for patients with KRAS G12C MT lung cancer (32,33). However, clinical trials, such as CodeBreaK100 and KRYSTAL-1 using single-agent sotorasib and adagrasib, respectively, have shown substantial differences in response rates between patients with lung cancer and colorectal cancer (10,34,35). On the basis of these studies, sotorasib was granted FDA approval, but only for patients with KRAS G12C MT lung cancer (36). Understanding the factors underlying intrinsic and/or acquired resistance to this new class of compounds is critical, in particular to benefit patients with colorectal cancer. There is a growing body of evidence suggesting that deficient apoptosis induction following targeted therapy treatments can lead to a lack of efficacy (37,38). We found that KRAS G12C inhibitor monotherapy was relatively ineffective at inducing apoptosis in vitro in KRAS G12C MT colorectal cancer.
There are, as of yet, no available predictive biomarkers for response to KRAS G12C inhibitors. In agreement with previous studies, we found that sensitivity to AZ'1569 was not predicted by the KRAS allele zygosity status, the presence of concomitant mutations (including TP53 mutations), or baseline levels of kinases within the EGFR/KRAS axis (32,39). We previously used DR_MOMP, an apoptosis predictor model that requires protein profiling of Bcl-2 family proteins to predict therapeutic response and prognosis in colorectal cancer (16,27).
Although not significant, we found that the Bcl-xL/BAK ratio correlated with response to AZ'1569 treatment. Previous in vitro and clinical studies have shown positive correlations between BIM expression and response to anti-EGFR and BRAF drugs in EGFR and BRAF addicted tumors (40,41). Interestingly, our study showed that pretreatment BIM levels were associated with sensitivity and apoptotic response to AZ'1569 in KRAS G12C MT colorectal cancer cells. Further biomarker analysis of tissue samples from patients treated with KRAS G12C inhibitors will be needed to confirm the predictive role of BIM.
Previous studies evidenced that RTK/kinase feedback activation is among the main mechanisms of adaptive resistance to KRAS G12C inhibitors and that therefore vertical combinations with RTK/SHP2 inhibitors (42) or kinase inhibitors (e.g., MEK1/2, PI3K/mTORC1/2; refs. 32,39,43) are the most attractive combination options. In support of these studies, we found consistent evidence of rapid ERK1/2, AKT, and/or S6 feedback reactivation following AZ'1569 treatment, although this was heterogeneous across the KRAS G12C MT models. Combinations of AZ'1569 with inhibitors of these feedback loops showed differential effects across the KRAS G12C MT colorectal cancer cell lines. Conversely, inhibition of EGFR markedly enhanced sensitivity to AZ'1569 in all KRAS G12C MT colorectal cancer cells, supporting the findings of a recent study (12). However, addition of cetuximab to AZ'1569 only resulted in potent increases in cell death in three of seven KRAS G12C MT colorectal cancer models.
Using RNA-seq, IPA, and a siRNA screening approach, we identified that BCL2L1 was a critical mediator of resistance to cell death following KRAS G12C inhibition in colorectal cancer cells. Moreover, using focused drug screens, we identified that the BCL-2/Bcl-xL inhibitor ABT-737 was an effective inducer of apoptosis when combined with AZ'1569 in the panel of KRAS G12C MT colorectal cancer cells. Treatment with AZ'1569 resulted in acute increases in the proapoptotic protein BIM, which may "prime" cells for death, but was insufficient to cause apoptosis in five of seven KRAS G12C MT colorectal cancer models due to the presence of inhibitory antiapoptotic proteins, such as Bcl-xL. Consistent with previous studies, we showed that ABT-737 abrogates the inhibitory complex between Bcl-xL and BIM (Supplementary Fig. S3D; ref. 44), leading to robust increases in apoptosis when ABT-737 was combined with KRAS G12C inhibition in our study. Alongside its pivotal role in regulating MOMP, Bcl-xL has been identified as a critical mediator of stem cell survival through the adeno-to-colon carcinoma sequence (45). In addition, a number of studies have shown that Bcl-xL plays an important role in regulating sensitivity to chemotherapy and other targeted therapies (45)(46)(47).
The importance of Bcl-xL as a mediator of acute resistance to KRAS G12C inhibitors was shown in vivo, where combined treatment of KRAS G12C MT colorectal cancer xenografts with the BCL-2/Bcl-xL and KRAS G12C inhibitors, navitoclax and AZ'8037, resulted in supraadditive reductions in tumor growth or regression. During this article's preparation, initial results of the phase Ib study of cetuximab with adagrasib were released, showing a response rate of 43% in patients with KRAS G12C MT colorectal cancer (48). Although initial results of this small study are encouraging, it also suggests that a major part of this population will not respond to this combination, indicating the need for alternative treatment combinations. Our data suggest that combined Bcl-xL/KRAS G12C inhibition is another potential novel treatment strategy for this molecular subgroup of patients with colorectal cancer. Although less well tolerated in our in vivo strain, combination treatments with navitoclax have been widely trialed in other in vivo strains and patients without major reported toxicities (46,49). In further support of our data, a recent study showed that the Bcl-xL-targeted PROTAC, DT2216, enhanced the therapeutic efficacy of sotorasib in the SW837 KRAS G12C MT model, and demonstrated also a good tolerability (50).
Acquired resistance is a major problem limiting clinical efficacy of targeted therapies. We observed amplification of the KRAS G12C allele in all three AZ'1569-acquired resistant clones, which also coincided with acquired bypass activations in a number of RTK. Interestingly, this is consistent with analysis of clinical samples from patients treated in phase I/II studies with adagrasib (28). Contrastingly, no acquired mutations affecting the switch II pocket of KRAS (R68S, H95D/Q/R, Y96C) or other pathogenic mutations in other RTK-RAS-MAPK pathway members were detected. Importantly, we also show that acquired resistance driven by KRAS amplification is reversible upon drug withdrawal, likely because KRAS amplification confers a selective disadvantage in the absence of KRAS G12C inhibition.
The AZ'1569-R cells showed a high level of resistance to a range of targeted therapies, particularly SOS1 and MEK/ERK inhibition. Importantly, AZ'1569-R cells also showed markedly reduced sensitivity to the three chemotherapies used in colorectal cancer treatment. Thus, our results would indicate that, at least in cases where KRAS G12C inhibitor resistance is driven by KRAS amplification, patients who progress following upfront treatment with KRAS G12C inhibition may be poor candidates for other targeted therapies or chemotherapies. A previous study has shown that sotorasib has an early impact on tumor immune cell infiltration (33). Interestingly, our acquired AZ'1569-R cells showed a markedly increased proinflammatory cytokine and/or chemokine profile, which resulted in increased lymphocytic infiltration. These data further suggest major changes in the immune microenvironment of KRAS G12C inhibitor-resistant tumors, which may affect their response to immune-targeted therapies.
In conclusion, using a systems biology approach, we identified Bcl-xL as an important mediator of intrinsic resistance to KRAS G12C inhibition in KRAS G12C MT colorectal cancer. We show that KRAS G12C inhibition primes cells for death through acute induction of BIM, with coneutralization of Bcl-xL resulting in potent increases in cell death. From a cancer therapeutics perspective, the substantial tumor growth inhibition observed in our xenografts provides a strong rationale to combine Bcl-xL blockade, using navitoclax or HDAC1-3 inhibition ( Supplementary Fig. S4C) with KRAS G12C inhibitors in patients with colorectal cancer. We also demonstrate the importance of drug holidays, to delay and/or overcome emergent resistance to KRAS G12C inhibition. Finally, cross-resistance to other targeted therapies and importantly conventional chemotherapy in the AZ'1569-R cells poses a challenge, with implications for the optimal use of KRAS G12C inhibitors as a second-or third-line option.