Comprehensive Clinical and Molecular Characterization of KRAS^G12C-Mutant Colorectal Cancer

Abstract

PURPOSE

KRAS p.G12C mutations occur in approximately 3% of metastatic colorectal cancers (mCRC). Recently, two allosteric inhibitors of KRAS p.G12C have demonstrated activity in early phase clinical trials. There are no robust studies examining the behavior of this newly targetable population.

METHODS

We queried the MD Anderson Cancer Center data set for patients with colorectal cancer who harbored KRAS p.G12C mutations between January 2003 and September 2019. Patients were analyzed for clinical characteristics, overall survival (OS), and progression-free survival (PFS) and compared against KRAS nonG12C. Next, we analyzed several internal and external data sets to assess immune signatures, gene expression profiles, hypermethylation, co-occurring mutations, and proteomics.

RESULTS

Among the 4,632 patients with comprehensive molecular profiling, 134 (2.9%) were found to have KRAS p.G12C mutations. An additional 53 patients with single gene sequencing were included in clinical data but excluded from prevalence analysis allowing for 187 total patients. Sixty-five patients had de novo metastatic disease and received a median of two lines of chemotherapy without surgical intervention. For the first three lines of chemotherapy, the median PFS was 6.4 months (n = 65; 95% CI, 5.0 to 7.4 months), 3.9 months (n = 47; 95% CI, 2.9 to 5.9 months), and 3.0 months (n = 21; 95% CI, 2.0 to 3.4 months), respectively. KRAS p.G12C demonstrated higher rates of basal EGFR activation compared with KRAS nonG12C. When compared with an internal cohort of KRAS nonG12C, KRAS p.G12C patients had worse OS.

CONCLUSION

PFS is poor for patients with KRAS p.G12C metastatic colorectal cancer. OS was worse in KRAS p.G12C compared with KRAS nonG12C patients. Our data highlight the innate resistance to chemotherapy for KRAS p.G12C patients and serve as a historical comparator for future clinical trials.

INTRODUCTION

Colorectal cancer (CRC) is the second leading cause of cancer mortality in the United States, with approximately 50%-60% of patients eventually developing metastatic disease.^1,2 Overall survival (OS) remains poor with 5-year survival rates for patients with metastatic disease estimated at about 15%.^3,4 KRAS, a small oncogenic GTPase, represents the most common activating mutation in human cancers, including approximately 50% of all metastatic colorectal cancer (mCRC), and often predicts resistant tumors.^5-7 Targeting oncogenic KRAS has proven difficult due to high affinity for GTP or GDP and lack of a clear binding pocket.⁸ Moreover, inhibition of downstream effectors in the RAF → MEK → ERK pathway has been ineffective, highlighting an unmet need of novel therapeutic approaches for these patients.^9,10

CONTEXT

• Key Objective

• KRAS-mutated colorectal cancer (CRC) portends a poor prognosis and is challenging to target. KRAS p.G12C (G12C) represents 3% of all CRC, but the natural history of this CRC subpopulation is unknown. Recently developed G12C inhibitors have an excellent safety profile but offer modest single-agent benefits, so combination therapies are needed. We describe the natural history of G12C patients and explore translational correlatives to define future directions.

• Knowledge Generated

• Progression-free survival is poor for patients in the second- and third-line settings. We highlight that overall survival is worse in G12C patients in comparison to KRAS nonG12C CRC patients. Our translational correlatives suggest higher rates of basal EGFR activation, enrichment with PIK3CA^Mut, and diminished immune signatures.

• Relevance

• We highlight the unique clinicopathologic and molecular profile, poor prognosis, and unmet need for this targetable subset of CRC, suggesting that EGFR, PIK3CA, and immune checkpoint inhibitors may serve as viable therapeutic partners.

A consistent research focus has led to new approaches for inhibition, with Ostrem et al¹¹ demonstrating the first selective KRAS p.G12C (glycine 12 to cysteine) inhibitor in 2013 by targeting the nucleophilicity of cystine. More recently, two preclinical studies have demonstrated activity in covalent inhibitors targeting KRAS p.G12C^12,13 allowing for a potentially targetable drug for 3% of mCRC with RAS mutations.¹⁴ Both sotorasib and adagrasib work by covalently binding to the C12 position in an allosteric pocket adjacent to switch II of KRAS p.G12C–mutant proteins, irreversibly locking KRAS in an inactive state.^15,16 However, robust description regarding the clinical and molecular features of KRAS p.G12C in patients with CRC is limited. To our knowledge, no previous studies have described progression-free survival (PFS) across sequential lines of chemotherapy among patients with KRAS p.G12C mCRC. Accordingly, we retrospectively evaluated the clinicopathologic features and survival of 187 patients with KRAS p.G12C CRC treated at The University of Texas MD Anderson Cancer Center (MDACC) and explored comprehensive molecular features to serve as support for future trials.

METHODS

Patient Descriptive Data

The MDACC institutional computer database was queried to identify patients with KRAS p.G12C mutations between January 1, 2003, and September 1, 2019, and medical records reviewed according to an established IRB protocol. All patients with CRC, longitudinal follow-up, and baseline KRAS p.G12C mutation were included. Patients with previous EGFR blockade or evidence of acquired KRAS p.G12C mutation were excluded. Among the 4,632 patients with comprehensive molecular profiling, 172 patients harbored KRAS p.G12C mutation that was narrowed to 134 (2.9%) after exclusions (Fig 1). An additional 53 patients with single gene sequencing were included in clinical data but excluded from prevalence analysis allowing for 187 total patients. We chart reviewed all patients for baseline characteristics, clinical history, treatment history, and outcome data including stage at time of diagnosis,¹⁷ surgical procedures, PFS, and OS. OS data were compared with patients with KRAS nonG12C mutations.

FIG 1.

Study diagram showing stratification of MDACC-CRC cohort into patients with and without KRAS p.G12C mutation.

Outside Data Sets

We used several data sets to assess immune signatures, gene expression profiles (GEPs), hypermethylation, comutations, and proteomics. Specifically, we obtained data from the following data sets: Integromics (an internal MDA data set curated by MDA bioinformaticians); The Cancer Gene Atlas (TCGA); the MARISA data set (GEO Accession: GSE39582)¹⁸; the Yeatman data set (GEO Accession: GSE86566)¹⁹; Memorial Sloan Kettering Cancer Center (MSKCC) data set, composed of two data sets: Brannon et al²⁰ and Total Cancer Care/MSKCC Project²¹; GENIE data set²²; and Nurses Health.²³

Statistical Analysis

Descriptive statistics were used for clinical characteristics. OS was defined as time from metastases to date of death or last known follow-up excluding patients who had undergone curative metastasectomy. PFS was defined as the time of treatment initiation to time of radiograph progression or date of death. Patient demographics and treatment characteristics were compared using a log-rank test. Survival curves and comparison between KRAS p.G12C and KRAS nonG12C patients were performed by using the Kaplan-Meier method (P < .05).

Differences in protein signatures were identified using a t-test (P < .05; corrected for multiple testing via Benjamini-Hochberg false discovery rate [BH-FDR]). Finally, we used the DISCOVER method, which is highly sensitive for controlling the FDR, that was used to assess the co-occurrence and mutual exclusivity across all pairs of the recurrently mutated genes in the cohort. Fisher's exact test was conducted to compute the odds ratio (OR) of the mutational events of every pair of mutated genes. A pair of genes was considered mutually exclusive (OR < 0.7) or co-occurrence (OR > 1.5) if the pairwise association of the genes was found significant (FDR < 0.3) in the DISCOVER method.²⁴

RESULTS

Patient Demographics

The median age at diagnosis was 56 (range, 21-83) with even distribution between male and female (female, n = 97, 51.9%) (Table 1). One hundred eighty-five (99.0%) patients were moderately or poorly differentiated with only two (1.1%) patients having well-differentiated tumors. One hundred fourteen (61.0%) patients had left-sided tumors at presentation, whereas two patients had two primary tumors at both the left and right sides. Six (3.2%) patients had evidence of microsatellite instability. Sixty-eight (36.4%) patients were current or former smokers with a median pack year history of 20 years. Fifty (26.7%) patients ultimately underwent metastasectomy.

TABLE 1.

Patient Demographic and Disease Characteristics

Characteristics of Patients With Stage I-III Disease

Ninety-nine of the 100 (99.0%) patients with stage I-III disease underwent surgical resection. One patient with stage I disease had an endoscopic resection at an outside institution and presented to MDACC with metastatic disease. The median OS for patients with stage I-III at diagnosis was 75.2 months (95% CI, 64.4 to 86.0) (Data Supplement). Patients with left-sided disease at diagnosis had longer OS in comparison to right-sided disease at 96.2 (95% CI, 84.9 to 107.5) and 61.8 (95% CI, 50.8 to 62.8) months, respectively (P = .0043). No differences in OS were observed on the basis of other clinicopathologic features. Seventy-six (76.0%) patients received chemotherapy in the adjuvant or neoadjuvant setting; all regimens were consistent with standard of care. Eighty-five (85.0%) patients with stage I-III disease eventually developed metastatic disease after surgical resection.

Characteristics and Survival Outcomes of Patients With Metastatic or Recurrent Disease

Eighty-seven patients had metastatic disease at the time of diagnosis (Data Supplement). Median OS for patients with stage IV disease at the time of diagnosis was 25.1 months (95% CI, 18.9 to 31.3). There was no significant difference in OS on the basis of age, sex, body mass index, smoking hx, histology, site of primary tumor, or T stage. Twenty-two (25.2%) patients in this population ultimately underwent metastasectomy with superior OS compared with those who did not undergo metastasectomy at 51.2 (95% CI, 40.6 to 61.8) versus 21.2 (95% CI, 17.3 to 25.1) months, respectively (P < .0001).

Of the 65 patients with metastatic disease who did not undergo metastasectomy, 63 patients received systemic therapy and continued treatment at MDACC (Data Supplement). The median number of lines of therapy was 2. The median PFS was not affected by the backbone chemotherapeutic agent in the first-line setting whether oxaliplatin-based or irinotecan-based (6.0 months v 7.6 months, respectively, P = .5608). Fifty-four of the 63 (85.7%) patients received an oxaliplatin doublet with or without bevacizumab in the first line, whereas six patients (9.5%) received an irinotecan doublet and three patients received FOLFOXIRI (4.8%). Forty-seven patients ultimately received second-line chemotherapy. Forty of the 47 (87.2%) patients received irinotecan-based therapy, whereas five (10.6%) patients received an oxaliplatin-based regimen, and one patient each received trifluridine and tipiracil and regorafenib, respectively. Twenty-one patients received third-line therapy. Seven (33.3%) patients received a rechallenge with FOLFOX or intensification with FOLFOXIRI. Seven (33.3%) patients received regorafenib, and three patients received trifluridine and tipiracil (14.2%). Four patients were enrolled in clinical trials in the third-line setting. Eleven patients received fourth-line therapy, and two patients went on to receive more than four lines. The median PFS for patients receiving the first-, second-, and third-line therapy was 6.4 (95% CI, 5.05 to 7.41) months, 3.9 (95% CI, 2.85 to 5.97) months, and 3.0 (95% CI, 2.00 to 3.44 months; Fig 2) months, respectively. In the second-line setting, 15 of 42 (35.7%) patients had PFS that exceeded 6 months, whereas in the third-line setting, only two of 21 (9.5%) patients had PFS that exceeded 6 months. Finally, we compared the OS of KRAS p.G12C patients with an internal cohort of 720 patients with KRAS nonG12C excluding patients who had undergone metastasectomy (Fig 3). Patients with KRAS p.G12C had shorter OS compared with patients with KRAS nonG12C at 21.2 (95% CI, 16.7 to 25.7) versus 31.6 (95% CI, 24.3 to 38.9) months, respectively (P = .003).

FIG 2.

PFS by line of chemotherapy. PFS, progression-free survival.

FIG 3.

Comparison of OS between patients with and without KRAS p.G12C mutation. OS, overall survival.

Translational Correlatives

Comutations.

Using the 134 patients with comprehensive profiling, we performed comutation analysis. Patients with KRAS p.G12C and other KRAS alleles were removed because of potential for acquired resistance. A binary mutation matrix was constructed to analyze the comutation or mutational exclusivity between every pair of genes. TP53, APC, and PIK3CA represented the highest percentage of comutations (Fig 4A). An oncoprint shows different groups of patients with mutually exclusive gene mutations within KRAS G12C cohort, such as TP53^WT or PIK3CA^Mut and TP53^Mut or PIK3CA^WT (Fig 4B).

FIG 4.

Comutation profile of patients with KRAS p.G12C. (A) Percentage of patients comutated with genes of interest. (B) An oncoprint plot of mutually exclusive genes comutated in patients with KRAS p.G12C, such as TP53^WT/PIK3CA^Mut and TP53^Mut/PIK3CA^WT.

To confirm our findings, we analyzed seven external data sets including 1,810 KRAS p.G12C and 10,219 KRAS nonG12C samples to curate comutated genes. We selected 15 genes that are broadly known to be important drivers of CRC and are also involved in protein-protein interactions and pathways critical to CRC development, targeted therapy, and drug resistance (Data Supplement). Analysis of mutual co-occurrence and exclusivity was done to determine which were significantly comutated when comparing KRAS p.G12C patients with KRAS nonG12C and wild-type patients. Regarding the former, NOTCH3 (0.05; 95% CI, 0.0 to 0.84) and PIK3CA (0.63; 95% CI, 0.42 to 0.94) (P < .05) were found to be significantly less likely to co-occur with KRAS p.G12C patients compared with KRAS nonG12C patients. When comparing KRAS p.G12C patients with KRAS^WT patients, we found that APC (2.8; 95% CI, 1.70 to 4.62) was significantly more likely to co-occur with KRAS p.G12C patients compared with KRAS nonG12C patients. Conversely, BRAF (0.03; 95% CI, 0.00 to 0.21), ERBB4 (0.31; 95% CI, 0.11 to 0.85), NOTCH3 (0.04; 95% 0.0-0.61), NRAS (0.04; 95% CI, 0.00 to 0.70), and TP53 (0.44; 95% CI, 0.31 to 0.61) were significantly less likely to co-occur with KRAS p.G12C patients (P < .05) (Fig 4B).

GEP.

Differences in GEPs from the TCGA data set were analyzed comparing KRAS p.G12C, KRAS nonG12C, and KRAS^WT patients yielded a common core immune signature. In both comparisons, KRAS p.G12C displayed a significantly decreased immune expression profile compared with KRAS nonG12C and KRAS^WT patients marked by lower levels of IL6/JAK/STAT3, IFN-γ, complement, and IL2/STAT5 (P < .0001) (Fig 5A).

FIG 5.

(A) Differences in GEPs between KRAS p.G12C, KRAS nonG12C, and KRAS wild-type patients. *P < .01. (B) Reverse phase protein array of KRAS p.G12C and non-G12C from MD Anderson Cancer Center and The Cancer Genome Atlas (TCGA) data sets for key proteins in MAPK pathway. Average z-scores are plotted. *P < .05, **P < .001. GEP, gene expression profile.

Protein expression MEK, RSK, and HER2.

We compared protein expression levels between KRAS p.G12C patients and KRAS nonG12C patients using TCGA and MDACC data sets (Fig 5B). Our analysis revealed that KRAS p.G12C patients harbor higher levels of basal EGFR activation (P = .004 for TCGA; P < .001 for MDACC). Additionally, our analysis showed that there is no change in the activation of MEK, RSK, and HER2 between KRAS p.G12C and KRAS non-G12C patients.

Consensus molecular subtypes (CMS) and hypermethylation.

The MDACC, TCGA, Marisa, and Yeatman data sets were used to analyze 24 KRAS p.G12C patients, 317 KRAS nonG12C patients, and 595 KRAS^WT patients. CMS2 was the predominant CMS subtype among KRAS p.G12C-mutated patients at 45.6% although there was no difference in CMS distribution between the three groups. We categorized patients into high (CIMP-H) and low (CIMP-L) levels of hypermethylation using the MDACC, TCGA, and Nurses Health data sets. KRAS p.G12C patients had significantly lower rates of hypermethylation compared with KRAS^WT patients (P = .0096).

DISCUSSION

This is the first study, to our knowledge, that delineates PFS among patients with KRAS p.G12C-mutated CRC and adding important correlative studies to the existing literature. Our population demonstrated a first-line PFS of 6.4 months demonstrating only modest benefit compared with expectations. Similarly, PFS was short in the second- and third-line settings. These data demonstrate the relative chemotherapy-resistant nature of these tumors and highlight the need for novel-targeted therapies. Additionally, although all RAS mutations portend a poor prognosis, KRAS p.G12C patients had a significantly lower OS compared with KRAS nonG12C patients in our internal cohort, confirming a recent study with similar findings.²⁵ Notably, patients who underwent metastasectomy had a significantly longer OS compared with those who did not, highlighting that metastasectomy should still be considered for appropriately selected patients despite poor prognosis. Marginal PFS in subsequent line settings should accelerate efforts for new therapies for these patients.

The potential for a targetable RAS mutation represents a new horizon in CRC where RAS as a biomarker has primarily been used to predict the lack of benefit from specific therapies such as EGFR inhibitors.^26-28 The emergence of KRAS p.G12C inhibitors sotorasib and adagrasib has changed the molecular landscape of a heretofore undruggable mutation in cancer. Preliminary phase I or II trial data demonstrate safety and efficacy in metastatic non–small-cell lung cancer (NSCLC) and CRC although with enhanced efficacy for patients with NSCLC compared with patients with CRC thus far.^14,18 This was confirmed in a recent phase I effort by Hong et al.²⁹ Although 32.2% of patients with NSCLC had a confirmed partial response, only three (7.1%) patients with CRC developed partial response. Isolated KRAS inhibition is insufficient for CRC, reminiscent of the stark differences appreciated with monotherapy for BRAF^V600E between melanoma and CRC. Unlike the success achieved in metastatic melanoma, early experience with single-agent BRAF inhibition yielded disappointing results with response rates of 5% and 0% in two early phase clinical trials for CRC.^30,31 Investigators discovered that BRAF inhibition results in feedback activation of EGFR^32,33—however, understanding this therapeutic vulnerability supported a biologically rational design for the landmark BEACON trial that demonstrated an OS benefit for dual inhibition with cetuximab and encorafenib.³⁴ These lessons highlight the complex biological heterogeneity in CRC, and early phase data confirm that KRAS G12C inhibitors will also require a combinatorial approach to enhance response and durability.

Preclinical work investigating adagrasib efficacy using KRAS p.G12C cell line and patient-derived xenograft models identified potential resistance mechanisms to KRAS p.G12C inhibition, most notably activation of receptor tyrosine kinases (RTKs) or changes in cell cycle regulators such as CDKN2A or CDK4.¹³ Antitumor efficacy of KRAS p.G12C inhibitors was restored through combinatorial therapies of novel inhibitors targeting RTKs, SHP2, mTOR, and CDK4/6, all of which represent viable next steps for clinical investigation. Moreover, our study demonstrated higher basal EGFR activation in KRAS p.G12C patients in comparison to KRAS nonG12C patients, providing critical evidence that KRAS p.G12C CRC patients harbor intrinsic resistance to KRAS p.G12C inhibition. This is further supported by recent work demonstrating EGFR as the dominant mechanism of resistance in CRC to KRAS G12C inhibitors,³⁵ suggesting novel combinatorial strategies with EGFR inhibition for the emerging class of KRAS p.G12C inhibitors as a logical next step.³⁵

Additionally, previous attempts to target the RAS-MAPK pathway have been significantly derailed by adaptive feedback mechanisms, resulting in resistance because of pathway reactivation. Unfortunately, this theme is inevitably maintained in the KRAS p.G12C story. Ryan et al³⁶ evaluated a panel of KRAS G12C cell lines treated with G12C inhibitors and noted RTK-mediated rapid feedback reactivation of wild-type RAS as a primary adaptive resistance mechanism. Although wild-type RAS cannot be targeted by KRAS p.G12C inhibitors, coinhibition with SHP2 maintained RAS pathway suppression in vitro and in vivo, thereby improving efficacy. We await the results of the ongoing phase I trial evaluating the safety of sotorasib as monotherapy and in combination with ICI and additional novel agents of promise (SHP2, MEK, pan-HER inhibitors) (ClinicalTrials.gov identifier: NCT04185883). As additional research investigates the combination of RAS inhibitors with chemotherapy, immunotherapy, and novel-targeted agents, circulating tumor DNA will play a critical role in real-time monitoring of response and identifying novel adaptive resistance mechanisms that may arise.

Furthermore, after APC and TP53, comutation with PIK3CA was noted in KRAS p.G12C patients (Fig 4). Previous experience has shown that patients with KRAS mutations do not respond to phosphoinositide 3-kinase (PI3K) inhibitors; therefore, investigating combinations with novel KRAS p.G12C inhibitors and PI3K inhibitors for comutated patients may overcome this underlying resistance. Although preclinical data suggest that the PI3K inhibitor in combination with MAPK inhibition is a rational combination, its use has been limited by low tolerability preventing later-stage clinical trials.³⁷ However, the excellent safety profile of novel KRAS p.G12C inhibitors may allow for an alternative partner to pursue in combination clinical trials.

Finally, our study demonstrated that KRAS p.G12C is associated with lower rates of immune signatures such as IL6/JAK/STAT pathway. This is important to highlight considering Canon et al’s report demonstrating that treatment with sotorasib in immune-competent mice was followed by an influx of proinflammatory microenvironment. Their group combined sotorasib with immune checkpoint inhibitors, which resulted in delayed tumor growth with a complete response in nine of 10 tumors which was durable at 112 days after treatment.¹² Although early, this may suggest that our findings are supportive of a rational combination with preclinical promise.

A limitation of our study is that our internal molecular panel profiling APC consists of only hot spots and does not characterize the full gene. A second is the need for a variety of data sets with differing annotations. Using multiple data sets has potential to add variable results as not all data sets are standardized. First, we sought to maintain consistency by using a normalization strategy to standardize results. Second, several data sets derive from an internal source, so we can analyze and ensure integrity of the data. Third, the outside data sets we used primarily consisted of binary results such as KRAS p.G12C mutated versus not mutated making likelihood of variance minimal.

In summary, our results demonstrate the most comprehensive clinical and molecular data analysis in patients with KRAS p.G12C-mutated CRC. Our findings highlight that KRAS p.G12C demonstrates a unique molecular subtype of KRAS-mutated patients with higher rates of basal EGFR activity, low rates of immune signatures, and hypermethylation. Most importantly, our findings highlight that PFS beyond the first line has limited activity serving as a historical control and prompting new efforts for development of novel targeted therapies in the refractory setting. Our study supports several biologically sound exploratory combination strategies that are consistent with other preclinical data and should propel future clinical trial design.

AUTHOR CONTRIBUTIONS

Conception and design: Jason T. Henry, Scott Kopetz, Benny Johnson

Administrative support: Jason T. Henry, Oluwadara Coker

Financial support: Scott Kopetz

Provision of study materials or patients: John Paul Shen, Van K. Morris, Arvind Dasari, Bryan Kee, Christine Parseghian, Shubham Pant, David Fogelman, Robert A. Wolff, David Hong, Michael J. Overman, JeanNicolas Vauthey, Scott Kopetz

Collection and assembly of data: Jason T. Henry, Oluwadara Coker, Saikat Chowdhury, John Paul Shen, Maliha Nusrat, Bryan Kee, Shubham Pant, Nikeshan Jeyakumar, Limin Zhu, Yujiro Nishioka, David Fogelman, JeanNicolas Vauthey, Benny Johnson

Data analysis and interpretation: Jason T. Henry, Oluwadara Coker, Saikat Chowdhury, John Paul Shen, Arvind Dasari, Kanwal Raghav, Christine Parseghian, Robert Wolff, David Hong, Michael J. Overman, Scott Kopetz, Benny Johnson

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO’s conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).