Same-Cell Co-Occurrence of RAS Hotspot and BRAF V600E Mutations in Treatment-Naive Colorectal Cancer

Abstract

PURPOSE

Mitogen-activated protein kinase pathway–activating mutations occur in the majority of colorectal cancer (CRC) cases and show mutual exclusivity. We identified 47 epidermal growth factor receptor/BRAF inhibitor-naive CRC patients with dual RAS hotspot/BRAF V600E mutations (CRC-DD) from a cohort of 4,561 CRC patients with clinical next-generation sequencing results. We aimed to define the molecular phenotypes of the CRC-DD and to test if the dual RAS hotspot/BRAF V600E mutations coexist within the same cell.

MATERIALS AND METHODS

We developed a single-cell genotyping method with a mutation detection rate of 96.3% and a genotype prediction accuracy of 92.1%. Mutations in the CRC-DD cohort were analyzed for clonality, allelic imbalance, copy number, and overall survival.

RESULTS

Application of single-cell genotyping to four CRC-DD revealed the co-occurrence of both mutations in the following percentages of cells per case: NRAS G13D/KRAS G12C, 95%; KRAS G12D/NRAS G12V, 48%; BRAF V600E/KRAS G12D, 44%; and KRAS G12D/NRAS G13V, 14%, respectively. Allelic imbalance favoring the oncogenic allele was less frequent in CRC-DD (24 of 76, 31.5%, somatic mutations) compared with a curated cohort of CRC with a single-driver mutation (CRC-SD; 119 of 232 mutations, 51.3%; P = .013). Microsatellite instability–high status was enriched in CRC-DD compared with CRC-SD (23% v 11.4%, P = .028). Of the seven CRC-DD cases with multiregional sequencing, five retained both driver mutations throughout all sequenced tumor sites. Both CRC-DD cases with discordant multiregional sequencing were microsatellite instability–high.

CONCLUSION

Our findings indicate that dual-driver mutations occur in a rare subset of CRC, often within the same tumor cells and across multiple tumor sites. Their presence and a lower rate of allelic imbalance may be related to dose-dependent signaling within the mitogen-activated protein kinase pathway.

INTRODUCTION

Colorectal cancer (CRC) is the third most common cancer in the United States, with nearly 150,000 new cases each year.¹ Approximately 60% of CRC cases are driven by hotspot driver mutations in the mitogen-activated protein kinase (MAPK) pathway including KRAS/NRAS hotspots and BRAF V600E mutations.^2,3 These alterations generally occur in a mutually exclusive nature,^4,5 and management of advanced CRC is dependent on the mutation status of these genes: Patients with advanced CRC with wild-type RAS are eligible for cetuximab or panitumumab, whereas patients with BRAF p.V600E–mutant CRC are eligible for combination therapy with encorafenib and cetuximab.⁶

CONTEXT

• Key Objective

• Mitogen-activated protein kinase pathway–activating mutations occur in colorectal cancer (CRC). We identified a subset of cases with dual RAS hotspot/BRAF V600E mutations. Here, we defined their molecular phenotypes and investigated whether these dual-driving mutations coexist within the same cell.

• Knowledge Generated

• Approximately 1% of CRC harbor dual RAS hotspot/BRAF V600E mutations, with both mutations often co-occurring in a proportion within the same tumor cell and both often present across multiple disease foci. Subclonal mutations may be lost in metastases, and liquid biopsy may be important to understand the clinical importance of these events.

• Relevance

• Roughly 1% of CRC have two driver alterations. While a false molecular result might be one's first thought, awareness of their existence and heterogeneity is important to understand the molecular results. Microsatellite stable, BRAF V600E–mutant CRC did not have a second driver mitogen-activated protein kinase alteration before therapy, which is important for BRAF inhibitor therapy selection.

The coexistence of two RAS hotspot/BRAF V600E mutations has been described but not systematically studied.⁷ It is also known that under selection pressure, such as anti-epidermal growth factor receptor (EGFR) or anti-BRAF combination therapy, multiple MAPK pathway driver mutations emerge.⁸ Experimental models have suggested that resistance alterations are likely already present at low frequency before treatment exposure.⁹ More recently, it has been suggested that multiple RAS mutations in CRC should be treated as separate primaries and hypothesized that these follow parallel evolution.

Here, we defined the clinicopathologic and molecular characteristics of dual RAS hotspots and BRAF p.V600E hotspot mutations from a large, well-curated cohort of targeted therapy-naive CRC (herein referred to as CRC-DD). Clinical DNA-based next-generation sequencing (NGS) data, multiregional, single-cell genotyping (SCG) of point mutations, tumor evolution, and clonality predictions inferred from NGS were used.

MATERIALS AND METHODS

Patient Cohort Description

After approval of this study by our institutional review board, we surveyed the NGS data of 4,561 patients with CRC subjected to Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT)¹⁰ to identify patients with CRC with dual hotspot mutations in KRAS and NRAS codons A146, A59, G12, G13, K117, Q61, and V14 and BRAF V600E (Data Supplement). The primary tumor site was classified as proximal from the cecum through the transverse colon and distal from the splenic flexure through the rectum.

Molecular Assays for Dual RAS/RAF Hotspot Mutation Discovery

MSK-IMPACT¹⁰ is an NGS assay that interrogates the entire coding region, select introns and promoters of 468 genes to identify somatic mutations, copy number alterations (CNA), select fusions, and microsatellite instability (MSI) status. Multiregional molecular testing was performed in seven CRC-DD on all available additional tumor and metastatic sites ranging from 2 to 4 sites using an MSKCC custom panel designed for AmpliSeq (ThermoFisher Scientific, Waltham, MA)¹¹ containing 98 cancer-associated genes. MSK-IMPACT and AmpliSeq are clinically validated assays and were performed in a Clinical Laboratory Improvement Amendments–approved molecular laboratory. Informed consent was obtained from all patients undergoing MSK-IMPACT testing.

Analysis of Clonality and Allelic Imbalance Analyses in the Dual RAS/RAF Driver Cohort

MSK-IMPACT data were analyzed with FACETS to assess the clonality of each RAS/BRAF mutation and the presence or absence of allelic imbalance (AI). AI included both copy number neutral and loss events. Mutations were considered clonal when clonal cell fraction (CCF) was ≥ 0.80 or if CCF ≥ 0.70 and the lower end of the 95% CI was ≥ 90%.¹² All other mutations were considered subclonal mutations. When tumor purity was < 80%, mutations were deemed indeterminate and excluded from clonality assessment. Four samples were excluded from FACETS analyses because of low tumor purity. To assess whole-genome duplication, data were analyzed as described by Bielski et al.^12,13

Comparison of the RAS/RAF Dual-Driver CRC With Previous Single-Driver CRC

Clinical and molecular characteristics from our CRC-DD cohort were compared with 271 CRC cases with a single MAPK pathway driver mutation (CRC-SD).¹⁴ CRC-DD versus CRC-SD comparisons were carried out using a Fisher's exact test or analysis of variance for tabulated or quantitative comparisons, respectively, using a two-tailed test. When comparing for enrichment, a greater-than one-tailed test was used. Overall survival was examined from date of diagnosis to date of death or last available follow-up. Kaplan-Meier curves were generated and compared using a log-rank test. Cox proportional hazards models were used to generate hazard ratios and CI.

SCG

SCG on four CRC-DDs was performed to ascertain if both RAS/BRAF mutations occurred within the same cells. Core punches of frozen tumor and matched normal were used. Nuclei were isolated by FACS on the basis of DNA content.¹⁵ We performed a multiplexed polymerase chain reaction to directly amplify single-cell DNA (Data Supplement). Gene-specific primers were tailed with M-13 F/R isothermic primers for a combinatorial indexing polymerase chain reaction (Data Supplement). Indices had a 16 or 10 nucleotide barcode at the 5′ end with a Hamming distance > 4 to identify each cell (Data Supplement). Genotypes were predicted using a β-binomial mixture model by training on a set composed of MCF7 and CAMA1 cells genotyped for 12 known polymorphic mutations. The genotype with the highest posterior probability was used as the predicted genotype. Because the genotyped mutations were only a subset of the mutations present in the tumor, CCF and clonality calls for the genotyped mutations in the four cases with SCG were re-estimated using ABSOLUTE.¹⁶ Tumor evolution was inferred with an infinite sites model using infSCITE.¹⁷ A detailed description of laboratory and computational methods is provided in the Data Supplement.

RESULTS

Prevalence of RAS/RAF Dual Drivers and Clinicopathologic Characteristics in CRC Patients With MSK-IMPACT Sequencing Results

Of 4,561 patients with CRC subjected to clinical MSK-IMPACT sequencing, 47 (1.03%) had dual RAS/BRAF driver mutations. CRC-DD samples were more often primary tumors (89.3%) compared with CRC-SD (59.4%; P = .0001; Fig 1A). Of these, 32 harbored dual KRAS/KRAS mutations, 11 harbored dual KRAS/NRAS mutations, two harbored KRAS/BRAF V600E mutations, one harbored an NRAS/BRAF V600E mutation, and one case harbored dual NRAS/NRAS mutations (Fig 1A). CRC-DD demonstrated higher rates of subclonal RAS hotspot mutations in comparison with CRC-SD (Fig 1B). The frequency of BRAF V600E mutations in CRC-DD was lower than that in CRC-SD: 3% versus 13.3%, respectively (P = .0058). Median age at diagnosis (56 v 54 years, P = .23) and proximal tumor location prevalence (53% v 47%, P = .43) did not differ significantly between CRC-DD and CRC-SD. Forty-three CRC-DD cases were adequate for clonality analyses using FACETS. Four groups of CRC-DD were observed after clonality prediction: (1) both mutations clonal (nine cases, 21%), (2) one mutation clonal and one mutation subclonal (17 cases, 40%), (3) both mutations subclonal (eight cases, 19%), and (4) with one or more indeterminate CCFs (nine cases, 21%; Fig 1C). Four cases (9%) showed one atypical KRAS V14I mutation in combination with a typical KRAS mutation (Data Supplement).

FIG 1.

Mutation summary of the CRC-DD cohort. (A) Oncoprint of KRAS, NRAS, and BRAF V600E hotspot mutations in 47 CRC-DD patients. (B) Comparison of clonal and subclonal frequencies shows an increased proportion of subclonal mutations in the CRC-DD compared with CRC-SD. (C) Cancer cell fraction estimates of dual RAS/BRAF V600E mutations in patients were used to separate CRC-DD into three multigroups: clonal-clonal, clonal-subclonal, subclonal-subclonal, and one containing cases with indeterminate clonal calls. With clonality status stratified by RAS/RAF genes, we observe an abundance of subclonal KRAS mutations in the CRC-DD as compared with CRC-SD, where most mutations are clonal. NRAS follows the same trend albeit with a much smaller sample size. Neither BRAF V600E mutation was subclonal. (D) Kaplan-Meier curves of CRC-DD versus CRC-SD and wild-type CRC suggest that CRC-DD have slightly longer overall survival than CRC-SD without reaching statistical significance. (E) Multivariate analysis suggesting single-mutant status, stage at diagnosis, and tumor location are the most influential variables contributing to the differences in survival. AI, allelic imbalance; CRC, colorectal cancer; CRC-DD, colorectal cancer dual-driver; CRC-SD, colorectal cancer single-driver; MSI, microsatellite instability.

No significant differences were observed in overall survival between CRC-DD and CRC-SD (P = .07; Fig 1D) nor between CRC-DD and RAS/RAF wild-type CRC (P = .80; Fig 1D). The median CRC-DD survival was 63.5 months (95% CI, 56.1 to not-reached), whereas the median CRC-SD survival was 42.9 months (95% CI, 38.9 to 53.1). Multivariate analysis showed that CRC-SD were linked with worse overall survival (hazard ratio = 1.74; 95% CI, 1.34 to 2.30; P < .001; Fig 1E).

Genetic Heterogeneity in CRC-DD Resolved by SCG

SCG performance tests showed a 96% mutation detection rate and 92% genotype prediction accuracy (Data Supplement). The presence of dual-clonal RAS/BRAF mutations in nine CRC-DD cases suggested that these mutations co-occur within the same cell. To confirm this, we performed SCG on 2,094 cells across four CRC-DD cases (P-0021518, P-0061563, C-XPMDEU, and P-0008729) and obtained genotypes on 1,170 cells with a median genotype posterior probability of (P(g)) ≥ 0.99 (interquartile range 0.98-1.00 for mutations and interquartile range 0.62-1 for frame shifts; Data Supplement).

Six mutations were detected by MSK-IMPACT for P-0021518: PTEN I135V, APC S1505Afs, APC I1008Yfs, KRAS G12C, NRAS G13D, and TP53 C275Y (Fig 2). A total of 480 cells were sorted from four tumor regions of P-0021518 (Q1→4) and one normal control region (Q5), of which 387 yielded genotypes for all mutations (83, 85, 71, 80, and 68 cells for Q1→5, respectively). The average number of tumor cells across all regions was 74 ± 5 (93.6% ± 6.6%). The 75 cells from non-neoplastic tissue had homozygous wild-type genotypes for all mutations. In the tumor sample, 21 cells were found with all wild-type genotypes from the tumor regions, and 298 cells harbored at least one genotyped mutation (Fig 2A). We observed a same-cell co-occurrence of KRAS G13D and NRAS G12C in 95% of tumor cells. These results are concordant with FACETS results from MSK-IMPACT data, which predicted the NRAS and KRAS mutations to be clonal without AI (CCF 1.0 and 0.96, respectively, total copy number = 2). A subset of 11 cells with SCG data was submitted for Sanger sequencing and confirmed the mutations (Fig 2B). We replicated the Sanger sequence of NRAS G12C and KRAS G13D in an independent set of 16 cells and confirmed the same-cell mutation co-occurrence (Figs 2B and 2C).

FIG 2.

MSK-IMPACT and single-cell sequencing results for P-0021518. (A) (top) Single-cell sequencing of tumor cells from four distinct tumor regions (Q1→ 4), adjacent normal tissue (N5) from the same patient, and (bottom) clinical MSK-IMPACT estimates for CCF by ABSOLUTE, VAF, TCN, EMC, FACETS number state (CN State), and AI are shown. Frequencies of all mutations assessed by single-cell sequencing were highly consistent with MSK-IMPACT results. The single-cell sequencing results were also consistent with the EMC, showing heterozygous genotypes for all mutations with TCN of 2 and EMC approximately 1. (B) Technical and biological validation of NRAS G13D and KRAS G12D genotypes by Sanger sequencing. Technical replicates matched the single-cell sequencing results. Biological replicates were independent cells isolated from regions Q1 and Q2 that were not previously processed and confirmed the co-occurrence of NRAS G13D and KRAS G12D mutations in the same tumor cells. (C) Representative Sanger electropherograms from single cells showing the genotypes for NRAS G13D- and G12C-mutant and wild-type cells. (D) (top) Comparison of across the four tumor regions (Q1→ 4) any significant difference in the frequency of co-occurrence NRAS/KRAS mutations. (E) Inferred tumor evolution tree using an infinite sites model on single-cell genotype suggests that APC, PTEN, and TP53 were first events, followed by NRAS mutation, a second APC mutation, and finally KRAS G12C. The number of cells used to infer each event is indicated below its corresponding event. In total, 306 cells were identified with dual NRAS/KRAS mutations. (F) Pie charts show the occurrence of both MAPK mutations on the total number of cells sequenced per case. AI, allelic imbalance; CCF, clonal cell fraction; EMC, expected number of mutated copies; M, mutant; MAPK, mitogen-activated protein kinase; NA, not available; TCN, total copy number; VAF, variant allele fraction; WT, wild-type.

To determine if P-0021518 displayed spatial heterogeneity in the proportion of tumor cells harboring both mutations, we sequenced several regions. Same-cell co-occurrence of NRAS G12C and KRAS G13D mutations were detected in 95%, 93%, 96%, and 99% of sequenced tumor cells for regions Q1 through Q4, respectively (Fig 2D). Tumor evolution inference with the SCG genotypes showed that the first mutations were APC S1505Afs, PTEN I35V, and TP53 C275Y (Fig 2E). The specific timing of the loss of heterozygosity could not be assessed. NRAS G13D occurred before a second inactivating mutation on APC (I1008Yfs) and finally the KRAS G12C mutation.

C-XPMDEU was assessed for KRAS G12D and NRAS G12V. Of 334 cells, 27 (7.9%) harbored at least one mutation (Data Supplement). Of the 27 tumor cells, 13 (48%) had mutations in both KRAS G12D and NRAS G12V (Fig 2F, Data Supplement).

P-0008729 was microsatellite instability–high (MSI-H) with BRAF V600E and KRAS G12D. Of 320 cells, 84 (26.3%) had at least one mutation (Data Supplement). Seventy-five percent of these cells harbored a KRAS G12D mutation, and 69.1% of tumor cells had a BRAF V600E mutation. KRAS and BRAF mutations co-occurred in 44% of tumor cells (Fig 2F, Data Supplement).

Finally, P-0061563 had KRAS G12D and NRAS G13V mutations. Of 129 cells, 86 had at least one mutation. KRAS G12D mutation occurred in 98.8% of cells and NRAS G13V was detected in 13.9% of tumor cells. All NRAS G13V were detected in cells with an existing KRAS G12D mutation (Fig 2F, Data Supplement).

Multiregional Sequencing in CRC-DD

Multiregional sequencing was performed in seven CRC-DD patients (Fig 3). Five patients had both driver mutations in the additional tumor sites. The 2 cases with discordant multiregional genotypes were MSI-H: P-0001237 harbored BRAF V600E and KRAS G13D in the primary CRC while the abdominal wall and liver metastasis showed only BRAF V600E, and P-0039163 harbored both KRAS G12D and K117N mutations in the primary tumor but only KRAS G12D in the liver metastasis. Interestingly, P-0039163 was MSI-H in the colon primary tumor, while the liver metastasis was microsatellite stability (MSS). The MSI status was confirmed by mismatch repair immunohistochemistry for both samples, and the clonal relationship was demonstrated by the presence of several shared somatic mutations of MSK-IMPACT. These results show that although the clonality of an individual sample of CRC-DD is variable, the majority of CRC-DD retain both driver mutations across multiple primary and metastatic sites.

FIG 3.

Multiregional sequencing of dual RAS hotspot/BRAF V600E mutations across tumor foci. Of seven CRC-DD patients, five MSS CRC-DD displayed both MAPK mutations across all samples sequenced. Two CRC-DD cases demonstrated discordance with only the primary tumor having both drivers and the metastasis retaining only one MAPK hotspot mutation. CRC, colorectal cancer; RC-DD, colorectal cancer dual-driver; MAPK, mitogen-activated protein kinase; MSI-H, microsatellite instability–high; MSS, microsatellite stability.

Copy Number and Allelic Imbalance Characteristics of Dual RAS/RAF-Driven CRC Tumors

Genome-wide copy number analysis revealed that 14% of CRC-DD and 26.6% of CRC-SD were polyploid. CRC-DD contained highly recurrent gains of chromosome eight in 23 (56%) cases, chromosomes 13 and 20q in 21 (51%) cases, chromosome 7 in 17 (41%) cases, and losses on/chromosomes 10q and 18 in 9 (21%) and 22 (47%) cases, respectively (Data Supplement).

Because AI favoring an oncogenic mutant allele is an important mechanism of cell fitness and corresponds with targeted therapy response,¹² we analyzed it in the CRC-DD cohort. AI occurred in 26 of the 86 RAS/BRAF V600E mutations of these tumors. At the gene level, AI was most prevalent in KRAS, accounting for 21 of 65 instances, followed by NRAS with five of 13 instances. AI was present in clonal and subclonal in equal proportions of NRAS and KRAS. No BRAF V600E mutations displayed AI. In the CRC-DD, AI was significantly enriched in cases where one driver was clonal and the second was subclonal (ie, clonal-subclonal): At least one MAPK driver mutation displayed AI in one of nine cases of the clonal-clonal group, nine of 17 cases in the clonal-subclonal group, and none in the subclonal-subclonal group (P = .0006). KRAS mutations displayed AI in two of 14 cases in the clonal-clonal group, 11 of 17 cases in the clonal-subclonal group, and none of 14 cases in the subclonal-subclonal group, confirming an enrichment of AI in at least one MAPK hotspot mutation in CRC-DD with one clonal and one subclonal MAPK hotspot mutation (P = .0074). Both CRC-DD with clonal KRAS and clonal BRAF V600E were MSI-H without mutant AI. Fourteen patients showed double KRAS-KRAS pairings: Each of the two mutations occurred on different reads, meaning the mutations either occurred in trans or in different cells but not on the same allele (Data Supplement).

Molecular Features of CRC-DD Versus CRC-SD

RAS hotspot and BRAF V600E mutations were significantly more likely to occur subclonally in CRC-DD than CRC-SD (P = 8 × 10^–9). Comparisons for the low-frequency mutations are shown in the Data Supplement. Mutation prevalence in recurrently mutated CRC genes including PIK3CA, APC, and TP53 was not significantly different between CRC-SD and CRC-DD (P > .05).

AI of the oncogenic mutant allele was significantly less prevalent in the CRC-DD (26 of 80 mutations) cohort in comparison with the CRC-SD (119 of 232 mutations) cohort (32.5% v 51.3%, respectively; P = .0026). AI was more frequently seen in KRAS mutations in CRC-DD compared with CRC-SD (P = .038). Neither of the two BRAF p.V600E mutations in the CRC-DD had AI while BRAF p.V600E mutations in the CRC-SD sometimes demonstrated AI (Data Supplement).

CRC-DD had lower genomic instability, or fraction of genome altered (FGA), as compared with CRC-SD (P = 7.99 × 10^–9). The differences persisted even when restricted to MSS tumors (P = 2.67 × 10^–7). The average FGA for the CRC-DD and CRC-SD cohorts was 0.325 ± 0.224 and 0.511 ± 0.244, respectively. The subclonal-subclonal group had the least amount of unstable genomes marked by lower FGA mean (0.163 ± 0.119), and the clonal-subclonal group had the largest variability (0.412 ± 0.261; P = .0007). CRC-SD with subclonal RAS hotspot/BRAF V600E mutations also had lower FGA mean (0.347 ± 0.251 v 0.505 ± 0.239 in subclonal v clonal, respectively; P = .0019; Data Supplement).

MSI-H status was enriched in CRC-DD (P = .028; Data Supplement). Twenty-seven percent of CRC-DD were MSI-H, and all BRAF V600E mutants in the CRC-DD group were MSI-H, compared with an MSI-H rate of 8.7% across the cohort by Yaeger et al (P = .0004). No CRC-DD harbored POLE exonuclease domain hotspot mutations. Because cytosine to thymine transition mutations are part of the signature of MSI-H/mismatch repair deficiency, we surveyed our MSI-H CRC-DD oncogene driver mutations (Data Supplement for methods). Of the 12 MSI-H CRC-DD, all had at least 1C>T or G>A (corresponding mutation on complementary strand) driver mutation, with both drivers consisting of C>T or G>A mutations in seven of the 12 MSI-H CRC-DD. Of the 33 MSS CRC-DD, 16 had one driver mutation with C>T or G>A mutation, 8 had C>T or G>A mutations as both driver mutations, and nine had neither oncogenic MAPK driver mutation with C>T or G>A mutation. These analyses show that the C>T/G>A transitions were enriched in oncogenic MAPK pathway mutations in MSI-H CRC-DD (P = .043, one-tailed test), with MSI-H chronologically preceding these mutations.

DISCUSSION

This study details an overlooked group of targeted therapy-naive CRC with co-occurring MAPK pathway driver mutations. Indeed, we found that the majority (99%) of CRC had either a single-driver or no canonical-driver mutation. Yet, CRC-DD cases account for 1% of advanced, targeted therapy-naive CRC. Intratumoral heterogeneity may be present; both mutations are usually present across multiple tumor sites; and mutant allelic imbalances which increase oncogenic mutant dose are less common in CRC-DD. We find that concurrent BRAF V600E and RAS hotspot mutations are extremely rare and are only seen in MSI-H CRC. Four cases (9%) showed co-occurrence between a typical KRAS mutation and an atypical KRAS V14I of cases, as has also been seen by Pietrantonio et al.¹⁸ This mutation likely confers higher MAPK activity than wild-type KRAS, as recently proposed by Loree et al.¹⁹ Furthermore, we show that dual MAPK pathway driver mutations co-occur within the same tumor cell at a relatively high frequency. Our findings contradict the classic belief that MAPK pathway driver mutations such as RAS hotspot mutations and BRAF V600E occur only in mutual exclusivity of each other.²⁰ While the mechanism behind how this occurs remains unknown, the biological reason for the existence of dual drivers might be related to the dosage of MAPK output and the increased fitness it provides.

Recently, oncogene mutation AI that increases the mutant MAPK pathway allele dose has been recognized as an important event in cancer progression. CNA of wild-type KRAS G12D is associated with higher grade pathology and metabolic reprogramming in non–small-cell lung cancer^21,22 and in vitro sensitivity to MEK inhibition in colon cancer cell lines.²³ AI through loss of wild-type allele or copy number gains is found in 55% of KRAS-mutant cancers.²³ Increasing KRAS-mutant dose in CRC is associated with MAPK pathway dependence, and oncogene mutation AI increases tumor cell fitness.¹²

While loss of the wild-type allele enhances the response to targeted therapy in melanoma,¹² amplification of the mutant allele and KRAS and NRAS hotspot mutations are mechanisms of acquired resistance to BRAF V600E inhibition in CRC.²⁴ The present study included only patients who were naive to targeted therapy, and it is likely this 1.1% of CRC patients would not benefit from targeted therapy as suggested by the aforementioned studies. In addition, our results show that BRAF V600E mutations can co-exist with typical RAS mutations, although this type of tumor is extremely rare: Three in 4,561 (0.06%) of CRC and CCFs range from clonal to subclonal in CRC-DD. Moreover, resistance to targeted therapies often develops within 6 months of treatment through the emergence of on-pathway MAPK alterations from EGFR, BRAF, HER2, and TRK inhibition.^2,8,25,26 These patient data suggest that a small reservoir of dual-driver cells are selected under targeted therapy, and the importance of dual drivers may be a broader phenomenon with implications for targeted therapy in CRC.

We observed that oncogene driver mutation AI was significantly decreased in CRC-DD compared with CRC-SD and that MSI-H status was enriched in CRC-DD: a unique feature of this disease subset. The lack of CNA in MSI-H CRC is due to the established inverse relationship between chromosomal instability and microsatellite instability.²⁷ Because oncogene driver AI requires copy number gain of the mutated allele or loss of the wild-type allele, a second MAPK driver mutation is an alternative mechanism to increase MAPK output in MSI cases intolerant of CNA.

Limitations of this study include the rarity of CRC-DD which limited our statistical power to analyze clinical outcomes. SCG was targeted to address the co-occurrence of the MAPK mutations which limited our ability to define clear evolutionary trajectories beyond these genes. SCG proved effective in defining cell populations with dual drivers, yet a baseline level of noise remains inherent to the technology. Our performance test suggests that our method has a 97% mutation detection rate, with an extremely low false-negative rate of 0.9%-3.7% and a false-positive rate ranging between 2% and 14%, heavily dependent on the training set. On average, 91.6 ± 3.9% of genotypes were correctly predicted. Discriminating between mutant allele dropout and wild-type calls cannot be resolved in rare instances, thus rendering a small margin of error despite stringent mutation-calling criteria.

In conclusion, approximately 1% of CRC harbor dual RAS hotspot/BRAF V600E mutations. These cases are enriched for MSI-H status and have a lower incidence of AI favoring the driver mutation. BRAF V600E and RAS hotspot comutant CRC are more subject than other dual-driver pairings to have only one driver mutation exist clonally. Both mutations often co-occur in at least a proportion of the same tumor cells and may be present across multiple disease foci. Subclonal mutations may be lost in metastases, and liquid biopsy may be important to understand the clinical importance of these events.

DATA SHARING STATEMENT

Data is publicly available through a dedicated study in the cBioPortal (https://www.cbioportal.org/study/summary?id=crc_dd_2022). Code and training/test data for the β-binomial mixture model is available via GitHub (https://github.com/RodrigoGM/bbmm).

AUTHOR CONTRIBUTIONS

Conception and design: Rodrigo Gularte Mérida, Britta Weigelt, Jorge S. Reis-Filho, Jaclyn F. Hechtman

Financial support: Jorge S. Reis-Filho, Jaclyn F. Hechtman

Provision of study material or patients: Jaclyn F. Hechtman

Collection and assembly of data: Rodrigo Gularte Mérida, Anita S. Bowman, Arnaud da Cruz Paula, Monika Vyas, Laetitia Borsu, Ahmet Zehir, Rona Yaeger, Fang Fang, Rui Gardner, Jaclyn F. Hechtman

Data analysis and interpretation: Rodrigo Gularte Mérida, Shaleigh Smith, Anita S. Bowman, Walid Chatila, Craig M. Bielski, Monika Vyas, Ahmet Zehir, Luciano G. Martelotto, Jinru Shia, Rona Yaeger, Ruibang Luo, Michael C. Schatz, Ronglai Shen, Britta Weigelt, Francisco Sánchez-Vega, Jorge S. Reis-Filho, Jaclyn F. Hechtman

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).