Clinical and molecular characteristics of Class II and III BRAF mutations in colorectal cancer

Abstract

Class II and III BRAF mutations are uncommon events, with limited data on their clinical and biological characteristics. We investigated clinical and molecular features of BRAF mutation classes in a total of 24,402 patients with mismatch repair proficient CRC (MMRp). Samples collected between 2006 and 2023 were profiled using next-generation sequencing and whole-transcriptome sequencing. We identified 1268 (5.2%), 132 (0.54%), and 323 (1.3%) patients with class I, II, and III BRAF-mutated CRC. Patients with class III mutations had significantly better median overall survival (OS) than those with class I mutations (23.6 vs 17.4 months; HR = 1.26, CI: 1.08–1.47, p = 0.004). Transcriptomic analyses revealed that the MAPK pathway score was significantly lower for class II and III BRAF mutations without concurrent RAS mutations than for those with RAS co-mutations. The cetuximab score, an RNA expression-based predictor of EGFR therapy response, was significantly better for class II and III BRAF mutations than for class I. The cetuximab score significantly improved for only class III BRAF mutations when those with concurrent RAS mutations were excluded. The results of this large multi-institutional analysis of BRAF mutation classes reveal the prognostic value of class II and III BRAF mutations, and their distinct clinical and molecular features.

Introduction

Colorectal cancer (CRC) remains one of the leading causes of cancer-related death in Western countries, with an increasing incidence of young-onset colorectal cancer¹. CRC consists of distinct molecular subgroups with different clinical behaviors resulting in various clinical outcomes². The use of next-generation sequencing has revealed several molecular targets for precision medicine and drug development³. Recently, several targeted therapeutic agents have been developed for patients with metastatic CRC, including HER2 amplification, KRAS G12C mutation, and BRAF V600E mutation^4–8. Among these alterations, BRAF mutations confer the most aggressive biology and are associated with significantly inferior survival outcomes, underscoring an unmet need for research to better understand their biology and clinical impact^8,9.

BRAF mutations represent a heterogeneous subgroup of molecular alterations seen in CRC. Class I BRAF mutations consist of V600 mutations. These mutations function as monomers and induce significant oncogenic activity, leading to carcinogenesis and aggressive biology¹⁰. The combination of cetuximab and encorafenib, a BRAF V600E inhibitor, resulted in significant improvement in outcomes for patients with BRAF V600E mutant CRC who experienced progression on systemic chemotherapy¹¹. More recently, the BREAKWATER trial demonstrated significant improvements in overall survival, progression-free survival, and objective response rate when encorafenib and cetuximab were combined with fluorouracil and oxaliplatin, compared to the standard of care, resulting in accelerated FDA approval^8,12. Class II and III BRAF mutants, however, represent biologically distinct classes of BRAF alterations compared to Class I BRAF mutants. Class II BRAF mutants, unlike class I BRAF mutants, function by forming homodimers and carry significant oncogenic activity with increased MAPK pathway activation¹³. Class II BRAF mutants do not respond to currently available BRAF inhibitors, and although they may be associated with modest response to EGFR blockade, the data regarding their EGFR responsiveness is very limited¹³. Class III BRAF mutants often exhibit reduced tyrosine kinase activity, leading to limited oncogenic potential. Their activation requires the formation of heterodimers with C-RAF and other BRAF oncoproteins, and they can be concurrently seen with KRAS mutations¹⁴. Limited data suggest that class III BRAF mutants are associated with better outcomes than those with BRAF class I mutants¹⁵. Some case series and case reports suggest that EGFR blockade therapy may have clinical benefits for non-class I mutations; however, the information is limited to institutional experiences and reports, and there is a significant knowledge gap on clinical and biological features of class II and III BRAF mutants.

In this large multi-institutional study, we investigated the molecular characteristics of BRAF mutation classes and their association with the clinical characteristics of patients with CRC. We also examined the responsiveness of BRAF mutation classes to EGFR blockade and investigated survival outcomes with and without EGFR blockade therapy in patients with BRAF-mutated CRC.

Results

Patient clinicopathologic characteristics

In our study, a total of 24,327 patients with metastatic MMRp CRC were included. Among those, 1268 (5.2%), 132 (0.5%), and 323 (1.3%) patients had class I, II, and III BRAF mutations, respectively. The incidence of class I BRAF mutations among peritoneal and retroperitoneal samples was 9.0%, while this was only 2.5% among those with lung samples (P < 0.0001). Class II and III BRAF mutations occurred at similar rates across samples from different sites (Table 1). Among 17,973 patients with known race information, 71.4% were white, 17.9% were black, and 4.2% were Asian or pacific islander. Notably, class I BRAF mutation rates were significantly lower among the Black population compared to the Caucasian population and in males compared to females (1.86% vs 6.55%, P < 0.0001, and 4.0% vs 6.7%, P < 0.0001, respectively; Table 1). While class I BRAF mutations were less common among patients aged <50 compared to patients ≥ 50 (3.2% vs. 5.7%, P < 0.001), class II and III BRAF mutation rates were similar in these age groups (0.4% vs 0.6%, P = 0.14; 1.1% vs 1.4%, P = 0.19, respectively). The incidence of BRAF mutation classes among age groups is shown in Table 1. While the majority of class I BRAF mutations were present in right-sided tumors, patients with class II and III BRAF mutations and BRAF WT had more left-sided cancers (Table 1).

Table 1.

Clinical and demographic characteristics of patients with BRAF class I, II, and III mutations

		MSS Class 1	MSS Class 2	MSS Class 3	MSS WT	Total	P values
Gender	Female	720 (6.73%)	60 (0.56%)	145 (1.36%)	9776 (91.36%)	10,701	<0.0001
	Male	548 (4.02%)	72 (0.53%)	178 (1.31%)	12,828 (94.14%)	13,626
Age	Median Age	66	64	63	62	62	<0.0001
	Age IQR	57–74	55–70	53–72	52–70	53–71
	<50 yrs old	138 (3.2%)	17 (0.4%)	49 (1.1%)	4128 (95%)	4332
	>=50 yrs old	1130 (5.7%)	115 (0.6%)	274 (1.4%)	18476 (92.4%)	19,995
Race	Asian or Pacific Islander	24 (3.19%)	3 (0.40%)	8 (1.06%)	718 (95.35%)	753	<0.0001
	White	840 (6.55%)	70 (0.55%)	175 (1.36%)	11744 (91.54%)	12,829
	Black or African American	60 (1.86%)	15 (0.47%)	43 (1.33%)	3104 (96.34%)	3222
	Other	51 (4.36%)	4 (0.34%)	17 (1.45%)	1097 (93.84%)	1169
	Unknown	293 (4.61%)	40 (0.63%)	80 (1.26%)	5941 (93.50%)	6354
Ethnicity	Hispanic or Latino	119 (4.22%)	17 (0.60%)	35 (1.24%)	2651 (93.94%)	2822	<0.0001
	Not Hispanic or Latino	865 (5.82%)	75 (0.50%)	201 (1.35%)	13727 (92.33%)	14,868
	Unknown	284 (4.28%)	40 (0.60%)	87 (1.31%)	6226 (93.81%)	6637
Sidedness	Left-sided	361 (28.47%)	59 (44.7%)	172 (53.25%)	13058 (57.77%)	13,650	<0.0001
	Right-Sided	541 (42.67%)	34 (25.76%)	83 (25.7%)	4617 (20.43%)	5275
	Transverse	130 (10.25%)	4 (3.03%)	7 (2.17%)	856 (3.79%)	997
	Other/Unclear	236 (18.61%)	35 (26.52%)	61 (18.89%)	4073 (18.02%)	4405
Specimen Sites	Colon	709 (7.05%)	54 (0.54%)	120 (1.19%)	9176 (91.22%)	10,059	<0.0001
	Liver	186 (3.60%)	35 (0.68%)	72 (1.39%)	4873 (94.33%)	5166
	Rectum	70 (2.14%)	12 (0.37%)	58 (1.78%)	3124 (95.71%)	3264
	Lung	33 (2.48%)	9 (0.68%)	18 (1.35%)	1270 (95.49%)	1330
	Peritoneum	85 (8.99%)	5 (0.53%)	16 (1.69%)	839 (88.78%)	945
	Lymph Node	38 (5.97%)	2 (0.31%)	12 (1.88%)	585 (91.84%)	637
	Unclear/Other	147 (5.02%)	15 (0.51%)	27 (0.92%)	2737 (93.54%)	2926

Molecular landscape of BRAF mutation classes

Patients with class I BRAF mutations were more likely to fall into the consensus molecular subtype 1 (CMS1) category than patients with class II or III mutations (class I, II, and III: 47% vs. 13% vs. 18%, respectively; Fig. 1A). Patients with class II and III mutations had more CMS2 subtypes than those with class I mutations (30% and 29% vs. 2%, respectively). Notably, the CMS subtype distribution of patients with class II and III mutations was similar to that of patients with BRAF wild-type (WT) CRC. The MPAS score, which estimates the magnitude of MAPK pathway activation, was significantly higher among patients with class I BRAF mutations than among those with class II or III mts (1.73 versus 0.38 versus 0.79, P < 0.05; Figs. 1B and C). When the MPAS was examined based on concurrent RAS mutation status, class II and III BRAF mutations without RAS mutations had significantly lower MPAS scores than those with RAS mutations (Fig. 1D and Fig. 1E). The T-cell inflamed score was higher in patients with class I and II BRAF mutations than in those with class III BRAF mutations or BRAF wild-type disease (-31.5, -31, -56, and -55, respectively; p < 0.05). A similar pattern was noted for interferon-gamma signature (-0.35, - 0.36, -0.39, and -0.41, respectively, P < 0.05). While KRAS and class I BRAF mutations were nearly mutually exclusive (0.5%), KRAS mutations were relatively common events among patients with class II and III mutations (12.8% and 27.4%, respectively, P < 0.01). Detailed allele-specific KRAS mutations are summarized in Table 2. A similar pattern was noted for NRAS mutations (Suppl. Figure 1). The incidence of PIK3CA mutation was relatively similar across BRAF mutation classes, though a lower incidence was noted in the class I BRAF group compared to class II, II BRAF, and BRAF WT groups (11.9% vs 21.2% vs 19.9%, 17.1%, respectively; p < 0.05). APC mutations were significantly lower among patients with class I BRAF mutations compared to those with other classes and BRAF WT (23.3% vs. 79.7% vs. 78.2% vs 81.6%, respectively, P < 0.001). However, RNF43, another regulator of the WNT pathway, was significantly more mutated among patients with class I BRAF mutations compared to those with BRAF WT (24.1% vs 1.5%). Other molecular alterations with significantly different incidences are provided in the supplementary material (Suppl. Table 1S A, B, C).

Fig. 1. The distribution of CMS classes and MPAS scores across class I, II, and III BRAF mutations with and without concurrent RAS mutations.

A CMS classification by BRAF status, B) MPAK pathway score (MPAS), Interferon-gamma score (IFN) score, T-Cell Inflamed Score by BRAF mutation classes, C) MPAS score across BRAF mutation classes and BRAF WT, D) MPAS score with and without RAS Mutations for Class II BRAF mutations, E) MPAS score with and without RAS Mutations for Class III BRAF mutations.

Table 2.

KRAS mutations in BRAF mutated colorectal cancer

KRAS mutations in Class I BRAF mutations			KRAS mutations in Class II BRAF mutations			KRAS mutations in Class III BRAF mutations
KRAS MT	N	%	KRAS MT	N	%	KRAS MT	N	%
G12V	3	50%	G12V	4	22.2%	A146V	23	26.1%
G13D	1	16.7%	Q61	3	16.7%	G12S	11	12.5%
G12A	1	16.7%	G13D	3	16.7%	Other	11	12.5%
G12D	1	16.7%	A146V	2	11.1%	G13D	9	10.2%
Total	6	100%	G12D	2	11.1%	A146T	8	9.1%
			G12C	1	5.6%	A59E	5	5.7%
			A146T	1	5.6%	G12D	5	5.7%
			G13C	1	5.6%	L19F	5	5.7%
			K117N	1	5.6%	K117N	3	3.4%
			Total	18	100%	G12V	2	2.3%
						Q61	2	2.3%
						A146P	1	1.1%
						A59T	1	1.1%
						G12A	1	1.1%
						G13C	1	1.1%
						Total	88	100%

The impact of BRAF mutation classes on the prognosis of patients with pMMR CRC

Median OS for patients with BRAF WT and BRAF mutation classes (class I, II, and III) was significantly different and more favorable for patients with class II and III BRAF mutations compared to those with class I (29.8, 17.3, 21.7, and 23.6 months, respectively, P < 0.01, Fig. 2A). A similar pattern of OS was also seen among patients who did not receive anti-EGFR therapy, with median OS of 28.3, 15.2, 20.8, and 21.9 months for patients with BRAF WT, BRAF Class I, II, and III mutations, respectively, P < 0.001 (Fig. 2B), indicating that class II and III BRAF mutations are prognostic biomarkers. The prognostic value of class II and III was more significant when the two groups were combined (Supplementary Fig. 2). Patients with class III mutations had significantly better median OS than patients with class I mutations (23.6 vs 17.4 months; HR = 1.26, CI: 1.08–1.47, P = 0.004; Fig. 2C). No significant difference was noted between patients with class I and class II mutations, though a ~4-month mOS difference was observed in patients with class II BRAF mutations, which was not statistically significant (21.7 vs 17.4 months; HR = 1.16, CI: 0.93–1.45, P = 0.177, Fig. 2D). When patients with class II and III were combined in an exploratory analysis to enhance statistical power, the median OS was significantly better than that of patients with class I BRAF mutations (22.8 vs 17.4 months; HR = 1.23, CI: 1.08–1.41, P = 0.002; Fig. 2E). Patients with BRAF WT disease had significantly better OS than those with any BRAF mutation class, including class III BRAF mutations (P < 0.001; Supplementary Fig. 2).

Fig. 2. Overall survival outcomes of patients with class I, II, and III BRAF mutations with and without anti-EGFR therapy exposure.

A Overall population regardless of anti-EGFR therapy. B Patients who never received anti-EGFR therapy. C Patients with class I vs class III BRAF mutations. D Patients with Class I vs Class II BRAF mutations. E Patients with class I vs class II and III BRAF mutations. F Patients with class II BRAF mutation with vs without RAS mutations. G Patients with class III BRAF mutation with vs without RAS mutations.

Distinctive survival outcomes with anti-EGFR therapy and related transcriptomic analyses

Among patients treated with anti-EGFR with or without encorafenib, patients with both class I as well as class II & III BRAF mutations had statistically inferior anti-EGFR-associated survival outcomes compared to patients with BRAF WT (10.5 vs 14.5 vs 26.3 months, P < 0.001, Fig. 3 A1). We performed further analyses by excluding patients who received encorafenib in combination with anti-EGFR therapy; patients with class II and III BRAF mutations had significantly better anti-EGFR-associated survival outcomes compared to those with class I BRAF mutations (14.2 months vs 7.8 months, respectively; P < 0.0001; Fig. 3A2). Notably, no significant difference between BRAF mutation classes was noted in overall population, although numerically different anti-EGFR-associated survival outcomes were seen among patients with class II and III mutations (Fig. 3B–E) without statistical significance (10.4, 21.3 vs 13.8 months, respectively, for class I, II, and III). Information on the timing of anti-EGFR therapy (first-line versus later-line setting) was unavailable for analysis, and the number of patients receiving anti-EGFR therapy in class II and III mutants was limited. We therefore evaluated the cetuximab score by Yang et al.¹⁶ to estimate anti-EGFR response, and notably, class II and III BRAF mutations had a significantly higher cetuximab score than class I BRAF mutations. This was similar to that observed in patients with BRAF WT, suggesting that anti-EGFR therapy may have activity among patients with class II and class III BRAF mutations (Fig. 3F). Notably. cetuximab score further improved for class III BRAF mutations when the analyses were performed after excluding the cases with concurrent KRAS mutations (Fig. 3F).

Fig. 3. Survival outcomes after anti-EFGR therapy across BRAF subclasses and cetuximab score for class I, II, and III BRAF mutations with and without concurrent RAS mutations.

A1 Overall population. A2 Patients who received anti-EGFR only without encorafenib. B Patients with class I vs class II BRAF mutations in the overall population. C Patients with class I vs class III BRAF mutations in the overall population. D Patients with class I vs class II and III BRAF mutations in the overall population. E Patients with class II vs class III BRAF mutations in the overall population. F Cetuximab Score for all BRAF mutation classes and cetuximab scores with and without RAS mutations for class II and III BRAF mutations.

Discussion

BRAF mutations are composed of molecularly and biologically distinct subgroups (Table 3). The clinical characteristics of class II and class III BRAF mutations in CRC are largely unknown due to their rarity. To our knowledge, this is the largest comprehensive study of the BRAF subclasses, in which we identified that class III BRAF mutations in CRC are associated with better prognosis than class I BRAF mutations. This was validated among patients who did not receive anti-EGFR therapy, confirming the prognostic value of BRAF mutation classes. Unlike the other two landmark studies^17,18, this study observed improved overall survival among patients with BRAF WT compared with those with class III BRAF mutations. This may be due to frequent concurrent KRAS and class III BRAF mutations (27.4%), which may explain the difference between our findings and previous reports. In fact, in our study, we demonstrated that patients with RAS wild-type class II and class III BRAF mutations have significantly better overall survival than those with concurrent RAS mutations, providing further insight into the impact of RAS mutations in class II and III BRAF-mutated CRC. It is also important to note that the study by Schirripa et al.¹⁷ reported outcomes of 13 patients with class III BRAF mutation, while in our study, this was 327, and the small cohort size included in other studies remains a major limitation to derive a definitive conclusion from other reports and may also explain the differences in overall outcomes in our study and other reports. We also observed a trend toward numerically longer overall survival in patients with class II BRAF mutations compared to class I (21.7 vs 17.3 months), which was not statistically significant, likely due to sample size.

Table 3.

BRAF mutation classes by clinical and molecular characteristics in colorectal cancer

	Class I	Class II	Class III
Alterations	V600E, V600K, V600D, V600M, V600R	G469; G464, N486, T599, P367, K601, L597, BRAF kinase duplication and fusions	D594, D596, G466, G469E, F595, G596, D387, N581, V459, S467, D287
Frequency*	6–8%	~0.5–1%	~1–1.5%
Kinase Activity	+++	++	Low (kinase-dead)
Upstream RAS activation dependence	Independent	Independent	Dependent
Monomer/dimer	Monomer	Homodimer/heterodimer	Heterodimer
CRAF heterodimerization	None	None	Present
EGFR activation	Not required	Not required	Required if no concurrent RAS mutation present
Response to currently FDA-approved BRAF inhibitors	Yes	No	No
Response to EGFR blockade alone	No	Limited	Modest
Tumor side association	Right (more likely)	Left (more likely)	Left (more likely)
Prognosis	Poor	Better than Class I	Better than Class I

*RTK receptor tyrosine kinase, *EGFR epidermal growth factor receptor, MMR-D mismatch repair deficient, MSI-H microsatellite instability high. *Frequency in colorectal cancer.

In our study, although distinct survival outcomes with anti-EGFR therapy were seen with class II and III BRAF mutations, it did not translate into a statistical difference, which may be due to sample size and inherent limitations of the database, including various time points for anti-EGFR therapy initiations based on practice patterns and data obtained from insurance claims, which may introduce lead-time bias. It is also important to note that our study is limited by ant-EGFR treatment heterogeneity, including the lines of therapy in which they were used, and that treatment-related outcomes should be interpreted within this context. Our study also revealed significant biological and clinical differences in BRAF mutation classes. Notably, peritoneal samples had higher rates of class I BRAF mutations, while this was notably lower among lung samples. We observed a lower incidence of BRAF mutations in the African American population (1.8%), whereas the rate among Caucasians was similar to that of the overall population. Similarly, we observed a significantly lower incidence of class I BRAF mutations among males. We identified that, unlike patients with class I BRAF mutations, patients with class II and III BRAF mutations are more likely to present with left-sided tumors, and they are more likely to have CMS2 subtype with concurrent KRAS/NRAS mutations (27.4%), and the incidence of KRAS mutation in class III BRAF mutations was not reported in previous studies. This novel information is particularly important, as RAS and BRAF oncogenes are both activators of the MAPK pathway, and co-occurrence may directly impact therapeutic interventions, including anti-EGFR therapy. Notably, our study is also the first to reveal the biological relevance of RAS co-mutations, providing transcriptomic analyses that shed light on their impact on the MAPK pathway (Fig. 1D, E) and anti-EGFR responsiveness (Fig. 3F). These findings are particularly important for future drug development targeting class II and III BRAF mutations, where concurrent RAS mutations may confer treatment resistance.

BRAF class I mutations (V600) have been well studied, and they are associated with significantly worse outcomes among patients with colorectal cancer^4,19,20. BRAF V600E mutations are now among actionable alterations based on evidence from the BEACON and BREAKWATER trials; however, the outcomes of patients with BRAF V600E mutations remain relatively unfavorable¹¹. Class II and III mutations are a biologically distinct subgroup of BRAF mutations, and their oncogenic activity also varies significantly. While class I BRAF mutations function as monomeric proteins, class II and III BRAF mutations undergo homo- or heterodimerization for activation and do not respond to currently available BRAF inhibitors^11,17. Class III BRAF mutations may result in reduced tyrosine kinase activity and, in some cases, produce kinase-dead proteins. They function by forming heterodimers with other oncoproteins, such as CRAF, and upstream RAS activation may be needed for their cellular activity¹⁴. Consistently, preclinical data suggest that tumors with class III BRAF mutations without concurrent RAS mutations may respond to anti-EGFR therapy, as class III BRAF mutations are dependent on EGFR-driven RAS signaling in colorectal cancer¹⁴. This preclinical evidence aligns with our results, in which we demonstrated a distinct MPAS score that is significantly higher in class I BRAF mutations than in class II and III BRAF mutations. A case series reported meaningful tumor regression among patients with class III BRAF mutations, whereas it was modest among those with class II BRAF mutations¹³. Consistently, in our study, we also demonstrated an improved cetuximab score in patients with class II and III BRAF mutations compared with those with class I BRAF mutations. Class II BRAF mutations function in an RAS-independent manner by forming homodimers and inducing oncogenic activity. Although currently available BRAF inhibitors have no activity against class II BRAF mutations, next-generation BRAF inhibitors that disrupt dimerization and are currently being investigated in clinical trials with MEK inhibitors²¹. More drug development is on the horizon with panRAF inhibitors that may offer more therapeutic avenues for patients with non-V600 BRAF mutations²².

Our study also uncovers fundamental biological and clinical differences between class I, II, and III BRAF mutations. We identified a highly distinct pattern of CMS classification in this group, and notably, non-class I BRAF mutations had a CMS signature similar to that of BRAF WT MSS CRC. Highly reduced canonical pathway enrichment was observed among patients with class I BRAF mutations. Consistently, the incidence of APC mutations, a hallmark mutation of the canonical molecular subtype of CRC (CMS2), was significantly lower among patients with class I BRAF mutant CRC. Patients with class II and III BRAF mutations had CMS distributions that were relatively similar to those with BRAF WT, revealing important molecular features of non-V600 BRAF mutations. Although this does not directly impact current practices, it provides further insight for future drug development where class II and III BRAF mutants are more likely to benefit from therapeutic approaches targeting canonical pathways. Unlike class I BRAF mutations, frequent RAS mutations in other BRAF mutation classes were notable, albeit lower than in patients with BRAF WT disease. This is particularly important for drug development as RAS mutations may activate bypass mechanisms for RAF/MEK inhibitors, given the ability of the KRAS oncoprotein to activate PIK3CA and the mTOR pathway in addition to the MAPK pathway²³. MPAS scores of each BRAF mutation class were consistent with the oncogenic MAPK activity of each class of BRAF mutations, where class II and III BRAF mutations demonstrated significantly lower MAPK activity as compared to class I BRAF mutations, and this was further notable when cases with RAS co-mutation were excluded (Fig. 1C–E).

Although our study represents the largest cohort of BRAF subclasses with several important clinical and transcriptomic data, there are limitations to our study. First, the treatment history was limited in our records and line of therapy, and the underlying reason for treatment discontinuation was not available, leading to the inability to assess anti-EGFR benefits precisely. Our analysis was also limited by the lack of detailed clinical information, including the tumor’s sidedness, the exact date of diagnosis for patients, other relevant treatment history, and the exact duration of individual therapies, which may affect the precise measurement of treatment-related survival outcomes.

Collectively, the evidence from this large, multi-institutional cohort study indicates that class III BRAF mutations are associated with better prognosis than class I BRAF mutations among patients with MSS CRC. Our study also suggests that clinical and biological characteristics of class II and III BRAF mutations are highly distinct from BRAF V600 mutations, and this difference should be considered for therapeutic and prognostic discussion. Lastly, differences seen in the molecular underpinnings of class II and class III BRAF mutations provide further insight into potential resistance mechanisms for agents in development targeting these uncommon BRAF alterations, as well as KRAS mutations.

Methods

Patient samples

A cohort of 24,327 formalin-fixed paraffin-embedded (FFPE) tumor samples from patients with MSS/MMRp (microsatellite stable/mismatch repair proficient) colorectal cancer were submitted for molecular profiling by a CLIA-certified laboratory between 2006 and 2023 (Caris Life Sciences, Phoenix, AZ, USA). This study was conducted in accordance with the guidelines of the Declaration of Helsinki, the Belmont Report, and the U.S. Common Rule. In keeping with 45 CFR 46.104(d)(4), this study was performed utilizing retrospective, de-identified clinical data. Therefore, this study is considered IRB-exempt, and no patient consent was necessary from the subject.

Tumor specimen processing

Molecular profiling was performed at Caris Life Sciences (Phoenix, AZ, USA), a College of American Pathologists (CAP)/Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory. Hematoxylin and eosin (H&E) stained formalin-fixed, paraffin-embedded (FFPE) slides of the patient’s tumor underwent review by a board-certified pathologist or trained pathologist assistant. Tumor enrichment was achieved by harvesting targeted tissue using manual microdissection techniques.

RNA analysis and signatures

For whole transcriptome sequencing (WTS), a minimum of 10% of tumor content in the area for microdissection was required. The Qiagen RNA FFPE tissue extraction kit was used for extraction, and the RNA quality and quantity were determined using the Agilent TapeStation. Biotinylated RNA baits were hybridized to the synthesized and purified cDNA targets, and the bait-target complexes were amplified in a post-capture PCR reaction. The Illumina NovaSeq 6500 was used to sequence the whole transcriptome from patients to an average of 60 M reads. Raw data were demultiplexed by Illumina Dragen BioIT accelerator, trimmed, counted, PCR duplicates removed, and aligned to the human reference genome hg19 by STAR aligner. For transcription counting, transcripts per million (TPM) values were generated using the Salmon expression pipeline, and four RNA expression-based scores were calculated: Interferon-gamma signature²⁴ and T-cell inflamed signature²⁵ were previously published to be associated with immune checkpoint inhibitor outcome in various cancer types, cetuximab sensitivity score (CTX-S)¹⁶, an RNA expression-based predictor of EGFR therapy response, which was first described by Yang et al as investigational research assay¹⁸ while MAPK Pathway Activity Score (MPAS) was derived from ten conserved MAPK pathways genes²⁶ and associated with MEK inhibitor sensitivity in cancer cells.

Next-generation sequencing (NGS)/whole exome sequencing (WES)

Genomic DNA was isolated from formalin-fixed paraffin-embedded (FFPE) tumor samples and sequenced using the NextSeq or NovaSeq 6000 platforms (Illumina, Inc., San Diego, CA). For NextSeq sequenced tumors, a custom-designed SureSelect XT assay was used to enrich 592 whole-gene targets (Agilent Technologies, Santa Clara, CA). For NovaSeq sequenced tumors, more than 700 clinically relevant genes at high coverage and high read-depth were used, along with another panel designed to enrich for an additional >20,000 genes at lower depth. All variants were detected with >99% confidence based on allele frequency and amplicon coverage, with an average sequencing depth of coverage of >500 and an analytic sensitivity of 5%. Genetic variants identified were interpreted by board-certified molecular geneticists according to the American College of Medical Genetics and Genomics (ACMG) standards.

MSS/MMRp status determination

Immunohistochemistry (IHC) and next-generation or whole-exome sequencing (NGS/WES) were used to assess MSS/MMRp (microsatellite stable/mismatch repair proficiency) status. These platforms generate highly concordant results as previously reported²⁷. In the rare cases of discordant results, status was determined in the order of IHC, then NGS.

For IHC, the tumor was considered mismatch repair proficient (MMRp) if expression of all four proteins was observed: MLH1 (M1, RRID: AB_2336022), MSH2 (G2191129, RRID: AB_2936886), MSH6 (44, RRID: AB_2336020), PMS2 (EPR3947, RRID: AB_2336010) (Roche (Ventana), Tucson, AZ, USA). IHC was performed on a Ventana Benchmark automated slide preparation system (Roche (Ventana), Tucson, AZ, USA).

For NGS/WES, MSI. MSI was determined from NGS data by analyzing INDEL mutations in an optimized number of microsatellite loci depending on the NGS assay utilized (592-gene panel and WES). The number of loci analyzed was >2000 for all NGS assays. The threshold to determine MSI-high also varied depending on the NGS assay used (592: ≥46, which was adjusted to ≥21 when the assay was optimized with fewer loci examined; WES: ≥116). Indeterminate results were reported for samples with low average depth of coverage (<500× for 592 and WES).

Consensus molecular subtype (CMS) classifier

The Caris CMS classifier²⁸ was developed using RNA sequencing data collected from a whole transcriptome sequencing (WTS) platform (NovaSeq 6000, Illumina, CA) (RRID: SCR_016387). A full 22,948-gene dataset of expression data was produced by the Salmon RNASeq pipeline (RRID: SCR_017036), which provides fast and bias-aware quantification of transcript expression²⁹. This pipeline yields discrete TPM (transcripts per million molecules) values for each gene transcript. A classifier was trained against the original CMS datasets (excluding a TCGA validation dataset of 512 cases) published by Guinney et al. using a classic SVM model as implemented in R (RRID: SCR_000432)². Six hundred genes were subsequently selected for each of the four CMS classifiers using One vs. Rest t-test to identify genes uniquely expressed in each of the four classes (N = 2400 total genes). Cross-validation was performed to optimize the model and finalize SVM parameters. Possible overtraining was evaluated by predicting CMS subtypes from an independent, blinded dataset (TCGA dataset of 512 samples) with an accuracy of 88.0%.

Survival analysis and statistics

Real-world overall survival (OS) data were obtained from insurance claims information and calculated from the time of tissue collection as a surrogate for diagnosis or start of treatment until last contact. Patients without contact/claims data for a period of at least 100 days were presumed deceased. Conversely, patients with a documented clinical activity within 100 days prior to the latest data update were censored in the analysis³⁰. Kaplan–Meier survival estimates were generated for cohorts defined by molecular characteristics. The hazard ratio (HR) was computed utilizing the Cox proportional hazards model, and the significance of differences (p-values) in survival times was assessed with the log-rank test. Molecular comparisons were performed using Chi-square or Fisher’s exact test, and p values were adjusted for multiple comparisons using Benjamini–Hochberg (q < 0.05).

Compliance statement

This study was conducted in accordance with the guidelines of the Declaration of Helsinki, Belmont Report, and U.S. Common Rule. In keeping with 45 CFR 46.101(b)(4), this study was performed utilizing retrospective, deidentified clinical data and is considered IRB exempt with waivers of patient-informed consent. Waiver of patient consent and exempt status was determined by WCG IRB.

Author contributions

I.H.S. (Conceptualization, investigation, methodology, project administration, writing— original draft), J.X., Y.B. (Data curation, formal analysis, writing). All authors (Investigation, writing, and review & editing).

Data availability

The data presented in this study are not publicly available due to data size constraints and patient privacy considerations, butmay be obtained from the corresponding author upon reasonable request. The NGS raw data are owned by Caris Life Sciences.Qualified researchers may apply for access to these data by contacting Joanne Xiu (jxiu@carisls.com) and executing a datausage agreement.

Competing interests

I.H.S. has declared Advisory Board fees from Seattle Genetics, Guardant, Amgen, Pfizer and GSK, and Lumanity; research funding from BAYER, Roche, and GSK. M.K. has disclosed consulting or advisory roles at Cardinal Health, Muerus, and TriSalus Life Sciences; stock and other ownership interests from Cardiff Oncology; Speakers' Bureau at Caris Life Sciences; and research funding from Abbvie (Inst), FOGPharma (Inst), IgM Biosciences (Inst), and Torl Biotherapeutics (Inst). A.G. has declared honoraria: Total Health Conferencing, Cardinal Health; consulting or advisory role: Bristol Myers Squibb (Inst), Lilly (Inst), Amgen (Inst), Daiichi Sankyo (Inst), OBI Pharma, Caris Life Sciences (Inst), Guardant Health (Inst), Natera (Inst), Replimune (Inst); research funding: Eisai (Inst), Lilly (Inst), Daiichi Sankyo (Inst), Replimune (Inst), Natera (Inst), Caris Life Sciences (Inst), Guardant Health (Inst); travel, accommodations, expenses: Cardinal Health, Total Health Conferencing. B.A.W. has declared employment: Apple (I); stock and other ownership interests: Apple (I); honoraria: Primum, Caris Life Sciences; consulting or advisory role: Merus NV, DoMore Diagnostics, Foundation Medicine, Agenus, Regeneron, CytoDyn; Speakers' Bureau: Sirtex Medical, Seagen, Merus NV, Daiichi Sankyo/AstraZeneca, Natera, Jazz Pharmaceuticals, Pfizer; research funding: Ipsen (Inst), Immune System Key (Inst), Merck (Inst), BioXCel therapeutics (Inst); expert testimony: AstraZeneca; travel, accommodations, expenses: Merus NV. M.A. has disclosed consulting or advisory roles at AstraZeneca, Curio Science, Eisai, Exelixis, Genentech, GlaxoSmithKline, Incyte, Ipsen, Isofol Medical, QED Therapeutics, Taiho Pharmaceutical; and research funding from AstraZeneca (Inst), Bayer (Inst), Bellicum Pharmaceuticals (Inst), Boehringer Ingelheim (Inst), Bristol-Myers Squibb-Ono Pharmaceutical (Inst), Eisai (Inst), IMPAC Medical Systems (Inst), Merck Sharp & Dohme (Inst), Pfizer (Inst), Polaris (Inst), RedHill Biopharma (Inst), Syros Pharmaceuticals (Inst), Tesaro (Inst), and Xencor (Inst). A.K. has received grants from Astellas, Panavance, Ability, Novita, Cardiff, Lisata Therapeutics, and Boundless Bio outside the submitted work. No other disclosures were reported. E.L. has disclosed equipment for laboratory-based research 2018-present, Novocure, Ltd; honorarium for panel discussion organized by Antidote Education for a CME module on diagnostics and treatment of HER2+ gastric and colorectal cancers, funded by Daiichi-Sankyo, 2021 (honorarium donated to lab); compensation for scientific review of proposed printed content; Elsevier Publishing and Johns Hopkins Press; consultant, Nomocan Pharmaceuticals (no financial compensation); Institutional Principal Investigator for clinical trials sponsored by Celgene, Novocure, Intima Bioscience, Inc., the National Cancer Institute, and the University of Minnesota membership in the Caris Life Sciences Precision Oncology Alliance (no financial compensation). A.S. has received research grants (to institution) from Merck, AstraZeneca, Bristol Myers Squibb, Exelixis, Clovis, Biontech, Astellas, Dragonfly Therapeutics, Innovent biologics, KAHR medical, Amgen, Celgene, Actuate Therapeutics, Incyte corporation, Seagen, Daiichi Sankyo, and Advisory Board fees from Bristol Myers Squibb, AstraZeneca, Exelixis, Pfizer, and Daiichi Sankyo. J.X., Y.B., E.P., M.O., and G.S. are employees of Caris Life Sciences. The authors declare no other financial or non-financial interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41698-026-01329-w.