Impact of BRAF Mutation Class on Disease Characteristics and Clinical Outcomes in BRAF-mutant Lung Cancer

Non–small cell lung cancer (NSCLC) is a heterogeneous disease composed of an expanding number of biologically distinct and clinically relevant molecular subsets (1, 2). Identification of these unique molecular drivers and development of highly effective genotype-specific therapies have transformed the natural history of disease for select subgroups of NSCLC (1). However, there is variability in clinical characteristics and the magnitude of benefit from identical therapies among patients with NSCLCs that share a common oncogenic driver (3–5). For example, studies suggest that patients with atypical EGFR mutations are more likely to be ever-smokers and less likely to have durable responses to EGFR tyrosine kinase inhibitors than patients with exon 19 deletions and L858R mutations (5, 6). Appreciation of the diversity within molecular subsets and insight into the mechanisms that underlie this heterogeneity may facilitate development of more effective therapeutic strategies.

BRAF mutations, present in 2%–4% of NSCLC, are emerging therapeutic targets in NSCLC (7). In contrast to other malignancies that are predominantly driven by the BRAF V600E mutation, alternative mutations distributed throughout exons 11 and 15 collectively account for one-half of BRAF mutations in NSCLC (8). Yet, current understanding of BRAF-mutant NSCLC is derived from a handful of small studies that predominantly included patients with V600E mutations (9–13). Preclinical studies suggest that BRAF mutations are biologically heterogeneous and activate the RAF kinases to differing degrees (14–16). These studies have been instrumental in categorizing BRAF mutations into 3 classes, (i) V600 mutations that signal as monomers (class I), (ii) kinase-activating non-V600 mutations that function as dimers (class II), and (iii) kinase-impaired non-V600 mutations that “amplify” ERK signaling in the presence of activated upstream receptor tyrosine kinases or coalterations that increase RAS activity (class III; refs. 14, 15).

To date, clinical studies have classified BRAF-mutant NSCLC as either V600-mutant or non-V600–mutant (9–13). Although there is consensus among these studies that V600 mutations are more likely to arise in never-smokers than other BRAF mutations, there is disagreement about whether disease outcomes differ if mutations occur within or outside the V600 locus (11, 12). None of these studies was designed to evaluate whether differences exist between the 3 functional classes of BRAF-mutant NSCLC. Here, we performed a retrospective analysis to describe clinicopathologic characteristics and evaluate clinical outcomes of patients with NSCLC with class I–III BRAF mutations.

Patients with BRAF-mutant NSCLC were identified from NSCLCs that underwent molecular analysis at Massachusetts General Hospital (Boston, MA) or the Dana-Farber Cancer Institute (Boston, MA) between 2005 and 2017. The systematic genotyping of patients at these institutions commenced in 2005. This report is limited to patients who were found to have a BRAF mutation at initial genotyping. BRAF mutations were organized into functional classes on the basis of mutation classification in published articles (15, 16). Tumors that acquired a BRAF mutation after developing resistance to therapies targeting another molecular driver and those with BRAF variants (n = 33) that were not previously reported or functionally characterized were excluded from the analysis. All patients in this study provided written informed consent. Patient studies were conducted according to the Declaration of Helsinki, the Belmont Report, and the U.S. Common Rule.

Medical records were retrospectively reviewed to extract data on treatment histories and clinical, molecular, and pathologic characteristics. Data were updated as of November 30, 2017. Sites of disease at initial metastatic presentation were recorded after review of clinic notes and imaging reports. Progression-free survival (PFS) was measured from the time of chemotherapy initiation to investigator-assessed radiographic progression or death. PFS was otherwise censored on the date of last follow-up or date that chemotherapy was discontinued if patients did not experience progression or death. Overall survival (OS) was measured from diagnosis of metastatic disease to death. Patients without a known date of death were censored at last follow-up. The Institutional Review Board at each institution approved this study.

BRAF, NF1, and RAS mutations were identified using SNaPshot or DFCI Oncopanel as described previously (17, 18). The current iterations of both assays utilize next-generation sequencing, whereas earlier versions of SNaPshot relied on multiplex PCR. The current version of SNaPshot interrogates exons 11 and 15 of BRAF, exons 2–5 of KRAS and NRAS, and exons 1–58 of NF1. Oncopanel detects alterations involving all exons of BRAF, KRAS, NRAS, and NF1. All patients included in this study provided consent for molecular testing.

Fisher exact test was used to compare categorical characteristics between BRAF mutation classes, while age was analyzed by Wilcoxon rank-sum test. PFS and OS curves were estimated by the Kaplan–Meier method. Survival differences between molecular groups were expressed as the HR estimated by the Cox proportional hazards model with significance assessed using the score test. All P values were based on a 2-sided hypothesis. Statistical computations were performed using SAS 9.4 (SAS Institute).

Between April 2005 and October 2017, we identified 236 patients with NSCLC with BRAF mutations belonging to the following classes: class 1 (n = 107, 45%); class 2 (n = 75, 32%); and class 3 (n = 54, 23%). There were 25 unique BRAF variants. In our cohort, 202 (86%) of the BRAF mutations occurred at 1 of the following 4 residues: V600 (n = 107; 45%); G469 (n = 51; 22%); G466 (n = 25; 11%); and D594 (n = 19; 8%). The breakdown of BRAF mutations is depicted in Fig. 1 and Table 1. Clinicopathologic characteristics of these patients are presented in Table 2. Most patients in the 3 classes were white, current or former smokers with adenocarcinoma. Notably, squamous histology was more common in class II than class I tumors (P = 0.020). The proportion of never-smokers was similar when patients with class II and III mutations were compared (n = 2/75, 3% vs. n = 3/54, 6%, P = 0.649), but patients with class I tumors were more likely to be never-smokers (n = 23/107, 22%; P < 0.001 vs. class II; P = 0.011 vs. class III). There was no difference in the age or sex distribution across mutation classes.

A total of 139 (59%; class I, n = 69; class II, n = 38; and class III, n = 32) patients had metastatic disease at diagnosis. The remaining patients were diagnosed with stage I (n = 37; 16%), stage II (n = 17; 7%), or stage III (n = 43; 18%) NSCLC (Table 2). Across the 3 classes, a majority of patients had metastatic disease at presentation, likely reflecting our institutional practices of preferentially genotyping metastatic lung cancers. In a subset of patients presenting with metastatic NSCLC, metastases were confined to the thoracic cavity [M1a; n = 38/69 (55%) class I, n = 10/38 (26%) class II, and n = 14/32 (44%) class III; Table 2]. The difference in incidence of M1a disease at presentation was statistically significant when class I was compared with class II (P = 0.005), but was not statistically different for comparisons between class I and III (P = 0.392) and class II and III (P = 0.139). All but 1 patient included in this series underwent brain imaging at diagnosis. When only patients with metastatic disease at diagnosis were assessed, 9% (n = 6/69) of patients with class I mutations had brain metastases compared with 29% (n = 11/38) and 31% (n = 10/32) of those with class II and III mutations, respectively. The incidence of brain metastases was not statistically different between class II and III (P = 1.00) but was significantly lower for patients with class I mutations (P = 0.011 vs. class II; P = 0.007 vs. class III).

Preclinical studies suggest that class I and II BRAF mutants signal in a RAS-independent manner, whereas class III BRAF mutants rely on RAS activation to overcome negative feedback from ERK (15). To test the hypothesis that genetic alterations leading to RAS activation are more likely to occur in lung tumors with class III BRAF mutations, we analyzed our cohort to identify RAS and NF1 coalterations.

We detected genetic alterations in KRAS or NRAS in specimens from 23 BRAF-mutant patients. These alterations included activating KRAS mutations (n = 17), activating NRAS mutations (n = 5), and 1 case of high-level KRAS amplification (>25 copies) that was identified by NGS and confirmed with FISH. We did not identify any HRAS mutations. Our findings are illustrated in Fig. 2. Apart from one class I BRAF-mutant tumor with a cooccurring KRAS G12C mutation, RAS alterations only coexisted with class II and III mutations. Ten (13%) of 75 NSCLCs with class II mutations contained concurrent RAS alterations. RAS and BRAF alterations coexisted in 12 (22%) of 54 tumors with class III BRAF mutations, including 8 (42%) of 19 cancers with kinase-dead D594 mutations. Although the frequency of RAS coalterations was numerically higher in class III than class II tumors, this difference did not reach statistical significance (P = 0.237). Compared with class I mutations, class II and III mutations were more likely to occur alongside RAS alterations (class I vs. class II, P = 0.001; class I vs. class III, P < 0.001). Taken together, our findings demonstrate that RAS alterations may cooccur with class II and III BRAF mutations but are rarely seen in NSCLCs with class I BRAF mutations.

To evaluate whether the increased frequency of RAS coalterations in class II and III NSCLCs accounted for the higher rates of brain metastases in these groups, we compared the prevalence of brain metastases in patients with RAS comutated tumors to that of patients without coaltered tumors within each mutation class. We limited our analysis to patients presenting with metastatic disease. Among patients with class II tumors, brain metastases were detected in 1 (25%) of 4 patients with concurrent RAS and BRAF mutations compared with 10 (29%) of 34 tumors without concomitant RAS alterations (P = 1.000). Three (38%) of 8 patients with metastatic NSCLC harboring a class III BRAF mutation and a RAS coalteration had brain metastases compared with 7 (29%) of 24 patients with only class III mutations (P = 0.681). These data suggest that the higher rate of brain metastases in class II and III tumors is not associated with concurrent RAS mutations.

In total, we detected 11 NF1 genetic alterations in specimens from 10 patients. None of the NF1 mutations overlapped with RAS mutations. However, as NF1 mutations were not assessed in earlier versions of SNaPshot, we cannot definitively exclude coexistence of NF1 and RAS mutations in tumors genotyped using earlier versions of the assay. With the exception of a single class I tumor that harbored an NF1 variant of unknown significance (c.7189+3delA), all other NF1 alterations cooccurred with class II and III mutations. Five (7%) class II tumors had NF1 coalterations, including 2 NSCLCs with missense mutations (E2195G and N1619S), 1 tumor with a concurrent splice variant and missense mutation (G1890C), and 2 NSCLCs with truncating mutations (E1889* and K2160fs). We identified an NF1 rearrangement, 2 truncating NF1 mutations (R2258* and Q535Ter), and 1 missense NF1 mutation (A12D) in 4 class III tumors. The overlap between NF1 and BRAF mutations is depicted in Supplementary Fig. S1.

We next assessed whether observed differences in clinical and molecular features of class I–III BRAF mutants translated into different clinical outcomes. Because of the limited number of patients with stage I–III NSCLC, we limited our analysis to the 139 patients with metastatic NSCLC at diagnosis.

Treatment histories for patients with metastatic disease are summarized in Supplementary Table S1. Across functional classes, the combination of carboplatin and pemetrexed was the most widely used first-line treatment regimen [n = 29 (42%) class I, n = 22 (58%) class II, and n = 11 (34%) class III]. A handful of patients were treated with carboplatin/pemetrexed/bevacizumab (n = 5 class I, n = 1 class II, and n = 3 class III) or pemetrexed monotherapy (n = 2 class I and n = 3 class III). The combination of carboplatin and paclitaxel was the most common non-pemetrexed–based regimen [n = 6 (9%) class I, n = 3 (8%) class II, and n = 3 (9%) class III]. A minority of patients were treated with checkpoint inhibitors. Five (7%) class I patients received a checkpoint inhibitor, including 1 patient who received first-line pembrolizumab. Immunotherapy use was higher in classes II and III, with 14 (37%, 5 first-line) and 7 (22%, 2 first-line) patients receiving checkpoint inhibitors, respectively. In contrast, a greater number of class I patients were treated with targeted therapies. Specifically, 6 class II or III patients (9%) received MAPK-directed therapy compared with 34 (49%) class I patients.

As carboplatin/pemetrexed was the most frequently used regimen, we limited our PFS analysis to patients who received this regimen as first-line therapy. Of 62 patients who were given this regimen, 59 had adequate follow-up to determine PFS. Median PFS was 6.2 months for 28 patients with a class I BRAF mutation. Median PFS was 3.3 months and 4.9 months, respectively, for 20 patients with class II and 11 patients with class III mutations. The difference between class II and III was not statistically significant (P = 0.452). There was a trend toward shorter PFS when class II was compared with class I [HR 1.80; 95% confidence interval (CI), 0.95–3.42; P = 0.069]. Patients with class III tumors had statistically shorter PFS on chemotherapy than their class I counterparts (HR 2.31; 95% CI, 1.04–5.15; P = 0.034). PFS by mutation class is depicted in Fig. 3A.

To determine whether PFS differences arose from disproportionate distribution of patients with M1a-only disease across functional classes, we repeated the analysis after excluding 23 patients (n = 14 class I, n = 5 class II, and n = 4 class III) with M1a-only disease. This analysis yielded a median PFS of 5.1 months for class I vs. 1.4 months for class II and 4.9 months for class III, again demonstrating that PFS favored class I over class II (class I vs. II, P = 0.031). The PFS difference between class I and III was no longer statistically significant (class I vs. III, P = 0.186). These findings support the notion that PFS differences were not exclusively due to a more favorable pattern of metastasis.

On the basis of our observation that patients with class II and III BRAF mutations may have shorter PFS on first-line chemotherapy, we hypothesized that patients with class II and III mutations would have shorter OS after diagnosis. With a median follow-up of 54.0 months, median OS was 40.1 months (95% CI, 17.5–56.1 months) for 69 patients with class I mutations. Median OS was 13.9 months (95% CI, 3.7–18.7 months) and 15.6 months (95% CI, 8.9–37.4 months) for 38 patients harboring class II mutations and 32 patients with class III mutations at a median follow-up of 32.7 and 24.6 months, respectively. OS was not statistically different when class II and class III were compared (HR = 0.75; 95% CI, 0.40–1.42; P = 0.375). Compared with class I, patients with class II and III NSCLC had shorter OS (class II vs. I, HR 2.50; 95% CI, 1.44–4.32; P < 0.001; and class III vs. I, HR 1.97; 95% CI, 1.09–3.56; P = 0.023). OS by functional class is depicted in Fig. 3B. To determine whether shorter OS was the result of concurrent RAS alterations, we independently evaluated the impact of RAS coalterations on survival of a pooled cohort of patients with NSCLC with class II and III mutations. The OS for the group of 12 patients with RAS comutations was nearly identical to that of 58 patients with class II and III mutations without RAS mutations (median 15.6 vs. 15.7 months; HR = 1.0; 95% CI, 0.4–2.3; P = 0.988).

Because targeted therapies have the potential to alter OS, we conducted a focused OS analysis for the subgroup of patients who had not received MAPK-directed therapy (Fig. 3C). OS was still lower for 35 patients with class II tumors than 35 patients with class I tumors (median 39.1 vs. 12.5 months; HR 2.24; 95% CI, 1.11–4.5; P = 0.021), whereas the difference between class I and III was no longer significant (median OS for 29 class III patients, 18.3 months; P = 0.187). In addition, we performed a separate analysis of patients with M1b disease who had not received targeted therapies to explore the possibility that the survival gap might result from the combination of exposure to targeted therapies and different patterns of metastases across functional classes. When we limited our analysis to this subgroup of patients, there was no difference in OS across the 3 classes (median class I, 9.4 months; class II, 7.9 months; and class III, 9.7 months; Fig. 3D).

BRAF mutations are among the newest molecular targets in NSCLC. Unlike other BRAF-driven malignancies, recurrent mutations outside of the V600 locus account for at least one-half of BRAF mutations in NSCLC (7). However, precision medicine efforts to date have largely focused on the V600 subset where repurposing melanoma drugs has led to significant and durable responses (19, 20). Preclinical studies suggest that classifying BRAF mutants by underlying kinase activity and RAF signaling mechanism may be more informative than grouping tumors according to V600 status (14, 16, 21). These preclinical observations are supported by case reports and small series demonstrating distinct activity of available BRAF inhibitors in lung cancers with class I, II, or III mutations (19, 22, 23). Here, we present the largest clinical cohort of patients with BRAF-mutant NSCLC to date. Our findings suggest that class II and III NSCLCs may have more aggressive clinical characteristics than class I NSCLCs.

In our series, differences between mutation classes were apparent from the time of diagnosis. As seen in smaller studies, approximately 20% of patients with class I mutations in our study were never-smokers and virtually all patients with class II or III mutations were current or former smokers (11, 12, 19, 24). Among those presenting with metastatic disease, patients with class I mutations had a 3-fold lower incidence of brain metastases. The propensity to develop brain metastases among patients with class II and III mutations has potential implications for drug development. Specifically, although the combination of dabrafenib and trametinib appears to be effective against some of these mutations in preclinical models (16, 21), the limited blood–brain barrier penetration of trametinib and the lack of sensitivity of most class II and III mutations to dabrafenib monotherapy suggest that alternate drugs may be necessary to produce durable intracranial responses (21). Notably, in a recent large retrospective study of BRAF-mutant colorectal cancer, non-V600 mutations were associated with lower-grade histology and improved survival compared with V600 mutations (25). In contrast, a single institution study suggests that clinical outcomes may be similar for melanoma patients with V600 and non-V600 BRAF mutations (26). Considering the above, it is possible that the impact of non-V600 mutations on disease characteristics and clinical outcomes may differ across tumor types.

In addition to having more aggressive disease at diagnosis, patients with class II and III mutations may have a less favorable clinical course than their class I counterparts. For example, we found that patients with class II and III mutations progressed earlier when treated with first-line chemotherapy (Fig. 3A). Of note, existing studies exploring the impact of BRAF mutations on chemotherapy outcomes have yielded inconsistent results, with some studies demonstrating longer PFS in the non-V600 group and others reporting the opposite (11, 12). However, interpretation of the conclusions from these studies is affected by suboptimal study design, specifically small numbers of patients and failure to distinguish between distinct chemotherapy regimens.

Considering the difference in outcomes to first-line treatment observed in our study and the fact that effective targeted therapies are approved only for class I BRAF mutations, we hypothesized that life expectancy from diagnosis might be shorter for patients with class II and III mutations. Interestingly, the disparity in OS (i.e., median OS of 3 years for class I vs. 1 year for class II) was largely maintained for the comparison between class I and II after we excluded patients who had received targeted therapy, suggesting that intrinsic rather than extrinsic variables might be accounting for the difference. Indeed, when we restricted the analysis to patients who had extrathoracic metastases (i.e., M1b disease) who had not received targeted therapies, OS was similar for all functional classes. Thus, the more favorable outcomes observed for class I may reflect the greater proportion of patients with thoracic-only metastases and presumably more indolent biological behavior. The improved OS of class I patients may be further magnified by the availability of effective targeted therapies.

In this study, we did not observe survival differences between NSCLCs with class II and III mutations treated with nontargeted therapies. However, it is possible that differences may ultimately emerge when patients are exposed to effective genotype-directed therapies. For example, preclinical findings suggest that reliance on upregulation of receptor tyrosine kinases may increase sensitivity of some class III NSCLCs to monotherapy with tyrosine kinase inhibitors targeting the activated upstream kinase (15, 27). In addition, a recent study using melanoma cell lines observed differences in sensitivity of distinct class II mutants to inhibition with approved BRAF and MEK inhibitors, with some mutants (L597 and K601) surprisingly displaying sensitivity to monotherapy with BRAF inhibitors, whereas others (G464 and G469) were generally resistant to BRAF and/or MEK inhibition (21). Notably, L597 and K601 are located in the active segment while G464 and G469 arise in the glycine-rich p-loop. These findings raise the possibility that subclasses exist within each functional class and suggest that subclass differences may lead to response heterogeneity.

We anticipate that cooccurrence of RAS alterations will also impact response to targeted therapies. Interestingly, we detected RAS or NF1 mutations in approximately one-half of tumors with kinase-dead BRAF mutations; this finding supports the notion that class III mutants rely on RAS activation and suggests that effectively suppressing RAS signaling is essential for treating class III tumors. Although present at a lower frequency, we observed RAS coalterations in 13% of class II tumors, suggesting that a subgroup of class II tumors may be particularly addicted to MAPK signaling. The clinical and biological implications of these double mutations are worthy of further exploration as pathway hyperactivation and RAS coactivation may impact sensitivity to therapy (28, 29).

There are several limitations of this study. In addition to those described above, limitations include the retrospective nature of the analysis, the relatively small number of class II and III patients who were assessed for outcomes, the limited number of patients included in subgroup analyses, and employment of multiple versions of the genotyping assays over time. A significant subset of patients in our study had metastases confined to the thoracic cavity. While this distribution of metastatic sites may be representative of BRAF-mutant NSCLC, it is also possible that our cohort was enriched for more favorable characteristics than is typical for this molecular group. Larger clinical studies are necessary to confirm our conclusions about clinicopathologic features and disease outcomes.

In summary, we have performed the largest clinical analysis of BRAF-mutant NSCLC to date. Our findings demonstrate that BRAF-mutant NSCLC is not a single disease, but rather a disease entity that encompasses 3 functional classes. Compared with class I, classes II and III have distinct clinical and molecular features including poorer clinical outcomes. The class-specific pathogenic mechanisms and disease characteristics highlight the importance of developing novel therapeutic strategies that can effectively target each BRAF functional class and ultimately improve survival.

I. Dagogo-Jack has served as compensated consultant or received honoraria from Boehringer Ingelheim and Foundation Medicine. P. Martinez has received grant support from the Sociedad Española de Oncologia Medica (SEOM) and is currently working for Medimmune (Frederick, MD, USA). A.J. Iafrate holds equity in ArcherDx and has served as a consultant for Roche, Pfizer, Chugai, DebioPharm, and Constellation. B.E. Johnson has research grants from Novartis and Toshiba. A.T. Shaw has served as a compensated consultant or received honoraria from Pfizer, Novartis, Genentech/Roche, Ariad/Takeda, Igntya, Loxo, Blueprint Medicines, KSQ Therapeutics, Daiichi Sankyo, EMD Serono, Taiho Pharmaceutical, TP Therapeutics, Foundation Medicine, and Natera. M.M. Awad has served as a compensated consultant or received honoraria from Abbvie, Ariad, AstraZeneca, MedImmune, Boehringer Ingelheim, Bristol-Myers Squibb, Clovis Oncology, Foundation Medicine, Genentech, Merck, Nektar, Novartis, Pfizer, and Syndax. No potential conflicts of interest were disclosed by the other authors.

Conception and design: I. Dagogo-Jack, P. Martinez, B.E. Johnson, A.T. Shaw, M.M. Awad

Development of methodology: I. Dagogo-Jack, P. Martinez, T. Nguyen, B.E. Johnson, J.K. Lennerz, A.T. Shaw, M.M. Awad

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I. Dagogo-Jack, P. Martinez, L.A. Ferris, T. Nguyen, A.J. Iafrate, J.K. Lennerz, A.T. Shaw, M.M. Awad

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I. Dagogo-Jack, B.Y. Yeap, A.J. Iafrate, B.E. Johnson, J.K. Lennerz, A.T. Shaw, M.M. Awad

Writing, review, and/or revision of the manuscript: I. Dagogo-Jack, P. Martinez, B.Y. Yeap, L.A. Ferris, A.J. Iafrate, B.E. Johnson, J.K. Lennerz, A.T. Shaw, M.M. Awad

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Ambrogio, L.A. Ferris, C. Lydon, T. Nguyen, N.A. Jessop, J.K. Lennerz, A.T. Shaw

Study supervision: A.T. Shaw

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.