The limited knowledge on the molecular profile of patients with BRAF-mutant non–small cell lung cancer (NSCLC) who progress under BRAF-targeted therapies (BRAF-TT) has hampered the development of subsequent therapeutic strategies for these patients. Here, we evaluated the clinical utility of circulating tumor DNA (ctDNA)-targeted sequencing to identify canonical BRAF mutations and genomic alterations potentially related to resistance to BRAF-TT, in a large cohort of patients with BRAF-mutant NSCLC.
This was a prospective study of 78 patients with advanced BRAF-mutant NSCLC, enrolled in 27 centers across France. Blood samples (n = 208) were collected from BRAF-TT–naïve patients (n = 47), patients nonprogressive under treatment (n = 115), or patients at disease progression (PD) to BRAF-TT (24/46 on BRAF monotherapy and 22/46 on BRAF/MEK combination therapy). ctDNA sequencing was performed using InVisionFirst-Lung. In silico structural modeling was used to predict the potential functional effect of the alterations found in ctDNA.
BRAFV600E ctDNA was detected in 74% of BRAF-TT–naïve patients, where alterations in genes related with the MAPK and PI3K pathways, signal transducers, and protein kinases were identified in 29% of the samples. ctDNA positivity at the first radiographic evaluation under treatment, as well as BRAF-mutant ctDNA positivity at PD were associated with poor survival. Potential drivers of resistance to either BRAF-TT monotherapy or BRAF/MEK combination were identified in 46% of patients and these included activating mutations in effectors of the MAPK and PI3K pathways, as well as alterations in U2AF1, IDH1, and CTNNB1.
ctDNA sequencing is clinically relevant for the detection of BRAF-activating mutations and the identification of alterations potentially related to resistance to BRAF-TT in BRAF-mutant NSCLC.
This study shows that targeted circulating tumor DNA (ctDNA) sequencing is clinically relevant for the identification of BRAF-activating mutations at diagnosis, and potential mechanisms of primary and acquired resistance to BRAF-targeted therapies (BRAF-TT) in patients with BRAF-mutant non–small cell lung cancer (NSCLC). We evidenced that detection of ctDNA at the first radiological examination was associated with poor survival. At disease progression, we detected alterations in key effectors of the MAPK or PI3K pathways and in signal transducers, which are potentially related to resistance to BRAF-TTs. Our data thus support the notion that reactivation of the MAPK pathway remains an important resistance mechanism to BRAF-targeted monotherapy or BRAF/MEK combination therapy in these patients. Our findings might be of broad interest to basic cancer scientists and medical oncologists alike, as they provide insights on the biology of BRAF-mutant NSCLC and may have implications for the development of subsequent therapeutic strategies for these patients.
BRAF-activating mutations occur in 2%–3% of patients with non–small cell lung cancer (NSCLC; ref. 1) and result in constitutive kinase activation, triggering downstream pathways regulating cancer cell proliferation and survival (2). In these patients, BRAF inhibitor monotherapy with vemurafenib or dabrafenib is associated with response rates of 42% (3, 4) and 33% (5), respectively, rising to 64% for the combination treatment with dabrafenib and trametinib (6). On the basis of these results, the FDA and the European Medicines Agency granted approval to the combination regimen of dabrafenib plus trametinib in this clinical setting. Despite these encouraging results, most responses are temporary. Almost all patients ultimately develop resistance to therapy within several months (3, 6). Currently, the limited knowledge of the molecular profile of patients with NSCLC who progress under BRAF-targeted therapies (BRAF-TT) has limited the development of subsequent targeted therapeutic strategies.
Sensitive methods for detecting cancer mutations in plasma have been used as an alternative to predict relapse in patients with early-stage lung cancer following surgery (7) and to monitor longitudinal tumor dynamics over time or to detect potential mechanisms of resistance to therapy in patients with metastatic lung cancer (8, 9).
Here, the clinical relevance of targeted circulating tumor DNA (ctDNA) sequencing to detect BRAF mutations was assessed with two aims: (i) to monitor response to treatment with BRAF inhibitors (BRAFi) in patients with NSCLC and (ii) to identify potential molecular mechanisms of resistance to BRAF-TT.
Materials and Methods
Eligible patients were ≥18 years old, with histologic/cytologic confirmation of advanced NSCLC, and harboring a BRAF mutation as determined by a validated test on tumor tissue. Patients were prospectively enrolled between 2014 and 2018 within the framework of the following studies: AcSé vemurafenib (NCT02304809), Liquid Biopsies for Lung from the Leon Bérard Cancer Center (LIBIL, NCT02511288), Gustave Roussy Institute (CEC-CTC Study, 2008-A00585–50), and Toulouse University Hospital (DC-2011-1382, AC-2013-1984, no. CNIL 1727608). These studies were conducted in accordance with the principles of the Declaration of Helsinki. All patients provided written informed consent for biomedical research and the institutional ethics committees approved the protocol.
Sample collection and ctDNA analysis
A total of 208 prospective samples were collected at diagnosis before starting BRAF-TT, under systemic therapy (e.g., BRAF-TT, chemotherapy, immunotherapy, etc.) at each disease evaluation (every 6–8 weeks) or at disease progression (PD). Blood samples (n = 208) were collected from BRAF-TT–naïve patients (n = 47), patients nonprogressive under treatment (n = 115), or patients at PD to BRAF-TT (24/46 on BRAF monotherapy and 22/46 on BRAF/MEK combination therapy). Plasma was isolated and ctDNA analysis was centralized using InVisionFirst-Lung (Inivata at Cambridge, United Kingdom and Research Triangle Park, NC), which identifies single-nucleotide variants (SNV), insertions and deletions, copy-number variations, and fusions, with whole gene and gene hotspots across a 36-gene panel (Supplementary Fig. S1; ref. 10; Supplementary Materials and Methods). The InVisionFirst-Lung Assay was shown to have an excellent limit of detection in analytic validation studies (10, 11), with 99.48% sensitivity for SNVs present at variant allele frequency (VAF) range 0.25%–0.33% and 56.25% sensitivity at VAF range 0.06%–0.08% while retaining high specificity (99.9997% per base). The ability to analyze more DNA molecules using an amplicon-based approach, coupled with high sequencing depths achieved with a focused lung cancer panel allows improved sensitivity over hybridization capture–based approaches, although with the limitation that panel size is reduced. Positivity for plasma-mutant ctDNA was defined by the detection of a mutant allelic fraction (AF) in replicates >0.01% in any gene and greater than a base-specific background model. BRAF-mutant ctDNA clearance was defined as undetectable levels of BRAF mutation in ctDNA within 100 days of treatment initiation.
Response assessment was performed as per RECIST v1.1 in the cohort of patients enrolled in the AcSé Vemurafenib Study, according to the protocol (4). In the population receiving BRAF-TT in clinical routine, the radiological assessment was mainly performed as per RECIST v1.1 (in 84% of cases), and in 16% of cases as per investigator's discretion.
Overall survival (OS) was calculated from the date of first administration of systemic therapy until death due to any cause. Progression-free survival (PFS) was calculated from the date of first administration of each systemic therapy until PD or death due to any cause. PD was assessed as per RESIST for most cases and as per investigator's criteria for 5 patients. Survival curves were estimated by the Kaplan–Meier method and were compared by the log-rank test. All P values were two-sided and values <0.05 were considered statistically significant. Data were processed and analyzed by using SAS version 9.4 (SAS Institute Inc.). The number of all included patients and recorded variables were reported using descriptive statistics.
The relationship between clinical characteristics and response was calculated using Fisher exact test. The association of BRAF-mutant ctDNA detection with response to treatment during follow-up was assessed using an unpaired t test with Welch correction. Finally, Wilcoxon matched-pairs signed rank test was used to assess the change in the AF of BRAF-mutant ctDNA between the plasma samples collected before and at PD.
Characteristics of the study population
Seventy-eight patients with advanced NSCLC and positive for an activating BRAF mutation, based on tissue biopsy, were included in this study. A summary of patients' characteristics is presented in Supplementary Table S1. Molecular testing of tumor tissue identified a BRAFV600E mutation in 72 (92%) patients and non-V600E alterations in 6 (8%) patients. The latter included alterations in the P-loop domain or the activating segment of the BRAF kinase domain, namely G469A, G466V, N581S, G596R, D594M, and K601E.
First-line treatment consisted of platinum-based chemotherapy in 47 of 78 (60%) patients. Thirty-six (46%) patients were treated with vemurafenib and 26 (33%) patients received dabrafenib in combination with trametinib. The remaining 21% patients received other regimens with or without targeted therapy (Supplementary Table S2; Supplementary Data).
Forty-seven blood samples were successfully collected from 44 BRAF-TT–naïve patients, 46 samples from 35 patients at PD to BRAF-TTs, and 115 additional samples during longitudinal follow-up under treatment (47 samples from patients on BRAF inhibitor monotherapy and 68 samples from patients on combination therapy), concomitant to routine radiological evaluation (Supplementary Fig. S2). The median number of plasma samples analyzed per patient was three (range 1–8).
Utility of ctDNA analysis to identify BRAF mutations in targeted therapy–naïve patients with NSCLC
Detection of mutant ctDNA (i.e., detection of a mutant AF in replicates >0.01% in any gene) was seen in 34 of 47 (72%) patient samples, before the initiation of targeted therapy. Circulating BRAFV600E was detected in 31 of 42 (74%) samples, whereas non-V600E mutations were detected in three of five samples (Fig. 1A). Overall, patients with detectable BRAF-mutant ctDNA presented mainly with systemic disease and more than two metastatic sites (Supplementary Fig. S3; Supplementary Data). We identified a median of two nonsynonymous genomic alterations per patient, within the targeted regions of the gene panel.
Patients with detectable circulating BRAF ctDNA at baseline showed a median PFS of 5.2 months, while patients with undetectable BRAF-mutant levels in plasma had a median PFS of 7.8 months, not reaching statistical significance (Fig. 1B). In-line with these observations, the median OS was 12.1 months in cases with detectable BRAF ctDNA and it was not reached in cases with undetectable BRAF mutation in plasma, although not reaching statistical significance (Fig. 1B). We observed similar results with regards to the detection of total ctDNA in plasma (Fig. 1C).
Cooccurring ctDNA alterations in targeted therapy–naïve BRAF-mutant lung cancers
Targeted therapy–naïve BRAF-mutant samples harbored 0–4 genetic alterations in addition to BRAF, including TP53 in 23 of 47 (49%), STK11 in 6 of 47 (15%), and CDKN2A in 2 of 47 (4%; Fig. 1A). Other genomic alterations that cooccurred with canonical BRAF driver mutations were detected in genes related to the PI3K–protein kinase pathway (10%) or the MAPK pathway (6%), or in genes coding for protein kinases (6%) or signal transducers (8%; Fig. 1A). In BRAFV600 plasma-positive cases, these alterations included KRASG12C (1 patient), PIK3CAE545K (1 patient), PIK3CAH1047R (1 patient), AKTE71K (1 patient), FGFR2A553D (1 patient), ERBB2 amplification (1 patient), IDH1I130T (1 patient), IDH1R132C (2 patients), CTNNB1G34V (1 patient), and UA2F1Q157P (1 patient). Of note, the frequency of concurrent mutations found in BRAFV600E-mutant patients at baseline was consistent with that reported in cBioPortal (Supplementary Fig. S5; Supplementary Data). Mutations in KRASG12C and PTENR14K (P79) or in NTRK3P612T (P45) were observed in 2 BRAFV600E tissue–positive patients who had undetectable levels of plasma BRAFV600E (Fig. 1A). These 2 patients presented with single-site extracranial metastasis.
Interestingly, molecular analysis of the tumor biopsy at baseline evidenced that the diagnostic tumor biopsy collected from P79 1 week before the liquid biopsy was wild-type for KRAS (Supplementary Table S6). Moreover, in P68, a tumor biopsy collected concomitantly with plasma confirmed the presence of BRAFV600E and TP53H168N mutations. This patient presented an isolated progression in the bone, which could explain why the BRAFV600E was not detected in plasma. Finally, the PIK3CAH1047R mutation detected in the baseline plasma sample from P57 was also reported in a tissue biopsy collected at PD on first-line chemotherapy (Supplementary Table S6).
The concurrent genetic alterations found in KRAS, PIK3CA, AKT1, ERBB2, and IDH1R132C were reported to be oncogenic, while CTNNB1G34V and UA2F1Q157P were described as likely oncogenic (OncoKB; ref. 12; Supplementary Table S3). To predict the potential functional effect of the mutations with unknown oncogenic potential, in silico structural modeling was used (Materials and Methods). This analysis revealed that FGFR2A553D is potentially activating, PTENR14K and IDH1I130T occur in key regulatory regions in the target proteins, while NTRK3P612T is predicted to have no influence in protein stability (Fig. 1D and E; Supplementary Table S3; Supplementary Fig. S4A and S4B; Supplementary Data).
The two cases with concurrent activating PIK3CA and BRAFV600E mutations presented stable disease (SD) for 6.4 months and PD after 2.1 months on vemurafenib treatment, respectively (Fig. 1A). Two other patients who carried an FGFR2A553D variant (P18) and ERBB2 amplification (P47), in addition to BRAFV600E, responded to vemurafenib for 3.3 and 3.7 months, respectively. The median time on BRAF-TT for these 4 patients was 3.5 months (range, 2.1–6.4).
In the subset of patients treated with the combination of dabrafenib and trametinib, those with baseline alterations in AKTE71K (P48) or NTRK3P612T (P45) exhibited SD as best response. The concomitant presence of KRASG12C and PTENR14K (P79) was associated with rapid progression to treatment (0.6 months). Conversely, patients presenting with IDH1R132C (P68) and U2AF1Q157P (P64) variants were treated for longer periods of time (16.5 and 13.3 months, respectively).
The prevalence of metastatic sites compared with BRAF ctDNA status (Supplementary Fig. S3), the mutations detected in tumor biopsies at diagnosis for the patients who harbored a comutation in ctDNA (Supplementary Table S6), as well as detailed information on the concurrent mutations found in non-V600E cases are shown and discussed in the Supplementary Data.
Overall, these results suggest that oncogenic mutations in PIK3CA are associated with poor responses to vemurafenib, while the presence of activating KRAS alterations and a likely oncogenic alteration in PTEN, at baseline, result in rapid progression to combination therapy.
Early plasma dynamics of BRAF mutations as a predictor of response to BRAF-TTs
A total of 19 patients had matched baseline and first evaluation plasma samples with evaluable response data (median time to first evaluation, 8 weeks). Complete clearance of BRAFV600E ctDNA was observed at the first radiological evaluation in 66% (4/6) and 62% (8/13) of patients treated with vemurafenib or dabrafenib and trametinib, respectively (Fig. 2A, left). In the remaining cases (7/19), the change in the AF of circulating BRAFV600E at first evaluation ranged from a decrease of 97% to an increase of 1,600%. Patients who presented partial response (PR) exhibited a greater BRAF-mutant ctDNA decrease (i.e., clearance) when compared with patients who presented SD under BRAF-TT (Fig. 2A, right; unpaired t test with Welch correction; P = 0.040). Furthermore, the AF of BRAFV600E in ctDNA assessed concomitantly with each radiographic evaluation was significantly lower in patients with PR or SD than in patients with PD, as determined in 161 serial longitudinal plasma samples collected from patients with BRAF-mutant NSCLC treated with BRAF-TTs (Fig. 2B). Patient 33 exemplifies this observation (Supplementary Fig. S6; Supplementary Data).
Detection of mutant ctDNA (i.e., detection of a mutant AF > 0.01% in any gene) at the first radiographic evaluation was associated with worse PFS (log-rank, P = 0.023; Fig. 2C, left) and worse OS (log-rank, P = 0.0004; Fig. 2C, right). Similarly, patients with detectable BRAF-mutant ctDNA at first follow-up showed shorter OS (log-rank, P = 0.021; Supplementary Fig. S7A) and a median PFS of 5.7 months, while patients with undetectable BRAF-mutant levels in plasma had a median PFS of 8.7 months, not reaching statistical significance (Supplementary Fig. S7B).
Among the 8 patients with a detectable potential driver of resistance at PD (i.e., a genomic alteration found in ctDNA at PD, excluding mutations in TP53, CDKN2A, or STK11) and a plasma sample collected before PD, serial longitudinal follow-up allowed the early detection of molecular PD in 3 patients. Mutant ctDNA detection preceded confirmation by radiological evaluation, with a median lead time of 57 days (range, 52–63 days), as evidenced by the early emergence of NRASQ61R, MAPKC121S, and GNASR201C activating mutations in plasma (Fig. 2D).
ctDNA profiles from patients with BRAF-mutant NSCLC at PD on BRAF-TT
Genomic ctDNA sequencing was performed on 46 plasma samples collected from 35 patients at PD on BRAFi treatment (24 samples) or on the combination of dabrafenib and trametinib treatment (22 samples). Repeated plasma samples were collected at PD to BRAF-TTs for patients P40, P55, P57, P58, P64, P65, and P69.
BRAF-activating mutations were identified at PD in 35% (16/46) of plasma samples (Fig. 3A). A consistent rebound in BRAFV600E was observed in 17 of 27 (63%) patients at PD, for whom a plasma sample was collected prior to the clinical assessment of PD [median, 81 days (42–265); Fig. 3B].
Potential drivers of resistance to BRAF-TTs (i.e., a genomic alteration found in ctDNA at PD, excluding mutations in TP53, CDKN2A, or STK11) were detected in ctDNA in 16 of 35 (46%) patients. These alterations mainly involved effectors of the MAPK pathway (e.g., KRAS, NRAS, MAP2K1, GNAS, or GNA11), and occurred in 19% of samples (Fig. 3A; Supplementary Fig. S10A; Supplementary Table S4). Oncogenic mutations in KRAS (G12V/C and Q61R) were observed in patients progressing on either vemurafenib (P17) or the combination therapy (P57 and P68). Other alterations included GNASR201C, GNASR201H, GNA11R214S, and MAP2K1C121S, which were observed at PD on BRAFi monotherapy, and NRASQ61R, which was detected at PD on the combination therapy (Fig. 3A). Notably, activating mutations in KRAS, NRAS, MAP2K1, and GNAS were not detected in plasma before the initiation of BRAF-TT (P17, P35, P57, and P68; Fig. 3A), within the detection cutoff of the assay (see Materials and Methods). The preexistence of the GNA11R214S alteration (P26) could not be established as a baseline plasma sample was not available for this patient.
Of note, KRASG12V and NRASQ61R (P57) as well as KRASQ61R and KRASG12V (P68) were not detected by targeted DNA sequencing in the tissue sample used for molecular diagnosis (Supplementary Table S6), suggesting that these alterations were most probably acquired during treatment. Because KRASG12C (P17), GNA11R214S (P26), and MAP2K1C121S (P35) were not tested at molecular diagnosis, their preexistence before BRAF-TT cannot be assessed.
Alterations in the PI3K pathway were observed at PD on either vemurafenib or the combination treatment (P1, P3, P55, and P57; Fig. 3A; Supplementary Fig. S10A; Supplementary Table S4). In 2 of these patients, these alterations were observed before the initiation of BRAF-TT (Fig. 1A; Supplementary Table S6). Finally, oncogenic or likely oncogenic alterations (OncoKB; ref. 12) in U2AF1 (P64 and P68), IDH1 (P27 and P68), or CTNNB1 (P58 and P59) were detected at PD on either monotherapy or combination therapy (Fig. 3A; Supplementary Fig. S10A; Supplementary Table S4). Heterogeneous potential mechanisms of resistance to BRAF-TTs were identified in 2 patients (Supplementary Fig. S10).
In silico structural modeling showed that PPP2R1AD265G, evidenced in P55 at PD on dabrafenib and trametinib, destabilizes the fold of the protein and potentially downregulates its activity. Similarly, U2AF1R156H (P68) was predicted to disrupt the binding of the protein to RNA, while NFE2L231–32:GV/X (P35) impacted the interaction with KEAP1 (Fig. 3C; Supplementary Fig. S11; Supplementary Data).
The median time on treatment for patients with KRAS mutations detected at PD was 2.7 months (0.9–16.5), 13 months (2.7–50) for cases with other alterations in the MAPK pathway, and 4.3 months (1.9–65) for patients with PI3K pathway mutations. A median time of treatment of 8 months (4.8–15.8) was observed in 3 patients who presented mutations in IDH1 or CTNNB1, in absence of other potential drivers of resistance.
A cosubmitted work on the genomic analysis of tumor rebiopsy samples at PD, in a subgroup of these patients (excluding P53), was recently published (13). These data are presented in light of our results in Supplementary Table S7. P53, P55, and P65 presented localized progression, which could explain the undetectable levels of ctDNA in plasma. Some alterations were found in ctDNA, but were not detected in the tumor rebiopsy, in patients with systemic disease (e.g., P58 and P68), potentially suggesting the existence of heterogeneous mechanisms of resistance. Of note, in P59, CTNNB1S37C was detected in ctDNA and in the rebiopsy at PD, but also in tumor biopsy prior to dabrafenib treatment.
Overall, the presence of BRAF ctDNA at PD was associated with a reduced OS compared with BRAF ctDNA–negative patients (log-rank, P = 0.01; Fig. 3D). Similar results were observed when the detection of total ctDNA at PD was analyzed (log rank, P = 0.02; Supplementary Fig. S9). The median OS for patients who presented concurrent mutations at PD (besides TP53, STK11, or CDKN2A) was 17.9 months, while for patients with no such alterations detected in ctDNA, the median OS was not reached (Fig. 3E).
Herein we report the feasibility and clinical relevance of targeted genomic profiling of ctDNA to detect canonical BRAF mutations at the time of diagnosis and relapse, to evidence genomic alterations potentially associated with resistance to therapy, and to monitor disease burden under treatment, in a large prospective cohort of 78 patients with BRAF-positive NSCLC.
Detection of BRAFV600E in ctDNA was reported previously in 5 patients with NSCLC, using droplet-digital PCR (14). Our findings highlight the feasibility of targeted ctDNA sequencing to detect BRAF-activating mutations in 34 of 47 (72%) samples from patients with BRAF-driven NSCLC, before the initiation of targeted therapy. In-line with previous reports in patients with other oncogene-driven NSCLC (11, 15), our results suggest that the presence of detectable BRAF mutations in plasma is correlated with tumor burden and the occurrence of tumor lesions in specific metastatic sites.
ctDNA dynamics in the early stages of treatment can be used to predict treatment outcome before tumor response is assessed by conventional imaging (16, 17). In this study, circulating BRAF-mutant clones showed an initial marked decrease in abundance, at the first radiological evaluation, in 63% (12/19) of the evaluable patients treated with BRAF-TT. Mutant ctDNA clearance was associated with longer PFS and OS relative to patients with detectable levels of early mutant ctDNA. These results indicate that early measurement of ctDNA may provide an adequate surrogate of the initial response to treatment and that serial monitoring of BRAF ctDNA levels during treatment might be a clinically useful marker of tumor response, similar to what has been observed in patients with EGFR-mutant NSCLC (18). Further studies are needed to assess the potential of liquid biopsies to enable real-time monitoring of disease during therapy to allow the identification and characterization of minimal residual disease (19, 20) in BRAF-mutant NSCLC.
Cooccurring genomic alterations detected in treatment-naïve patients constitute major determinants of both tumor cell intrinsic and noncell autonomous response to therapy, as reported in EGFR-mutant (21) or in ALK-translocated lung cancers (22). We provide evidence that cooccurring genetic alterations commonly exist in treatment-naïve BRAF-mutant NSCLC, and include oncogenic or likely activating alterations in the MAPK and PI3K pathways, and in FGFR2, ERBB2, U2AF1, IDH1, and CTNNB1. Some of these alterations might constitute a determinant of primary resistance to BRAF-TT. Interestingly, cooccurrence of BRAFV600E with alterations in FGFR2, ERBB2, CTNNB1, or AKT1 is not reported in The Cancer Genome Atlas (www.cbioportal.org), but was identified in 4 patients in this cohort, likely highlighting the increased heterogeneity of metastatic BRAF-mutant NSCLC.
Although BRAF-TT may provide tumor control in patients with BRAF-mutant advanced NSCLC (3, 6, 23, 24), acquired resistance ultimately develops through mechanisms that are largely unknown. A detailed understanding of the molecular alterations associated with resistance to BRAF-TT may help inform future therapeutic strategies in subsequent treatment lines. Genomic ctDNA profiling in BRAF-mutant NSCLC provided insights into the potential mechanisms of acquired resistance to BRAF-TT in 16 of 35 (46%) patients. These alterations were related to the MAPK pathway (n = 9, 19%), PI3K pathway (n = 5, 11%), and signal transducers (n = 10, 21%). MAPK pathway reactivation constituted a key pathway of acquired resistance to BRAFi monotherapy, even in combination with MEK inhibition, as evidenced by the detection of activating mutations in KRAS and NRAS. Likely activating mutations in MAP2K1, GNAS, and GNA11 were only detected in patients who progressed to BRAFi, while activating mutations in key effectors of the PI3K pathway were evidenced at PD to both monotherapy and the combination therapy. Finally, likely activating alterations in U2AF1, IDH1, and CTNNB1 were observed at PD to both monotherapy and combination therapy, in patients who presented rather favorable outcomes on BRAF-TT.
In acquired BRAFi resistance, the proportion of cases harboring MAPK-reactivating alterations was greater than that harboring PI3K-AKT–activating mutations. KRAS mutations were detected at PD on BRAFi (KRASG12C) or BRAFi/MEK inhibitor (MEKi, KRASG12V, and KRASQ61R) treatment, in-line with reports on 3 patients with BRAF-mutant NSCLC (Supplementary Table S9; refs. 25–27). In BRAFV600E melanoma, KRASG12C and KRASQ61H were associated with PD on BRAFi (28), the latter being also detected at PD on BRAFi/MEKi combination (29). Comparably, KRASG12D/C/V/R, KRASQ61H/L, and KRASG13D were observed in patients with colorectal cancer upon PD on BRAFi/MEKi/EGFRi triple combination (30, 31). Importantly, exogenous expression of KRASG13D in VACO432 colorectal cancer cells recapitulated resistance to combined BRAFi/MEKi (32). Moreover, in BRAFV600E-mutant colorectal cancer cells, ERK inhibition suppressed MAPK signaling despite expression of KRASG12D or KRASG13D (32).
P57 presented heterogeneous potential mechanisms of resistance to the combination therapy (Supplementary Fig. S10C), including NRASQ61K. NRASQ61K is often detected in patients with melanoma resistant to BRAFi (28, 33–37) and is found at PD on BRAFi/MEKi (29, 38). In this setting, overexpression of NRASQ61K in vitro conferred resistance to single-agent BRAFi or combined BRAFi/MEKi (29, 34, 35).
We detected an activating mutation in GNASR201C in P24 at PD on vemurafenib. This mutation is known to promote intestinal tumorigenesis in vivo through activation of Wnt and ERK1/2 MAPK pathways (39). GNASR201H (P52, PD on dabrafenib) has been also identified in a patient-derived xenograft (PDX) from a melanoma case who progressed under BRAFi (40). MAP2K1C121S (P35, PD on vemurafenib) is linked with PD to BRAFi alone or in combination with MEKi, in patients with melanoma (28, 41–43). Overexpression of MEK1C121S in BRAFV600E melanoma cell lines (e.g., A375) conferred profound resistance to BRAFi alone or in combination with MEKi, with >100-fold changes in GI50 (half-maximal inhibitor concentration). The resistant phenotype caused by this mutant was reverted in vitro with the ERK inhibitor, VRT11E (42). In advanced uveal melanoma, mutant GNA11 induces constitutive activation of (ERK)1/2 and YES-associated protein (44). Uveal melanoma cells with a Gα-protein mutation (GNAQ or GNA11) are mildly sensitive to MEKi, but completely resistant to BRAFi. The Akt inhibitor, MK2206, sensitized Gα-mutant cells to MEKi, but did not overcome the BRAFi resistance (45).
Other alterations in NRAS (28, 29, 31, 33, 38, 42) and MAP2K1 (28–30, 33, 38, 40, 42, 43) are described at PD to BRAF-TT in patients with BRAF-driven melanoma or in patients with colorectal cancer (Supplementary Table S9), but were not detected in this study.
As demonstrated in BRAF-mutant colorectal cancer (30), targeting a convergent signaling node (i.e., MAPK signaling) as upfront therapy, can suppress the outgrowth of heterogeneous clones harboring a variety of clinically relevant resistance alterations. This might be a promising approach for future clinical trials in BRAF-mutant NSCLC.
The potential activation of the PI3K pathway, prior to therapy and at acquired resistance, was evidenced by the presence of PI3KCA, AKT1, PTEN, or PPP2R1A mutations. Similar to our observations, AKT1E17K was recently reported prior to BRAF/MEK inhibitor therapy in 2 patients with BRAF-mutant NSCLC who subsequently presented oligoprogression after 5.2 and 3 months on treatment (25). This alteration was discovered in progressive tumors under BRAF-TT in patients with BRAF-mutant melanoma (28, 38). Of note, stable overexpression of AKT1E17K in the M229 BRAFV600E human melanoma cell line conferred vemurafenib resistance (28). PIK3CA-activating mutations and alterations in PTEN, a suppressor of PI3K activity, were associated with a decreased response to BRAF-TT, similar to what has been observed in EGFR-mutant NSCLC (21). Of note, a preliminary analysis of a subgroup of patients from the phase II dabrafenib plus trametinib study showed that PIK3CA-activating mutations are associated with poor PFS and OS (46).
In BRAFV600E melanoma, genomic alterations in PIK3CA and PTEN have been implicated in resistance to BRAF-TT (Supplementary Table S9; refs. 28, 36, 42, 47, 48), and PTEN loss is associated with shorter median PFS in patients treated with dabrafenib (36). In vitro, overexpression of PIK3CAE545K (28) or PTEN knockdown (28, 42, 47) conferred resistance to BRAFi monotherapy in melanoma models. In addition, PTEN knockdown also contributes to resistance to combined BRAFi/MEKi (29). Functional analysis of melanoma cells from the subcutaneous metastasis (37) revealed that PIK3CAH1047R clones are resistant to BRAFi, and combined treatment of BRAFi with either GDC-0941 or LY294002 (PI3K inhibitors) only slightly improved growth inhibition in two of the three clones. Mao and colleagues (49) showed that colorectal cancer cell lines with mutations in PTEN or PIK3CA were less sensitive to BRAFi. Combined BRAFi/PI3K inhibitor treatment resulted in synergistic growth inhibition in BRAF-mutant colorectal cancer cells with both primary and secondary resistance, counter to the results obtained in melanoma models (37).
In vitro studies clearly show the potential of using AKT, pan-PI3K, or dual PI3K/mTOR inhibitors in combination with BRAF/MEK inhibitor; however, the relevance of these therapeutic strategies is hampered by their enhanced systemic toxicities (50). For cases presenting concurrent K/NRAS and PIK3CA mutations at PD (e.g., P57), preclinical evidence in melanoma PDXs suggests that triple combination encorafenib/binimetinib/BKM120 or double combination VX-11e/BKM120 could improve tumor growth control (40).
This study also demonstrated the presence of concurrent alterations in protein kinases and signal transducers, in addition to BRAF, at baseline or at PD to BRAF-TT. Coalterations in FGFR2, ERBB2, CTNNB1, IDH1, and UA2F1 have been previously reported in treatment-naïve EGFR-mutant NSCLC (21). Interestingly, lentiviral expression library screens in melanoma cell lines revealed that FGFR2 is a resistance effector of vemurafenib alone or in combination with MEK inhibition (51). HER2-mediated bypass signaling and CTNNB1 mutations are described as resistance mechanisms in ROS1-driven NSCLC (52). Blakely and colleagues (21) showed that CTNNB1S37F, an alteration found in P59 in our study, promoted an increase in the number of invading cells and sustained AKT signaling in EGFR-mutant NSCLC cells treated with EGFR inhibitors (21). In BRAFV600E melanoma, CTNNB1S45F was reported in a PDX from a patient who progressed to vemurafenib (40), and CTNNB1 alterations in codons outside S37 and S34 were seen exclusively in BRAFi-resistant tumors (i.e., undetected in pretreatment tumors; Supplementary Table S9; ref. 42). The development of therapeutic strategies to target these alterations is limited to preclinical evidence suggesting that targeting the HER3/HER2 pathway will likely improve the efficacy of BRAFis and might help overcome resistance (53).
In our study, IDH1 mutations were detected in ctDNA at baseline (P2) and at PD on BRAF-TT (P27, P68) in patients who presented a rather long therapeutic benefit. IDHR132C is linked with resistance to BRAFi alone or in combination with MEKi in melanoma (38, 40), and has been described as an oncogenic driver in hematologic and central nervous system malignancies (54, 55). Genomic alterations in UA2F1 are known to be critical for RNA binding activity (56) and have been reported to confer resistance to clinically relevant doses of X-rays (57). Finally, inactivation of PP2A through siRNA-mediated silencing of the A subunit results in MEKi resistance in KRAS-mutant lung cancer cell lines (58).
These results raise the possibility that BRAF-mutant NSCLC tumors may have elaborated as-yet uncharacterized genetic BRAF/MEK resistance mechanisms. Future functional studies will investigate the molecular mechanisms by which mutations in FGFR2, CTNNB1, PPP2R1A, U2AF1, and other candidate genes can contribute to resistance to BRAF/MEK inhibition.
Our results are consistent with what has been shown in BRAF-mutant melanoma, where the main mechanisms of cross-resistance to both monotherapy and combination therapy involve reactivation of the MAPK pathway (28, 42, 43), followed by parallel activation of the PI3K/AKT pathway (59). However, unlike what has been reported in melanoma, we did not detect on-target alterations (i.e., BRAF amplification or BRAF splicing variants; refs. 28, 42, 43) at PD in BRAF-mutant NSCLC. To our knowledge, mutations in PPP2R1A, U2AF1, and NFE2L2, found in our study, are not reported in progressive BRAF-mutant melanoma. The presence of genomic alterations in MAP2K2, NF1, PIK3R2, AKT3, RICTOR, or RAC1 (28, 29, 33, 42, 43) could not be assessed because these genes were not covered by the sequencing panel used in our study.
The limitations of this study that may impact the interpretation of the survival analyses include the heterogeneity of sample collection and treatment regimens, as well as the lack of RECIST tumor response evaluation for 5 patients for which PD was determined according to the investigator's criteria. Furthermore, our study is limited to a relatively small cohort of patients that display either intrinsic or acquired potential mechanisms of resistance to BRAF-TT. Larger, prospective cohorts will be needed to validate our observations about the role of these concurrent alterations in predicting response to treatment.
The use of broader gene panels (60) or whole-exome ctDNA sequencing (61, 62), might widen the information obtained on the presence of concurrent alterations at diagnosis or on the emergence of de novo mutations in these patients under therapy. ctDNA can be derived from lesions that were not or cannot be biopsied and may provide the means to capture the molecular heterogeneity present in multiple tumor lesions in an individual patient (Supplementary Fig. S10; ref. 63). However, the sequencing of matched tumor samples could contribute to determine the concordance in the detection of genomic alterations in tumor and plasma samples (64) and could further illuminate the processes behind tumor evolution and resistance to therapy (e.g., truncal or subclonal nature of PIK3CA, FGFR2, or KRAS mutations). In addition, sequencing of white blood cells would inform on the detection of any false-positive mutations that appear to be somatic, but may indeed be a result of clonal hematopoiesis.
To our knowledge, there is currently a lack of translational NSCLC models to functionally evaluate candidate resistance effectors found in patients treated with BRAF-TT and to study precision medicine approaches to tackle these resistance mechanisms. The HCC364 cell line, the only available in vitro model of BRAFV600E lung adenocarcinoma, displays limited sensitivity to BRAFi monotherapy (IC50 = 800 nmol/L; ref. 65). In contrast, BRAF-mutant melanoma cell lines (e.g., M229, M238, and M249) display higher, and in some cases exquisite (e.g., A375), sensitivity to BRAF-TTs (28, 35, 66). The generation of alterative NSCLC patient-derived models (e.g., PDXs; ref. 40) that pertinently mirror the sensitivity to BRAF-TT observed in patients might provide a relevant tool to functionally characterize the role of these alterations in resistance to BRAF-TT and to identify efficacious combination therapies to delay or revert the emergence of resistance.
Although the mechanistic relevance of the genomic alterations described here was not evaluated in this study, in silico structural modeling might serve as a complementary tool to predict the potential role of these alterations in resistance to therapy. Our study paves the way for investigating the biological and clinical impact of the concurrent genetic alterations observed in treatment-naïve and treatment-resistant patients with BRAF-mutant NSCLC.
This work underlines the ability of genomic ctDNA profiling to characterize the alterations that might potentially impact treatment response to BRAF-TT, using a minimally invasive approach. Further work is needed to determine the biological relevance and the full clinical utility of these mutations in ctDNA in patients with BRAF-mutant lung cancer. Finally, subsequent studies are needed to evaluate the integration of ctDNA profiling in the routine setting in this patient population, and in other BRAF-mutant tumors treated with BRAF-TT.
Disclosure of Potential Conflicts of Interest
L. Mezquita reports grants and personal fees from Roche Diagnostics (consulting, advisory role), Bristol-Myers Squibb, Tecnofarma, Roche, AstraZeneca (lectures and educational activities), Chugai, Roche (travel, accommodations, expenses), and AstraZeneca (mentorship program with key opinion leaders funded) outside the submitted work. J. Mazieres reports grants from Novartis during the conduct of the study, and grants from Roche, AstraZeneca, BMS, MSD, and Pierre Fabre outside the submitted work. Y. Loriot reports personal fees and non-financial support from Roche, Sanofi, and AstraZeneca; personal fees from Astellas, Seattle Genetics, BMS, Pfizer, and Ipsen; and grants, personal fees, and non-financial support from Janssen and MSD outside the submitted work. V. Westeel reports personal fees and other from Roche (support for meeting attendance), BMS (support for meeting attendance), AstraZeneca (support for meeting attendance), and MSD (support for meeting attendance); personal fees from Takeda; and other from Pfizer (support for meeting attendance) and Boehringer Ingelheim (support for meeting attendance) outside the submitted work. C. Tissot reports personal fees from AstraZeneca, Bristol-Myers Squibb, MSD, and Roche outside the submitted work. E. Brain reports grants from Amgen, HalioDX, and TEVA; personal fees and other from AstraZeneca (travel support), Pfizer (travel support), and Roche (travel support); other from Novartis (travel support), Pierre Fabre (travel support), and Sandoz (travel support); and personal fees from BMS, Celgene, Clinigen, G1 Therapeutics, Hospira, Janssen, Mylan, OBI Pharma, PUMA, and Samsung outside the submitted work. I. Monnet reports non-financial support from Pfizer (congress ESMO 2019) and Roche (congress ESMO 2018) outside the submitted work. E. Giroux Leprieur reports personal fees and non-financial support from AstraZeneca and Boehringer-Ingelheim; grants, personal fees, and non-financial support from Bristol-Myers Squibb, Roche, and MSD; and personal fees from Novartis outside the submitted work. C. Caramella reports personal fees from BMS, MSD, Pfizer, and AstraZeneca outside the submitted work. F. de Kievit reports employment with Inivata Ltd. K. Howarth reports other from Inivata Ltd (employment and shareholder) during the conduct of the study and outside the submitted work. C. Morris reports other from Inivata (employment and stock holder) during the conduct of the study. E. Green reports personal fees from Inivata (employment and shareholder) during the conduct of the study and outside the submitted work. L. Friboulet reports grants from Incyte and DebioPharm outside the submitted work. M. Perol reports grants from Novartis and AstraZeneca and non-financial support from Inivata during the conduct of the study, as well as grants, personal fees, and non-financial support from Roche, AstraZeneca, Boehringer Ingelheim, and Takeda, personal fees and non-financial support from Pfizer, Bristol-Myers Squibb, MSD, and Chugai, and personal fees from Amgen and Lilly outside the submitted work. B. Besse reports grants from AbbVie, Amgen, AstraZeneca, BeiGene, Blueprint Medicines, BMS, Boehringer Ingelheim, Celgene, Cristal Therapeutics, Daiichi-Sankyo, Eli Lilly, GSK, Ignyta, IPSEN, Inivata, Janssen, Merck KGaA, MSD, Nektar, Onxeo, OSE Immunotherapeutics, Pfizer, PharmaMar, Roche-Genentech, Sanofi, Servier, Spectrum Pharmaceuticals, Takeda, Tiziana Pharma, and Tolero Pharmaceuticals during the conduct of the study. J.-Y. Blay reports grants and personal fees from Roche, Novartis, and GSK during the conduct of the study, as well as grants and personal fees from BMS, MSD, Deciphera, and PharmaMar outside the submitted work. P. Saintigny reports grants and non-financial support from AstraZeneca (grant and travel) during the conduct of the study; grants from Roche (grant and travel) and Novartis (grant); other from HTG Diagnostics (research and development) and Cellenion (research and development); and grants and non-financial support from Hitachi (biomarker development) outside the submitted work. D. Planchard reports non-financial support from Inivata Ltd (institutional contracted research) during the conduct of the study; personal fees from AstraZeneca, Bristol-Myers Squibb, Boehringer Ingelheim, Celgene, Daiichi Sankyo, Eli Lilly, Merck, Novartis, Pfizer, prIME Oncology, Peer CME, Roche, and Samsung (consulting, advisory role, or lectures) outside the submitted work; clinical trials research as principal or coinvestigator (institutional financial interests) from AstraZeneca, Bristol-Myers Squibb, Boehringer Ingelheim, Eli Lilly, Merck, Novartis, Pfizer, Roche, Medimmune, Sanofi-Aventis, Taiho Pharma, Novocure, and Daiichi Sankyo; and travel, accommodations, expenses from AstraZeneca, Roche, Novartis, prIME Oncology, and Pfizer. No potential conflicts of interest were disclosed by the other authors.
S. Ortiz-Cuaran: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. L. Mezquita: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. A. Swalduz: Resources, data curation, formal analysis, investigation, methodology. M. Aldea: Resources, data curation, investigation, methodology. J. Mazieres: Conceptualization, resources, data curation, formal analysis, investigation, writing-review and editing. C. Leonce: Resources, data curation, investigation. C. Jovelet: Resources, data curation, investigation. A. Pradines: Resources, data curation, formal analysis, investigation. V. Avrillon: Resources, data curation, investigation. W.R. Chumbi Flores: Resources, data curation, investigation. L. Lacroix: Resources, data curation, investigation. Y. Loriot: Resources, data curation, investigation. V. Westeel: Resources, data curation, investigation. M. Ngo-Camus: Resources, data curation, investigation. C. Tissot: Resources, data curation, formal analysis, investigation, project administration. C. Raynaud: Resources, data curation, investigation. R. Gervais: Resources, data curation, investigation. E. Brain: Resources, data curation, investigation. I. Monnet: Resources, data curation, investigation. E. Giroux Leprieur: Resources, data curation, investigation. C. Caramella: Resources, data curation, investigation. C. Mahier-Aït Oukhatar: Resources, data curation, investigation. N. Hoog-Labouret: Resources, data curation, investigation, project administration. F. de Kievit: Resources, data curation, formal analysis, investigation, methodology. K. Howarth: Conceptualization, resources, data curation, formal analysis, investigation, visualization, methodology, writing-review and editing. C. Morris: Resources, data curation, investigation, methodology. E. Green: Conceptualization, resources, data curation, formal analysis, investigation, methodology, writing-original draft. L. Friboulet: Resources, data curation, investigation. S. Chabaud: Resources, data curation, formal analysis, investigation, visualization, methodology. J.-F. Guichou: Resources, data curation, formal analysis, investigation, visualization, methodology. M. Perol: Conceptualization, resources, data curation, investigation, methodology, writing-review and editing. B. Besse: Conceptualization, resources, investigation, methodology. J.-Y. Blay: Resources, data curation, investigation, writing-review and editing. P. Saintigny: Conceptualization, resources, formal analysis, investigation, methodology, writing-review and editing. D. Planchard: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, writing-original draft, writing-review and editing.
S. Ortiz-Cuaran was supported by a research grant from Fondation ARC (PJA 2017–1206573). L. Mezquita was recipient of IASLC Translational Research Fellowship 2018, ESMO Translational Research Fellowship 2019, and SEOM Retorno de Investigadores 2019. This work was also supported by funding from the Integrated Cancer Research Site LYriCAN (INCa-DGOS-Inserm_12563 to J.-Y. Blay, P. Saintigny, and S. Ortiz-Cuaran) and by the Agence National de la Recherche (ANR-10-BINF-03–03 to J-.F. Guichou). We are grateful to the patients and families involved in this study. The authors thank the investigator teams from AcSé Vemurafenib, Unicancer, and the Respiratory Disease Department from the Larrey Hospital (University Hospital of Toulouse); the biobanks from the Centre Léon Bérard (BB-0033-00050, CRB Centre Léon Bérard, Lyon, France) and the University Hospital of Toulouse (DC-2011-1382, AC-2013-1984, and CNIL 1727608); as well as the team of the Liquid Biopsy Program at Gustave Roussy (M. Ngo-Camus, C. Nicotra, K. Kais, S. Bellahoues-Senane) and the secretary team of the Thoracic Oncology Group (L. Mezquita, C. Caramella, and L. Soulier) for the support. The authors also thank the teams involved in data collection and analyses at each center, and particularly Camille Schiffler at the Department of Clinical Research (Centre Léon Bérard).
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.