Entrectinib in ROS1-positive advanced non-small cell lung cancer: the phase 2/3 BFAST trial

Targeted therapies have improved outcomes in oncogene-driven advanced NSCLC and are standard of care, making comprehensive biomarker testing essential before first-line treatment. However, real-world uptake of broad NGS testing remains suboptimal, and tissue availability, feasibility, and yield can limit molecular testing. Liquid biopsies analyzing circulating tumor DNA offer a faster, less invasive alternative with high tissue–liquid concordance and demonstrated clinical utility, and can also inform mechanisms of resistance. ROS1 rearrangements occur in ~1–2% of NSCLC, with frequent brain metastases, highlighting the need for CNS-penetrant therapies. Entrectinib is a potent ROS1/TRK/ALK TKI designed for CNS penetration and has shown durable systemic and intracranial activity in tissue-selected ROS1-positive NSCLC. The research question for BFAST Cohort D was whether efficacy and safety of entrectinib in ROS1-positive NSCLC identified solely by liquid biopsy are consistent with outcomes from tissue-based selection, thereby supporting liquid biopsy as a basis for first-line targeted therapy selection.

Prior integrated analyses across ALKA-372-001, STARTRK-1, and STARTRK-2 reported ORR ~67% with median DoR ~20.4 months and median PFS ~16.8 months in ROS1-positive NSCLC treated with entrectinib, including durable intracranial responses. Liquid biopsies have shown high concordance with tissue testing, faster turnaround, and clinical benefit, and can elucidate resistance mechanisms. Brain metastases occur in ~40% of ROS1-positive NSCLC, necessitating CNS-active agents. Other ROS1 inhibitors include crizotinib (ORR ~72%), and next-generation inhibitors (lorlatinib, taletrectinib, repotrectinib) with emerging CNS activity, though cross-trial comparisons are limited and typically tissue-selected. Prior reports variably suggested prognostic roles for specific fusion partners (CD74, EZR) and TP53 co-mutations; evidence is mixed regarding their impact on entrectinib outcomes. Studies also indicate ctDNA dynamics, including early clearance, may correlate with improved outcomes across settings.

Design: BFAST (NCT03178552) is a global, open-label, multicohort trial enrolling patients with advanced/metastatic NSCLC harboring actionable alterations detected by liquid biopsy alone. Cohort D evaluated entrectinib in treatment-naive ROS1-positive NSCLC. Eligibility: Age ≥18; previously untreated, unresectable stage IIIB/IV NSCLC; ECOG PS 0–2; life expectancy ≥12 weeks; measurable disease by RECIST v1.1. Asymptomatic and/or previously treated brain metastases allowed; brain radiotherapy must have been completed ≥14 days pre-treatment. Prior (neo/adjuvant) therapy allowed if treatment-free ≥6 months. Biomarker screening: Central blood-based NGS using FoundationOne Liquid CDx clinical trial assay or FoundationACT (Foundation Medicine). ROS1 rearrangements defined as ROS1 fusions with known partners (any frame) or in-frame fusions with novel partners, with breakpoints before the kinase domain. Enrollment was based solely on liquid biopsy; tissue testing was not required. Treatment: Entrectinib 600 mg orally once daily until RECIST v1.1 progression, unacceptable toxicity, withdrawal, sponsor termination, or death. Tumor assessments at baseline and every 8 weeks. Dose reductions in 200 mg steps; treatment interruptions up to 28 days permitted. Brain imaging beyond baseline was not mandated in patients without baseline CNS disease. Endpoints: Primary—confirmed ORR per investigator (RECIST v1.1; confirmation required). Secondary—CBR, DoR, PFS per investigator; ORR, CBR, DoR, PFS per independent review facility (IRF); OS; time to CNS progression (by investigator and IRF); and safety. Exploratory—investigator-assessed ORR in patients with baseline CNS metastases and multiple post hoc biomarker analyses (fusion partners, TP53 status, circulating tumor fraction (cTF), ROS1 ctDNA clearance, resistance mechanisms). Statistical analysis: Sample size targeted preservation of 75% of the historical ORR from integrated tissue-selected analyses; with n=54 measurable, ORR ≥70.4% (95% CI: 56.0–82.0) was required to meet the primary endpoint. Kaplan–Meier methods estimated medians and 95% CIs for DoR, PFS, OS, and time to CNS progression. Assay concordance between FoundationACT and FoundationOne Liquid CDx was assessed via positive predictive agreement. Clinical analyses used SAS v9.04; exploratory biomarker analyses used R v3.5.2. Safety: AEs graded per NCI CTCAE v4.0. All patients receiving ≥1 dose comprised the safety population.

Screening and population: From 11 Jan 2018 to 9 Dec 2020, 5,220 patients were screened; 92 (1.8%) had ROS1-positive NSCLC by liquid biopsy. Fifty-five treatment-naive patients were enrolled; 54 had measurable disease. Median age 56 years (22–83); 58% female; 75% never-smokers. Four patients (7.3%) had asymptomatic and/or previously treated CNS metastases at baseline (investigator). Efficacy (primary and secondary): In 54 measurable patients, confirmed ORR was 81.5% (44/54; 95% CI: 68.6–90.8) by both investigator and IRF. By investigator: CR 3.7% (2/54), PR 77.8% (42/54); SD 13.0% (7/54); PD 5.6% (3/54). CBR 87.0% (47/54; 95% CI: 75.1–94.6). Median DoR 13.0 months (95% CI: 6.3–18.4) by investigator; 16.7 months (95% CI: 5.6–24.0) by IRF. Median PFS (n=55) 12.9 months (95% CI: 8.7–18.5) by investigator; 14.8 months (95% CI: 7.2–24.0) by IRF. OS data immature; 12-month OS 79.0%; Kaplan–Meier median OS 31.2 months (95% CI: 20.2–NE). Time to CNS progression (n=54): median not reached; 12-month CNS progression-free rate 83.5% (investigator) and 86.4% (IRF). In 4 patients with baseline CNS metastases, 2 achieved PR. Safety (n=55): Median treatment duration 12.8 months (range 1–33). Patients with ≥1 treatment-related AE (TRAE): 92.7% (51/55); serious TRAEs: 12.7% (7/55). Grade 3–5 AEs in 56.4% (31/55); most common grade ≥3 AE: weight increase 7.3% (all grade 3). Two grade 5 AEs (COVID-19; unexplained) deemed not related to study treatment. TRAE-related dose reduction 36.4%, interruption 20.0%, discontinuation 5.5%. Median dose intensity 97.5% (range 31.8–103.2). Biomarker findings: High assay concordance—96.9% positive predictive agreement between FoundationOne Liquid CDx and FoundationACT. ROS1 fusion partners (n=55): CD74 56.4% (31), EZR 23.6% (13), TPM3 7.3% (4), ROS1 self-rearrangement 3.6% (2), and others (FAM91A1, LRIG3, RFC4, SDC4, ZCCHC8; each 1.8%). Outcomes did not differ between CD74 versus non-CD74, or EZR versus non-EZR fusion partners. Co-mutations (n=54): TP53 most common (40.7%, 22/54); mTP53 associated with numerically shorter DoR and PFS versus wtTP53. Baseline circulating tumor fraction (cTF) was not associated with DoR or PFS; weak positive correlation between baseline cTF and tumor burden (SLD) (Pearson r=0.3, P=0.031). ctDNA (ROS1) clearance by cycle 3 day 1 (subset with samples): 86.1% (31/36) cleared; clearance associated with longer median DoR and PFS versus no clearance. Resistance mechanisms at progression (n=20 evaluable): 14 (70%) had ROS1 fusions detectable at discontinuation; emerging mutations in 12 patients (30 total; 29 unique), including ROS1 G2032R in 2 patients and other alterations (e.g., ROS1 F2004L/F2004C).

Cohort D achieved its primary objective, demonstrating an investigator-assessed ORR of 81.5%—above the protocol-defined consistency threshold—and consistent with prior integrated analyses of entrectinib in tissue-selected ROS1-positive NSCLC. Durability of response and PFS were in the range of historical tissue-selected data, supporting that liquid biopsy–based identification can effectively guide first-line entrectinib therapy. Although intracranial efficacy could not be robustly assessed due to the low incidence of baseline CNS metastases, time to CNS progression findings suggest potential for delaying CNS events; definitive conclusions require further data and dedicated assessments. Comparisons with other ROS1 inhibitors are limited by differences in study populations and selection methods; ongoing randomized head-to-head evaluation of entrectinib versus crizotinib (NCT04603807) should clarify comparative systemic and intracranial efficacy. Exploratory biomarker analyses suggest that early ROS1 ctDNA clearance may be prognostic for improved outcomes, whereas baseline cTF was not prognostic in this cohort. The frequent TP53 co-mutation was associated with numerically poorer outcomes, aligning with prior reports, though prevalence may be influenced by ctDNA selection and potential CHIP contributions. Identified resistance mutations, including ROS1 G2032R, are consistent with known mechanisms of on-target resistance to ROS1 TKIs; broader genomic profiling at progression may inform sequencing strategies.

Entrectinib demonstrated high response rates, durable disease control, and a manageable safety profile in treatment-naive, ROS1-positive advanced/metastatic NSCLC identified solely via liquid biopsy, with outcomes consistent with tissue-based selection. These results support incorporating liquid biopsy into clinical workflows to expedite biomarker-driven decisions when tissue is limited or turnaround time is critical. Future research should include randomized comparisons with other ROS1 inhibitors (including CNS endpoints), validation of ctDNA clearance as a prognostic biomarker, systematic characterization of resistance mechanisms, and evaluation of longitudinal ctDNA monitoring to anticipate progression and guide therapy sequencing.

Key limitations include the single-arm design without a comparator, small sample size, and relatively short follow-up (median 18.3 months; last patient ~13 months), with 27% still on treatment at cutoff. Intracranial efficacy could not be definitively assessed due to few patients with baseline CNS disease and non-mandated CNS follow-up beyond baseline for those without CNS metastases. Two different liquid biopsy assays (FoundationOne Liquid CDx and FoundationACT) were used, potentially introducing variability, although high concordance (96.9%) was observed. Liquid biopsy sensitivity depends on tumor shedding; patients with low tumor burden may be ctDNA-negative and not assessable by this method. Selection based on detectable ctDNA may enrich for higher tumor burden or certain co-mutations (e.g., TP53), potentially affecting generalizability.