For the MDACC Cohort, genomic profiling information was obtained through Clinical Laboratory Improvement Amendments certified laboratory assay results annotated in the GEMINI database. For the CGDB and FMI biomarker cohorts, the information was obtained from respective data repositories and sequencing was performed by FoundationOne assay.

Oncogene alterations included in this study are: BRAF: codons 457–723, including V600E and non-V600E; EGFR exon 20: codons 762–823 (T790M mutation not included); HER2: codons 755 and 770–785; KRAS: codons 12, 13, 33, and 61; classic EGFR: exon 19 deletions and exon 21 p.L858R mutation (±T790M mutation); MET: exon 14 skipping mutations; and ALK, ROS1, and RET: gene fusion/rearrangements.

For the CGDB cohorts and for the FMI biomarker cohort, TMB was determined through tissue-based next-generation sequencing (FoundationOne). TMB was defined as all mutations (including all non-synonymous and synonymous mutations with germline variants and driver mutations excluded) divided by the total covered exonic regions16 and presented as number of mutations/megabase (Mb). High TMB is defined as ≥10 mutations/Mb, corresponding to FMI diagnostic test definition of high TMB or ≥16 mutations/Mb based on OAK and POPLAR trials.17 TMB data was not available for the MDACC cohort.

For the MDACC cohort, PD-L1 staining was obtained from pathology reports. For the CGDB cohorts, PD-L1 status was obtained from data repository. For the FMI biomarker cohort, PD-L1 staining was assessed through immunohistochemistry using the 22C3 PharmDx assay (Dako North America). PD-L1 expression was quantified by tumor proportional score (TPS) and defined as positive (≥1%) or negative (<1%), and high (≥50%) or low (1%–49%).5 6 18

Progression-free survival (PFS) was defined as time from starting treatment until disease progression or death. Overall survival (OS) was defined as time from starting treatment until death. Patients without an event were censored at last follow-up. Left truncation method was applied to real-world PFS and OS analysis to adjust for situations where molecular profiling occurred after the start of treatment.19

Kaplan-Meier method was used to estimate PFS and OS, and differences were assessed by log-rank test. HRs and 95% CIs were determined through Cox proportional hazards model. Multivariable analysis was performed using the Cox regression method. Best response to ICB treatment determined through RECIST V.1.1 was available only for MDACC Cohort. The KRAS group was used as the reference comparator because: (1) it was the most commonly mutated oncogene in these cohorts of NSCLCs; and (2) KRAS-mutant NSCLCs constitute a heterogeneous patient population that have been shown to have similar prognosis compared with KRAS wild-type patients20–23 and to derive similar benefit from ICB therapy when compared with KRAS wild-type NSCLC patients.1 24

Differences in categorical variables were assessed through χ² or Fisher’s exact test with Bonferroni method for between group comparisons. Continuous variables were tested for normality using Shapiro-Wilk test, and Dunn’s test with Benjamini-Hochberg correction was used for between group comparisons. Significance was established at p value ≤0.05. Statistical analysis was performed on GraphPad Prism V.7.0 (La Jolla, California, USA), IBM SPSS Statistics V.24.0 (Armonk, New York, USA), SAS (Cary, North Carolina, USA), and RStudio V.3.5.5 (Boston, Massachusetts, USA).

We first analyzed the CGDB immunotherapy cohort (median follow-up 17.8 months (95% CI 15.6 to 19.8)) to assess the predictive performance of TMB or PD-L1 for benefit from ICB in oncogene-driven NSCLC. We observed that higher TMB using either 10 or 16 mut/Mb as the cut-off was associated with longer PFS (TMB ≥10 vs <10 mut/Mb: HR 0.61, 95% CI 0.50 to 0.75, p<0.001; TMB ≥16 vs <16 mut/Mb: HR 0.52, 95% CI 0.39 to 0.69, p<0.001), and longer OS (TMB ≥10 vs <10 mut/Mb: HR 0.79, 95% CI 0.64 to 0.97, p<0.001; TMB ≥16 vs <16 mut/Mb: HR 0.69, 95% CI 0.51 to 0.92, p<0.001) consistent with previous findings.7 8 10 17 Positive PD-L1 expression was associated with longer PFS (PD-L1 positive vs negative: HR 0.48, 95% CI 0.30 to 0.77, p<0.001), but the difference did not reach statistical significance for OS (PD-L1 positive vs negative: HR 0.66, 95% CI 0.39 to 1.10, p=0.095). In a multivariable analysis, higher TMB was significantly associated with longer PFS (TMB ≥10 vs <10 mut/Mb: HR 0.66, 95% CI 0.53 to 0.82, p<0.001; TMB ≥16 vs <16 mut/Mb: HR 0.56, 95% CI 0.42 to 0.76, p<0.001) (online supplemental table 1A,B), and longer OS (TMB ≥10 vs <10 mut/Mb: HR 0.76, 95% CI 0.61 to 0.96, p=0.019; TMB ≥16 vs <16 mut/Mb: HR 0.66, 95% CI 0.49 to 0.89, p=0.006) (online supplemental table 2A,B). PD-L1 expression was also associated with longer PFS (TMB cut-off of 10: PD-L1 positive vs negative: HR 0.57, 95% CI 0.35 to 0.92, p=0.022) (online supplemental table 1A).

We next assessed the impact of oncogene alterations on clinical outcome from ICB treatment using 1066 oncogene-altered advanced NSCLC patients (n=172 from MDACC cohort: median follow-up 29.1 months (95% CI 26.2 to 33.6); and n=894 from CGDB immunotherapy cohort: median follow-up 17.8 months (95% CI 15.6 to 19.8)) (online supplemental figure 1A,B). As outlined in online supplemental table 3, BRAF and KRAS groups were enriched for smokers (p<0.001) in both MDACC cohort and CGDB immunotherapy cohort, while patients in the classic EGFR group received more prior treatment lines (p<0.001) compared with other genomic groups, likely due to higher likelihood of prior exposure to TKIs (p<0.001). Since KRAS-mutant NSCLC constituted the largest population in both MDACC cohort (n=87) and CGDB immunotherapy cohort (n=601) and have been previously shown to have similar prognosis compared with KRAS wild-type patients20–23 and to derive similar benefit from ICB therapy when compared with KRAS wild-type NSCLC patients,1 24 we used the KRAS group as the reference for outcome analysis.

In the MDACC cohort, the BRAF group had the longest PFS among all genomic groups (BRAF 7.4 vs KRAS 2.8 months, HR 0.36, 95% CI 0.14 to 0.88, p=0.026) (figure 1A), while classic EGFR and HER2 groups had the shortest PFS (1.8 and 1.9 months, respectively) (figure 1A). The same trend remained in subsequent multivariable analysis adjusting for PD-L1 expression and smoking status (TMB data were not available for MDACC cohort) (p=0.032 for BRAF group, p=0.022 for Classic EGFR group, and p=0.116 for HER-2 group, online supplemental table 4). Compared with the KRAS group, OS was numerically longer in BRAF group, but the difference was not statistically significant (35.6 vs 16.8 months, HR 0.65, 95% CI (0.26 to 1.63), p=0.363). Meanwhile patients with classic EGFR mutations had the shortest OS (11.3 months, 95% CI 5.8 to 16.9, HR 2.01, p=0.006) (figure 1B). Similar trends were observed in a multivariable analysis adjusting for PD-L1 expression, although the difference was not significant (BRAF vs KRAS: HR 0.69, p=0.454; classic EGFR vs KRAS: HR 1.49, p=0.263; online supplemental table 5). Furthermore, the BRAF group also had highest objective response rate (ORR) (62%) among all oncogene groups (Fisher’s exact test p<0.001). Compared with the KRAS group, the common comparator, this difference was not statistically significant (62% vs 24%, p=0.364) (figure 1C). Importantly, two of three patients with BRAF V600E mutations had recently progressed on vemurafenib and achieved durable responses at 35+ and 20+ months post treatment on ICB. There were seven patients harboring ALK fusion (n=2), RET fusion (n=2) or MET exon 14 skipping mutations (n=3) in the MDACC cohort. Except for two patients with MET mutations who had stable disease, all remaining five patients had progressive disease as best response. Results for PFS, OS and ORR are summarized in table 1.

Consistent with the results from the MDACC Cohort, the BRAF groups in the CGDB immunotherapy cohort had the longest PFS (V600E 9.8 and non-V600E 5.4 months) (figure 1D and table 1) and OS (V600E 20.8 and non-V600E 14.9 months) (figure 1E and table 1) among all genomic groups although the difference did not reach statistical difference. Meanwhile, patients with classic EGFR mutations had significantly shorter PFS (figure 1D and table 1). Similar to classic EGFR group, patients with oncogene fusions (ALK: n=19; ROS1: n=3; RET: n=14) had poor outcomes with a trend for shorter PFS compared with the KRAS group. The MET exon 14 mutation group (n=34) also had a short PFS of 2.7 months, but the difference was not significant compared with the KRAS group (HR 1.13, 95% CI 0.69 to 1.84, p=0.59) (figure 1D and table 1). In a multivariable analysis adjusting for TMB, PD-L1 expression, and smoking status, BRAF V600E had longer PFS compared with the KRAS group (TMB cut-off of 10: HR 0.58, 95% CI 0.35 to 0.98, p=0.041; TMB cut-off of 16: HR 0.63, 95% CI 0.38 to 1.05, p=0.074) (online supplemental table 1A,B).

KRAS-mutant group is the largest group in our cohorts, and we therefore assessed whether different KRAS-mutant alleles are associated with distinct clinical outcome from ICB treatment. As shown in online supplemental figure 2A–E, ORR and survival were similar between G12C, G12D and G12V subgroups.

To understand whether the impact of these oncogene alterations was only present on clinical outcome from ICB, we analyzed the CGDB chemotherapy cohort (n=933) for clinical outcomes on treatment with chemotherapy across the same NSCLC oncogene groups (online supplemental figure 1C, online supplemental table 6). These analyses demonstrated that the fusion group (ALK: n=39; ROS1: n=12; RET: n=14) had the longest OS with chemotherapy (27.2 months in fusion group vs 11.7 months in KRAS group, HR 0.50, 95% CI 0.34 to 0.74, p<0.001) (online supplemental figure 3A). This difference remained significant in a multivariable analysis adjusting for prior TKI therapy (HR 0.48, 95% CI 0.33 to 0.71, p<0.001). Otherwise, PFS or OS was not different between oncogene groups (online supplemental figure 3A,B).

We leveraged the 4017 NSCLC patients from FMI biomarker cohort to assess the impact of oncogene alterations on PD-L1 expression and TMB. High prevalence of PD-L1 positivity (TPS ≥1%) was observed in MET (115/145, 79.3%), RET (42/54, 77.8%), BRAF V600E (89/118, 75.4%), ROS1 (40/55, 72.7%) and ALK (136/194, 70.1%) groups, and low prevalence was observed in tumors harboring EGFR exon 20 (75/166, 45.2%), HER2 (50/105, 47.6%), classic EGFR (exon 19 deletion and exon 21 L858R) (368/732, 50.3%), and BRAF non-V600E (116/208, 55.8%). Compared with KRAS-mutant NSCLCs, MET-mutant tumors had higher PD-L1 positivity rate (TPS ≥1%), while EGFR-mutant tumors had significantly lower PD-L1 positivity rate (figure 2A) consistent with our previous findings.25 Furthermore, high PD-L1 expression (TPS ≥50%) was common in tumors harboring alterations in MET (80/145, 55.2%), BRAF V600E (57/118, 48.3%), ROS1 fusion (23/55, 41.8%) and RET fusion (20/54, 37.0%). Among these, tumors harboring MET alterations had significantly higher prevalence of high PD-L1 expression (TPS ≥50%), while classic EGFR, EGFR exon 20, and HER2 groups had lower prevalence of high PD-L1 compared with KRAS group (pairwise comparison vs KRAS p<0.050) (figure 2B).

Compared with the KRAS group, TMB was higher for BRAF non-V600E group (9.6 mut/Mb (n=208) vs 7.8 mut/Mb (n=2240); adjusted p=0.003), while all other oncogene groups had significantly lower TMB (adjusted p<0.001). Of note, ALK (n=194), classic EGFR (n=732), HER2 (n=105), RET (n=54), and ROS1 (n=55) groups had the lowest TMB with median <3 mut/Mb (figure 2C and online supplemental table 7).