Between 2 April 2019 and 5 August 2020, 465 patients were randomized in a 2:1 ratio to either the toripalimab group (n = 309) or the placebo group (n = 156). Stratification was performed based on PD-L1 expression status, histological classification (SCC vs. non-SCC), and smoking status (Fig. 1). Baseline demographic and disease characteristics were evenly distributed across both treatment groups (Table 1).

CONSORT diagram of the CHOICE-01 study. Between April 2, 2019 and August 5, 2020, 835 patients with advanced non-small-cell lung cancer were screened for eligibility from 59 centers across China. 465 were enrolled and randomly assigned in a 2:1 ratio to the toripalimab plus standard first-line chemotherapy group (n = 309) or the placebo plus standard first-line chemotherapy group (n = 156). By the cutoff date of August 31, 2022, one patient from the toripalimab group and seven patients from the placebo group were still under study treatment. Among the patients who ended study treatment, 127 patients from the toripalimab group and 45 patients from the placebo group were still under study. Among patients in the placebo group, 81 crossed over to toripalimab treatment after disease progression. ITT, intention-to-treat; PPS, per protocol set; SS, safety set

As of the data cutoff date of August 31, 2022, in the intention-to-treat (ITT) population, for one in the toripalimab group and seven in the placebo group (who crossed over to receive toripalimab as per the protocol), had completed all required study treatments. Notably, 45 patients (14.6%) in the toripalimab group completed the full two-year treatment course. Furthermore, 81 patients (51.9%) in the placebo group were switched to active crossover treatment following disease progression as assessed by the investigator, with 74% of these patients initiating treatment within three weeks. In total, 17.2% (53/309) of toripalimab patients and 65.4% (102/156) of placebo patients received subsequent anti-PD-1/PD-L1 therapy, including both active crossover and post-study treatments (Supplementary Table 1).

As noted in previous reports, the interim analysis had exceeded the pre-specified efficacy boundary,5 rendering the pre-specified final analysis descriptive. By the cutoff date of August 31, 2022, with a median survival follow-up of 21.2 months, the final analysis of overall survival (OS) was triggered by 283 death events. Consistent with interim results, a notable enhancement in OS was observed in the toripalimab group than that in the placebo group [median OS: 23.8 vs. 17.0 months, hazard ratio (HR) 0.69 (95% CI: 0.57–0.93), nominal P = 0.01]. The 1-, 2-, and 3-year OS rates were 74.0% vs. 72.8%, 49.8% vs. 37.5%, and 32.5% vs. 18.4%, respectively (Fig. 2a). The effect of toripalimab on OS was consistent across all major subgroups, with favorable outcomes observed in toripalimab-treated patients (Fig. 2b). Within subgroups based on PD-L1 expression (TC < 1% (n = 139), 1–49% (n = 203), ≥50% (n = 100), and not evaluable (n = 23)), the HRs were 0.79 (95% CI: 0.52–1.24), 0.72 (95% CI: 0.51–1.03), 0.91 (95% CI: 0.49–1.80), and 0.40 (95% CI: 0.15–1.10), respectively (Supplementary Fig. 1).

OS in the Intention-to-Treat Population. a The Kaplan-Meier estimates of OS comparing the toripalimab plus chemotherapy group with the placebo plus chemotherapy group as first-line treatment for patients with advanced NSCLC. Censored patients are marked with “┃” in the graph. Numbers of patients at risk at indicated time points are shown below the x-axis. Number of events, median OS, 1-, 2- and 3-year OS rates, and stratified hazard ratio for death are shown to the right of Kaplan-Meier curves. b Overall survivals in key subgroups. All hazard ratios were computed using the Cox proportional hazards model. All P values were two-sided with no adjustment of multiplicity. The P values of comparing the Kaplan-Meier curves were computed using the log-rank test stratified by the baseline PD-L1 expression status, histology, and smoking status. The P values of testing the interaction of the subgroup variables with the treatment were computed using the Cox proportional hazards regression model with the treatment group, the subgroup variable and their interaction as the covariates. HR, hazard ratio; NA, not available; NSCLC, non–small-cell lung cancer; PD-L1, programmed death ligand-1; PFS, progression-free survival

A significant survival benefit was seen in the non-SCC subgroup [median OS 27.8 vs. 15.9 months, HR 0.49 (95% CI: 0.35–0.69), P < 0.001]. However, no such difference was observed in the SCC subgroup, where the final OS analysis showed no statistically significant difference [median OS 19.6 vs. 18.1 months, HR 1.09 (95% CI: 0.77–1.56), P = 0.65] (Supplementary Fig. 2).

As of August 31, 2022, patients in the toripalimab group received a median of 9 cycles of treatment, while those in the placebo group received a median of 8 cycles (Supplementary Table 2). No new safety concerns emerged following prolonged toripalimab exposure for up to two years. The incidence of Grade ≥3 adverse events (AEs) (78.9% vs. 82.1%) was comparable between the two groups (78.9% vs. 82.1%), as was the rate of serious adverse events (SAEs) (46.4% vs. 35.3%). AEs leading to treatment discontinuation were more frequent in the toripalimab group (15.3% vs. 3.2%), as were investigator-assessed immune-related adverse events (irAEs) (50.6% vs. 21.2%) and Grade ≥3 irAEs (16.9% vs. 3.2%) (Supplementary Table 3). Supplementary Table 4 lists the most common AEs observed. Increased alanine aminotransferase (ALT), hypo- and hyperthyroidism, diarrhea, and rash were more frequently reported in the toripalimab group. The irAEs were consistent with those seen in patients treated with checkpoint inhibitors (Supplementary Table 5).

Among the 465 eligible enrolled patients, whole exome sequencing was successfully performed on tumor tissue samples from 394 patients as previously described.15 Following the updated OS analysis, patients with mutations in the FA-PI3K-Akt pathway and IL-7 signaling pathway were associated with improved OS in the toripalimab group (interaction P = 0.006, and 0.001 respectively) compared to the interim analysis results (Supplementary Fig. 3). To further validate the previous findings from tumor tissue profiling, ctDNA were extracted from 460 pretreatment plasma samples and subjected to deep panel-based sequencing (520-gene panel). Among them, 387 patients (84.1%) tested positive for somatic alterations (ctDNA positive). The absence of baseline ctDNA correlated with improved PFS and OS but did not predict a response to toripalimab treatment (Supplementary Fig. 4).

Notably, the overall mutational landscape of ctDNA greatly exhibited a striking resemblance to the results obtained from tissue-based analysis. The mutation prevalence of the top mutated genes from ctDNA profiling was highly correlated with that from the tissue profiling (rho=0.81, P < 3.7e-13, Fig. 3a) in matched tissue and pre-treatment blood samples. Mutations of individual genes were also strongly concordant between ctDNA and tissue samples (median Cohen’s Kappa: 0.71, median PPV: 0.82, Fig. 3b) although the mutation frequencies of the majority of genes were slightly lower in ctDNA likely due to a lower amount of tumor DNA present in the blood compared to the tumor tissue. Similarly, bTMB also exhibited a strong correlation with tissue-based TMB (Supplementary Fig. 5). Patients with high bTMB showed significantly better PFS in the toripalimab group (interaction P = 0.02) (Fig. 3c). High TMB—whether measured in tissue or blood—was associated with significantly longer PFS in the toripalimab group only among non-squamous NSCLC patients (Supplementary Fig. 6), but did not have predictive value for OS (Supplementary Fig. 7)

The concordance of mutational landscape between ctDNA and tissue WES data. a The matched frequency of top mutated genes derived from blood-based ctDNA and tumor WES data, ordered by the prevalence rate in ctDNA. b The mutations of top mutated genes between ctDNA profiling and matched tissue were strongly concordant. cThe Kaplan-Meier estimates of PFS stratified by bTMB status in ITT patients. Low ctDNA shedders (<5% tumor fraction) were removed to ensure the specimens had subclone mutations adequately captured by the sequencing assay. ctDNA, circulating tumor DNA; WES, whole exome sequencing; TMB, tumor mutational burden; bTMB, blood tumor mutational burden; PFS, progression-free survival; ITT, intention-to-trea;JS001: toripalimab

Gene Set Enrichment Analysis on the ctDNA data was also performed independently on significantly interacted genes, as was done in tissue WES data,15 to identify over-represented biologic pathways. The FA-PI3K-Ak and IL-7 signaling pathways remained in the top enriched pathways despite the number of genes being substantially limited by the designed panel (520 genes in total) (Supplementary Fig. 8). Altered FA-PI3K-Akt and IL-7 signaling pathways by ctDNA analysis were also predictive of PFS (interaction P = 0.008 and 0.002 respectively) (Supplementary Fig. 8b, c), even though some genes were absent in the NGS panel compared with WES. Notably, mutations in the FA-PI3K-Akt and IL-7 signaling pathways displayed minimal predictive value for SCC (Supplementary Fig. 3), and we observed a distinct biomarker profile in SCC (Supplementary Fig. 9).

To explore the potential of dynamic ctDNA profiling, we conducted ctDNA analysis on blood samples collected at cycle 3, day 1 (C3D1) from 113 patients (71 in the toripalimab group and 42 in the placebo group). Interestingly, 82.1% (32/39) of the ctDNA-negative patients in the toripalimab group showed a clinical response, while only 57.1% (8/14) in the placebo group responded (Fig. 4a). Among the 32 responders in the toripalimab group, 56.3% (18/32) demonstrated a response at the initial RECIST evaluation, suggesting that ctDNA changes occurred prior to radiographic identification of clinical responses. Furthermore, among the 32 responders in the toripalimab arm, only 56.3% (18/32) displayed a response at the initial RECIST evaluation, suggesting that ctDNA changes preceded radiographic identification of clinical responses in some cases. Significantly, tumor burden, as assessed by the tumor fraction in blood, notably decreased at C3D1 in both toripalimab and placebo arms (nominal P < 0.001), with a more pronounced reduction observed in the toripalimab arm(Fig. 4b). Moreover, a significant reduction in tumor burden (defined as either a six-fold decrease or reaching an undetectable level) predicted improved PFS and OS for toripalimab-treated patients in both the ITT and non-SCC populations (Fig. 4c, d and Supplementary Fig. 10a–e).

Analyses of ctDNA dynamics at C3D1. a Distribution of tumor load changes in the two arms and association with the best overall response. Tumor load dynamics were categorized into four groups: Negative in both baseline and C3D1 timepoints(BothNeg), Decrease at C3D1(Decrease), Increase at C3D1(Increase), and Reduction to undetectable levels at C3D1(ToNeg). The decrease in the toripalimab arm was more prominent. A higher percentage of patients exhibiting partial response was also observed in the toripalimab group. b Tumor load, as estimated by tumor fraction in blood, decreased at C3D1 for both toripalimab and placebo arms. Response of ctDNA at C3D1 is predictive of PFS (c) and OS (d) in ITT patients. The ctDNA response was defined as >85% reduction of tumor load from C1D1 or no detection of ctDNA at C3D1. ctDNA, circulating tumor DNA, ITT, intention-to-treat. JS001: toripalimab

To further explore the mechanism possible leads to the survival different between SCC and non-SCC, we analyzed the genomic profiling and ctDNA dynamics. We observed distinct mutational patterns between SCC and non-SCC for the top mutated genes (Supplementary Fig. 11a, b). More patients in SCC harbored PIK3CA, PTEN, and NFE2L2. By contrast, non-SCC patients carried more mutations of KRAS, SMARCA4, and STK11. In addition, patients with SCC had higher ctDNA+ rate at the baseline comparing to ones with non-SCC. Furthermore, reduction of ctDNA at C3D1 was more frequent in SCC placebo group (79.0%, 15 out of 19) than in non-squamous placebo group (56.5%, 13 out of 23) (Supplementary Fig. 11c).